Initial commit; kernel source import

arch/tile/kernel/Makefile

@@ -0,0 +1,25 @@
+#
+# Makefile for the Linux/TILE kernel.
+#
+
+extra-y := vmlinux.lds head_$(BITS).o
+obj-y := backtrace.o entry.o irq.o messaging.o \
+	pci-dma.o proc.o process.o ptrace.o reboot.o \
+	setup.o signal.o single_step.o stack.o sys.o \
+	sysfs.o time.o traps.o vdso.o \
+	intvec_$(BITS).o regs_$(BITS).o tile-desc_$(BITS).o
+
+obj-$(CONFIG_HARDWALL)		+= hardwall.o
+obj-$(CONFIG_COMPAT)		+= compat.o compat_signal.o
+obj-$(CONFIG_SMP)		+= smpboot.o smp.o tlb.o
+obj-$(CONFIG_MODULES)		+= module.o
+obj-$(CONFIG_EARLY_PRINTK)	+= early_printk.o
+obj-$(CONFIG_KEXEC)		+= machine_kexec.o relocate_kernel_$(BITS).o
+ifdef CONFIG_TILEGX
+obj-$(CONFIG_PCI)		+= pci_gx.o
+else
+obj-$(CONFIG_PCI)		+= pci.o
+endif
+obj-$(CONFIG_TILE_USB)		+= usb.o
+
+obj-y				+= vdso/

arch/tile/kernel/asm-offsets.c

@@ -0,0 +1,76 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * Generates definitions from c-type structures used by assembly sources.
+ */
+
+#include <linux/kbuild.h>
+#include <linux/thread_info.h>
+#include <linux/sched.h>
+#include <linux/hardirq.h>
+#include <linux/ptrace.h>
+#include <hv/hypervisor.h>
+
+/* Check for compatible compiler early in the build. */
+#ifdef CONFIG_TILEGX
+# ifndef __tilegx__
+#  error Can only build TILE-Gx configurations with tilegx compiler
+# endif
+# ifndef __LP64__
+#  error Must not specify -m32 when building the TILE-Gx kernel
+# endif
+#else
+# ifdef __tilegx__
+#  error Can not build TILEPro/TILE64 configurations with tilegx compiler
+# endif
+#endif
+
+void foo(void)
+{
+	DEFINE(SINGLESTEP_STATE_BUFFER_OFFSET, \
+	       offsetof(struct single_step_state, buffer));
+	DEFINE(SINGLESTEP_STATE_FLAGS_OFFSET, \
+	       offsetof(struct single_step_state, flags));
+	DEFINE(SINGLESTEP_STATE_ORIG_PC_OFFSET, \
+	       offsetof(struct single_step_state, orig_pc));
+	DEFINE(SINGLESTEP_STATE_NEXT_PC_OFFSET, \
+	       offsetof(struct single_step_state, next_pc));
+	DEFINE(SINGLESTEP_STATE_BRANCH_NEXT_PC_OFFSET, \
+	       offsetof(struct single_step_state, branch_next_pc));
+	DEFINE(SINGLESTEP_STATE_UPDATE_VALUE_OFFSET, \
+	       offsetof(struct single_step_state, update_value));
+
+	DEFINE(THREAD_INFO_TASK_OFFSET, \
+	       offsetof(struct thread_info, task));
+	DEFINE(THREAD_INFO_FLAGS_OFFSET, \
+	       offsetof(struct thread_info, flags));
+	DEFINE(THREAD_INFO_STATUS_OFFSET, \
+	       offsetof(struct thread_info, status));
+	DEFINE(THREAD_INFO_HOMECACHE_CPU_OFFSET, \
+	       offsetof(struct thread_info, homecache_cpu));
+	DEFINE(THREAD_INFO_STEP_STATE_OFFSET, \
+	       offsetof(struct thread_info, step_state));
+
+	DEFINE(TASK_STRUCT_THREAD_KSP_OFFSET,
+	       offsetof(struct task_struct, thread.ksp));
+	DEFINE(TASK_STRUCT_THREAD_PC_OFFSET,
+	       offsetof(struct task_struct, thread.pc));
+
+	DEFINE(HV_TOPOLOGY_WIDTH_OFFSET, \
+	       offsetof(HV_Topology, width));
+	DEFINE(HV_TOPOLOGY_HEIGHT_OFFSET, \
+	       offsetof(HV_Topology, height));
+
+	DEFINE(IRQ_CPUSTAT_SYSCALL_COUNT_OFFSET, \
+	       offsetof(irq_cpustat_t, irq_syscall_count));
+}

arch/tile/kernel/backtrace.c

@@ -0,0 +1,683 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <asm/byteorder.h>
+#include <asm/backtrace.h>
+#include <asm/tile-desc.h>
+#include <arch/abi.h>
+
+#ifdef __tilegx__
+#define TILE_MAX_INSTRUCTIONS_PER_BUNDLE TILEGX_MAX_INSTRUCTIONS_PER_BUNDLE
+#define tile_decoded_instruction tilegx_decoded_instruction
+#define tile_mnemonic tilegx_mnemonic
+#define parse_insn_tile parse_insn_tilegx
+#define TILE_OPC_IRET TILEGX_OPC_IRET
+#define TILE_OPC_ADDI TILEGX_OPC_ADDI
+#define TILE_OPC_ADDLI TILEGX_OPC_ADDLI
+#define TILE_OPC_INFO TILEGX_OPC_INFO
+#define TILE_OPC_INFOL TILEGX_OPC_INFOL
+#define TILE_OPC_JRP TILEGX_OPC_JRP
+#define TILE_OPC_MOVE TILEGX_OPC_MOVE
+#define OPCODE_STORE TILEGX_OPC_ST
+typedef long long bt_int_reg_t;
+#else
+#define TILE_MAX_INSTRUCTIONS_PER_BUNDLE TILEPRO_MAX_INSTRUCTIONS_PER_BUNDLE
+#define tile_decoded_instruction tilepro_decoded_instruction
+#define tile_mnemonic tilepro_mnemonic
+#define parse_insn_tile parse_insn_tilepro
+#define TILE_OPC_IRET TILEPRO_OPC_IRET
+#define TILE_OPC_ADDI TILEPRO_OPC_ADDI
+#define TILE_OPC_ADDLI TILEPRO_OPC_ADDLI
+#define TILE_OPC_INFO TILEPRO_OPC_INFO
+#define TILE_OPC_INFOL TILEPRO_OPC_INFOL
+#define TILE_OPC_JRP TILEPRO_OPC_JRP
+#define TILE_OPC_MOVE TILEPRO_OPC_MOVE
+#define OPCODE_STORE TILEPRO_OPC_SW
+typedef int bt_int_reg_t;
+#endif
+
+/* A decoded bundle used for backtracer analysis. */
+struct BacktraceBundle {
+	tile_bundle_bits bits;
+	int num_insns;
+	struct tile_decoded_instruction
+	insns[TILE_MAX_INSTRUCTIONS_PER_BUNDLE];
+};
+
+
+/* Locates an instruction inside the given bundle that
+ * has the specified mnemonic, and whose first 'num_operands_to_match'
+ * operands exactly match those in 'operand_values'.
+ */
+static const struct tile_decoded_instruction *find_matching_insn(
+	const struct BacktraceBundle *bundle,
+	tile_mnemonic mnemonic,
+	const int *operand_values,
+	int num_operands_to_match)
+{
+	int i, j;
+	bool match;
+
+	for (i = 0; i < bundle->num_insns; i++) {
+		const struct tile_decoded_instruction *insn =
+			&bundle->insns[i];
+
+		if (insn->opcode->mnemonic != mnemonic)
+			continue;
+
+		match = true;
+		for (j = 0; j < num_operands_to_match; j++) {
+			if (operand_values[j] != insn->operand_values[j]) {
+				match = false;
+				break;
+			}
+		}
+
+		if (match)
+			return insn;
+	}
+
+	return NULL;
+}
+
+/* Does this bundle contain an 'iret' instruction? */
+static inline bool bt_has_iret(const struct BacktraceBundle *bundle)
+{
+	return find_matching_insn(bundle, TILE_OPC_IRET, NULL, 0) != NULL;
+}
+
+/* Does this bundle contain an 'addi sp, sp, OFFSET' or
+ * 'addli sp, sp, OFFSET' instruction, and if so, what is OFFSET?
+ */
+static bool bt_has_addi_sp(const struct BacktraceBundle *bundle, int *adjust)
+{
+	static const int vals[2] = { TREG_SP, TREG_SP };
+
+	const struct tile_decoded_instruction *insn =
+		find_matching_insn(bundle, TILE_OPC_ADDI, vals, 2);
+	if (insn == NULL)
+		insn = find_matching_insn(bundle, TILE_OPC_ADDLI, vals, 2);
+#ifdef __tilegx__
+	if (insn == NULL)
+		insn = find_matching_insn(bundle, TILEGX_OPC_ADDXLI, vals, 2);
+	if (insn == NULL)
+		insn = find_matching_insn(bundle, TILEGX_OPC_ADDXI, vals, 2);
+#endif
+	if (insn == NULL)
+		return false;
+
+	*adjust = insn->operand_values[2];
+	return true;
+}
+
+/* Does this bundle contain any 'info OP' or 'infol OP'
+ * instruction, and if so, what are their OP?  Note that OP is interpreted
+ * as an unsigned value by this code since that's what the caller wants.
+ * Returns the number of info ops found.
+ */
+static int bt_get_info_ops(const struct BacktraceBundle *bundle,
+		int operands[MAX_INFO_OPS_PER_BUNDLE])
+{
+	int num_ops = 0;
+	int i;
+
+	for (i = 0; i < bundle->num_insns; i++) {
+		const struct tile_decoded_instruction *insn =
+			&bundle->insns[i];
+
+		if (insn->opcode->mnemonic == TILE_OPC_INFO ||
+		    insn->opcode->mnemonic == TILE_OPC_INFOL) {
+			operands[num_ops++] = insn->operand_values[0];
+		}
+	}
+
+	return num_ops;
+}
+
+/* Does this bundle contain a jrp instruction, and if so, to which
+ * register is it jumping?
+ */
+static bool bt_has_jrp(const struct BacktraceBundle *bundle, int *target_reg)
+{
+	const struct tile_decoded_instruction *insn =
+		find_matching_insn(bundle, TILE_OPC_JRP, NULL, 0);
+	if (insn == NULL)
+		return false;
+
+	*target_reg = insn->operand_values[0];
+	return true;
+}
+
+/* Does this bundle modify the specified register in any way? */
+static bool bt_modifies_reg(const struct BacktraceBundle *bundle, int reg)
+{
+	int i, j;
+	for (i = 0; i < bundle->num_insns; i++) {
+		const struct tile_decoded_instruction *insn =
+			&bundle->insns[i];
+
+		if (insn->opcode->implicitly_written_register == reg)
+			return true;
+
+		for (j = 0; j < insn->opcode->num_operands; j++)
+			if (insn->operands[j]->is_dest_reg &&
+			    insn->operand_values[j] == reg)
+				return true;
+	}
+
+	return false;
+}
+
+/* Does this bundle modify sp? */
+static inline bool bt_modifies_sp(const struct BacktraceBundle *bundle)
+{
+	return bt_modifies_reg(bundle, TREG_SP);
+}
+
+/* Does this bundle modify lr? */
+static inline bool bt_modifies_lr(const struct BacktraceBundle *bundle)
+{
+	return bt_modifies_reg(bundle, TREG_LR);
+}
+
+/* Does this bundle contain the instruction 'move fp, sp'? */
+static inline bool bt_has_move_r52_sp(const struct BacktraceBundle *bundle)
+{
+	static const int vals[2] = { 52, TREG_SP };
+	return find_matching_insn(bundle, TILE_OPC_MOVE, vals, 2) != NULL;
+}
+
+/* Does this bundle contain a store of lr to sp? */
+static inline bool bt_has_sw_sp_lr(const struct BacktraceBundle *bundle)
+{
+	static const int vals[2] = { TREG_SP, TREG_LR };
+	return find_matching_insn(bundle, OPCODE_STORE, vals, 2) != NULL;
+}
+
+#ifdef __tilegx__
+/* Track moveli values placed into registers. */
+static inline void bt_update_moveli(const struct BacktraceBundle *bundle,
+				    int moveli_args[])
+{
+	int i;
+	for (i = 0; i < bundle->num_insns; i++) {
+		const struct tile_decoded_instruction *insn =
+			&bundle->insns[i];
+
+		if (insn->opcode->mnemonic == TILEGX_OPC_MOVELI) {
+			int reg = insn->operand_values[0];
+			moveli_args[reg] = insn->operand_values[1];
+		}
+	}
+}
+
+/* Does this bundle contain an 'add sp, sp, reg' instruction
+ * from a register that we saw a moveli into, and if so, what
+ * is the value in the register?
+ */
+static bool bt_has_add_sp(const struct BacktraceBundle *bundle, int *adjust,
+			  int moveli_args[])
+{
+	static const int vals[2] = { TREG_SP, TREG_SP };
+
+	const struct tile_decoded_instruction *insn =
+		find_matching_insn(bundle, TILEGX_OPC_ADDX, vals, 2);
+	if (insn) {
+		int reg = insn->operand_values[2];
+		if (moveli_args[reg]) {
+			*adjust = moveli_args[reg];
+			return true;
+		}
+	}
+	return false;
+}
+#endif
+
+/* Locates the caller's PC and SP for a program starting at the
+ * given address.
+ */
+static void find_caller_pc_and_caller_sp(CallerLocation *location,
+					 const unsigned long start_pc,
+					 BacktraceMemoryReader read_memory_func,
+					 void *read_memory_func_extra)
+{
+	/* Have we explicitly decided what the sp is,
+	 * rather than just the default?
+	 */
+	bool sp_determined = false;
+
+	/* Has any bundle seen so far modified lr? */
+	bool lr_modified = false;
+
+	/* Have we seen a move from sp to fp? */
+	bool sp_moved_to_r52 = false;
+
+	/* Have we seen a terminating bundle? */
+	bool seen_terminating_bundle = false;
+
+	/* Cut down on round-trip reading overhead by reading several
+	 * bundles at a time.
+	 */
+	tile_bundle_bits prefetched_bundles[32];
+	int num_bundles_prefetched = 0;
+	int next_bundle = 0;
+	unsigned long pc;
+
+#ifdef __tilegx__
+	/* Naively try to track moveli values to support addx for -m32. */
+	int moveli_args[TILEGX_NUM_REGISTERS] = { 0 };
+#endif
+
+	/* Default to assuming that the caller's sp is the current sp.
+	 * This is necessary to handle the case where we start backtracing
+	 * right at the end of the epilog.
+	 */
+	location->sp_location = SP_LOC_OFFSET;
+	location->sp_offset = 0;
+
+	/* Default to having no idea where the caller PC is. */
+	location->pc_location = PC_LOC_UNKNOWN;
+
+	/* Don't even try if the PC is not aligned. */
+	if (start_pc % TILE_BUNDLE_ALIGNMENT_IN_BYTES != 0)
+		return;
+
+	for (pc = start_pc;; pc += sizeof(tile_bundle_bits)) {
+
+		struct BacktraceBundle bundle;
+		int num_info_ops, info_operands[MAX_INFO_OPS_PER_BUNDLE];
+		int one_ago, jrp_reg;
+		bool has_jrp;
+
+		if (next_bundle >= num_bundles_prefetched) {
+			/* Prefetch some bytes, but don't cross a page
+			 * boundary since that might cause a read failure we
+			 * don't care about if we only need the first few
+			 * bytes. Note: we don't care what the actual page
+			 * size is; using the minimum possible page size will
+			 * prevent any problems.
+			 */
+			unsigned int bytes_to_prefetch = 4096 - (pc & 4095);
+			if (bytes_to_prefetch > sizeof prefetched_bundles)
+				bytes_to_prefetch = sizeof prefetched_bundles;
+
+			if (!read_memory_func(prefetched_bundles, pc,
+					      bytes_to_prefetch,
+					      read_memory_func_extra)) {
+				if (pc == start_pc) {
+					/* The program probably called a bad
+					 * address, such as a NULL pointer.
+					 * So treat this as if we are at the
+					 * start of the function prolog so the
+					 * backtrace will show how we got here.
+					 */
+					location->pc_location = PC_LOC_IN_LR;
+					return;
+				}
+
+				/* Unreadable address. Give up. */
+				break;
+			}
+
+			next_bundle = 0;
+			num_bundles_prefetched =
+				bytes_to_prefetch / sizeof(tile_bundle_bits);
+		}
+
+		/*
+		 * Decode the next bundle.
+		 * TILE always stores instruction bundles in little-endian
+		 * mode, even when the chip is running in big-endian mode.
+		 */
+		bundle.bits = le64_to_cpu(prefetched_bundles[next_bundle++]);
+		bundle.num_insns =
+			parse_insn_tile(bundle.bits, pc, bundle.insns);
+		num_info_ops = bt_get_info_ops(&bundle, info_operands);
+
+		/* First look at any one_ago info ops if they are interesting,
+		 * since they should shadow any non-one-ago info ops.
+		 */
+		for (one_ago = (pc != start_pc) ? 1 : 0;
+		     one_ago >= 0; one_ago--) {
+			int i;
+			for (i = 0; i < num_info_ops; i++) {
+				int info_operand = info_operands[i];
+				if (info_operand < CALLER_UNKNOWN_BASE)	{
+					/* Weird; reserved value, ignore it. */
+					continue;
+				}
+
+				/* Skip info ops which are not in the
+				 * "one_ago" mode we want right now.
+				 */
+				if (((info_operand & ONE_BUNDLE_AGO_FLAG) != 0)
+				    != (one_ago != 0))
+					continue;
+
+				/* Clear the flag to make later checking
+				 * easier. */
+				info_operand &= ~ONE_BUNDLE_AGO_FLAG;
+
+				/* Default to looking at PC_IN_LR_FLAG. */
+				if (info_operand & PC_IN_LR_FLAG)
+					location->pc_location =
+						PC_LOC_IN_LR;
+				else
+					location->pc_location =
+						PC_LOC_ON_STACK;
+
+				switch (info_operand) {
+				case CALLER_UNKNOWN_BASE:
+					location->pc_location = PC_LOC_UNKNOWN;
+					location->sp_location = SP_LOC_UNKNOWN;
+					return;
+
+				case CALLER_SP_IN_R52_BASE:
+				case CALLER_SP_IN_R52_BASE | PC_IN_LR_FLAG:
+					location->sp_location = SP_LOC_IN_R52;
+					return;
+
+				default:
+				{
+					const unsigned int val = info_operand
+						- CALLER_SP_OFFSET_BASE;
+					const unsigned int sp_offset =
+						(val >> NUM_INFO_OP_FLAGS) * 8;
+					if (sp_offset < 32768) {
+						/* This is a properly encoded
+						 * SP offset. */
+						location->sp_location =
+							SP_LOC_OFFSET;
+						location->sp_offset =
+							sp_offset;
+						return;
+					} else {
+						/* This looked like an SP
+						 * offset, but it's outside
+						 * the legal range, so this
+						 * must be an unrecognized
+						 * info operand.  Ignore it.
+						 */
+					}
+				}
+				break;
+				}
+			}
+		}
+
+		if (seen_terminating_bundle) {
+			/* We saw a terminating bundle during the previous
+			 * iteration, so we were only looking for an info op.
+			 */
+			break;
+		}
+
+		if (bundle.bits == 0) {
+			/* Wacky terminating bundle. Stop looping, and hope
+			 * we've already seen enough to find the caller.
+			 */
+			break;
+		}
+
+		/*
+		 * Try to determine caller's SP.
+		 */
+
+		if (!sp_determined) {
+			int adjust;
+			if (bt_has_addi_sp(&bundle, &adjust)
+#ifdef __tilegx__
+			    || bt_has_add_sp(&bundle, &adjust, moveli_args)
+#endif
+				) {
+				location->sp_location = SP_LOC_OFFSET;
+
+				if (adjust <= 0) {
+					/* We are in prolog about to adjust
+					 * SP. */
+					location->sp_offset = 0;
+				} else {
+					/* We are in epilog restoring SP. */
+					location->sp_offset = adjust;
+				}
+
+				sp_determined = true;
+			} else {
+				if (bt_has_move_r52_sp(&bundle)) {
+					/* Maybe in prolog, creating an
+					 * alloca-style frame.  But maybe in
+					 * the middle of a fixed-size frame
+					 * clobbering r52 with SP.
+					 */
+					sp_moved_to_r52 = true;
+				}
+
+				if (bt_modifies_sp(&bundle)) {
+					if (sp_moved_to_r52) {
+						/* We saw SP get saved into
+						 * r52 earlier (or now), which
+						 * must have been in the
+						 * prolog, so we now know that
+						 * SP is still holding the
+						 * caller's sp value.
+						 */
+						location->sp_location =
+							SP_LOC_OFFSET;
+						location->sp_offset = 0;
+					} else {
+						/* Someone must have saved
+						 * aside the caller's SP value
+						 * into r52, so r52 holds the
+						 * current value.
+						 */
+						location->sp_location =
+							SP_LOC_IN_R52;
+					}
+					sp_determined = true;
+				}
+			}
+
+#ifdef __tilegx__
+			/* Track moveli arguments for -m32 mode. */
+			bt_update_moveli(&bundle, moveli_args);
+#endif
+		}
+
+		if (bt_has_iret(&bundle)) {
+			/* This is a terminating bundle. */
+			seen_terminating_bundle = true;
+			continue;
+		}
+
+		/*
+		 * Try to determine caller's PC.
+		 */
+
+		jrp_reg = -1;
+		has_jrp = bt_has_jrp(&bundle, &jrp_reg);
+		if (has_jrp)
+			seen_terminating_bundle = true;
+
+		if (location->pc_location == PC_LOC_UNKNOWN) {
+			if (has_jrp) {
+				if (jrp_reg == TREG_LR && !lr_modified) {
+					/* Looks like a leaf function, or else
+					 * lr is already restored. */
+					location->pc_location =
+						PC_LOC_IN_LR;
+				} else {
+					location->pc_location =
+						PC_LOC_ON_STACK;
+				}
+			} else if (bt_has_sw_sp_lr(&bundle)) {
+				/* In prolog, spilling initial lr to stack. */
+				location->pc_location = PC_LOC_IN_LR;
+			} else if (bt_modifies_lr(&bundle)) {
+				lr_modified = true;
+			}
+		}
+	}
+}
+
+/* Initializes a backtracer to start from the given location.
+ *
+ * If the frame pointer cannot be determined it is set to -1.
+ *
+ * state: The state to be filled in.
+ * read_memory_func: A callback that reads memory.
+ * read_memory_func_extra: An arbitrary argument to read_memory_func.
+ * pc: The current PC.
+ * lr: The current value of the 'lr' register.
+ * sp: The current value of the 'sp' register.
+ * r52: The current value of the 'r52' register.
+ */
+void backtrace_init(BacktraceIterator *state,
+		    BacktraceMemoryReader read_memory_func,
+		    void *read_memory_func_extra,
+		    unsigned long pc, unsigned long lr,
+		    unsigned long sp, unsigned long r52)
+{
+	CallerLocation location;
+	unsigned long fp, initial_frame_caller_pc;
+
+	/* Find out where we are in the initial frame. */
+	find_caller_pc_and_caller_sp(&location, pc,
+				     read_memory_func, read_memory_func_extra);
+
+	switch (location.sp_location) {
+	case SP_LOC_UNKNOWN:
+		/* Give up. */
+		fp = -1;
+		break;
+
+	case SP_LOC_IN_R52:
+		fp = r52;
+		break;
+
+	case SP_LOC_OFFSET:
+		fp = sp + location.sp_offset;
+		break;
+
+	default:
+		/* Give up. */
+		fp = -1;
+		break;
+	}
+
+	/* If the frame pointer is not aligned to the basic word size
+	 * something terrible happened and we should mark it as invalid.
+	 */
+	if (fp % sizeof(bt_int_reg_t) != 0)
+		fp = -1;
+
+	/* -1 means "don't know initial_frame_caller_pc". */
+	initial_frame_caller_pc = -1;
+
+	switch (location.pc_location) {
+	case PC_LOC_UNKNOWN:
+		/* Give up. */
+		fp = -1;
+		break;
+
+	case PC_LOC_IN_LR:
+		if (lr == 0 || lr % TILE_BUNDLE_ALIGNMENT_IN_BYTES != 0) {
+			/* Give up. */
+			fp = -1;
+		} else {
+			initial_frame_caller_pc = lr;
+		}
+		break;
+
+	case PC_LOC_ON_STACK:
+		/* Leave initial_frame_caller_pc as -1,
+		 * meaning check the stack.
+		 */
+		break;
+
+	default:
+		/* Give up. */
+		fp = -1;
+		break;
+	}
+
+	state->pc = pc;
+	state->sp = sp;
+	state->fp = fp;
+	state->initial_frame_caller_pc = initial_frame_caller_pc;
+	state->read_memory_func = read_memory_func;
+	state->read_memory_func_extra = read_memory_func_extra;
+}
+
+/* Handle the case where the register holds more bits than the VA. */
+static bool valid_addr_reg(bt_int_reg_t reg)
+{
+	return ((unsigned long)reg == reg);
+}
+
+/* Advances the backtracing state to the calling frame, returning
+ * true iff successful.
+ */
+bool backtrace_next(BacktraceIterator *state)
+{
+	unsigned long next_fp, next_pc;
+	bt_int_reg_t next_frame[2];
+
+	if (state->fp == -1) {
+		/* No parent frame. */
+		return false;
+	}
+
+	/* Try to read the frame linkage data chaining to the next function. */
+	if (!state->read_memory_func(&next_frame, state->fp, sizeof next_frame,
+				     state->read_memory_func_extra)) {
+		return false;
+	}
+
+	next_fp = next_frame[1];
+	if (!valid_addr_reg(next_frame[1]) ||
+	    next_fp % sizeof(bt_int_reg_t) != 0) {
+		/* Caller's frame pointer is suspect, so give up. */
+		return false;
+	}
+
+	if (state->initial_frame_caller_pc != -1) {
+		/* We must be in the initial stack frame and already know the
+		 * caller PC.
+		 */
+		next_pc = state->initial_frame_caller_pc;
+
+		/* Force reading stack next time, in case we were in the
+		 * initial frame.  We don't do this above just to paranoidly
+		 * avoid changing the struct at all when we return false.
+		 */
+		state->initial_frame_caller_pc = -1;
+	} else {
+		/* Get the caller PC from the frame linkage area. */
+		next_pc = next_frame[0];
+		if (!valid_addr_reg(next_frame[0]) || next_pc == 0 ||
+		    next_pc % TILE_BUNDLE_ALIGNMENT_IN_BYTES != 0) {
+			/* The PC is suspect, so give up. */
+			return false;
+		}
+	}
+
+	/* Update state to become the caller's stack frame. */
+	state->pc = next_pc;
+	state->sp = state->fp;
+	state->fp = next_fp;
+
+	return true;
+}

arch/tile/kernel/compat.c

@@ -0,0 +1,108 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+/* Adjust unistd.h to provide 32-bit numbers and functions. */
+#define __SYSCALL_COMPAT
+
+#include <linux/compat.h>
+#include <linux/syscalls.h>
+#include <linux/kdev_t.h>
+#include <linux/fs.h>
+#include <linux/fcntl.h>
+#include <linux/uaccess.h>
+#include <linux/signal.h>
+#include <asm/syscalls.h>
+
+/*
+ * Syscalls that take 64-bit numbers traditionally take them in 32-bit
+ * "high" and "low" value parts on 32-bit architectures.
+ * In principle, one could imagine passing some register arguments as
+ * fully 64-bit on TILE-Gx in 32-bit mode, but it seems easier to
+ * adapt the usual convention.
+ */
+
+COMPAT_SYSCALL_DEFINE4(truncate64, char __user *, filename, u32, dummy,
+                       u32, low, u32, high)
+{
+	return sys_truncate(filename, ((loff_t)high << 32) | low);
+}
+
+COMPAT_SYSCALL_DEFINE4(ftruncate64, unsigned int, fd, u32, dummy,
+                       u32, low, u32, high)
+{
+	return sys_ftruncate(fd, ((loff_t)high << 32) | low);
+}
+
+COMPAT_SYSCALL_DEFINE6(pread64, unsigned int, fd, char __user *, ubuf,
+                       size_t, count, u32, dummy, u32, low, u32, high)
+{
+	return sys_pread64(fd, ubuf, count, ((loff_t)high << 32) | low);
+}
+
+COMPAT_SYSCALL_DEFINE6(pwrite64, unsigned int, fd, char __user *, ubuf,
+                       size_t, count, u32, dummy, u32, low, u32, high)
+{
+	return sys_pwrite64(fd, ubuf, count, ((loff_t)high << 32) | low);
+}
+
+COMPAT_SYSCALL_DEFINE6(sync_file_range2, int, fd, unsigned int, flags,
+                       u32, offset_lo, u32, offset_hi,
+                       u32, nbytes_lo, u32, nbytes_hi)
+{
+	return sys_sync_file_range(fd, ((loff_t)offset_hi << 32) | offset_lo,
+				   ((loff_t)nbytes_hi << 32) | nbytes_lo,
+				   flags);
+}
+
+COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode,
+                       u32, offset_lo, u32, offset_hi,
+                       u32, len_lo, u32, len_hi)
+{
+	return sys_fallocate(fd, mode, ((loff_t)offset_hi << 32) | offset_lo,
+			     ((loff_t)len_hi << 32) | len_lo);
+}
+
+/*
+ * Avoid bug in generic sys_llseek() that specifies offset_high and
+ * offset_low as "unsigned long", thus making it possible to pass
+ * a sign-extended high 32 bits in offset_low.
+ */
+COMPAT_SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned int, offset_high,
+		       unsigned int, offset_low, loff_t __user *, result,
+		       unsigned int, origin)
+{
+	return sys_llseek(fd, offset_high, offset_low, result, origin);
+}
+ 
+/* Provide the compat syscall number to call mapping. */
+#undef __SYSCALL
+#define __SYSCALL(nr, call) [nr] = (call),
+
+/* See comments in sys.c */
+#define compat_sys_fadvise64_64 sys32_fadvise64_64
+#define compat_sys_readahead sys32_readahead
+#define sys_llseek compat_sys_llseek
+
+/* Call the assembly trampolines where necessary. */
+#define compat_sys_rt_sigreturn _compat_sys_rt_sigreturn
+#define sys_clone _sys_clone
+
+/*
+ * Note that we can't include <linux/unistd.h> here since the header
+ * guard will defeat us; <asm/unistd.h> checks for __SYSCALL as well.
+ */
+void *compat_sys_call_table[__NR_syscalls] = {
+	[0 ... __NR_syscalls-1] = sys_ni_syscall,
+#include <asm/unistd.h>
+};

arch/tile/kernel/compat_signal.c

@@ -0,0 +1,255 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/smp.h>
+#include <linux/kernel.h>
+#include <linux/signal.h>
+#include <linux/errno.h>
+#include <linux/wait.h>
+#include <linux/unistd.h>
+#include <linux/stddef.h>
+#include <linux/personality.h>
+#include <linux/suspend.h>
+#include <linux/ptrace.h>
+#include <linux/elf.h>
+#include <linux/compat.h>
+#include <linux/syscalls.h>
+#include <linux/uaccess.h>
+#include <asm/processor.h>
+#include <asm/ucontext.h>
+#include <asm/sigframe.h>
+#include <asm/syscalls.h>
+#include <asm/vdso.h>
+#include <arch/interrupts.h>
+
+struct compat_ucontext {
+	compat_ulong_t	  uc_flags;
+	compat_uptr_t     uc_link;
+	struct compat_sigaltstack	  uc_stack;
+	struct sigcontext uc_mcontext;
+	sigset_t	  uc_sigmask;	/* mask last for extensibility */
+};
+
+struct compat_rt_sigframe {
+	unsigned char save_area[C_ABI_SAVE_AREA_SIZE]; /* caller save area */
+	struct compat_siginfo info;
+	struct compat_ucontext uc;
+};
+
+int copy_siginfo_to_user32(struct compat_siginfo __user *to, siginfo_t *from)
+{
+	int err;
+
+	if (!access_ok(VERIFY_WRITE, to, sizeof(struct compat_siginfo)))
+		return -EFAULT;
+
+	/* If you change siginfo_t structure, please make sure that
+	   this code is fixed accordingly.
+	   It should never copy any pad contained in the structure
+	   to avoid security leaks, but must copy the generic
+	   3 ints plus the relevant union member.  */
+	err = __put_user(from->si_signo, &to->si_signo);
+	err |= __put_user(from->si_errno, &to->si_errno);
+	err |= __put_user((short)from->si_code, &to->si_code);
+
+	if (from->si_code < 0) {
+		err |= __put_user(from->si_pid, &to->si_pid);
+		err |= __put_user(from->si_uid, &to->si_uid);
+		err |= __put_user(ptr_to_compat(from->si_ptr), &to->si_ptr);
+	} else {
+		/*
+		 * First 32bits of unions are always present:
+		 * si_pid === si_band === si_tid === si_addr(LS half)
+		 */
+		err |= __put_user(from->_sifields._pad[0],
+				  &to->_sifields._pad[0]);
+		switch (from->si_code >> 16) {
+		case __SI_FAULT >> 16:
+			break;
+		case __SI_CHLD >> 16:
+			err |= __put_user(from->si_utime, &to->si_utime);
+			err |= __put_user(from->si_stime, &to->si_stime);
+			err |= __put_user(from->si_status, &to->si_status);
+			/* FALL THROUGH */
+		default:
+		case __SI_KILL >> 16:
+			err |= __put_user(from->si_uid, &to->si_uid);
+			break;
+		case __SI_POLL >> 16:
+			err |= __put_user(from->si_fd, &to->si_fd);
+			break;
+		case __SI_TIMER >> 16:
+			err |= __put_user(from->si_overrun, &to->si_overrun);
+			err |= __put_user(ptr_to_compat(from->si_ptr),
+					  &to->si_ptr);
+			break;
+			 /* This is not generated by the kernel as of now.  */
+		case __SI_RT >> 16:
+		case __SI_MESGQ >> 16:
+			err |= __put_user(from->si_uid, &to->si_uid);
+			err |= __put_user(from->si_int, &to->si_int);
+			break;
+		}
+	}
+	return err;
+}
+
+int copy_siginfo_from_user32(siginfo_t *to, struct compat_siginfo __user *from)
+{
+	int err;
+	u32 ptr32;
+
+	if (!access_ok(VERIFY_READ, from, sizeof(struct compat_siginfo)))
+		return -EFAULT;
+
+	err = __get_user(to->si_signo, &from->si_signo);
+	err |= __get_user(to->si_errno, &from->si_errno);
+	err |= __get_user(to->si_code, &from->si_code);
+
+	err |= __get_user(to->si_pid, &from->si_pid);
+	err |= __get_user(to->si_uid, &from->si_uid);
+	err |= __get_user(ptr32, &from->si_ptr);
+	to->si_ptr = compat_ptr(ptr32);
+
+	return err;
+}
+
+/* The assembly shim for this function arranges to ignore the return value. */
+long compat_sys_rt_sigreturn(void)
+{
+	struct pt_regs *regs = current_pt_regs();
+	struct compat_rt_sigframe __user *frame =
+		(struct compat_rt_sigframe __user *) compat_ptr(regs->sp);
+	sigset_t set;
+
+	if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
+		goto badframe;
+	if (__copy_from_user(&set, &frame->uc.uc_sigmask, sizeof(set)))
+		goto badframe;
+
+	set_current_blocked(&set);
+
+	if (restore_sigcontext(regs, &frame->uc.uc_mcontext))
+		goto badframe;
+
+	if (compat_restore_altstack(&frame->uc.uc_stack))
+		goto badframe;
+
+	return 0;
+
+badframe:
+	signal_fault("bad sigreturn frame", regs, frame, 0);
+	return 0;
+}
+
+/*
+ * Determine which stack to use..
+ */
+static inline void __user *compat_get_sigframe(struct k_sigaction *ka,
+					       struct pt_regs *regs,
+					       size_t frame_size)
+{
+	unsigned long sp;
+
+	/* Default to using normal stack */
+	sp = (unsigned long)compat_ptr(regs->sp);
+
+	/*
+	 * If we are on the alternate signal stack and would overflow
+	 * it, don't.  Return an always-bogus address instead so we
+	 * will die with SIGSEGV.
+	 */
+	if (on_sig_stack(sp) && !likely(on_sig_stack(sp - frame_size)))
+		return (void __user __force *)-1UL;
+
+	/* This is the X/Open sanctioned signal stack switching.  */
+	if (ka->sa.sa_flags & SA_ONSTACK) {
+		if (sas_ss_flags(sp) == 0)
+			sp = current->sas_ss_sp + current->sas_ss_size;
+	}
+
+	sp -= frame_size;
+	/*
+	 * Align the stack pointer according to the TILE ABI,
+	 * i.e. so that on function entry (sp & 15) == 0.
+	 */
+	sp &= -16UL;
+	return (void __user *) sp;
+}
+
+int compat_setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
+			  sigset_t *set, struct pt_regs *regs)
+{
+	unsigned long restorer;
+	struct compat_rt_sigframe __user *frame;
+	int err = 0;
+	int usig;
+
+	frame = compat_get_sigframe(ka, regs, sizeof(*frame));
+
+	if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
+		goto give_sigsegv;
+
+	usig = current_thread_info()->exec_domain
+		&& current_thread_info()->exec_domain->signal_invmap
+		&& sig < 32
+		? current_thread_info()->exec_domain->signal_invmap[sig]
+		: sig;
+
+	/* Always write at least the signal number for the stack backtracer. */
+	if (ka->sa.sa_flags & SA_SIGINFO) {
+		/* At sigreturn time, restore the callee-save registers too. */
+		err |= copy_siginfo_to_user32(&frame->info, info);
+		regs->flags |= PT_FLAGS_RESTORE_REGS;
+	} else {
+		err |= __put_user(info->si_signo, &frame->info.si_signo);
+	}
+
+	/* Create the ucontext.  */
+	err |= __clear_user(&frame->save_area, sizeof(frame->save_area));
+	err |= __put_user(0, &frame->uc.uc_flags);
+	err |= __put_user(0, &frame->uc.uc_link);
+	err |= __compat_save_altstack(&frame->uc.uc_stack, regs->sp);
+	err |= setup_sigcontext(&frame->uc.uc_mcontext, regs);
+	err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
+	if (err)
+		goto give_sigsegv;
+
+	restorer = VDSO_SYM(&__vdso_rt_sigreturn);
+	if (ka->sa.sa_flags & SA_RESTORER)
+		restorer = ptr_to_compat_reg(ka->sa.sa_restorer);
+
+	/*
+	 * Set up registers for signal handler.
+	 * Registers that we don't modify keep the value they had from
+	 * user-space at the time we took the signal.
+	 * We always pass siginfo and mcontext, regardless of SA_SIGINFO,
+	 * since some things rely on this (e.g. glibc's debug/segfault.c).
+	 */
+	regs->pc = ptr_to_compat_reg(ka->sa.sa_handler);
+	regs->ex1 = PL_ICS_EX1(USER_PL, 1); /* set crit sec in handler */
+	regs->sp = ptr_to_compat_reg(frame);
+	regs->lr = restorer;
+	regs->regs[0] = (unsigned long) usig;
+	regs->regs[1] = ptr_to_compat_reg(&frame->info);
+	regs->regs[2] = ptr_to_compat_reg(&frame->uc);
+	regs->flags |= PT_FLAGS_CALLER_SAVES;
+	return 0;
+
+give_sigsegv:
+	signal_fault("bad setup frame", regs, frame, sig);
+	return -EFAULT;
+}

arch/tile/kernel/early_printk.c

@@ -0,0 +1,93 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/console.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/irqflags.h>
+#include <linux/printk.h>
+#include <asm/setup.h>
+#include <hv/hypervisor.h>
+
+static void early_hv_write(struct console *con, const char *s, unsigned n)
+{
+	hv_console_write((HV_VirtAddr) s, n);
+}
+
+static struct console early_hv_console = {
+	.name =		"earlyhv",
+	.write =	early_hv_write,
+	.flags =	CON_PRINTBUFFER,
+	.index =	-1,
+};
+
+/* Direct interface for emergencies */
+static int early_console_complete;
+
+void early_panic(const char *fmt, ...)
+{
+	va_list ap;
+	arch_local_irq_disable_all();
+	va_start(ap, fmt);
+	early_printk("Kernel panic - not syncing: ");
+	early_vprintk(fmt, ap);
+	early_console->write(early_console, "\n", 1);
+	va_end(ap);
+	dump_stack();
+	hv_halt();
+}
+
+static int __initdata keep_early;
+
+static int __init setup_early_printk(char *str)
+{
+	if (early_console)
+		return 1;
+
+	if (str != NULL && strncmp(str, "keep", 4) == 0)
+		keep_early = 1;
+
+	early_console = &early_hv_console;
+	register_console(early_console);
+
+	return 0;
+}
+
+void __init disable_early_printk(void)
+{
+	early_console_complete = 1;
+	if (!early_console)
+		return;
+	if (!keep_early) {
+		early_printk("disabling early console\n");
+		unregister_console(early_console);
+		early_console = NULL;
+	} else {
+		early_printk("keeping early console\n");
+	}
+}
+
+void warn_early_printk(void)
+{
+	if (early_console_complete || early_console)
+		return;
+	early_printk("\
+Machine shutting down before console output is fully initialized.\n\
+You may wish to reboot and add the option 'earlyprintk' to your\n\
+boot command line to see any diagnostic early console output.\n\
+");
+}
+
+early_param("earlyprintk", setup_early_printk);

arch/tile/kernel/entry.S

@@ -0,0 +1,71 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/linkage.h>
+#include <linux/unistd.h>
+#include <asm/irqflags.h>
+#include <asm/processor.h>
+#include <arch/abi.h>
+#include <arch/spr_def.h>
+
+#ifdef __tilegx__
+#define bnzt bnezt
+#endif
+
+STD_ENTRY(current_text_addr)
+	{ move r0, lr; jrp lr }
+	STD_ENDPROC(current_text_addr)
+
+STD_ENTRY(dump_stack)
+	{ move r2, lr; lnk r1 }
+	{ move r4, r52; addli r1, r1, dump_stack - . }
+	{ move r3, sp; j _dump_stack }
+	jrp lr   /* keep backtracer happy */
+	STD_ENDPROC(dump_stack)
+
+STD_ENTRY(KBacktraceIterator_init_current)
+	{ move r2, lr; lnk r1 }
+	{ move r4, r52; addli r1, r1, KBacktraceIterator_init_current - . }
+	{ move r3, sp; j _KBacktraceIterator_init_current }
+	jrp lr   /* keep backtracer happy */
+	STD_ENDPROC(KBacktraceIterator_init_current)
+
+/* Loop forever on a nap during SMP boot. */
+STD_ENTRY(smp_nap)
+	nap
+	nop       /* avoid provoking the icache prefetch with a jump */
+	j smp_nap /* we are not architecturally guaranteed not to exit nap */
+	jrp lr    /* clue in the backtracer */
+	STD_ENDPROC(smp_nap)
+
+/*
+ * Enable interrupts racelessly and then nap until interrupted.
+ * Architecturally, we are guaranteed that enabling interrupts via
+ * mtspr to INTERRUPT_CRITICAL_SECTION only interrupts at the next PC.
+ * This function's _cpu_idle_nap address is special; see intvec.S.
+ * When interrupted at _cpu_idle_nap, we bump the PC forward 8, and
+ * as a result return to the function that called _cpu_idle().
+ */
+STD_ENTRY(_cpu_idle)
+	movei r1, 1
+	IRQ_ENABLE_LOAD(r2, r3)
+	mtspr INTERRUPT_CRITICAL_SECTION, r1
+	IRQ_ENABLE_APPLY(r2, r3)       /* unmask, but still with ICS set */
+	mtspr INTERRUPT_CRITICAL_SECTION, zero
+	.global _cpu_idle_nap
+_cpu_idle_nap:
+	nap
+	nop       /* avoid provoking the icache prefetch with a jump */
+	jrp lr
+	STD_ENDPROC(_cpu_idle)

arch/tile/kernel/futex_64.S

@@ -0,0 +1,55 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * Atomically access user memory, but use MMU to avoid propagating
+ * kernel exceptions.
+ */
+
+#include <linux/linkage.h>
+#include <asm/errno.h>
+#include <asm/futex.h>
+#include <asm/page.h>
+#include <asm/processor.h>
+
+/*
+ * Provide a set of atomic memory operations supporting <asm/futex.h>.
+ *
+ * r0: user address to manipulate
+ * r1: new value to write, or for cmpxchg, old value to compare against
+ * r2: (cmpxchg only) new value to write
+ *
+ * Return __get_user struct, r0 with value, r1 with error.
+ */
+#define FUTEX_OP(name, ...) \
+STD_ENTRY(futex_##name)			\
+	__VA_ARGS__;			\
+	{				\
+	 move   r1, zero;		\
+	 jrp    lr			\
+	};				\
+	STD_ENDPROC(futex_##name);	\
+	.pushsection __ex_table,"a";	\
+	.quad 1b, get_user_fault;	\
+	.popsection
+
+	.pushsection .fixup,"ax"
+get_user_fault:
+	{ movei r1, -EFAULT; jrp lr }
+	ENDPROC(get_user_fault)
+	.popsection
+
+FUTEX_OP(cmpxchg, mtspr CMPEXCH_VALUE, r1; 1: cmpexch4 r0, r0, r2)
+FUTEX_OP(set, 1: exch4 r0, r0, r1)
+FUTEX_OP(add, 1: fetchadd4 r0, r0, r1)
+FUTEX_OP(or, 1: fetchor4 r0, r0, r1)
+FUTEX_OP(andn, nor r1, r1, zero; 1: fetchand4 r0, r0, r1)

arch/tile/kernel/head_32.S

@@ -0,0 +1,184 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * TILE startup code.
+ */
+
+#include <linux/linkage.h>
+#include <linux/init.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/thread_info.h>
+#include <asm/processor.h>
+#include <asm/asm-offsets.h>
+#include <hv/hypervisor.h>
+#include <arch/chip.h>
+#include <arch/spr_def.h>
+
+/*
+ * This module contains the entry code for kernel images. It performs the
+ * minimal setup needed to call the generic C routines.
+ */
+
+	__HEAD
+ENTRY(_start)
+	/* Notify the hypervisor of what version of the API we want */
+	{
+	  movei r1, TILE_CHIP
+	  movei r2, TILE_CHIP_REV
+	}
+	{
+	  moveli r0, _HV_VERSION_OLD_HV_INIT
+	  jal hv_init
+	}
+	/* Get a reasonable default ASID in r0 */
+	{
+	  move r0, zero
+	  jal hv_inquire_asid
+	}
+	/* Install the default page table */
+	{
+	  moveli r6, lo16(swapper_pgprot - PAGE_OFFSET)
+	  move r4, r0     /* use starting ASID of range for this page table */
+	}
+	{
+	  moveli r0, lo16(swapper_pg_dir - PAGE_OFFSET)
+	  auli r6, r6, ha16(swapper_pgprot - PAGE_OFFSET)
+	}
+	{
+	  lw r2, r6
+	  addi r6, r6, 4
+	}
+	{
+	  lw r3, r6
+	  auli r0, r0, ha16(swapper_pg_dir - PAGE_OFFSET)
+	}
+	{
+	  inv r6
+	  move r1, zero   /* high 32 bits of CPA is zero */
+	}
+	{
+	  moveli lr, lo16(1f)
+	  moveli r5, CTX_PAGE_FLAG
+	}
+	{
+	  auli lr, lr, ha16(1f)
+	  j hv_install_context
+	}
+1:
+
+	/* Get our processor number and save it away in SAVE_K_0. */
+	jal hv_inquire_topology
+	mulll_uu r4, r1, r2        /* r1 == y, r2 == width */
+	add r4, r4, r0             /* r0 == x, so r4 == cpu == y*width + x */
+
+#ifdef CONFIG_SMP
+	/*
+	 * Load up our per-cpu offset.  When the first (master) tile
+	 * boots, this value is still zero, so we will load boot_pc
+	 * with start_kernel, and boot_sp with init_stack + THREAD_SIZE.
+	 * The master tile initializes the per-cpu offset array, so that
+	 * when subsequent (secondary) tiles boot, they will instead load
+	 * from their per-cpu versions of boot_sp and boot_pc.
+	 */
+	moveli r5, lo16(__per_cpu_offset)
+	auli r5, r5, ha16(__per_cpu_offset)
+	s2a r5, r4, r5
+	lw r5, r5
+	bnz r5, 1f
+
+	/*
+	 * Save the width and height to the smp_topology variable
+	 * for later use.
+	 */
+	moveli r0, lo16(smp_topology + HV_TOPOLOGY_WIDTH_OFFSET)
+	auli r0, r0, ha16(smp_topology + HV_TOPOLOGY_WIDTH_OFFSET)
+	{
+	  sw r0, r2
+	  addi r0, r0, (HV_TOPOLOGY_HEIGHT_OFFSET - HV_TOPOLOGY_WIDTH_OFFSET)
+	}
+	sw r0, r3
+1:
+#else
+	move r5, zero
+#endif
+
+	/* Load and go with the correct pc and sp. */
+	{
+	  addli r1, r5, lo16(boot_sp)
+	  addli r0, r5, lo16(boot_pc)
+	}
+	{
+	  auli r1, r1, ha16(boot_sp)
+	  auli r0, r0, ha16(boot_pc)
+	}
+	lw r0, r0
+	lw sp, r1
+	or r4, sp, r4
+	mtspr SPR_SYSTEM_SAVE_K_0, r4  /* save ksp0 + cpu */
+	addi sp, sp, -STACK_TOP_DELTA
+	{
+	  move lr, zero   /* stop backtraces in the called function */
+	  jr r0
+	}
+	ENDPROC(_start)
+
+__PAGE_ALIGNED_BSS
+	.align PAGE_SIZE
+ENTRY(empty_zero_page)
+	.fill PAGE_SIZE,1,0
+	END(empty_zero_page)
+
+	.macro PTE va, cpa, bits1, no_org=0
+	.ifeq \no_org
+	.org swapper_pg_dir + PGD_INDEX(\va) * HV_PTE_SIZE
+	.endif
+	.word HV_PTE_PAGE | HV_PTE_DIRTY | HV_PTE_PRESENT | HV_PTE_ACCESSED | \
+	      (HV_PTE_MODE_CACHE_NO_L3 << HV_PTE_INDEX_MODE)
+	.word (\bits1) | (HV_CPA_TO_PTFN(\cpa) << (HV_PTE_INDEX_PTFN - 32))
+	.endm
+
+__PAGE_ALIGNED_DATA
+	.align PAGE_SIZE
+ENTRY(swapper_pg_dir)
+	/*
+	 * All data pages from PAGE_OFFSET to MEM_USER_INTRPT are mapped as
+	 * VA = PA + PAGE_OFFSET.  We remap things with more precise access
+	 * permissions and more respect for size of RAM later.
+	 */
+	.set addr, 0
+	.rept (MEM_USER_INTRPT - PAGE_OFFSET) >> PGDIR_SHIFT
+	PTE addr + PAGE_OFFSET, addr, (1 << (HV_PTE_INDEX_READABLE - 32)) | \
+				      (1 << (HV_PTE_INDEX_WRITABLE - 32))
+	.set addr, addr + PGDIR_SIZE
+	.endr
+
+	/* The true text VAs are mapped as VA = PA + MEM_SV_INTRPT */
+	PTE MEM_SV_INTRPT, 0, (1 << (HV_PTE_INDEX_READABLE - 32)) | \
+			      (1 << (HV_PTE_INDEX_EXECUTABLE - 32))
+	.org swapper_pg_dir + PGDIR_SIZE
+	END(swapper_pg_dir)
+
+	/*
+	 * Isolate swapper_pgprot to its own cache line, since each cpu
+	 * starting up will read it using VA-is-PA and local homing.
+	 * This would otherwise likely conflict with other data on the cache
+	 * line, once we have set its permanent home in the page tables.
+	 */
+	__INITDATA
+	.align CHIP_L2_LINE_SIZE()
+ENTRY(swapper_pgprot)
+	PTE	0, 0, (1 << (HV_PTE_INDEX_READABLE - 32)) | \
+		      (1 << (HV_PTE_INDEX_WRITABLE - 32)), 1
+	.align CHIP_L2_LINE_SIZE()
+	END(swapper_pgprot)

arch/tile/kernel/head_64.S

@@ -0,0 +1,275 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * TILE startup code.
+ */
+
+#include <linux/linkage.h>
+#include <linux/init.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/thread_info.h>
+#include <asm/processor.h>
+#include <asm/asm-offsets.h>
+#include <hv/hypervisor.h>
+#include <arch/chip.h>
+#include <arch/spr_def.h>
+
+/*
+ * This module contains the entry code for kernel images. It performs the
+ * minimal setup needed to call the generic C routines.
+ */
+
+	__HEAD
+ENTRY(_start)
+	/* Notify the hypervisor of what version of the API we want */
+	{
+#if KERNEL_PL == 1 && _HV_VERSION == 13
+	  /* Support older hypervisors by asking for API version 12. */
+	  movei r0, _HV_VERSION_OLD_HV_INIT
+#else
+	  movei r0, _HV_VERSION
+#endif
+	  movei r1, TILE_CHIP
+	}
+	{
+	  movei r2, TILE_CHIP_REV
+	  movei r3, KERNEL_PL
+	}
+	jal hv_init
+	/* Get a reasonable default ASID in r0 */
+	{
+	  move r0, zero
+	  jal hv_inquire_asid
+	}
+
+	/*
+	 * Install the default page table.  The relocation required to
+	 * statically define the table is a bit too complex, so we have
+	 * to plug in the pointer from the L0 to the L1 table by hand.
+	 * We only do this on the first cpu to boot, though, since the
+	 * other CPUs should see a properly-constructed page table.
+	 */
+	{
+	  v4int_l r2, zero, r0    /* ASID for hv_install_context */
+	  moveli r4, hw1_last(swapper_pgprot - PAGE_OFFSET)
+	}
+	{
+	  shl16insli r4, r4, hw0(swapper_pgprot - PAGE_OFFSET)
+	}
+	{
+	  ld r1, r4               /* access_pte for hv_install_context */
+	}
+	{
+	  moveli r0, hw1_last(.Lsv_data_pmd - PAGE_OFFSET)
+	  moveli r6, hw1_last(temp_data_pmd - PAGE_OFFSET)
+	}
+	{
+	  /* After initializing swapper_pgprot, HV_PTE_GLOBAL is set. */
+	  bfextu r7, r1, HV_PTE_INDEX_GLOBAL, HV_PTE_INDEX_GLOBAL
+	  inv r4
+	}
+	bnez r7, .Lno_write
+	{
+	  shl16insli r0, r0, hw0(.Lsv_data_pmd - PAGE_OFFSET)
+	  shl16insli r6, r6, hw0(temp_data_pmd - PAGE_OFFSET)
+	}
+	{
+	  /* Cut off the low bits of the PT address. */
+	  shrui r6, r6, HV_LOG2_PAGE_TABLE_ALIGN
+	  /* Start with our access pte. */
+	  move r5, r1
+	}
+	{
+	  /* Stuff the address into the page table pointer slot of the PTE. */
+	  bfins r5, r6, HV_PTE_INDEX_PTFN, \
+			HV_PTE_INDEX_PTFN + HV_PTE_PTFN_BITS - 1
+	}
+	{
+	  /* Store the L0 data PTE. */
+	  st r0, r5
+	  addli r6, r6, (temp_code_pmd - temp_data_pmd) >> \
+			HV_LOG2_PAGE_TABLE_ALIGN
+	}
+	{
+	  addli r0, r0, .Lsv_code_pmd - .Lsv_data_pmd
+	  bfins r5, r6, HV_PTE_INDEX_PTFN, \
+			HV_PTE_INDEX_PTFN + HV_PTE_PTFN_BITS - 1
+	}
+	/* Store the L0 code PTE. */
+	st r0, r5
+
+.Lno_write:
+	moveli lr, hw2_last(1f)
+	{
+	  shl16insli lr, lr, hw1(1f)
+	  moveli r0, hw1_last(swapper_pg_dir - PAGE_OFFSET)
+	}
+	{
+	  shl16insli lr, lr, hw0(1f)
+	  shl16insli r0, r0, hw0(swapper_pg_dir - PAGE_OFFSET)
+	}
+	{
+	  moveli r3, CTX_PAGE_FLAG
+	  j hv_install_context
+	}
+1:
+
+	/* Install the interrupt base. */
+	moveli r0, hw2_last(MEM_SV_START)
+	shl16insli r0, r0, hw1(MEM_SV_START)
+	shl16insli r0, r0, hw0(MEM_SV_START)
+	mtspr SPR_INTERRUPT_VECTOR_BASE_K, r0
+
+	/*
+	 * Get our processor number and save it away in SAVE_K_0.
+	 * Extract stuff from the topology structure: r4 = y, r6 = x,
+	 * r5 = width.  FIXME: consider whether we want to just make these
+	 * 64-bit values (and if so fix smp_topology write below, too).
+	 */
+	jal hv_inquire_topology
+	{
+	  v4int_l r5, zero, r1    /* r5 = width */
+	  shrui r4, r0, 32        /* r4 = y */
+	}
+	{
+	  v4int_l r6, zero, r0    /* r6 = x */
+	  mul_lu_lu r4, r4, r5
+	}
+	{
+	  add r4, r4, r6          /* r4 == cpu == y*width + x */
+	}
+
+#ifdef CONFIG_SMP
+	/*
+	 * Load up our per-cpu offset.  When the first (master) tile
+	 * boots, this value is still zero, so we will load boot_pc
+	 * with start_kernel, and boot_sp with init_stack + THREAD_SIZE.
+	 * The master tile initializes the per-cpu offset array, so that
+	 * when subsequent (secondary) tiles boot, they will instead load
+	 * from their per-cpu versions of boot_sp and boot_pc.
+	 */
+	moveli r5, hw2_last(__per_cpu_offset)
+	shl16insli r5, r5, hw1(__per_cpu_offset)
+	shl16insli r5, r5, hw0(__per_cpu_offset)
+	shl3add r5, r4, r5
+	ld r5, r5
+	bnez r5, 1f
+
+	/*
+	 * Save the width and height to the smp_topology variable
+	 * for later use.
+	 */
+	moveli r0, hw2_last(smp_topology + HV_TOPOLOGY_WIDTH_OFFSET)
+	shl16insli r0, r0, hw1(smp_topology + HV_TOPOLOGY_WIDTH_OFFSET)
+	shl16insli r0, r0, hw0(smp_topology + HV_TOPOLOGY_WIDTH_OFFSET)
+	st r0, r1
+1:
+#else
+	move r5, zero
+#endif
+
+	/* Load and go with the correct pc and sp. */
+	{
+	  moveli r1, hw2_last(boot_sp)
+	  moveli r0, hw2_last(boot_pc)
+	}
+	{
+	  shl16insli r1, r1, hw1(boot_sp)
+	  shl16insli r0, r0, hw1(boot_pc)
+	}
+	{
+	  shl16insli r1, r1, hw0(boot_sp)
+	  shl16insli r0, r0, hw0(boot_pc)
+	}
+	{
+	  add r1, r1, r5
+	  add r0, r0, r5
+	}
+	ld r0, r0
+	ld sp, r1
+	or r4, sp, r4
+	mtspr SPR_SYSTEM_SAVE_K_0, r4  /* save ksp0 + cpu */
+	addi sp, sp, -STACK_TOP_DELTA
+	{
+	  move lr, zero   /* stop backtraces in the called function */
+	  jr r0
+	}
+	ENDPROC(_start)
+
+__PAGE_ALIGNED_BSS
+	.align PAGE_SIZE
+ENTRY(empty_zero_page)
+	.fill PAGE_SIZE,1,0
+	END(empty_zero_page)
+
+	.macro PTE cpa, bits1
+	.quad HV_PTE_PAGE | HV_PTE_DIRTY | HV_PTE_PRESENT | HV_PTE_ACCESSED |\
+	      HV_PTE_GLOBAL | (HV_PTE_MODE_CACHE_NO_L3 << HV_PTE_INDEX_MODE) |\
+	      (\bits1) | (HV_CPA_TO_PTFN(\cpa) << HV_PTE_INDEX_PTFN)
+	.endm
+
+__PAGE_ALIGNED_DATA
+	.align PAGE_SIZE
+ENTRY(swapper_pg_dir)
+	.org swapper_pg_dir + PGD_INDEX(PAGE_OFFSET) * HV_PTE_SIZE
+.Lsv_data_pmd:
+	.quad 0  /* PTE temp_data_pmd - PAGE_OFFSET, 0 */
+	.org swapper_pg_dir + PGD_INDEX(MEM_SV_START) * HV_PTE_SIZE
+.Lsv_code_pmd:
+	.quad 0  /* PTE temp_code_pmd - PAGE_OFFSET, 0 */
+	.org swapper_pg_dir + SIZEOF_PGD
+	END(swapper_pg_dir)
+
+	.align HV_PAGE_TABLE_ALIGN
+ENTRY(temp_data_pmd)
+	/*
+	 * We fill the PAGE_OFFSET pmd with huge pages with
+	 * VA = PA + PAGE_OFFSET.  We remap things with more precise access
+	 * permissions later.
+	 */
+	.set addr, 0
+	.rept PTRS_PER_PMD
+	PTE addr, HV_PTE_READABLE | HV_PTE_WRITABLE
+	.set addr, addr + HPAGE_SIZE
+	.endr
+	.org temp_data_pmd + SIZEOF_PMD
+	END(temp_data_pmd)
+
+	.align HV_PAGE_TABLE_ALIGN
+ENTRY(temp_code_pmd)
+	/*
+	 * We fill the MEM_SV_START pmd with huge pages with
+	 * VA = PA + PAGE_OFFSET.  We remap things with more precise access
+	 * permissions later.
+	 */
+	.set addr, 0
+	.rept PTRS_PER_PMD
+	PTE addr, HV_PTE_READABLE | HV_PTE_EXECUTABLE
+	.set addr, addr + HPAGE_SIZE
+	.endr
+	.org temp_code_pmd + SIZEOF_PMD
+	END(temp_code_pmd)
+
+	/*
+	 * Isolate swapper_pgprot to its own cache line, since each cpu
+	 * starting up will read it using VA-is-PA and local homing.
+	 * This would otherwise likely conflict with other data on the cache
+	 * line, once we have set its permanent home in the page tables.
+	 */
+	__INITDATA
+	.align CHIP_L2_LINE_SIZE()
+ENTRY(swapper_pgprot)
+	.quad HV_PTE_PRESENT | (HV_PTE_MODE_CACHE_NO_L3 << HV_PTE_INDEX_MODE)
+	.align CHIP_L2_LINE_SIZE()
+	END(swapper_pgprot)

arch/tile/kernel/hvglue.lds

@@ -0,0 +1,59 @@
+/* Hypervisor call vector addresses; see <hv/hypervisor.h> */
+hv_init = TEXT_OFFSET + 0x10020;
+hv_install_context = TEXT_OFFSET + 0x10040;
+hv_sysconf = TEXT_OFFSET + 0x10060;
+hv_get_rtc = TEXT_OFFSET + 0x10080;
+hv_set_rtc = TEXT_OFFSET + 0x100a0;
+hv_flush_asid = TEXT_OFFSET + 0x100c0;
+hv_flush_page = TEXT_OFFSET + 0x100e0;
+hv_flush_pages = TEXT_OFFSET + 0x10100;
+hv_restart = TEXT_OFFSET + 0x10120;
+hv_halt = TEXT_OFFSET + 0x10140;
+hv_power_off = TEXT_OFFSET + 0x10160;
+hv_inquire_physical = TEXT_OFFSET + 0x10180;
+hv_inquire_memory_controller = TEXT_OFFSET + 0x101a0;
+hv_inquire_virtual = TEXT_OFFSET + 0x101c0;
+hv_inquire_asid = TEXT_OFFSET + 0x101e0;
+hv_nanosleep = TEXT_OFFSET + 0x10200;
+hv_console_read_if_ready = TEXT_OFFSET + 0x10220;
+hv_console_write = TEXT_OFFSET + 0x10240;
+hv_downcall_dispatch = TEXT_OFFSET + 0x10260;
+hv_inquire_topology = TEXT_OFFSET + 0x10280;
+hv_fs_findfile = TEXT_OFFSET + 0x102a0;
+hv_fs_fstat = TEXT_OFFSET + 0x102c0;
+hv_fs_pread = TEXT_OFFSET + 0x102e0;
+hv_physaddr_read64 = TEXT_OFFSET + 0x10300;
+hv_physaddr_write64 = TEXT_OFFSET + 0x10320;
+hv_get_command_line = TEXT_OFFSET + 0x10340;
+hv_set_caching = TEXT_OFFSET + 0x10360;
+hv_bzero_page = TEXT_OFFSET + 0x10380;
+hv_register_message_state = TEXT_OFFSET + 0x103a0;
+hv_send_message = TEXT_OFFSET + 0x103c0;
+hv_receive_message = TEXT_OFFSET + 0x103e0;
+hv_inquire_context = TEXT_OFFSET + 0x10400;
+hv_start_all_tiles = TEXT_OFFSET + 0x10420;
+hv_dev_open = TEXT_OFFSET + 0x10440;
+hv_dev_close = TEXT_OFFSET + 0x10460;
+hv_dev_pread = TEXT_OFFSET + 0x10480;
+hv_dev_pwrite = TEXT_OFFSET + 0x104a0;
+hv_dev_poll = TEXT_OFFSET + 0x104c0;
+hv_dev_poll_cancel = TEXT_OFFSET + 0x104e0;
+hv_dev_preada = TEXT_OFFSET + 0x10500;
+hv_dev_pwritea = TEXT_OFFSET + 0x10520;
+hv_flush_remote = TEXT_OFFSET + 0x10540;
+hv_console_putc = TEXT_OFFSET + 0x10560;
+hv_inquire_tiles = TEXT_OFFSET + 0x10580;
+hv_confstr = TEXT_OFFSET + 0x105a0;
+hv_reexec = TEXT_OFFSET + 0x105c0;
+hv_set_command_line = TEXT_OFFSET + 0x105e0;
+hv_clear_intr = TEXT_OFFSET + 0x10600;
+hv_enable_intr = TEXT_OFFSET + 0x10620;
+hv_disable_intr = TEXT_OFFSET + 0x10640;
+hv_raise_intr = TEXT_OFFSET + 0x10660;
+hv_trigger_ipi = TEXT_OFFSET + 0x10680;
+hv_store_mapping = TEXT_OFFSET + 0x106a0;
+hv_inquire_realpa = TEXT_OFFSET + 0x106c0;
+hv_flush_all = TEXT_OFFSET + 0x106e0;
+hv_get_ipi_pte = TEXT_OFFSET + 0x10700;
+hv_set_pte_super_shift = TEXT_OFFSET + 0x10720;
+hv_glue_internals = TEXT_OFFSET + 0x10740;

arch/tile/kernel/irq.c

@@ -0,0 +1,295 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/module.h>
+#include <linux/seq_file.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/kernel_stat.h>
+#include <linux/uaccess.h>
+#include <hv/drv_pcie_rc_intf.h>
+#include <arch/spr_def.h>
+#include <asm/traps.h>
+
+/* Bit-flag stored in irq_desc->chip_data to indicate HW-cleared irqs. */
+#define IS_HW_CLEARED 1
+
+/*
+ * The set of interrupts we enable for arch_local_irq_enable().
+ * This is initialized to have just a single interrupt that the kernel
+ * doesn't actually use as a sentinel.  During kernel init,
+ * interrupts are added as the kernel gets prepared to support them.
+ * NOTE: we could probably initialize them all statically up front.
+ */
+DEFINE_PER_CPU(unsigned long long, interrupts_enabled_mask) =
+  INITIAL_INTERRUPTS_ENABLED;
+EXPORT_PER_CPU_SYMBOL(interrupts_enabled_mask);
+
+/* Define per-tile device interrupt statistics state. */
+DEFINE_PER_CPU(irq_cpustat_t, irq_stat) ____cacheline_internodealigned_in_smp;
+EXPORT_PER_CPU_SYMBOL(irq_stat);
+
+/*
+ * Define per-tile irq disable mask; the hardware/HV only has a single
+ * mask that we use to implement both masking and disabling.
+ */
+static DEFINE_PER_CPU(unsigned long, irq_disable_mask)
+	____cacheline_internodealigned_in_smp;
+
+/*
+ * Per-tile IRQ nesting depth.  Used to make sure we enable newly
+ * enabled IRQs before exiting the outermost interrupt.
+ */
+static DEFINE_PER_CPU(int, irq_depth);
+
+/* State for allocating IRQs on Gx. */
+#if CHIP_HAS_IPI()
+static unsigned long available_irqs = ~(1UL << IRQ_RESCHEDULE);
+static DEFINE_SPINLOCK(available_irqs_lock);
+#endif
+
+#if CHIP_HAS_IPI()
+/* Use SPRs to manipulate device interrupts. */
+#define mask_irqs(irq_mask) __insn_mtspr(SPR_IPI_MASK_SET_K, irq_mask)
+#define unmask_irqs(irq_mask) __insn_mtspr(SPR_IPI_MASK_RESET_K, irq_mask)
+#define clear_irqs(irq_mask) __insn_mtspr(SPR_IPI_EVENT_RESET_K, irq_mask)
+#else
+/* Use HV to manipulate device interrupts. */
+#define mask_irqs(irq_mask) hv_disable_intr(irq_mask)
+#define unmask_irqs(irq_mask) hv_enable_intr(irq_mask)
+#define clear_irqs(irq_mask) hv_clear_intr(irq_mask)
+#endif
+
+/*
+ * The interrupt handling path, implemented in terms of HV interrupt
+ * emulation on TILE64 and TILEPro, and IPI hardware on TILE-Gx.
+ */
+void tile_dev_intr(struct pt_regs *regs, int intnum)
+{
+	int depth = __get_cpu_var(irq_depth)++;
+	unsigned long original_irqs;
+	unsigned long remaining_irqs;
+	struct pt_regs *old_regs;
+
+#if CHIP_HAS_IPI()
+	/*
+	 * Pending interrupts are listed in an SPR.  We might be
+	 * nested, so be sure to only handle irqs that weren't already
+	 * masked by a previous interrupt.  Then, mask out the ones
+	 * we're going to handle.
+	 */
+	unsigned long masked = __insn_mfspr(SPR_IPI_MASK_K);
+	original_irqs = __insn_mfspr(SPR_IPI_EVENT_K) & ~masked;
+	__insn_mtspr(SPR_IPI_MASK_SET_K, original_irqs);
+#else
+	/*
+	 * Hypervisor performs the equivalent of the Gx code above and
+	 * then puts the pending interrupt mask into a system save reg
+	 * for us to find.
+	 */
+	original_irqs = __insn_mfspr(SPR_SYSTEM_SAVE_K_3);
+#endif
+	remaining_irqs = original_irqs;
+
+	/* Track time spent here in an interrupt context. */
+	old_regs = set_irq_regs(regs);
+	irq_enter();
+
+#ifdef CONFIG_DEBUG_STACKOVERFLOW
+	/* Debugging check for stack overflow: less than 1/8th stack free? */
+	{
+		long sp = stack_pointer - (long) current_thread_info();
+		if (unlikely(sp < (sizeof(struct thread_info) + STACK_WARN))) {
+			pr_emerg("tile_dev_intr: "
+			       "stack overflow: %ld\n",
+			       sp - sizeof(struct thread_info));
+			dump_stack();
+		}
+	}
+#endif
+	while (remaining_irqs) {
+		unsigned long irq = __ffs(remaining_irqs);
+		remaining_irqs &= ~(1UL << irq);
+
+		/* Count device irqs; Linux IPIs are counted elsewhere. */
+		if (irq != IRQ_RESCHEDULE)
+			__get_cpu_var(irq_stat).irq_dev_intr_count++;
+
+		generic_handle_irq(irq);
+	}
+
+	/*
+	 * If we weren't nested, turn on all enabled interrupts,
+	 * including any that were reenabled during interrupt
+	 * handling.
+	 */
+	if (depth == 0)
+		unmask_irqs(~__get_cpu_var(irq_disable_mask));
+
+	__get_cpu_var(irq_depth)--;
+
+	/*
+	 * Track time spent against the current process again and
+	 * process any softirqs if they are waiting.
+	 */
+	irq_exit();
+	set_irq_regs(old_regs);
+}
+
+
+/*
+ * Remove an irq from the disabled mask.  If we're in an interrupt
+ * context, defer enabling the HW interrupt until we leave.
+ */
+static void tile_irq_chip_enable(struct irq_data *d)
+{
+	get_cpu_var(irq_disable_mask) &= ~(1UL << d->irq);
+	if (__get_cpu_var(irq_depth) == 0)
+		unmask_irqs(1UL << d->irq);
+	put_cpu_var(irq_disable_mask);
+}
+
+/*
+ * Add an irq to the disabled mask.  We disable the HW interrupt
+ * immediately so that there's no possibility of it firing.  If we're
+ * in an interrupt context, the return path is careful to avoid
+ * unmasking a newly disabled interrupt.
+ */
+static void tile_irq_chip_disable(struct irq_data *d)
+{
+	get_cpu_var(irq_disable_mask) |= (1UL << d->irq);
+	mask_irqs(1UL << d->irq);
+	put_cpu_var(irq_disable_mask);
+}
+
+/* Mask an interrupt. */
+static void tile_irq_chip_mask(struct irq_data *d)
+{
+	mask_irqs(1UL << d->irq);
+}
+
+/* Unmask an interrupt. */
+static void tile_irq_chip_unmask(struct irq_data *d)
+{
+	unmask_irqs(1UL << d->irq);
+}
+
+/*
+ * Clear an interrupt before processing it so that any new assertions
+ * will trigger another irq.
+ */
+static void tile_irq_chip_ack(struct irq_data *d)
+{
+	if ((unsigned long)irq_data_get_irq_chip_data(d) != IS_HW_CLEARED)
+		clear_irqs(1UL << d->irq);
+}
+
+/*
+ * For per-cpu interrupts, we need to avoid unmasking any interrupts
+ * that we disabled via disable_percpu_irq().
+ */
+static void tile_irq_chip_eoi(struct irq_data *d)
+{
+	if (!(__get_cpu_var(irq_disable_mask) & (1UL << d->irq)))
+		unmask_irqs(1UL << d->irq);
+}
+
+static struct irq_chip tile_irq_chip = {
+	.name = "tile_irq_chip",
+	.irq_enable = tile_irq_chip_enable,
+	.irq_disable = tile_irq_chip_disable,
+	.irq_ack = tile_irq_chip_ack,
+	.irq_eoi = tile_irq_chip_eoi,
+	.irq_mask = tile_irq_chip_mask,
+	.irq_unmask = tile_irq_chip_unmask,
+};
+
+void __init init_IRQ(void)
+{
+	ipi_init();
+}
+
+void __cpuinit setup_irq_regs(void)
+{
+	/* Enable interrupt delivery. */
+	unmask_irqs(~0UL);
+#if CHIP_HAS_IPI()
+	arch_local_irq_unmask(INT_IPI_K);
+#endif
+}
+
+void tile_irq_activate(unsigned int irq, int tile_irq_type)
+{
+	/*
+	 * We use handle_level_irq() by default because the pending
+	 * interrupt vector (whether modeled by the HV on TILE64 and
+	 * TILEPro or implemented in hardware on TILE-Gx) has
+	 * level-style semantics for each bit.  An interrupt fires
+	 * whenever a bit is high, not just at edges.
+	 */
+	irq_flow_handler_t handle = handle_level_irq;
+	if (tile_irq_type == TILE_IRQ_PERCPU)
+		handle = handle_percpu_irq;
+	irq_set_chip_and_handler(irq, &tile_irq_chip, handle);
+
+	/*
+	 * Flag interrupts that are hardware-cleared so that ack()
+	 * won't clear them.
+	 */
+	if (tile_irq_type == TILE_IRQ_HW_CLEAR)
+		irq_set_chip_data(irq, (void *)IS_HW_CLEARED);
+}
+EXPORT_SYMBOL(tile_irq_activate);
+
+
+void ack_bad_irq(unsigned int irq)
+{
+	pr_err("unexpected IRQ trap at vector %02x\n", irq);
+}
+
+/*
+ * Generic, controller-independent functions:
+ */
+
+#if CHIP_HAS_IPI()
+int create_irq(void)
+{
+	unsigned long flags;
+	int result;
+
+	spin_lock_irqsave(&available_irqs_lock, flags);
+	if (available_irqs == 0)
+		result = -ENOMEM;
+	else {
+		result = __ffs(available_irqs);
+		available_irqs &= ~(1UL << result);
+		dynamic_irq_init(result);
+	}
+	spin_unlock_irqrestore(&available_irqs_lock, flags);
+
+	return result;
+}
+EXPORT_SYMBOL(create_irq);
+
+void destroy_irq(unsigned int irq)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&available_irqs_lock, flags);
+	available_irqs |= (1UL << irq);
+	dynamic_irq_cleanup(irq);
+	spin_unlock_irqrestore(&available_irqs_lock, flags);
+}
+EXPORT_SYMBOL(destroy_irq);
+#endif

arch/tile/kernel/machine_kexec.c

@@ -0,0 +1,304 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * based on machine_kexec.c from other architectures in linux-2.6.18
+ */
+
+#include <linux/mm.h>
+#include <linux/kexec.h>
+#include <linux/delay.h>
+#include <linux/reboot.h>
+#include <linux/errno.h>
+#include <linux/vmalloc.h>
+#include <linux/cpumask.h>
+#include <linux/kernel.h>
+#include <linux/elf.h>
+#include <linux/highmem.h>
+#include <linux/mmu_context.h>
+#include <linux/io.h>
+#include <linux/timex.h>
+#include <asm/pgtable.h>
+#include <asm/pgalloc.h>
+#include <asm/cacheflush.h>
+#include <asm/checksum.h>
+#include <asm/tlbflush.h>
+#include <asm/homecache.h>
+#include <hv/hypervisor.h>
+
+
+/*
+ * This stuff is not in elf.h and is not in any other kernel include.
+ * This stuff is needed below in the little boot notes parser to
+ * extract the command line so we can pass it to the hypervisor.
+ */
+struct Elf32_Bhdr {
+	Elf32_Word b_signature;
+	Elf32_Word b_size;
+	Elf32_Half b_checksum;
+	Elf32_Half b_records;
+};
+#define ELF_BOOT_MAGIC		0x0E1FB007
+#define EBN_COMMAND_LINE	0x00000004
+#define roundupsz(X) (((X) + 3) & ~3)
+
+/* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
+
+
+void machine_shutdown(void)
+{
+	/*
+	 * Normally we would stop all the other processors here, but
+	 * the check in machine_kexec_prepare below ensures we'll only
+	 * get this far if we've been booted with "nosmp" on the
+	 * command line or without CONFIG_SMP so there's nothing to do
+	 * here (for now).
+	 */
+}
+
+void machine_crash_shutdown(struct pt_regs *regs)
+{
+	/*
+	 * Cannot happen.  This type of kexec is disabled on this
+	 * architecture (and enforced in machine_kexec_prepare below).
+	 */
+}
+
+
+int machine_kexec_prepare(struct kimage *image)
+{
+	if (num_online_cpus() > 1) {
+		pr_warning("%s: detected attempt to kexec "
+		       "with num_online_cpus() > 1\n",
+		       __func__);
+		return -ENOSYS;
+	}
+	if (image->type != KEXEC_TYPE_DEFAULT) {
+		pr_warning("%s: detected attempt to kexec "
+		       "with unsupported type: %d\n",
+		       __func__,
+		       image->type);
+		return -ENOSYS;
+	}
+	return 0;
+}
+
+void machine_kexec_cleanup(struct kimage *image)
+{
+	/*
+	 * We did nothing in machine_kexec_prepare,
+	 * so we have nothing to do here.
+	 */
+}
+
+/*
+ * If we can find elf boot notes on this page, return the command
+ * line.  Otherwise, silently return null.  Somewhat kludgy, but no
+ * good way to do this without significantly rearchitecting the
+ * architecture-independent kexec code.
+ */
+
+static unsigned char *kexec_bn2cl(void *pg)
+{
+	struct Elf32_Bhdr *bhdrp;
+	Elf32_Nhdr *nhdrp;
+	unsigned char *desc;
+	unsigned char *command_line;
+	__sum16 csum;
+
+	bhdrp = (struct Elf32_Bhdr *) pg;
+
+	/*
+	 * This routine is invoked for every source page, so make
+	 * sure to quietly ignore every impossible page.
+	 */
+	if (bhdrp->b_signature != ELF_BOOT_MAGIC ||
+	    bhdrp->b_size > PAGE_SIZE)
+		return 0;
+
+	/*
+	 * If we get a checksum mismatch, warn with the checksum
+	 * so we can diagnose better.
+	 */
+	csum = ip_compute_csum(pg, bhdrp->b_size);
+	if (csum != 0) {
+		pr_warning("%s: bad checksum %#x (size %d)\n",
+			   __func__, csum, bhdrp->b_size);
+		return 0;
+	}
+
+	nhdrp = (Elf32_Nhdr *) (bhdrp + 1);
+
+	while (nhdrp->n_type != EBN_COMMAND_LINE) {
+
+		desc = (unsigned char *) (nhdrp + 1);
+		desc += roundupsz(nhdrp->n_descsz);
+
+		nhdrp = (Elf32_Nhdr *) desc;
+
+		/* still in bounds? */
+		if ((unsigned char *) (nhdrp + 1) >
+		    ((unsigned char *) pg) + bhdrp->b_size) {
+
+			pr_info("%s: out of bounds\n", __func__);
+			return 0;
+		}
+	}
+
+	command_line = (unsigned char *) (nhdrp + 1);
+	desc = command_line;
+
+	while (*desc != '\0') {
+		desc++;
+		if (((unsigned long)desc & PAGE_MASK) != (unsigned long)pg) {
+			pr_info("%s: ran off end of page\n",
+			       __func__);
+			return 0;
+		}
+	}
+
+	return command_line;
+}
+
+static void kexec_find_and_set_command_line(struct kimage *image)
+{
+	kimage_entry_t *ptr, entry;
+
+	unsigned char *command_line = 0;
+	unsigned char *r;
+	HV_Errno hverr;
+
+	for (ptr = &image->head;
+	     (entry = *ptr) && !(entry & IND_DONE);
+	     ptr = (entry & IND_INDIRECTION) ?
+		     phys_to_virt((entry & PAGE_MASK)) : ptr + 1) {
+
+		if ((entry & IND_SOURCE)) {
+			void *va =
+				kmap_atomic_pfn(entry >> PAGE_SHIFT);
+			r = kexec_bn2cl(va);
+			if (r) {
+				command_line = r;
+				break;
+			}
+			kunmap_atomic(va);
+		}
+	}
+
+	if (command_line != 0) {
+		pr_info("setting new command line to \"%s\"\n",
+		       command_line);
+
+		hverr = hv_set_command_line(
+			(HV_VirtAddr) command_line, strlen(command_line));
+		kunmap_atomic(command_line);
+	} else {
+		pr_info("%s: no command line found; making empty\n",
+		       __func__);
+		hverr = hv_set_command_line((HV_VirtAddr) command_line, 0);
+	}
+	if (hverr)
+		pr_warning("%s: hv_set_command_line returned error: %d\n",
+			   __func__, hverr);
+}
+
+/*
+ * The kexec code range-checks all its PAs, so to avoid having it run
+ * amok and allocate memory and then sequester it from every other
+ * controller, we force it to come from controller zero.  We also
+ * disable the oom-killer since if we do end up running out of memory,
+ * that almost certainly won't help.
+ */
+struct page *kimage_alloc_pages_arch(gfp_t gfp_mask, unsigned int order)
+{
+	gfp_mask |= __GFP_THISNODE | __GFP_NORETRY;
+	return alloc_pages_node(0, gfp_mask, order);
+}
+
+/*
+ * Address range in which pa=va mapping is set in setup_quasi_va_is_pa().
+ * For tilepro, PAGE_OFFSET is used since this is the largest possbile value
+ * for tilepro, while for tilegx, we limit it to entire middle level page
+ * table which we assume has been allocated and is undoubtedly large enough.
+ */
+#ifndef __tilegx__
+#define	QUASI_VA_IS_PA_ADDR_RANGE PAGE_OFFSET
+#else
+#define	QUASI_VA_IS_PA_ADDR_RANGE PGDIR_SIZE
+#endif
+
+static void setup_quasi_va_is_pa(void)
+{
+	HV_PTE pte;
+	unsigned long i;
+
+	/*
+	 * Flush our TLB to prevent conflicts between the previous contents
+	 * and the new stuff we're about to add.
+	 */
+	local_flush_tlb_all();
+
+	/*
+	 * setup VA is PA, at least up to QUASI_VA_IS_PA_ADDR_RANGE.
+	 * Note here we assume that level-1 page table is defined by
+	 * HPAGE_SIZE.
+	 */
+	pte = hv_pte(_PAGE_KERNEL | _PAGE_HUGE_PAGE);
+	pte = hv_pte_set_mode(pte, HV_PTE_MODE_CACHE_NO_L3);
+	for (i = 0; i < (QUASI_VA_IS_PA_ADDR_RANGE >> HPAGE_SHIFT); i++) {
+		unsigned long vaddr = i << HPAGE_SHIFT;
+		pgd_t *pgd = pgd_offset(current->mm, vaddr);
+		pud_t *pud = pud_offset(pgd, vaddr);
+		pte_t *ptep = (pte_t *) pmd_offset(pud, vaddr);
+		unsigned long pfn = i << (HPAGE_SHIFT - PAGE_SHIFT);
+
+		if (pfn_valid(pfn))
+			__set_pte(ptep, pfn_pte(pfn, pte));
+	}
+}
+
+
+void machine_kexec(struct kimage *image)
+{
+	void *reboot_code_buffer;
+	pte_t *ptep;
+	void (*rnk)(unsigned long, void *, unsigned long)
+		__noreturn;
+
+	/* Mask all interrupts before starting to reboot. */
+	interrupt_mask_set_mask(~0ULL);
+
+	kexec_find_and_set_command_line(image);
+
+	/*
+	 * Adjust the home caching of the control page to be cached on
+	 * this cpu, and copy the assembly helper into the control
+	 * code page, which we map in the vmalloc area.
+	 */
+	homecache_change_page_home(image->control_code_page, 0,
+				   smp_processor_id());
+	reboot_code_buffer = page_address(image->control_code_page);
+	BUG_ON(reboot_code_buffer == NULL);
+	ptep = virt_to_pte(NULL, (unsigned long)reboot_code_buffer);
+	__set_pte(ptep, pte_mkexec(*ptep));
+	memcpy(reboot_code_buffer, relocate_new_kernel,
+	       relocate_new_kernel_size);
+	__flush_icache_range(
+		(unsigned long) reboot_code_buffer,
+		(unsigned long) reboot_code_buffer + relocate_new_kernel_size);
+
+	setup_quasi_va_is_pa();
+
+	/* now call it */
+	rnk = reboot_code_buffer;
+	(*rnk)(image->head, reboot_code_buffer, image->start);
+}

arch/tile/kernel/messaging.c

@@ -0,0 +1,116 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/percpu.h>
+#include <linux/smp.h>
+#include <linux/hardirq.h>
+#include <linux/ptrace.h>
+#include <asm/hv_driver.h>
+#include <asm/irq_regs.h>
+#include <asm/traps.h>
+#include <hv/hypervisor.h>
+#include <arch/interrupts.h>
+
+/* All messages are stored here */
+static DEFINE_PER_CPU(HV_MsgState, msg_state);
+
+void __cpuinit init_messaging(void)
+{
+	/* Allocate storage for messages in kernel space */
+	HV_MsgState *state = &__get_cpu_var(msg_state);
+	int rc = hv_register_message_state(state);
+	if (rc != HV_OK)
+		panic("hv_register_message_state: error %d", rc);
+
+	/* Make sure downcall interrupts will be enabled. */
+	arch_local_irq_unmask(INT_INTCTRL_K);
+}
+
+void hv_message_intr(struct pt_regs *regs, int intnum)
+{
+	/*
+	 * We enter with interrupts disabled and leave them disabled,
+	 * to match expectations of called functions (e.g.
+	 * do_ccupdate_local() in mm/slab.c).  This is also consistent
+	 * with normal call entry for device interrupts.
+	 */
+
+	int message[HV_MAX_MESSAGE_SIZE/sizeof(int)];
+	HV_RcvMsgInfo rmi;
+	int nmsgs = 0;
+
+	/* Track time spent here in an interrupt context */
+	struct pt_regs *old_regs = set_irq_regs(regs);
+	irq_enter();
+
+#ifdef CONFIG_DEBUG_STACKOVERFLOW
+	/* Debugging check for stack overflow: less than 1/8th stack free? */
+	{
+		long sp = stack_pointer - (long) current_thread_info();
+		if (unlikely(sp < (sizeof(struct thread_info) + STACK_WARN))) {
+			pr_emerg("hv_message_intr: "
+			       "stack overflow: %ld\n",
+			       sp - sizeof(struct thread_info));
+			dump_stack();
+		}
+	}
+#endif
+
+	while (1) {
+		rmi = hv_receive_message(__get_cpu_var(msg_state),
+					 (HV_VirtAddr) message,
+					 sizeof(message));
+		if (rmi.msglen == 0)
+			break;
+
+		if (rmi.msglen < 0)
+			panic("hv_receive_message failed: %d", rmi.msglen);
+
+		++nmsgs;
+
+		if (rmi.source == HV_MSG_TILE) {
+			int tag;
+
+			/* we just send tags for now */
+			BUG_ON(rmi.msglen != sizeof(int));
+
+			tag = message[0];
+#ifdef CONFIG_SMP
+			evaluate_message(message[0]);
+#else
+			panic("Received IPI message %d in UP mode", tag);
+#endif
+		} else if (rmi.source == HV_MSG_INTR) {
+			HV_IntrMsg *him = (HV_IntrMsg *)message;
+			struct hv_driver_cb *cb =
+				(struct hv_driver_cb *)him->intarg;
+			cb->callback(cb, him->intdata);
+			__get_cpu_var(irq_stat).irq_hv_msg_count++;
+		}
+	}
+
+	/*
+	 * We shouldn't have gotten a message downcall with no
+	 * messages available.
+	 */
+	if (nmsgs == 0)
+		panic("Message downcall invoked with no messages!");
+
+	/*
+	 * Track time spent against the current process again and
+	 * process any softirqs if they are waiting.
+	 */
+	irq_exit();
+	set_irq_regs(old_regs);
+}

arch/tile/kernel/module.c

@@ -0,0 +1,232 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * Based on i386 version, copyright (C) 2001 Rusty Russell.
+ */
+
+#include <linux/moduleloader.h>
+#include <linux/elf.h>
+#include <linux/vmalloc.h>
+#include <linux/fs.h>
+#include <linux/string.h>
+#include <linux/kernel.h>
+#include <asm/pgtable.h>
+#include <asm/homecache.h>
+#include <arch/opcode.h>
+
+#ifdef MODULE_DEBUG
+#define DEBUGP printk
+#else
+#define DEBUGP(fmt...)
+#endif
+
+/*
+ * Allocate some address space in the range MEM_MODULE_START to
+ * MEM_MODULE_END and populate it with memory.
+ */
+void *module_alloc(unsigned long size)
+{
+	struct page **pages;
+	pgprot_t prot_rwx = __pgprot(_PAGE_KERNEL | _PAGE_KERNEL_EXEC);
+	struct vm_struct *area;
+	int i = 0;
+	int npages;
+
+	npages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
+	pages = kmalloc(npages * sizeof(struct page *), GFP_KERNEL);
+	if (pages == NULL)
+		return NULL;
+	for (; i < npages; ++i) {
+		pages[i] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
+		if (!pages[i])
+			goto error;
+	}
+
+	area = __get_vm_area(size, VM_ALLOC, MEM_MODULE_START, MEM_MODULE_END);
+	if (!area)
+		goto error;
+	area->nr_pages = npages;
+	area->pages = pages;
+
+	if (map_vm_area(area, prot_rwx, &pages)) {
+		vunmap(area->addr);
+		goto error;
+	}
+
+	return area->addr;
+
+error:
+	while (--i >= 0)
+		__free_page(pages[i]);
+	kfree(pages);
+	return NULL;
+}
+
+
+/* Free memory returned from module_alloc */
+void module_free(struct module *mod, void *module_region)
+{
+	vfree(module_region);
+
+	/* Globally flush the L1 icache. */
+	flush_remote(0, HV_FLUSH_EVICT_L1I, cpu_online_mask,
+		     0, 0, 0, NULL, NULL, 0);
+
+	/*
+	 * FIXME: If module_region == mod->module_init, trim exception
+	 * table entries.
+	 */
+}
+
+#ifdef __tilegx__
+/*
+ * Validate that the high 16 bits of "value" is just the sign-extension of
+ * the low 48 bits.
+ */
+static int validate_hw2_last(long value, struct module *me)
+{
+	if (((value << 16) >> 16) != value) {
+		pr_warning("module %s: Out of range HW2_LAST value %#lx\n",
+			   me->name, value);
+		return 0;
+	}
+	return 1;
+}
+
+/*
+ * Validate that "value" isn't too big to hold in a JumpOff relocation.
+ */
+static int validate_jumpoff(long value)
+{
+	/* Determine size of jump offset. */
+	int shift = __builtin_clzl(get_JumpOff_X1(create_JumpOff_X1(-1)));
+
+	/* Check to see if it fits into the relocation slot. */
+	long f = get_JumpOff_X1(create_JumpOff_X1(value));
+	f = (f << shift) >> shift;
+
+	return f == value;
+}
+#endif
+
+int apply_relocate_add(Elf_Shdr *sechdrs,
+		       const char *strtab,
+		       unsigned int symindex,
+		       unsigned int relsec,
+		       struct module *me)
+{
+	unsigned int i;
+	Elf_Rela *rel = (void *)sechdrs[relsec].sh_addr;
+	Elf_Sym *sym;
+	u64 *location;
+	unsigned long value;
+
+	DEBUGP("Applying relocate section %u to %u\n", relsec,
+	       sechdrs[relsec].sh_info);
+	for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
+		/* This is where to make the change */
+		location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
+			+ rel[i].r_offset;
+		/*
+		 * This is the symbol it is referring to.
+		 * Note that all undefined symbols have been resolved.
+		 */
+		sym = (Elf_Sym *)sechdrs[symindex].sh_addr
+			+ ELF_R_SYM(rel[i].r_info);
+		value = sym->st_value + rel[i].r_addend;
+
+		switch (ELF_R_TYPE(rel[i].r_info)) {
+
+#ifdef __LITTLE_ENDIAN
+# define MUNGE(func) \
+	(*location = ((*location & ~func(-1)) | func(value)))
+#else
+/*
+ * Instructions are always little-endian, so when we read them as data,
+ * we have to swap them around before and after modifying them.
+ */
+# define MUNGE(func) \
+	(*location = swab64((swab64(*location) & ~func(-1)) | func(value)))
+#endif
+
+#ifndef __tilegx__
+		case R_TILE_32:
+			*(uint32_t *)location = value;
+			break;
+		case R_TILE_IMM16_X0_HA:
+			value = (value + 0x8000) >> 16;
+			/*FALLTHROUGH*/
+		case R_TILE_IMM16_X0_LO:
+			MUNGE(create_Imm16_X0);
+			break;
+		case R_TILE_IMM16_X1_HA:
+			value = (value + 0x8000) >> 16;
+			/*FALLTHROUGH*/
+		case R_TILE_IMM16_X1_LO:
+			MUNGE(create_Imm16_X1);
+			break;
+		case R_TILE_JOFFLONG_X1:
+			value -= (unsigned long) location;  /* pc-relative */
+			value = (long) value >> 3;     /* count by instrs */
+			MUNGE(create_JOffLong_X1);
+			break;
+#else
+		case R_TILEGX_64:
+			*location = value;
+			break;
+		case R_TILEGX_IMM16_X0_HW2_LAST:
+			if (!validate_hw2_last(value, me))
+				return -ENOEXEC;
+			value >>= 16;
+			/*FALLTHROUGH*/
+		case R_TILEGX_IMM16_X0_HW1:
+			value >>= 16;
+			/*FALLTHROUGH*/
+		case R_TILEGX_IMM16_X0_HW0:
+			MUNGE(create_Imm16_X0);
+			break;
+		case R_TILEGX_IMM16_X1_HW2_LAST:
+			if (!validate_hw2_last(value, me))
+				return -ENOEXEC;
+			value >>= 16;
+			/*FALLTHROUGH*/
+		case R_TILEGX_IMM16_X1_HW1:
+			value >>= 16;
+			/*FALLTHROUGH*/
+		case R_TILEGX_IMM16_X1_HW0:
+			MUNGE(create_Imm16_X1);
+			break;
+		case R_TILEGX_JUMPOFF_X1:
+			value -= (unsigned long) location;  /* pc-relative */
+			value = (long) value >> 3;     /* count by instrs */
+			if (!validate_jumpoff(value)) {
+				pr_warning("module %s: Out of range jump to"
+					   " %#llx at %#llx (%p)\n", me->name,
+					   sym->st_value + rel[i].r_addend,
+					   rel[i].r_offset, location);
+				return -ENOEXEC;
+			}
+			MUNGE(create_JumpOff_X1);
+			break;
+#endif
+
+#undef MUNGE
+
+		default:
+			pr_err("module %s: Unknown relocation: %d\n",
+			       me->name, (int) ELF_R_TYPE(rel[i].r_info));
+			return -ENOEXEC;
+		}
+	}
+	return 0;
+}

arch/tile/kernel/pci-dma.c

@@ -0,0 +1,586 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/mm.h>
+#include <linux/dma-mapping.h>
+#include <linux/swiotlb.h>
+#include <linux/vmalloc.h>
+#include <linux/export.h>
+#include <asm/tlbflush.h>
+#include <asm/homecache.h>
+
+/* Generic DMA mapping functions: */
+
+/*
+ * Allocate what Linux calls "coherent" memory.  On TILEPro this is
+ * uncached memory; on TILE-Gx it is hash-for-home memory.
+ */
+#ifdef __tilepro__
+#define PAGE_HOME_DMA PAGE_HOME_UNCACHED
+#else
+#define PAGE_HOME_DMA PAGE_HOME_HASH
+#endif
+
+static void *tile_dma_alloc_coherent(struct device *dev, size_t size,
+				     dma_addr_t *dma_handle, gfp_t gfp,
+				     struct dma_attrs *attrs)
+{
+	u64 dma_mask = dev->coherent_dma_mask ?: DMA_BIT_MASK(32);
+	int node = dev_to_node(dev);
+	int order = get_order(size);
+	struct page *pg;
+	dma_addr_t addr;
+
+	gfp |= __GFP_ZERO;
+
+	/*
+	 * If the mask specifies that the memory be in the first 4 GB, then
+	 * we force the allocation to come from the DMA zone.  We also
+	 * force the node to 0 since that's the only node where the DMA
+	 * zone isn't empty.  If the mask size is smaller than 32 bits, we
+	 * may still not be able to guarantee a suitable memory address, in
+	 * which case we will return NULL.  But such devices are uncommon.
+	 */
+	if (dma_mask <= DMA_BIT_MASK(32)) {
+		gfp |= GFP_DMA;
+		node = 0;
+	}
+
+	pg = homecache_alloc_pages_node(node, gfp, order, PAGE_HOME_DMA);
+	if (pg == NULL)
+		return NULL;
+
+	addr = page_to_phys(pg);
+	if (addr + size > dma_mask) {
+		__homecache_free_pages(pg, order);
+		return NULL;
+	}
+
+	*dma_handle = addr;
+
+	return page_address(pg);
+}
+
+/*
+ * Free memory that was allocated with tile_dma_alloc_coherent.
+ */
+static void tile_dma_free_coherent(struct device *dev, size_t size,
+				   void *vaddr, dma_addr_t dma_handle,
+				   struct dma_attrs *attrs)
+{
+	homecache_free_pages((unsigned long)vaddr, get_order(size));
+}
+
+/*
+ * The map routines "map" the specified address range for DMA
+ * accesses.  The memory belongs to the device after this call is
+ * issued, until it is unmapped with dma_unmap_single.
+ *
+ * We don't need to do any mapping, we just flush the address range
+ * out of the cache and return a DMA address.
+ *
+ * The unmap routines do whatever is necessary before the processor
+ * accesses the memory again, and must be called before the driver
+ * touches the memory.  We can get away with a cache invalidate if we
+ * can count on nothing having been touched.
+ */
+
+/* Set up a single page for DMA access. */
+static void __dma_prep_page(struct page *page, unsigned long offset,
+			    size_t size, enum dma_data_direction direction)
+{
+	/*
+	 * Flush the page from cache if necessary.
+	 * On tilegx, data is delivered to hash-for-home L3; on tilepro,
+	 * data is delivered direct to memory.
+	 *
+	 * NOTE: If we were just doing DMA_TO_DEVICE we could optimize
+	 * this to be a "flush" not a "finv" and keep some of the
+	 * state in cache across the DMA operation, but it doesn't seem
+	 * worth creating the necessary flush_buffer_xxx() infrastructure.
+	 */
+	int home = page_home(page);
+	switch (home) {
+	case PAGE_HOME_HASH:
+#ifdef __tilegx__
+		return;
+#endif
+		break;
+	case PAGE_HOME_UNCACHED:
+#ifdef __tilepro__
+		return;
+#endif
+		break;
+	case PAGE_HOME_IMMUTABLE:
+		/* Should be going to the device only. */
+		BUG_ON(direction == DMA_FROM_DEVICE ||
+		       direction == DMA_BIDIRECTIONAL);
+		return;
+	case PAGE_HOME_INCOHERENT:
+		/* Incoherent anyway, so no need to work hard here. */
+		return;
+	default:
+		BUG_ON(home < 0 || home >= NR_CPUS);
+		break;
+	}
+	homecache_finv_page(page);
+
+#ifdef DEBUG_ALIGNMENT
+	/* Warn if the region isn't cacheline aligned. */
+	if (offset & (L2_CACHE_BYTES - 1) || (size & (L2_CACHE_BYTES - 1)))
+		pr_warn("Unaligned DMA to non-hfh memory: PA %#llx/%#lx\n",
+			PFN_PHYS(page_to_pfn(page)) + offset, size);
+#endif
+}
+
+/* Make the page ready to be read by the core. */
+static void __dma_complete_page(struct page *page, unsigned long offset,
+				size_t size, enum dma_data_direction direction)
+{
+#ifdef __tilegx__
+	switch (page_home(page)) {
+	case PAGE_HOME_HASH:
+		/* I/O device delivered data the way the cpu wanted it. */
+		break;
+	case PAGE_HOME_INCOHERENT:
+		/* Incoherent anyway, so no need to work hard here. */
+		break;
+	case PAGE_HOME_IMMUTABLE:
+		/* Extra read-only copies are not a problem. */
+		break;
+	default:
+		/* Flush the bogus hash-for-home I/O entries to memory. */
+		homecache_finv_map_page(page, PAGE_HOME_HASH);
+		break;
+	}
+#endif
+}
+
+static void __dma_prep_pa_range(dma_addr_t dma_addr, size_t size,
+				enum dma_data_direction direction)
+{
+	struct page *page = pfn_to_page(PFN_DOWN(dma_addr));
+	unsigned long offset = dma_addr & (PAGE_SIZE - 1);
+	size_t bytes = min(size, (size_t)(PAGE_SIZE - offset));
+
+	while (size != 0) {
+		__dma_prep_page(page, offset, bytes, direction);
+		size -= bytes;
+		++page;
+		offset = 0;
+		bytes = min((size_t)PAGE_SIZE, size);
+	}
+}
+
+static void __dma_complete_pa_range(dma_addr_t dma_addr, size_t size,
+				    enum dma_data_direction direction)
+{
+	struct page *page = pfn_to_page(PFN_DOWN(dma_addr));
+	unsigned long offset = dma_addr & (PAGE_SIZE - 1);
+	size_t bytes = min(size, (size_t)(PAGE_SIZE - offset));
+
+	while (size != 0) {
+		__dma_complete_page(page, offset, bytes, direction);
+		size -= bytes;
+		++page;
+		offset = 0;
+		bytes = min((size_t)PAGE_SIZE, size);
+	}
+}
+
+static int tile_dma_map_sg(struct device *dev, struct scatterlist *sglist,
+			   int nents, enum dma_data_direction direction,
+			   struct dma_attrs *attrs)
+{
+	struct scatterlist *sg;
+	int i;
+
+	BUG_ON(!valid_dma_direction(direction));
+
+	WARN_ON(nents == 0 || sglist->length == 0);
+
+	for_each_sg(sglist, sg, nents, i) {
+		sg->dma_address = sg_phys(sg);
+		__dma_prep_pa_range(sg->dma_address, sg->length, direction);
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+		sg->dma_length = sg->length;
+#endif
+	}
+
+	return nents;
+}
+
+static void tile_dma_unmap_sg(struct device *dev, struct scatterlist *sglist,
+			      int nents, enum dma_data_direction direction,
+			      struct dma_attrs *attrs)
+{
+	struct scatterlist *sg;
+	int i;
+
+	BUG_ON(!valid_dma_direction(direction));
+	for_each_sg(sglist, sg, nents, i) {
+		sg->dma_address = sg_phys(sg);
+		__dma_complete_pa_range(sg->dma_address, sg->length,
+					direction);
+	}
+}
+
+static dma_addr_t tile_dma_map_page(struct device *dev, struct page *page,
+				    unsigned long offset, size_t size,
+				    enum dma_data_direction direction,
+				    struct dma_attrs *attrs)
+{
+	BUG_ON(!valid_dma_direction(direction));
+
+	BUG_ON(offset + size > PAGE_SIZE);
+	__dma_prep_page(page, offset, size, direction);
+
+	return page_to_pa(page) + offset;
+}
+
+static void tile_dma_unmap_page(struct device *dev, dma_addr_t dma_address,
+				size_t size, enum dma_data_direction direction,
+				struct dma_attrs *attrs)
+{
+	BUG_ON(!valid_dma_direction(direction));
+
+	__dma_complete_page(pfn_to_page(PFN_DOWN(dma_address)),
+			    dma_address & PAGE_OFFSET, size, direction);
+}
+
+static void tile_dma_sync_single_for_cpu(struct device *dev,
+					 dma_addr_t dma_handle,
+					 size_t size,
+					 enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+
+	__dma_complete_pa_range(dma_handle, size, direction);
+}
+
+static void tile_dma_sync_single_for_device(struct device *dev,
+					    dma_addr_t dma_handle, size_t size,
+					    enum dma_data_direction direction)
+{
+	__dma_prep_pa_range(dma_handle, size, direction);
+}
+
+static void tile_dma_sync_sg_for_cpu(struct device *dev,
+				     struct scatterlist *sglist, int nelems,
+				     enum dma_data_direction direction)
+{
+	struct scatterlist *sg;
+	int i;
+
+	BUG_ON(!valid_dma_direction(direction));
+	WARN_ON(nelems == 0 || sglist->length == 0);
+
+	for_each_sg(sglist, sg, nelems, i) {
+		dma_sync_single_for_cpu(dev, sg->dma_address,
+					sg_dma_len(sg), direction);
+	}
+}
+
+static void tile_dma_sync_sg_for_device(struct device *dev,
+					struct scatterlist *sglist, int nelems,
+					enum dma_data_direction direction)
+{
+	struct scatterlist *sg;
+	int i;
+
+	BUG_ON(!valid_dma_direction(direction));
+	WARN_ON(nelems == 0 || sglist->length == 0);
+
+	for_each_sg(sglist, sg, nelems, i) {
+		dma_sync_single_for_device(dev, sg->dma_address,
+					   sg_dma_len(sg), direction);
+	}
+}
+
+static inline int
+tile_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
+{
+	return 0;
+}
+
+static inline int
+tile_dma_supported(struct device *dev, u64 mask)
+{
+	return 1;
+}
+
+static struct dma_map_ops tile_default_dma_map_ops = {
+	.alloc = tile_dma_alloc_coherent,
+	.free = tile_dma_free_coherent,
+	.map_page = tile_dma_map_page,
+	.unmap_page = tile_dma_unmap_page,
+	.map_sg = tile_dma_map_sg,
+	.unmap_sg = tile_dma_unmap_sg,
+	.sync_single_for_cpu = tile_dma_sync_single_for_cpu,
+	.sync_single_for_device = tile_dma_sync_single_for_device,
+	.sync_sg_for_cpu = tile_dma_sync_sg_for_cpu,
+	.sync_sg_for_device = tile_dma_sync_sg_for_device,
+	.mapping_error = tile_dma_mapping_error,
+	.dma_supported = tile_dma_supported
+};
+
+struct dma_map_ops *tile_dma_map_ops = &tile_default_dma_map_ops;
+EXPORT_SYMBOL(tile_dma_map_ops);
+
+/* Generic PCI DMA mapping functions */
+
+static void *tile_pci_dma_alloc_coherent(struct device *dev, size_t size,
+					 dma_addr_t *dma_handle, gfp_t gfp,
+					 struct dma_attrs *attrs)
+{
+	int node = dev_to_node(dev);
+	int order = get_order(size);
+	struct page *pg;
+	dma_addr_t addr;
+
+	gfp |= __GFP_ZERO;
+
+	pg = homecache_alloc_pages_node(node, gfp, order, PAGE_HOME_DMA);
+	if (pg == NULL)
+		return NULL;
+
+	addr = page_to_phys(pg);
+
+	*dma_handle = phys_to_dma(dev, addr);
+
+	return page_address(pg);
+}
+
+/*
+ * Free memory that was allocated with tile_pci_dma_alloc_coherent.
+ */
+static void tile_pci_dma_free_coherent(struct device *dev, size_t size,
+				       void *vaddr, dma_addr_t dma_handle,
+				       struct dma_attrs *attrs)
+{
+	homecache_free_pages((unsigned long)vaddr, get_order(size));
+}
+
+static int tile_pci_dma_map_sg(struct device *dev, struct scatterlist *sglist,
+			       int nents, enum dma_data_direction direction,
+			       struct dma_attrs *attrs)
+{
+	struct scatterlist *sg;
+	int i;
+
+	BUG_ON(!valid_dma_direction(direction));
+
+	WARN_ON(nents == 0 || sglist->length == 0);
+
+	for_each_sg(sglist, sg, nents, i) {
+		sg->dma_address = sg_phys(sg);
+		__dma_prep_pa_range(sg->dma_address, sg->length, direction);
+
+		sg->dma_address = phys_to_dma(dev, sg->dma_address);
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+		sg->dma_length = sg->length;
+#endif
+	}
+
+	return nents;
+}
+
+static void tile_pci_dma_unmap_sg(struct device *dev,
+				  struct scatterlist *sglist, int nents,
+				  enum dma_data_direction direction,
+				  struct dma_attrs *attrs)
+{
+	struct scatterlist *sg;
+	int i;
+
+	BUG_ON(!valid_dma_direction(direction));
+	for_each_sg(sglist, sg, nents, i) {
+		sg->dma_address = sg_phys(sg);
+		__dma_complete_pa_range(sg->dma_address, sg->length,
+					direction);
+	}
+}
+
+static dma_addr_t tile_pci_dma_map_page(struct device *dev, struct page *page,
+					unsigned long offset, size_t size,
+					enum dma_data_direction direction,
+					struct dma_attrs *attrs)
+{
+	BUG_ON(!valid_dma_direction(direction));
+
+	BUG_ON(offset + size > PAGE_SIZE);
+	__dma_prep_page(page, offset, size, direction);
+
+	return phys_to_dma(dev, page_to_pa(page) + offset);
+}
+
+static void tile_pci_dma_unmap_page(struct device *dev, dma_addr_t dma_address,
+				    size_t size,
+				    enum dma_data_direction direction,
+				    struct dma_attrs *attrs)
+{
+	BUG_ON(!valid_dma_direction(direction));
+
+	dma_address = dma_to_phys(dev, dma_address);
+
+	__dma_complete_page(pfn_to_page(PFN_DOWN(dma_address)),
+			    dma_address & PAGE_OFFSET, size, direction);
+}
+
+static void tile_pci_dma_sync_single_for_cpu(struct device *dev,
+					     dma_addr_t dma_handle,
+					     size_t size,
+					     enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+
+	dma_handle = dma_to_phys(dev, dma_handle);
+
+	__dma_complete_pa_range(dma_handle, size, direction);
+}
+
+static void tile_pci_dma_sync_single_for_device(struct device *dev,
+						dma_addr_t dma_handle,
+						size_t size,
+						enum dma_data_direction
+						direction)
+{
+	dma_handle = dma_to_phys(dev, dma_handle);
+
+	__dma_prep_pa_range(dma_handle, size, direction);
+}
+
+static void tile_pci_dma_sync_sg_for_cpu(struct device *dev,
+					 struct scatterlist *sglist,
+					 int nelems,
+					 enum dma_data_direction direction)
+{
+	struct scatterlist *sg;
+	int i;
+
+	BUG_ON(!valid_dma_direction(direction));
+	WARN_ON(nelems == 0 || sglist->length == 0);
+
+	for_each_sg(sglist, sg, nelems, i) {
+		dma_sync_single_for_cpu(dev, sg->dma_address,
+					sg_dma_len(sg), direction);
+	}
+}
+
+static void tile_pci_dma_sync_sg_for_device(struct device *dev,
+					    struct scatterlist *sglist,
+					    int nelems,
+					    enum dma_data_direction direction)
+{
+	struct scatterlist *sg;
+	int i;
+
+	BUG_ON(!valid_dma_direction(direction));
+	WARN_ON(nelems == 0 || sglist->length == 0);
+
+	for_each_sg(sglist, sg, nelems, i) {
+		dma_sync_single_for_device(dev, sg->dma_address,
+					   sg_dma_len(sg), direction);
+	}
+}
+
+static inline int
+tile_pci_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
+{
+	return 0;
+}
+
+static inline int
+tile_pci_dma_supported(struct device *dev, u64 mask)
+{
+	return 1;
+}
+
+static struct dma_map_ops tile_pci_default_dma_map_ops = {
+	.alloc = tile_pci_dma_alloc_coherent,
+	.free = tile_pci_dma_free_coherent,
+	.map_page = tile_pci_dma_map_page,
+	.unmap_page = tile_pci_dma_unmap_page,
+	.map_sg = tile_pci_dma_map_sg,
+	.unmap_sg = tile_pci_dma_unmap_sg,
+	.sync_single_for_cpu = tile_pci_dma_sync_single_for_cpu,
+	.sync_single_for_device = tile_pci_dma_sync_single_for_device,
+	.sync_sg_for_cpu = tile_pci_dma_sync_sg_for_cpu,
+	.sync_sg_for_device = tile_pci_dma_sync_sg_for_device,
+	.mapping_error = tile_pci_dma_mapping_error,
+	.dma_supported = tile_pci_dma_supported
+};
+
+struct dma_map_ops *gx_pci_dma_map_ops = &tile_pci_default_dma_map_ops;
+EXPORT_SYMBOL(gx_pci_dma_map_ops);
+
+/* PCI DMA mapping functions for legacy PCI devices */
+
+#ifdef CONFIG_SWIOTLB
+static void *tile_swiotlb_alloc_coherent(struct device *dev, size_t size,
+					 dma_addr_t *dma_handle, gfp_t gfp,
+					 struct dma_attrs *attrs)
+{
+	gfp |= GFP_DMA;
+	return swiotlb_alloc_coherent(dev, size, dma_handle, gfp);
+}
+
+static void tile_swiotlb_free_coherent(struct device *dev, size_t size,
+				       void *vaddr, dma_addr_t dma_addr,
+				       struct dma_attrs *attrs)
+{
+	swiotlb_free_coherent(dev, size, vaddr, dma_addr);
+}
+
+static struct dma_map_ops pci_swiotlb_dma_ops = {
+	.alloc = tile_swiotlb_alloc_coherent,
+	.free = tile_swiotlb_free_coherent,
+	.map_page = swiotlb_map_page,
+	.unmap_page = swiotlb_unmap_page,
+	.map_sg = swiotlb_map_sg_attrs,
+	.unmap_sg = swiotlb_unmap_sg_attrs,
+	.sync_single_for_cpu = swiotlb_sync_single_for_cpu,
+	.sync_single_for_device = swiotlb_sync_single_for_device,
+	.sync_sg_for_cpu = swiotlb_sync_sg_for_cpu,
+	.sync_sg_for_device = swiotlb_sync_sg_for_device,
+	.dma_supported = swiotlb_dma_supported,
+	.mapping_error = swiotlb_dma_mapping_error,
+};
+
+struct dma_map_ops *gx_legacy_pci_dma_map_ops = &pci_swiotlb_dma_ops;
+#else
+struct dma_map_ops *gx_legacy_pci_dma_map_ops;
+#endif
+EXPORT_SYMBOL(gx_legacy_pci_dma_map_ops);
+
+#ifdef CONFIG_ARCH_HAS_DMA_SET_COHERENT_MASK
+int dma_set_coherent_mask(struct device *dev, u64 mask)
+{
+	struct dma_map_ops *dma_ops = get_dma_ops(dev);
+
+	/* Handle legacy PCI devices with limited memory addressability. */
+	if (((dma_ops == gx_pci_dma_map_ops) ||
+	    (dma_ops == gx_legacy_pci_dma_map_ops)) &&
+	    (mask <= DMA_BIT_MASK(32))) {
+		if (mask > dev->archdata.max_direct_dma_addr)
+			mask = dev->archdata.max_direct_dma_addr;
+	}
+
+	if (!dma_supported(dev, mask))
+		return -EIO;
+	dev->coherent_dma_mask = mask;
+	return 0;
+}
+EXPORT_SYMBOL(dma_set_coherent_mask);
+#endif

arch/tile/kernel/pci.c

@@ -0,0 +1,596 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/delay.h>
+#include <linux/string.h>
+#include <linux/init.h>
+#include <linux/capability.h>
+#include <linux/sched.h>
+#include <linux/errno.h>
+#include <linux/bootmem.h>
+#include <linux/irq.h>
+#include <linux/io.h>
+#include <linux/uaccess.h>
+#include <linux/export.h>
+
+#include <asm/processor.h>
+#include <asm/sections.h>
+#include <asm/byteorder.h>
+#include <asm/hv_driver.h>
+#include <hv/drv_pcie_rc_intf.h>
+
+
+/*
+ * Initialization flow and process
+ * -------------------------------
+ *
+ * This files contains the routines to search for PCI buses,
+ * enumerate the buses, and configure any attached devices.
+ *
+ * There are two entry points here:
+ * 1) tile_pci_init
+ *    This sets up the pci_controller structs, and opens the
+ *    FDs to the hypervisor.  This is called from setup_arch() early
+ *    in the boot process.
+ * 2) pcibios_init
+ *    This probes the PCI bus(es) for any attached hardware.  It's
+ *    called by subsys_initcall.  All of the real work is done by the
+ *    generic Linux PCI layer.
+ *
+ */
+
+/*
+ * This flag tells if the platform is TILEmpower that needs
+ * special configuration for the PLX switch chip.
+ */
+int __write_once tile_plx_gen1;
+
+static struct pci_controller controllers[TILE_NUM_PCIE];
+static int num_controllers;
+static int pci_scan_flags[TILE_NUM_PCIE];
+
+static struct pci_ops tile_cfg_ops;
+
+
+/*
+ * We don't need to worry about the alignment of resources.
+ */
+resource_size_t pcibios_align_resource(void *data, const struct resource *res,
+			    resource_size_t size, resource_size_t align)
+{
+	return res->start;
+}
+EXPORT_SYMBOL(pcibios_align_resource);
+
+/*
+ * Open a FD to the hypervisor PCI device.
+ *
+ * controller_id is the controller number, config type is 0 or 1 for
+ * config0 or config1 operations.
+ */
+static int tile_pcie_open(int controller_id, int config_type)
+{
+	char filename[32];
+	int fd;
+
+	sprintf(filename, "pcie/%d/config%d", controller_id, config_type);
+
+	fd = hv_dev_open((HV_VirtAddr)filename, 0);
+
+	return fd;
+}
+
+
+/*
+ * Get the IRQ numbers from the HV and set up the handlers for them.
+ */
+static int tile_init_irqs(int controller_id, struct pci_controller *controller)
+{
+	char filename[32];
+	int fd;
+	int ret;
+	int x;
+	struct pcie_rc_config rc_config;
+
+	sprintf(filename, "pcie/%d/ctl", controller_id);
+	fd = hv_dev_open((HV_VirtAddr)filename, 0);
+	if (fd < 0) {
+		pr_err("PCI: hv_dev_open(%s) failed\n", filename);
+		return -1;
+	}
+	ret = hv_dev_pread(fd, 0, (HV_VirtAddr)(&rc_config),
+			   sizeof(rc_config), PCIE_RC_CONFIG_MASK_OFF);
+	hv_dev_close(fd);
+	if (ret != sizeof(rc_config)) {
+		pr_err("PCI: wanted %zd bytes, got %d\n",
+		       sizeof(rc_config), ret);
+		return -1;
+	}
+	/* Record irq_base so that we can map INTx to IRQ # later. */
+	controller->irq_base = rc_config.intr;
+
+	for (x = 0; x < 4; x++)
+		tile_irq_activate(rc_config.intr + x,
+				  TILE_IRQ_HW_CLEAR);
+
+	if (rc_config.plx_gen1)
+		controller->plx_gen1 = 1;
+
+	return 0;
+}
+
+/*
+ * First initialization entry point, called from setup_arch().
+ *
+ * Find valid controllers and fill in pci_controller structs for each
+ * of them.
+ *
+ * Returns the number of controllers discovered.
+ */
+int __init tile_pci_init(void)
+{
+	int i;
+
+	pr_info("PCI: Searching for controllers...\n");
+
+	/* Re-init number of PCIe controllers to support hot-plug feature. */
+	num_controllers = 0;
+
+	/* Do any configuration we need before using the PCIe */
+
+	for (i = 0; i < TILE_NUM_PCIE; i++) {
+		/*
+		 * To see whether we need a real config op based on
+		 * the results of pcibios_init(), to support PCIe hot-plug.
+		 */
+		if (pci_scan_flags[i] == 0) {
+			int hv_cfg_fd0 = -1;
+			int hv_cfg_fd1 = -1;
+			int hv_mem_fd = -1;
+			char name[32];
+			struct pci_controller *controller;
+
+			/*
+			 * Open the fd to the HV.  If it fails then this
+			 * device doesn't exist.
+			 */
+			hv_cfg_fd0 = tile_pcie_open(i, 0);
+			if (hv_cfg_fd0 < 0)
+				continue;
+			hv_cfg_fd1 = tile_pcie_open(i, 1);
+			if (hv_cfg_fd1 < 0) {
+				pr_err("PCI: Couldn't open config fd to HV "
+				    "for controller %d\n", i);
+				goto err_cont;
+			}
+
+			sprintf(name, "pcie/%d/mem", i);
+			hv_mem_fd = hv_dev_open((HV_VirtAddr)name, 0);
+			if (hv_mem_fd < 0) {
+				pr_err("PCI: Could not open mem fd to HV!\n");
+				goto err_cont;
+			}
+
+			pr_info("PCI: Found PCI controller #%d\n", i);
+
+			controller = &controllers[i];
+
+			controller->index = i;
+			controller->hv_cfg_fd[0] = hv_cfg_fd0;
+			controller->hv_cfg_fd[1] = hv_cfg_fd1;
+			controller->hv_mem_fd = hv_mem_fd;
+			controller->first_busno = 0;
+			controller->last_busno = 0xff;
+			controller->ops = &tile_cfg_ops;
+
+			num_controllers++;
+			continue;
+
+err_cont:
+			if (hv_cfg_fd0 >= 0)
+				hv_dev_close(hv_cfg_fd0);
+			if (hv_cfg_fd1 >= 0)
+				hv_dev_close(hv_cfg_fd1);
+			if (hv_mem_fd >= 0)
+				hv_dev_close(hv_mem_fd);
+			continue;
+		}
+	}
+
+	/*
+	 * Before using the PCIe, see if we need to do any platform-specific
+	 * configuration, such as the PLX switch Gen 1 issue on TILEmpower.
+	 */
+	for (i = 0; i < num_controllers; i++) {
+		struct pci_controller *controller = &controllers[i];
+
+		if (controller->plx_gen1)
+			tile_plx_gen1 = 1;
+	}
+
+	return num_controllers;
+}
+
+/*
+ * (pin - 1) converts from the PCI standard's [1:4] convention to
+ * a normal [0:3] range.
+ */
+static int tile_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
+{
+	struct pci_controller *controller =
+		(struct pci_controller *)dev->sysdata;
+	return (pin - 1) + controller->irq_base;
+}
+
+
+static void fixup_read_and_payload_sizes(void)
+{
+	struct pci_dev *dev = NULL;
+	int smallest_max_payload = 0x1; /* Tile maxes out at 256 bytes. */
+	int max_read_size = 0x2; /* Limit to 512 byte reads. */
+	u16 new_values;
+
+	/* Scan for the smallest maximum payload size. */
+	for_each_pci_dev(dev) {
+		u32 devcap;
+		int max_payload;
+
+		if (!pci_is_pcie(dev))
+			continue;
+
+		pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &devcap);
+		max_payload = devcap & PCI_EXP_DEVCAP_PAYLOAD;
+		if (max_payload < smallest_max_payload)
+			smallest_max_payload = max_payload;
+	}
+
+	/* Now, set the max_payload_size for all devices to that value. */
+	new_values = (max_read_size << 12) | (smallest_max_payload << 5);
+	for_each_pci_dev(dev)
+		pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
+				PCI_EXP_DEVCTL_PAYLOAD | PCI_EXP_DEVCTL_READRQ,
+				new_values);
+}
+
+
+/*
+ * Second PCI initialization entry point, called by subsys_initcall.
+ *
+ * The controllers have been set up by the time we get here, by a call to
+ * tile_pci_init.
+ */
+int __init pcibios_init(void)
+{
+	int i;
+
+	pr_info("PCI: Probing PCI hardware\n");
+
+	/*
+	 * Delay a bit in case devices aren't ready.  Some devices are
+	 * known to require at least 20ms here, but we use a more
+	 * conservative value.
+	 */
+	mdelay(250);
+
+	/* Scan all of the recorded PCI controllers.  */
+	for (i = 0; i < TILE_NUM_PCIE; i++) {
+		/*
+		 * Do real pcibios init ops if the controller is initialized
+		 * by tile_pci_init() successfully and not initialized by
+		 * pcibios_init() yet to support PCIe hot-plug.
+		 */
+		if (pci_scan_flags[i] == 0 && controllers[i].ops != NULL) {
+			struct pci_controller *controller = &controllers[i];
+			struct pci_bus *bus;
+			LIST_HEAD(resources);
+
+			if (tile_init_irqs(i, controller)) {
+				pr_err("PCI: Could not initialize IRQs\n");
+				continue;
+			}
+
+			pr_info("PCI: initializing controller #%d\n", i);
+
+			/*
+			 * This comes from the generic Linux PCI driver.
+			 *
+			 * It reads the PCI tree for this bus into the Linux
+			 * data structures.
+			 *
+			 * This is inlined in linux/pci.h and calls into
+			 * pci_scan_bus_parented() in probe.c.
+			 */
+			pci_add_resource(&resources, &ioport_resource);
+			pci_add_resource(&resources, &iomem_resource);
+			bus = pci_scan_root_bus(NULL, 0, controller->ops, controller, &resources);
+			controller->root_bus = bus;
+			controller->last_busno = bus->busn_res.end;
+		}
+	}
+
+	/* Do machine dependent PCI interrupt routing */
+	pci_fixup_irqs(pci_common_swizzle, tile_map_irq);
+
+	/*
+	 * This comes from the generic Linux PCI driver.
+	 *
+	 * It allocates all of the resources (I/O memory, etc)
+	 * associated with the devices read in above.
+	 */
+	pci_assign_unassigned_resources();
+
+	/* Configure the max_read_size and max_payload_size values. */
+	fixup_read_and_payload_sizes();
+
+	/* Record the I/O resources in the PCI controller structure. */
+	for (i = 0; i < TILE_NUM_PCIE; i++) {
+		/*
+		 * Do real pcibios init ops if the controller is initialized
+		 * by tile_pci_init() successfully and not initialized by
+		 * pcibios_init() yet to support PCIe hot-plug.
+		 */
+		if (pci_scan_flags[i] == 0 && controllers[i].ops != NULL) {
+			struct pci_bus *root_bus = controllers[i].root_bus;
+			struct pci_bus *next_bus;
+			struct pci_dev *dev;
+
+			list_for_each_entry(dev, &root_bus->devices, bus_list) {
+				/*
+				 * Find the PCI host controller, ie. the 1st
+				 * bridge.
+				 */
+				if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI &&
+					(PCI_SLOT(dev->devfn) == 0)) {
+					next_bus = dev->subordinate;
+					controllers[i].mem_resources[0] =
+						*next_bus->resource[0];
+					controllers[i].mem_resources[1] =
+						 *next_bus->resource[1];
+					controllers[i].mem_resources[2] =
+						 *next_bus->resource[2];
+
+					/* Setup flags. */
+					pci_scan_flags[i] = 1;
+
+					break;
+				}
+			}
+		}
+	}
+
+	return 0;
+}
+subsys_initcall(pcibios_init);
+
+/*
+ * No bus fixups needed.
+ */
+void pcibios_fixup_bus(struct pci_bus *bus)
+{
+	/* Nothing needs to be done. */
+}
+
+void pcibios_set_master(struct pci_dev *dev)
+{
+	/* No special bus mastering setup handling. */
+}
+
+/*
+ * Enable memory and/or address decoding, as appropriate, for the
+ * device described by the 'dev' struct.
+ *
+ * This is called from the generic PCI layer, and can be called
+ * for bridges or endpoints.
+ */
+int pcibios_enable_device(struct pci_dev *dev, int mask)
+{
+	u16 cmd, old_cmd;
+	u8 header_type;
+	int i;
+	struct resource *r;
+
+	pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
+
+	pci_read_config_word(dev, PCI_COMMAND, &cmd);
+	old_cmd = cmd;
+	if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) {
+		/*
+		 * For bridges, we enable both memory and I/O decoding
+		 * in call cases.
+		 */
+		cmd |= PCI_COMMAND_IO;
+		cmd |= PCI_COMMAND_MEMORY;
+	} else {
+		/*
+		 * For endpoints, we enable memory and/or I/O decoding
+		 * only if they have a memory resource of that type.
+		 */
+		for (i = 0; i < 6; i++) {
+			r = &dev->resource[i];
+			if (r->flags & IORESOURCE_UNSET) {
+				pr_err("PCI: Device %s not available "
+				       "because of resource collisions\n",
+				       pci_name(dev));
+				return -EINVAL;
+			}
+			if (r->flags & IORESOURCE_IO)
+				cmd |= PCI_COMMAND_IO;
+			if (r->flags & IORESOURCE_MEM)
+				cmd |= PCI_COMMAND_MEMORY;
+		}
+	}
+
+	/*
+	 * We only write the command if it changed.
+	 */
+	if (cmd != old_cmd)
+		pci_write_config_word(dev, PCI_COMMAND, cmd);
+	return 0;
+}
+
+/****************************************************************
+ *
+ * Tile PCI config space read/write routines
+ *
+ ****************************************************************/
+
+/*
+ * These are the normal read and write ops
+ * These are expanded with macros from  pci_bus_read_config_byte() etc.
+ *
+ * devfn is the combined PCI slot & function.
+ *
+ * offset is in bytes, from the start of config space for the
+ * specified bus & slot.
+ */
+
+static int tile_cfg_read(struct pci_bus *bus, unsigned int devfn, int offset,
+			 int size, u32 *val)
+{
+	struct pci_controller *controller = bus->sysdata;
+	int busnum = bus->number & 0xff;
+	int slot = (devfn >> 3) & 0x1f;
+	int function = devfn & 0x7;
+	u32 addr;
+	int config_mode = 1;
+
+	/*
+	 * There is no bridge between the Tile and bus 0, so we
+	 * use config0 to talk to bus 0.
+	 *
+	 * If we're talking to a bus other than zero then we
+	 * must have found a bridge.
+	 */
+	if (busnum == 0) {
+		/*
+		 * We fake an empty slot for (busnum == 0) && (slot > 0),
+		 * since there is only one slot on bus 0.
+		 */
+		if (slot) {
+			*val = 0xFFFFFFFF;
+			return 0;
+		}
+		config_mode = 0;
+	}
+
+	addr = busnum << 20;		/* Bus in 27:20 */
+	addr |= slot << 15;		/* Slot (device) in 19:15 */
+	addr |= function << 12;		/* Function is in 14:12 */
+	addr |= (offset & 0xFFF);	/* byte address in 0:11 */
+
+	return hv_dev_pread(controller->hv_cfg_fd[config_mode], 0,
+			    (HV_VirtAddr)(val), size, addr);
+}
+
+
+/*
+ * See tile_cfg_read() for relevant comments.
+ * Note that "val" is the value to write, not a pointer to that value.
+ */
+static int tile_cfg_write(struct pci_bus *bus, unsigned int devfn, int offset,
+			  int size, u32 val)
+{
+	struct pci_controller *controller = bus->sysdata;
+	int busnum = bus->number & 0xff;
+	int slot = (devfn >> 3) & 0x1f;
+	int function = devfn & 0x7;
+	u32 addr;
+	int config_mode = 1;
+	HV_VirtAddr valp = (HV_VirtAddr)&val;
+
+	/*
+	 * For bus 0 slot 0 we use config 0 accesses.
+	 */
+	if (busnum == 0) {
+		/*
+		 * We fake an empty slot for (busnum == 0) && (slot > 0),
+		 * since there is only one slot on bus 0.
+		 */
+		if (slot)
+			return 0;
+		config_mode = 0;
+	}
+
+	addr = busnum << 20;		/* Bus in 27:20 */
+	addr |= slot << 15;		/* Slot (device) in 19:15 */
+	addr |= function << 12;		/* Function is in 14:12 */
+	addr |= (offset & 0xFFF);	/* byte address in 0:11 */
+
+#ifdef __BIG_ENDIAN
+	/* Point to the correct part of the 32-bit "val". */
+	valp += 4 - size;
+#endif
+
+	return hv_dev_pwrite(controller->hv_cfg_fd[config_mode], 0,
+			     valp, size, addr);
+}
+
+
+static struct pci_ops tile_cfg_ops = {
+	.read =         tile_cfg_read,
+	.write =        tile_cfg_write,
+};
+
+
+/*
+ * In the following, each PCI controller's mem_resources[1]
+ * represents its (non-prefetchable) PCI memory resource.
+ * mem_resources[0] and mem_resources[2] refer to its PCI I/O and
+ * prefetchable PCI memory resources, respectively.
+ * For more details, see pci_setup_bridge() in setup-bus.c.
+ * By comparing the target PCI memory address against the
+ * end address of controller 0, we can determine the controller
+ * that should accept the PCI memory access.
+ */
+#define TILE_READ(size, type)						\
+type _tile_read##size(unsigned long addr)				\
+{									\
+	type val;							\
+	int idx = 0;							\
+	if (addr > controllers[0].mem_resources[1].end &&		\
+	    addr > controllers[0].mem_resources[2].end)			\
+		idx = 1;                                                \
+	if (hv_dev_pread(controllers[idx].hv_mem_fd, 0,			\
+			 (HV_VirtAddr)(&val), sizeof(type), addr))	\
+		pr_err("PCI: read %zd bytes at 0x%lX failed\n",		\
+		       sizeof(type), addr);				\
+	return val;							\
+}									\
+EXPORT_SYMBOL(_tile_read##size)
+
+TILE_READ(b, u8);
+TILE_READ(w, u16);
+TILE_READ(l, u32);
+TILE_READ(q, u64);
+
+#define TILE_WRITE(size, type)						\
+void _tile_write##size(type val, unsigned long addr)			\
+{									\
+	int idx = 0;							\
+	if (addr > controllers[0].mem_resources[1].end &&		\
+	    addr > controllers[0].mem_resources[2].end)			\
+		idx = 1;                                                \
+	if (hv_dev_pwrite(controllers[idx].hv_mem_fd, 0,		\
+			  (HV_VirtAddr)(&val), sizeof(type), addr))	\
+		pr_err("PCI: write %zd bytes at 0x%lX failed\n",	\
+		       sizeof(type), addr);				\
+}									\
+EXPORT_SYMBOL(_tile_write##size)
+
+TILE_WRITE(b, u8);
+TILE_WRITE(w, u16);
+TILE_WRITE(l, u32);
+TILE_WRITE(q, u64);

arch/tile/kernel/proc.c

@@ -0,0 +1,163 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/smp.h>
+#include <linux/seq_file.h>
+#include <linux/threads.h>
+#include <linux/cpumask.h>
+#include <linux/timex.h>
+#include <linux/delay.h>
+#include <linux/fs.h>
+#include <linux/proc_fs.h>
+#include <linux/sysctl.h>
+#include <linux/hardirq.h>
+#include <linux/hugetlb.h>
+#include <linux/mman.h>
+#include <asm/unaligned.h>
+#include <asm/pgtable.h>
+#include <asm/processor.h>
+#include <asm/sections.h>
+#include <asm/homecache.h>
+#include <asm/hardwall.h>
+#include <arch/chip.h>
+
+
+/*
+ * Support /proc/cpuinfo
+ */
+
+#define cpu_to_ptr(n) ((void *)((long)(n)+1))
+#define ptr_to_cpu(p) ((long)(p) - 1)
+
+static int show_cpuinfo(struct seq_file *m, void *v)
+{
+	int n = ptr_to_cpu(v);
+
+	if (n == 0) {
+		char buf[NR_CPUS*5];
+		cpulist_scnprintf(buf, sizeof(buf), cpu_online_mask);
+		seq_printf(m, "cpu count\t: %d\n", num_online_cpus());
+		seq_printf(m, "cpu list\t: %s\n", buf);
+		seq_printf(m, "model name\t: %s\n", chip_model);
+		seq_printf(m, "flags\t\t:\n");  /* nothing for now */
+		seq_printf(m, "cpu MHz\t\t: %llu.%06llu\n",
+			   get_clock_rate() / 1000000,
+			   (get_clock_rate() % 1000000));
+		seq_printf(m, "bogomips\t: %lu.%02lu\n\n",
+			   loops_per_jiffy/(500000/HZ),
+			   (loops_per_jiffy/(5000/HZ)) % 100);
+	}
+
+#ifdef CONFIG_SMP
+	if (!cpu_online(n))
+		return 0;
+#endif
+
+	seq_printf(m, "processor\t: %d\n", n);
+
+	/* Print only num_online_cpus() blank lines total. */
+	if (cpumask_next(n, cpu_online_mask) < nr_cpu_ids)
+		seq_printf(m, "\n");
+
+	return 0;
+}
+
+static void *c_start(struct seq_file *m, loff_t *pos)
+{
+	return *pos < nr_cpu_ids ? cpu_to_ptr(*pos) : NULL;
+}
+static void *c_next(struct seq_file *m, void *v, loff_t *pos)
+{
+	++*pos;
+	return c_start(m, pos);
+}
+static void c_stop(struct seq_file *m, void *v)
+{
+}
+const struct seq_operations cpuinfo_op = {
+	.start	= c_start,
+	.next	= c_next,
+	.stop	= c_stop,
+	.show	= show_cpuinfo,
+};
+
+/*
+ * Support /proc/tile directory
+ */
+
+static int __init proc_tile_init(void)
+{
+	struct proc_dir_entry *root = proc_mkdir("tile", NULL);
+	if (root == NULL)
+		return 0;
+
+	proc_tile_hardwall_init(root);
+
+	return 0;
+}
+
+arch_initcall(proc_tile_init);
+
+/*
+ * Support /proc/sys/tile directory
+ */
+
+#ifndef __tilegx__  /* FIXME: GX: no support for unaligned access yet */
+static ctl_table unaligned_subtable[] = {
+	{
+		.procname	= "enabled",
+		.data		= &unaligned_fixup,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec
+	},
+	{
+		.procname	= "printk",
+		.data		= &unaligned_printk,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec
+	},
+	{
+		.procname	= "count",
+		.data		= &unaligned_fixup_count,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec
+	},
+	{}
+};
+
+static ctl_table unaligned_table[] = {
+	{
+		.procname	= "unaligned_fixup",
+		.mode		= 0555,
+		.child		= unaligned_subtable
+	},
+	{}
+};
+
+static struct ctl_path tile_path[] = {
+	{ .procname = "tile" },
+	{ }
+};
+
+static int __init proc_sys_tile_init(void)
+{
+	register_sysctl_paths(tile_path, unaligned_table);
+	return 0;
+}
+
+arch_initcall(proc_sys_tile_init);
+#endif

arch/tile/kernel/process.c

@@ -0,0 +1,598 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/sched.h>
+#include <linux/preempt.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/kprobes.h>
+#include <linux/elfcore.h>
+#include <linux/tick.h>
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <linux/compat.h>
+#include <linux/hardirq.h>
+#include <linux/syscalls.h>
+#include <linux/kernel.h>
+#include <linux/tracehook.h>
+#include <linux/signal.h>
+#include <asm/stack.h>
+#include <asm/switch_to.h>
+#include <asm/homecache.h>
+#include <asm/syscalls.h>
+#include <asm/traps.h>
+#include <asm/setup.h>
+#ifdef CONFIG_HARDWALL
+#include <asm/hardwall.h>
+#endif
+#include <arch/chip.h>
+#include <arch/abi.h>
+#include <arch/sim_def.h>
+
+/*
+ * Use the (x86) "idle=poll" option to prefer low latency when leaving the
+ * idle loop over low power while in the idle loop, e.g. if we have
+ * one thread per core and we want to get threads out of futex waits fast.
+ */
+static int __init idle_setup(char *str)
+{
+	if (!str)
+		return -EINVAL;
+
+	if (!strcmp(str, "poll")) {
+		pr_info("using polling idle threads.\n");
+		cpu_idle_poll_ctrl(true);
+		return 0;
+	} else if (!strcmp(str, "halt")) {
+		return 0;
+	}
+	return -1;
+}
+early_param("idle", idle_setup);
+
+void arch_cpu_idle(void)
+{
+	__get_cpu_var(irq_stat).idle_timestamp = jiffies;
+	_cpu_idle();
+}
+
+/*
+ * Release a thread_info structure
+ */
+void arch_release_thread_info(struct thread_info *info)
+{
+	struct single_step_state *step_state = info->step_state;
+
+#ifdef CONFIG_HARDWALL
+	/*
+	 * We free a thread_info from the context of the task that has
+	 * been scheduled next, so the original task is already dead.
+	 * Calling deactivate here just frees up the data structures.
+	 * If the task we're freeing held the last reference to a
+	 * hardwall fd, it would have been released prior to this point
+	 * anyway via exit_files(), and the hardwall_task.info pointers
+	 * would be NULL by now.
+	 */
+	hardwall_deactivate_all(info->task);
+#endif
+
+	if (step_state) {
+
+		/*
+		 * FIXME: we don't munmap step_state->buffer
+		 * because the mm_struct for this process (info->task->mm)
+		 * has already been zeroed in exit_mm().  Keeping a
+		 * reference to it here seems like a bad move, so this
+		 * means we can't munmap() the buffer, and therefore if we
+		 * ptrace multiple threads in a process, we will slowly
+		 * leak user memory.  (Note that as soon as the last
+		 * thread in a process dies, we will reclaim all user
+		 * memory including single-step buffers in the usual way.)
+		 * We should either assign a kernel VA to this buffer
+		 * somehow, or we should associate the buffer(s) with the
+		 * mm itself so we can clean them up that way.
+		 */
+		kfree(step_state);
+	}
+}
+
+static void save_arch_state(struct thread_struct *t);
+
+int copy_thread(unsigned long clone_flags, unsigned long sp,
+		unsigned long arg, struct task_struct *p)
+{
+	struct pt_regs *childregs = task_pt_regs(p);
+	unsigned long ksp;
+	unsigned long *callee_regs;
+
+	/*
+	 * Set up the stack and stack pointer appropriately for the
+	 * new child to find itself woken up in __switch_to().
+	 * The callee-saved registers must be on the stack to be read;
+	 * the new task will then jump to assembly support to handle
+	 * calling schedule_tail(), etc., and (for userspace tasks)
+	 * returning to the context set up in the pt_regs.
+	 */
+	ksp = (unsigned long) childregs;
+	ksp -= C_ABI_SAVE_AREA_SIZE;   /* interrupt-entry save area */
+	((long *)ksp)[0] = ((long *)ksp)[1] = 0;
+	ksp -= CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long);
+	callee_regs = (unsigned long *)ksp;
+	ksp -= C_ABI_SAVE_AREA_SIZE;   /* __switch_to() save area */
+	((long *)ksp)[0] = ((long *)ksp)[1] = 0;
+	p->thread.ksp = ksp;
+
+	/* Record the pid of the task that created this one. */
+	p->thread.creator_pid = current->pid;
+
+	if (unlikely(p->flags & PF_KTHREAD)) {
+		/* kernel thread */
+		memset(childregs, 0, sizeof(struct pt_regs));
+		memset(&callee_regs[2], 0,
+		       (CALLEE_SAVED_REGS_COUNT - 2) * sizeof(unsigned long));
+		callee_regs[0] = sp;   /* r30 = function */
+		callee_regs[1] = arg;  /* r31 = arg */
+		childregs->ex1 = PL_ICS_EX1(KERNEL_PL, 0);
+		p->thread.pc = (unsigned long) ret_from_kernel_thread;
+		return 0;
+	}
+
+	/*
+	 * Start new thread in ret_from_fork so it schedules properly
+	 * and then return from interrupt like the parent.
+	 */
+	p->thread.pc = (unsigned long) ret_from_fork;
+
+	/*
+	 * Do not clone step state from the parent; each thread
+	 * must make its own lazily.
+	 */
+	task_thread_info(p)->step_state = NULL;
+
+	/*
+	 * Copy the registers onto the kernel stack so the
+	 * return-from-interrupt code will reload it into registers.
+	 */
+	*childregs = *current_pt_regs();
+	childregs->regs[0] = 0;         /* return value is zero */
+	if (sp)
+		childregs->sp = sp;  /* override with new user stack pointer */
+	memcpy(callee_regs, &childregs->regs[CALLEE_SAVED_FIRST_REG],
+	       CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long));
+
+	/* Save user stack top pointer so we can ID the stack vm area later. */
+	p->thread.usp0 = childregs->sp;
+
+	/*
+	 * If CLONE_SETTLS is set, set "tp" in the new task to "r4",
+	 * which is passed in as arg #5 to sys_clone().
+	 */
+	if (clone_flags & CLONE_SETTLS)
+		childregs->tp = childregs->regs[4];
+
+
+#if CHIP_HAS_TILE_DMA()
+	/*
+	 * No DMA in the new thread.  We model this on the fact that
+	 * fork() clears the pending signals, alarms, and aio for the child.
+	 */
+	memset(&p->thread.tile_dma_state, 0, sizeof(struct tile_dma_state));
+	memset(&p->thread.dma_async_tlb, 0, sizeof(struct async_tlb));
+#endif
+
+#if CHIP_HAS_SN_PROC()
+	/* Likewise, the new thread is not running static processor code. */
+	p->thread.sn_proc_running = 0;
+	memset(&p->thread.sn_async_tlb, 0, sizeof(struct async_tlb));
+#endif
+
+#if CHIP_HAS_PROC_STATUS_SPR()
+	/* New thread has its miscellaneous processor state bits clear. */
+	p->thread.proc_status = 0;
+#endif
+
+#ifdef CONFIG_HARDWALL
+	/* New thread does not own any networks. */
+	memset(&p->thread.hardwall[0], 0,
+	       sizeof(struct hardwall_task) * HARDWALL_TYPES);
+#endif
+
+
+	/*
+	 * Start the new thread with the current architecture state
+	 * (user interrupt masks, etc.).
+	 */
+	save_arch_state(&p->thread);
+
+	return 0;
+}
+
+/*
+ * Return "current" if it looks plausible, or else a pointer to a dummy.
+ * This can be helpful if we are just trying to emit a clean panic.
+ */
+struct task_struct *validate_current(void)
+{
+	static struct task_struct corrupt = { .comm = "<corrupt>" };
+	struct task_struct *tsk = current;
+	if (unlikely((unsigned long)tsk < PAGE_OFFSET ||
+		     (high_memory && (void *)tsk > high_memory) ||
+		     ((unsigned long)tsk & (__alignof__(*tsk) - 1)) != 0)) {
+		pr_err("Corrupt 'current' %p (sp %#lx)\n", tsk, stack_pointer);
+		tsk = &corrupt;
+	}
+	return tsk;
+}
+
+/* Take and return the pointer to the previous task, for schedule_tail(). */
+struct task_struct *sim_notify_fork(struct task_struct *prev)
+{
+	struct task_struct *tsk = current;
+	__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK_PARENT |
+		     (tsk->thread.creator_pid << _SIM_CONTROL_OPERATOR_BITS));
+	__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK |
+		     (tsk->pid << _SIM_CONTROL_OPERATOR_BITS));
+	return prev;
+}
+
+int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
+{
+	struct pt_regs *ptregs = task_pt_regs(tsk);
+	elf_core_copy_regs(regs, ptregs);
+	return 1;
+}
+
+#if CHIP_HAS_TILE_DMA()
+
+/* Allow user processes to access the DMA SPRs */
+void grant_dma_mpls(void)
+{
+#if CONFIG_KERNEL_PL == 2
+	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
+	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
+#else
+	__insn_mtspr(SPR_MPL_DMA_CPL_SET_0, 1);
+	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0, 1);
+#endif
+}
+
+/* Forbid user processes from accessing the DMA SPRs */
+void restrict_dma_mpls(void)
+{
+#if CONFIG_KERNEL_PL == 2
+	__insn_mtspr(SPR_MPL_DMA_CPL_SET_2, 1);
+	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_2, 1);
+#else
+	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
+	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
+#endif
+}
+
+/* Pause the DMA engine, then save off its state registers. */
+static void save_tile_dma_state(struct tile_dma_state *dma)
+{
+	unsigned long state = __insn_mfspr(SPR_DMA_USER_STATUS);
+	unsigned long post_suspend_state;
+
+	/* If we're running, suspend the engine. */
+	if ((state & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK)
+		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__SUSPEND_MASK);
+
+	/*
+	 * Wait for the engine to idle, then save regs.  Note that we
+	 * want to record the "running" bit from before suspension,
+	 * and the "done" bit from after, so that we can properly
+	 * distinguish a case where the user suspended the engine from
+	 * the case where the kernel suspended as part of the context
+	 * swap.
+	 */
+	do {
+		post_suspend_state = __insn_mfspr(SPR_DMA_USER_STATUS);
+	} while (post_suspend_state & SPR_DMA_STATUS__BUSY_MASK);
+
+	dma->src = __insn_mfspr(SPR_DMA_SRC_ADDR);
+	dma->src_chunk = __insn_mfspr(SPR_DMA_SRC_CHUNK_ADDR);
+	dma->dest = __insn_mfspr(SPR_DMA_DST_ADDR);
+	dma->dest_chunk = __insn_mfspr(SPR_DMA_DST_CHUNK_ADDR);
+	dma->strides = __insn_mfspr(SPR_DMA_STRIDE);
+	dma->chunk_size = __insn_mfspr(SPR_DMA_CHUNK_SIZE);
+	dma->byte = __insn_mfspr(SPR_DMA_BYTE);
+	dma->status = (state & SPR_DMA_STATUS__RUNNING_MASK) |
+		(post_suspend_state & SPR_DMA_STATUS__DONE_MASK);
+}
+
+/* Restart a DMA that was running before we were context-switched out. */
+static void restore_tile_dma_state(struct thread_struct *t)
+{
+	const struct tile_dma_state *dma = &t->tile_dma_state;
+
+	/*
+	 * The only way to restore the done bit is to run a zero
+	 * length transaction.
+	 */
+	if ((dma->status & SPR_DMA_STATUS__DONE_MASK) &&
+	    !(__insn_mfspr(SPR_DMA_USER_STATUS) & SPR_DMA_STATUS__DONE_MASK)) {
+		__insn_mtspr(SPR_DMA_BYTE, 0);
+		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK);
+		while (__insn_mfspr(SPR_DMA_USER_STATUS) &
+		       SPR_DMA_STATUS__BUSY_MASK)
+			;
+	}
+
+	__insn_mtspr(SPR_DMA_SRC_ADDR, dma->src);
+	__insn_mtspr(SPR_DMA_SRC_CHUNK_ADDR, dma->src_chunk);
+	__insn_mtspr(SPR_DMA_DST_ADDR, dma->dest);
+	__insn_mtspr(SPR_DMA_DST_CHUNK_ADDR, dma->dest_chunk);
+	__insn_mtspr(SPR_DMA_STRIDE, dma->strides);
+	__insn_mtspr(SPR_DMA_CHUNK_SIZE, dma->chunk_size);
+	__insn_mtspr(SPR_DMA_BYTE, dma->byte);
+
+	/*
+	 * Restart the engine if we were running and not done.
+	 * Clear a pending async DMA fault that we were waiting on return
+	 * to user space to execute, since we expect the DMA engine
+	 * to regenerate those faults for us now.  Note that we don't
+	 * try to clear the TIF_ASYNC_TLB flag, since it's relatively
+	 * harmless if set, and it covers both DMA and the SN processor.
+	 */
+	if ((dma->status & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK) {
+		t->dma_async_tlb.fault_num = 0;
+		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK);
+	}
+}
+
+#endif
+
+static void save_arch_state(struct thread_struct *t)
+{
+#if CHIP_HAS_SPLIT_INTR_MASK()
+	t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0_0) |
+		((u64)__insn_mfspr(SPR_INTERRUPT_MASK_0_1) << 32);
+#else
+	t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0);
+#endif
+	t->ex_context[0] = __insn_mfspr(SPR_EX_CONTEXT_0_0);
+	t->ex_context[1] = __insn_mfspr(SPR_EX_CONTEXT_0_1);
+	t->system_save[0] = __insn_mfspr(SPR_SYSTEM_SAVE_0_0);
+	t->system_save[1] = __insn_mfspr(SPR_SYSTEM_SAVE_0_1);
+	t->system_save[2] = __insn_mfspr(SPR_SYSTEM_SAVE_0_2);
+	t->system_save[3] = __insn_mfspr(SPR_SYSTEM_SAVE_0_3);
+	t->intctrl_0 = __insn_mfspr(SPR_INTCTRL_0_STATUS);
+#if CHIP_HAS_PROC_STATUS_SPR()
+	t->proc_status = __insn_mfspr(SPR_PROC_STATUS);
+#endif
+#if !CHIP_HAS_FIXED_INTVEC_BASE()
+	t->interrupt_vector_base = __insn_mfspr(SPR_INTERRUPT_VECTOR_BASE_0);
+#endif
+#if CHIP_HAS_TILE_RTF_HWM()
+	t->tile_rtf_hwm = __insn_mfspr(SPR_TILE_RTF_HWM);
+#endif
+#if CHIP_HAS_DSTREAM_PF()
+	t->dstream_pf = __insn_mfspr(SPR_DSTREAM_PF);
+#endif
+}
+
+static void restore_arch_state(const struct thread_struct *t)
+{
+#if CHIP_HAS_SPLIT_INTR_MASK()
+	__insn_mtspr(SPR_INTERRUPT_MASK_0_0, (u32) t->interrupt_mask);
+	__insn_mtspr(SPR_INTERRUPT_MASK_0_1, t->interrupt_mask >> 32);
+#else
+	__insn_mtspr(SPR_INTERRUPT_MASK_0, t->interrupt_mask);
+#endif
+	__insn_mtspr(SPR_EX_CONTEXT_0_0, t->ex_context[0]);
+	__insn_mtspr(SPR_EX_CONTEXT_0_1, t->ex_context[1]);
+	__insn_mtspr(SPR_SYSTEM_SAVE_0_0, t->system_save[0]);
+	__insn_mtspr(SPR_SYSTEM_SAVE_0_1, t->system_save[1]);
+	__insn_mtspr(SPR_SYSTEM_SAVE_0_2, t->system_save[2]);
+	__insn_mtspr(SPR_SYSTEM_SAVE_0_3, t->system_save[3]);
+	__insn_mtspr(SPR_INTCTRL_0_STATUS, t->intctrl_0);
+#if CHIP_HAS_PROC_STATUS_SPR()
+	__insn_mtspr(SPR_PROC_STATUS, t->proc_status);
+#endif
+#if !CHIP_HAS_FIXED_INTVEC_BASE()
+	__insn_mtspr(SPR_INTERRUPT_VECTOR_BASE_0, t->interrupt_vector_base);
+#endif
+#if CHIP_HAS_TILE_RTF_HWM()
+	__insn_mtspr(SPR_TILE_RTF_HWM, t->tile_rtf_hwm);
+#endif
+#if CHIP_HAS_DSTREAM_PF()
+	__insn_mtspr(SPR_DSTREAM_PF, t->dstream_pf);
+#endif
+}
+
+
+void _prepare_arch_switch(struct task_struct *next)
+{
+#if CHIP_HAS_SN_PROC()
+	int snctl;
+#endif
+#if CHIP_HAS_TILE_DMA()
+	struct tile_dma_state *dma = &current->thread.tile_dma_state;
+	if (dma->enabled)
+		save_tile_dma_state(dma);
+#endif
+#if CHIP_HAS_SN_PROC()
+	/*
+	 * Suspend the static network processor if it was running.
+	 * We do not suspend the fabric itself, just like we don't
+	 * try to suspend the UDN.
+	 */
+	snctl = __insn_mfspr(SPR_SNCTL);
+	current->thread.sn_proc_running =
+		(snctl & SPR_SNCTL__FRZPROC_MASK) == 0;
+	if (current->thread.sn_proc_running)
+		__insn_mtspr(SPR_SNCTL, snctl | SPR_SNCTL__FRZPROC_MASK);
+#endif
+}
+
+
+struct task_struct *__sched _switch_to(struct task_struct *prev,
+				       struct task_struct *next)
+{
+	/* DMA state is already saved; save off other arch state. */
+	save_arch_state(&prev->thread);
+
+#if CHIP_HAS_TILE_DMA()
+	/*
+	 * Restore DMA in new task if desired.
+	 * Note that it is only safe to restart here since interrupts
+	 * are disabled, so we can't take any DMATLB miss or access
+	 * interrupts before we have finished switching stacks.
+	 */
+	if (next->thread.tile_dma_state.enabled) {
+		restore_tile_dma_state(&next->thread);
+		grant_dma_mpls();
+	} else {
+		restrict_dma_mpls();
+	}
+#endif
+
+	/* Restore other arch state. */
+	restore_arch_state(&next->thread);
+
+#if CHIP_HAS_SN_PROC()
+	/*
+	 * Restart static network processor in the new process
+	 * if it was running before.
+	 */
+	if (next->thread.sn_proc_running) {
+		int snctl = __insn_mfspr(SPR_SNCTL);
+		__insn_mtspr(SPR_SNCTL, snctl & ~SPR_SNCTL__FRZPROC_MASK);
+	}
+#endif
+
+#ifdef CONFIG_HARDWALL
+	/* Enable or disable access to the network registers appropriately. */
+	hardwall_switch_tasks(prev, next);
+#endif
+
+	/*
+	 * Switch kernel SP, PC, and callee-saved registers.
+	 * In the context of the new task, return the old task pointer
+	 * (i.e. the task that actually called __switch_to).
+	 * Pass the value to use for SYSTEM_SAVE_K_0 when we reset our sp.
+	 */
+	return __switch_to(prev, next, next_current_ksp0(next));
+}
+
+/*
+ * This routine is called on return from interrupt if any of the
+ * TIF_WORK_MASK flags are set in thread_info->flags.  It is
+ * entered with interrupts disabled so we don't miss an event
+ * that modified the thread_info flags.  If any flag is set, we
+ * handle it and return, and the calling assembly code will
+ * re-disable interrupts, reload the thread flags, and call back
+ * if more flags need to be handled.
+ *
+ * We return whether we need to check the thread_info flags again
+ * or not.  Note that we don't clear TIF_SINGLESTEP here, so it's
+ * important that it be tested last, and then claim that we don't
+ * need to recheck the flags.
+ */
+int do_work_pending(struct pt_regs *regs, u32 thread_info_flags)
+{
+	/* If we enter in kernel mode, do nothing and exit the caller loop. */
+	if (!user_mode(regs))
+		return 0;
+
+	/* Enable interrupts; they are disabled again on return to caller. */
+	local_irq_enable();
+
+	if (thread_info_flags & _TIF_NEED_RESCHED) {
+		schedule();
+		return 1;
+	}
+#if CHIP_HAS_TILE_DMA() || CHIP_HAS_SN_PROC()
+	if (thread_info_flags & _TIF_ASYNC_TLB) {
+		do_async_page_fault(regs);
+		return 1;
+	}
+#endif
+	if (thread_info_flags & _TIF_SIGPENDING) {
+		do_signal(regs);
+		return 1;
+	}
+	if (thread_info_flags & _TIF_NOTIFY_RESUME) {
+		clear_thread_flag(TIF_NOTIFY_RESUME);
+		tracehook_notify_resume(regs);
+		return 1;
+	}
+	if (thread_info_flags & _TIF_SINGLESTEP) {
+		single_step_once(regs);
+		return 0;
+	}
+	panic("work_pending: bad flags %#x\n", thread_info_flags);
+}
+
+unsigned long get_wchan(struct task_struct *p)
+{
+	struct KBacktraceIterator kbt;
+
+	if (!p || p == current || p->state == TASK_RUNNING)
+		return 0;
+
+	for (KBacktraceIterator_init(&kbt, p, NULL);
+	     !KBacktraceIterator_end(&kbt);
+	     KBacktraceIterator_next(&kbt)) {
+		if (!in_sched_functions(kbt.it.pc))
+			return kbt.it.pc;
+	}
+
+	return 0;
+}
+
+/* Flush thread state. */
+void flush_thread(void)
+{
+	/* Nothing */
+}
+
+/*
+ * Free current thread data structures etc..
+ */
+void exit_thread(void)
+{
+	/* Nothing */
+}
+
+void show_regs(struct pt_regs *regs)
+{
+	struct task_struct *tsk = validate_current();
+	int i;
+
+	pr_err("\n");
+	show_regs_print_info(KERN_ERR);
+#ifdef __tilegx__
+	for (i = 0; i < 51; i += 3)
+		pr_err(" r%-2d: "REGFMT" r%-2d: "REGFMT" r%-2d: "REGFMT"\n",
+		       i, regs->regs[i], i+1, regs->regs[i+1],
+		       i+2, regs->regs[i+2]);
+	pr_err(" r51: "REGFMT" r52: "REGFMT" tp : "REGFMT"\n",
+	       regs->regs[51], regs->regs[52], regs->tp);
+	pr_err(" sp : "REGFMT" lr : "REGFMT"\n", regs->sp, regs->lr);
+#else
+	for (i = 0; i < 52; i += 4)
+		pr_err(" r%-2d: "REGFMT" r%-2d: "REGFMT
+		       " r%-2d: "REGFMT" r%-2d: "REGFMT"\n",
+		       i, regs->regs[i], i+1, regs->regs[i+1],
+		       i+2, regs->regs[i+2], i+3, regs->regs[i+3]);
+	pr_err(" r52: "REGFMT" tp : "REGFMT" sp : "REGFMT" lr : "REGFMT"\n",
+	       regs->regs[52], regs->tp, regs->sp, regs->lr);
+#endif
+	pr_err(" pc : "REGFMT" ex1: %ld     faultnum: %ld\n",
+	       regs->pc, regs->ex1, regs->faultnum);
+
+	dump_stack_regs(regs);
+}

arch/tile/kernel/ptrace.c

@@ -0,0 +1,292 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * Copied from i386: Ross Biro 1/23/92
+ */
+
+#include <linux/kernel.h>
+#include <linux/ptrace.h>
+#include <linux/kprobes.h>
+#include <linux/compat.h>
+#include <linux/uaccess.h>
+#include <linux/regset.h>
+#include <linux/elf.h>
+#include <linux/tracehook.h>
+#include <asm/traps.h>
+#include <arch/chip.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/syscalls.h>
+
+void user_enable_single_step(struct task_struct *child)
+{
+	set_tsk_thread_flag(child, TIF_SINGLESTEP);
+}
+
+void user_disable_single_step(struct task_struct *child)
+{
+	clear_tsk_thread_flag(child, TIF_SINGLESTEP);
+}
+
+/*
+ * Called by kernel/ptrace.c when detaching..
+ */
+void ptrace_disable(struct task_struct *child)
+{
+	clear_tsk_thread_flag(child, TIF_SINGLESTEP);
+
+	/*
+	 * These two are currently unused, but will be set by arch_ptrace()
+	 * and used in the syscall assembly when we do support them.
+	 */
+	clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
+}
+
+/*
+ * Get registers from task and ready the result for userspace.
+ * Note that we localize the API issues to getregs() and putregs() at
+ * some cost in performance, e.g. we need a full pt_regs copy for
+ * PEEKUSR, and two copies for POKEUSR.  But in general we expect
+ * GETREGS/PUTREGS to be the API of choice anyway.
+ */
+static char *getregs(struct task_struct *child, struct pt_regs *uregs)
+{
+	*uregs = *task_pt_regs(child);
+
+	/* Set up flags ABI bits. */
+	uregs->flags = 0;
+#ifdef CONFIG_COMPAT
+	if (task_thread_info(child)->status & TS_COMPAT)
+		uregs->flags |= PT_FLAGS_COMPAT;
+#endif
+
+	return (char *)uregs;
+}
+
+/* Put registers back to task. */
+static void putregs(struct task_struct *child, struct pt_regs *uregs)
+{
+	struct pt_regs *regs = task_pt_regs(child);
+
+	/* Don't allow overwriting the kernel-internal flags word. */
+	uregs->flags = regs->flags;
+
+	/* Only allow setting the ICS bit in the ex1 word. */
+	uregs->ex1 = PL_ICS_EX1(USER_PL, EX1_ICS(uregs->ex1));
+
+	*regs = *uregs;
+}
+
+enum tile_regset {
+	REGSET_GPR,
+};
+
+static int tile_gpr_get(struct task_struct *target,
+			  const struct user_regset *regset,
+			  unsigned int pos, unsigned int count,
+			  void *kbuf, void __user *ubuf)
+{
+	struct pt_regs regs;
+
+	getregs(target, &regs);
+
+	return user_regset_copyout(&pos, &count, &kbuf, &ubuf, &regs, 0,
+				   sizeof(regs));
+}
+
+static int tile_gpr_set(struct task_struct *target,
+			  const struct user_regset *regset,
+			  unsigned int pos, unsigned int count,
+			  const void *kbuf, const void __user *ubuf)
+{
+	int ret;
+	struct pt_regs regs;
+
+	ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &regs, 0,
+				 sizeof(regs));
+	if (ret)
+		return ret;
+
+	putregs(target, &regs);
+
+	return 0;
+}
+
+static const struct user_regset tile_user_regset[] = {
+	[REGSET_GPR] = {
+		.core_note_type = NT_PRSTATUS,
+		.n = ELF_NGREG,
+		.size = sizeof(elf_greg_t),
+		.align = sizeof(elf_greg_t),
+		.get = tile_gpr_get,
+		.set = tile_gpr_set,
+	},
+};
+
+static const struct user_regset_view tile_user_regset_view = {
+	.name = CHIP_ARCH_NAME,
+	.e_machine = ELF_ARCH,
+	.ei_osabi = ELF_OSABI,
+	.regsets = tile_user_regset,
+	.n = ARRAY_SIZE(tile_user_regset),
+};
+
+const struct user_regset_view *task_user_regset_view(struct task_struct *task)
+{
+	return &tile_user_regset_view;
+}
+
+long arch_ptrace(struct task_struct *child, long request,
+		 unsigned long addr, unsigned long data)
+{
+	unsigned long __user *datap = (long __user __force *)data;
+	unsigned long tmp;
+	long ret = -EIO;
+	char *childreg;
+	struct pt_regs copyregs;
+
+	switch (request) {
+
+	case PTRACE_PEEKUSR:  /* Read register from pt_regs. */
+		if (addr >= PTREGS_SIZE)
+			break;
+		childreg = getregs(child, &copyregs) + addr;
+#ifdef CONFIG_COMPAT
+		if (is_compat_task()) {
+			if (addr & (sizeof(compat_long_t)-1))
+				break;
+			ret = put_user(*(compat_long_t *)childreg,
+				       (compat_long_t __user *)datap);
+		} else
+#endif
+		{
+			if (addr & (sizeof(long)-1))
+				break;
+			ret = put_user(*(long *)childreg, datap);
+		}
+		break;
+
+	case PTRACE_POKEUSR:  /* Write register in pt_regs. */
+		if (addr >= PTREGS_SIZE)
+			break;
+		childreg = getregs(child, &copyregs) + addr;
+#ifdef CONFIG_COMPAT
+		if (is_compat_task()) {
+			if (addr & (sizeof(compat_long_t)-1))
+				break;
+			*(compat_long_t *)childreg = data;
+		} else
+#endif
+		{
+			if (addr & (sizeof(long)-1))
+				break;
+			*(long *)childreg = data;
+		}
+		putregs(child, &copyregs);
+		ret = 0;
+		break;
+
+	case PTRACE_GETREGS:  /* Get all registers from the child. */
+		ret = copy_regset_to_user(child, &tile_user_regset_view,
+					  REGSET_GPR, 0,
+					  sizeof(struct pt_regs), datap);
+		break;
+
+	case PTRACE_SETREGS:  /* Set all registers in the child. */
+		ret = copy_regset_from_user(child, &tile_user_regset_view,
+					    REGSET_GPR, 0,
+					    sizeof(struct pt_regs), datap);
+		break;
+
+	case PTRACE_GETFPREGS:  /* Get the child FPU state. */
+	case PTRACE_SETFPREGS:  /* Set the child FPU state. */
+		break;
+
+	case PTRACE_SETOPTIONS:
+		/* Support TILE-specific ptrace options. */
+		BUILD_BUG_ON(PTRACE_O_MASK_TILE & PTRACE_O_MASK);
+		tmp = data & PTRACE_O_MASK_TILE;
+		data &= ~PTRACE_O_MASK_TILE;
+		ret = ptrace_request(child, request, addr, data);
+		if (ret == 0) {
+			unsigned int flags = child->ptrace;
+			flags &= ~(PTRACE_O_MASK_TILE << PT_OPT_FLAG_SHIFT);
+			flags |= (tmp << PT_OPT_FLAG_SHIFT);
+			child->ptrace = flags;
+		}
+		break;
+
+	default:
+#ifdef CONFIG_COMPAT
+		if (task_thread_info(current)->status & TS_COMPAT) {
+			ret = compat_ptrace_request(child, request,
+						    addr, data);
+			break;
+		}
+#endif
+		ret = ptrace_request(child, request, addr, data);
+		break;
+	}
+
+	return ret;
+}
+
+#ifdef CONFIG_COMPAT
+/* Not used; we handle compat issues in arch_ptrace() directly. */
+long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
+			       compat_ulong_t addr, compat_ulong_t data)
+{
+	BUG();
+}
+#endif
+
+int do_syscall_trace_enter(struct pt_regs *regs)
+{
+	if (test_thread_flag(TIF_SYSCALL_TRACE)) {
+		if (tracehook_report_syscall_entry(regs))
+			regs->regs[TREG_SYSCALL_NR] = -1;
+	}
+
+	if (test_thread_flag(TIF_SYSCALL_TRACEPOINT))
+		trace_sys_enter(regs, regs->regs[TREG_SYSCALL_NR]);
+
+	return regs->regs[TREG_SYSCALL_NR];
+}
+
+void do_syscall_trace_exit(struct pt_regs *regs)
+{
+	if (test_thread_flag(TIF_SYSCALL_TRACE))
+		tracehook_report_syscall_exit(regs, 0);
+
+	if (test_thread_flag(TIF_SYSCALL_TRACEPOINT))
+		trace_sys_exit(regs, regs->regs[0]);
+}
+
+void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
+{
+	struct siginfo info;
+
+	memset(&info, 0, sizeof(info));
+	info.si_signo = SIGTRAP;
+	info.si_code  = TRAP_BRKPT;
+	info.si_addr  = (void __user *) regs->pc;
+
+	/* Send us the fakey SIGTRAP */
+	force_sig_info(SIGTRAP, &info, tsk);
+}
+
+/* Handle synthetic interrupt delivered only by the simulator. */
+void __kprobes do_breakpoint(struct pt_regs* regs, int fault_num)
+{
+	send_sigtrap(current, regs, fault_num);
+}

arch/tile/kernel/reboot.c

@@ -0,0 +1,53 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/stddef.h>
+#include <linux/reboot.h>
+#include <linux/smp.h>
+#include <linux/pm.h>
+#include <linux/export.h>
+#include <asm/page.h>
+#include <asm/setup.h>
+#include <hv/hypervisor.h>
+
+#ifndef CONFIG_SMP
+#define smp_send_stop()
+#endif
+
+void machine_halt(void)
+{
+	warn_early_printk();
+	arch_local_irq_disable_all();
+	smp_send_stop();
+	hv_halt();
+}
+
+void machine_power_off(void)
+{
+	warn_early_printk();
+	arch_local_irq_disable_all();
+	smp_send_stop();
+	hv_power_off();
+}
+
+void machine_restart(char *cmd)
+{
+	arch_local_irq_disable_all();
+	smp_send_stop();
+	hv_restart((HV_VirtAddr) "vmlinux", (HV_VirtAddr) cmd);
+}
+
+/* No interesting distinction to be made here. */
+void (*pm_power_off)(void) = NULL;
+EXPORT_SYMBOL(pm_power_off);

arch/tile/kernel/regs_32.S

@@ -0,0 +1,145 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/linkage.h>
+#include <asm/ptrace.h>
+#include <asm/asm-offsets.h>
+#include <arch/spr_def.h>
+#include <asm/processor.h>
+#include <asm/switch_to.h>
+
+/*
+ * See <asm/system.h>; called with prev and next task_struct pointers.
+ * "prev" is returned in r0 for _switch_to and also for ret_from_fork.
+ *
+ * We want to save pc/sp in "prev", and get the new pc/sp from "next".
+ * We also need to save all the callee-saved registers on the stack.
+ *
+ * Intel enables/disables access to the hardware cycle counter in
+ * seccomp (secure computing) environments if necessary, based on
+ * has_secure_computing().  We might want to do this at some point,
+ * though it would require virtualizing the other SPRs under WORLD_ACCESS.
+ *
+ * Since we're saving to the stack, we omit sp from this list.
+ * And for parallels with other architectures, we save lr separately,
+ * in the thread_struct itself (as the "pc" field).
+ *
+ * This code also needs to be aligned with process.c copy_thread()
+ */
+
+#if CALLEE_SAVED_REGS_COUNT != 24
+# error Mismatch between <asm/system.h> and kernel/entry.S
+#endif
+#define FRAME_SIZE ((2 + CALLEE_SAVED_REGS_COUNT) * 4)
+
+#define SAVE_REG(r) { sw r12, r; addi r12, r12, 4 }
+#define LOAD_REG(r) { lw r, r12; addi r12, r12, 4 }
+#define FOR_EACH_CALLEE_SAVED_REG(f)					\
+							f(r30); f(r31); \
+	f(r32); f(r33); f(r34); f(r35);	f(r36); f(r37); f(r38); f(r39); \
+	f(r40); f(r41); f(r42); f(r43); f(r44); f(r45); f(r46); f(r47); \
+	f(r48); f(r49); f(r50); f(r51); f(r52);
+
+STD_ENTRY_SECTION(__switch_to, .sched.text)
+	{
+	  move r10, sp
+	  sw sp, lr
+	  addi sp, sp, -FRAME_SIZE
+	}
+	{
+	  addi r11, sp, 4
+	  addi r12, sp, 8
+	}
+	{
+	  sw r11, r10
+	  addli r4, r1, TASK_STRUCT_THREAD_KSP_OFFSET
+	}
+	{
+	  lw r13, r4   /* Load new sp to a temp register early. */
+	  addli r3, r0, TASK_STRUCT_THREAD_KSP_OFFSET
+	}
+	FOR_EACH_CALLEE_SAVED_REG(SAVE_REG)
+	{
+	  sw r3, sp
+	  addli r3, r0, TASK_STRUCT_THREAD_PC_OFFSET
+	}
+	{
+	  sw r3, lr
+	  addli r4, r1, TASK_STRUCT_THREAD_PC_OFFSET
+	}
+	{
+	  lw lr, r4
+	  addi r12, r13, 8
+	}
+	{
+	  /* Update sp and ksp0 simultaneously to avoid backtracer warnings. */
+	  move sp, r13
+	  mtspr SPR_SYSTEM_SAVE_K_0, r2
+	}
+	FOR_EACH_CALLEE_SAVED_REG(LOAD_REG)
+.L__switch_to_pc:
+	{
+	  addi sp, sp, FRAME_SIZE
+	  jrp lr   /* r0 is still valid here, so return it */
+	}
+	STD_ENDPROC(__switch_to)
+
+/* Return a suitable address for the backtracer for suspended threads */
+STD_ENTRY_SECTION(get_switch_to_pc, .sched.text)
+	lnk r0
+	{
+	  addli r0, r0, .L__switch_to_pc - .
+	  jrp lr
+	}
+	STD_ENDPROC(get_switch_to_pc)
+
+STD_ENTRY(get_pt_regs)
+	.irp reg, r0, r1, r2, r3, r4, r5, r6, r7, \
+		 r8, r9, r10, r11, r12, r13, r14, r15, \
+		 r16, r17, r18, r19, r20, r21, r22, r23, \
+		 r24, r25, r26, r27, r28, r29, r30, r31, \
+		 r32, r33, r34, r35, r36, r37, r38, r39, \
+		 r40, r41, r42, r43, r44, r45, r46, r47, \
+		 r48, r49, r50, r51, r52, tp, sp
+	{
+	 sw r0, \reg
+	 addi r0, r0, 4
+	}
+	.endr
+	{
+	 sw r0, lr
+	 addi r0, r0, PTREGS_OFFSET_PC - PTREGS_OFFSET_LR
+	}
+	lnk r1
+	{
+	 sw r0, r1
+	 addi r0, r0, PTREGS_OFFSET_EX1 - PTREGS_OFFSET_PC
+	}
+	mfspr r1, INTERRUPT_CRITICAL_SECTION
+	shli r1, r1, SPR_EX_CONTEXT_1_1__ICS_SHIFT
+	ori r1, r1, KERNEL_PL
+	{
+	 sw r0, r1
+	 addi r0, r0, PTREGS_OFFSET_FAULTNUM - PTREGS_OFFSET_EX1
+	}
+	{
+	 sw r0, zero       /* clear faultnum */
+	 addi r0, r0, PTREGS_OFFSET_ORIG_R0 - PTREGS_OFFSET_FAULTNUM
+	}
+	{
+	 sw r0, zero       /* clear orig_r0 */
+	 addli r0, r0, -PTREGS_OFFSET_ORIG_R0    /* restore r0 to base */
+	}
+	jrp lr
+	STD_ENDPROC(get_pt_regs)

arch/tile/kernel/regs_64.S

@@ -0,0 +1,145 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/linkage.h>
+#include <asm/ptrace.h>
+#include <asm/asm-offsets.h>
+#include <arch/spr_def.h>
+#include <asm/processor.h>
+#include <asm/switch_to.h>
+
+/*
+ * See <asm/system.h>; called with prev and next task_struct pointers.
+ * "prev" is returned in r0 for _switch_to and also for ret_from_fork.
+ *
+ * We want to save pc/sp in "prev", and get the new pc/sp from "next".
+ * We also need to save all the callee-saved registers on the stack.
+ *
+ * Intel enables/disables access to the hardware cycle counter in
+ * seccomp (secure computing) environments if necessary, based on
+ * has_secure_computing().  We might want to do this at some point,
+ * though it would require virtualizing the other SPRs under WORLD_ACCESS.
+ *
+ * Since we're saving to the stack, we omit sp from this list.
+ * And for parallels with other architectures, we save lr separately,
+ * in the thread_struct itself (as the "pc" field).
+ *
+ * This code also needs to be aligned with process.c copy_thread()
+ */
+
+#if CALLEE_SAVED_REGS_COUNT != 24
+# error Mismatch between <asm/system.h> and kernel/entry.S
+#endif
+#define FRAME_SIZE ((2 + CALLEE_SAVED_REGS_COUNT) * 8)
+
+#define SAVE_REG(r) { st r12, r; addi r12, r12, 8 }
+#define LOAD_REG(r) { ld r, r12; addi r12, r12, 8 }
+#define FOR_EACH_CALLEE_SAVED_REG(f)					\
+							f(r30); f(r31); \
+	f(r32); f(r33); f(r34); f(r35);	f(r36); f(r37); f(r38); f(r39); \
+	f(r40); f(r41); f(r42); f(r43); f(r44); f(r45); f(r46); f(r47); \
+	f(r48); f(r49); f(r50); f(r51); f(r52);
+
+STD_ENTRY_SECTION(__switch_to, .sched.text)
+	{
+	  move r10, sp
+	  st sp, lr
+	}
+	{
+	  addli r11, sp, -FRAME_SIZE + 8
+	  addli sp, sp, -FRAME_SIZE
+	}
+	{
+	  st r11, r10
+	  addli r4, r1, TASK_STRUCT_THREAD_KSP_OFFSET
+	}
+	{
+	  ld r13, r4   /* Load new sp to a temp register early. */
+	  addi r12, sp, 16
+	}
+	FOR_EACH_CALLEE_SAVED_REG(SAVE_REG)
+	addli r3, r0, TASK_STRUCT_THREAD_KSP_OFFSET
+	{
+	  st r3, sp
+	  addli r3, r0, TASK_STRUCT_THREAD_PC_OFFSET
+	}
+	{
+	  st r3, lr
+	  addli r4, r1, TASK_STRUCT_THREAD_PC_OFFSET
+	}
+	{
+	  ld lr, r4
+	  addi r12, r13, 16
+	}
+	{
+	  /* Update sp and ksp0 simultaneously to avoid backtracer warnings. */
+	  move sp, r13
+	  mtspr SPR_SYSTEM_SAVE_K_0, r2
+	}
+	FOR_EACH_CALLEE_SAVED_REG(LOAD_REG)
+.L__switch_to_pc:
+	{
+	  addli sp, sp, FRAME_SIZE
+	  jrp lr   /* r0 is still valid here, so return it */
+	}
+	STD_ENDPROC(__switch_to)
+
+/* Return a suitable address for the backtracer for suspended threads */
+STD_ENTRY_SECTION(get_switch_to_pc, .sched.text)
+	lnk r0
+	{
+	  addli r0, r0, .L__switch_to_pc - .
+	  jrp lr
+	}
+	STD_ENDPROC(get_switch_to_pc)
+
+STD_ENTRY(get_pt_regs)
+	.irp reg, r0, r1, r2, r3, r4, r5, r6, r7, \
+		 r8, r9, r10, r11, r12, r13, r14, r15, \
+		 r16, r17, r18, r19, r20, r21, r22, r23, \
+		 r24, r25, r26, r27, r28, r29, r30, r31, \
+		 r32, r33, r34, r35, r36, r37, r38, r39, \
+		 r40, r41, r42, r43, r44, r45, r46, r47, \
+		 r48, r49, r50, r51, r52, tp, sp
+	{
+	 st r0, \reg
+	 addi r0, r0, 8
+	}
+	.endr
+	{
+	 st r0, lr
+	 addi r0, r0, PTREGS_OFFSET_PC - PTREGS_OFFSET_LR
+	}
+	lnk r1
+	{
+	 st r0, r1
+	 addi r0, r0, PTREGS_OFFSET_EX1 - PTREGS_OFFSET_PC
+	}
+	mfspr r1, INTERRUPT_CRITICAL_SECTION
+	shli r1, r1, SPR_EX_CONTEXT_1_1__ICS_SHIFT
+	ori r1, r1, KERNEL_PL
+	{
+	 st r0, r1
+	 addi r0, r0, PTREGS_OFFSET_FAULTNUM - PTREGS_OFFSET_EX1
+	}
+	{
+	 st r0, zero       /* clear faultnum */
+	 addi r0, r0, PTREGS_OFFSET_ORIG_R0 - PTREGS_OFFSET_FAULTNUM
+	}
+	{
+	 st r0, zero       /* clear orig_r0 */
+	 addli r0, r0, -PTREGS_OFFSET_ORIG_R0    /* restore r0 to base */
+	}
+	jrp lr
+	STD_ENDPROC(get_pt_regs)

arch/tile/kernel/relocate_kernel_32.S

@@ -0,0 +1,280 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * copy new kernel into place and then call hv_reexec
+ *
+ */
+
+#include <linux/linkage.h>
+#include <arch/chip.h>
+#include <asm/page.h>
+#include <hv/hypervisor.h>
+
+#define ___hvb	MEM_SV_INTRPT + HV_GLUE_START_CPA
+
+#define ___hv_dispatch(f) (___hvb + (HV_DISPATCH_ENTRY_SIZE * f))
+
+#define ___hv_console_putc ___hv_dispatch(HV_DISPATCH_CONSOLE_PUTC)
+#define ___hv_halt         ___hv_dispatch(HV_DISPATCH_HALT)
+#define ___hv_reexec       ___hv_dispatch(HV_DISPATCH_REEXEC)
+#define ___hv_flush_remote ___hv_dispatch(HV_DISPATCH_FLUSH_REMOTE)
+
+#undef RELOCATE_NEW_KERNEL_VERBOSE
+
+STD_ENTRY(relocate_new_kernel)
+
+	move	r30, r0		/* page list */
+	move	r31, r1		/* address of page we are on */
+	move	r32, r2		/* start address of new kernel */
+
+	shri	r1, r1, PAGE_SHIFT
+	addi	r1, r1, 1
+	shli	sp, r1, PAGE_SHIFT
+	addi	sp, sp, -8
+	/* we now have a stack (whether we need one or not) */
+
+	moveli	r40, lo16(___hv_console_putc)
+	auli	r40, r40, ha16(___hv_console_putc)
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 'r'
+	jalr	r40
+
+	moveli	r0, '_'
+	jalr	r40
+
+	moveli	r0, 'n'
+	jalr	r40
+
+	moveli	r0, '_'
+	jalr	r40
+
+	moveli	r0, 'k'
+	jalr	r40
+
+	moveli	r0, '\n'
+	jalr	r40
+#endif
+
+	/*
+	 * Throughout this code r30 is pointer to the element of page
+	 * list we are working on.
+	 *
+	 * Normally we get to the next element of the page list by
+	 * incrementing r30 by four.  The exception is if the element
+	 * on the page list is an IND_INDIRECTION in which case we use
+	 * the element with the low bits masked off as the new value
+	 * of r30.
+	 *
+	 * To get this started, we need the value passed to us (which
+	 * will always be an IND_INDIRECTION) in memory somewhere with
+	 * r30 pointing at it.  To do that, we push the value passed
+	 * to us on the stack and make r30 point to it.
+	 */
+
+	sw	sp, r30
+	move	r30, sp
+	addi	sp, sp, -8
+
+#if CHIP_HAS_CBOX_HOME_MAP()
+	/*
+	 * On TILEPro, we need to flush all tiles' caches, since we may
+	 * have been doing hash-for-home caching there.  Note that we
+	 * must do this _after_ we're completely done modifying any memory
+	 * other than our output buffer (which we know is locally cached).
+	 * We want the caches to be fully clean when we do the reexec,
+	 * because the hypervisor is going to do this flush again at that
+	 * point, and we don't want that second flush to overwrite any memory.
+	 */
+	{
+	 move	r0, zero	 /* cache_pa */
+	 move	r1, zero
+	}
+	{
+	 auli	r2, zero, ha16(HV_FLUSH_EVICT_L2) /* cache_control */
+	 movei	r3, -1		 /* cache_cpumask; -1 means all client tiles */
+	}
+	{
+	 move	r4, zero	 /* tlb_va */
+	 move	r5, zero	 /* tlb_length */
+	}
+	{
+	 move	r6, zero	 /* tlb_pgsize */
+	 move	r7, zero	 /* tlb_cpumask */
+	}
+	{
+	 move	r8, zero	 /* asids */
+	 moveli	r20, lo16(___hv_flush_remote)
+	}
+	{
+	 move	r9, zero	 /* asidcount */
+	 auli	r20, r20, ha16(___hv_flush_remote)
+	}
+
+	jalr	r20
+#endif
+
+	/* r33 is destination pointer, default to zero */
+
+	moveli	r33, 0
+
+.Lloop:	lw	r10, r30
+
+	andi	r9, r10, 0xf	/* low 4 bits tell us what type it is */
+	xor	r10, r10, r9	/* r10 is now value with low 4 bits stripped */
+
+	seqi	r0, r9, 0x1	/* IND_DESTINATION */
+	bzt	r0, .Ltry2
+
+	move	r33, r10
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 'd'
+	jalr	r40
+#endif
+
+	addi	r30, r30, 4
+	j	.Lloop
+
+.Ltry2:
+	seqi	r0, r9, 0x2	/* IND_INDIRECTION */
+	bzt	r0, .Ltry4
+
+	move	r30, r10
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 'i'
+	jalr	r40
+#endif
+
+	j	.Lloop
+
+.Ltry4:
+	seqi	r0, r9, 0x4	/* IND_DONE */
+	bzt	r0, .Ltry8
+
+	mf
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 'D'
+	jalr	r40
+	moveli	r0, '\n'
+	jalr	r40
+#endif
+
+	move	r0, r32
+	moveli	r1, 0		/* arg to hv_reexec is 64 bits */
+
+	moveli	r41, lo16(___hv_reexec)
+	auli	r41, r41, ha16(___hv_reexec)
+
+	jalr	r41
+
+	/* we should not get here */
+
+	moveli	r0, '?'
+	jalr	r40
+	moveli	r0, '\n'
+	jalr	r40
+
+	j	.Lhalt
+
+.Ltry8:	seqi	r0, r9, 0x8	/* IND_SOURCE */
+	bz	r0, .Lerr	/* unknown type */
+
+	/* copy page at r10 to page at r33 */
+
+	move	r11, r33
+
+	moveli	r0, lo16(PAGE_SIZE)
+	auli	r0, r0, ha16(PAGE_SIZE)
+	add	r33, r33, r0
+
+	/* copy word at r10 to word at r11 until r11 equals r33 */
+
+	/* We know page size must be multiple of 16, so we can unroll
+	 * 16 times safely without any edge case checking.
+	 *
+	 * Issue a flush of the destination every 16 words to avoid
+	 * incoherence when starting the new kernel.  (Now this is
+	 * just good paranoia because the hv_reexec call will also
+	 * take care of this.)
+	 */
+
+1:
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0; addi	r11, r11, 4 }
+	{ lw	r0, r10; addi	r10, r10, 4 }
+	{ sw	r11, r0 }
+	{ flush r11    ; addi	r11, r11, 4 }
+
+	seq	r0, r33, r11
+	bzt	r0, 1b
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 's'
+	jalr	r40
+#endif
+
+	addi	r30, r30, 4
+	j	.Lloop
+
+
+.Lerr:	moveli	r0, 'e'
+	jalr	r40
+	moveli	r0, 'r'
+	jalr	r40
+	moveli	r0, 'r'
+	jalr	r40
+	moveli	r0, '\n'
+	jalr	r40
+.Lhalt:
+	moveli	r41, lo16(___hv_halt)
+	auli	r41, r41, ha16(___hv_halt)
+
+	jalr	r41
+	STD_ENDPROC(relocate_new_kernel)
+
+	.section .rodata,"a"
+
+	.globl relocate_new_kernel_size
+relocate_new_kernel_size:
+	.long .Lend_relocate_new_kernel - relocate_new_kernel

arch/tile/kernel/relocate_kernel_64.S

@@ -0,0 +1,260 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * copy new kernel into place and then call hv_reexec
+ *
+ */
+
+#include <linux/linkage.h>
+#include <arch/chip.h>
+#include <asm/page.h>
+#include <hv/hypervisor.h>
+
+#undef RELOCATE_NEW_KERNEL_VERBOSE
+
+STD_ENTRY(relocate_new_kernel)
+
+	move	r30, r0		/* page list */
+	move	r31, r1		/* address of page we are on */
+	move	r32, r2		/* start address of new kernel */
+
+	shrui	r1, r1, PAGE_SHIFT
+	addi	r1, r1, 1
+	shli	sp, r1, PAGE_SHIFT
+	addi	sp, sp, -8
+	/* we now have a stack (whether we need one or not) */
+
+	moveli	r40, hw2_last(hv_console_putc)
+	shl16insli r40, r40, hw1(hv_console_putc)
+	shl16insli r40, r40, hw0(hv_console_putc)
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 'r'
+	jalr	r40
+
+	moveli	r0, '_'
+	jalr	r40
+
+	moveli	r0, 'n'
+	jalr	r40
+
+	moveli	r0, '_'
+	jalr	r40
+
+	moveli	r0, 'k'
+	jalr	r40
+
+	moveli	r0, '\n'
+	jalr	r40
+#endif
+
+	/*
+	 * Throughout this code r30 is pointer to the element of page
+	 * list we are working on.
+	 *
+	 * Normally we get to the next element of the page list by
+	 * incrementing r30 by eight.  The exception is if the element
+	 * on the page list is an IND_INDIRECTION in which case we use
+	 * the element with the low bits masked off as the new value
+	 * of r30.
+	 *
+	 * To get this started, we need the value passed to us (which
+	 * will always be an IND_INDIRECTION) in memory somewhere with
+	 * r30 pointing at it.  To do that, we push the value passed
+	 * to us on the stack and make r30 point to it.
+	 */
+
+	st	sp, r30
+	move	r30, sp
+	addi	sp, sp, -16
+
+#if CHIP_HAS_CBOX_HOME_MAP()
+	/*
+	 * On TILE-GX, we need to flush all tiles' caches, since we may
+	 * have been doing hash-for-home caching there.  Note that we
+	 * must do this _after_ we're completely done modifying any memory
+	 * other than our output buffer (which we know is locally cached).
+	 * We want the caches to be fully clean when we do the reexec,
+	 * because the hypervisor is going to do this flush again at that
+	 * point, and we don't want that second flush to overwrite any memory.
+	 */
+	{
+	 move	r0, zero	 /* cache_pa */
+	 moveli	r1, hw2_last(HV_FLUSH_EVICT_L2)
+	}
+	{
+	 shl16insli	r1, r1, hw1(HV_FLUSH_EVICT_L2)
+	 movei	r2, -1		 /* cache_cpumask; -1 means all client tiles */
+	}
+	{
+	 shl16insli	r1, r1, hw0(HV_FLUSH_EVICT_L2)  /* cache_control */
+	 move	r3, zero	 /* tlb_va */
+	}
+	{
+	 move	r4, zero	 /* tlb_length */
+	 move	r5, zero	 /* tlb_pgsize */
+	}
+	{
+	 move	r6, zero	 /* tlb_cpumask */
+	 move	r7, zero	 /* asids */
+	}
+	{
+	 moveli	r20, hw2_last(hv_flush_remote)
+	 move	r8, zero	 /* asidcount */
+	}
+	shl16insli	r20, r20, hw1(hv_flush_remote)
+	shl16insli	r20, r20, hw0(hv_flush_remote)
+
+	jalr	r20
+#endif
+
+	/* r33 is destination pointer, default to zero */
+
+	moveli	r33, 0
+
+.Lloop:	ld	r10, r30
+
+	andi	r9, r10, 0xf	/* low 4 bits tell us what type it is */
+	xor	r10, r10, r9	/* r10 is now value with low 4 bits stripped */
+
+	cmpeqi	r0, r9, 0x1	/* IND_DESTINATION */
+	beqzt	r0, .Ltry2
+
+	move	r33, r10
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 'd'
+	jalr	r40
+#endif
+
+	addi	r30, r30, 8
+	j	.Lloop
+
+.Ltry2:
+	cmpeqi	r0, r9, 0x2	/* IND_INDIRECTION */
+	beqzt	r0, .Ltry4
+
+	move	r30, r10
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 'i'
+	jalr	r40
+#endif
+
+	j	.Lloop
+
+.Ltry4:
+	cmpeqi	r0, r9, 0x4	/* IND_DONE */
+	beqzt	r0, .Ltry8
+
+	mf
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 'D'
+	jalr	r40
+	moveli	r0, '\n'
+	jalr	r40
+#endif
+
+	move	r0, r32
+
+	moveli	r41, hw2_last(hv_reexec)
+	shl16insli	r41, r41, hw1(hv_reexec)
+	shl16insli	r41, r41, hw0(hv_reexec)
+
+	jalr	r41
+
+	/* we should not get here */
+
+	moveli	r0, '?'
+	jalr	r40
+	moveli	r0, '\n'
+	jalr	r40
+
+	j	.Lhalt
+
+.Ltry8:	cmpeqi	r0, r9, 0x8	/* IND_SOURCE */
+	beqz	r0, .Lerr	/* unknown type */
+
+	/* copy page at r10 to page at r33 */
+
+	move	r11, r33
+
+	moveli	r0, hw2_last(PAGE_SIZE)
+	shl16insli	r0, r0, hw1(PAGE_SIZE)
+	shl16insli	r0, r0, hw0(PAGE_SIZE)
+	add	r33, r33, r0
+
+	/* copy word at r10 to word at r11 until r11 equals r33 */
+
+	/* We know page size must be multiple of 8, so we can unroll
+	 * 8 times safely without any edge case checking.
+	 *
+	 * Issue a flush of the destination every 8 words to avoid
+	 * incoherence when starting the new kernel.  (Now this is
+	 * just good paranoia because the hv_reexec call will also
+	 * take care of this.)
+	 */
+
+1:
+	{ ld	r0, r10; addi	r10, r10, 8 }
+	{ st	r11, r0; addi	r11, r11, 8 }
+	{ ld	r0, r10; addi	r10, r10, 8 }
+	{ st	r11, r0; addi	r11, r11, 8 }
+	{ ld	r0, r10; addi	r10, r10, 8 }
+	{ st	r11, r0; addi	r11, r11, 8 }
+	{ ld	r0, r10; addi	r10, r10, 8 }
+	{ st	r11, r0; addi	r11, r11, 8 }
+	{ ld	r0, r10; addi	r10, r10, 8 }
+	{ st	r11, r0; addi	r11, r11, 8 }
+	{ ld	r0, r10; addi	r10, r10, 8 }
+	{ st	r11, r0; addi	r11, r11, 8 }
+	{ ld	r0, r10; addi	r10, r10, 8 }
+	{ st	r11, r0; addi	r11, r11, 8 }
+	{ ld	r0, r10; addi	r10, r10, 8 }
+	{ st	r11, r0 }
+	{ flush r11    ; addi	r11, r11, 8 }
+
+	cmpeq	r0, r33, r11
+	beqzt	r0, 1b
+
+#ifdef RELOCATE_NEW_KERNEL_VERBOSE
+	moveli	r0, 's'
+	jalr	r40
+#endif
+
+	addi	r30, r30, 8
+	j	.Lloop
+
+
+.Lerr:	moveli	r0, 'e'
+	jalr	r40
+	moveli	r0, 'r'
+	jalr	r40
+	moveli	r0, 'r'
+	jalr	r40
+	moveli	r0, '\n'
+	jalr	r40
+.Lhalt:
+	moveli r41, hw2_last(hv_halt)
+	shl16insli r41, r41, hw1(hv_halt)
+	shl16insli r41, r41, hw0(hv_halt)
+
+	jalr	r41
+	STD_ENDPROC(relocate_new_kernel)
+
+	.section .rodata,"a"
+
+	.globl relocate_new_kernel_size
+relocate_new_kernel_size:
+	.long .Lend_relocate_new_kernel - relocate_new_kernel

arch/tile/kernel/signal.c

@@ -0,0 +1,433 @@
+/*
+ * Copyright (C) 1991, 1992  Linus Torvalds
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/smp.h>
+#include <linux/kernel.h>
+#include <linux/signal.h>
+#include <linux/errno.h>
+#include <linux/wait.h>
+#include <linux/unistd.h>
+#include <linux/stddef.h>
+#include <linux/personality.h>
+#include <linux/suspend.h>
+#include <linux/ptrace.h>
+#include <linux/elf.h>
+#include <linux/compat.h>
+#include <linux/syscalls.h>
+#include <linux/uaccess.h>
+#include <asm/processor.h>
+#include <asm/ucontext.h>
+#include <asm/sigframe.h>
+#include <asm/syscalls.h>
+#include <asm/vdso.h>
+#include <arch/interrupts.h>
+
+#define DEBUG_SIG 0
+
+/*
+ * Do a signal return; undo the signal stack.
+ */
+
+int restore_sigcontext(struct pt_regs *regs,
+		       struct sigcontext __user *sc)
+{
+	int err = 0;
+	int i;
+
+	/* Always make any pending restarted system calls return -EINTR */
+	current_thread_info()->restart_block.fn = do_no_restart_syscall;
+
+	/*
+	 * Enforce that sigcontext is like pt_regs, and doesn't mess
+	 * up our stack alignment rules.
+	 */
+	BUILD_BUG_ON(sizeof(struct sigcontext) != sizeof(struct pt_regs));
+	BUILD_BUG_ON(sizeof(struct sigcontext) % 8 != 0);
+
+	for (i = 0; i < sizeof(struct pt_regs)/sizeof(long); ++i)
+		err |= __get_user(regs->regs[i], &sc->gregs[i]);
+
+	/* Ensure that the PL is always set to USER_PL. */
+	regs->ex1 = PL_ICS_EX1(USER_PL, EX1_ICS(regs->ex1));
+
+	regs->faultnum = INT_SWINT_1_SIGRETURN;
+
+	return err;
+}
+
+void signal_fault(const char *type, struct pt_regs *regs,
+		  void __user *frame, int sig)
+{
+	trace_unhandled_signal(type, regs, (unsigned long)frame, SIGSEGV);
+	force_sigsegv(sig, current);
+}
+
+/* The assembly shim for this function arranges to ignore the return value. */
+SYSCALL_DEFINE0(rt_sigreturn)
+{
+	struct pt_regs *regs = current_pt_regs();
+	struct rt_sigframe __user *frame =
+		(struct rt_sigframe __user *)(regs->sp);
+	sigset_t set;
+
+	if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
+		goto badframe;
+	if (__copy_from_user(&set, &frame->uc.uc_sigmask, sizeof(set)))
+		goto badframe;
+
+	set_current_blocked(&set);
+
+	if (restore_sigcontext(regs, &frame->uc.uc_mcontext))
+		goto badframe;
+
+	if (restore_altstack(&frame->uc.uc_stack))
+		goto badframe;
+
+	return 0;
+
+badframe:
+	signal_fault("bad sigreturn frame", regs, frame, 0);
+	return 0;
+}
+
+/*
+ * Set up a signal frame.
+ */
+
+int setup_sigcontext(struct sigcontext __user *sc, struct pt_regs *regs)
+{
+	int i, err = 0;
+
+	for (i = 0; i < sizeof(struct pt_regs)/sizeof(long); ++i)
+		err |= __put_user(regs->regs[i], &sc->gregs[i]);
+
+	return err;
+}
+
+/*
+ * Determine which stack to use..
+ */
+static inline void __user *get_sigframe(struct k_sigaction *ka,
+					struct pt_regs *regs,
+					size_t frame_size)
+{
+	unsigned long sp;
+
+	/* Default to using normal stack */
+	sp = regs->sp;
+
+	/*
+	 * If we are on the alternate signal stack and would overflow
+	 * it, don't.  Return an always-bogus address instead so we
+	 * will die with SIGSEGV.
+	 */
+	if (on_sig_stack(sp) && !likely(on_sig_stack(sp - frame_size)))
+		return (void __user __force *)-1UL;
+
+	/* This is the X/Open sanctioned signal stack switching.  */
+	if (ka->sa.sa_flags & SA_ONSTACK) {
+		if (sas_ss_flags(sp) == 0)
+			sp = current->sas_ss_sp + current->sas_ss_size;
+	}
+
+	sp -= frame_size;
+	/*
+	 * Align the stack pointer according to the TILE ABI,
+	 * i.e. so that on function entry (sp & 15) == 0.
+	 */
+	sp &= -16UL;
+	return (void __user *) sp;
+}
+
+static int setup_rt_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
+			   sigset_t *set, struct pt_regs *regs)
+{
+	unsigned long restorer;
+	struct rt_sigframe __user *frame;
+	int err = 0;
+	int usig;
+
+	frame = get_sigframe(ka, regs, sizeof(*frame));
+
+	if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
+		goto give_sigsegv;
+
+	usig = current_thread_info()->exec_domain
+		&& current_thread_info()->exec_domain->signal_invmap
+		&& sig < 32
+		? current_thread_info()->exec_domain->signal_invmap[sig]
+		: sig;
+
+	/* Always write at least the signal number for the stack backtracer. */
+	if (ka->sa.sa_flags & SA_SIGINFO) {
+		/* At sigreturn time, restore the callee-save registers too. */
+		err |= copy_siginfo_to_user(&frame->info, info);
+		regs->flags |= PT_FLAGS_RESTORE_REGS;
+	} else {
+		err |= __put_user(info->si_signo, &frame->info.si_signo);
+	}
+
+	/* Create the ucontext.  */
+	err |= __clear_user(&frame->save_area, sizeof(frame->save_area));
+	err |= __put_user(0, &frame->uc.uc_flags);
+	err |= __put_user(NULL, &frame->uc.uc_link);
+	err |= __save_altstack(&frame->uc.uc_stack, regs->sp);
+	err |= setup_sigcontext(&frame->uc.uc_mcontext, regs);
+	err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
+	if (err)
+		goto give_sigsegv;
+
+	restorer = VDSO_SYM(&__vdso_rt_sigreturn);
+	if (ka->sa.sa_flags & SA_RESTORER)
+		restorer = (unsigned long) ka->sa.sa_restorer;
+
+	/*
+	 * Set up registers for signal handler.
+	 * Registers that we don't modify keep the value they had from
+	 * user-space at the time we took the signal.
+	 * We always pass siginfo and mcontext, regardless of SA_SIGINFO,
+	 * since some things rely on this (e.g. glibc's debug/segfault.c).
+	 */
+	regs->pc = (unsigned long) ka->sa.sa_handler;
+	regs->ex1 = PL_ICS_EX1(USER_PL, 1); /* set crit sec in handler */
+	regs->sp = (unsigned long) frame;
+	regs->lr = restorer;
+	regs->regs[0] = (unsigned long) usig;
+	regs->regs[1] = (unsigned long) &frame->info;
+	regs->regs[2] = (unsigned long) &frame->uc;
+	regs->flags |= PT_FLAGS_CALLER_SAVES;
+	return 0;
+
+give_sigsegv:
+	signal_fault("bad setup frame", regs, frame, sig);
+	return -EFAULT;
+}
+
+/*
+ * OK, we're invoking a handler
+ */
+
+static void handle_signal(unsigned long sig, siginfo_t *info,
+			 struct k_sigaction *ka,
+			 struct pt_regs *regs)
+{
+	sigset_t *oldset = sigmask_to_save();
+	int ret;
+
+	/* Are we from a system call? */
+	if (regs->faultnum == INT_SWINT_1) {
+		/* If so, check system call restarting.. */
+		switch (regs->regs[0]) {
+		case -ERESTART_RESTARTBLOCK:
+		case -ERESTARTNOHAND:
+			regs->regs[0] = -EINTR;
+			break;
+
+		case -ERESTARTSYS:
+			if (!(ka->sa.sa_flags & SA_RESTART)) {
+				regs->regs[0] = -EINTR;
+				break;
+			}
+			/* fallthrough */
+		case -ERESTARTNOINTR:
+			/* Reload caller-saves to restore r0..r5 and r10. */
+			regs->flags |= PT_FLAGS_CALLER_SAVES;
+			regs->regs[0] = regs->orig_r0;
+			regs->pc -= 8;
+		}
+	}
+
+	/* Set up the stack frame */
+#ifdef CONFIG_COMPAT
+	if (is_compat_task())
+		ret = compat_setup_rt_frame(sig, ka, info, oldset, regs);
+	else
+#endif
+		ret = setup_rt_frame(sig, ka, info, oldset, regs);
+	if (ret)
+		return;
+	signal_delivered(sig, info, ka, regs,
+			test_thread_flag(TIF_SINGLESTEP));
+}
+
+/*
+ * Note that 'init' is a special process: it doesn't get signals it doesn't
+ * want to handle. Thus you cannot kill init even with a SIGKILL even by
+ * mistake.
+ */
+void do_signal(struct pt_regs *regs)
+{
+	siginfo_t info;
+	int signr;
+	struct k_sigaction ka;
+
+	/*
+	 * i386 will check if we're coming from kernel mode and bail out
+	 * here.  In my experience this just turns weird crashes into
+	 * weird spin-hangs.  But if we find a case where this seems
+	 * helpful, we can reinstate the check on "!user_mode(regs)".
+	 */
+
+	signr = get_signal_to_deliver(&info, &ka, regs, NULL);
+	if (signr > 0) {
+		/* Whee! Actually deliver the signal.  */
+		handle_signal(signr, &info, &ka, regs);
+		goto done;
+	}
+
+	/* Did we come from a system call? */
+	if (regs->faultnum == INT_SWINT_1) {
+		/* Restart the system call - no handlers present */
+		switch (regs->regs[0]) {
+		case -ERESTARTNOHAND:
+		case -ERESTARTSYS:
+		case -ERESTARTNOINTR:
+			regs->flags |= PT_FLAGS_CALLER_SAVES;
+			regs->regs[0] = regs->orig_r0;
+			regs->pc -= 8;
+			break;
+
+		case -ERESTART_RESTARTBLOCK:
+			regs->flags |= PT_FLAGS_CALLER_SAVES;
+			regs->regs[TREG_SYSCALL_NR] = __NR_restart_syscall;
+			regs->pc -= 8;
+			break;
+		}
+	}
+
+	/* If there's no signal to deliver, just put the saved sigmask back. */
+	restore_saved_sigmask();
+
+done:
+	/* Avoid double syscall restart if there are nested signals. */
+	regs->faultnum = INT_SWINT_1_SIGRETURN;
+}
+
+int show_unhandled_signals = 1;
+
+static int __init crashinfo(char *str)
+{
+	unsigned long val;
+	const char *word;
+
+	if (*str == '\0')
+		val = 2;
+	else if (*str != '=' || strict_strtoul(++str, 0, &val) != 0)
+		return 0;
+	show_unhandled_signals = val;
+	switch (show_unhandled_signals) {
+	case 0:
+		word = "No";
+		break;
+	case 1:
+		word = "One-line";
+		break;
+	default:
+		word = "Detailed";
+		break;
+	}
+	pr_info("%s crash reports will be generated on the console\n", word);
+	return 1;
+}
+__setup("crashinfo", crashinfo);
+
+static void dump_mem(void __user *address)
+{
+	void __user *addr;
+	enum { region_size = 256, bytes_per_line = 16 };
+	int i, j, k;
+	int found_readable_mem = 0;
+
+	pr_err("\n");
+	if (!access_ok(VERIFY_READ, address, 1)) {
+		pr_err("Not dumping at address 0x%lx (kernel address)\n",
+		       (unsigned long)address);
+		return;
+	}
+
+	addr = (void __user *)
+		(((unsigned long)address & -bytes_per_line) - region_size/2);
+	if (addr > address)
+		addr = NULL;
+	for (i = 0; i < region_size;
+	     addr += bytes_per_line, i += bytes_per_line) {
+		unsigned char buf[bytes_per_line];
+		char line[100];
+		if (copy_from_user(buf, addr, bytes_per_line))
+			continue;
+		if (!found_readable_mem) {
+			pr_err("Dumping memory around address 0x%lx:\n",
+			       (unsigned long)address);
+			found_readable_mem = 1;
+		}
+		j = sprintf(line, REGFMT":", (unsigned long)addr);
+		for (k = 0; k < bytes_per_line; ++k)
+			j += sprintf(&line[j], " %02x", buf[k]);
+		pr_err("%s\n", line);
+	}
+	if (!found_readable_mem)
+		pr_err("No readable memory around address 0x%lx\n",
+		       (unsigned long)address);
+}
+
+void trace_unhandled_signal(const char *type, struct pt_regs *regs,
+			    unsigned long address, int sig)
+{
+	struct task_struct *tsk = current;
+
+	if (show_unhandled_signals == 0)
+		return;
+
+	/* If the signal is handled, don't show it here. */
+	if (!is_global_init(tsk)) {
+		void __user *handler =
+			tsk->sighand->action[sig-1].sa.sa_handler;
+		if (handler != SIG_IGN && handler != SIG_DFL)
+			return;
+	}
+
+	/* Rate-limit the one-line output, not the detailed output. */
+	if (show_unhandled_signals <= 1 && !printk_ratelimit())
+		return;
+
+	printk("%s%s[%d]: %s at %lx pc "REGFMT" signal %d",
+	       task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
+	       tsk->comm, task_pid_nr(tsk), type, address, regs->pc, sig);
+
+	print_vma_addr(KERN_CONT " in ", regs->pc);
+
+	printk(KERN_CONT "\n");
+
+	if (show_unhandled_signals > 1) {
+		switch (sig) {
+		case SIGILL:
+		case SIGFPE:
+		case SIGSEGV:
+		case SIGBUS:
+			pr_err("User crash: signal %d,"
+			       " trap %ld, address 0x%lx\n",
+			       sig, regs->faultnum, address);
+			show_regs(regs);
+			dump_mem((void __user *)address);
+			break;
+		default:
+			pr_err("User crash: signal %d, trap %ld\n",
+			       sig, regs->faultnum);
+			break;
+		}
+	}
+}

arch/tile/kernel/single_step.c

@@ -0,0 +1,776 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * A code-rewriter that enables instruction single-stepping.
+ * Derived from iLib's single-stepping code.
+ */
+
+#ifndef __tilegx__   /* Hardware support for single step unavailable. */
+
+/* These functions are only used on the TILE platform */
+#include <linux/slab.h>
+#include <linux/thread_info.h>
+#include <linux/uaccess.h>
+#include <linux/mman.h>
+#include <linux/types.h>
+#include <linux/err.h>
+#include <asm/cacheflush.h>
+#include <asm/unaligned.h>
+#include <arch/abi.h>
+#include <arch/opcode.h>
+
+#define signExtend17(val) sign_extend((val), 17)
+#define TILE_X1_MASK (0xffffffffULL << 31)
+
+int unaligned_printk;
+
+static int __init setup_unaligned_printk(char *str)
+{
+	long val;
+	if (strict_strtol(str, 0, &val) != 0)
+		return 0;
+	unaligned_printk = val;
+	pr_info("Printk for each unaligned data accesses is %s\n",
+		unaligned_printk ? "enabled" : "disabled");
+	return 1;
+}
+__setup("unaligned_printk=", setup_unaligned_printk);
+
+unsigned int unaligned_fixup_count;
+
+enum mem_op {
+	MEMOP_NONE,
+	MEMOP_LOAD,
+	MEMOP_STORE,
+	MEMOP_LOAD_POSTINCR,
+	MEMOP_STORE_POSTINCR
+};
+
+static inline tile_bundle_bits set_BrOff_X1(tile_bundle_bits n, s32 offset)
+{
+	tile_bundle_bits result;
+
+	/* mask out the old offset */
+	tile_bundle_bits mask = create_BrOff_X1(-1);
+	result = n & (~mask);
+
+	/* or in the new offset */
+	result |= create_BrOff_X1(offset);
+
+	return result;
+}
+
+static inline tile_bundle_bits move_X1(tile_bundle_bits n, int dest, int src)
+{
+	tile_bundle_bits result;
+	tile_bundle_bits op;
+
+	result = n & (~TILE_X1_MASK);
+
+	op = create_Opcode_X1(SPECIAL_0_OPCODE_X1) |
+		create_RRROpcodeExtension_X1(OR_SPECIAL_0_OPCODE_X1) |
+		create_Dest_X1(dest) |
+		create_SrcB_X1(TREG_ZERO) |
+		create_SrcA_X1(src) ;
+
+	result |= op;
+	return result;
+}
+
+static inline tile_bundle_bits nop_X1(tile_bundle_bits n)
+{
+	return move_X1(n, TREG_ZERO, TREG_ZERO);
+}
+
+static inline tile_bundle_bits addi_X1(
+	tile_bundle_bits n, int dest, int src, int imm)
+{
+	n &= ~TILE_X1_MASK;
+
+	n |=  (create_SrcA_X1(src) |
+	       create_Dest_X1(dest) |
+	       create_Imm8_X1(imm) |
+	       create_S_X1(0) |
+	       create_Opcode_X1(IMM_0_OPCODE_X1) |
+	       create_ImmOpcodeExtension_X1(ADDI_IMM_0_OPCODE_X1));
+
+	return n;
+}
+
+static tile_bundle_bits rewrite_load_store_unaligned(
+	struct single_step_state *state,
+	tile_bundle_bits bundle,
+	struct pt_regs *regs,
+	enum mem_op mem_op,
+	int size, int sign_ext)
+{
+	unsigned char __user *addr;
+	int val_reg, addr_reg, err, val;
+
+	/* Get address and value registers */
+	if (bundle & TILEPRO_BUNDLE_Y_ENCODING_MASK) {
+		addr_reg = get_SrcA_Y2(bundle);
+		val_reg = get_SrcBDest_Y2(bundle);
+	} else if (mem_op == MEMOP_LOAD || mem_op == MEMOP_LOAD_POSTINCR) {
+		addr_reg = get_SrcA_X1(bundle);
+		val_reg  = get_Dest_X1(bundle);
+	} else {
+		addr_reg = get_SrcA_X1(bundle);
+		val_reg  = get_SrcB_X1(bundle);
+	}
+
+	/*
+	 * If registers are not GPRs, don't try to handle it.
+	 *
+	 * FIXME: we could handle non-GPR loads by getting the real value
+	 * from memory, writing it to the single step buffer, using a
+	 * temp_reg to hold a pointer to that memory, then executing that
+	 * instruction and resetting temp_reg.  For non-GPR stores, it's a
+	 * little trickier; we could use the single step buffer for that
+	 * too, but we'd have to add some more state bits so that we could
+	 * call back in here to copy that value to the real target.  For
+	 * now, we just handle the simple case.
+	 */
+	if ((val_reg >= PTREGS_NR_GPRS &&
+	     (val_reg != TREG_ZERO ||
+	      mem_op == MEMOP_LOAD ||
+	      mem_op == MEMOP_LOAD_POSTINCR)) ||
+	    addr_reg >= PTREGS_NR_GPRS)
+		return bundle;
+
+	/* If it's aligned, don't handle it specially */
+	addr = (void __user *)regs->regs[addr_reg];
+	if (((unsigned long)addr % size) == 0)
+		return bundle;
+
+	/*
+	 * Return SIGBUS with the unaligned address, if requested.
+	 * Note that we return SIGBUS even for completely invalid addresses
+	 * as long as they are in fact unaligned; this matches what the
+	 * tilepro hardware would be doing, if it could provide us with the
+	 * actual bad address in an SPR, which it doesn't.
+	 */
+	if (unaligned_fixup == 0) {
+		siginfo_t info = {
+			.si_signo = SIGBUS,
+			.si_code = BUS_ADRALN,
+			.si_addr = addr
+		};
+		trace_unhandled_signal("unaligned trap", regs,
+				       (unsigned long)addr, SIGBUS);
+		force_sig_info(info.si_signo, &info, current);
+		return (tilepro_bundle_bits) 0;
+	}
+
+	/* Handle unaligned load/store */
+	if (mem_op == MEMOP_LOAD || mem_op == MEMOP_LOAD_POSTINCR) {
+		unsigned short val_16;
+		switch (size) {
+		case 2:
+			err = copy_from_user(&val_16, addr, sizeof(val_16));
+			val = sign_ext ? ((short)val_16) : val_16;
+			break;
+		case 4:
+			err = copy_from_user(&val, addr, sizeof(val));
+			break;
+		default:
+			BUG();
+		}
+		if (err == 0) {
+			state->update_reg = val_reg;
+			state->update_value = val;
+			state->update = 1;
+		}
+	} else {
+		unsigned short val_16;
+		val = (val_reg == TREG_ZERO) ? 0 : regs->regs[val_reg];
+		switch (size) {
+		case 2:
+			val_16 = val;
+			err = copy_to_user(addr, &val_16, sizeof(val_16));
+			break;
+		case 4:
+			err = copy_to_user(addr, &val, sizeof(val));
+			break;
+		default:
+			BUG();
+		}
+	}
+
+	if (err) {
+		siginfo_t info = {
+			.si_signo = SIGSEGV,
+			.si_code = SEGV_MAPERR,
+			.si_addr = addr
+		};
+		trace_unhandled_signal("segfault", regs,
+				       (unsigned long)addr, SIGSEGV);
+		force_sig_info(info.si_signo, &info, current);
+		return (tile_bundle_bits) 0;
+	}
+
+	if (unaligned_printk || unaligned_fixup_count == 0) {
+		pr_info("Process %d/%s: PC %#lx: Fixup of"
+			" unaligned %s at %#lx.\n",
+			current->pid, current->comm, regs->pc,
+			(mem_op == MEMOP_LOAD ||
+			 mem_op == MEMOP_LOAD_POSTINCR) ?
+			"load" : "store",
+			(unsigned long)addr);
+		if (!unaligned_printk) {
+#define P pr_info
+P("\n");
+P("Unaligned fixups in the kernel will slow your application considerably.\n");
+P("To find them, write a \"1\" to /proc/sys/tile/unaligned_fixup/printk,\n");
+P("which requests the kernel show all unaligned fixups, or write a \"0\"\n");
+P("to /proc/sys/tile/unaligned_fixup/enabled, in which case each unaligned\n");
+P("access will become a SIGBUS you can debug. No further warnings will be\n");
+P("shown so as to avoid additional slowdown, but you can track the number\n");
+P("of fixups performed via /proc/sys/tile/unaligned_fixup/count.\n");
+P("Use the tile-addr2line command (see \"info addr2line\") to decode PCs.\n");
+P("\n");
+#undef P
+		}
+	}
+	++unaligned_fixup_count;
+
+	if (bundle & TILEPRO_BUNDLE_Y_ENCODING_MASK) {
+		/* Convert the Y2 instruction to a prefetch. */
+		bundle &= ~(create_SrcBDest_Y2(-1) |
+			    create_Opcode_Y2(-1));
+		bundle |= (create_SrcBDest_Y2(TREG_ZERO) |
+			   create_Opcode_Y2(LW_OPCODE_Y2));
+	/* Replace the load postincr with an addi */
+	} else if (mem_op == MEMOP_LOAD_POSTINCR) {
+		bundle = addi_X1(bundle, addr_reg, addr_reg,
+				 get_Imm8_X1(bundle));
+	/* Replace the store postincr with an addi */
+	} else if (mem_op == MEMOP_STORE_POSTINCR) {
+		bundle = addi_X1(bundle, addr_reg, addr_reg,
+				 get_Dest_Imm8_X1(bundle));
+	} else {
+		/* Convert the X1 instruction to a nop. */
+		bundle &= ~(create_Opcode_X1(-1) |
+			    create_UnShOpcodeExtension_X1(-1) |
+			    create_UnOpcodeExtension_X1(-1));
+		bundle |= (create_Opcode_X1(SHUN_0_OPCODE_X1) |
+			   create_UnShOpcodeExtension_X1(
+				   UN_0_SHUN_0_OPCODE_X1) |
+			   create_UnOpcodeExtension_X1(
+				   NOP_UN_0_SHUN_0_OPCODE_X1));
+	}
+
+	return bundle;
+}
+
+/*
+ * Called after execve() has started the new image.  This allows us
+ * to reset the info state.  Note that the the mmap'ed memory, if there
+ * was any, has already been unmapped by the exec.
+ */
+void single_step_execve(void)
+{
+	struct thread_info *ti = current_thread_info();
+	kfree(ti->step_state);
+	ti->step_state = NULL;
+}
+
+/**
+ * single_step_once() - entry point when single stepping has been triggered.
+ * @regs: The machine register state
+ *
+ *  When we arrive at this routine via a trampoline, the single step
+ *  engine copies the executing bundle to the single step buffer.
+ *  If the instruction is a condition branch, then the target is
+ *  reset to one past the next instruction. If the instruction
+ *  sets the lr, then that is noted. If the instruction is a jump
+ *  or call, then the new target pc is preserved and the current
+ *  bundle instruction set to null.
+ *
+ *  The necessary post-single-step rewriting information is stored in
+ *  single_step_state->  We use data segment values because the
+ *  stack will be rewound when we run the rewritten single-stepped
+ *  instruction.
+ */
+void single_step_once(struct pt_regs *regs)
+{
+	extern tile_bundle_bits __single_step_ill_insn;
+	extern tile_bundle_bits __single_step_j_insn;
+	extern tile_bundle_bits __single_step_addli_insn;
+	extern tile_bundle_bits __single_step_auli_insn;
+	struct thread_info *info = (void *)current_thread_info();
+	struct single_step_state *state = info->step_state;
+	int is_single_step = test_ti_thread_flag(info, TIF_SINGLESTEP);
+	tile_bundle_bits __user *buffer, *pc;
+	tile_bundle_bits bundle;
+	int temp_reg;
+	int target_reg = TREG_LR;
+	int err;
+	enum mem_op mem_op = MEMOP_NONE;
+	int size = 0, sign_ext = 0;  /* happy compiler */
+
+	asm(
+"    .pushsection .rodata.single_step\n"
+"    .align 8\n"
+"    .globl    __single_step_ill_insn\n"
+"__single_step_ill_insn:\n"
+"    ill\n"
+"    .globl    __single_step_addli_insn\n"
+"__single_step_addli_insn:\n"
+"    { nop; addli r0, zero, 0 }\n"
+"    .globl    __single_step_auli_insn\n"
+"__single_step_auli_insn:\n"
+"    { nop; auli r0, r0, 0 }\n"
+"    .globl    __single_step_j_insn\n"
+"__single_step_j_insn:\n"
+"    j .\n"
+"    .popsection\n"
+	);
+
+	/*
+	 * Enable interrupts here to allow touching userspace and the like.
+	 * The callers expect this: do_trap() already has interrupts
+	 * enabled, and do_work_pending() handles functions that enable
+	 * interrupts internally.
+	 */
+	local_irq_enable();
+
+	if (state == NULL) {
+		/* allocate a page of writable, executable memory */
+		state = kmalloc(sizeof(struct single_step_state), GFP_KERNEL);
+		if (state == NULL) {
+			pr_err("Out of kernel memory trying to single-step\n");
+			return;
+		}
+
+		/* allocate a cache line of writable, executable memory */
+		buffer = (void __user *) vm_mmap(NULL, 0, 64,
+					  PROT_EXEC | PROT_READ | PROT_WRITE,
+					  MAP_PRIVATE | MAP_ANONYMOUS,
+					  0);
+
+		if (IS_ERR((void __force *)buffer)) {
+			kfree(state);
+			pr_err("Out of kernel pages trying to single-step\n");
+			return;
+		}
+
+		state->buffer = buffer;
+		state->is_enabled = 0;
+
+		info->step_state = state;
+
+		/* Validate our stored instruction patterns */
+		BUG_ON(get_Opcode_X1(__single_step_addli_insn) !=
+		       ADDLI_OPCODE_X1);
+		BUG_ON(get_Opcode_X1(__single_step_auli_insn) !=
+		       AULI_OPCODE_X1);
+		BUG_ON(get_SrcA_X1(__single_step_addli_insn) != TREG_ZERO);
+		BUG_ON(get_Dest_X1(__single_step_addli_insn) != 0);
+		BUG_ON(get_JOffLong_X1(__single_step_j_insn) != 0);
+	}
+
+	/*
+	 * If we are returning from a syscall, we still haven't hit the
+	 * "ill" for the swint1 instruction.  So back the PC up to be
+	 * pointing at the swint1, but we'll actually return directly
+	 * back to the "ill" so we come back in via SIGILL as if we
+	 * had "executed" the swint1 without ever being in kernel space.
+	 */
+	if (regs->faultnum == INT_SWINT_1)
+		regs->pc -= 8;
+
+	pc = (tile_bundle_bits __user *)(regs->pc);
+	if (get_user(bundle, pc) != 0) {
+		pr_err("Couldn't read instruction at %p trying to step\n", pc);
+		return;
+	}
+
+	/* We'll follow the instruction with 2 ill op bundles */
+	state->orig_pc = (unsigned long)pc;
+	state->next_pc = (unsigned long)(pc + 1);
+	state->branch_next_pc = 0;
+	state->update = 0;
+
+	if (!(bundle & TILEPRO_BUNDLE_Y_ENCODING_MASK)) {
+		/* two wide, check for control flow */
+		int opcode = get_Opcode_X1(bundle);
+
+		switch (opcode) {
+		/* branches */
+		case BRANCH_OPCODE_X1:
+		{
+			s32 offset = signExtend17(get_BrOff_X1(bundle));
+
+			/*
+			 * For branches, we use a rewriting trick to let the
+			 * hardware evaluate whether the branch is taken or
+			 * untaken.  We record the target offset and then
+			 * rewrite the branch instruction to target 1 insn
+			 * ahead if the branch is taken.  We then follow the
+			 * rewritten branch with two bundles, each containing
+			 * an "ill" instruction. The supervisor examines the
+			 * pc after the single step code is executed, and if
+			 * the pc is the first ill instruction, then the
+			 * branch (if any) was not taken.  If the pc is the
+			 * second ill instruction, then the branch was
+			 * taken. The new pc is computed for these cases, and
+			 * inserted into the registers for the thread.  If
+			 * the pc is the start of the single step code, then
+			 * an exception or interrupt was taken before the
+			 * code started processing, and the same "original"
+			 * pc is restored.  This change, different from the
+			 * original implementation, has the advantage of
+			 * executing a single user instruction.
+			 */
+			state->branch_next_pc = (unsigned long)(pc + offset);
+
+			/* rewrite branch offset to go forward one bundle */
+			bundle = set_BrOff_X1(bundle, 2);
+		}
+		break;
+
+		/* jumps */
+		case JALB_OPCODE_X1:
+		case JALF_OPCODE_X1:
+			state->update = 1;
+			state->next_pc =
+				(unsigned long) (pc + get_JOffLong_X1(bundle));
+			break;
+
+		case JB_OPCODE_X1:
+		case JF_OPCODE_X1:
+			state->next_pc =
+				(unsigned long) (pc + get_JOffLong_X1(bundle));
+			bundle = nop_X1(bundle);
+			break;
+
+		case SPECIAL_0_OPCODE_X1:
+			switch (get_RRROpcodeExtension_X1(bundle)) {
+			/* jump-register */
+			case JALRP_SPECIAL_0_OPCODE_X1:
+			case JALR_SPECIAL_0_OPCODE_X1:
+				state->update = 1;
+				state->next_pc =
+					regs->regs[get_SrcA_X1(bundle)];
+				break;
+
+			case JRP_SPECIAL_0_OPCODE_X1:
+			case JR_SPECIAL_0_OPCODE_X1:
+				state->next_pc =
+					regs->regs[get_SrcA_X1(bundle)];
+				bundle = nop_X1(bundle);
+				break;
+
+			case LNK_SPECIAL_0_OPCODE_X1:
+				state->update = 1;
+				target_reg = get_Dest_X1(bundle);
+				break;
+
+			/* stores */
+			case SH_SPECIAL_0_OPCODE_X1:
+				mem_op = MEMOP_STORE;
+				size = 2;
+				break;
+
+			case SW_SPECIAL_0_OPCODE_X1:
+				mem_op = MEMOP_STORE;
+				size = 4;
+				break;
+			}
+			break;
+
+		/* loads and iret */
+		case SHUN_0_OPCODE_X1:
+			if (get_UnShOpcodeExtension_X1(bundle) ==
+			    UN_0_SHUN_0_OPCODE_X1) {
+				switch (get_UnOpcodeExtension_X1(bundle)) {
+				case LH_UN_0_SHUN_0_OPCODE_X1:
+					mem_op = MEMOP_LOAD;
+					size = 2;
+					sign_ext = 1;
+					break;
+
+				case LH_U_UN_0_SHUN_0_OPCODE_X1:
+					mem_op = MEMOP_LOAD;
+					size = 2;
+					sign_ext = 0;
+					break;
+
+				case LW_UN_0_SHUN_0_OPCODE_X1:
+					mem_op = MEMOP_LOAD;
+					size = 4;
+					break;
+
+				case IRET_UN_0_SHUN_0_OPCODE_X1:
+				{
+					unsigned long ex0_0 = __insn_mfspr(
+						SPR_EX_CONTEXT_0_0);
+					unsigned long ex0_1 = __insn_mfspr(
+						SPR_EX_CONTEXT_0_1);
+					/*
+					 * Special-case it if we're iret'ing
+					 * to PL0 again.  Otherwise just let
+					 * it run and it will generate SIGILL.
+					 */
+					if (EX1_PL(ex0_1) == USER_PL) {
+						state->next_pc = ex0_0;
+						regs->ex1 = ex0_1;
+						bundle = nop_X1(bundle);
+					}
+				}
+				}
+			}
+			break;
+
+#if CHIP_HAS_WH64()
+		/* postincrement operations */
+		case IMM_0_OPCODE_X1:
+			switch (get_ImmOpcodeExtension_X1(bundle)) {
+			case LWADD_IMM_0_OPCODE_X1:
+				mem_op = MEMOP_LOAD_POSTINCR;
+				size = 4;
+				break;
+
+			case LHADD_IMM_0_OPCODE_X1:
+				mem_op = MEMOP_LOAD_POSTINCR;
+				size = 2;
+				sign_ext = 1;
+				break;
+
+			case LHADD_U_IMM_0_OPCODE_X1:
+				mem_op = MEMOP_LOAD_POSTINCR;
+				size = 2;
+				sign_ext = 0;
+				break;
+
+			case SWADD_IMM_0_OPCODE_X1:
+				mem_op = MEMOP_STORE_POSTINCR;
+				size = 4;
+				break;
+
+			case SHADD_IMM_0_OPCODE_X1:
+				mem_op = MEMOP_STORE_POSTINCR;
+				size = 2;
+				break;
+
+			default:
+				break;
+			}
+			break;
+#endif /* CHIP_HAS_WH64() */
+		}
+
+		if (state->update) {
+			/*
+			 * Get an available register.  We start with a
+			 * bitmask with 1's for available registers.
+			 * We truncate to the low 32 registers since
+			 * we are guaranteed to have set bits in the
+			 * low 32 bits, then use ctz to pick the first.
+			 */
+			u32 mask = (u32) ~((1ULL << get_Dest_X0(bundle)) |
+					   (1ULL << get_SrcA_X0(bundle)) |
+					   (1ULL << get_SrcB_X0(bundle)) |
+					   (1ULL << target_reg));
+			temp_reg = __builtin_ctz(mask);
+			state->update_reg = temp_reg;
+			state->update_value = regs->regs[temp_reg];
+			regs->regs[temp_reg] = (unsigned long) (pc+1);
+			regs->flags |= PT_FLAGS_RESTORE_REGS;
+			bundle = move_X1(bundle, target_reg, temp_reg);
+		}
+	} else {
+		int opcode = get_Opcode_Y2(bundle);
+
+		switch (opcode) {
+		/* loads */
+		case LH_OPCODE_Y2:
+			mem_op = MEMOP_LOAD;
+			size = 2;
+			sign_ext = 1;
+			break;
+
+		case LH_U_OPCODE_Y2:
+			mem_op = MEMOP_LOAD;
+			size = 2;
+			sign_ext = 0;
+			break;
+
+		case LW_OPCODE_Y2:
+			mem_op = MEMOP_LOAD;
+			size = 4;
+			break;
+
+		/* stores */
+		case SH_OPCODE_Y2:
+			mem_op = MEMOP_STORE;
+			size = 2;
+			break;
+
+		case SW_OPCODE_Y2:
+			mem_op = MEMOP_STORE;
+			size = 4;
+			break;
+		}
+	}
+
+	/*
+	 * Check if we need to rewrite an unaligned load/store.
+	 * Returning zero is a special value meaning we need to SIGSEGV.
+	 */
+	if (mem_op != MEMOP_NONE && unaligned_fixup >= 0) {
+		bundle = rewrite_load_store_unaligned(state, bundle, regs,
+						      mem_op, size, sign_ext);
+		if (bundle == 0)
+			return;
+	}
+
+	/* write the bundle to our execution area */
+	buffer = state->buffer;
+	err = __put_user(bundle, buffer++);
+
+	/*
+	 * If we're really single-stepping, we take an INT_ILL after.
+	 * If we're just handling an unaligned access, we can just
+	 * jump directly back to where we were in user code.
+	 */
+	if (is_single_step) {
+		err |= __put_user(__single_step_ill_insn, buffer++);
+		err |= __put_user(__single_step_ill_insn, buffer++);
+	} else {
+		long delta;
+
+		if (state->update) {
+			/* We have some state to update; do it inline */
+			int ha16;
+			bundle = __single_step_addli_insn;
+			bundle |= create_Dest_X1(state->update_reg);
+			bundle |= create_Imm16_X1(state->update_value);
+			err |= __put_user(bundle, buffer++);
+			bundle = __single_step_auli_insn;
+			bundle |= create_Dest_X1(state->update_reg);
+			bundle |= create_SrcA_X1(state->update_reg);
+			ha16 = (state->update_value + 0x8000) >> 16;
+			bundle |= create_Imm16_X1(ha16);
+			err |= __put_user(bundle, buffer++);
+			state->update = 0;
+		}
+
+		/* End with a jump back to the next instruction */
+		delta = ((regs->pc + TILE_BUNDLE_SIZE_IN_BYTES) -
+			(unsigned long)buffer) >>
+			TILE_LOG2_BUNDLE_ALIGNMENT_IN_BYTES;
+		bundle = __single_step_j_insn;
+		bundle |= create_JOffLong_X1(delta);
+		err |= __put_user(bundle, buffer++);
+	}
+
+	if (err) {
+		pr_err("Fault when writing to single-step buffer\n");
+		return;
+	}
+
+	/*
+	 * Flush the buffer.
+	 * We do a local flush only, since this is a thread-specific buffer.
+	 */
+	__flush_icache_range((unsigned long)state->buffer,
+			     (unsigned long)buffer);
+
+	/* Indicate enabled */
+	state->is_enabled = is_single_step;
+	regs->pc = (unsigned long)state->buffer;
+
+	/* Fault immediately if we are coming back from a syscall. */
+	if (regs->faultnum == INT_SWINT_1)
+		regs->pc += 8;
+}
+
+#else
+#include <linux/smp.h>
+#include <linux/ptrace.h>
+#include <arch/spr_def.h>
+
+static DEFINE_PER_CPU(unsigned long, ss_saved_pc);
+
+
+/*
+ * Called directly on the occasion of an interrupt.
+ *
+ * If the process doesn't have single step set, then we use this as an
+ * opportunity to turn single step off.
+ *
+ * It has been mentioned that we could conditionally turn off single stepping
+ * on each entry into the kernel and rely on single_step_once to turn it
+ * on for the processes that matter (as we already do), but this
+ * implementation is somewhat more efficient in that we muck with registers
+ * once on a bum interrupt rather than on every entry into the kernel.
+ *
+ * If SINGLE_STEP_CONTROL_K has CANCELED set, then an interrupt occurred,
+ * so we have to run through this process again before we can say that an
+ * instruction has executed.
+ *
+ * swint will set CANCELED, but it's a legitimate instruction.  Fortunately
+ * it changes the PC.  If it hasn't changed, then we know that the interrupt
+ * wasn't generated by swint and we'll need to run this process again before
+ * we can say an instruction has executed.
+ *
+ * If either CANCELED == 0 or the PC's changed, we send out SIGTRAPs and get
+ * on with our lives.
+ */
+
+void gx_singlestep_handle(struct pt_regs *regs, int fault_num)
+{
+	unsigned long *ss_pc = &__get_cpu_var(ss_saved_pc);
+	struct thread_info *info = (void *)current_thread_info();
+	int is_single_step = test_ti_thread_flag(info, TIF_SINGLESTEP);
+	unsigned long control = __insn_mfspr(SPR_SINGLE_STEP_CONTROL_K);
+
+	if (is_single_step == 0) {
+		__insn_mtspr(SPR_SINGLE_STEP_EN_K_K, 0);
+
+	} else if ((*ss_pc != regs->pc) ||
+		   (!(control & SPR_SINGLE_STEP_CONTROL_1__CANCELED_MASK))) {
+
+		ptrace_notify(SIGTRAP);
+		control |= SPR_SINGLE_STEP_CONTROL_1__CANCELED_MASK;
+		control |= SPR_SINGLE_STEP_CONTROL_1__INHIBIT_MASK;
+		__insn_mtspr(SPR_SINGLE_STEP_CONTROL_K, control);
+	}
+}
+
+
+/*
+ * Called from need_singlestep.  Set up the control registers and the enable
+ * register, then return back.
+ */
+
+void single_step_once(struct pt_regs *regs)
+{
+	unsigned long *ss_pc = &__get_cpu_var(ss_saved_pc);
+	unsigned long control = __insn_mfspr(SPR_SINGLE_STEP_CONTROL_K);
+
+	*ss_pc = regs->pc;
+	control |= SPR_SINGLE_STEP_CONTROL_1__CANCELED_MASK;
+	control |= SPR_SINGLE_STEP_CONTROL_1__INHIBIT_MASK;
+	__insn_mtspr(SPR_SINGLE_STEP_CONTROL_K, control);
+	__insn_mtspr(SPR_SINGLE_STEP_EN_K_K, 1 << USER_PL);
+}
+
+void single_step_execve(void)
+{
+	/* Nothing */
+}
+
+#endif /* !__tilegx__ */

arch/tile/kernel/smp.c

@@ -0,0 +1,244 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * TILE SMP support routines.
+ */
+
+#include <linux/smp.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/irq.h>
+#include <linux/module.h>
+#include <asm/cacheflush.h>
+
+HV_Topology smp_topology __write_once;
+EXPORT_SYMBOL(smp_topology);
+
+#if CHIP_HAS_IPI()
+static unsigned long __iomem *ipi_mappings[NR_CPUS];
+#endif
+
+
+/*
+ * Top-level send_IPI*() functions to send messages to other cpus.
+ */
+
+/* Set by smp_send_stop() to avoid recursive panics. */
+static int stopping_cpus;
+
+static void __send_IPI_many(HV_Recipient *recip, int nrecip, int tag)
+{
+	int sent = 0;
+	while (sent < nrecip) {
+		int rc = hv_send_message(recip, nrecip,
+					 (HV_VirtAddr)&tag, sizeof(tag));
+		if (rc < 0) {
+			if (!stopping_cpus)  /* avoid recursive panic */
+				panic("hv_send_message returned %d", rc);
+			break;
+		}
+		WARN_ONCE(rc == 0, "hv_send_message() returned zero\n");
+		sent += rc;
+	}
+}
+
+void send_IPI_single(int cpu, int tag)
+{
+	HV_Recipient recip = {
+		.y = cpu / smp_width,
+		.x = cpu % smp_width,
+		.state = HV_TO_BE_SENT
+	};
+	__send_IPI_many(&recip, 1, tag);
+}
+
+void send_IPI_many(const struct cpumask *mask, int tag)
+{
+	HV_Recipient recip[NR_CPUS];
+	int cpu;
+	int nrecip = 0;
+	int my_cpu = smp_processor_id();
+	for_each_cpu(cpu, mask) {
+		HV_Recipient *r;
+		BUG_ON(cpu == my_cpu);
+		r = &recip[nrecip++];
+		r->y = cpu / smp_width;
+		r->x = cpu % smp_width;
+		r->state = HV_TO_BE_SENT;
+	}
+	__send_IPI_many(recip, nrecip, tag);
+}
+
+void send_IPI_allbutself(int tag)
+{
+	struct cpumask mask;
+	cpumask_copy(&mask, cpu_online_mask);
+	cpumask_clear_cpu(smp_processor_id(), &mask);
+	send_IPI_many(&mask, tag);
+}
+
+/*
+ * Functions related to starting/stopping cpus.
+ */
+
+/* Handler to start the current cpu. */
+static void smp_start_cpu_interrupt(void)
+{
+	get_irq_regs()->pc = start_cpu_function_addr;
+}
+
+/* Handler to stop the current cpu. */
+static void smp_stop_cpu_interrupt(void)
+{
+	set_cpu_online(smp_processor_id(), 0);
+	arch_local_irq_disable_all();
+	for (;;)
+		asm("nap; nop");
+}
+
+/* This function calls the 'stop' function on all other CPUs in the system. */
+void smp_send_stop(void)
+{
+	stopping_cpus = 1;
+	send_IPI_allbutself(MSG_TAG_STOP_CPU);
+}
+
+/* On panic, just wait; we may get an smp_send_stop() later on. */
+void panic_smp_self_stop(void)
+{
+	while (1)
+		asm("nap; nop");
+}
+
+/*
+ * Dispatch code called from hv_message_intr() for HV_MSG_TILE hv messages.
+ */
+void evaluate_message(int tag)
+{
+	switch (tag) {
+	case MSG_TAG_START_CPU: /* Start up a cpu */
+		smp_start_cpu_interrupt();
+		break;
+
+	case MSG_TAG_STOP_CPU: /* Sent to shut down slave CPU's */
+		smp_stop_cpu_interrupt();
+		break;
+
+	case MSG_TAG_CALL_FUNCTION_MANY: /* Call function on cpumask */
+		generic_smp_call_function_interrupt();
+		break;
+
+	case MSG_TAG_CALL_FUNCTION_SINGLE: /* Call function on one other CPU */
+		generic_smp_call_function_single_interrupt();
+		break;
+
+	default:
+		panic("Unknown IPI message tag %d", tag);
+		break;
+	}
+}
+
+
+/*
+ * flush_icache_range() code uses smp_call_function().
+ */
+
+struct ipi_flush {
+	unsigned long start;
+	unsigned long end;
+};
+
+static void ipi_flush_icache_range(void *info)
+{
+	struct ipi_flush *flush = (struct ipi_flush *) info;
+	__flush_icache_range(flush->start, flush->end);
+}
+
+void flush_icache_range(unsigned long start, unsigned long end)
+{
+	struct ipi_flush flush = { start, end };
+	preempt_disable();
+	on_each_cpu(ipi_flush_icache_range, &flush, 1);
+	preempt_enable();
+}
+
+
+/* Called when smp_send_reschedule() triggers IRQ_RESCHEDULE. */
+static irqreturn_t handle_reschedule_ipi(int irq, void *token)
+{
+	__get_cpu_var(irq_stat).irq_resched_count++;
+	scheduler_ipi();
+
+	return IRQ_HANDLED;
+}
+
+static struct irqaction resched_action = {
+	.handler = handle_reschedule_ipi,
+	.name = "resched",
+	.dev_id = handle_reschedule_ipi /* unique token */,
+};
+
+void __init ipi_init(void)
+{
+#if CHIP_HAS_IPI()
+	int cpu;
+	/* Map IPI trigger MMIO addresses. */
+	for_each_possible_cpu(cpu) {
+		HV_Coord tile;
+		HV_PTE pte;
+		unsigned long offset;
+
+		tile.x = cpu_x(cpu);
+		tile.y = cpu_y(cpu);
+		if (hv_get_ipi_pte(tile, KERNEL_PL, &pte) != 0)
+			panic("Failed to initialize IPI for cpu %d\n", cpu);
+
+		offset = PFN_PHYS(pte_pfn(pte));
+		ipi_mappings[cpu] = ioremap_prot(offset, PAGE_SIZE, pte);
+	}
+#endif
+
+	/* Bind handle_reschedule_ipi() to IRQ_RESCHEDULE. */
+	tile_irq_activate(IRQ_RESCHEDULE, TILE_IRQ_PERCPU);
+	BUG_ON(setup_irq(IRQ_RESCHEDULE, &resched_action));
+}
+
+#if CHIP_HAS_IPI()
+
+void smp_send_reschedule(int cpu)
+{
+	WARN_ON(cpu_is_offline(cpu));
+
+	/*
+	 * We just want to do an MMIO store.  The traditional writeq()
+	 * functions aren't really correct here, since they're always
+	 * directed at the PCI shim.  For now, just do a raw store,
+	 * casting away the __iomem attribute.
+	 */
+	((unsigned long __force *)ipi_mappings[cpu])[IRQ_RESCHEDULE] = 0;
+}
+
+#else
+
+void smp_send_reschedule(int cpu)
+{
+	HV_Coord coord;
+
+	WARN_ON(cpu_is_offline(cpu));
+
+	coord.y = cpu_y(cpu);
+	coord.x = cpu_x(cpu);
+	hv_trigger_ipi(coord, IRQ_RESCHEDULE);
+}
+
+#endif /* CHIP_HAS_IPI() */

arch/tile/kernel/smpboot.c

@@ -0,0 +1,267 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/sched.h>
+#include <linux/kernel_stat.h>
+#include <linux/bootmem.h>
+#include <linux/notifier.h>
+#include <linux/cpu.h>
+#include <linux/percpu.h>
+#include <linux/delay.h>
+#include <linux/err.h>
+#include <linux/irq.h>
+#include <asm/mmu_context.h>
+#include <asm/tlbflush.h>
+#include <asm/sections.h>
+
+/* State of each CPU. */
+static DEFINE_PER_CPU(int, cpu_state) = { 0 };
+
+/* The messaging code jumps to this pointer during boot-up */
+unsigned long start_cpu_function_addr;
+
+/* Called very early during startup to mark boot cpu as online */
+void __init smp_prepare_boot_cpu(void)
+{
+	int cpu = smp_processor_id();
+	set_cpu_online(cpu, 1);
+	set_cpu_present(cpu, 1);
+	__get_cpu_var(cpu_state) = CPU_ONLINE;
+
+	init_messaging();
+}
+
+static void start_secondary(void);
+
+/*
+ * Called at the top of init() to launch all the other CPUs.
+ * They run free to complete their initialization and then wait
+ * until they get an IPI from the boot cpu to come online.
+ */
+void __init smp_prepare_cpus(unsigned int max_cpus)
+{
+	long rc;
+	int cpu, cpu_count;
+	int boot_cpu = smp_processor_id();
+
+	current_thread_info()->cpu = boot_cpu;
+
+	/*
+	 * Pin this task to the boot CPU while we bring up the others,
+	 * just to make sure we don't uselessly migrate as they come up.
+	 */
+	rc = sched_setaffinity(current->pid, cpumask_of(boot_cpu));
+	if (rc != 0)
+		pr_err("Couldn't set init affinity to boot cpu (%ld)\n", rc);
+
+	/* Print information about disabled and dataplane cpus. */
+	print_disabled_cpus();
+
+	/*
+	 * Tell the messaging subsystem how to respond to the
+	 * startup message.  We use a level of indirection to avoid
+	 * confusing the linker with the fact that the messaging
+	 * subsystem is calling __init code.
+	 */
+	start_cpu_function_addr = (unsigned long) &online_secondary;
+
+	/* Set up thread context for all new processors. */
+	cpu_count = 1;
+	for (cpu = 0; cpu < NR_CPUS; ++cpu)	{
+		struct task_struct *idle;
+
+		if (cpu == boot_cpu)
+			continue;
+
+		if (!cpu_possible(cpu)) {
+			/*
+			 * Make this processor do nothing on boot.
+			 * Note that we don't give the boot_pc function
+			 * a stack, so it has to be assembly code.
+			 */
+			per_cpu(boot_sp, cpu) = 0;
+			per_cpu(boot_pc, cpu) = (unsigned long) smp_nap;
+			continue;
+		}
+
+		/* Create a new idle thread to run start_secondary() */
+		idle = fork_idle(cpu);
+		if (IS_ERR(idle))
+			panic("failed fork for CPU %d", cpu);
+		idle->thread.pc = (unsigned long) start_secondary;
+
+		/* Make this thread the boot thread for this processor */
+		per_cpu(boot_sp, cpu) = task_ksp0(idle);
+		per_cpu(boot_pc, cpu) = idle->thread.pc;
+
+		++cpu_count;
+	}
+	BUG_ON(cpu_count > (max_cpus ? max_cpus : 1));
+
+	/* Fire up the other tiles, if any */
+	init_cpu_present(cpu_possible_mask);
+	if (cpumask_weight(cpu_present_mask) > 1) {
+		mb();  /* make sure all data is visible to new processors */
+		hv_start_all_tiles();
+	}
+}
+
+static __initdata struct cpumask init_affinity;
+
+static __init int reset_init_affinity(void)
+{
+	long rc = sched_setaffinity(current->pid, &init_affinity);
+	if (rc != 0)
+		pr_warning("couldn't reset init affinity (%ld)\n",
+		       rc);
+	return 0;
+}
+late_initcall(reset_init_affinity);
+
+static struct cpumask cpu_started __cpuinitdata;
+
+/*
+ * Activate a secondary processor.  Very minimal; don't add anything
+ * to this path without knowing what you're doing, since SMP booting
+ * is pretty fragile.
+ */
+static void __cpuinit start_secondary(void)
+{
+	int cpuid = smp_processor_id();
+
+	/* Set our thread pointer appropriately. */
+	set_my_cpu_offset(__per_cpu_offset[cpuid]);
+
+	preempt_disable();
+
+	/*
+	 * In large machines even this will slow us down, since we
+	 * will be contending for for the printk spinlock.
+	 */
+	/* printk(KERN_DEBUG "Initializing CPU#%d\n", cpuid); */
+
+	/* Initialize the current asid for our first page table. */
+	__get_cpu_var(current_asid) = min_asid;
+
+	/* Set up this thread as another owner of the init_mm */
+	atomic_inc(&init_mm.mm_count);
+	current->active_mm = &init_mm;
+	if (current->mm)
+		BUG();
+	enter_lazy_tlb(&init_mm, current);
+
+	/* Allow hypervisor messages to be received */
+	init_messaging();
+	local_irq_enable();
+
+	/* Indicate that we're ready to come up. */
+	/* Must not do this before we're ready to receive messages */
+	if (cpumask_test_and_set_cpu(cpuid, &cpu_started)) {
+		pr_warning("CPU#%d already started!\n", cpuid);
+		for (;;)
+			local_irq_enable();
+	}
+
+	smp_nap();
+}
+
+/*
+ * Bring a secondary processor online.
+ */
+void __cpuinit online_secondary(void)
+{
+	/*
+	 * low-memory mappings have been cleared, flush them from
+	 * the local TLBs too.
+	 */
+	local_flush_tlb();
+
+	BUG_ON(in_interrupt());
+
+	/* This must be done before setting cpu_online_mask */
+	wmb();
+
+	notify_cpu_starting(smp_processor_id());
+
+	set_cpu_online(smp_processor_id(), 1);
+	__get_cpu_var(cpu_state) = CPU_ONLINE;
+
+	/* Set up tile-specific state for this cpu. */
+	setup_cpu(0);
+
+	/* Set up tile-timer clock-event device on this cpu */
+	setup_tile_timer();
+
+	cpu_startup_entry(CPUHP_ONLINE);
+}
+
+int __cpuinit __cpu_up(unsigned int cpu, struct task_struct *tidle)
+{
+	/* Wait 5s total for all CPUs for them to come online */
+	static int timeout;
+	for (; !cpumask_test_cpu(cpu, &cpu_started); timeout++) {
+		if (timeout >= 50000) {
+			pr_info("skipping unresponsive cpu%d\n", cpu);
+			local_irq_enable();
+			return -EIO;
+		}
+		udelay(100);
+	}
+
+	local_irq_enable();
+	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
+
+	/* Unleash the CPU! */
+	send_IPI_single(cpu, MSG_TAG_START_CPU);
+	while (!cpumask_test_cpu(cpu, cpu_online_mask))
+		cpu_relax();
+	return 0;
+}
+
+static void panic_start_cpu(void)
+{
+	panic("Received a MSG_START_CPU IPI after boot finished.");
+}
+
+void __init smp_cpus_done(unsigned int max_cpus)
+{
+	int cpu, next, rc;
+
+	/* Reset the response to a (now illegal) MSG_START_CPU IPI. */
+	start_cpu_function_addr = (unsigned long) &panic_start_cpu;
+
+	cpumask_copy(&init_affinity, cpu_online_mask);
+
+	/*
+	 * Pin ourselves to a single cpu in the initial affinity set
+	 * so that kernel mappings for the rootfs are not in the dataplane,
+	 * if set, and to avoid unnecessary migrating during bringup.
+	 * Use the last cpu just in case the whole chip has been
+	 * isolated from the scheduler, to keep init away from likely
+	 * more useful user code.  This also ensures that work scheduled
+	 * via schedule_delayed_work() in the init routines will land
+	 * on this cpu.
+	 */
+	for (cpu = cpumask_first(&init_affinity);
+	     (next = cpumask_next(cpu, &init_affinity)) < nr_cpu_ids;
+	     cpu = next)
+		;
+	rc = sched_setaffinity(current->pid, cpumask_of(cpu));
+	if (rc != 0)
+		pr_err("Couldn't set init affinity to cpu %d (%d)\n", cpu, rc);
+}

arch/tile/kernel/stack.c

@@ -0,0 +1,493 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/kprobes.h>
+#include <linux/module.h>
+#include <linux/pfn.h>
+#include <linux/kallsyms.h>
+#include <linux/stacktrace.h>
+#include <linux/uaccess.h>
+#include <linux/mmzone.h>
+#include <linux/dcache.h>
+#include <linux/fs.h>
+#include <asm/backtrace.h>
+#include <asm/page.h>
+#include <asm/ucontext.h>
+#include <asm/switch_to.h>
+#include <asm/sigframe.h>
+#include <asm/stack.h>
+#include <asm/vdso.h>
+#include <arch/abi.h>
+#include <arch/interrupts.h>
+
+#define KBT_ONGOING	0  /* Backtrace still ongoing */
+#define KBT_DONE	1  /* Backtrace cleanly completed */
+#define KBT_RUNNING	2  /* Can't run backtrace on a running task */
+#define KBT_LOOP	3  /* Backtrace entered a loop */
+
+/* Is address on the specified kernel stack? */
+static int in_kernel_stack(struct KBacktraceIterator *kbt, unsigned long sp)
+{
+	ulong kstack_base = (ulong) kbt->task->stack;
+	if (kstack_base == 0)  /* corrupt task pointer; just follow stack... */
+		return sp >= PAGE_OFFSET && sp < (unsigned long)high_memory;
+	return sp >= kstack_base && sp < kstack_base + THREAD_SIZE;
+}
+
+/* Callback for backtracer; basically a glorified memcpy */
+static bool read_memory_func(void *result, unsigned long address,
+			     unsigned int size, void *vkbt)
+{
+	int retval;
+	struct KBacktraceIterator *kbt = (struct KBacktraceIterator *)vkbt;
+
+	if (address == 0)
+		return 0;
+	if (__kernel_text_address(address)) {
+		/* OK to read kernel code. */
+	} else if (address >= PAGE_OFFSET) {
+		/* We only tolerate kernel-space reads of this task's stack */
+		if (!in_kernel_stack(kbt, address))
+			return 0;
+	} else if (!kbt->is_current) {
+		return 0;	/* can't read from other user address spaces */
+	}
+	pagefault_disable();
+	retval = __copy_from_user_inatomic(result,
+					   (void __user __force *)address,
+					   size);
+	pagefault_enable();
+	return (retval == 0);
+}
+
+/* Return a pt_regs pointer for a valid fault handler frame */
+static struct pt_regs *valid_fault_handler(struct KBacktraceIterator* kbt)
+{
+	const char *fault = NULL;  /* happy compiler */
+	char fault_buf[64];
+	unsigned long sp = kbt->it.sp;
+	struct pt_regs *p;
+
+	if (sp % sizeof(long) != 0)
+		return NULL;
+	if (!in_kernel_stack(kbt, sp))
+		return NULL;
+	if (!in_kernel_stack(kbt, sp + C_ABI_SAVE_AREA_SIZE + PTREGS_SIZE-1))
+		return NULL;
+	p = (struct pt_regs *)(sp + C_ABI_SAVE_AREA_SIZE);
+	if (p->faultnum == INT_SWINT_1 || p->faultnum == INT_SWINT_1_SIGRETURN)
+		fault = "syscall";
+	else {
+		if (kbt->verbose) {     /* else we aren't going to use it */
+			snprintf(fault_buf, sizeof(fault_buf),
+				 "interrupt %ld", p->faultnum);
+			fault = fault_buf;
+		}
+	}
+	if (EX1_PL(p->ex1) == KERNEL_PL &&
+	    __kernel_text_address(p->pc) &&
+	    in_kernel_stack(kbt, p->sp) &&
+	    p->sp >= sp) {
+		if (kbt->verbose)
+			pr_err("  <%s while in kernel mode>\n", fault);
+	} else if (EX1_PL(p->ex1) == USER_PL &&
+	    p->pc < PAGE_OFFSET &&
+	    p->sp < PAGE_OFFSET) {
+		if (kbt->verbose)
+			pr_err("  <%s while in user mode>\n", fault);
+	} else if (kbt->verbose) {
+		pr_err("  (odd fault: pc %#lx, sp %#lx, ex1 %#lx?)\n",
+		       p->pc, p->sp, p->ex1);
+		p = NULL;
+	}
+	if (!kbt->profile || ((1ULL << p->faultnum) & QUEUED_INTERRUPTS) == 0)
+		return p;
+	return NULL;
+}
+
+/* Is the pc pointing to a sigreturn trampoline? */
+static int is_sigreturn(unsigned long pc)
+{
+	return current->mm && (pc == VDSO_SYM(&__vdso_rt_sigreturn));
+}
+
+/* Return a pt_regs pointer for a valid signal handler frame */
+static struct pt_regs *valid_sigframe(struct KBacktraceIterator* kbt,
+				      struct rt_sigframe* kframe)
+{
+	BacktraceIterator *b = &kbt->it;
+
+	if (is_sigreturn(b->pc) && b->sp < PAGE_OFFSET &&
+	    b->sp % sizeof(long) == 0) {
+		int retval;
+		pagefault_disable();
+		retval = __copy_from_user_inatomic(
+			kframe, (void __user __force *)b->sp,
+			sizeof(*kframe));
+		pagefault_enable();
+		if (retval != 0 ||
+		    (unsigned int)(kframe->info.si_signo) >= _NSIG)
+			return NULL;
+		if (kbt->verbose) {
+			pr_err("  <received signal %d>\n",
+			       kframe->info.si_signo);
+		}
+		return (struct pt_regs *)&kframe->uc.uc_mcontext;
+	}
+	return NULL;
+}
+
+static int KBacktraceIterator_is_sigreturn(struct KBacktraceIterator *kbt)
+{
+	return is_sigreturn(kbt->it.pc);
+}
+
+static int KBacktraceIterator_restart(struct KBacktraceIterator *kbt)
+{
+	struct pt_regs *p;
+	struct rt_sigframe kframe;
+
+	p = valid_fault_handler(kbt);
+	if (p == NULL)
+		p = valid_sigframe(kbt, &kframe);
+	if (p == NULL)
+		return 0;
+	backtrace_init(&kbt->it, read_memory_func, kbt,
+		       p->pc, p->lr, p->sp, p->regs[52]);
+	kbt->new_context = 1;
+	return 1;
+}
+
+/* Find a frame that isn't a sigreturn, if there is one. */
+static int KBacktraceIterator_next_item_inclusive(
+	struct KBacktraceIterator *kbt)
+{
+	for (;;) {
+		do {
+			if (!KBacktraceIterator_is_sigreturn(kbt))
+				return KBT_ONGOING;
+		} while (backtrace_next(&kbt->it));
+
+		if (!KBacktraceIterator_restart(kbt))
+			return KBT_DONE;
+	}
+}
+
+/*
+ * If the current sp is on a page different than what we recorded
+ * as the top-of-kernel-stack last time we context switched, we have
+ * probably blown the stack, and nothing is going to work out well.
+ * If we can at least get out a warning, that may help the debug,
+ * though we probably won't be able to backtrace into the code that
+ * actually did the recursive damage.
+ */
+static void validate_stack(struct pt_regs *regs)
+{
+	int cpu = smp_processor_id();
+	unsigned long ksp0 = get_current_ksp0();
+	unsigned long ksp0_base = ksp0 - THREAD_SIZE;
+	unsigned long sp = stack_pointer;
+
+	if (EX1_PL(regs->ex1) == KERNEL_PL && regs->sp >= ksp0) {
+		pr_err("WARNING: cpu %d: kernel stack page %#lx underrun!\n"
+		       "  sp %#lx (%#lx in caller), caller pc %#lx, lr %#lx\n",
+		       cpu, ksp0_base, sp, regs->sp, regs->pc, regs->lr);
+	}
+
+	else if (sp < ksp0_base + sizeof(struct thread_info)) {
+		pr_err("WARNING: cpu %d: kernel stack page %#lx overrun!\n"
+		       "  sp %#lx (%#lx in caller), caller pc %#lx, lr %#lx\n",
+		       cpu, ksp0_base, sp, regs->sp, regs->pc, regs->lr);
+	}
+}
+
+void KBacktraceIterator_init(struct KBacktraceIterator *kbt,
+			     struct task_struct *t, struct pt_regs *regs)
+{
+	unsigned long pc, lr, sp, r52;
+	int is_current;
+
+	/*
+	 * Set up callback information.  We grab the kernel stack base
+	 * so we will allow reads of that address range.
+	 */
+	is_current = (t == NULL || t == current);
+	kbt->is_current = is_current;
+	if (is_current)
+		t = validate_current();
+	kbt->task = t;
+	kbt->verbose = 0;   /* override in caller if desired */
+	kbt->profile = 0;   /* override in caller if desired */
+	kbt->end = KBT_ONGOING;
+	kbt->new_context = 1;
+	if (is_current)
+		validate_stack(regs);
+
+	if (regs == NULL) {
+		if (is_current || t->state == TASK_RUNNING) {
+			/* Can't do this; we need registers */
+			kbt->end = KBT_RUNNING;
+			return;
+		}
+		pc = get_switch_to_pc();
+		lr = t->thread.pc;
+		sp = t->thread.ksp;
+		r52 = 0;
+	} else {
+		pc = regs->pc;
+		lr = regs->lr;
+		sp = regs->sp;
+		r52 = regs->regs[52];
+	}
+
+	backtrace_init(&kbt->it, read_memory_func, kbt, pc, lr, sp, r52);
+	kbt->end = KBacktraceIterator_next_item_inclusive(kbt);
+}
+EXPORT_SYMBOL(KBacktraceIterator_init);
+
+int KBacktraceIterator_end(struct KBacktraceIterator *kbt)
+{
+	return kbt->end != KBT_ONGOING;
+}
+EXPORT_SYMBOL(KBacktraceIterator_end);
+
+void KBacktraceIterator_next(struct KBacktraceIterator *kbt)
+{
+	unsigned long old_pc = kbt->it.pc, old_sp = kbt->it.sp;
+	kbt->new_context = 0;
+	if (!backtrace_next(&kbt->it) && !KBacktraceIterator_restart(kbt)) {
+		kbt->end = KBT_DONE;
+		return;
+	}
+	kbt->end = KBacktraceIterator_next_item_inclusive(kbt);
+	if (old_pc == kbt->it.pc && old_sp == kbt->it.sp) {
+		/* Trapped in a loop; give up. */
+		kbt->end = KBT_LOOP;
+	}
+}
+EXPORT_SYMBOL(KBacktraceIterator_next);
+
+static void describe_addr(struct KBacktraceIterator *kbt,
+			  unsigned long address,
+			  int have_mmap_sem, char *buf, size_t bufsize)
+{
+	struct vm_area_struct *vma;
+	size_t namelen, remaining;
+	unsigned long size, offset, adjust;
+	char *p, *modname;
+	const char *name;
+	int rc;
+
+	/*
+	 * Look one byte back for every caller frame (i.e. those that
+	 * aren't a new context) so we look up symbol data for the
+	 * call itself, not the following instruction, which may be on
+	 * a different line (or in a different function).
+	 */
+	adjust = !kbt->new_context;
+	address -= adjust;
+
+	if (address >= PAGE_OFFSET) {
+		/* Handle kernel symbols. */
+		BUG_ON(bufsize < KSYM_NAME_LEN);
+		name = kallsyms_lookup(address, &size, &offset,
+				       &modname, buf);
+		if (name == NULL) {
+			buf[0] = '\0';
+			return;
+		}
+		namelen = strlen(buf);
+		remaining = (bufsize - 1) - namelen;
+		p = buf + namelen;
+		rc = snprintf(p, remaining, "+%#lx/%#lx ",
+			      offset + adjust, size);
+		if (modname && rc < remaining)
+			snprintf(p + rc, remaining - rc, "[%s] ", modname);
+		buf[bufsize-1] = '\0';
+		return;
+	}
+
+	/* If we don't have the mmap_sem, we can't show any more info. */
+	buf[0] = '\0';
+	if (!have_mmap_sem)
+		return;
+
+	/* Find vma info. */
+	vma = find_vma(kbt->task->mm, address);
+	if (vma == NULL || address < vma->vm_start) {
+		snprintf(buf, bufsize, "[unmapped address] ");
+		return;
+	}
+
+	if (vma->vm_file) {
+		char *s;
+		p = d_path(&vma->vm_file->f_path, buf, bufsize);
+		if (IS_ERR(p))
+			p = "?";
+		s = strrchr(p, '/');
+		if (s)
+			p = s+1;
+	} else {
+		p = "anon";
+	}
+
+	/* Generate a string description of the vma info. */
+	namelen = strlen(p);
+	remaining = (bufsize - 1) - namelen;
+	memmove(buf, p, namelen);
+	snprintf(buf + namelen, remaining, "[%lx+%lx] ",
+		 vma->vm_start, vma->vm_end - vma->vm_start);
+}
+
+/*
+ * This method wraps the backtracer's more generic support.
+ * It is only invoked from the architecture-specific code; show_stack()
+ * and dump_stack() (in entry.S) are architecture-independent entry points.
+ */
+void tile_show_stack(struct KBacktraceIterator *kbt, int headers)
+{
+	int i;
+	int have_mmap_sem = 0;
+
+	if (headers) {
+		/*
+		 * Add a blank line since if we are called from panic(),
+		 * then bust_spinlocks() spit out a space in front of us
+		 * and it will mess up our KERN_ERR.
+		 */
+		pr_err("\n");
+		pr_err("Starting stack dump of tid %d, pid %d (%s)"
+		       " on cpu %d at cycle %lld\n",
+		       kbt->task->pid, kbt->task->tgid, kbt->task->comm,
+		       smp_processor_id(), get_cycles());
+	}
+	kbt->verbose = 1;
+	i = 0;
+	for (; !KBacktraceIterator_end(kbt); KBacktraceIterator_next(kbt)) {
+		char namebuf[KSYM_NAME_LEN+100];
+		unsigned long address = kbt->it.pc;
+
+		/* Try to acquire the mmap_sem as we pass into userspace. */
+		if (address < PAGE_OFFSET && !have_mmap_sem && kbt->task->mm)
+			have_mmap_sem =
+				down_read_trylock(&kbt->task->mm->mmap_sem);
+
+		describe_addr(kbt, address, have_mmap_sem,
+			      namebuf, sizeof(namebuf));
+
+		pr_err("  frame %d: 0x%lx %s(sp 0x%lx)\n",
+		       i++, address, namebuf, (unsigned long)(kbt->it.sp));
+
+		if (i >= 100) {
+			pr_err("Stack dump truncated"
+			       " (%d frames)\n", i);
+			break;
+		}
+	}
+	if (kbt->end == KBT_LOOP)
+		pr_err("Stack dump stopped; next frame identical to this one\n");
+	if (headers)
+		pr_err("Stack dump complete\n");
+	if (have_mmap_sem)
+		up_read(&kbt->task->mm->mmap_sem);
+}
+EXPORT_SYMBOL(tile_show_stack);
+
+
+/* This is called from show_regs() and _dump_stack() */
+void dump_stack_regs(struct pt_regs *regs)
+{
+	struct KBacktraceIterator kbt;
+	KBacktraceIterator_init(&kbt, NULL, regs);
+	tile_show_stack(&kbt, 1);
+}
+EXPORT_SYMBOL(dump_stack_regs);
+
+static struct pt_regs *regs_to_pt_regs(struct pt_regs *regs,
+				       ulong pc, ulong lr, ulong sp, ulong r52)
+{
+	memset(regs, 0, sizeof(struct pt_regs));
+	regs->pc = pc;
+	regs->lr = lr;
+	regs->sp = sp;
+	regs->regs[52] = r52;
+	return regs;
+}
+
+/* This is called from dump_stack() and just converts to pt_regs */
+void _dump_stack(int dummy, ulong pc, ulong lr, ulong sp, ulong r52)
+{
+	struct pt_regs regs;
+	dump_stack_regs(regs_to_pt_regs(&regs, pc, lr, sp, r52));
+}
+
+/* This is called from KBacktraceIterator_init_current() */
+void _KBacktraceIterator_init_current(struct KBacktraceIterator *kbt, ulong pc,
+				      ulong lr, ulong sp, ulong r52)
+{
+	struct pt_regs regs;
+	KBacktraceIterator_init(kbt, NULL,
+				regs_to_pt_regs(&regs, pc, lr, sp, r52));
+}
+
+/* This is called only from kernel/sched.c, with esp == NULL */
+void show_stack(struct task_struct *task, unsigned long *esp)
+{
+	struct KBacktraceIterator kbt;
+	if (task == NULL || task == current)
+		KBacktraceIterator_init_current(&kbt);
+	else
+		KBacktraceIterator_init(&kbt, task, NULL);
+	tile_show_stack(&kbt, 0);
+}
+
+#ifdef CONFIG_STACKTRACE
+
+/* Support generic Linux stack API too */
+
+void save_stack_trace_tsk(struct task_struct *task, struct stack_trace *trace)
+{
+	struct KBacktraceIterator kbt;
+	int skip = trace->skip;
+	int i = 0;
+
+	if (task == NULL || task == current)
+		KBacktraceIterator_init_current(&kbt);
+	else
+		KBacktraceIterator_init(&kbt, task, NULL);
+	for (; !KBacktraceIterator_end(&kbt); KBacktraceIterator_next(&kbt)) {
+		if (skip) {
+			--skip;
+			continue;
+		}
+		if (i >= trace->max_entries || kbt.it.pc < PAGE_OFFSET)
+			break;
+		trace->entries[i++] = kbt.it.pc;
+	}
+	trace->nr_entries = i;
+}
+EXPORT_SYMBOL(save_stack_trace_tsk);
+
+void save_stack_trace(struct stack_trace *trace)
+{
+	save_stack_trace_tsk(NULL, trace);
+}
+EXPORT_SYMBOL_GPL(save_stack_trace);
+
+#endif
+
+/* In entry.S */
+EXPORT_SYMBOL(KBacktraceIterator_init_current);

arch/tile/kernel/sys.c

@@ -0,0 +1,121 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * This file contains various random system calls that
+ * have a non-standard calling sequence on the Linux/TILE
+ * platform.
+ */
+
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/smp.h>
+#include <linux/syscalls.h>
+#include <linux/mman.h>
+#include <linux/file.h>
+#include <linux/mempolicy.h>
+#include <linux/binfmts.h>
+#include <linux/fs.h>
+#include <linux/compat.h>
+#include <linux/uaccess.h>
+#include <linux/signal.h>
+#include <asm/syscalls.h>
+#include <asm/pgtable.h>
+#include <asm/homecache.h>
+#include <asm/cachectl.h>
+#include <arch/chip.h>
+
+SYSCALL_DEFINE3(cacheflush, unsigned long, addr, unsigned long, len,
+		unsigned long, flags)
+{
+	if (flags & DCACHE)
+		homecache_evict(cpumask_of(smp_processor_id()));
+	if (flags & ICACHE)
+		flush_remote(0, HV_FLUSH_EVICT_L1I, mm_cpumask(current->mm),
+			     0, 0, 0, NULL, NULL, 0);
+	return 0;
+}
+
+/*
+ * Syscalls that pass 64-bit values on 32-bit systems normally
+ * pass them as (low,high) word packed into the immediately adjacent
+ * registers.  If the low word naturally falls on an even register,
+ * our ABI makes it work correctly; if not, we adjust it here.
+ * Handling it here means we don't have to fix uclibc AND glibc AND
+ * any other standard libcs we want to support.
+ */
+
+#if !defined(__tilegx__) || defined(CONFIG_COMPAT)
+
+ssize_t sys32_readahead(int fd, u32 offset_lo, u32 offset_hi, u32 count)
+{
+	return sys_readahead(fd, ((loff_t)offset_hi << 32) | offset_lo, count);
+}
+
+int sys32_fadvise64_64(int fd, u32 offset_lo, u32 offset_hi,
+		       u32 len_lo, u32 len_hi, int advice)
+{
+	return sys_fadvise64_64(fd, ((loff_t)offset_hi << 32) | offset_lo,
+				((loff_t)len_hi << 32) | len_lo, advice);
+}
+
+#endif /* 32-bit syscall wrappers */
+
+/* Note: used by the compat code even in 64-bit Linux. */
+SYSCALL_DEFINE6(mmap2, unsigned long, addr, unsigned long, len,
+		unsigned long, prot, unsigned long, flags,
+		unsigned long, fd, unsigned long, off_4k)
+{
+#define PAGE_ADJUST (PAGE_SHIFT - 12)
+	if (off_4k & ((1 << PAGE_ADJUST) - 1))
+		return -EINVAL;
+	return sys_mmap_pgoff(addr, len, prot, flags, fd,
+			      off_4k >> PAGE_ADJUST);
+}
+
+#ifdef __tilegx__
+SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
+		unsigned long, prot, unsigned long, flags,
+		unsigned long, fd, off_t, offset)
+{
+	if (offset & ((1 << PAGE_SHIFT) - 1))
+		return -EINVAL;
+	return sys_mmap_pgoff(addr, len, prot, flags, fd,
+			      offset >> PAGE_SHIFT);
+}
+#endif
+
+
+/* Provide the actual syscall number to call mapping. */
+#undef __SYSCALL
+#define __SYSCALL(nr, call) [nr] = (call),
+
+#ifndef __tilegx__
+/* See comments at the top of the file. */
+#define sys_fadvise64_64 sys32_fadvise64_64
+#define sys_readahead sys32_readahead
+#endif
+
+/* Call the assembly trampolines where necessary. */
+#undef sys_rt_sigreturn
+#define sys_rt_sigreturn _sys_rt_sigreturn
+#define sys_clone _sys_clone
+
+/*
+ * Note that we can't include <linux/unistd.h> here since the header
+ * guard will defeat us; <asm/unistd.h> checks for __SYSCALL as well.
+ */
+void *sys_call_table[__NR_syscalls] = {
+	[0 ... __NR_syscalls-1] = sys_ni_syscall,
+#include <asm/unistd.h>
+};

arch/tile/kernel/sysfs.c

@@ -0,0 +1,193 @@
+/*
+ * Copyright 2011 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * /sys entry support.
+ */
+
+#include <linux/device.h>
+#include <linux/cpu.h>
+#include <linux/slab.h>
+#include <linux/smp.h>
+#include <linux/stat.h>
+#include <hv/hypervisor.h>
+
+/* Return a string queried from the hypervisor, truncated to page size. */
+static ssize_t get_hv_confstr(char *page, int query)
+{
+	ssize_t n = hv_confstr(query, (unsigned long)page, PAGE_SIZE - 1);
+	n = n < 0 ? 0 : min(n, (ssize_t)PAGE_SIZE - 1) - 1;
+	if (n)
+		page[n++] = '\n';
+	page[n] = '\0';
+	return n;
+}
+
+static ssize_t chip_width_show(struct device *dev,
+			       struct device_attribute *attr,
+			       char *page)
+{
+	return sprintf(page, "%u\n", smp_width);
+}
+static DEVICE_ATTR(chip_width, 0444, chip_width_show, NULL);
+
+static ssize_t chip_height_show(struct device *dev,
+				struct device_attribute *attr,
+				char *page)
+{
+	return sprintf(page, "%u\n", smp_height);
+}
+static DEVICE_ATTR(chip_height, 0444, chip_height_show, NULL);
+
+static ssize_t chip_serial_show(struct device *dev,
+				struct device_attribute *attr,
+				char *page)
+{
+	return get_hv_confstr(page, HV_CONFSTR_CHIP_SERIAL_NUM);
+}
+static DEVICE_ATTR(chip_serial, 0444, chip_serial_show, NULL);
+
+static ssize_t chip_revision_show(struct device *dev,
+				  struct device_attribute *attr,
+				  char *page)
+{
+	return get_hv_confstr(page, HV_CONFSTR_CHIP_REV);
+}
+static DEVICE_ATTR(chip_revision, 0444, chip_revision_show, NULL);
+
+
+static ssize_t type_show(struct device *dev,
+			    struct device_attribute *attr,
+			    char *page)
+{
+	return sprintf(page, "tilera\n");
+}
+static DEVICE_ATTR(type, 0444, type_show, NULL);
+
+#define HV_CONF_ATTR(name, conf)					\
+	static ssize_t name ## _show(struct device *dev,		\
+				     struct device_attribute *attr, \
+				     char *page)			\
+	{								\
+		return get_hv_confstr(page, conf);			\
+	}								\
+	static DEVICE_ATTR(name, 0444, name ## _show, NULL);
+
+HV_CONF_ATTR(version,		HV_CONFSTR_HV_SW_VER)
+HV_CONF_ATTR(config_version,	HV_CONFSTR_HV_CONFIG_VER)
+
+HV_CONF_ATTR(board_part,	HV_CONFSTR_BOARD_PART_NUM)
+HV_CONF_ATTR(board_serial,	HV_CONFSTR_BOARD_SERIAL_NUM)
+HV_CONF_ATTR(board_revision,	HV_CONFSTR_BOARD_REV)
+HV_CONF_ATTR(board_description,	HV_CONFSTR_BOARD_DESC)
+HV_CONF_ATTR(mezz_part,		HV_CONFSTR_MEZZ_PART_NUM)
+HV_CONF_ATTR(mezz_serial,	HV_CONFSTR_MEZZ_SERIAL_NUM)
+HV_CONF_ATTR(mezz_revision,	HV_CONFSTR_MEZZ_REV)
+HV_CONF_ATTR(mezz_description,	HV_CONFSTR_MEZZ_DESC)
+HV_CONF_ATTR(cpumod_part,	HV_CONFSTR_CPUMOD_PART_NUM)
+HV_CONF_ATTR(cpumod_serial,	HV_CONFSTR_CPUMOD_SERIAL_NUM)
+HV_CONF_ATTR(cpumod_revision,	HV_CONFSTR_CPUMOD_REV)
+HV_CONF_ATTR(cpumod_description,HV_CONFSTR_CPUMOD_DESC)
+HV_CONF_ATTR(switch_control,	HV_CONFSTR_SWITCH_CONTROL)
+
+static struct attribute *board_attrs[] = {
+	&dev_attr_board_part.attr,
+	&dev_attr_board_serial.attr,
+	&dev_attr_board_revision.attr,
+	&dev_attr_board_description.attr,
+	&dev_attr_mezz_part.attr,
+	&dev_attr_mezz_serial.attr,
+	&dev_attr_mezz_revision.attr,
+	&dev_attr_mezz_description.attr,
+	&dev_attr_cpumod_part.attr,
+	&dev_attr_cpumod_serial.attr,
+	&dev_attr_cpumod_revision.attr,
+	&dev_attr_cpumod_description.attr,
+	&dev_attr_switch_control.attr,
+	NULL
+};
+
+static struct attribute_group board_attr_group = {
+	.name   = "board",
+	.attrs  = board_attrs,
+};
+
+
+static struct bin_attribute hvconfig_bin;
+
+static ssize_t
+hvconfig_bin_read(struct file *filp, struct kobject *kobj,
+		  struct bin_attribute *bin_attr,
+		  char *buf, loff_t off, size_t count)
+{
+	static size_t size;
+
+	/* Lazily learn the true size (minus the trailing NUL). */
+	if (size == 0)
+		size = hv_confstr(HV_CONFSTR_HV_CONFIG, 0, 0) - 1;
+
+	/* Check and adjust input parameters. */
+	if (off > size)
+		return -EINVAL;
+	if (count > size - off)
+		count = size - off;
+
+	if (count) {
+		/* Get a copy of the hvc and copy out the relevant portion. */
+		char *hvc;
+
+		size = off + count;
+		hvc = kmalloc(size, GFP_KERNEL);
+		if (hvc == NULL)
+			return -ENOMEM;
+		hv_confstr(HV_CONFSTR_HV_CONFIG, (unsigned long)hvc, size);
+		memcpy(buf, hvc + off, count);
+		kfree(hvc);
+	}
+
+	return count;
+}
+
+static int __init create_sysfs_entries(void)
+{
+	int err = 0;
+
+#define create_cpu_attr(name)						\
+	if (!err)							\
+		err = device_create_file(cpu_subsys.dev_root, &dev_attr_##name);
+	create_cpu_attr(chip_width);
+	create_cpu_attr(chip_height);
+	create_cpu_attr(chip_serial);
+	create_cpu_attr(chip_revision);
+
+#define create_hv_attr(name)						\
+	if (!err)							\
+		err = sysfs_create_file(hypervisor_kobj, &dev_attr_##name.attr);
+	create_hv_attr(type);
+	create_hv_attr(version);
+	create_hv_attr(config_version);
+
+	if (!err)
+		err = sysfs_create_group(hypervisor_kobj, &board_attr_group);
+
+	if (!err) {
+		sysfs_bin_attr_init(&hvconfig_bin);
+		hvconfig_bin.attr.name = "hvconfig";
+		hvconfig_bin.attr.mode = S_IRUGO;
+		hvconfig_bin.read = hvconfig_bin_read;
+		hvconfig_bin.size = PAGE_SIZE;
+		err = sysfs_create_bin_file(hypervisor_kobj, &hvconfig_bin);
+	}
+
+	return err;
+}
+subsys_initcall(create_sysfs_entries);

arch/tile/kernel/time.c

@@ -0,0 +1,274 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * Support the cycle counter clocksource and tile timer clock event device.
+ */
+
+#include <linux/time.h>
+#include <linux/timex.h>
+#include <linux/clocksource.h>
+#include <linux/clockchips.h>
+#include <linux/hardirq.h>
+#include <linux/sched.h>
+#include <linux/smp.h>
+#include <linux/delay.h>
+#include <linux/module.h>
+#include <linux/timekeeper_internal.h>
+#include <asm/irq_regs.h>
+#include <asm/traps.h>
+#include <asm/vdso.h>
+#include <hv/hypervisor.h>
+#include <arch/interrupts.h>
+#include <arch/spr_def.h>
+
+
+/*
+ * Define the cycle counter clock source.
+ */
+
+/* How many cycles per second we are running at. */
+static cycles_t cycles_per_sec __write_once;
+
+cycles_t get_clock_rate(void)
+{
+	return cycles_per_sec;
+}
+
+#if CHIP_HAS_SPLIT_CYCLE()
+cycles_t get_cycles(void)
+{
+	unsigned int high = __insn_mfspr(SPR_CYCLE_HIGH);
+	unsigned int low = __insn_mfspr(SPR_CYCLE_LOW);
+	unsigned int high2 = __insn_mfspr(SPR_CYCLE_HIGH);
+
+	while (unlikely(high != high2)) {
+		low = __insn_mfspr(SPR_CYCLE_LOW);
+		high = high2;
+		high2 = __insn_mfspr(SPR_CYCLE_HIGH);
+	}
+
+	return (((cycles_t)high) << 32) | low;
+}
+EXPORT_SYMBOL(get_cycles);
+#endif
+
+/*
+ * We use a relatively small shift value so that sched_clock()
+ * won't wrap around very often.
+ */
+#define SCHED_CLOCK_SHIFT 10
+
+static unsigned long sched_clock_mult __write_once;
+
+static cycles_t clocksource_get_cycles(struct clocksource *cs)
+{
+	return get_cycles();
+}
+
+static struct clocksource cycle_counter_cs = {
+	.name = "cycle counter",
+	.rating = 300,
+	.read = clocksource_get_cycles,
+	.mask = CLOCKSOURCE_MASK(64),
+	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
+};
+
+/*
+ * Called very early from setup_arch() to set cycles_per_sec.
+ * We initialize it early so we can use it to set up loops_per_jiffy.
+ */
+void __init setup_clock(void)
+{
+	cycles_per_sec = hv_sysconf(HV_SYSCONF_CPU_SPEED);
+	sched_clock_mult =
+		clocksource_hz2mult(cycles_per_sec, SCHED_CLOCK_SHIFT);
+}
+
+void __init calibrate_delay(void)
+{
+	loops_per_jiffy = get_clock_rate() / HZ;
+	pr_info("Clock rate yields %lu.%02lu BogoMIPS (lpj=%lu)\n",
+		loops_per_jiffy/(500000/HZ),
+		(loops_per_jiffy/(5000/HZ)) % 100, loops_per_jiffy);
+}
+
+/* Called fairly late in init/main.c, but before we go smp. */
+void __init time_init(void)
+{
+	/* Initialize and register the clock source. */
+	clocksource_register_hz(&cycle_counter_cs, cycles_per_sec);
+
+	/* Start up the tile-timer interrupt source on the boot cpu. */
+	setup_tile_timer();
+}
+
+/*
+ * Define the tile timer clock event device.  The timer is driven by
+ * the TILE_TIMER_CONTROL register, which consists of a 31-bit down
+ * counter, plus bit 31, which signifies that the counter has wrapped
+ * from zero to (2**31) - 1.  The INT_TILE_TIMER interrupt will be
+ * raised as long as bit 31 is set.
+ *
+ * The TILE_MINSEC value represents the largest range of real-time
+ * we can possibly cover with the timer, based on MAX_TICK combined
+ * with the slowest reasonable clock rate we might run at.
+ */
+
+#define MAX_TICK 0x7fffffff   /* we have 31 bits of countdown timer */
+#define TILE_MINSEC 5         /* timer covers no more than 5 seconds */
+
+static int tile_timer_set_next_event(unsigned long ticks,
+				     struct clock_event_device *evt)
+{
+	BUG_ON(ticks > MAX_TICK);
+	__insn_mtspr(SPR_TILE_TIMER_CONTROL, ticks);
+	arch_local_irq_unmask_now(INT_TILE_TIMER);
+	return 0;
+}
+
+/*
+ * Whenever anyone tries to change modes, we just mask interrupts
+ * and wait for the next event to get set.
+ */
+static void tile_timer_set_mode(enum clock_event_mode mode,
+				struct clock_event_device *evt)
+{
+	arch_local_irq_mask_now(INT_TILE_TIMER);
+}
+
+/*
+ * Set min_delta_ns to 1 microsecond, since it takes about
+ * that long to fire the interrupt.
+ */
+static DEFINE_PER_CPU(struct clock_event_device, tile_timer) = {
+	.name = "tile timer",
+	.features = CLOCK_EVT_FEAT_ONESHOT,
+	.min_delta_ns = 1000,
+	.rating = 100,
+	.irq = -1,
+	.set_next_event = tile_timer_set_next_event,
+	.set_mode = tile_timer_set_mode,
+};
+
+void __cpuinit setup_tile_timer(void)
+{
+	struct clock_event_device *evt = &__get_cpu_var(tile_timer);
+
+	/* Fill in fields that are speed-specific. */
+	clockevents_calc_mult_shift(evt, cycles_per_sec, TILE_MINSEC);
+	evt->max_delta_ns = clockevent_delta2ns(MAX_TICK, evt);
+
+	/* Mark as being for this cpu only. */
+	evt->cpumask = cpumask_of(smp_processor_id());
+
+	/* Start out with timer not firing. */
+	arch_local_irq_mask_now(INT_TILE_TIMER);
+
+	/* Register tile timer. */
+	clockevents_register_device(evt);
+}
+
+/* Called from the interrupt vector. */
+void do_timer_interrupt(struct pt_regs *regs, int fault_num)
+{
+	struct pt_regs *old_regs = set_irq_regs(regs);
+	struct clock_event_device *evt = &__get_cpu_var(tile_timer);
+
+	/*
+	 * Mask the timer interrupt here, since we are a oneshot timer
+	 * and there are now by definition no events pending.
+	 */
+	arch_local_irq_mask(INT_TILE_TIMER);
+
+	/* Track time spent here in an interrupt context */
+	irq_enter();
+
+	/* Track interrupt count. */
+	__get_cpu_var(irq_stat).irq_timer_count++;
+
+	/* Call the generic timer handler */
+	evt->event_handler(evt);
+
+	/*
+	 * Track time spent against the current process again and
+	 * process any softirqs if they are waiting.
+	 */
+	irq_exit();
+
+	set_irq_regs(old_regs);
+}
+
+/*
+ * Scheduler clock - returns current time in nanosec units.
+ * Note that with LOCKDEP, this is called during lockdep_init(), and
+ * we will claim that sched_clock() is zero for a little while, until
+ * we run setup_clock(), above.
+ */
+unsigned long long sched_clock(void)
+{
+	return clocksource_cyc2ns(get_cycles(),
+				  sched_clock_mult, SCHED_CLOCK_SHIFT);
+}
+
+int setup_profiling_timer(unsigned int multiplier)
+{
+	return -EINVAL;
+}
+
+/*
+ * Use the tile timer to convert nsecs to core clock cycles, relying
+ * on it having the same frequency as SPR_CYCLE.
+ */
+cycles_t ns2cycles(unsigned long nsecs)
+{
+	/*
+	 * We do not have to disable preemption here as each core has the same
+	 * clock frequency.
+	 */
+	struct clock_event_device *dev = &__raw_get_cpu_var(tile_timer);
+	return ((u64)nsecs * dev->mult) >> dev->shift;
+}
+
+void update_vsyscall_tz(void)
+{
+	/* Userspace gettimeofday will spin while this value is odd. */
+	++vdso_data->tz_update_count;
+	smp_wmb();
+	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
+	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
+	smp_wmb();
+	++vdso_data->tz_update_count;
+}
+
+void update_vsyscall(struct timekeeper *tk)
+{
+	struct timespec wall_time = tk_xtime(tk);
+	struct timespec *wtm = &tk->wall_to_monotonic;
+	struct clocksource *clock = tk->tkr.clock;
+
+	if (clock != &cycle_counter_cs)
+		return;
+
+	/* Userspace gettimeofday will spin while this value is odd. */
+	++vdso_data->tb_update_count;
+	smp_wmb();
+	vdso_data->xtime_tod_stamp = tk->tkr.cycle_last;
+	vdso_data->xtime_clock_sec = tk->xtime_sec;
+	vdso_data->xtime_clock_nsec = tk->tkr.xtime_nsec;
+	vdso_data->wtom_clock_sec = wtm->tv_sec;
+	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
+	vdso_data->mult = tk->tkr.mult;
+	vdso_data->shift = tk->tkr.shift;
+	smp_wmb();
+	++vdso_data->tb_update_count;
+}

arch/tile/kernel/tlb.c

@@ -0,0 +1,98 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ */
+
+#include <linux/cpumask.h>
+#include <linux/module.h>
+#include <linux/hugetlb.h>
+#include <asm/tlbflush.h>
+#include <asm/homecache.h>
+#include <hv/hypervisor.h>
+
+/* From tlbflush.h */
+DEFINE_PER_CPU(int, current_asid);
+int min_asid, max_asid;
+
+/*
+ * Note that we flush the L1I (for VM_EXEC pages) as well as the TLB
+ * so that when we are unmapping an executable page, we also flush it.
+ * Combined with flushing the L1I at context switch time, this means
+ * we don't have to do any other icache flushes.
+ */
+
+void flush_tlb_mm(struct mm_struct *mm)
+{
+	HV_Remote_ASID asids[NR_CPUS];
+	int i = 0, cpu;
+	for_each_cpu(cpu, mm_cpumask(mm)) {
+		HV_Remote_ASID *asid = &asids[i++];
+		asid->y = cpu / smp_topology.width;
+		asid->x = cpu % smp_topology.width;
+		asid->asid = per_cpu(current_asid, cpu);
+	}
+	flush_remote(0, HV_FLUSH_EVICT_L1I, mm_cpumask(mm),
+		     0, 0, 0, NULL, asids, i);
+}
+
+void flush_tlb_current_task(void)
+{
+	flush_tlb_mm(current->mm);
+}
+
+void flush_tlb_page_mm(struct vm_area_struct *vma, struct mm_struct *mm,
+		       unsigned long va)
+{
+	unsigned long size = vma_kernel_pagesize(vma);
+	int cache = (vma->vm_flags & VM_EXEC) ? HV_FLUSH_EVICT_L1I : 0;
+	flush_remote(0, cache, mm_cpumask(mm),
+		     va, size, size, mm_cpumask(mm), NULL, 0);
+}
+
+void flush_tlb_page(struct vm_area_struct *vma, unsigned long va)
+{
+	flush_tlb_page_mm(vma, vma->vm_mm, va);
+}
+EXPORT_SYMBOL(flush_tlb_page);
+
+void flush_tlb_range(struct vm_area_struct *vma,
+		     unsigned long start, unsigned long end)
+{
+	unsigned long size = vma_kernel_pagesize(vma);
+	struct mm_struct *mm = vma->vm_mm;
+	int cache = (vma->vm_flags & VM_EXEC) ? HV_FLUSH_EVICT_L1I : 0;
+	flush_remote(0, cache, mm_cpumask(mm), start, end - start, size,
+		     mm_cpumask(mm), NULL, 0);
+}
+
+void flush_tlb_all(void)
+{
+	int i;
+	for (i = 0; ; ++i) {
+		HV_VirtAddrRange r = hv_inquire_virtual(i);
+		if (r.size == 0)
+			break;
+		flush_remote(0, HV_FLUSH_EVICT_L1I, cpu_online_mask,
+			     r.start, r.size, PAGE_SIZE, cpu_online_mask,
+			     NULL, 0);
+		flush_remote(0, 0, NULL,
+			     r.start, r.size, HPAGE_SIZE, cpu_online_mask,
+			     NULL, 0);
+	}
+}
+
+void flush_tlb_kernel_range(unsigned long start, unsigned long end)
+{
+	flush_remote(0, HV_FLUSH_EVICT_L1I, cpu_online_mask,
+		     start, end - start, PAGE_SIZE, cpu_online_mask, NULL, 0);
+}

arch/tile/kernel/traps.c

@@ -0,0 +1,355 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/kprobes.h>
+#include <linux/module.h>
+#include <linux/reboot.h>
+#include <linux/uaccess.h>
+#include <linux/ptrace.h>
+#include <asm/stack.h>
+#include <asm/traps.h>
+#include <asm/setup.h>
+
+#include <arch/interrupts.h>
+#include <arch/spr_def.h>
+#include <arch/opcode.h>
+
+void __init trap_init(void)
+{
+	/* Nothing needed here since we link code at .intrpt1 */
+}
+
+int unaligned_fixup = 1;
+
+static int __init setup_unaligned_fixup(char *str)
+{
+	/*
+	 * Say "=-1" to completely disable it.  If you just do "=0", we
+	 * will still parse the instruction, then fire a SIGBUS with
+	 * the correct address from inside the single_step code.
+	 */
+	long val;
+	if (strict_strtol(str, 0, &val) != 0)
+		return 0;
+	unaligned_fixup = val;
+	pr_info("Fixups for unaligned data accesses are %s\n",
+	       unaligned_fixup >= 0 ?
+	       (unaligned_fixup ? "enabled" : "disabled") :
+	       "completely disabled");
+	return 1;
+}
+__setup("unaligned_fixup=", setup_unaligned_fixup);
+
+#if CHIP_HAS_TILE_DMA()
+
+static int dma_disabled;
+
+static int __init nodma(char *str)
+{
+	pr_info("User-space DMA is disabled\n");
+	dma_disabled = 1;
+	return 1;
+}
+__setup("nodma", nodma);
+
+/* How to decode SPR_GPV_REASON */
+#define IRET_ERROR (1U << 31)
+#define MT_ERROR   (1U << 30)
+#define MF_ERROR   (1U << 29)
+#define SPR_INDEX  ((1U << 15) - 1)
+#define SPR_MPL_SHIFT  9  /* starting bit position for MPL encoded in SPR */
+
+/*
+ * See if this GPV is just to notify the kernel of SPR use and we can
+ * retry the user instruction after adjusting some MPLs suitably.
+ */
+static int retry_gpv(unsigned int gpv_reason)
+{
+	int mpl;
+
+	if (gpv_reason & IRET_ERROR)
+		return 0;
+
+	BUG_ON((gpv_reason & (MT_ERROR|MF_ERROR)) == 0);
+	mpl = (gpv_reason & SPR_INDEX) >> SPR_MPL_SHIFT;
+	if (mpl == INT_DMA_NOTIFY && !dma_disabled) {
+		/* User is turning on DMA. Allow it and retry. */
+		printk(KERN_DEBUG "Process %d/%s is now enabled for DMA\n",
+		       current->pid, current->comm);
+		BUG_ON(current->thread.tile_dma_state.enabled);
+		current->thread.tile_dma_state.enabled = 1;
+		grant_dma_mpls();
+		return 1;
+	}
+
+	return 0;
+}
+
+#endif /* CHIP_HAS_TILE_DMA() */
+
+#ifdef __tilegx__
+#define bundle_bits tilegx_bundle_bits
+#else
+#define bundle_bits tile_bundle_bits
+#endif
+
+extern bundle_bits bpt_code;
+
+asm(".pushsection .rodata.bpt_code,\"a\";"
+    ".align 8;"
+    "bpt_code: bpt;"
+    ".size bpt_code,.-bpt_code;"
+    ".popsection");
+
+static int special_ill(bundle_bits bundle, int *sigp, int *codep)
+{
+	int sig, code, maxcode;
+
+	if (bundle == bpt_code) {
+		*sigp = SIGTRAP;
+		*codep = TRAP_BRKPT;
+		return 1;
+	}
+
+	/* If it's a "raise" bundle, then "ill" must be in pipe X1. */
+#ifdef __tilegx__
+	if ((bundle & TILEGX_BUNDLE_MODE_MASK) != 0)
+		return 0;
+	if (get_Opcode_X1(bundle) != RRR_0_OPCODE_X1)
+		return 0;
+	if (get_RRROpcodeExtension_X1(bundle) != UNARY_RRR_0_OPCODE_X1)
+		return 0;
+	if (get_UnaryOpcodeExtension_X1(bundle) != ILL_UNARY_OPCODE_X1)
+		return 0;
+#else
+	if (bundle & TILEPRO_BUNDLE_Y_ENCODING_MASK)
+		return 0;
+	if (get_Opcode_X1(bundle) != SHUN_0_OPCODE_X1)
+		return 0;
+	if (get_UnShOpcodeExtension_X1(bundle) != UN_0_SHUN_0_OPCODE_X1)
+		return 0;
+	if (get_UnOpcodeExtension_X1(bundle) != ILL_UN_0_SHUN_0_OPCODE_X1)
+		return 0;
+#endif
+
+	/* Check that the magic distinguishers are set to mean "raise". */
+	if (get_Dest_X1(bundle) != 29 || get_SrcA_X1(bundle) != 37)
+		return 0;
+
+	/* There must be an "addli zero, zero, VAL" in X0. */
+	if (get_Opcode_X0(bundle) != ADDLI_OPCODE_X0)
+		return 0;
+	if (get_Dest_X0(bundle) != TREG_ZERO)
+		return 0;
+	if (get_SrcA_X0(bundle) != TREG_ZERO)
+		return 0;
+
+	/*
+	 * Validate the proposed signal number and si_code value.
+	 * Note that we embed these in the static instruction itself
+	 * so that we perturb the register state as little as possible
+	 * at the time of the actual fault; it's unlikely you'd ever
+	 * need to dynamically choose which kind of fault to raise
+	 * from user space.
+	 */
+	sig = get_Imm16_X0(bundle) & 0x3f;
+	switch (sig) {
+	case SIGILL:
+		maxcode = NSIGILL;
+		break;
+	case SIGFPE:
+		maxcode = NSIGFPE;
+		break;
+	case SIGSEGV:
+		maxcode = NSIGSEGV;
+		break;
+	case SIGBUS:
+		maxcode = NSIGBUS;
+		break;
+	case SIGTRAP:
+		maxcode = NSIGTRAP;
+		break;
+	default:
+		return 0;
+	}
+	code = (get_Imm16_X0(bundle) >> 6) & 0xf;
+	if (code <= 0 || code > maxcode)
+		return 0;
+
+	/* Make it the requested signal. */
+	*sigp = sig;
+	*codep = code | __SI_FAULT;
+	return 1;
+}
+
+static const char *const int_name[] = {
+	[INT_MEM_ERROR] = "Memory error",
+	[INT_ILL] = "Illegal instruction",
+	[INT_GPV] = "General protection violation",
+	[INT_UDN_ACCESS] = "UDN access",
+	[INT_IDN_ACCESS] = "IDN access",
+#if CHIP_HAS_SN()
+	[INT_SN_ACCESS] = "SN access",
+#endif
+	[INT_SWINT_3] = "Software interrupt 3",
+	[INT_SWINT_2] = "Software interrupt 2",
+	[INT_SWINT_0] = "Software interrupt 0",
+	[INT_UNALIGN_DATA] = "Unaligned data",
+	[INT_DOUBLE_FAULT] = "Double fault",
+#ifdef __tilegx__
+	[INT_ILL_TRANS] = "Illegal virtual address",
+#endif
+};
+
+void __kprobes do_trap(struct pt_regs *regs, int fault_num,
+		       unsigned long reason)
+{
+	siginfo_t info = { 0 };
+	int signo, code;
+	unsigned long address = 0;
+	bundle_bits instr;
+
+	/* Re-enable interrupts. */
+	local_irq_enable();
+
+	/*
+	 * If it hits in kernel mode and we can't fix it up, just exit the
+	 * current process and hope for the best.
+	 */
+	if (!user_mode(regs)) {
+		const char *name;
+		if (fixup_exception(regs))  /* only UNALIGN_DATA in practice */
+			return;
+		if (fault_num >= 0 &&
+		    fault_num < sizeof(int_name)/sizeof(int_name[0]) &&
+		    int_name[fault_num] != NULL)
+			name = int_name[fault_num];
+		else
+			name = "Unknown interrupt";
+		pr_alert("Kernel took bad trap %d (%s) at PC %#lx\n",
+			 fault_num, name, regs->pc);
+		if (fault_num == INT_GPV)
+			pr_alert("GPV_REASON is %#lx\n", reason);
+		show_regs(regs);
+		do_exit(SIGKILL);  /* FIXME: implement i386 die() */
+		return;
+	}
+
+	switch (fault_num) {
+	case INT_MEM_ERROR:
+		signo = SIGBUS;
+		code = BUS_OBJERR;
+		break;
+	case INT_ILL:
+		if (copy_from_user(&instr, (void __user *)regs->pc,
+				   sizeof(instr))) {
+			pr_err("Unreadable instruction for INT_ILL:"
+			       " %#lx\n", regs->pc);
+			do_exit(SIGKILL);
+			return;
+		}
+		if (!special_ill(instr, &signo, &code)) {
+			signo = SIGILL;
+			code = ILL_ILLOPC;
+		}
+		address = regs->pc;
+		break;
+	case INT_GPV:
+#if CHIP_HAS_TILE_DMA()
+		if (retry_gpv(reason))
+			return;
+#endif
+		/*FALLTHROUGH*/
+	case INT_UDN_ACCESS:
+	case INT_IDN_ACCESS:
+#if CHIP_HAS_SN()
+	case INT_SN_ACCESS:
+#endif
+		signo = SIGILL;
+		code = ILL_PRVREG;
+		address = regs->pc;
+		break;
+	case INT_SWINT_3:
+	case INT_SWINT_2:
+	case INT_SWINT_0:
+		signo = SIGILL;
+		code = ILL_ILLTRP;
+		address = regs->pc;
+		break;
+	case INT_UNALIGN_DATA:
+#ifndef __tilegx__  /* Emulated support for single step debugging */
+		if (unaligned_fixup >= 0) {
+			struct single_step_state *state =
+				current_thread_info()->step_state;
+			if (!state ||
+			    (void __user *)(regs->pc) != state->buffer) {
+				single_step_once(regs);
+				return;
+			}
+		}
+#endif
+		signo = SIGBUS;
+		code = BUS_ADRALN;
+		address = 0;
+		break;
+	case INT_DOUBLE_FAULT:
+		/*
+		 * For double fault, "reason" is actually passed as
+		 * SYSTEM_SAVE_K_2, the hypervisor's double-fault info, so
+		 * we can provide the original fault number rather than
+		 * the uninteresting "INT_DOUBLE_FAULT" so the user can
+		 * learn what actually struck while PL0 ICS was set.
+		 */
+		fault_num = reason;
+		signo = SIGILL;
+		code = ILL_DBLFLT;
+		address = regs->pc;
+		break;
+#ifdef __tilegx__
+	case INT_ILL_TRANS: {
+		/* Avoid a hardware erratum with the return address stack. */
+		fill_ra_stack();
+
+		signo = SIGSEGV;
+		code = SEGV_MAPERR;
+		if (reason & SPR_ILL_TRANS_REASON__I_STREAM_VA_RMASK)
+			address = regs->pc;
+		else
+			address = 0;  /* FIXME: GX: single-step for address */
+		break;
+	}
+#endif
+	default:
+		panic("Unexpected do_trap interrupt number %d", fault_num);
+		return;
+	}
+
+	info.si_signo = signo;
+	info.si_code = code;
+	info.si_addr = (void __user *)address;
+	if (signo == SIGILL)
+		info.si_trapno = fault_num;
+	if (signo != SIGTRAP)
+		trace_unhandled_signal("trap", regs, address, signo);
+	force_sig_info(signo, &info, current);
+}
+
+void kernel_double_fault(int dummy, ulong pc, ulong lr, ulong sp, ulong r52)
+{
+	_dump_stack(dummy, pc, lr, sp, r52);
+	pr_emerg("Double fault: exiting\n");
+	machine_halt();
+}

arch/tile/kernel/usb.c

@@ -0,0 +1,69 @@
+/*
+ * Copyright 2012 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * Register the Tile-Gx USB interfaces as platform devices.
+ *
+ * The actual USB driver is just some glue (in
+ * drivers/usb/host/[eo]hci-tilegx.c) which makes the registers available
+ * to the standard kernel EHCI and OHCI drivers.
+ */
+
+#include <linux/dma-mapping.h>
+#include <linux/platform_device.h>
+#include <linux/usb/tilegx.h>
+#include <linux/types.h>
+
+static u64 ehci_dmamask = DMA_BIT_MASK(32);
+
+#define USB_HOST_DEF(unit, type, dmamask) \
+	static struct \
+	    tilegx_usb_platform_data tilegx_usb_platform_data_ ## type ## \
+		hci ## unit = { \
+		.dev_index = unit, \
+	}; \
+	\
+	static struct platform_device tilegx_usb_ ## type ## hci ## unit = { \
+		.name		= "tilegx-" #type "hci", \
+		.id		= unit, \
+		.dev = { \
+			.dma_mask		= dmamask, \
+			.coherent_dma_mask	= DMA_BIT_MASK(32), \
+			.platform_data = \
+				&tilegx_usb_platform_data_ ## type ## hci ## \
+				unit, \
+		}, \
+	};
+
+USB_HOST_DEF(0, e, &ehci_dmamask)
+USB_HOST_DEF(0, o, NULL)
+USB_HOST_DEF(1, e, &ehci_dmamask)
+USB_HOST_DEF(1, o, NULL)
+
+#undef USB_HOST_DEF
+
+static struct platform_device *tilegx_usb_devices[] __initdata = {
+	&tilegx_usb_ehci0,
+	&tilegx_usb_ehci1,
+	&tilegx_usb_ohci0,
+	&tilegx_usb_ohci1,
+};
+
+/** Add our set of possible USB devices. */
+static int __init tilegx_usb_init(void)
+{
+	platform_add_devices(tilegx_usb_devices,
+			     ARRAY_SIZE(tilegx_usb_devices));
+
+	return 0;
+}
+arch_initcall(tilegx_usb_init);

arch/tile/kernel/vdso.c

@@ -0,0 +1,212 @@
+/*
+ * Copyright 2012 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/binfmts.h>
+#include <linux/compat.h>
+#include <linux/elf.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+
+#include <asm/vdso.h>
+#include <asm/mman.h>
+#include <asm/sections.h>
+
+#include <arch/sim.h>
+
+/* The alignment of the vDSO. */
+#define VDSO_ALIGNMENT  PAGE_SIZE
+
+
+static unsigned int vdso_pages;
+static struct page **vdso_pagelist;
+
+#ifdef CONFIG_COMPAT
+static unsigned int vdso32_pages;
+static struct page **vdso32_pagelist;
+#endif
+static int vdso_ready;
+
+/*
+ * The vdso data page.
+ */
+static union {
+	struct vdso_data	data;
+	u8			page[PAGE_SIZE];
+} vdso_data_store __page_aligned_data;
+
+struct vdso_data *vdso_data = &vdso_data_store.data;
+
+static unsigned int __read_mostly vdso_enabled = 1;
+
+static struct page **vdso_setup(void *vdso_kbase, unsigned int pages)
+{
+	int i;
+	struct page **pagelist;
+
+	pagelist = kzalloc(sizeof(struct page *) * (pages + 1), GFP_KERNEL);
+	BUG_ON(pagelist == NULL);
+	for (i = 0; i < pages - 1; i++) {
+		struct page *pg = virt_to_page(vdso_kbase + i*PAGE_SIZE);
+		ClearPageReserved(pg);
+		pagelist[i] = pg;
+	}
+	pagelist[pages - 1] = virt_to_page(vdso_data);
+	pagelist[pages] = NULL;
+
+	return pagelist;
+}
+
+static int __init vdso_init(void)
+{
+	int data_pages = sizeof(vdso_data_store) >> PAGE_SHIFT;
+
+	/*
+	 * We can disable vDSO support generally, but we need to retain
+	 * one page to support the two-bundle (16-byte) rt_sigreturn path.
+	 */
+	if (!vdso_enabled) {
+		size_t offset = (unsigned long)&__vdso_rt_sigreturn;
+		static struct page *sigret_page;
+		sigret_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+		BUG_ON(sigret_page == NULL);
+		vdso_pagelist = &sigret_page;
+		vdso_pages = 1;
+		BUG_ON(offset >= PAGE_SIZE);
+		memcpy(page_address(sigret_page) + offset,
+		       vdso_start + offset, 16);
+#ifdef CONFIG_COMPAT
+		vdso32_pages = vdso_pages;
+		vdso32_pagelist = vdso_pagelist;
+#endif
+		vdso_ready = 1;
+		return 0;
+	}
+
+	vdso_pages = (vdso_end - vdso_start) >> PAGE_SHIFT;
+	vdso_pages += data_pages;
+	vdso_pagelist = vdso_setup(vdso_start, vdso_pages);
+
+#ifdef CONFIG_COMPAT
+	vdso32_pages = (vdso32_end - vdso32_start) >> PAGE_SHIFT;
+	vdso32_pages += data_pages;
+	vdso32_pagelist = vdso_setup(vdso32_start, vdso32_pages);
+#endif
+
+	smp_wmb();
+	vdso_ready = 1;
+
+	return 0;
+}
+arch_initcall(vdso_init);
+
+const char *arch_vma_name(struct vm_area_struct *vma)
+{
+	if (vma->vm_mm && vma->vm_start == VDSO_BASE)
+		return "[vdso]";
+#ifndef __tilegx__
+	if (vma->vm_start == MEM_USER_INTRPT)
+		return "[intrpt]";
+#endif
+	return NULL;
+}
+
+struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
+{
+	return NULL;
+}
+
+int in_gate_area(struct mm_struct *mm, unsigned long address)
+{
+	return 0;
+}
+
+int in_gate_area_no_mm(unsigned long address)
+{
+	return 0;
+}
+
+int setup_vdso_pages(void)
+{
+	struct page **pagelist;
+	unsigned long pages;
+	struct mm_struct *mm = current->mm;
+	unsigned long vdso_base = 0;
+	int retval = 0;
+
+	if (!vdso_ready)
+		return 0;
+
+	mm->context.vdso_base = 0;
+
+	pagelist = vdso_pagelist;
+	pages = vdso_pages;
+#ifdef CONFIG_COMPAT
+	if (is_compat_task()) {
+		pagelist = vdso32_pagelist;
+		pages = vdso32_pages;
+	}
+#endif
+
+	/*
+	 * vDSO has a problem and was disabled, just don't "enable" it for the
+	 * process.
+	 */
+	if (pages == 0)
+		return 0;
+
+	vdso_base = get_unmapped_area(NULL, vdso_base,
+				      (pages << PAGE_SHIFT) +
+				      ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
+				      0, 0);
+	if (IS_ERR_VALUE(vdso_base)) {
+		retval = vdso_base;
+		return retval;
+	}
+
+	/* Add required alignment. */
+	vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
+
+	/*
+	 * Put vDSO base into mm struct. We need to do this before calling
+	 * install_special_mapping or the perf counter mmap tracking code
+	 * will fail to recognise it as a vDSO (since arch_vma_name fails).
+	 */
+	mm->context.vdso_base = vdso_base;
+
+	/*
+	 * our vma flags don't have VM_WRITE so by default, the process isn't
+	 * allowed to write those pages.
+	 * gdb can break that with ptrace interface, and thus trigger COW on
+	 * those pages but it's then your responsibility to never do that on
+	 * the "data" page of the vDSO or you'll stop getting kernel updates
+	 * and your nice userland gettimeofday will be totally dead.
+	 * It's fine to use that for setting breakpoints in the vDSO code
+	 * pages though
+	 */
+	retval = install_special_mapping(mm, vdso_base,
+					 pages << PAGE_SHIFT,
+					 VM_READ|VM_EXEC |
+					 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC,
+					 pagelist);
+	if (retval)
+		mm->context.vdso_base = 0;
+
+	return retval;
+}
+
+static __init int vdso_func(char *s)
+{
+	return kstrtouint(s, 0, &vdso_enabled);
+}
+__setup("vdso=", vdso_func);

arch/tile/kernel/vdso/Makefile

@@ -0,0 +1,118 @@
+# Symbols present in the vdso
+vdso-syms = rt_sigreturn gettimeofday
+
+# Files to link into the vdso
+obj-vdso = $(patsubst %, v%.o, $(vdso-syms))
+
+# Build rules
+targets := $(obj-vdso) vdso.so vdso.so.dbg vdso.lds
+obj-vdso := $(addprefix $(obj)/, $(obj-vdso))
+
+# vdso32 is only for tilegx -m32 compat task.
+VDSO32-$(CONFIG_COMPAT) := y
+
+obj-y += vdso.o
+obj-$(VDSO32-y) += vdso32.o
+extra-y += vdso.lds
+CPPFLAGS_vdso.lds += -P -C -U$(ARCH)
+
+# vDSO code runs in userspace and -pg doesn't help with profiling anyway.
+CFLAGS_REMOVE_vdso.o = -pg
+CFLAGS_REMOVE_vdso32.o = -pg
+CFLAGS_REMOVE_vrt_sigreturn.o = -pg
+CFLAGS_REMOVE_vrt_sigreturn32.o = -pg
+CFLAGS_REMOVE_vgettimeofday.o = -pg
+CFLAGS_REMOVE_vgettimeofday32.o = -pg
+
+ifdef CONFIG_FEEDBACK_COLLECT
+# vDSO code runs in userspace, not collecting feedback data.
+CFLAGS_REMOVE_vdso.o = -ffeedback-generate
+CFLAGS_REMOVE_vdso32.o = -ffeedback-generate
+CFLAGS_REMOVE_vrt_sigreturn.o = -ffeedback-generate
+CFLAGS_REMOVE_vrt_sigreturn32.o = -ffeedback-generate
+CFLAGS_REMOVE_vgettimeofday.o = -ffeedback-generate
+CFLAGS_REMOVE_vgettimeofday32.o = -ffeedback-generate
+endif
+
+# Disable gcov profiling for VDSO code
+GCOV_PROFILE := n
+
+# Force dependency
+$(obj)/vdso.o: $(obj)/vdso.so
+
+# link rule for the .so file, .lds has to be first
+SYSCFLAGS_vdso.so.dbg = $(c_flags)
+$(obj)/vdso.so.dbg: $(src)/vdso.lds $(obj-vdso)
+	$(call if_changed,vdsold)
+
+
+# We also create a special relocatable object that should mirror the symbol
+# table and layout of the linked DSO.  With ld -R we can then refer to
+# these symbols in the kernel code rather than hand-coded addresses.
+extra-y += vdso-syms.o
+$(obj)/built-in.o: $(obj)/vdso-syms.o
+$(obj)/built-in.o: ld_flags += -R $(obj)/vdso-syms.o
+
+SYSCFLAGS_vdso.so.dbg = -shared -s -Wl,-soname=linux-vdso.so.1 \
+                            $(call cc-ldoption, -Wl$(comma)--hash-style=sysv)
+SYSCFLAGS_vdso_syms.o = -r
+$(obj)/vdso-syms.o: $(src)/vdso.lds $(obj)/vrt_sigreturn.o FORCE
+	$(call if_changed,vdsold)
+
+
+# strip rule for the .so file
+$(obj)/%.so: OBJCOPYFLAGS := -S
+$(obj)/%.so: $(obj)/%.so.dbg FORCE
+	$(call if_changed,objcopy)
+
+# actual build commands
+# The DSO images are built using a special linker script
+# Add -lgcc so tilepro gets static muldi3 and lshrdi3 definitions.
+# Make sure only to export the intended __vdso_xxx symbol offsets.
+quiet_cmd_vdsold = VDSOLD  $@
+      cmd_vdsold = $(CC) $(KCFLAGS) -nostdlib $(SYSCFLAGS_$(@F)) \
+                           -Wl,-T,$(filter-out FORCE,$^) -o $@.tmp -lgcc && \
+                   $(CROSS_COMPILE)objcopy \
+                           $(patsubst %, -G __vdso_%, $(vdso-syms)) $@.tmp $@
+
+# install commands for the unstripped file
+quiet_cmd_vdso_install = INSTALL $@
+      cmd_vdso_install = cp $(obj)/$@.dbg $(MODLIB)/vdso/$@
+
+vdso.so: $(obj)/vdso.so.dbg
+	@mkdir -p $(MODLIB)/vdso
+	$(call cmd,vdso_install)
+
+vdso32.so: $(obj)/vdso32.so.dbg
+	$(call cmd,vdso_install)
+
+vdso_install: vdso.so
+vdso32_install: vdso32.so
+
+
+KBUILD_AFLAGS_32 := $(filter-out -m64,$(KBUILD_AFLAGS))
+KBUILD_AFLAGS_32 += -m32 -s
+KBUILD_CFLAGS_32 := $(filter-out -m64,$(KBUILD_CFLAGS))
+KBUILD_CFLAGS_32 += -m32 -fPIC -shared
+
+obj-vdso32 = $(patsubst %, v%32.o, $(vdso-syms))
+obj-vdso32 := $(addprefix $(obj)/, $(obj-vdso32))
+
+targets += $(obj-vdso32) vdso32.so vdso32.so.dbg
+
+$(obj-vdso32:%=%): KBUILD_AFLAGS = $(KBUILD_AFLAGS_32)
+$(obj-vdso32:%=%): KBUILD_CFLAGS = $(KBUILD_CFLAGS_32)
+
+$(obj)/vgettimeofday32.o: $(obj)/vgettimeofday.c
+	$(call if_changed,cc_o_c)
+
+$(obj)/vrt_sigreturn32.o: $(obj)/vrt_sigreturn.S
+	$(call if_changed,as_o_S)
+
+# Force dependency
+$(obj)/vdso32.o: $(obj)/vdso32.so
+
+SYSCFLAGS_vdso32.so.dbg = -m32 -shared -s -Wl,-soname=linux-vdso32.so.1 \
+			    $(call cc-ldoption, -Wl$(comma)--hash-style=sysv)
+$(obj)/vdso32.so.dbg: $(src)/vdso.lds $(obj-vdso32)
+	$(call if_changed,vdsold)

arch/tile/kernel/vdso/vdso.S

@@ -0,0 +1,28 @@
+/*
+ * Copyright 2012 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/init.h>
+#include <linux/linkage.h>
+#include <asm/page.h>
+
+	__PAGE_ALIGNED_DATA
+
+	.global vdso_start, vdso_end
+	.align PAGE_SIZE
+vdso_start:
+	.incbin "arch/tile/kernel/vdso/vdso.so"
+	.align PAGE_SIZE
+vdso_end:
+
+	.previous

arch/tile/kernel/vdso/vdso.lds.S

@@ -0,0 +1,87 @@
+/*
+ * Copyright 2012 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#define VDSO_VERSION_STRING	LINUX_2.6
+
+
+OUTPUT_ARCH(tile)
+
+/* The ELF entry point can be used to set the AT_SYSINFO value. */
+ENTRY(__vdso_rt_sigreturn);
+
+
+SECTIONS
+{
+	. = SIZEOF_HEADERS;
+
+	.hash		: { *(.hash) }			:text
+	.gnu.hash	: { *(.gnu.hash) }
+	.dynsym		: { *(.dynsym) }
+	.dynstr		: { *(.dynstr) }
+	.gnu.version	: { *(.gnu.version) }
+	.gnu.version_d	: { *(.gnu.version_d) }
+	.gnu.version_r	: { *(.gnu.version_r) }
+
+	.note		: { *(.note.*) }		:text	:note
+	.dynamic	: { *(.dynamic) }		:text	:dynamic
+
+	.eh_frame_hdr	: { *(.eh_frame_hdr) }		:text	:eh_frame_hdr
+	.eh_frame	: { KEEP (*(.eh_frame)) }	:text
+
+	.rodata	 : { *(.rodata .rodata.* .gnu.linkonce.r.*) }
+
+	/*
+	 * This linker script is used both with -r and with -shared.
+	 * For the layouts to match, we need to skip more than enough
+	 * space for the dynamic symbol table et al. If this amount
+	 * is insufficient, ld -shared will barf. Just increase it here.
+	 */
+	. = 0x1000;
+	.text		: { *(.text .text.*) }		:text
+
+	.data		: {
+		*(.got.plt) *(.got)
+		*(.data .data.* .gnu.linkonce.d.*)
+		*(.dynbss)
+		*(.bss .bss.* .gnu.linkonce.b.*)
+	}
+}
+
+
+/*
+ * We must supply the ELF program headers explicitly to get just one
+ * PT_LOAD segment, and set the flags explicitly to make segments read-only.
+ */
+PHDRS
+{
+	text		PT_LOAD		FLAGS(5) FILEHDR PHDRS;	/* PF_R|PF_X */
+	dynamic		PT_DYNAMIC	FLAGS(4);		/* PF_R */
+	note		PT_NOTE		FLAGS(4);		/* PF_R */
+	eh_frame_hdr	PT_GNU_EH_FRAME;
+}
+
+
+/*
+ * This controls what userland symbols we export from the vDSO.
+ */
+VERSION
+{
+	VDSO_VERSION_STRING {
+	global:
+		__vdso_rt_sigreturn;
+		__vdso_gettimeofday;
+		gettimeofday;
+	local:*;
+	};
+}

arch/tile/kernel/vdso/vdso32.S

@@ -0,0 +1,28 @@
+/*
+ * Copyright 2013 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/init.h>
+#include <linux/linkage.h>
+#include <asm/page.h>
+
+	__PAGE_ALIGNED_DATA
+
+	.global vdso32_start, vdso32_end
+	.align PAGE_SIZE
+vdso32_start:
+	.incbin "arch/tile/kernel/vdso/vdso32.so"
+	.align PAGE_SIZE
+vdso32_end:
+
+	.previous

arch/tile/kernel/vdso/vgettimeofday.c

@@ -0,0 +1,107 @@
+/*
+ * Copyright 2012 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#define VDSO_BUILD  /* avoid some shift warnings for -m32 in <asm/page.h> */
+#include <linux/time.h>
+#include <asm/timex.h>
+#include <asm/vdso.h>
+
+#if CHIP_HAS_SPLIT_CYCLE()
+static inline cycles_t get_cycles_inline(void)
+{
+	unsigned int high = __insn_mfspr(SPR_CYCLE_HIGH);
+	unsigned int low = __insn_mfspr(SPR_CYCLE_LOW);
+	unsigned int high2 = __insn_mfspr(SPR_CYCLE_HIGH);
+
+	while (unlikely(high != high2)) {
+		low = __insn_mfspr(SPR_CYCLE_LOW);
+		high = high2;
+		high2 = __insn_mfspr(SPR_CYCLE_HIGH);
+	}
+
+	return (((cycles_t)high) << 32) | low;
+}
+#define get_cycles get_cycles_inline
+#endif
+
+/*
+ * Find out the vDSO data page address in the process address space.
+ */
+inline unsigned long get_datapage(void)
+{
+	unsigned long ret;
+
+	/* vdso data page located in the 2nd vDSO page. */
+	asm volatile ("lnk %0" : "=r"(ret));
+	ret &= ~(PAGE_SIZE - 1);
+	ret += PAGE_SIZE;
+
+	return ret;
+}
+
+int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
+{
+	cycles_t cycles;
+	unsigned long count, sec, ns;
+	volatile struct vdso_data *vdso_data;
+
+	vdso_data = (struct vdso_data *)get_datapage();
+	/* The use of the timezone is obsolete, normally tz is NULL. */
+	if (unlikely(tz != NULL)) {
+		while (1) {
+			/* Spin until the update finish. */
+			count = vdso_data->tz_update_count;
+			if (count & 1)
+				continue;
+
+			tz->tz_minuteswest = vdso_data->tz_minuteswest;
+			tz->tz_dsttime = vdso_data->tz_dsttime;
+
+			/* Check whether updated, read again if so. */
+			if (count == vdso_data->tz_update_count)
+				break;
+		}
+	}
+
+	if (unlikely(tv == NULL))
+		return 0;
+
+	while (1) {
+		/* Spin until the update finish. */
+		count = vdso_data->tb_update_count;
+		if (count & 1)
+			continue;
+
+		cycles = (get_cycles() - vdso_data->xtime_tod_stamp);
+		ns = (cycles * vdso_data->mult) >> vdso_data->shift;
+		sec = vdso_data->xtime_clock_sec;
+		ns += vdso_data->xtime_clock_nsec;
+		if (ns >= NSEC_PER_SEC) {
+			ns -= NSEC_PER_SEC;
+			sec += 1;
+		}
+
+		/* Check whether updated, read again if so. */
+		if (count == vdso_data->tb_update_count)
+			break;
+	}
+
+	tv->tv_sec = sec;
+	tv->tv_usec = ns / 1000;
+
+	return 0;
+}
+
+int gettimeofday(struct timeval *tv, struct timezone *tz)
+	__attribute__((weak, alias("__vdso_gettimeofday")));

arch/tile/kernel/vdso/vrt_sigreturn.S

@@ -0,0 +1,30 @@
+/*
+ * Copyright 2012 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/linkage.h>
+#include <arch/abi.h>
+#include <asm/unistd.h>
+
+/*
+ * Note that libc has a copy of this function that it uses to compare
+ * against the PC when a stack backtrace ends, so if this code is
+ * changed, the libc implementation(s) should also be updated.
+ */
+ENTRY(__vdso_rt_sigreturn)
+	moveli TREG_SYSCALL_NR_NAME, __NR_rt_sigreturn
+	swint1
+	/* We don't use ENDPROC to avoid tagging this symbol as FUNC,
+	 * which confuses the perf tool.
+	 */
+	END(__vdso_rt_sigreturn)

arch/tile/kernel/vmlinux.lds.S

@@ -0,0 +1,95 @@
+#include <asm-generic/vmlinux.lds.h>
+#include <asm/page.h>
+#include <asm/cache.h>
+#include <asm/thread_info.h>
+#include <hv/hypervisor.h>
+
+/* Text loads starting from the supervisor interrupt vector address. */
+#define TEXT_OFFSET MEM_SV_INTRPT
+
+OUTPUT_ARCH(tile)
+ENTRY(_start)
+jiffies = jiffies_64;
+
+PHDRS
+{
+  intrpt1 PT_LOAD ;
+  text PT_LOAD ;
+  data PT_LOAD ;
+}
+SECTIONS
+{
+  /* Text is loaded with a different VA than data; start with text. */
+  #undef LOAD_OFFSET
+  #define LOAD_OFFSET TEXT_OFFSET
+
+  /* Interrupt vectors */
+  .intrpt1 (LOAD_OFFSET) : AT ( 0 )   /* put at the start of physical memory */
+  {
+    _text = .;
+    _stext = .;
+    *(.intrpt1)
+  } :intrpt1 =0
+
+  /* Hypervisor call vectors */
+  #include "hvglue.lds"
+
+  /* Now the real code */
+  . = ALIGN(0x20000);
+  .text : AT (ADDR(.text) - LOAD_OFFSET) {
+    HEAD_TEXT
+    SCHED_TEXT
+    LOCK_TEXT
+    __fix_text_end = .;   /* tile-cpack won't rearrange before this */
+    TEXT_TEXT
+    *(.text.*)
+    *(.coldtext*)
+    *(.fixup)
+    *(.gnu.warning)
+  } :text =0
+  _etext = .;
+
+  /* "Init" is divided into two areas with very different virtual addresses. */
+  INIT_TEXT_SECTION(PAGE_SIZE)
+
+  /* Now we skip back to PAGE_OFFSET for the data. */
+  . = (. - TEXT_OFFSET + PAGE_OFFSET);
+  #undef LOAD_OFFSET
+  #define LOAD_OFFSET PAGE_OFFSET
+
+  . = ALIGN(PAGE_SIZE);
+  VMLINUX_SYMBOL(_sinitdata) = .;
+  INIT_DATA_SECTION(16) :data =0
+  PERCPU_SECTION(L2_CACHE_BYTES)
+  . = ALIGN(PAGE_SIZE);
+  VMLINUX_SYMBOL(_einitdata) = .;
+
+  _sdata = .;                   /* Start of data section */
+
+  RO_DATA_SECTION(PAGE_SIZE)
+
+  /* initially writeable, then read-only */
+  . = ALIGN(PAGE_SIZE);
+  __w1data_begin = .;
+  .w1data : AT(ADDR(.w1data) - LOAD_OFFSET) {
+    VMLINUX_SYMBOL(__w1data_begin) = .;
+    *(.w1data)
+    VMLINUX_SYMBOL(__w1data_end) = .;
+  }
+
+  RW_DATA_SECTION(L2_CACHE_BYTES, PAGE_SIZE, THREAD_SIZE)
+
+  _edata = .;
+
+  EXCEPTION_TABLE(L2_CACHE_BYTES)
+  NOTES
+
+
+  BSS_SECTION(8, PAGE_SIZE, 1)
+  _end = . ;
+
+  STABS_DEBUG
+  DWARF_DEBUG
+
+  DISCARDS
+}