From: Mathieu Desnoyers [bunk@stusta.de: marker exports must be EXPORT_SYMBOL_GPL] Signed-off-by: Mathieu Desnoyers Cc: Andi Kleen Signed-off-by: Andrew Morton --- arch/i386/kernel/Makefile | 1 arch/i386/kernel/marker.c | 94 ++++++++++++++++++++++++++++++++++++ include/asm-i386/marker.h | 63 ++++++++++++++++++++++++ 3 files changed, 158 insertions(+) diff -puN arch/i386/kernel/Makefile~linux-kernel-markers-i386-optimization arch/i386/kernel/Makefile --- a/arch/i386/kernel/Makefile~linux-kernel-markers-i386-optimization +++ a/arch/i386/kernel/Makefile @@ -34,6 +34,7 @@ obj-$(CONFIG_MODULES) += module.o obj-y += sysenter.o vsyscall.o obj-$(CONFIG_ACPI_SRAT) += srat.o obj-$(CONFIG_EFI) += efi.o efi_stub.o +obj-$(CONFIG_MARKERS_ENABLE_OPTIMIZATION) += marker.o obj-$(CONFIG_DOUBLEFAULT) += doublefault.o obj-$(CONFIG_VM86) += vm86.o obj-$(CONFIG_EARLY_PRINTK) += early_printk.o diff -puN /dev/null arch/i386/kernel/marker.c --- /dev/null +++ a/arch/i386/kernel/marker.c @@ -0,0 +1,94 @@ +/* marker.c + * + * Erratum 49 fix for Intel PIII and higher. + * + * Permits marker activation by XMC with correct serialization. + * + * Reentrant for NMI and trap handler instrumentation. :-) + * + * Mathieu Desnoyers + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define BREAKPOINT_INSTRUCTION 0xcc +#define BREAKPOINT_INS_LEN 1 + +static DEFINE_MUTEX(mark_mutex); +static long target_eip = 0; + +static void mark_synchronize_core(void *info) +{ + sync_core(); /* use cpuid to stop speculative execution */ +} + +/* We simply skip the 2 bytes load immediate here, leaving the register in an + * undefined state. We don't care about the content (0 or !0), because we are + * changing the value 0->1 or 1->0. This small window of undefined value + * doesn't matter. + */ +static int mark_notifier(struct notifier_block *nb, + unsigned long val, void *data) +{ + enum die_val die_val = (enum die_val) val; + struct die_args *args = (struct die_args *)data; + + if (!args->regs || user_mode_vm(args->regs)) + return NOTIFY_DONE; + + if (die_val == DIE_INT3 && args->regs->eip == target_eip) { + args->regs->eip += 1; /* Skip the next byte of load immediate */ + return NOTIFY_STOP; + } + return NOTIFY_DONE; +} + +static struct notifier_block mark_notify = { + .notifier_call = mark_notifier, + .priority = 0x7fffffff, /* we need to be notified first */ +}; + +int marker_optimized_set_enable(void *address, char enable) +{ + char saved_byte; + int ret; + char *dest = address; + + mutex_lock(&mark_mutex); + BUG_ON(!(enable ^ dest[1])); /* Must be a state change 0<->1 */ + target_eip = (long)address + BREAKPOINT_INS_LEN; + /* register_die_notifier has memory barriers */ + register_die_notifier(&mark_notify); + saved_byte = *dest; + *dest = BREAKPOINT_INSTRUCTION; + wmb(); + /* Execute serializing instruction on each CPU. + * Acts as a memory barrier. */ + ret = on_each_cpu(mark_synchronize_core, NULL, 1, 1); + BUG_ON(ret != 0); + + dest[1] = enable; + wmb(); + *dest = saved_byte; + /* Wait for all int3 handlers to end + (interrupts are disabled in int3). + This CPU is clearly not in a int3 handler + (not preemptible). + synchronize_sched has memory barriers */ + synchronize_sched(); + unregister_die_notifier(&mark_notify); + /* unregister_die_notifier has memory barriers */ + target_eip = 0; + mutex_unlock(&mark_mutex); + flush_icache_range(address, size); + return 0; +} +EXPORT_SYMBOL_GPL(marker_optimized_set_enable); diff -puN /dev/null include/asm-i386/marker.h --- /dev/null +++ a/include/asm-i386/marker.h @@ -0,0 +1,63 @@ +/* + * marker.h + * + * Code markup for dynamic and static tracing. i386 architecture optimisations. + * + * (C) Copyright 2006 Mathieu Desnoyers + * + * This file is released under the GPLv2. + * See the file COPYING for more details. + */ + + +#ifdef CONFIG_MARKERS + +#define _MF_DEFAULT (_MF_OPTIMIZED | _MF_LOCKDEP | _MF_PRINTK) + +#define MARK_OPTIMIZED(flags, name, format, args...) \ + do { \ + static marker_probe_func *__mark_call_##name = \ + __mark_empty_function; \ + static const struct __mark_marker_c __mark_c_##name \ + __attribute__((section(".markers.c"))) = \ + { #name, &__mark_call_##name, format, \ + (flags) | _MF_OPTIMIZED } ; \ + char condition; \ + asm volatile( ".section .markers, \"a\", @progbits;\n\t" \ + ".long %1, 0f;\n\t" \ + ".previous;\n\t" \ + ".align 2\n\t" \ + "0:\n\t" \ + "movb $0,%0;\n\t" \ + : "=r" (condition) \ + : "m" (__mark_c_##name)); \ + __mark_check_format(format, ## args); \ + if (unlikely(condition)) { \ + preempt_disable(); \ + (*__mark_call_##name)(format, ## args); \ + preempt_enable(); \ + } \ + } while (0) + +#define _MARK(flags, format, args...) \ +do { \ + if (((flags) & _MF_LOCKDEP) && ((flags) & _MF_OPTIMIZED)) \ + MARK_OPTIMIZED(flags, format, ## args); \ + else \ + MARK_GENERIC(flags, format, ## args); \ +} while (0) + +#define MARK(format, args...) _MARK(_MF_DEFAULT, format, ## args) + +/* Offset of the immediate value from the start of the movb instruction, in + * bytes. */ +#define MARK_OPTIMIZED_ENABLE_IMMEDIATE_OFFSET 1 +#define MARK_OPTIMIZED_ENABLE_TYPE char +/* Dereference enable as lvalue from a pointer to its instruction */ +#define MARK_OPTIMIZED_ENABLE(a) \ + *(MARK_OPTIMIZED_ENABLE_TYPE*) \ + ((char*)a+MARK_OPTIMIZED_ENABLE_IMMEDIATE_OFFSET) + +extern int marker_optimized_set_enable(void *address, char enable); + +#endif _