This is the mail archive of the
systemtap@sources.redhat.com
mailing list for the systemtap project.
[RFC PATCH 6/6] kprobes: use per_cpu data areas - x86_64 changes
x86_64 specific per_cpu changes. Compile tested only.
Signed-off-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
arch/x86_64/kernel/kprobes.c | 142 +++++++++++++++++++++++++------------------
include/asm-x86_64/kprobes.h | 19 +++++
2 files changed, 103 insertions(+), 58 deletions(-)
Index: linux-2.6.13-rc3/arch/x86_64/kernel/kprobes.c
===================================================================
--- linux-2.6.13-rc3.orig/arch/x86_64/kernel/kprobes.c 2005-07-29 12:44:57.000000000 -0400
+++ linux-2.6.13-rc3/arch/x86_64/kernel/kprobes.c 2005-07-29 12:58:35.000000000 -0400
@@ -45,16 +45,26 @@
static DECLARE_MUTEX(kprobe_mutex);
-static struct kprobe *current_kprobe;
-static unsigned long kprobe_status, kprobe_old_rflags, kprobe_saved_rflags;
-static struct kprobe *kprobe_prev;
-static unsigned long kprobe_status_prev, kprobe_old_rflags_prev, kprobe_saved_rflags_prev;
-static struct pt_regs jprobe_saved_regs;
-static long *jprobe_saved_rsp;
void jprobe_return_end(void);
-/* copy of the kernel stack at the probe fire time */
-static kprobe_opcode_t jprobes_stack[MAX_STACK_SIZE];
+static DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
+static DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
+
+/* kprobe_running() will just return the current_kprobe on this CPU */
+static inline struct kprobe *kprobe_running(void)
+{
+ return (__get_cpu_var(current_kprobe));
+}
+
+static inline void reset_current_kprobe(void)
+{
+ __get_cpu_var(current_kprobe) = NULL;
+}
+
+static inline struct kprobe_ctlblk *get_kprobe_ctlblk(void)
+{
+ return (&__get_cpu_var(kprobe_ctlblk));
+}
/*
* returns non-zero if opcode modifies the interrupt flag.
@@ -236,29 +246,30 @@ void __kprobes arch_remove_kprobe(struct
down(&kprobe_mutex);
}
-static inline void save_previous_kprobe(void)
+static inline void save_previous_kprobe(struct kprobe_ctlblk *kcb)
{
- kprobe_prev = current_kprobe;
- kprobe_status_prev = kprobe_status;
- kprobe_old_rflags_prev = kprobe_old_rflags;
- kprobe_saved_rflags_prev = kprobe_saved_rflags;
+ kcb->prev_kprobe.kp = kprobe_running();
+ kcb->prev_kprobe.status = kcb->kprobe_status;
+ kcb->prev_kprobe.old_rflags = kcb->kprobe_old_rflags;
+ kcb->prev_kprobe.saved_rflags = kcb->kprobe_saved_rflags;
}
-static inline void restore_previous_kprobe(void)
+static inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb)
{
- current_kprobe = kprobe_prev;
- kprobe_status = kprobe_status_prev;
- kprobe_old_rflags = kprobe_old_rflags_prev;
- kprobe_saved_rflags = kprobe_saved_rflags_prev;
+ __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
+ kcb->kprobe_status = kcb->prev_kprobe.status;
+ kcb->kprobe_old_rflags = kcb->prev_kprobe.old_rflags;
+ kcb->kprobe_saved_rflags = kcb->prev_kprobe.saved_rflags;
}
-static inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs)
+static inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
+ struct kprobe_ctlblk *kcb)
{
- current_kprobe = p;
- kprobe_saved_rflags = kprobe_old_rflags
+ __get_cpu_var(current_kprobe) = p;
+ kcb->kprobe_saved_rflags = kcb->kprobe_old_rflags
= (regs->eflags & (TF_MASK | IF_MASK));
if (is_IF_modifier(p->ainsn.insn))
- kprobe_saved_rflags &= ~IF_MASK;
+ kcb->kprobe_saved_rflags &= ~IF_MASK;
}
static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
@@ -301,6 +312,7 @@ int __kprobes kprobe_handler(struct pt_r
struct kprobe *p;
int ret = 0;
kprobe_opcode_t *addr = (kprobe_opcode_t *)(regs->rip - sizeof(kprobe_opcode_t));
+ struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
/* Check we're not actually recursing */
if (kprobe_running()) {
@@ -308,12 +320,12 @@ int __kprobes kprobe_handler(struct pt_r
Disarm the probe we just hit, and ignore it. */
p = get_kprobe(addr);
if (p) {
- if (kprobe_status == KPROBE_HIT_SS) {
+ if (kcb->kprobe_status == KPROBE_HIT_SS) {
regs->eflags &= ~TF_MASK;
- regs->eflags |= kprobe_saved_rflags;
+ regs->eflags |= kcb->kprobe_saved_rflags;
unlock_kprobes();
goto no_kprobe;
- } else if (kprobe_status == KPROBE_HIT_SSDONE) {
+ } else if (kcb->kprobe_status == KPROBE_HIT_SSDONE) {
/* TODO: Provide re-entrancy from
* post_kprobes_handler() and avoid exception
* stack corruption while single-stepping on
@@ -321,6 +333,7 @@ int __kprobes kprobe_handler(struct pt_r
*/
arch_disarm_kprobe(p);
regs->rip = (unsigned long)p->addr;
+ reset_current_kprobe();
ret = 1;
} else {
/* We have reentered the kprobe_handler(), since
@@ -330,15 +343,15 @@ int __kprobes kprobe_handler(struct pt_r
* of the new probe without calling any user
* handlers.
*/
- save_previous_kprobe();
- set_current_kprobe(p, regs);
+ save_previous_kprobe(kcb);
+ set_current_kprobe(p, regs, kcb);
p->nmissed++;
prepare_singlestep(p, regs);
- kprobe_status = KPROBE_REENTER;
+ kcb->kprobe_status = KPROBE_REENTER;
return 1;
}
} else {
- p = current_kprobe;
+ p = __get_cpu_var(current_kprobe);
if (p->break_handler && p->break_handler(p, regs)) {
goto ss_probe;
}
@@ -365,8 +378,8 @@ int __kprobes kprobe_handler(struct pt_r
goto no_kprobe;
}
- kprobe_status = KPROBE_HIT_ACTIVE;
- set_current_kprobe(p, regs);
+ set_current_kprobe(p, regs, kcb);
+ kcb->kprobe_status = KPROBE_HIT_ACTIVE;
if (p->pre_handler && p->pre_handler(p, regs))
/* handler has already set things up, so skip ss setup */
@@ -374,7 +387,7 @@ int __kprobes kprobe_handler(struct pt_r
ss_probe:
prepare_singlestep(p, regs);
- kprobe_status = KPROBE_HIT_SS;
+ kcb->kprobe_status = KPROBE_HIT_SS;
/*
* This preempt_disable() matches the preempt_enable_no_resched()
* in post_kprobe_handler()
@@ -480,7 +493,8 @@ int __kprobes trampoline_probe_handler(s
* that is atop the stack is the address following the copied instruction.
* We need to make it the address following the original instruction.
*/
-static void __kprobes resume_execution(struct kprobe *p, struct pt_regs *regs)
+static void __kprobes resume_execution(struct kprobe *p,
+ struct pt_regs *regs, struct kprobe_ctlblk *kcb)
{
unsigned long *tos = (unsigned long *)regs->rsp;
unsigned long next_rip = 0;
@@ -495,7 +509,7 @@ static void __kprobes resume_execution(s
switch (*insn) {
case 0x9c: /* pushfl */
*tos &= ~(TF_MASK | IF_MASK);
- *tos |= kprobe_old_rflags;
+ *tos |= kcb->kprobe_old_rflags;
break;
case 0xc3: /* ret/lret */
case 0xcb:
@@ -540,20 +554,23 @@ static void __kprobes resume_execution(s
*/
int __kprobes post_kprobe_handler(struct pt_regs *regs)
{
- if (!kprobe_running())
+ struct kprobe *cur = kprobe_running();
+ struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+
+ if (!cur)
return 0;
- if ((kprobe_status != KPROBE_REENTER) && current_kprobe->post_handler) {
- kprobe_status = KPROBE_HIT_SSDONE;
- current_kprobe->post_handler(current_kprobe, regs, 0);
+ if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
+ kcb->kprobe_status = KPROBE_HIT_SSDONE;
+ cur->post_handler(cur, regs, 0);
}
- resume_execution(current_kprobe, regs);
- regs->eflags |= kprobe_saved_rflags;
+ resume_execution(cur, regs, kcb);
+ regs->eflags |= kcb->kprobe_saved_rflags;
/* Restore the original saved kprobes variables and continue. */
- if (kprobe_status == KPROBE_REENTER) {
- restore_previous_kprobe();
+ if (kcb->kprobe_status == KPROBE_REENTER) {
+ restore_previous_kprobe(kcb);
goto out;
} else {
unlock_kprobes();
@@ -575,14 +592,17 @@ out:
/* Interrupts disabled, kprobe_lock held. */
int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
{
- if (current_kprobe->fault_handler
- && current_kprobe->fault_handler(current_kprobe, regs, trapnr))
+ struct kprobe *cur = kprobe_running();
+ struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+
+ if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
return 1;
- if (kprobe_status & KPROBE_HIT_SS) {
- resume_execution(current_kprobe, regs);
- regs->eflags |= kprobe_old_rflags;
+ if (kcb->kprobe_status & KPROBE_HIT_SS) {
+ resume_execution(cur, regs, kcb);
+ regs->eflags |= kcb->kprobe_old_rflags;
+ reset_current_kprobe();
unlock_kprobes();
preempt_enable_no_resched();
}
@@ -625,10 +645,11 @@ int __kprobes setjmp_pre_handler(struct
{
struct jprobe *jp = container_of(p, struct jprobe, kp);
unsigned long addr;
+ struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
- jprobe_saved_regs = *regs;
- jprobe_saved_rsp = (long *) regs->rsp;
- addr = (unsigned long)jprobe_saved_rsp;
+ kcb->jprobe_saved_regs = *regs;
+ kcb->jprobe_saved_rsp = (long *) regs->rsp;
+ addr = (unsigned long)(kcb->jprobe_saved_rsp);
/*
* As Linus pointed out, gcc assumes that the callee
* owns the argument space and could overwrite it, e.g.
@@ -636,7 +657,8 @@ int __kprobes setjmp_pre_handler(struct
* we also save and restore enough stack bytes to cover
* the argument area.
*/
- memcpy(jprobes_stack, (kprobe_opcode_t *) addr, MIN_STACK_SIZE(addr));
+ memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
+ MIN_STACK_SIZE(addr));
regs->eflags &= ~IF_MASK;
regs->rip = (unsigned long)(jp->entry);
return 1;
@@ -644,34 +666,38 @@ int __kprobes setjmp_pre_handler(struct
void __kprobes jprobe_return(void)
{
+ struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+
asm volatile (" xchg %%rbx,%%rsp \n"
" int3 \n"
" .globl jprobe_return_end \n"
" jprobe_return_end: \n"
" nop \n"::"b"
- (jprobe_saved_rsp):"memory");
+ (kcb->jprobe_saved_rsp):"memory");
}
int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
{
+ struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
u8 *addr = (u8 *) (regs->rip - 1);
- unsigned long stack_addr = (unsigned long)jprobe_saved_rsp;
+ unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_rsp);
struct jprobe *jp = container_of(p, struct jprobe, kp);
if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
- if ((long *)regs->rsp != jprobe_saved_rsp) {
+ if ((long *)regs->rsp != kcb->jprobe_saved_rsp) {
struct pt_regs *saved_regs =
- container_of(jprobe_saved_rsp, struct pt_regs, rsp);
+ container_of(kcb->jprobe_saved_rsp,
+ struct pt_regs, rsp);
printk("current rsp %p does not match saved rsp %p\n",
- (long *)regs->rsp, jprobe_saved_rsp);
+ (long *)regs->rsp, kcb->jprobe_saved_rsp);
printk("Saved registers for jprobe %p\n", jp);
show_registers(saved_regs);
printk("Current registers\n");
show_registers(regs);
BUG();
}
- *regs = jprobe_saved_regs;
- memcpy((kprobe_opcode_t *) stack_addr, jprobes_stack,
+ *regs = kcb->jprobe_saved_regs;
+ memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
MIN_STACK_SIZE(stack_addr));
return 1;
}
Index: linux-2.6.13-rc3/include/asm-x86_64/kprobes.h
===================================================================
--- linux-2.6.13-rc3.orig/include/asm-x86_64/kprobes.h 2005-07-29 12:44:57.000000000 -0400
+++ linux-2.6.13-rc3/include/asm-x86_64/kprobes.h 2005-07-29 12:58:17.000000000 -0400
@@ -25,6 +25,7 @@
*/
#include <linux/types.h>
#include <linux/ptrace.h>
+#include <linux/percpu.h>
struct pt_regs;
@@ -48,6 +49,24 @@ struct arch_specific_insn {
kprobe_opcode_t *insn;
};
+struct prev_kprobe {
+ struct kprobe *kp;
+ unsigned long status;
+ unsigned long old_rflags;
+ unsigned long saved_rflags;
+};
+
+/* per-cpu kprobe control block */
+struct kprobe_ctlblk {
+ unsigned long kprobe_status;
+ unsigned long kprobe_old_rflags;
+ unsigned long kprobe_saved_rflags;
+ long *jprobe_saved_rsp;
+ struct pt_regs jprobe_saved_regs;
+ kprobe_opcode_t jprobes_stack[MAX_STACK_SIZE];
+ struct prev_kprobe prev_kprobe;
+};
+
/* trap3/1 are intr gates for kprobes. So, restore the status of IF,
* if necessary, before executing the original int3/1 (trap) handler.
*/