From: Jiri Olsa <jolsa@kernel.org>
To: Oleg Nesterov <oleg@redhat.com>,
Peter Zijlstra <peterz@infradead.org>,
Ingo Molnar <mingo@kernel.org>,
Masami Hiramatsu <mhiramat@kernel.org>,
Andrii Nakryiko <andrii@kernel.org>
Cc: bpf@vger.kernel.org, linux-trace-kernel@vger.kernel.org
Subject: [PATCH 1/7] uprobes/x86: Move optimized uprobe from nop5 to nop10
Date: Thu, 14 May 2026 15:53:36 +0200 [thread overview]
Message-ID: <20260514135342.22130-2-jolsa@kernel.org> (raw)
In-Reply-To: <20260514135342.22130-1-jolsa@kernel.org>
Andrii reported an issue with optimized uprobes [1] that can clobber
redzone area with call instruction storing return address on stack
where user code may keep temporary data without adjusting rsp.
Fixing this by moving the optimized uprobes on top of 10-bytes nop
instruction, so we can squeeze another instruction to escape the
redzone area before doing the call, like:
lea -0x80(%rsp), %rsp
call tramp
Note the lea instruction is used to adjust the rsp register without
changing the flags.
The optimized uprobe performance stays the same:
uprobe-nop : 3.129 ± 0.013M/s
uprobe-push : 3.045 ± 0.006M/s
uprobe-ret : 1.095 ± 0.004M/s
--> uprobe-nop10 : 7.170 ± 0.020M/s
uretprobe-nop : 2.143 ± 0.021M/s
uretprobe-push : 2.090 ± 0.000M/s
uretprobe-ret : 0.942 ± 0.000M/s
--> uretprobe-nop10: 3.381 ± 0.003M/s
usdt-nop : 3.245 ± 0.004M/s
--> usdt-nop10 : 7.256 ± 0.023M/s
[1] https://lore.kernel.org/bpf/20260509003146.976844-1-andrii@kernel.org/
Reported-by: Andrii Nakryiko <andrii@kernel.org>
Closes: https://lore.kernel.org/bpf/20260509003146.976844-1-andrii@kernel.org/
Fixes: ba2bfc97b462 ("uprobes/x86: Add support to optimize uprobes")
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
arch/x86/kernel/uprobes.c | 121 +++++++++++++++++++++++++++-----------
1 file changed, 86 insertions(+), 35 deletions(-)
diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c
index ebb1baf1eb1d..f7c4101a4039 100644
--- a/arch/x86/kernel/uprobes.c
+++ b/arch/x86/kernel/uprobes.c
@@ -636,9 +636,21 @@ struct uprobe_trampoline {
unsigned long vaddr;
};
+#define LEA_INSN_SIZE 5
+#define OPT_INSN_SIZE (LEA_INSN_SIZE + CALL_INSN_SIZE)
+#define OPT_JMP8_OFFSET (OPT_INSN_SIZE - JMP8_INSN_SIZE)
+#define REDZONE_SIZE 0x80
+
+static const u8 lea_rsp[] = { 0x48, 0x8d, 0x64, 0x24, 0x80 };
+
+static bool is_lea_insn(const uprobe_opcode_t *insn)
+{
+ return !memcmp(insn, lea_rsp, LEA_INSN_SIZE);
+}
+
static bool is_reachable_by_call(unsigned long vtramp, unsigned long vaddr)
{
- long delta = (long)(vaddr + 5 - vtramp);
+ long delta = (long)(vaddr + OPT_INSN_SIZE - vtramp);
return delta >= INT_MIN && delta <= INT_MAX;
}
@@ -651,7 +663,7 @@ static unsigned long find_nearest_trampoline(unsigned long vaddr)
};
unsigned long low_limit, high_limit;
unsigned long low_tramp, high_tramp;
- unsigned long call_end = vaddr + 5;
+ unsigned long call_end = vaddr + OPT_INSN_SIZE;
if (check_add_overflow(call_end, INT_MIN, &low_limit))
low_limit = PAGE_SIZE;
@@ -826,8 +838,8 @@ SYSCALL_DEFINE0(uprobe)
regs->ax = args.ax;
regs->r11 = args.r11;
regs->cx = args.cx;
- regs->ip = args.retaddr - 5;
- regs->sp += sizeof(args);
+ regs->ip = args.retaddr - OPT_INSN_SIZE;
+ regs->sp += sizeof(args) + REDZONE_SIZE;
regs->orig_ax = -1;
sp = regs->sp;
@@ -844,12 +856,12 @@ SYSCALL_DEFINE0(uprobe)
*/
if (regs->sp != sp) {
/* skip the trampoline call */
- if (args.retaddr - 5 == regs->ip)
- regs->ip += 5;
+ if (args.retaddr - OPT_INSN_SIZE == regs->ip)
+ regs->ip += OPT_INSN_SIZE;
return regs->ax;
}
- regs->sp -= sizeof(args);
+ regs->sp -= sizeof(args) + REDZONE_SIZE;
/* for the case uprobe_consumer has changed ax/r11/cx */
args.ax = regs->ax;
@@ -857,7 +869,7 @@ SYSCALL_DEFINE0(uprobe)
args.cx = regs->cx;
/* keep return address unless we are instructed otherwise */
- if (args.retaddr - 5 != regs->ip)
+ if (args.retaddr - OPT_INSN_SIZE != regs->ip)
args.retaddr = regs->ip;
if (shstk_push(args.retaddr) == -EFAULT)
@@ -891,7 +903,7 @@ asm (
"pop %rax\n"
"pop %r11\n"
"pop %rcx\n"
- "ret\n"
+ "ret $" __stringify(REDZONE_SIZE) "\n"
"int3\n"
".balign " __stringify(PAGE_SIZE) "\n"
".popsection\n"
@@ -909,7 +921,7 @@ late_initcall(arch_uprobes_init);
enum {
EXPECT_SWBP,
- EXPECT_CALL,
+ EXPECT_OPTIMIZED,
};
struct write_opcode_ctx {
@@ -930,17 +942,18 @@ static int verify_insn(struct page *page, unsigned long vaddr, uprobe_opcode_t *
int nbytes, void *data)
{
struct write_opcode_ctx *ctx = data;
- uprobe_opcode_t old_opcode[5];
+ uprobe_opcode_t old_opcode[OPT_INSN_SIZE];
- uprobe_copy_from_page(page, ctx->base, (uprobe_opcode_t *) &old_opcode, 5);
+ uprobe_copy_from_page(page, ctx->base, old_opcode, OPT_INSN_SIZE);
switch (ctx->expect) {
case EXPECT_SWBP:
if (is_swbp_insn(&old_opcode[0]))
return 1;
break;
- case EXPECT_CALL:
- if (is_call_insn(&old_opcode[0]))
+ case EXPECT_OPTIMIZED:
+ if (is_lea_insn(&old_opcode[0]) &&
+ is_call_insn(&old_opcode[LEA_INSN_SIZE]))
return 1;
break;
}
@@ -963,7 +976,7 @@ static int verify_insn(struct page *page, unsigned long vaddr, uprobe_opcode_t *
* - SMP sync all CPUs
*/
static int int3_update(struct arch_uprobe *auprobe, struct vm_area_struct *vma,
- unsigned long vaddr, char *insn, bool optimize)
+ unsigned long vaddr, char *insn, int size, bool optimize)
{
uprobe_opcode_t int3 = UPROBE_SWBP_INSN;
struct write_opcode_ctx ctx = {
@@ -978,7 +991,7 @@ static int int3_update(struct arch_uprobe *auprobe, struct vm_area_struct *vma,
* so we can skip this step for optimize == true.
*/
if (!optimize) {
- ctx.expect = EXPECT_CALL;
+ ctx.expect = EXPECT_OPTIMIZED;
err = uprobe_write(auprobe, vma, vaddr, &int3, 1, verify_insn,
true /* is_register */, false /* do_update_ref_ctr */,
&ctx);
@@ -990,7 +1003,7 @@ static int int3_update(struct arch_uprobe *auprobe, struct vm_area_struct *vma,
/* Write all but the first byte of the patched range. */
ctx.expect = EXPECT_SWBP;
- err = uprobe_write(auprobe, vma, vaddr + 1, insn + 1, 4, verify_insn,
+ err = uprobe_write(auprobe, vma, vaddr + 1, insn + 1, size - 1, verify_insn,
true /* is_register */, false /* do_update_ref_ctr */,
&ctx);
if (err)
@@ -1017,17 +1030,32 @@ static int int3_update(struct arch_uprobe *auprobe, struct vm_area_struct *vma,
static int swbp_optimize(struct arch_uprobe *auprobe, struct vm_area_struct *vma,
unsigned long vaddr, unsigned long tramp)
{
- u8 call[5];
+ u8 insn[OPT_INSN_SIZE], *call = &insn[LEA_INSN_SIZE];
- __text_gen_insn(call, CALL_INSN_OPCODE, (const void *) vaddr,
+ /*
+ * We have nop10 instruction (with first byte overwritten to int3),
+ * changing it to:
+ * lea -0x80(%rsp), %rsp
+ * call tramp
+ */
+ memcpy(insn, lea_rsp, LEA_INSN_SIZE);
+ __text_gen_insn(call, CALL_INSN_OPCODE,
+ (const void *) (vaddr + LEA_INSN_SIZE),
(const void *) tramp, CALL_INSN_SIZE);
- return int3_update(auprobe, vma, vaddr, call, true /* optimize */);
+ return int3_update(auprobe, vma, vaddr, insn, OPT_INSN_SIZE, true /* optimize */);
}
static int swbp_unoptimize(struct arch_uprobe *auprobe, struct vm_area_struct *vma,
unsigned long vaddr)
{
- return int3_update(auprobe, vma, vaddr, auprobe->insn, false /* optimize */);
+ /*
+ * We have optimized nop10 (lea, call), changing it to 'jmp rel8' to
+ * end of the 10-byte slot instead of restoring the original nop10,
+ * because we could have thread already inside lea instruction.
+ */
+ u8 jmp[OPT_INSN_SIZE] = { JMP8_INSN_OPCODE, OPT_JMP8_OFFSET };
+
+ return int3_update(auprobe, vma, vaddr, jmp, JMP8_INSN_SIZE, false /* optimize */);
}
static int copy_from_vaddr(struct mm_struct *mm, unsigned long vaddr, void *dst, int len)
@@ -1049,19 +1077,21 @@ static bool __is_optimized(uprobe_opcode_t *insn, unsigned long vaddr)
struct __packed __arch_relative_insn {
u8 op;
s32 raddr;
- } *call = (struct __arch_relative_insn *) insn;
+ } *call = (struct __arch_relative_insn *)(insn + LEA_INSN_SIZE);
- if (!is_call_insn(insn))
+ if (!is_lea_insn(insn))
+ return false;
+ if (!is_call_insn(insn + LEA_INSN_SIZE))
return false;
- return __in_uprobe_trampoline(vaddr + 5 + call->raddr);
+ return __in_uprobe_trampoline(vaddr + OPT_INSN_SIZE + call->raddr);
}
static int is_optimized(struct mm_struct *mm, unsigned long vaddr)
{
- uprobe_opcode_t insn[5];
+ uprobe_opcode_t insn[OPT_INSN_SIZE];
int err;
- err = copy_from_vaddr(mm, vaddr, &insn, 5);
+ err = copy_from_vaddr(mm, vaddr, &insn, OPT_INSN_SIZE);
if (err)
return err;
return __is_optimized((uprobe_opcode_t *)&insn, vaddr);
@@ -1095,14 +1125,25 @@ int set_orig_insn(struct arch_uprobe *auprobe, struct vm_area_struct *vma,
unsigned long vaddr)
{
if (test_bit(ARCH_UPROBE_FLAG_CAN_OPTIMIZE, &auprobe->flags)) {
- int ret = is_optimized(vma->vm_mm, vaddr);
- if (ret < 0)
+ uprobe_opcode_t insn[OPT_INSN_SIZE];
+ int ret;
+
+ ret = copy_from_vaddr(vma->vm_mm, vaddr, &insn, OPT_INSN_SIZE);
+ if (ret)
return ret;
- if (ret) {
+ if (__is_optimized((uprobe_opcode_t *)&insn, vaddr)) {
ret = swbp_unoptimize(auprobe, vma, vaddr);
WARN_ON_ONCE(ret);
return ret;
}
+ /*
+ * We can have re-attached probe on top of jmp8 instruction,
+ * which did not get optimized. We need to restore the jmp8
+ * instruction, instead of the original instruction (nop10).
+ */
+ if (is_swbp_insn(&insn[0]) && insn[1] == OPT_JMP8_OFFSET)
+ return uprobe_write_opcode(auprobe, vma, vaddr, JMP8_INSN_OPCODE,
+ false /* is_register */);
}
return uprobe_write_opcode(auprobe, vma, vaddr, *(uprobe_opcode_t *)&auprobe->insn,
false /* is_register */);
@@ -1131,7 +1172,7 @@ static int __arch_uprobe_optimize(struct arch_uprobe *auprobe, struct mm_struct
void arch_uprobe_optimize(struct arch_uprobe *auprobe, unsigned long vaddr)
{
struct mm_struct *mm = current->mm;
- uprobe_opcode_t insn[5];
+ uprobe_opcode_t insn[OPT_INSN_SIZE];
if (!should_optimize(auprobe))
return;
@@ -1142,7 +1183,7 @@ void arch_uprobe_optimize(struct arch_uprobe *auprobe, unsigned long vaddr)
* Check if some other thread already optimized the uprobe for us,
* if it's the case just go away silently.
*/
- if (copy_from_vaddr(mm, vaddr, &insn, 5))
+ if (copy_from_vaddr(mm, vaddr, &insn, OPT_INSN_SIZE))
goto unlock;
if (!is_swbp_insn((uprobe_opcode_t*) &insn))
goto unlock;
@@ -1160,14 +1201,24 @@ void arch_uprobe_optimize(struct arch_uprobe *auprobe, unsigned long vaddr)
static bool can_optimize(struct insn *insn, unsigned long vaddr)
{
- if (!insn->x86_64 || insn->length != 5)
+ if (!insn->x86_64)
return false;
- if (!insn_is_nop(insn))
+ /* We can't do cross page atomic writes yet. */
+ if (PAGE_SIZE - (vaddr & ~PAGE_MASK) < OPT_INSN_SIZE)
return false;
- /* We can't do cross page atomic writes yet. */
- return PAGE_SIZE - (vaddr & ~PAGE_MASK) >= 5;
+ /* We can optimize on top of nop10.. */
+ if (insn->length == OPT_INSN_SIZE && insn_is_nop(insn))
+ return true;
+
+ /* .. and JMP rel8 to end of slot — check swbp_unoptimize. */
+ if (insn->length == 2 &&
+ insn->opcode.bytes[0] == JMP8_INSN_OPCODE &&
+ insn->immediate.value == OPT_JMP8_OFFSET)
+ return true;
+
+ return false;
}
#else /* 32-bit: */
/*
--
2.53.0
next prev parent reply other threads:[~2026-05-14 13:53 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-05-14 13:53 [PATCH 0/7] uprobes/x86: Fix red zone issue for optimized uprobes Jiri Olsa
2026-05-14 13:53 ` Jiri Olsa [this message]
2026-05-14 16:54 ` [PATCH 1/7] uprobes/x86: Move optimized uprobe from nop5 to nop10 Jakub Sitnicki
2026-05-14 13:53 ` [PATCH 2/7] libbpf: Change has_nop_combo to work on top of nop10 Jiri Olsa
2026-05-14 14:55 ` bot+bpf-ci
2026-05-14 13:53 ` [PATCH 3/7] selftests/bpf: Emit nop,nop10 instructions combo for x86_64 arch Jiri Olsa
2026-05-14 13:53 ` [PATCH 4/7] selftests/bpf: Change uprobe syscall tests to use nop10 Jiri Olsa
2026-05-14 13:53 ` [PATCH 5/7] selftests/bpf: Change uprobe/usdt trigger bench code " Jiri Olsa
2026-05-14 13:53 ` [PATCH 6/7] selftests/bpf: Add reattach tests for uprobe syscall Jiri Olsa
2026-05-14 13:53 ` [PATCH 7/7] selftests/bpf: Add tests for uprobe nop10 red zone clobbering Jiri Olsa
2026-05-14 14:55 ` bot+bpf-ci
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260514135342.22130-2-jolsa@kernel.org \
--to=jolsa@kernel.org \
--cc=andrii@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=linux-trace-kernel@vger.kernel.org \
--cc=mhiramat@kernel.org \
--cc=mingo@kernel.org \
--cc=oleg@redhat.com \
--cc=peterz@infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox