* [PATCH 6.12] x86/fgraph: Fix return_to_handler regs.rsp value
@ 2026-05-26 19:23 Gyokhan Kochmarla
2026-05-26 19:50 ` sashiko-bot
2026-05-27 19:49 ` Sasha Levin
0 siblings, 2 replies; 5+ messages in thread
From: Gyokhan Kochmarla @ 2026-05-26 19:23 UTC (permalink / raw)
To: stable, gregkh
Cc: jolsa, rostedt, mhiramat, tglx, mingo, bp, x86,
linux-trace-kernel, bpf, Andrii Nakryiko, Gyokhan Kochmarla
From: Jiri Olsa <jolsa@kernel.org>
commit 8bc11700e0d23d4fdb7d8d5a73b2e95de427cabc upstream.
The previous change (Fixes commit) messed up the rsp register value,
which is wrong because it's already adjusted with FRAME_SIZE, we need
the original rsp value.
This change does not affect fprobe current kernel unwind, the !perf_hw_regs
path perf_callchain_kernel:
if (perf_hw_regs(regs)) {
if (perf_callchain_store(entry, regs->ip))
return;
unwind_start(&state, current, regs, NULL);
} else {
unwind_start(&state, current, NULL, (void *)regs->sp);
}
which uses pt_regs.sp as first_frame boundary (FRAME_SIZE shift makes
no difference, unwind stil stops at the right frame).
This change fixes the other path when we want to unwind directly from
pt_regs sp/fp/ip state, which is coming in following change.
Fixes: 20a0bc10272f ("x86/fgraph,bpf: Fix stack ORC unwind from kprobe_multi return probe")
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Link: https://lore.kernel.org/bpf/20260126211837.472802-2-jolsa@kernel.org
Signed-off-by: Gyokhan Kochmarla <gyokhan@amazon.de>
---
arch/x86/kernel/ftrace_64.S | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/arch/x86/kernel/ftrace_64.S b/arch/x86/kernel/ftrace_64.S
index 8a3cff618692..143fc62bf6f8 100644
--- a/arch/x86/kernel/ftrace_64.S
+++ b/arch/x86/kernel/ftrace_64.S
@@ -349,6 +349,9 @@ SYM_CODE_START(return_to_handler)
UNWIND_HINT_UNDEFINED
ANNOTATE_NOENDBR
+ /* Store original rsp for pt_regs.sp value. */
+ movq %rsp, %rdi
+
/* Restore return_to_handler value that got eaten by previous ret instruction. */
subq $8, %rsp
UNWIND_HINT_FUNC
@@ -359,7 +362,7 @@ SYM_CODE_START(return_to_handler)
movq %rax, RAX(%rsp)
movq %rdx, RDX(%rsp)
movq %rbp, RBP(%rsp)
- movq %rsp, RSP(%rsp)
+ movq %rdi, RSP(%rsp)
movq %rsp, %rdi
call ftrace_return_to_handler
--
2.47.3
Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH 6.12] x86/fgraph: Fix return_to_handler regs.rsp value
2026-05-26 19:23 [PATCH 6.12] x86/fgraph: Fix return_to_handler regs.rsp value Gyokhan Kochmarla
@ 2026-05-26 19:50 ` sashiko-bot
2026-05-27 7:36 ` Jiri Olsa
2026-05-27 19:49 ` Sasha Levin
1 sibling, 1 reply; 5+ messages in thread
From: sashiko-bot @ 2026-05-26 19:50 UTC (permalink / raw)
To: Gyokhan Kochmarla; +Cc: bpf
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
Pre-existing issues:
- [High] The `return_to_handler` assembly routine creates a `pt_regs` (or `ftrace_regs`) structure on the kernel stack but only initializes four registers (`rax`, `rdx`, `rbp`, `rsp`), leaving the rest of the structure uninitialized. This partially initialized structure is passed to BPF kretprobe programs, leading to an information leak of kernel stack memory.
--
commit ff6f1024b92860971d84c833c6dac53501145715
Author: Jiri Olsa <jolsa@kernel.org>
x86/fgraph: Fix return_to_handler regs.rsp value
This commit fixes the unwinding rsp register value in return_to_handler. It
saves the original rsp before the FRAME_SIZE adjustment, replacing the
previous incorrect value. This ensures fprobe and direct unwinders see the
right stack pointer.
Link: https://lore.kernel.org/bpf/20260126211837.472802-2-jolsa@kernel.org
> diff --git a/arch/x86/kernel/ftrace_64.S b/arch/x86/kernel/ftrace_64.S
> --- a/arch/x86/kernel/ftrace_64.S
> +++ b/arch/x86/kernel/ftrace_64.S
> @@ -364,6 +364,9 @@ SYM_CODE_START(return_to_handler)
> UNWIND_HINT_UNDEFINED
> ANNOTATE_NOENDBR
>
> + /* Store original rsp for pt_regs.sp value. */
> + movq %rsp, %rdi
> +
> /* Restore return_to_handler value that got eaten by previous ret instruction. */
> subq $8, %rsp
> UNWIND_HINT_FUNC
> @@ -374,7 +377,7 @@ SYM_CODE_START(return_to_handler)
> movq %rax, RAX(%rsp)
> movq %rdx, RDX(%rsp)
> movq %rbp, RBP(%rsp)
> - movq %rsp, RSP(%rsp)
> + movq %rdi, RSP(%rsp)
[Severity: High]
This is a pre-existing issue, but could this code leak kernel stack memory
to BPF programs?
This assembly creates a pt_regs structure on the stack by allocating
FRAME_SIZE bytes, but only explicitly saves rax, rdx, rbp, and rsp. Later in
ftrace_return_to_handler and arch_ftrace_partial_regs, only a few additional
fields are populated.
The remaining fields (like rdi, rsi, rbx, and r12-r15) remain uninitialized.
Because this partially initialized pt_regs structure is passed to BPF
kprobe_multi return programs via kprobe_multi_link_exit_handler(), and the
BPF verifier allows reading any field of the pt_regs context, does this
allow BPF programs to extract residual kernel stack data?
> movq %rsp, %rdi
>
> call ftrace_return_to_handler
--
Sashiko AI review · https://sashiko.dev/#/patchset/20260526192324.79459-1-gyokhan@amazon.de?part=1
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 6.12] x86/fgraph: Fix return_to_handler regs.rsp value
2026-05-26 19:50 ` sashiko-bot
@ 2026-05-27 7:36 ` Jiri Olsa
2026-05-27 13:39 ` Steven Rostedt
0 siblings, 1 reply; 5+ messages in thread
From: Jiri Olsa @ 2026-05-27 7:36 UTC (permalink / raw)
To: sashiko-reviews; +Cc: Gyokhan Kochmarla, bpf, Masami Hiramatsu, Steven Rostedt
On Tue, May 26, 2026 at 07:50:07PM +0000, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
>
> Pre-existing issues:
> - [High] The `return_to_handler` assembly routine creates a `pt_regs` (or `ftrace_regs`) structure on the kernel stack but only initializes four registers (`rax`, `rdx`, `rbp`, `rsp`), leaving the rest of the structure uninitialized. This partially initialized structure is passed to BPF kretprobe programs, leading to an information leak of kernel stack memory.
> --
>
> commit ff6f1024b92860971d84c833c6dac53501145715
> Author: Jiri Olsa <jolsa@kernel.org>
>
> x86/fgraph: Fix return_to_handler regs.rsp value
>
> This commit fixes the unwinding rsp register value in return_to_handler. It
> saves the original rsp before the FRAME_SIZE adjustment, replacing the
> previous incorrect value. This ensures fprobe and direct unwinders see the
> right stack pointer.
>
> Link: https://lore.kernel.org/bpf/20260126211837.472802-2-jolsa@kernel.org
>
> > diff --git a/arch/x86/kernel/ftrace_64.S b/arch/x86/kernel/ftrace_64.S
> > --- a/arch/x86/kernel/ftrace_64.S
> > +++ b/arch/x86/kernel/ftrace_64.S
> > @@ -364,6 +364,9 @@ SYM_CODE_START(return_to_handler)
> > UNWIND_HINT_UNDEFINED
> > ANNOTATE_NOENDBR
> >
> > + /* Store original rsp for pt_regs.sp value. */
> > + movq %rsp, %rdi
> > +
> > /* Restore return_to_handler value that got eaten by previous ret instruction. */
> > subq $8, %rsp
> > UNWIND_HINT_FUNC
> > @@ -374,7 +377,7 @@ SYM_CODE_START(return_to_handler)
> > movq %rax, RAX(%rsp)
> > movq %rdx, RDX(%rsp)
> > movq %rbp, RBP(%rsp)
> > - movq %rsp, RSP(%rsp)
> > + movq %rdi, RSP(%rsp)
>
> [Severity: High]
> This is a pre-existing issue, but could this code leak kernel stack memory
> to BPF programs?
>
> This assembly creates a pt_regs structure on the stack by allocating
> FRAME_SIZE bytes, but only explicitly saves rax, rdx, rbp, and rsp. Later in
> ftrace_return_to_handler and arch_ftrace_partial_regs, only a few additional
> fields are populated.
>
> The remaining fields (like rdi, rsi, rbx, and r12-r15) remain uninitialized.
>
> Because this partially initialized pt_regs structure is passed to BPF
> kprobe_multi return programs via kprobe_multi_link_exit_handler(), and the
> BPF verifier allows reading any field of the pt_regs context, does this
> allow BPF programs to extract residual kernel stack data?
I think because kprobe/bpf is root only, it's not really an issue,
cc-ing Steven and Masami
jirka
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 6.12] x86/fgraph: Fix return_to_handler regs.rsp value
2026-05-27 7:36 ` Jiri Olsa
@ 2026-05-27 13:39 ` Steven Rostedt
0 siblings, 0 replies; 5+ messages in thread
From: Steven Rostedt @ 2026-05-27 13:39 UTC (permalink / raw)
To: Jiri Olsa; +Cc: sashiko-reviews, Gyokhan Kochmarla, bpf, Masami Hiramatsu
On Wed, 27 May 2026 09:36:14 +0200
Jiri Olsa <olsajiri@gmail.com> wrote:
> > Because this partially initialized pt_regs structure is passed to BPF
> > kprobe_multi return programs via kprobe_multi_link_exit_handler(), and the
> > BPF verifier allows reading any field of the pt_regs context, does this
> > allow BPF programs to extract residual kernel stack data?
>
> I think because kprobe/bpf is root only, it's not really an issue,
Correct. If you can add kprobes to bpf programs, there's a hell of a lot
more you can "leak" than stack addresses :-p
-- Steve
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 6.12] x86/fgraph: Fix return_to_handler regs.rsp value
2026-05-26 19:23 [PATCH 6.12] x86/fgraph: Fix return_to_handler regs.rsp value Gyokhan Kochmarla
2026-05-26 19:50 ` sashiko-bot
@ 2026-05-27 19:49 ` Sasha Levin
1 sibling, 0 replies; 5+ messages in thread
From: Sasha Levin @ 2026-05-27 19:49 UTC (permalink / raw)
To: stable, gregkh
Cc: Sasha Levin, jolsa, rostedt, mhiramat, tglx, mingo, bp, x86,
linux-trace-kernel, bpf, Andrii Nakryiko, Gyokhan Kochmarla
> commit 8bc11700e0d23d4fdb7d8d5a73b2e95de427cabc upstream.
Queued for 6.12.y, thanks.
--
Thanks,
Sasha
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2026-05-27 19:49 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-05-26 19:23 [PATCH 6.12] x86/fgraph: Fix return_to_handler regs.rsp value Gyokhan Kochmarla
2026-05-26 19:50 ` sashiko-bot
2026-05-27 7:36 ` Jiri Olsa
2026-05-27 13:39 ` Steven Rostedt
2026-05-27 19:49 ` Sasha Levin
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox