From: Stephen Brennan <stephen.s.brennan@oracle.com>
To: Alexander Gordeev <agordeev@linux.ibm.com>,
Vasily Gorbik <gor@linux.ibm.com>,
Heiko Carstens <hca@linux.ibm.com>
Cc: Janosch Frank <frankja@linux.ibm.com>,
Christian Borntraeger <borntraeger@linux.ibm.com>,
Sven Schnelle <svens@linux.ibm.com>,
Claudio Imbrenda <imbrenda@linux.ibm.com>,
Andrew Donnellan <andrew+kernel@donnellan.id.au>,
linux-s390@vger.kernel.org, linux-debuggers@vger.kernel.org,
linux-kernel@vger.kernel.org
Subject: Re: [PATCH] s390: add CFI directives for call_on_stack
Date: Mon, 16 Nov 2026 10:02:57 -0800 [thread overview]
Message-ID: <87qzgkpsby.fsf@oracle.com> (raw)
In-Reply-To: <20261030231714.1319773-1-stephen.s.brennan@oracle.com>
Stephen Brennan <stephen.s.brennan@oracle.com> writes:
> When debugging s390 on drgn, we experience two issues. First, when a
> panic is triggered on CPU 0, drgn is unable to unwind past a stack
> switching event[1,2]. Second, drgn is unable to unwind through an
> interrupt exception frame.
>
> It turns out that the reason for both is the generated call_on_stack()
> assembly. It modifies %r15 (the stack pointer) without any CFI
> directives describing this change. Thus, the compiler-generated CFI is
> wrong, and drgn tries to unwind using rules based on the wrong SP value.
>
> To resolve this, use %r10 (a callee-saved register) to store the CFA in
> a temporary register. Then, emit a CFI directive describing the change.
> So long as DWARF CFI is used to unwind all leaf functions, the value of
> %r10 should be recoverable, and unwinders will be able to use it to
> determine the CFA and continue to the new stack.
>
> I've tested this only in a QEMU VM using drgn's vmtest framework.
> With the patch, drgn is able to unwind through two stack changes: the
> change to the IRQ stack, and the change to the CPU0 nodat_stack during a
> panic on CPU 0.
>
> [1]: https://github.com/osandov/drgn/blob/97b613c23bd4d100a61b3b06fb1b5097fcef9f61/vmtest/enter_kdump.py#L69-L74
> [2]: https://github.com/osandov/drgn/blob/97b613c23bd4d100a61b3b06fb1b5097fcef9f61/tests/linux_kernel/vmcore/test_vmcore.py#L19-L25
>
> Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
Hello,
Following up on this patch. Is there any additional verification I can
work on to help facilitate review?
thanks,
Stephen
> ---
> arch/s390/include/asm/stacktrace.h | 18 +++++++++++++++++-
> 1 file changed, 17 insertions(+), 1 deletion(-)
>
> diff --git a/arch/s390/include/asm/stacktrace.h b/arch/s390/include/asm/stacktrace.h
> index ac3606c3babe4..ed9ae00fa339c 100644
> --- a/arch/s390/include/asm/stacktrace.h
> +++ b/arch/s390/include/asm/stacktrace.h
> @@ -170,6 +170,12 @@ static __always_inline unsigned long get_stack_pointer(struct task_struct *task,
> #define CALL_PARM_5(t, a, ...) t, CALL_PARM_4(__VA_ARGS__)
> #define CALL_PARM_6(t, a, ...) t, CALL_PARM_5(__VA_ARGS__)
>
> +#ifdef CONFIG_DEBUG_INFO
> +#define CALL_ON_STACK_CFI(str) str
> +#else
> +#define CALL_ON_STACK_CFI(str)
> +#endif
> +
> /*
> * Use call_on_stack() to call a function switching to a specified
> * stack. Proper sign and zero extension of function arguments is
> @@ -177,6 +183,10 @@ static __always_inline unsigned long get_stack_pointer(struct task_struct *task,
> *
> * rc = call_on_stack(nr, stack, rettype, fn, t1, a1, t2, a2, ...)
> *
> + * Keep the original CFA in callee-saved register 10 while running on the
> + * alternate stack so that DWARF register save rules still refer to the
> + * original stack.
> + *
> * - nr specifies the number of function arguments of fn.
> * - stack specifies the stack to be used.
> * - fn is the function to be called.
> @@ -189,6 +199,8 @@ static __always_inline unsigned long get_stack_pointer(struct task_struct *task,
> ({ \
> rettype (*__fn)(CALL_PARM_##nr(__VA_ARGS__)) = fn; \
> unsigned long frame = current_frame_address(); \
> + register unsigned long cfa asm("10") = \
> + frame + STACK_FRAME_OVERHEAD; \
> unsigned long __stack = stack; \
> unsigned long prev; \
> CALL_LARGS_##nr(__VA_ARGS__); \
> @@ -196,15 +208,19 @@ static __always_inline unsigned long get_stack_pointer(struct task_struct *task,
> \
> CALL_TYPECHECK_##nr(__VA_ARGS__); \
> asm volatile( \
> + CALL_ON_STACK_CFI(".cfi_remember_state\n") \
> " lgr %[_prev],15\n" \
> + CALL_ON_STACK_CFI(".cfi_def_cfa %[_cfa],0\n") \
> " lg 15,%[_stack]\n" \
> " stg %[_frame],%[_bc](15)\n" \
> " brasl 14,%[_fn]\n" \
> - " lgr 15,%[_prev]" \
> + " lgr 15,%[_prev]\n" \
> + CALL_ON_STACK_CFI(".cfi_restore_state\n") \
> : [_prev] "=&d" (prev), CALL_FMT_##nr \
> : [_stack] "R" (__stack), \
> [_bc] "i" (offsetof(struct stack_frame, back_chain)), \
> [_frame] "d" (frame), \
> + [_cfa] "d" (cfa), \
> [_fn] "X" (__fn) : CALL_CLOBBER_##nr); \
> (rettype)r2; \
> })
> --
> 2.52.0
prev parent reply other threads:[~2026-11-16 18:04 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-30 23:17 [PATCH] s390: add CFI directives for call_on_stack Stephen Brennan
2026-11-16 18:02 ` Stephen Brennan [this message]
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=87qzgkpsby.fsf@oracle.com \
--to=stephen.s.brennan@oracle.com \
--cc=agordeev@linux.ibm.com \
--cc=andrew+kernel@donnellan.id.au \
--cc=borntraeger@linux.ibm.com \
--cc=frankja@linux.ibm.com \
--cc=gor@linux.ibm.com \
--cc=hca@linux.ibm.com \
--cc=imbrenda@linux.ibm.com \
--cc=linux-debuggers@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-s390@vger.kernel.org \
--cc=svens@linux.ibm.com \
/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;
as well as URLs for NNTP newsgroup(s).