arm64: stacktrace: Clarify documentation of EL0 entry frame skipping
PATCH 1/1 arm64: stacktrace: Clarify documentation of EL0 entry frame skipping
The unwinder suppresses reporting of frames created on entry from EL0 where both FP and PC are NULL in order to avoid having the final NULL PC cluttering traces and potentially creating confusion. There is a comment explaining this which doesn't explicitly call out the case where we have a valid PC but NULL FP, do so in order to ensure that things are as clear as possible to readers unfamiliar with the code.
--- arch/arm64/kernel/stacktrace.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
arch/arm64/kernel/stacktrace.c +3 -2
@@ -107,8 +107,9 @@ int notrace unwind_frame(struct task_struct *tsk, struct stackframe *frame)/* * Frames created upon entry from EL0 have NULL FP and PC values, so * don't bother reporting these. Frames created by __noreturn functions -* might have a valid FP even if PC is bogus, so only terminate where -* both are NULL. +* might have a valid FP even if PC is bogus and the last frame in a +* normal stack will have a valid PC but NULL FP, so only terminate +* early where both are NULL. */ if (!frame->fp && !frame->pc)return -EINVAL ;
Signature · 7 lines
2.20.1 _______________________________________________ linux-arm-kernel mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Mark Rutland
reviewerHi Mark,
Quoted 17 lines · Mark Brown
The unwinder suppresses reporting of frames created on entry from EL0 where both FP and PC are NULL in order to avoid having the final NULL PC cluttering traces and potentially creating confusion. There is a comment explaining this which doesn't explicitly call out the case where we have a valid PC but NULL FP, do so in order to ensure that things are as clear as possible to readers unfamiliar with the code.
Signed-off-by Mark Brown <[email protected]>--- arch/arm64/kernel/stacktrace.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)arch/arm64/kernel/stacktrace.c +0
-0
arch/arm64/kernel/stacktrace.c int notrace unwind_frame(struct task_struct *tsk, struct stackframe *frame) · line 113
Quoted 9 lines
arch/arm64/kernel/stacktrace.c +3
-2@@ -107,8 +107,9 @@ int notrace unwind_frame(struct task_struct *tsk, struct stackframe *frame)/* * Frames created upon entry from EL0 have NULL FP and PC values, so * don't bother reporting these. Frames created by __noreturn functions -* might have a valid FP even if PC is bogus, so only terminate where -* both are NULL. +* might have a valid FP even if PC is bogus and the last frame in a +* normal stack will have a valid PC but NULL FP, so only terminate +* early where both are NULL. */
I reckon we should remove the special case entirely rather than trying to shore up the comment here -- the related comment in entry.S is stale and misleading, and this will be a mystery to unfamiliar readers anyhow.
Does the below patch look ok to you? I suspect the explicit check for the terminal case may help with the reliable stacktrace stuff? e.g. we could make that return -ENOENT to distinguish a terminal case from a broken frame record.
Thanks,
Mark.
---->8----
From 61cb79bd9d926744190a43a5e1ec9aefda7ba052 Mon Sep 17 00:00:00 2001
From: Mark Rutland <[email protected]>
Date: Thu, 8 Oct 2020 10:47:59 +0100
Subject: [PATCH] arm64: remove EL0 exception frame record
When entering an exception from EL0, the entry code creates a synthetic frame record with a NULL PC. This was used by the code introduced in commit:
7326749801396105 ("arm64: unwind: reference pt_regs via embedded stack frame")
... to discover exception entries on the stack and dump the associated pt_regs. Since the NULL PC was undesirable for the stacktrace, we added a special case to unwind_frame() to prevent the NULL PC from being logged.
Since commit:
a25ffd3a6302a678 ("arm64: traps: Don't print stack or raw PC/LR values in backtraces")
... we no longer try to dump the pt_regs as part of a stacktrace, and hence no longer need the synthetic exception record.
This patch removes the synthetic exception record and the associated special case in unwind_frame(). Instead, EL0 exceptions set the FP to NULL, as is the case for other terminal records (e.g. when a kernel thread starts). The synthetic record for exceptions from EL1 is retrained as this has useful unwind information for the interrupted context.
To make the terminal case a bit clearer, an explicit check is added to the start of unwind_frame(). This would otherwise be caught implicitly by the on_accessible_stack() checks.
--- arch/arm64/kernel/entry.S | 10 +++++----- arch/arm64/kernel/stacktrace.c | 13 ++++--------- 2 files changed, 9 insertions(+), 14 deletions(-)
arch/arm64/kernel/entry.S +5 -5
@@ -228,16 +228,16 @@ alternative_else_nop_endif stp lr, x21, [sp, #S_LR] /* - * In order to be able to dump the contents of struct pt_regs at the - * time the exception was taken (in case we attempt to walk the call - * stack later), chain it together with the stack frames. + * For exceptions from EL0, terminate the callchain here. + * For exceptions from EL1, create a synthetic frame record so the + * interrupted code shows up in the backtrace. */ .if \el == 0 - stp xzr, xzr, [sp, #S_STACKFRAME] + mov x29, xzr .else stp x29, x22, [sp, #S_STACKFRAME] - .endif add x29, sp, #S_STACKFRAME + .endif #ifdef CONFIG_ARM64_SW_TTBR0_PAN alternative_if_not ARM64_HAS_PAN
arch/arm64/kernel/stacktrace.c +4 -9
@@ -44,6 +44,10 @@ int notrace unwind_frame(struct task_struct *tsk, struct stackframe *frame)unsigned long fp = frame->fp;struct stack_info info; +/* Terminal record; nothing to unwind */ +if (!fp) +return -EINVAL ; +if (fp &0xf )return -EINVAL ;
@@ -104,15 +108,6 @@ int notrace unwind_frame(struct task_struct *tsk, struct stackframe *frame) frame->pc =ptrauth_strip_insn_pac (frame->pc); -/* -* Frames created upon entry from EL0 have NULL FP and PC values, so -* don't bother reporting these. Frames created by __noreturn functions -* might have a valid FP even if PC is bogus, so only terminate where -* both are NULL. -*/ -if (!frame->fp && !frame->pc) -return -EINVAL ; -return 0 ; }NOKPROBE_SYMBOL (unwind_frame);
Signature · 7 lines
2.11.0 _______________________________________________ linux-arm-kernel mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Quoted 3 lines · Mark Rutland
I reckon we should remove the special case entirely rather than trying to shore up the comment here -- the related comment in
entry.Sis stale and misleading, and this will be a mystery to unfamiliar readers anyhow.
That works too, so long as the check isn't really doing anything it can just be deleted and the comment with it - it's certainly one less weird special case. So long as all the things relying on it are already gone.
Quoted 4 lines
Does the below patch look ok to you? I suspect the explicit check for the terminal case may help with the reliable stacktrace stuff? e.g. we could make that return
-ENOENTto distinguish a terminal case from a broken frame record.
That's almost exactly what I've got in my reliable stacktrace queue, down to the choice of -ENOENT for reporting the end of an intact stack. I'll pick this up for my changes if it doesn't get picked up separately.
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel