Message ID | 1444915314-13649-1-git-send-email-jungseoklee85@gmail.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Thu, Oct 15, 2015 at 01:21:54PM +0000, Jungseok Lee wrote: > dump_backtrace() has its own backtrace logic unlike perf callchain which > relies on walk_stackframe(). They behave differently when a symbol is > recorded. Perf writes it down *before* calling unwind_frame(), but > dump_backtrace() prints it out *after* unwind_frame(). As a result, the > last valid symbol is not added to a list in case of dump_backtrace(). > > This patch catches up the last symbol as synchronising dump_backtrace() > with perf callchain. However, the patch does not cover a case where MMU > is disabled. That is, a physical address can be stored in stack frame, > but it's not handled. For example, a swapper process falls into this case. > Unlike a swapper from a secondary core, a swapper on a boot cpu, which > starting from __mmap_switched(), can't be tracked down with a simple > conversion, phys_to_virt(), because PC is retrieved from LR - 4, not LR. It would be good to have an example backtrace before and after this patch is applied, to show what it fixes. > It is a big tradeoff to change both head.S and unwind_frame() structure > for a few of symbols in *.S, so this hunk does not take care of the case. > > Cc: AKASHI Takahiro <takahiro.akashi@linaro.org> > Cc: James Morse <james.morse@arm.com> > Cc: Mark Rutland <mark.rutland@arm.com> > Signed-off-by: Jungseok Lee <jungseoklee85@gmail.com> > --- > arch/arm64/kernel/traps.c | 16 +++++++++++----- > 1 file changed, 11 insertions(+), 5 deletions(-) > > diff --git a/arch/arm64/kernel/traps.c b/arch/arm64/kernel/traps.c > index f93aae5..4ddb928 100644 > --- a/arch/arm64/kernel/traps.c > +++ b/arch/arm64/kernel/traps.c > @@ -103,12 +103,13 @@ static void dump_mem(const char *lvl, const char *str, unsigned long bottom, > set_fs(fs); > } > > -static void dump_backtrace_entry(unsigned long where, unsigned long stack) > +static void dump_backtrace_entry(unsigned long where) > { > + /* > + * The highest stack frame of a swapper process stores PC in a form > + * of physical address, but this case is not handled. > + */ > print_ip_sym(where); > - if (in_exception_text(where)) > - dump_mem("", "Exception stack", stack, > - stack + sizeof(struct pt_regs), false); > } > > static void dump_instr(const char *lvl, struct pt_regs *regs) > @@ -172,12 +173,17 @@ static void dump_backtrace(struct pt_regs *regs, struct task_struct *tsk) > pr_emerg("Call trace:\n"); > while (1) { > unsigned long where = frame.pc; > + unsigned long stack; > int ret; > > + dump_backtrace_entry(where); > ret = unwind_frame(&frame); > if (ret < 0) > break; > - dump_backtrace_entry(where, frame.sp); > + stack = frame.sp; > + if (in_exception_text(where)) > + dump_mem("", "Exception stack", stack, > + stack + sizeof(struct pt_regs), false); AFAICT, the original code is all based on unwind_backtrace in arch/arm/kernel/unwind.c. Does that need updating too (as a separate patch)? Will
On Oct 16, 2015, at 2:26 AM, Will Deacon wrote: Hi Will, > On Thu, Oct 15, 2015 at 01:21:54PM +0000, Jungseok Lee wrote: >> dump_backtrace() has its own backtrace logic unlike perf callchain which >> relies on walk_stackframe(). They behave differently when a symbol is >> recorded. Perf writes it down *before* calling unwind_frame(), but >> dump_backtrace() prints it out *after* unwind_frame(). As a result, the >> last valid symbol is not added to a list in case of dump_backtrace(). >> >> This patch catches up the last symbol as synchronising dump_backtrace() >> with perf callchain. However, the patch does not cover a case where MMU >> is disabled. That is, a physical address can be stored in stack frame, >> but it's not handled. For example, a swapper process falls into this case. >> Unlike a swapper from a secondary core, a swapper on a boot cpu, which >> starting from __mmap_switched(), can't be tracked down with a simple >> conversion, phys_to_virt(), because PC is retrieved from LR - 4, not LR. > > It would be good to have an example backtrace before and after this patch > is applied, to show what it fixes. Agreed. I will add a call trace data to the commit message. >> It is a big tradeoff to change both head.S and unwind_frame() structure >> for a few of symbols in *.S, so this hunk does not take care of the case. >> >> Cc: AKASHI Takahiro <takahiro.akashi@linaro.org> >> Cc: James Morse <james.morse@arm.com> >> Cc: Mark Rutland <mark.rutland@arm.com> >> Signed-off-by: Jungseok Lee <jungseoklee85@gmail.com> >> --- >> arch/arm64/kernel/traps.c | 16 +++++++++++----- >> 1 file changed, 11 insertions(+), 5 deletions(-) >> >> diff --git a/arch/arm64/kernel/traps.c b/arch/arm64/kernel/traps.c >> index f93aae5..4ddb928 100644 >> --- a/arch/arm64/kernel/traps.c >> +++ b/arch/arm64/kernel/traps.c >> @@ -103,12 +103,13 @@ static void dump_mem(const char *lvl, const char *str, unsigned long bottom, >> set_fs(fs); >> } >> >> -static void dump_backtrace_entry(unsigned long where, unsigned long stack) >> +static void dump_backtrace_entry(unsigned long where) >> { >> + /* >> + * The highest stack frame of a swapper process stores PC in a form >> + * of physical address, but this case is not handled. >> + */ >> print_ip_sym(where); >> - if (in_exception_text(where)) >> - dump_mem("", "Exception stack", stack, >> - stack + sizeof(struct pt_regs), false); >> } >> >> static void dump_instr(const char *lvl, struct pt_regs *regs) >> @@ -172,12 +173,17 @@ static void dump_backtrace(struct pt_regs *regs, struct task_struct *tsk) >> pr_emerg("Call trace:\n"); >> while (1) { >> unsigned long where = frame.pc; >> + unsigned long stack; >> int ret; >> >> + dump_backtrace_entry(where); >> ret = unwind_frame(&frame); >> if (ret < 0) >> break; >> - dump_backtrace_entry(where, frame.sp); >> + stack = frame.sp; >> + if (in_exception_text(where)) >> + dump_mem("", "Exception stack", stack, >> + stack + sizeof(struct pt_regs), false); > > AFAICT, the original code is all based on unwind_backtrace in > arch/arm/kernel/unwind.c. Does that need updating too (as a separate patch)? I think so, but I don't have any evidence from a real hardware.. Best Regards Jungseok Lee
diff --git a/arch/arm64/kernel/traps.c b/arch/arm64/kernel/traps.c index f93aae5..4ddb928 100644 --- a/arch/arm64/kernel/traps.c +++ b/arch/arm64/kernel/traps.c @@ -103,12 +103,13 @@ static void dump_mem(const char *lvl, const char *str, unsigned long bottom, set_fs(fs); } -static void dump_backtrace_entry(unsigned long where, unsigned long stack) +static void dump_backtrace_entry(unsigned long where) { + /* + * The highest stack frame of a swapper process stores PC in a form + * of physical address, but this case is not handled. + */ print_ip_sym(where); - if (in_exception_text(where)) - dump_mem("", "Exception stack", stack, - stack + sizeof(struct pt_regs), false); } static void dump_instr(const char *lvl, struct pt_regs *regs) @@ -172,12 +173,17 @@ static void dump_backtrace(struct pt_regs *regs, struct task_struct *tsk) pr_emerg("Call trace:\n"); while (1) { unsigned long where = frame.pc; + unsigned long stack; int ret; + dump_backtrace_entry(where); ret = unwind_frame(&frame); if (ret < 0) break; - dump_backtrace_entry(where, frame.sp); + stack = frame.sp; + if (in_exception_text(where)) + dump_mem("", "Exception stack", stack, + stack + sizeof(struct pt_regs), false); } }
dump_backtrace() has its own backtrace logic unlike perf callchain which relies on walk_stackframe(). They behave differently when a symbol is recorded. Perf writes it down *before* calling unwind_frame(), but dump_backtrace() prints it out *after* unwind_frame(). As a result, the last valid symbol is not added to a list in case of dump_backtrace(). This patch catches up the last symbol as synchronising dump_backtrace() with perf callchain. However, the patch does not cover a case where MMU is disabled. That is, a physical address can be stored in stack frame, but it's not handled. For example, a swapper process falls into this case. Unlike a swapper from a secondary core, a swapper on a boot cpu, which starting from __mmap_switched(), can't be tracked down with a simple conversion, phys_to_virt(), because PC is retrieved from LR - 4, not LR. It is a big tradeoff to change both head.S and unwind_frame() structure for a few of symbols in *.S, so this hunk does not take care of the case. Cc: AKASHI Takahiro <takahiro.akashi@linaro.org> Cc: James Morse <james.morse@arm.com> Cc: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Jungseok Lee <jungseoklee85@gmail.com> --- arch/arm64/kernel/traps.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)