Re: 2.6.19-rc2-mm1 unwinder issues ?

[Jan Beulich]

>>> Badari Pulavarty <pbadari@xxxxxxxxxx> 19.10.06 00:36 >>>
>Hi Andi,
>
>I am not getting stack traces properly on 2.6.19-rc2-mm1 again 
>(on my amd64 box).
>
>Wondering, if the unwinder code changed again ??
>
>Thanks,
>Badari

This patch:

>Annotate interrupt frame backlink in interrupt handlers
>
>Add correct CFI annotation to the backlink on top of the interrupt stack.
>
>Signed-off-by: Andi Kleen <ak@xxxxxxx>
>
>---
> arch/x86_64/kernel/entry.S |    3 +++
> 1 files changed, 3 insertions(+)
>
>Index: linux/arch/x86_64/kernel/entry.S
>===================================================================
>--- linux.orig/arch/x86_64/kernel/entry.S
>+++ linux/arch/x86_64/kernel/entry.S
>@@ -535,6 +535,8 @@ END(stub_rt_sigreturn)
> 1:	incl	%gs:pda_irqcount
> 	cmoveq %gs:pda_irqstackptr,%rsp
> 	push    %rbp			# backlink for old unwinder
>+	CFI_ADJUST_CFA_OFFSET 8
>+	CFI_REL_OFFSET rbp,0
> 	/*
> 	 * We entered an interrupt context - irqs are off:
> 	 */
>@@ -1174,6 +1176,7 @@ ENTRY(call_softirq)
> 	incl %gs:pda_irqcount
> 	cmove %gs:pda_irqstackptr,%rsp
> 	push  %rbp			# backlink for old unwinder
>+	CFI_ADJUST_CFA_OFFSET    8
> 	call __do_softirq
> 	leaveq
> 	CFI_DEF_CFA_REGISTER	rsp

must be reverted for things to work again. Andi, what did you
want to cure with it? Clearly, when rSP isn't the CFA register
anymore, there must not (normally) be adjustments to the
CFA offset. Similarly, when a register was saved already and
it's not its spill location that changes, it must not be marked
as being saved a second time.

Jan
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/