Andreas Jaeger <aj@suse.de> writes:

> Jakub Jelinek <jakub@redhat.com> writes:
>
>> On Sat, Jan 10, 2004 at 08:49:49PM +0100, Andreas Jaeger wrote:
>>> Andreas Jaeger <aj@suse.de> writes:
>>> 
>>> > Jakub Jelinek <jakub@redhat.com> writes:
>>> > [...]
>>> >>> libc/sysdeps/x86_64/__longjmp.S
>>> >
>>> > What's the best way to handle the destroying of the registers with
>>> > CFI?  Should we just mark the registers with "cfi_undefined"?
>>> 
>>> I'm appending a proposed patch for __longjmp.S.  Or can anybody
>>> envision a better way doing this?
>>
>> Cannot you from the first movq ***,%rbx pretend you're
>> already at the jmpq %*rdx instruction, ie. unwind to the
>> longjmp target?
>> Write in the unwind info where to fetch %rbp, %r12..%r15,
>> ra etc. from?
>
> You mean something like the following?
>
> cfi_def_cfa(%rdi,0)
> cfi_offset(%rbx,JB_RBX*8)
> cfi_offset(%r12,JB_R12*8)
> ...
>
> I'm not sure what the right way is to handle this here at all ...

I've done it now completely:

ENTRY(__longjmp)
	/* Restore registers.  */
	/* We add unwind information for the target here.  */
	cfi_def_cfa(%rdi, 0)
	cfi_offset(%rbx,JB_RBX*8)
	cfi_offset(%r12,JB_R12*8)
	cfi_offset(%r13,JB_R13*8)
	cfi_offset(%r14,JB_R14*8)
	cfi_offset(%r15,JB_R15*8)
	cfi_offset(%rsp,JB_RSP*8)
	movq (JB_RBX*8)(%rdi),%rbx
	movq (JB_RBP*8)(%rdi),%rbp
	movq (JB_R12*8)(%rdi),%r12
	movq (JB_R13*8)(%rdi),%r13
	movq (JB_R14*8)(%rdi),%r14
	movq (JB_R15*8)(%rdi),%r15
	/* Set return value for setjmp.  */
	test %esi,%esi
	mov $01,%eax
	cmove %eax,%esi
	mov %esi, %eax
	movq (JB_PC*8)(%rdi),%rdx
	movq (JB_RSP*8)(%rdi),%rsp
	jmpq *%rdx
END (BP_SYM (__longjmp))

Does this look ok? In that case I'll commit it (and will fix
setcontext/swapcontext in a similar way)...

Andreas
-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SuSE Linux AG, Maxfeldstr. 5, 90409 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126