Re: [PATCH v12 02/14] unwind_user: Add frame pointer support

[Jens Remus <jremus@linux.ibm.com>]

On 01.07.2025 04:56, Steven Rostedt wrote:
> On Mon, 30 Jun 2025 19:10:09 -0700
> Linus Torvalds <torvalds@linux-foundation.org> wrote:
>> On Mon, 30 Jun 2025 at 17:54, Steven Rostedt <rostedt@goodmis.org> wrote:
>>>
>>> +       /* stack going in wrong direction? */
>>> +       if (cfa <= state->sp)
>>> +               goto done;  
>>
>> I suspect this should do a lot more testing.
> 
> Sure.

The above assumes:

	curr_frame_sp = state->sp
	prev_frame_sp = cfa   // for arches that define CFA as SP at call site

On s390 the prev_frame_sp may be equal to curr_frame_sp for the topmost
frame, as long as the topmost function did not allocate any stack.  For
instance when early in the prologue or when in a leaf function that does
not require any stack space.  My s390 sframe support patches would
therefore currently change above check to:

	/* stack going in wrong direction? */
	if (sp <= state->sp - topmost)
		goto done;

I assume that should be the case for all architectures whose function
call instruction does not modify the SP, unlike x86-64's CALL does.

>>> +       if (frame->fp_off && get_user(fp, (unsigned long __user *)(cfa + frame->fp_off)))
>>> +               goto done;  
>>
>> .. and this should check the frame for validity too.  At a minimum it
>> should be properly aligned, but things like "it had better be above
>> the current frame" to avoid having some loop would seem to be a good
>> idea.
> 
> Makes sense.

On s390 the FP (register) value does not necessarily need to be above
the SP value, as the s390x ELF ABI does only designate a "preferred" FP
register, so that the FP register may be used for other purposes, when a
FP is not required in a function.

So the output FP value cannot be checked on s390.  But the input FP
value could probably be checked as follows before computing the CFA:

	if (frame->use_fp && state->fp < state->sp)
		goto done;

	/* Get the Canonical Frame Address (CFA) */
	cfa = (frame->use_fp ? state->fp : state->sp) + frame->cfa_off;

>> Maybe even check that it's the same vma?
> 
> Hmm, I call on to Jens Remus and ask if s390 can do anything whacky here?
> Where something that isn't allowed on other architectures may be allowed
> there? I know s390 has some strange type of stack usage.

On s390, if a FP is required for dynamic stack allocation, it is only
initialized as late as possible, that is usually after static stack
allocation.  Therefore SP == FP is rather seldom the case.

Regards,
Jens
-- 
Jens Remus
Linux on Z Development (D3303)
+49-7031-16-1128 Office
jremus@de.ibm.com

IBM

IBM Deutschland Research & Development GmbH; Vorsitzender des Aufsichtsrats: Wolfgang Wendt; Geschäftsführung: David Faller; Sitz der Gesellschaft: Böblingen; Registergericht: Amtsgericht Stuttgart, HRB 243294
IBM Data Privacy Statement: https://www.ibm.com/privacy/