Re: [PATCH] strub: Only unbias stack point for SPARC_STACK_BOUNDARY_HACK [PR113100]

[Alexandre Oliva <oliva@adacore.com>]

On Jan  7, 2024, "Kewen.Lin" <linkw@linux.ibm.com> wrote:

> As PR113100 shows, the unbiasing introduced by r14-6737 can
> cause the scrubbing to overrun and screw some critical data
> on stack like saved toc base consequently cause segfault on
> Power.

Ugh.  Sorry about the breakage, and thanks for addressing it during my
absence.  Happy GNU Year! :-)

> By checking PR112917, IMHO we should keep this unbiasing
> guarded under SPARC_STACK_BOUNDARY_HACK (TARGET_ARCH64 &&
> TARGET_STACK_BIAS), similar to some existing code special
> treating SPARC stack bias.

I'm afraid this change will most certainly regress 32-bit sparc, because
of the large register save area.

I had been hesitant to introduce yet another target configuration knob,
but it looks like this is what we're going to have to do to accommodate
all targets.

> I also expect the culprit commit can
> affect those ports with nonzero STACK_POINTER_OFFSET.

IMHO it really shouldn't.  STACK_POINTER_OFFSET should be the "Offset
from the stack pointer register to the first location at which outgoing
arguments are placed", which suggests to me that no data that the callee
couldn't change should go in the area below (or above) %sp+S_P_O.

ISTM that PPC sets up a save area between the outgoing args and the
stack pointer; I don't think that's very common, but I suppose other
targets that do so would also define STACK_POINTER_OFFSET to nonzero so
as to reserve those bits.  But whether they should be cleared by stack
scrubbing, as on sparc, or preserved, as on ppc, depends on the ABI
conventions, so we probably can't help yet another knob :-/

I'll take care of that, and update the corresponding documentation.

Thanks,

-- 
Alexandre Oliva, happy hacker            https://FSFLA.org/blogs/lxo/
   Free Software Activist                   GNU Toolchain Engineer
More tolerance and less prejudice are key for inclusion and diversity
Excluding neuro-others for not behaving ""normal"" is *not* inclusive