Hello.
Thank you for pointing this out. I didn't know this was related to ABI calling
conventions in this particular case. Simple removal of the `sub` instruction is
causing tests regressions in our development branch together with other
regressions introduced by rebase the to current upstream master branch (induced
by recent changes to pthread). It will take some time to investigate and
validate this change.
Thank you for your patience,
Radek
________________________________________
From: Jeremy Drake <cyg...@jdrake.com>
Sent: Saturday, June 21, 2025 8:47 PM
To: Radek Barton <radek.bar...@microsoft.com>
Cc: cygwin-patches@cygwin.com <cygwin-patches@cygwin.com>
Subject: [EXTERNAL] Re: [PATCH] Cygwin: stack base initialization for AArch64
On Wed, 18 Jun 2025, Jeremy Drake via Cygwin-patches wrote:
> On Wed, 18 Jun 2025, Radek Barton via Cygwin-patches wrote:
>
> > -#ifdef __x86_64__
> > /* Set stack pointer to new address. Set frame pointer to
> > stack pointer and subtract 32 bytes for shadow space. */
> > +#if defined(__x86_64__)
> > __asm__ ("\n\
> > movq %[ADDR], %%rsp \n\
> > movq %%rsp, %%rbp \n\
> > subq $32,%%rsp \n"
> > : : [ADDR] "r" (stackaddr));
> > +#elif defined(__aarch64__)
> > + __asm__ ("\n\
> > + mov fp, %[ADDR] \n\
> > + sub sp, fp, #32 \n"
>
> Is the 32-byte shadow space part of the aarch64 calling convention spec,
> or is this just copying what x86_64 was doing? My impression is that this
> space was part of the x86_64 calling convention.
The patch for pthread stack initialization dropped the 32-byte shadow
space, and I believe this patch should as well.
