Re: [PATCH] x86/signal: Remove pax argument from restore_sigcontext

[Ingo Molnar <mingo@kernel.org>]

* Brian Gerst <brgerst@gmail.com> wrote:

> On Mon, Apr 6, 2015 at 3:03 AM, Ingo Molnar <mingo@kernel.org> wrote:
> >
> > * Brian Gerst <brgerst@gmail.com> wrote:
> >
> >> > I'm wondering what the original reason for adding the extra
> >> > handling of regs->ax was. Maybe something changed regs->ax - but I
> >> > cannot find such code path anymore.
> >> >
> >> > It would be nice to try to do a bit of Git archeology to figure
> >> > out the origins of this complication - maybe it's something subtle
> >> > - or it's something that has changed meanwhile.
> >>
> >> It goes all the way back to 2.1.106pre1, when restore_sigcontext()
> >> was changed to return an error code instead of EAX directly.
> >>
> >> https://git.kernel.org/cgit/linux/kernel/git/history/history.git/diff/arch/i386/kernel/signal.c?id=9a8f8b7ca3f319bd668298d447bdf32730e51174
> >
> > Indeed: restore_sigcontext() used to return eax as a return value,
> > without copying it into regs->ax.
> >
> > Then in 2007
> 
> Version 2.1.106 was released on Jun 13, 1998.

Sigh, the Git timestamp of the historic tree threw me off :-)

> > sigaltstack syscall support was added, where the return value of 
> > restore_sigcontext() was changed to carry the memory-copying 
> > failure code. But instead of putting 'ax' into regs->ax, it was 
> > carried in via a pointer and then returned, where the generic 
> > syscall return code copied it to regs->ax.
> >
> > So there was never any deeper reason for this suboptimal pattern, 
> > it was simply never noticed after being introduced.
> >
> > (Btw., the regs->ax we return will be copied back to regs->ax 
> > after the syscall straight away once again - but I guess this 
> > cannot be helped.)
> 
> The 64-bit stub could skip saving it back to regs.

Yeah, but at the cost of having a duplicated entry stub, right?

Thanks,

	Ingo