Re: [PATCH 3/9] x86/entry/32: fix the end of the stack for newly forked tasks

[Andy Lutomirski]

On Sep 20, 2016 5:25 PM, "Josh Poimboeuf" <jpoimboe@xxxxxxxxxx> wrote:
>
> On Tue, Sep 20, 2016 at 09:10:55PM -0400, Brian Gerst wrote:
> > Dropping asmlinkage from schedule_tail() would be a better option if possible.
>
> My understanding is that it's still needed for ia64.  AFAICT, ia64
> relies on schedule_tail() having the syscall_linkage function attribute.
> From the gcc manual:
>
>   This attribute is used to modify the IA64 calling convention by
>   marking all input registers as live at all function exits. This makes
>   it possible to restart a system call after an interrupt without having
>   to save/restore the input registers. This also prevents kernel data
>   from leaking into application code.

/me needs to excise this from i386.  The amount of BS code involved to
avoid a whopping *six* register saves per syscall was absurd.

>
> And the ia64 entry code has some similar language:
>
>                 /*
>                  * Invoke schedule_tail(task) while preserving in0-in7, which may be needed
>                  * in case a system call gets restarted.
>                  */
>         GLOBAL_ENTRY(ia64_invoke_schedule_tail)
>                 ...

That comment has to be wrong.  What syscall could possibly be
restarted across schedule_tail()?  It's a brand new thread and has
literally never done a syscall.

There may be another reason that the registers are live there, but I
generally do my best to never look at ia64 asm code.