Re: [RFC] de-asmify the x86-64 system call slowpath

[Al Viro <viro@ZenIV.linux.org.uk>]

On Mon, Jan 27, 2014 at 02:43:14PM -0800, Linus Torvalds wrote:
> On Mon, Jan 27, 2014 at 2:32 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> >
> > do_signal() is also a place where arbitrary changes to regs might've
> > been done by tracer, so regs->cs might need to be checked in the same
> > place where we validate regs->rip ;-/
> 
> Fair enough. But it would still be really easy, and make the common
> case signal delivery a bit faster.
> 
> Now, sadly, most signal delivery is then followed by sigreturn (the
> exceptions being dying or doing a longjmp), so we'd still get the
> iretq then. But it would cut the iretq's related to signals in half.
> 
> We *could* try to do sigreturn with sysret and a small trampoline too,
> of course. But I'm not sure how far I'd want to take it.

The problem with validation is that we'll suddenly become sensitive to
hard-to-estimate pile of hardware bugs ;-/  E.g. which AMD-specific
errata is that comment in entry_64.S about?  The one I kinda-sorta
remember is Intel-specific, and that was about uncanonical RIP; looking
for AMD one has turned #353 (with suggested workaround being "have bit 16 set
in whatever you put into R11"), but I've no idea whether that's the only
potential issue there...