On Mon, Mar 05, 2018 at 10:23:59AM -0800, Linus Torvalds wrote:
> On Mon, Mar 5, 2018 at 5:12 AM, Joerg Roedel <j...@8bytes.org> wrote:
> >
> >> The things is, we *know* that we will restore two segment registers with
> >> the
> >> user cr3 already loaded: CS and SS get restored with the final iret.
> >
> > Yeah, I know, but the iret-exception path is fine because it will
> > deliver a SIGILL and doesn't return to the faulting iret.
>
> That's not so much my worry, as just getting %cr3 wrong. The fact is,
> we still take the exception, and we still have to handle it, and that
> still needs to get the user<->kernel cr3 right.
Right, as I said, up to v2 of this series I thought I could avoid the
whole from-kernel-with-user-cr3 game, but that turned out to be wrong.
Now I added the necessary check and handling for it, as at least the
#DB handler needs it.
> So then the whole "restore segments early" must be wrong, because
> *that* path must get it all right too, no?
>
> And it appears that the code *does* get it right, and you can just
> avoid this patch entirely?
Right, I will drop this patch.
>
> > The iret-exception case is tested by the ldt_gdt selftest (the
> > do_multicpu_tests subtest). But I didn't actually tested single-stepping
> > through sysenter yet. I just re-ran the same tests I did with v2 on this
> > patch-set.
>
> Ok. Maybe we should have a test for the "take DB on first instruction
> of sysenter".
I put a selftest for that on my list of things to look into. I'll have
no idea how difficult this will be, but I certainly find out :)
Regards,
Joerg
