Re: BUG: unable to handle kernel paging request in __switch_to

[Dmitry Vyukov]

On Thu, Dec 14, 2017 at 10:39 PM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Thu, Dec 14, 2017 at 1:27 PM, Andy Lutomirski <luto@xxxxxxxxxx> wrote:
>> On Thu, Dec 14, 2017 at 11:28 AM, Linus Torvalds
>> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>>> I don't think that's the case. "int3" is entirely synchronous, and
>>> doesn't have the same odd issues as a breakpoint trap (which honors RF
>>> etc). It's literally just a one-byte shorthand for "int $3".
>>
>> The SDM says precisely the same thing about INT N, so, whichever way
>> you dice it, int3 is a benign exception.
>
> That just means that it doesn't double-fault when it takes the page fault.
>
> Which we already know, because we see a page fault, not a double fault.
>
>> 0xfffffffffffffff8 is *exactly* where the fault would be if the
>> microcoded push of SS faulted if the IST contained zeros.
>
> Yes, I suspect it's the stack that is buggered for some reason.
>
>>> Plus I think the instruction that gets overwritten is just a 5-byte
>>> nop isn't it? So it really shouldn't take a fault without the "int3"
>>> overwriting.
>>
>> Unless it was being overwritten the other way and the oops hit while
>> tracing was being turned *off*.
>
> Doesn't really matter. The two forms of that instruction are "5-byte
> nop" and "unconditional branch".
>
> Neither of them will write to anything - the only page fault they
> could take is for instruction fetch.
>
> So it really must be the "int3" that fails. Unless we're looking at
> some odd CPU errata, which sounds very very unlikely.

FTR the commit is:

commit d127129e85a020879f334154300ddd3f7ec21c1e (HEAD, tag: next-20171129)
Author: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
Date:   Wed Nov 29 14:09:56 2017 +1100
    Add linux-next specific files for 20171129

You can get it from
git://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next-history.git
Compiler is this: https://storage.googleapis.com/syzkaller/gcc-7.tar.gz
Config was attached.

I've built this exact kernel and here is __switch_to disasm:
https://gist.githubusercontent.com/dvyukov/8137559f7da08fbe32f9018972a4498c/raw/0ef2abf723b117f0d0f0306fd50e216d50c5cecb/gistfile1.txt

__switch_to+0x95b seems to point to (?):

ffffffff81252f6b: 0f 1f 44 00 00        nopl   0x0(%rax,%rax,1)

which is branch target alignment nop.

We have a bunch of semi-similar non-sense crashes on syzbot:

https://groups.google.com/forum/#!msg/syzkaller-bugs/zGz7AVtMBV0/X_-CPbjNAgAJ
https://groups.google.com/forum/#!msg/syzkaller-bugs/9nMSJo9jmGs/tkRYgZ-XAwAJ
https://groups.google.com/forum/#!msg/syzkaller-bugs/04-q4OZrerA/XfYdNnWXAwAJ
https://groups.google.com/forum/#!msg/syzkaller-bugs/6iC6rPtAHKQ/UiZ4fnWXAwAJ
https://groups.google.com/forum/#!msg/syzkaller-bugs/2zSDbzRIH_k/SLCMqmeXAwAJ
https://groups.google.com/forum/#!msg/syzkaller-bugs/uEsjx8VISco/Mwu_pbGWAwAJ
https://groups.google.com/forum/#!msg/syzkaller-bugs/kZ6Z7UQLbCQ/JHpjTGeXAwAJ
https://groups.google.com/forum/#!msg/syzkaller-bugs/UjYsJxiGxwU/mponQq2XAwAJ

Lots of them are on 0xfffffffffffffff8 address.

I have some suspicion towards KVM. Potentially a nested KVM messed
host processor state (CRn or page tables) so that then we get these
weird crashes.

One question: how would triple-fault look like? I am asking because we
have hundreds of cases where kernel just starts silently rebooting
while running some unprivileged syscalls:
https://groups.google.com/forum/#!msg/syzkaller-bugs/w8dkVNrgzrc/4mLJLOAbCgAJ
Can these be triple faults? Reproducer for that one also seems to be
related to KVM.