[CRIU] Bug report: a process restored with criu crashes on SIGFPE

[Andrei Vagin]

On Fri, Jan 26, 2018 at 02:02:35PM +0200, Shlomi Matichin wrote:
>    0x00007fc88f5c7788: jmpq   0x7fc88f5c7706
>    0x00007fc88f5c778d: nopl   (%rax)
>    0x00007fc88f5c7790: movl   $0x0,0x4(%r13)
>    0x00007fc88f5c7798: jmpq   0x7fc88f5c76fc
>    0x00007fc88f5c779d: nopl   (%rax)
>    0x00007fc88f5c77a0: movl   $0x0,0x0(%r13)
>    0x00007fc88f5c77a8: jmpq   0x7fc88f5c76ee
>    0x00007fc88f5c77ad: pxor   %xmm3,%xmm3
> => 0x00007fc88f5c77b1: cvtsd2ss %xmm0,%xmm3

Looks like we incorrectly save fpu state...

I tried to reproduce this issue on my laptop, in digital ocean with the same
ubuntu version, and I even created an aws account. But in all my attempts,
everything works as expected.

Could you write a detailed instruction how you reproduce this bug on a
ec2 instance? What image do you use? Maybe you can create there an
instance and give us access to it?

Could you show registers before and after dump/restore?

For that you need to attach to a target process by gdb and call " info
all-registers", then you need to detach from a process, do dump/restore
and attach again to get registers. Pls, don't forget to check that the
issue is reproduced in this case.

Thanks,
Andrei