On 25/07/2022 16:51, Bertrand Marquis wrote:
> Hi,
>
> On our CI we have randomly a crash during guest boot on x86.
>
> We are running on qemu x86_64 using Xen staging.
> The crash is happening randomly (something like 1 out of 20 times).
>
> This is always happening on the first guest we start, we never got it after 
> first guest was successfully started.
>
> Please tell me if you need any other info.
>
> Here is the guest kernel log:
> <snip>
> [    0.000000] Hypervisor detected: Xen PV
> <snip>
> [    6.679020] general protection fault, maybe for address 0x8800: 0000 [#1] 
> PREEMPT SMP NOPTI
> [    6.679020] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.17.6 #1
> [    6.679020] RIP: e030:error_entry+0xaf/0xe0
> [    6.679020] Code: 29 89 c8 48 39 84 24 88 00 00 00 74 15 48 81 bc 24 88 00 
> 00 00 63 10 e0 81 75 03 0f 01 f8 90 90 90 c3 48 89 8c 24 88 00 00 00 <0f> 01 
> f8 90 90 90 eb 11 0f 20 d8 90 90 90 90 90 48 25 ff e7 ff ff
> [    6.679020] RSP: e02b:ffffffff82803a90 EFLAGS: 00000046
> [    6.679020] RAX: 0000000000008800 RBX: 0000000000000000 RCX: 
> ffffffff81e00fa7
> [    6.679020] RDX: 0000000000000000 RSI: ffffffff81e009f8 RDI: 
> 00000000000000eb
> [    6.679020] RBP: 0000000000000000 R08: 0000000000000000 R09: 
> 0000000000000000
> [    6.679020] R10: 0000000000000000 R11: 0000000000000000 R12: 
> 0000000000000000
> [    6.679020] R13: 0000000000000000 R14: 0000000000000000 R15: 
> 0000000000000000
> [    6.679020] FS:  0000000000000000(0000) GS:ffff88801f200000(0000) 
> knlGS:0000000000000000
> [    6.679020] CS:  10000e030 DS: 0000 ES: 0000 CR0: 0000000080050033
> [    6.679020] CR2: 0000000000000000 CR3: 000000000280c000 CR4: 
> 0000000000050660
> [    6.679020] Call Trace:
> [    6.679020]  <TASK>
>

0f 01 f8 is SWAPGS

This is a privileged instruction, and has never been permitted under Xen
PV.  This should have been excluded by pvops.

This is a Linux bug, not a Xen bug.

I can't explain why you're only seeing it intermittently.  Perhaps
error_entry is broken by default, and pvops fixes things up, but an
error early enough takes a non-fixed-up path :-/

~Andrew

Reply via email to