Re: [Xen-devel] xen crash with 4.17 kernel on Fedora

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 01/07/18 17:43, Michael Young wrote:
> I am seeing crash on boot and DomU (pv) on Fedora with the 4.17 kernel
> (eg. kernel-4.17.2-200.fc28.x86_64 and kernel-4.17.3-200.fc28.x86_64)
> which
> didn't occur with 4.16 kernel (eg. kernel-4.16.16-300.fc28.x86_64)
>
> The backtrace for a Dom0 boot of xen-4.10.1-5.fc28.x86_64 running
> kernel-4.17.2-200.fc28.x86_64 is
>
> (XEN) d0v0 Unhandled general protection fault fault/trap [#13, ec=0000]
> (XEN) domain_crash_sync called from entry.S: fault at ffff82d08035557c
> x86_64/entry.S#create_bounce_frame+0x135/0x159
> (XEN) Domain 0 (vcpu#0) crashed on cpu#0:
> (XEN) ----[ Xen-4.10.1  x86_64  debug=n   Not tainted ]----
> (XEN) CPU:    0
> (XEN) RIP:    e033:[<ffffffff81062330>]
> (XEN) RFLAGS: 0000000000000246   EM: 1   CONTEXT: pv guest (d0v0)
> (XEN) rax: 0000000000000246   rbx: 00000000ffffffff   rcx:
> 0000000000000000
> (XEN) rdx: 0000000000000000   rsi: 00000000ffffffff   rdi:
> 0000000000000000
> (XEN) rbp: 0000000000000000   rsp: ffffffff82203d90   r8: 
> ffffffff820bb698
> (XEN) r9:  ffffffff82203e38   r10: 0000000000000000   r11:
> 0000000000000000
> (XEN) r12: 0000000000000000   r13: ffffffff820bb698   r14:
> ffffffff82203e38
> (XEN) r15: 0000000000000000   cr0: 0000000080050033   cr4:
> 00000000000006e0
> (XEN) cr3: 000000001aacf000   cr2: 0000000000000000
> (XEN) fsb: 0000000000000000   gsb: ffffffff82731000   gss:
> 0000000000000000
> (XEN) ds: 0000   es: 0000   fs: 0000   gs: 0000   ss: e02b   cs: e033
> (XEN) Guest stack trace from rsp=ffffffff82203d90:
> (XEN)    0000000000000000 0000000000000000 0000000000000000
> ffffffff81062330
> (XEN)    000000010000e030 0000000000010046 ffffffff82203dd8
> 000000000000e02b
> (XEN)    0000000000000246 ffffffff8110e019 0000000000000000
> 0000000000000246
> (XEN)    0000000000000000 0000000000000000 ffffffff820a6cd8
> ffffffff82203e88
> (XEN)    ffffffff82739000 8000000000000061 0000000000000000
> 0000000000000000
> (XEN)    ffffffff8110ecb6 0000000000000008 ffffffff82203e98
> ffffffff82203e58
> (XEN)    0000000000000000 0000000000000000 8000000000000161
> 0000000000000100
> (XEN)    fffffffffffffeff 0000000000000000 0000000000000000
> ffffffff82203ef0
> (XEN)    ffffffff810ac990 0000000000000000 0000000000000000
> 0000000000000000
> (XEN)    0000000000000000 0000000000000000 8000000000000161
> 0000000000000100
> (XEN)    fffffffffffffeff 0000000000000000 0000000000000000
> 0000000002739000
> (XEN)    0000000000000080 ffffffff8275db62 000000000001a739
> 0000000000000000
> (XEN)    0000000000000000 0000000000000000 0000000000000000
> ffffffff81037c80
> (XEN)    007fffff8275efe7 ffffffff82739000 ffffffff81037f18
> ffffffff8102aaf0
> (XEN)    ffffffff8275dc8c 0000000000000000 0000000000000000
> 0000000000000000
> (XEN)    0000000000000000 0000000000000000 0000000000000000
> 0000000000000000
> (XEN)    0000000000000000 0000000000000000 0000000000000000
> 0000000000000000
> (XEN)    0000000000000000 0000000000000000 0000000000000000
> 0000000000000000
> (XEN)    0000000000000000 0000000000000000 0000000000000000
> 0000000000000000
> (XEN)    0000000000000000 0000000000000000 0f00000060c0c748
> ccccccccccccc305
>
> where
> addr2line -f -e vmlinux ffffffff81062330
> gives
> native_irq_disable
> /usr/src/debug/kernel-4.17.fc28/linux-4.17.2-200.fc28.x86_64/./arch/x86/include/asm/irqflags.h:44
>
>
> What is the problem or how might it be debugged?

The guest is executing a native `cli` instruction which is privileged
and we don't allow (we could trap & emulate, but we can't provide proper
STI-shadow behaviour, and such a guest might also expect popf to work,
which is very much doesnt).  In Linux, that codepath should be using a
pvop, rather than a native op.

It is either a subsystem which should be skipped when virtualised, or a
poorly coded subsystem, or a buggy setup path.

Can you see about trying to boot the old kernel as dom0, and the new
kernel as a domU with pause on crash configured? 
/usr/libexec/xen/bin/xenctx should be able to pull a backtrace out of
the crashed domain state if you pass the appropriate symbol table in.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel