Re: [Xen-devel] [xen-unstable-smoke test] 127247: regressions - FAIL

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 04/09/18 09:53, Wei Liu wrote:
> On Tue, Sep 04, 2018 at 09:49:11AM +0100, Andrew Cooper wrote:
>> On 04/09/18 09:45, Wei Liu wrote:
>>> On Tue, Sep 04, 2018 at 07:55:04AM +0000, osstest service owner wrote:
>>>> flight 127247 xen-unstable-smoke real [real]
>>>> http://logs.test-lab.xenproject.org/osstest/logs/127247/
>>>>
>>>> Regressions :-(
>>>>
>>>> Tests which did not succeed and are blocking,
>>>> including tests which could not be run:
>>>>  test-amd64-amd64-xl-qemuu-debianhvm-i386 10 debian-hvm-install fail REGR. 
>>>> vs. 127212
>>> Sep  4 06:49:12.366727 (XEN) MMIO emulation failed (1): d107v0 32bit @ 
>>> 0000:00100000 -> 55 89 e5 53 83 ec 10 8b 55 0c b8 20 02 08 00 8b
>>> Sep  4 06:49:12.378690 (XEN) d107v0 Triple fault - invoking HVM shutdown 
>>> action 1
>>> Sep  4 06:49:12.378727 (XEN) *** Dumping Dom107 vcpu#0 state: ***
>>> Sep  4 06:49:12.378755 (XEN) ----[ Xen-4.12-unstable  x86_64  debug=y   Not 
>>> tainted ]----
>>> Sep  4 06:49:12.390738 (XEN) CPU:    0
>>> Sep  4 06:49:12.390796 (XEN) RIP:    0000:[<0000000000100000>]
>>> Sep  4 06:49:12.390846 (XEN) RFLAGS: 0000000000010002   CONTEXT: hvm guest 
>>> (d107v0)
>>> Sep  4 06:49:12.402729 (XEN) rax: 0000000000000000   rbx: 00000000001a8000  
>>>  rcx: 0000000000000000
>>> Sep  4 06:49:12.402796 (XEN) rdx: 0000000000000000   rsi: 0000000000000000  
>>>  rdi: 0000000000000000
>>> Sep  4 06:49:12.414742 (XEN) rbp: 0000000000000000   rsp: 0000000000000000  
>>>  r8:  0000000000000000
>>> Sep  4 06:49:12.426728 (XEN) r9:  0000000000000000   r10: 0000000000000000  
>>>  r11: 0000000000000000
>>> Sep  4 06:49:12.426795 (XEN) r12: 0000000000000000   r13: 0000000000000000  
>>>  r14: 0000000000000000
>>> Sep  4 06:49:12.438672 (XEN) r15: 0000000000000000   cr0: 0000000000000011  
>>>  cr4: 0000000000000000
>>> Sep  4 06:49:12.450724 (XEN) cr3: 0000000000000000   cr2: 0000000000000000
>>> Sep  4 06:49:12.450787 (XEN) fsb: 0000000000000000   gsb: 0000000000000000  
>>>  gss: 0000000000000000
>>> Sep  4 06:49:12.462739 (XEN) ds: 0000   es: 0000   fs: 0000   gs: 0000   
>>> ss: 0000   cs: 0000
>>> Sep  4 06:49:12.462805 [  647.449914] xenbr0: port 3(vif107.0-emu) entered 
>>> disabled state
>>>
>>> Looking at the range of changeset there is one potential culprit:
>>>
>>>     x86/hvm: Fix mapping corner case during task switching
>> You've already R-by the patch which fixes this bug.
>>
>> HVMLoader doesn't work too well when it tries to execute
>> hypercall_xen_version first, without a working stack.
>>
>> There is a separate bug here which is xl/libxl's approach to crashing
>> guests.  If you notice, it reboots ad-infinitum until OSSTest times
>> out.  It is nigh on impossible to stop a runaway domain like this with
>> libxl.  The best I've come up with is to mv the kernel binary sideways
>> so the next reboot gets ESRCH.
> I'm not sure how to fix that. The problem is libxl is doing exactly what
> the user asked for: reboot the domain after it crashed. It is just that
> xl/libxl does it very quickly.
>
> Maybe we can add some adaptive delay into xl to avoid it going too
> quickly?

Xapi has a blanket "if it crashed in less than 5 seconds, never reboot"
setting.

Alternatively, set HVM_PARAM_TRIPLE_FAULT_REASON to crash.  Despite
being non-architectural, I'm debating doing this in the hypervisor,
because there is no reason a guest should need to resort to a triple
fault reboot.

~Andrew

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