Re: [Xen-devel] Regression in guest destruction caused by altp2m

["Sahita, Ravi" <ravi.sahita@xxxxxxxxx>]

>From: Andrew Cooper [mailto:amc96@xxxxxxxxxxxxxxxx] On Behalf Of
>Andrew Cooper
>Sent: Monday, July 27, 2015 4:22 PM
>
>On 27/07/2015 19:09, Wei Liu wrote:
>> On Mon, Jul 27, 2015 at 04:18:15PM +0100, Wei Liu wrote:
>>> Found this when I did "xl destroy" to a hvm guest *without* altp2m
>>> turned on.
>>>
>>> Current staging branch.
>>>
>>> (XEN) ----[ Xen-4.6-unstable  x86_64  debug=y  Tainted:    C ]----
>>> (XEN) CPU:    0
>>> (XEN) RIP:    e008:[<ffff82d0801fce0d>] vmx_vmenter_helper+0x263/0x2f3
>>> (XEN) RFLAGS: 0000000000010242   CONTEXT: hypervisor (d0v0)
>>> (XEN) rax: 000000000000201a   rbx: ffff8300cf5f9000   rcx: 0000000000000000
>>> (XEN) rdx: 0000000000000200   rsi: ffff830225fc5000   rdi: ffff8300cf5f9000
>>> (XEN) rbp: ffff8300cf30fde8   rsp: ffff8300cf30fdd8   r8:  ffff8300cf0fc5e0
>>> (XEN) r9:  00000016e7ea5ca3   r10: 0000000000000000   r11:
>00000016f1000b66
>>> (XEN) r12: ffff830227b18740   r13: ffff830227bda250   r14: ffff830227bda240
>>> (XEN) r15: 0000000000000000   cr0: 0000000080050033   cr4:
>00000000000026e0
>>> (XEN) cr3: 000000000f03a000   cr2: 00007fb13dc65150
>>> (XEN) ds: 0000   es: 0000   fs: 0000   gs: 0000   ss: 0000   cs: e008
>>> (XEN) Xen stack trace from rsp=ffff8300cf30fdd8:
>>> (XEN)    ffff8300cf5f9000 ffff8300cf5f9000 ffff8300cf30fdf8 ffff82d0801d95fb
>>> (XEN)    ffff8300cf30fe18 ffff82d0802145eb ffff82d08012d455
>ffff830227bda250
>>> (XEN)    ffff8300cf30fe48 ffff82d0801d3e0c ffff8300cf5f9000
>0000000000000001
>>> (XEN)    ffff830227bdaf28 ffff830227bda000 ffff8300cf30fe68
>ffff82d080174c6f
>>> (XEN)    ffff8300cf5f9000 ffff8300cf5f9000 ffff8300cf30fe98
>ffff82d080105e07
>>> (XEN)    ffff82d0803545c0 0000000000000000 0000000000000000
>ffff8300cf308000
>>> (XEN)    ffff8300cf30fec8 ffff82d08012148a ffff82d080328080
>ffff82d080328080
>>> (XEN)    ffff82d080328080 ffffffffffffffff ffff8300cf30fef8 ffff82d08012cbe7
>>> (XEN)    ffff8300ce68e000 00007fb13dc65150 ffffea00000715a0
>0000000000000000
>>> (XEN)    ffff8300cf30ff08 ffff82d08012cc3f 00007cff30cf00c7
>ffff82d08024dd51
>>> (XEN)    0000000000000000 0000000000000000 ffffea00000715a0
>00007fb13dc65150
>>> (XEN)    ffff8800cd4ff3d8 ffffea0000267a50 ffff880000000328
>00000000000000a9
>>> (XEN)    ffffea00000715a0 ffff8800cd4ff3d8 ffff880207c997b9
>0000000000267a50
>>> (XEN)    00007fb13dc65150 ffff8800cd4ff3d8 ffffea0000267a50
>000000fa00000000
>>> (XEN)    ffffffff8113d09d 000000000000e033 0000000000000202
>ffff88000b807d68
>>> (XEN)    000000000000e02b 000000000000beef 000000000000beef
>000000000000beef
>>> (XEN)    000000000000beef 0000000000000000 ffff8300ce68e000
>0000000000000000
>>> (XEN)    0000000000000000
>>> (XEN) Xen call trace:
>>> (XEN)    [<ffff82d0801fce0d>] vmx_vmenter_helper+0x263/0x2f3
>>> (XEN)    [<ffff82d0801d95fb>] altp2m_vcpu_update_p2m+0x12/0x15
>>> (XEN)    [<ffff82d0802145eb>] altp2m_vcpu_destroy+0x6b/0x94
>>> (XEN)    [<ffff82d0801d3e0c>] hvm_vcpu_destroy+0x5a/0xa7
>>> (XEN)    [<ffff82d080174c6f>] vcpu_destroy+0x5d/0x72
>>> (XEN)    [<ffff82d080105e07>] complete_domain_destroy+0x49/0x186
>>> (XEN)    [<ffff82d08012148a>] rcu_process_callbacks+0x144/0x1a5
>>> (XEN)    [<ffff82d08012cbe7>] __do_softirq+0x82/0x8d
>>> (XEN)    [<ffff82d08012cc3f>] do_softirq+0x13/0x15
>>> (XEN)    [<ffff82d08024dd51>] process_softirqs+0x21/0x30
>>> (XEN)
>>> (XEN)
>>> (XEN) ****************************************
>>> (XEN) Panic on CPU 0:
>>> (XEN) FATAL TRAP: vector = 6 (invalid opcode)
>>> (XEN) ****************************************
>>> (XEN)
>> This makes me think it doesn't matter if I do "xl destroy" or just
>> normal shutdown. And another test to normally shutdown guest confirmed
>> that.
>>
>> The machine I have is quite old and doesn't support EPT. The real
>> issues seems to be that you failed to properly disable that feature
>> for old machines.
>
>The altp2m feature is designed (and implemented) to be emulated on
>systems lacking hardware capabilities.
>
>As such, teardown is necessary even on older systems, but I would agree that
>there appears to be a bug.

Yes - 
the check for hvm_altp2m_supported() before altp2m_vcpu_destroy() was missing - 
will send a patch tonight.

Ravi


>
>~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel