Re: [PATCH v2] KVM: trace the events of mmu_notifier

[Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>]

On 09/16/2012 05:58 PM, Avi Kivity wrote:
> On 09/14/2012 08:59 AM, Xiao Guangrong wrote:
>> On 09/10/2012 05:26 PM, Xiao Guangrong wrote:
>>> On 09/10/2012 05:09 PM, Avi Kivity wrote:
>>>> On 09/07/2012 09:16 AM, Xiao Guangrong wrote:
>>>>> mmu_notifier is the interface to broadcast the mm events to KVM, the
>>>>> tracepoints introduced in this patch can trace all these events, it is
>>>>> very helpful for us to notice and fix the bug caused by mm
>>>>
>>>> There is nothing kvm specific here.  Perhaps this can be made generic
>>>> (with a mm parameter so we can filter by process).
>>>
>>> Hmm, i would like to put these tracepoints in the mmu-lock then we can clearly
>>> know the sequence between mm and kvm mmu. It is useful for us to detect the
>>> issue/race between them.
>>>
>>
>> Ping...?
> 
> Sorry.  Yes you are right, knowing the exact sequence is valuable.  Yet
> it will be hard to associate these events with the mmu since we don't
> have gpas there.
> 

Avi,

I have some patches in my local queue which use tracepoints instead of
rmap_printk, they can track rmap_add/rmap_remove/rmap_write_protect.

Though we can not directly get the gfn from these mmu-notifier events, it
can be got from the later events. We can see what mmu is doing when the
notifier events are triggered.