Re: [Xen-devel] [PATCH] x86/hvm: Set the emulation context correctly in hvmemul_cmpxchg()

[Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>]

On 10/05/2016 05:05 PM, Jan Beulich wrote:
>>>> On 05.10.16 at 15:54, <rcojocaru@xxxxxxxxxxxxxxx> wrote:
>> On 10/05/2016 04:43 PM, Jan Beulich wrote:
>>> So with this I wonder btw. why your patch (mostly) fixing this
>>> shortcoming (while adding proper LOCK handling) never made it
>>> to a version that could be committed.
>>
>> I was under the impression that your stand on the rwlock patch had
>> remained that you prefer a stub version to it, for possible performance
>> reasons, hence I've not pressed the issue. If I've misunderstood I'm
>> happy to try to rework it for staging.
>>
>> I thought that the only acceptable solution was adding an actual stub
>> running on the physical VCPU, and unfortunately I didn't get to work one
>> out, in part because I had to tackle other issues, and partly because
>> it's not very clear how to go about that in this case.
> 
> Hmm, I have to admit I don't recall any stubs to be in the picture
> here. What I recall is that the locked region was too large, and
> covered cases which don't need a lock in the first place.

Andrew I think suggested a stub first:

https://lists.xenproject.org/archives/html/xen-devel/2016-04/msg02050.html

then George brought it up again:

https://lists.xenproject.org/archives/html/xen-devel/2016-04/msg03294.html

Andrew also talked about XenServer's performance testing with the
original patch:

https://lists.xenproject.org/archives/html/xen-devel/2016-04/msg03354.html

There's actually a version of it in XenServer's patch queue for 4.7:

https://github.com/xenserver/xen-4.7.pg/blob/master/master/xen-x86-emulate-syncrhonise-LOCKed-instruction-emulation.patch

If there's no impediment, I'm happy to start working on it again.


Thanks,
Razvan

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