Re: [PATCH 0 of 6 V5] amd iommu: support ats/gpgpu passthru on iommuv2 systems

[Wei Wang <wei.wang2@amd.com>]

On 02/13/2012 05:54 PM, Ian Jackson wrote:
> Wei Wang writes ("[PATCH 0 of 6 V5] amd iommu: support ats/gpgpu passthru on iommuv2 systems"):
>> This is patch set v5. It includes all pending patches that are needed to enable gpgpu passthrough and heterogeneous computing in guest OSes. Basically, this patch set gives guest VM the same capability of running openCL applications on amd platforms as native OSes. Upstream Linux 3.3 rc2 with amd iommuv2 kernel driver has been tested well as guest OS, and since last submission, lots of regression tests have been done to make sure this does not break non-iommuv2 systems. Please review it, feedbacks are appreciated.
>
> Thanks.  I'm not qualified to review the hypervisor parts, but the
> explanation seems to make sense and the tools parts look OK.
>
> So as for these three:
>
>   [PATCH 4 of 6 V5] libxc: add wrappers for new hypercalls
>   [PATCH 5 of 6 V5] libxl: bind virtual bdf to physical bdf ...
>   [PATCH 6 of 6 V5] libxl: Introduce a new guest config file parameter
>
> Acked-by: Ian Jackson<ian.jackson@eu.citrix.com>
>

Cool! thanks a lot for reviewing it and moving this forward.
Issues you mentioned will be fixed in my next post.

Thanks,
Wei

> I do have a couple of minor niggles which you might like to address if
> you repost:
>
>> libxl: Introduce a new guest config file parameter
>> Use guest_iommu = {1,0} to enable or disable guest iommu emulation.
>> Default value is 0. Regression tests have been done to make sure
>> it does not break non-iommuv2 systems.
>
> It's conventional to leave a blank line between the summary line and
> the bulk of the description.
>
>> diff -r c39f5736e364 -r 09721a5ff844 docs/man/xl.cfg.pod.5
>> --- a/docs/man/xl.cfg.pod.5	Fri Feb 10 15:49:19 2012 +0100
>> +++ b/docs/man/xl.cfg.pod.5	Fri Feb 10 15:49:20 2012 +0100
>> @@ -820,6 +820,10 @@ certainly belong in a more appropriate s
> ...
>> +=item B<guest_iommu=BOOLEAN>
>> +
>> +Enable virtual iommu device for hvm guest. It should be enabled to passthrough AMD GPGPU.
>> +
>
> It would be nice for that line to be wrapped to fit within 75-80
> columns.
>
> Thanks,
> Ian.
>