lkml.org 
[lkml]   [2020]   [Feb]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 0/2] virtio: decouple protected guest RAM form VIRTIO_F_IOMMU_PLATFORM
From
Date

On 2020/2/24 下午9:56, Halil Pasic wrote:
> On Mon, 24 Feb 2020 17:26:20 +0800
> Jason Wang <jasowang@redhat.com> wrote:
>
>> That's better.
>>
>> How about attached?
>>
>> Thanks
> Thanks Jason! It does avoid the translation overhead in vhost.
>
> Tested-by: Halil Pasic <pasic@linux.ibm.com>
>
> Regarding the code, you fence it in virtio-net.c, but AFAIU this feature
> has relevance for other vhost devices as well. E.g. what about vhost
> user? Would it be the responsibility of each virtio device to fence this
> on its own?


Yes, it looks to me it's better to do that in virtio_set_features_nocheck()


>
> I'm also a bit confused about the semantics of the vhost feature bit
> F_ACCESS_PLATFORM. What we have specified on virtio level is:
> """
> This feature indicates that the device can be used on a platform where
> device access to data in memory is limited and/or translated. E.g. this
> is the case if the device can be located behind an IOMMU that translates
> bus addresses from the device into physical addresses in memory, if the
> device can be limited to only access certain memory addresses or if
> special commands such as a cache flush can be needed to synchronise data
> in memory with the device. Whether accesses are actually limited or
> translated is described by platform-specific means. If this feature bit
> is set to 0, then the device has same access to memory addresses
> supplied to it as the driver has. In particular, the device will always
> use physical addresses matching addresses used by the driver (typically
> meaning physical addresses used by the CPU) and not translated further,
> and can access any address supplied to it by the driver. When clear,
> this overrides any platform-specific description of whether device
> access is limited or translated in any way, e.g. whether an IOMMU may be
> present.
> """
>
> I read this like the addresses may be IOVAs which require
> IMMU translation or GPAs which don't.
>
> On the vhost level however, it seems that F_IOMMU_PLATFORM means that
> vhost has to do the translation (via IOTLB API).


Yes.


>
> Do I understand this correctly? If yes, I believe we should document
> this properly.


Good point. I think it was probably wrong to tie F_IOMMU_PLATFORM to
IOTLB API. Technically IOTLB can work with GPA->HVA mapping. I
originally use a dedicated feature bit (you can see that from commit
log), but for some reason Michael tweak it to virtio feature bit. I
guess it was probably because at that time there's no way to specify e.g
backend capability but now we have VHOST_GET_BACKEND_FEATURES.

For now, it was probably too late to fix that but document or we can add
the support of enabling IOTLB via new backend features.


>
> BTW I'm still not 100% on the purpose and semantics of the
> F_ACCESS_PLATFORM feature bit. But that is a different problem.


Yes, I aggree that we should decouple the features that does not belongs
to device (protected, encrypted, swiotlb etc) from F_IOMMU_PLATFORM. But
Michael and other have their points as well.

Thanks


>
> Regards,
> Halil
>

\
 
 \ /
  Last update: 2020-02-25 04:31    [W:0.147 / U:0.192 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site