Re: [RFC PATCH 0/7] A General Accelerator Framework, WarpDrive

[Kenneth Lee]

On Thu, Aug 09, 2018 at 08:31:31AM +0000, Tian, Kevin wrote:
> Date: Thu, 9 Aug 2018 08:31:31 +0000
> From: "Tian, Kevin" <kevin.tian@xxxxxxxxx>
> To: Kenneth Lee <liguozhu@xxxxxxxxxxxxx>, Jerome Glisse <jglisse@xxxxxxxxxx>
> CC: Kenneth Lee <nek.in.cn@xxxxxxxxx>, Alex Williamson
>  <alex.williamson@xxxxxxxxxx>, Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>,
>  "kvm@xxxxxxxxxxxxxxx" <kvm@xxxxxxxxxxxxxxx>, Jonathan Corbet
>  <corbet@xxxxxxx>, Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>, Zaibo
>  Xu <xuzaibo@xxxxxxxxxx>, "linux-doc@xxxxxxxxxxxxxxx"
>  <linux-doc@xxxxxxxxxxxxxxx>, "Kumar, Sanjay K" <sanjay.k.kumar@xxxxxxxxx>,
>  Hao Fang <fanghao11@xxxxxxxxxx>, "linux-kernel@xxxxxxxxxxxxxxx"
>  <linux-kernel@xxxxxxxxxxxxxxx>, "linuxarm@xxxxxxxxxx"
>  <linuxarm@xxxxxxxxxx>, "iommu@xxxxxxxxxxxxxxxxxxxxxxxxxx"
>  <iommu@xxxxxxxxxxxxxxxxxxxxxxxxxx>, "linux-crypto@xxxxxxxxxxxxxxx"
>  <linux-crypto@xxxxxxxxxxxxxxx>, Philippe Ombredanne
>  <pombredanne@xxxxxxxx>, Thomas Gleixner <tglx@xxxxxxxxxxxxx>, "David S .
>  Miller" <davem@xxxxxxxxxxxxx>, "linux-accelerators@xxxxxxxxxxxxxxxx"
>  <linux-accelerators@xxxxxxxxxxxxxxxx>
> Subject: RE: [RFC PATCH 0/7] A General Accelerator Framework, WarpDrive
> Message-ID: <AADFC41AFE54684AB9EE6CBC0274A5D1912B39B3@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>
> 
> > From: Kenneth Lee [mailto:liguozhu@xxxxxxxxxxxxx]
> > Sent: Thursday, August 9, 2018 4:04 PM
> > 
> > But we have another requirement which is to combine some device
> > together to
> > share the same address space. This is a little like these kinds of solution:
> > 
> > http://tce.technion.ac.il/wp-content/uploads/sites/8/2015/06/SC-7.2-M.-
> > Silberstein.pdf
> > 
> > With that, the application can directly pass the NiC packet pointer to the
> > decryption accelerator, and get the bare data in place. This is the feature
> > that
> > the VFIO container can provide.
> 
> above is not a good argument, at least in the context of your discussion.
> If each device has their own interface (similar to GPU) for process to bind 
> with, then having the process binding to multiple devices one-by-one then
> you still get same address space shared cross them...

If we consider this from the VFIO container perspective, with a container, you
can do DMA to the container applying it to all devices, even the device is added
after the DMA operation.  

So your argument remains true only when SVM is enabled and the whole process space
is devoted to the devices. 

Yes, the process can do the same all by itself. But if we agree with that, it
makes no sense to keep the container concept in VFIO;)

> 
> Thanks
> Kevin

-- 
			-Kenneth(Hisilicon)

================================================================================
æéäååéäåæåäååçäåäæïäéäåéçäéååäååçääæççãç
æääåäääääååäçïåæääéäåéæéååæéãååãææåïæéää
çäæãåææéæäæéäïèæçåçèæéäéçåääååéæéäï
This e-mail and its attachments contain confidential information from HUAWEI,
which is intended only for the person or entity whose address is listed above.
Any use of the 
information contained herein in any way (including, but not limited to, total or
partial disclosure, reproduction, or dissemination) by persons other than the
intended 
recipient(s) is prohibited. If you receive this e-mail in error, please notify
the sender by phone or email immediately and delete it!