question on "powerpc/pseries/dma: Allow SWIOTLB"

question on "powerpc/pseries/dma: Allow SWIOTLB"

[Christoph Hellwig]

Hey Alexey,

what is the use case for the above commit?  Shouldn't we handle all
addressing limits using the iommu?

Re: question on "powerpc/pseries/dma: Allow SWIOTLB"

[Alexey Kardashevskiy]

On 19/07/2019 17:10, Christoph Hellwig wrote:
> Hey Alexey,
> 
> what is the use case for the above commit?  Shouldn't we handle all
> addressing limits using the iommu?

Our secure VMs is the use case, when only a fraction of system memory is 
available for DMA.


-- 
Alexey

Re: question on "powerpc/pseries/dma: Allow SWIOTLB"

[Christoph Hellwig]

On Fri, Jul 19, 2019 at 05:52:37PM +1000, Alexey Kardashevskiy wrote:
> 
> 
> On 19/07/2019 17:10, Christoph Hellwig wrote:
> > Hey Alexey,
> > 
> > what is the use case for the above commit?  Shouldn't we handle all
> > addressing limits using the iommu?
> 
> Our secure VMs is the use case, when only a fraction of system memory is
> available for DMA.

But shouldn't we force usage of the direct ops in that case as the
IOMMU is not neededed at all?  Also isn't that support non-upstream so
far?

Re: question on "powerpc/pseries/dma: Allow SWIOTLB"

[Alexey Kardashevskiy]

On 19/07/2019 17:53, Christoph Hellwig wrote:
> On Fri, Jul 19, 2019 at 05:52:37PM +1000, Alexey Kardashevskiy wrote:
>>
>>
>> On 19/07/2019 17:10, Christoph Hellwig wrote:
>>> Hey Alexey,
>>>
>>> what is the use case for the above commit?  Shouldn't we handle all
>>> addressing limits using the iommu?
>>
>> Our secure VMs is the use case, when only a fraction of system memory is
>> available for DMA.
> 
> But shouldn't we force usage of the direct ops in that case as the
> IOMMU is not neededed at all?

We do, for mappings, but not unmappings and syncing.

> Also isn't that support non-upstream so far?

How is this relevant? I expect the existing "swiotlb=force" just work.


-- 
Alexey

Re: question on "powerpc/pseries/dma: Allow SWIOTLB"

[Christoph Hellwig]

On Fri, Jul 19, 2019 at 06:00:25PM +1000, Alexey Kardashevskiy wrote:
> > But shouldn't we force usage of the direct ops in that case as the
> > IOMMU is not neededed at all?
> 
> We do, for mappings, but not unmappings and syncing.

Well, I mean as in literally not setting a dma_ops so that the
dma_direct code is used without the indirection through the iommu ops.
This is not only more obvious, but also faster as you avoid the
indirect call (although that probably doesn't matter much if you
are bounce buffering anyway).

> > Also isn't that support non-upstream so far?
> 
> How is this relevant? I expect the existing "swiotlb=force" just work.

I though the whole secure VM support was still not upstream.

Re: question on "powerpc/pseries/dma: Allow SWIOTLB"

[Alexey Kardashevskiy]

On 19/07/2019 18:05, Christoph Hellwig wrote:
> On Fri, Jul 19, 2019 at 06:00:25PM +1000, Alexey Kardashevskiy wrote:
>>> But shouldn't we force usage of the direct ops in that case as the
>>> IOMMU is not neededed at all?
>>
>> We do, for mappings, but not unmappings and syncing.
> 
> Well, I mean as in literally not setting a dma_ops so that the
> dma_direct code is used without the indirection through the iommu ops.
> This is not only more obvious, but also faster as you avoid the
> indirect call (although that probably doesn't matter much if you
> are bounce buffering anyway).

I was not precise. We cannot avoid IOMMU in the guest for passed through 
devices.


>>> Also isn't that support non-upstream so far?
>>
>> How is this relevant? I expect the existing "swiotlb=force" just work.
> 
> I though the whole secure VM support was still not upstream.

It is getting there and I still do not see why "swiotlb=force" should 
not work if chosed in the cmdline.


-- 
Alexey

Re: question on "powerpc/pseries/dma: Allow SWIOTLB"

[Christoph Hellwig]

On Fri, Jul 19, 2019 at 06:23:59PM +1000, Alexey Kardashevskiy wrote:
> It is getting there and I still do not see why "swiotlb=force" should not
> work if chosed in the cmdline.

Ok, makes sense.  But that means we also have the issue in a few
other places..

Re: question on "powerpc/pseries/dma: Allow SWIOTLB"

[Alexey Kardashevskiy]

On 19/07/2019 22:25, Christoph Hellwig wrote:
> On Fri, Jul 19, 2019 at 06:23:59PM +1000, Alexey Kardashevskiy wrote:
>> It is getting there and I still do not see why "swiotlb=force" should not
>> work if chosed in the cmdline.
> 
> Ok, makes sense.  But that means we also have the issue in a few
> other places..

Hmm, where? I got broadcom ethernet working with this.


-- 
Alexey

Re: question on "powerpc/pseries/dma: Allow SWIOTLB"

[Christoph Hellwig]

On Sat, Jul 20, 2019 at 09:22:49PM +1000, Alexey Kardashevskiy wrote:
> 
> 
> On 19/07/2019 22:25, Christoph Hellwig wrote:
> > On Fri, Jul 19, 2019 at 06:23:59PM +1000, Alexey Kardashevskiy wrote:
> >> It is getting there and I still do not see why "swiotlb=force" should not
> >> work if chosed in the cmdline.
> > 
> > Ok, makes sense.  But that means we also have the issue in a few
> > other places..
> 
> Hmm, where? I got broadcom ethernet working with this.

In all the other IOMMU drivers that conditionally forward to the dma
direct ops.  So it shouldn't affect powerpc.