Re: [PATCH] iommu/dma: Handle MSI mappings separately

From: Robin Murphy <robin.murphy@arm.com>
To: Christoph Hellwig <hch@infradead.org>
Cc: maz@kernel.org, iommu@lists.linux-foundation.org,
	linux-arm-kernel@lists.infradead.org,
	Andre Przywara <andre.przywara@arm.com>
Subject: Re: [PATCH] iommu/dma: Handle MSI mappings separately
Date: Tue, 30 Jul 2019 11:43:25 +0100	[thread overview]
Message-ID: <90566814-ed36-2eb9-49e2-98a4a949c9f6@arm.com> (raw)
In-Reply-To: <20190730062843.GA1400@infradead.org>

On 30/07/2019 07:28, Christoph Hellwig wrote:
> On Mon, Jul 29, 2019 at 04:32:38PM +0100, Robin Murphy wrote:
>> MSI pages must always be mapped into a device's *current* domain, which
>> *might* be the default DMA domain, but might instead be a VFIO domain
>> with its own MSI cookie. This subtlety got accidentally lost in the
>> streamlining of __iommu_dma_map(), but rather than reintroduce more
>> complexity and/or special-casing, it turns out neater to just split this
>> path out entirely.
>>
>> Since iommu_dma_get_msi_page() already duplicates much of what
>> __iommu_dma_map() does, it can easily just make the allocation and
>> mapping calls directly as well. That way we can further streamline the
>> helper back to exclusively operating on DMA domains.
>>
>> Fixes: b61d271e59d7 ("iommu/dma: Move domain lookup into __iommu_dma_{map,unmap}")
>> Reported-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com>
>> Reported-by: Andre Przywara <andre.przywara@arm.com>
>> Signed-off-by: Robin Murphy <robin.murphy@arm.com>
> 
> Hmm.  I remember proposing this patch and you didn't like it because
> we could also have msis for a !IOMMU_DMA_IOVA_COOKIE cookie type.
> Or did we talk past each other?

Do you have a pointer? That sparks the vaguest of memories, but I can't 
seem to turn anything up in my inbox. If that was my objection, though, 
it sounds like your patch was probably trying to go a step or two 
further than this one.

> Note that if this change turns out to be valid we should also
> clean up the iommu_dma_free_iova() side.

We're not touching the iommu_dma_{alloc,free}_iova() path here; those 
are designed to cope with both types of cookie, and I think that's a 
reasonable abstraction to keep. This is just getting rid of the 
asymmetry - and now bug - caused by trying to keep the MSI page flow 
going through a special case in __iommu_dma_map() despite that having 
evolved into a more specific DMA domain fastpath (there's no 
corresponding unmap special case since MSI mappings just persist and get 
recycled until the domain is destroyed).

Robin.
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu