[PATCH v3 25/25] iommu: Allow enabling non-strict mode dynamically

[Robin Murphy]

On 2021-08-09 13:49, Will Deacon wrote:
> On Wed, Aug 04, 2021 at 06:15:53PM +0100, Robin Murphy wrote:
>> Allocating and enabling a flush queue is in fact something we can
>> reasonably do while a DMA domain is active, without having to rebuild it
>> from scratch. Thus we can allow a strict -> non-strict transition from
>> sysfs without requiring to unbind the device's driver, which is of
>> particular interest to users who want to make selective relaxations to
>> critical devices like the one serving their root filesystem.
>>
>> Disabling and draining a queue also seems technically possible to
>> achieve without rebuilding the whole domain, but would certainly be more
>> involved. Furthermore there's not such a clear use-case for tightening
>> up security *after* the device may already have done whatever it is that
>> you don't trust it not to do, so we only consider the relaxation case.
>>
>> CC: Sai Praneeth Prakhya <sai.praneeth.prakhya at intel.com>
>> Signed-off-by: Robin Murphy <robin.murphy at arm.com>
>>
>> ---
>>
>> v3: Actually think about concurrency, rework most of the fq data
>>      accesses to be (hopefully) safe and comment it all
>> ---
>>   drivers/iommu/dma-iommu.c | 25 ++++++++++++++++++-------
>>   drivers/iommu/iommu.c     | 16 ++++++++++++----
>>   drivers/iommu/iova.c      |  9 ++++++---
>>   3 files changed, 36 insertions(+), 14 deletions(-)
> 
> I failed to break this, so hopefully you've caught everything now.
> 
> Only thing I wasn't sure of is why we still need the smp_wmb() in
> init_iova_flush_queue(). Can we remove it now that we have one before
> assigning into the cookie?

Mostly because I failed to spot it, I think :)

Indeed now that we don't have any callers other than iommu_dma_init_fq() 
to worry about, I don't think that one matters any more. It would if 
were testing cookie->iovad->fq directly as our indicator instead of 
cookie->fq_domain, but then we'd still need the new barrier to ensure 
iommu_dma_flush_iotlb_all() properly observes the latter, so we may as 
well rely on that everywhere and let it fully replace the old one.

Thanks,
Robin.