Re: [PATCH v2 1/2] iommu/amd: Add PerfOpt IOMMU performance optimization support

[Mario Limonciello <superm1@kernel.org>]

On 9/24/26 5:50 PM, Jason Gunthorpe wrote:
> On Mon, Sep 07, 2026 at 11:12:06PM -0500, Mario Limonciello wrote:
>> +EXPORT_SYMBOL_GPL(amd_iommu_enable_perfopt);
> 
> Last time we hardwired the GPU and IOMMU drivers together it was a
> huge mess to undo.
> 
> Why does the GPU driver have to request this? Why can't the iommu
> driver know that this is a special device that can use this magic fast
> path and then auto set it when identity is asked for?

The reason was to allow an "opt-out" path on the GPU driver.

> 
> How does the GPU driver even know it is magic special?

See patch 2/2.  GPU driver looks if it's an APU and it's in identity 
mode.  The special mode is only for the GPU in an APU.  It shouldn't be 
applied to anything else.
> 
>> +	/*
>> +	 * Restore ATS/PRI/PASID (and thus SVA) by re-homing the device onto its
>> +	 * identity domain with the flag cleared, so a later bind without PerfOpt
>> +	 * sees a normally-capable device. See the locking note in
>> +	 * amd_iommu_enable_perfopt().
>> +	 */
> 
> And this is sort of a wrong thing in the driver, it shouldn't have ATS
> turned on for identitiy mappings, and it certainly shouldn't have PRI
> turned on until a PRI capable domain is attached.
> 
> Jason
> 

amd_iommu_disable_perfopt() only re-applies the policy that was 
previously set before PerfOpt.

If you would like I'll send a follow up patch to clarify this comment.