Re: [PATCH 1/1] iommu/vt-d: Avoid WARNING in sva unbind path

[Baolu Lu <baolu.lu@linux.intel.com>]

On 5/20/26 21:12, Yi Liu wrote:
> On 5/19/26 13:29, Lu Baolu wrote:
>> The Intel IOMMU driver allows SVA on devices even if they do not support
>> PCI/PRI. Commit 39c20c4e83b9 ("iommu/vt-d: Only handle IOPF for SVA when
>> PRI is supported") modified the SVA bind path to allow this configuration
>> by skipping IOPF enablement when PRI is missing. However, it failed to
>> update the unbind path.
>>
>> This creates an imbalance: the unbind path attempts to disable IOPF for
>> a device that never had it enabled, triggering a WARNING in
>> intel_iommu_disable_iopf():
>>
>>   WARNING: drivers/iommu/intel/iommu.c:3475 at 
>> intel_iommu_disable_iopf+0x4f/0x90d
>>   Call Trace:
>>    <TASK>
>>    blocking_domain_set_dev_pasid+0x50/0x70
>>    iommu_detach_device_pasid+0x89/0xc0
>>    iommu_sva_unbind_device+0x73/0x150
>>    xe_vm_close_and_put+0x4d2/0x1200 [xe]
>>
>> Fix this by bypassing IOPF operations for SVA domains on non-PRI hardware
>> in both the bind and unbind paths.
>>
>> Fixes: 39c20c4e83b9 ("iommu/vt-d: Only handle IOPF for SVA when PRI is 
>> supported")
>> Cc: stable@vger.kernel.org
>> Reported-by: Nareshkumar Gollakoti <naresh.kumar.g@intel.com>
>> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
>> ---
>>   drivers/iommu/intel/iommu.h | 11 +++++++++++
>>   drivers/iommu/intel/svm.c   | 12 ++++--------
>>   2 files changed, 15 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/iommu/intel/iommu.h b/drivers/iommu/intel/iommu.h
>> index ef145560aa98..775f1c4ae346 100644
>> --- a/drivers/iommu/intel/iommu.h
>> +++ b/drivers/iommu/intel/iommu.h
>> @@ -1254,18 +1254,29 @@ void intel_iommu_disable_iopf(struct device 
>> *dev);
>>   static inline int iopf_for_domain_set(struct iommu_domain *domain,
>>                         struct device *dev)
>>   {
>> +    struct device_domain_info *info = dev_iommu_priv_get(dev);
>> +
>>       if (!domain || !domain->iopf_handler)
>>           return 0;
>> +    /* SVA with non-IOMMU/PRI IOPF handling is allowed. */
>> +    if (domain->type == IOMMU_DOMAIN_SVA && !info->pri_supported)
>> +        return 0;
>> +
> 
> Looked into the history a bit, and this story begins with commit
> a86fb7717320 ("iommu/vt-d: Allow SVA with device-specific IOPF"). This
> commit enabled devices that support their own IOPF mechanism to use SVA
> even when the platform IOMMU doesn't support IOPF.
> 
> However, SVA isn't the only fault-capable domain type. Other fault-capable
> domain types (e.g., paging domains) should also be able to leverage
> device-specific IOPF capabilities.
> 
> My question is: can we drop the domain type check to support other types
> of fault-capable domains that rely on device-specific IOPF?

The Intel IOMMU driver treats the SVA case as a special case. It allows
an SVA domain with a registered iopf_handler to attach to a device for
SVA functionality. Broadly speaking, it would be better to handle this
in the IOMMU core — for example, by preventing the attachment of a
fault-capable domain to a device that lacks IOMMU-backed page fault
capabilities.

Thanks,
baolu