Re: [bug report] iommu/arm-smmu-v3: Event cannot be printed in some scenarios

[Mostafa Saleh <smostafa@google.com>]

Hi Kunkun,

On Wed, Jul 24, 2024 at 09:42:09AM +0800, Kunkun Jiang wrote:
> Hi all,
> 
> drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> 1797                 while (!queue_remove_raw(q, evt)) {
> 1798                         u8 id = FIELD_GET(EVTQ_0_ID, evt[0]);
> 1799
> 1800                         ret = arm_smmu_handle_evt(smmu, evt);
> 1801                         if (!ret || !__ratelimit(&rs))
> 1802                                 continue;
> 1803
> 1804                         dev_info(smmu->dev, "event 0x%02x received:\n",
> id);
> 1805                         for (i = 0; i < ARRAY_SIZE(evt); ++i)
> 1806                                 dev_info(smmu->dev, "\t0x%016llx\n",
> 1807                                          (unsigned long long)evt[i]);
> 1808
> 1809                         cond_resched();
> 1810                 }
> 
> The smmu-v3 driver cannot print event information when "ret" is 0.
> Unfortunately due to commit 3dfa64aecbaf
> ("iommu: Make iommu_report_device_fault() return void"), the default
> return value in arm_smmu_handle_evt() is 0. Maybe a trace should
> be added here?

In my opinion this change should be reverted as it’s very useful to
print event faults (for debugging crashes) as not always tracing is
available and enabled, and as we don’t want to print paging events
we should know if a device handled it or not.

Otherwise, as best effort from the driver, we can only skip printing
for IOPF enabled devices something like this? (or may be use
arm_smmu_master_sva_enabled() as I see only SVA uses it)

diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
index a31460f9f3d4..57f3a7d0e40f 100644
--- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
+++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
@@ -1777,7 +1777,11 @@ static int arm_smmu_handle_evt(struct arm_smmu_device *smmu, u64 *evt)
 		goto out_unlock;
 	}

-	iommu_report_device_fault(master->dev, &fault_evt);
+	if (master->iopf_enabled)
+		iommu_report_device_fault(master->dev, &fault_evt);
+	else
+		ret = -ENODEV;
+
 out_unlock:
 	mutex_unlock(&smmu->streams_mutex);
 	return ret;

But let’s see what others think.

Thanks,
Mostafa

> 
> Thanks,
> Kunkun Jiang
>