[RFC PATCH 4/5] iommu/arm-smmu-v3: Add stream table directory structure to debugfs

[Nicolin Chen]

On Fri, Mar 13, 2026 at 06:43:50PM +0800, Qinxin Xia wrote:
>   * The capabilities file provides detailed information about:
>   * - Architecture version and translation stage support (Stage1/Stage2)
> @@ -324,5 +327,78 @@ static int smmu_debugfs_ste_show(struct seq_file *seq, void *v)
>  	smmu_debug_dump_ste(seq, dev);
>  	return 0;
>  }
> -
>  DEFINE_SHOW_ATTRIBUTE(smmu_debugfs_ste);

nit: let's not add that line in the first place

> +
> +/**
> + * smmu_debugfs_create_stream_table() - Create debugfs entries for stream table
> + * @dev: device to create entries for
> + * @smmu: SMMU device
> + *
> + * Return: 0 on success, negative error code on failure
> + */
> +int smmu_debugfs_create_stream_table(struct device *dev,
> +				     struct arm_smmu_device *smmu)
> +{
> +	struct dentry *stream_dir, *cd_dir, *dev_dir;
> +	struct dentry *ste_file, *all_cds_file;
> +	struct iommu_fwspec *fwspec;
> +	char name[64];
> +	int i, ret = 0;
> +
> +	if (!smmu->debugfs->stream_dir) {
> +		stream_dir = debugfs_create_dir("stream_table",
> +						smmu->debugfs->smmu_dir);
> +		if (!stream_dir)
> +			return -ENOMEM;
> +		smmu->debugfs->stream_dir = stream_dir;
> +	} else {
> +		stream_dir = smmu->debugfs->stream_dir;
> +	}
> +
> +	fwspec = dev_iommu_fwspec_get(dev);
> +	if (!fwspec)
> +		return -ENODEV;
> +
> +	for (i = 0; i < fwspec->num_ids; i++) {
> +		u32 sid = fwspec->ids[i];
> +
> +		if (dev_is_pci(dev)) {
> +			struct pci_dev *pdev = to_pci_dev(dev);
> +
> +			snprintf(name, sizeof(name), "%04x:%02x:%02x.%d:%u",
> +				 pci_domain_nr(pdev->bus), pdev->bus->number,
> +				 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
> +				 sid);

Doesn't dev_name(dev) print the BDF numbers as well?

> +		} else {
> +			snprintf(name, sizeof(name), "%s:%u", dev_name(dev),
> +				 sid);
> +		}
> +
> +		dev_dir = debugfs_create_dir(name, stream_dir);
> +		if (!dev_dir) {
> +			ret = -ENOMEM;
> +			goto cleanup;
> +		}

A device can have multiple streams. So, dev_dir should be the
parent directory?

> +
> +		/* Success for this stream ID, continue to next */
> +		continue;
> +
> +cleanup_dev:
> +		debugfs_remove_recursive(dev_dir);
> +cleanup:
> +		if (ret) {
> +			debugfs_remove_recursive(stream_dir);

One is the other's parent directory. So it should be good enough
with a single debugfs_remove_recursive()?

> 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 211a0c87507a..c57897e5f644 100644
> --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> @@ -3133,6 +3133,9 @@ static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev,
>  
>  	arm_smmu_attach_commit(&state);
>  	mutex_unlock(&arm_smmu_asid_lock);
> +#ifdef CONFIG_ARM_SMMU_V3_DEBUGFS
> +	smmu_debugfs_create_stream_table(dev, smmu);
> +#endif

Calling in attach_dev doesn't seem correct to me. And device/master
can be unplugged. So, we would need to unwind those dev/stream_dir.

Maybe probe_device/release_device are the places to go?

Nicolin