[PATH v3 0/3] iommu/vt-d: debugfs: Enhancements to IOMMU debugfs

[PATH v3 0/3] iommu/vt-d: debugfs: Enhancements to IOMMU debugfs

[Jingqi Liu]

The original debugfs only dumps all IOMMU page tables without pasid
supported. It traverses all devices on the pci bus, then dumps all
page tables based on device domains. This traversal is from software
perspective.

This series dumps page tables whose mappings are created and destroyed
by the iommu_map/unmap() interfaces, by traversing root tables,
context tables, pasid directories and pasid tables from hardware
perspective. It supports dumping a specified page table in legacy mode
or scalable mode with or without a specified pasid.

It adds a debugfs directory per pair of {device, pasid} when attaching
device with or without pasid, i.e.
/sys/kernel/debug/iommu/intel/<device source id>/<pasid>.
And create a debugfs file in the directory for users to dump the page
table corresponding to {device, pasid}. e.g.
/sys/kernel/debug/iommu/intel/0000:00:02.0/0/domain_translation_struct.
Each device has a PASID#0, i.e. RID_PASID. Remove the corresponding
debugfs directory and file when detaching or releasing a device.

For legacy mode, according to bus number and DEVFN, traverse the root
table and context table to get the pointer of page table in the
context table entry, then dump the specified page table.

For scalable mode, according to bus number, DEVFN and pasid, traverse
the root table, context table, pasid directory and pasid table to get
the pointer of page table in the pasid table entry, then dump the
specified page table.

Examples are as follows:
1) Dump the page table of device "0000:00:01.0" that only supports
    legacy mode.
    $ sudo cat
    /sys/kernel/debug/iommu/intel/0000:00:01.0/0/domain_translation_struct

2) Dump the page table of device "0000:00:02.0" with PASID "1" that
   supports scalable mode.
   $ sudo cat
   /sys/kernel/debug/iommu/intel/0000:00:0a.0/1/domain_translation_struct

Change log:

v3:
 - Add domain as a parameter for creating debugfs pasid directory
   per Baolu's review.
 - dput() the dentry after debugfs_lookup() per Baolu's review.
 - debugfs device and pasid directory are managed separately per
   Baolu's review.

v2: https://lore.kernel.org/linux-iommu/20230922151636.77139-1-Jingqi.liu@intel.com
 - Add a debugfs directory per {dev, pasid} as suggested by Kevin.
 - Create the debugfs directory when attaching device as suggested by Baolu.
 - Only dump the page tables whose mappings are created and destroyed
   by the iommu_map/unmap() interfaces per Baolu's review.
 - Rename the helpers for creating/removing debugfs directory/file and
   merge patch 2,3,4,5 to one patch per Baolu's review.

v1: https://lore.kernel.org/linux-iommu/20230625150442.42197-1-Jingqi.liu@intel.com

Jingqi Liu (3):
  iommu/vt-d: debugfs: Dump entry pointing to huge page
  iommu/vt-d: debugfs: Create/remove debugfs file per {device, pasid}
  iommu/vt-d: debugfs: Support dumping a specified page table

 drivers/iommu/intel/debugfs.c | 307 ++++++++++++++++++++++++++++------
 drivers/iommu/intel/iommu.c   |  29 ++++
 drivers/iommu/intel/iommu.h   |  12 ++
 3 files changed, 298 insertions(+), 50 deletions(-)

-- 
2.21.3

[PATH v3 1/3] iommu/vt-d: debugfs: Dump entry pointing to huge page

[Jingqi Liu]

For the page table entry pointing to a huge page, the data below the
level of the huge page is meaningless and does not need to be dumped.

Signed-off-by: Jingqi Liu <Jingqi.liu@intel.com>
---
 drivers/iommu/intel/debugfs.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/drivers/iommu/intel/debugfs.c b/drivers/iommu/intel/debugfs.c
index 0255c4a23269..e2a3c37943a0 100644
--- a/drivers/iommu/intel/debugfs.c
+++ b/drivers/iommu/intel/debugfs.c
@@ -311,9 +311,14 @@ static inline unsigned long level_to_directory_size(int level)
 static inline void
 dump_page_info(struct seq_file *m, unsigned long iova, u64 *path)
 {
-	seq_printf(m, "0x%013lx |\t0x%016llx\t0x%016llx\t0x%016llx\t0x%016llx\t0x%016llx\n",
-		   iova >> VTD_PAGE_SHIFT, path[5], path[4],
-		   path[3], path[2], path[1]);
+	seq_printf(m, "0x%013lx |\t0x%016llx\t0x%016llx\t0x%016llx",
+		   iova >> VTD_PAGE_SHIFT, path[5], path[4], path[3]);
+	if (path[2]) {
+		seq_printf(m, "\t0x%016llx", path[2]);
+		if (path[1])
+			seq_printf(m, "\t0x%016llx", path[1]);
+	}
+	seq_putc(m, '\n');
 }
 
 static void pgtable_walk_level(struct seq_file *m, struct dma_pte *pde,
-- 
2.21.3

[PATH v3 2/3] iommu/vt-d: debugfs: Create/remove debugfs file per {device, pasid}

[Jingqi Liu]

Add a debugfs directory per pair of {device, pasid} if the mappings of
its page table are created and destroyed by the iommu_map/unmap()
interfaces. i.e. /sys/kernel/debug/iommu/intel/<device source id>/<pasid>.
Create a debugfs file in the directory for users to dump the page
table corresponding to {device, pasid}. e.g.
/sys/kernel/debug/iommu/intel/0000:00:02.0/0/domain_translation_struct.

When attaching device without pasid, create a debugfs file with
PASID#0, i.e. RID_PASID. When attaching a domain to a pasid of device,
create a debugfs file with the specified pasid.

When detaching without pasid, remove the directory and file for
PASID#0. When detaching with pasid, remove the debugfs directory and
file of the specified pasid. Remove the entire debugfs directory of
the specified device for releasing device.
e.g. /sys/kernel/debug/iommu/intel/0000:00:01.0

Signed-off-by: Jingqi Liu <Jingqi.liu@intel.com>
---
 drivers/iommu/intel/debugfs.c | 133 ++++++++++++++++++++++++++++++++--
 drivers/iommu/intel/iommu.c   |  29 ++++++++
 drivers/iommu/intel/iommu.h   |  12 +++
 3 files changed, 169 insertions(+), 5 deletions(-)

diff --git a/drivers/iommu/intel/debugfs.c b/drivers/iommu/intel/debugfs.c
index e2a3c37943a0..50dea53ba87c 100644
--- a/drivers/iommu/intel/debugfs.c
+++ b/drivers/iommu/intel/debugfs.c
@@ -32,6 +32,11 @@ struct iommu_regset {
 	const char *regs;
 };
 
+struct show_domain_info {
+	struct device *dev;
+	ioasid_t pasid;
+};
+
 #define DEBUG_BUFFER_SIZE	1024
 static char debug_buf[DEBUG_BUFFER_SIZE];
 
@@ -111,6 +116,8 @@ static const struct iommu_regset iommu_regs_64[] = {
 	IOMMU_REGSET_ENTRY(VCRSP),
 };
 
+static struct dentry *intel_iommu_debug;
+
 static int iommu_regset_show(struct seq_file *m, void *unused)
 {
 	struct dmar_drhd_unit *drhd;
@@ -673,16 +680,12 @@ static const struct file_operations dmar_perf_latency_fops = {
 
 void __init intel_iommu_debugfs_init(void)
 {
-	struct dentry *intel_iommu_debug = debugfs_create_dir("intel",
-						iommu_debugfs_dir);
+	intel_iommu_debug = debugfs_create_dir("intel", iommu_debugfs_dir);
 
 	debugfs_create_file("iommu_regset", 0444, intel_iommu_debug, NULL,
 			    &iommu_regset_fops);
 	debugfs_create_file("dmar_translation_struct", 0444, intel_iommu_debug,
 			    NULL, &dmar_translation_struct_fops);
-	debugfs_create_file("domain_translation_struct", 0444,
-			    intel_iommu_debug, NULL,
-			    &domain_translation_struct_fops);
 	debugfs_create_file("invalidation_queue", 0444, intel_iommu_debug,
 			    NULL, &invalidation_queue_fops);
 #ifdef CONFIG_IRQ_REMAP
@@ -692,3 +695,123 @@ void __init intel_iommu_debugfs_init(void)
 	debugfs_create_file("dmar_perf_latency", 0644, intel_iommu_debug,
 			    NULL, &dmar_perf_latency_fops);
 }
+
+/* Create a debugfs directory for each device. */
+void intel_iommu_debugfs_create_dev(struct device *dev)
+{
+	struct dentry *dev_dir;
+
+	dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
+	if (!dev_dir) {
+		dev_dir = debugfs_create_dir(dev_name(dev), intel_iommu_debug);
+		if (IS_ERR(dev_dir))
+			pr_info("%s: Failed to create debugfs directory.\n",
+				dev_name(dev));
+	} else
+		dput(dev_dir);
+}
+
+void intel_iommu_debugfs_remove_dev(struct device *dev)
+{
+	struct dentry *dev_dir, *sub_dir, *dentry;
+	struct list_head *plist;
+
+	dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
+	if (!dev_dir)
+		return;
+
+	list_for_each(plist, &(dev_dir->d_subdirs)) {
+		sub_dir = list_entry(plist, struct dentry, d_child);
+		if(sub_dir) {
+			dentry = debugfs_lookup("domain_translation_struct",
+						sub_dir);
+			if (!dentry)
+				continue;
+
+			if (dentry->d_inode->i_private)
+				kfree(dentry->d_inode->i_private);
+
+			dput(dentry);
+		}
+	}
+
+	debugfs_remove_recursive(dev_dir);
+	dput(dev_dir);
+}
+
+/*
+ * Create a debugfs directory per pair of {device, pasid},
+ * then create the corresponding debugfs file in this directory
+ * for user to dump its page table. e.g.
+ * /sys/kernel/debug/iommu/intel/0000:00:01.0/0/domain_translation_struct
+ */
+void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
+					  struct device *dev, u32 pasid)
+{
+	struct dentry *dev_dir, *pasid_dir;
+	struct show_domain_info *sinfo;
+	char dir_name[10];
+
+	/*
+	 * The debugfs only dumps the page tables whose mappings are created
+	 * and destroyed by the iommu_map/unmap() interfaces. Check the
+	 * mapping type of the domain before creating debugfs directory.
+	 */
+	if (!domain || !(domain->type & __IOMMU_DOMAIN_PAGING))
+		return;
+
+	dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
+	if (!dev_dir)
+		return;
+
+	sprintf(dir_name, "%x", pasid);
+	pasid_dir = debugfs_create_dir(dir_name, dev_dir);
+	if (IS_ERR(pasid_dir))
+		goto dput_out;
+
+	sinfo = kzalloc(sizeof(*sinfo), GFP_KERNEL);
+	if (!sinfo)
+		goto dput_out;
+
+	sinfo->dev = dev;
+	sinfo->pasid = pasid;
+	debugfs_create_file("domain_translation_struct", 0444,
+			    pasid_dir, sinfo,
+			    &domain_translation_struct_fops);
+dput_out:
+	dput(dev_dir);
+}
+
+/*
+ * Remove the debugfs directory and file corresponding to each pair of
+ * {device, pasid}.
+ */
+void intel_iommu_debugfs_remove_dev_pasid(struct device *dev, u32 pasid)
+{
+	struct dentry *dev_dir, *pasid_dir, *dentry;
+	char dir_name[10];
+
+	dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
+	if (!dev_dir)
+		return;
+
+	sprintf(dir_name, "%x", pasid);
+	pasid_dir = debugfs_lookup(dir_name, dev_dir);
+	if (!pasid_dir)
+		goto dput_dev;
+
+	dentry = debugfs_lookup("domain_translation_struct", pasid_dir);
+	if (!dentry)
+		goto dput_pasid;
+
+	if (dentry->d_inode->i_private)
+		kfree(dentry->d_inode->i_private);
+
+	debugfs_remove_recursive(pasid_dir);
+
+	dput(dentry);
+dput_pasid:
+	dput(pasid_dir);
+dput_dev:
+	dput(dev_dir);
+}
diff --git a/drivers/iommu/intel/iommu.c b/drivers/iommu/intel/iommu.c
index dd8ff358867d..af9c989035a2 100644
--- a/drivers/iommu/intel/iommu.c
+++ b/drivers/iommu/intel/iommu.c
@@ -2488,6 +2488,13 @@ static int dmar_domain_attach_device(struct dmar_domain *domain,
 
 	iommu_enable_pci_caps(info);
 
+	/*
+	 * Create a debugfs directory specified by RID_PASID
+	 * in the debugfs device directory.
+	 */
+	intel_iommu_debugfs_create_dev_pasid(&info->domain->domain,
+			                     dev, IOMMU_NO_PASID);
+
 	return 0;
 }
 
@@ -3997,6 +4004,12 @@ static void device_block_translation(struct device *dev)
 
 	domain_detach_iommu(info->domain, iommu);
 	info->domain = NULL;
+
+	/*
+	 * Remove the debugfs directory specified by RID_PASID
+	 * in the device debugfs directory.
+	 */
+	intel_iommu_debugfs_remove_dev_pasid(dev, IOMMU_NO_PASID);
 }
 
 static int md_domain_init(struct dmar_domain *domain, int guest_width)
@@ -4424,6 +4437,9 @@ static struct iommu_device *intel_iommu_probe_device(struct device *dev)
 		}
 	}
 
+	/* Create a debugfs directory per device. */
+	intel_iommu_debugfs_create_dev(dev);
+
 	return &iommu->iommu;
 }
 
@@ -4436,6 +4452,7 @@ static void intel_iommu_release_device(struct device *dev)
 	dev_iommu_priv_set(dev, NULL);
 	kfree(info);
 	set_dma_ops(dev, NULL);
+	intel_iommu_debugfs_remove_dev(dev);
 }
 
 static void intel_iommu_probe_finalize(struct device *dev)
@@ -4729,6 +4746,12 @@ static void intel_iommu_remove_dev_pasid(struct device *dev, ioasid_t pasid)
 out_tear_down:
 	intel_pasid_tear_down_entry(iommu, dev, pasid, false);
 	intel_drain_pasid_prq(dev, pasid);
+
+	/*
+	 * Remove the debugfs directory of the specified pasid
+	 * in the device debugfs directory.
+	 */
+	intel_iommu_debugfs_remove_dev_pasid(dev, pasid);
 }
 
 static int intel_iommu_set_dev_pasid(struct iommu_domain *domain,
@@ -4777,6 +4800,12 @@ static int intel_iommu_set_dev_pasid(struct iommu_domain *domain,
 	list_add(&dev_pasid->link_domain, &dmar_domain->dev_pasids);
 	spin_unlock_irqrestore(&dmar_domain->lock, flags);
 
+	/*
+	 * Create a debugfs directory with specified pasid
+	 * in the debugfs device directory.
+	 */
+	intel_iommu_debugfs_create_dev_pasid(domain, dev, pasid);
+
 	return 0;
 out_detach_iommu:
 	domain_detach_iommu(dmar_domain, iommu);
diff --git a/drivers/iommu/intel/iommu.h b/drivers/iommu/intel/iommu.h
index c18fb699c87a..f1c27e928886 100644
--- a/drivers/iommu/intel/iommu.h
+++ b/drivers/iommu/intel/iommu.h
@@ -883,8 +883,20 @@ static inline void intel_svm_remove_dev_pasid(struct device *dev, ioasid_t pasid
 
 #ifdef CONFIG_INTEL_IOMMU_DEBUGFS
 void intel_iommu_debugfs_init(void);
+void intel_iommu_debugfs_create_dev(struct device *dev);
+void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
+					  struct device *dev, u32 pasid);
+void intel_iommu_debugfs_remove_dev(struct device *dev);
+void intel_iommu_debugfs_remove_dev_pasid(struct device *dev, u32 pasid);
 #else
 static inline void intel_iommu_debugfs_init(void) {}
+static inline void intel_iommu_debugfs_create_dev(struct device *dev) {}
+static inline
+void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
+					  struct device *dev, u32 pasid) {}
+static inline void intel_iommu_debugfs_remove_dev(struct device *dev) {}
+static inline
+void intel_iommu_debugfs_remove_dev_pasid(struct device *dev, u32 pasid) {}
 #endif /* CONFIG_INTEL_IOMMU_DEBUGFS */
 
 extern const struct attribute_group *intel_iommu_groups[];
-- 
2.21.3

Re: [PATH v3 2/3] iommu/vt-d: debugfs: Create/remove debugfs file per {device, pasid}

[Baolu Lu]

On 9/27/23 11:15 PM, Jingqi Liu wrote:
> Add a debugfs directory per pair of {device, pasid} if the mappings of
> its page table are created and destroyed by the iommu_map/unmap()
> interfaces. i.e. /sys/kernel/debug/iommu/intel/<device source id>/<pasid>.
> Create a debugfs file in the directory for users to dump the page
> table corresponding to {device, pasid}. e.g.
> /sys/kernel/debug/iommu/intel/0000:00:02.0/0/domain_translation_struct.
> 
> When attaching device without pasid, create a debugfs file with
> PASID#0, i.e. RID_PASID. When attaching a domain to a pasid of device,
> create a debugfs file with the specified pasid.
> 
> When detaching without pasid, remove the directory and file for
> PASID#0. When detaching with pasid, remove the debugfs directory and
> file of the specified pasid. Remove the entire debugfs directory of
> the specified device for releasing device.
> e.g. /sys/kernel/debug/iommu/intel/0000:00:01.0
> 
> Signed-off-by: Jingqi Liu <Jingqi.liu@intel.com>
> ---
>   drivers/iommu/intel/debugfs.c | 133 ++++++++++++++++++++++++++++++++--
>   drivers/iommu/intel/iommu.c   |  29 ++++++++
>   drivers/iommu/intel/iommu.h   |  12 +++
>   3 files changed, 169 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/iommu/intel/debugfs.c b/drivers/iommu/intel/debugfs.c
> index e2a3c37943a0..50dea53ba87c 100644
> --- a/drivers/iommu/intel/debugfs.c
> +++ b/drivers/iommu/intel/debugfs.c
> @@ -32,6 +32,11 @@ struct iommu_regset {
>   	const char *regs;
>   };
>   
> +struct show_domain_info {
> +	struct device *dev;
> +	ioasid_t pasid;
> +};
> +
>   #define DEBUG_BUFFER_SIZE	1024
>   static char debug_buf[DEBUG_BUFFER_SIZE];
>   
> @@ -111,6 +116,8 @@ static const struct iommu_regset iommu_regs_64[] = {
>   	IOMMU_REGSET_ENTRY(VCRSP),
>   };
>   
> +static struct dentry *intel_iommu_debug;
> +
>   static int iommu_regset_show(struct seq_file *m, void *unused)
>   {
>   	struct dmar_drhd_unit *drhd;
> @@ -673,16 +680,12 @@ static const struct file_operations dmar_perf_latency_fops = {
>   
>   void __init intel_iommu_debugfs_init(void)
>   {
> -	struct dentry *intel_iommu_debug = debugfs_create_dir("intel",
> -						iommu_debugfs_dir);
> +	intel_iommu_debug = debugfs_create_dir("intel", iommu_debugfs_dir);
>   
>   	debugfs_create_file("iommu_regset", 0444, intel_iommu_debug, NULL,
>   			    &iommu_regset_fops);
>   	debugfs_create_file("dmar_translation_struct", 0444, intel_iommu_debug,
>   			    NULL, &dmar_translation_struct_fops);
> -	debugfs_create_file("domain_translation_struct", 0444,
> -			    intel_iommu_debug, NULL,
> -			    &domain_translation_struct_fops);
>   	debugfs_create_file("invalidation_queue", 0444, intel_iommu_debug,
>   			    NULL, &invalidation_queue_fops);
>   #ifdef CONFIG_IRQ_REMAP
> @@ -692,3 +695,123 @@ void __init intel_iommu_debugfs_init(void)
>   	debugfs_create_file("dmar_perf_latency", 0644, intel_iommu_debug,
>   			    NULL, &dmar_perf_latency_fops);
>   }
> +
> +/* Create a debugfs directory for each device. */
> +void intel_iommu_debugfs_create_dev(struct device *dev)
> +{
> +	struct dentry *dev_dir;
> +
> +	dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
> +	if (!dev_dir) {
> +		dev_dir = debugfs_create_dir(dev_name(dev), intel_iommu_debug);
> +		if (IS_ERR(dev_dir))
> +			pr_info("%s: Failed to create debugfs directory.\n",
> +				dev_name(dev));
> +	} else
> +		dput(dev_dir);
> +}

Above could simply be like this:

void intel_iommu_debugfs_create_dev(struct device *dev)
{
	struct device_domain_info *info = dev_iommu_priv_get(dev);

	info->debugfs_entry = debugfs_create_dir(dev_name(dev),
			intel_iommu_debug);
}

Isn't it?

> +
> +void intel_iommu_debugfs_remove_dev(struct device *dev)
> +{
> +	struct dentry *dev_dir, *sub_dir, *dentry;
> +	struct list_head *plist;
> +
> +	dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
> +	if (!dev_dir)
> +		return;
> +
> +	list_for_each(plist, &(dev_dir->d_subdirs)) {
> +		sub_dir = list_entry(plist, struct dentry, d_child);
> +		if(sub_dir) {
> +			dentry = debugfs_lookup("domain_translation_struct",
> +						sub_dir);
> +			if (!dentry)
> +				continue;
> +
> +			if (dentry->d_inode->i_private)
> +				kfree(dentry->d_inode->i_private);
> +
> +			dput(dentry);
> +		}
> +	}
> +
> +	debugfs_remove_recursive(dev_dir);
> +	dput(dev_dir);
> +}

And this could simply be like this:

void intel_iommu_debugfs_remove_dev(struct device *dev)
{
	struct device_domain_info *info = dev_iommu_priv_get(dev);

	debugfs_remove(info->debugfs_entry);
}

> +
> +/*
> + * Create a debugfs directory per pair of {device, pasid},
> + * then create the corresponding debugfs file in this directory
> + * for user to dump its page table. e.g.
> + * /sys/kernel/debug/iommu/intel/0000:00:01.0/0/domain_translation_struct
> + */
> +void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
> +					  struct device *dev, u32 pasid)
> +{
> +	struct dentry *dev_dir, *pasid_dir;
> +	struct show_domain_info *sinfo;
> +	char dir_name[10];
> +
> +	/*
> +	 * The debugfs only dumps the page tables whose mappings are created
> +	 * and destroyed by the iommu_map/unmap() interfaces. Check the
> +	 * mapping type of the domain before creating debugfs directory.
> +	 */
> +	if (!domain || !(domain->type & __IOMMU_DOMAIN_PAGING))
> +		return;
> +
> +	dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
> +	if (!dev_dir)
> +		return;
> +
> +	sprintf(dir_name, "%x", pasid);
> +	pasid_dir = debugfs_create_dir(dir_name, dev_dir);
> +	if (IS_ERR(pasid_dir))
> +		goto dput_out;
> +
> +	sinfo = kzalloc(sizeof(*sinfo), GFP_KERNEL);
> +	if (!sinfo)
> +		goto dput_out;
> +
> +	sinfo->dev = dev;
> +	sinfo->pasid = pasid;
> +	debugfs_create_file("domain_translation_struct", 0444,
> +			    pasid_dir, sinfo,
> +			    &domain_translation_struct_fops);
> +dput_out:
> +	dput(dev_dir);
> +}

And here,

void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
					  struct device *dev, u32 pasid)
{
	sprintf(dir_name, "%x", pasid);
	dev_pasid->debugfs_entry = debugfs_create_dir(dir_name,
			info->debugfs_entry);

	debugfs_create_file("domain_translation_struct", 0444,
			    dev_pasid->debugfs_entry, dev_pasid,
			    &domain_translation_struct_fops);
}

> +
> +/*
> + * Remove the debugfs directory and file corresponding to each pair of
> + * {device, pasid}.
> + */
> +void intel_iommu_debugfs_remove_dev_pasid(struct device *dev, u32 pasid)
> +{
> +	struct dentry *dev_dir, *pasid_dir, *dentry;
> +	char dir_name[10];
> +
> +	dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
> +	if (!dev_dir)
> +		return;
> +
> +	sprintf(dir_name, "%x", pasid);
> +	pasid_dir = debugfs_lookup(dir_name, dev_dir);
> +	if (!pasid_dir)
> +		goto dput_dev;
> +
> +	dentry = debugfs_lookup("domain_translation_struct", pasid_dir);
> +	if (!dentry)
> +		goto dput_pasid;
> +
> +	if (dentry->d_inode->i_private)
> +		kfree(dentry->d_inode->i_private);
> +
> +	debugfs_remove_recursive(pasid_dir);
> +
> +	dput(dentry);
> +dput_pasid:
> +	dput(pasid_dir);
> +dput_dev:
> +	dput(dev_dir);
> +}

The same thing here:

	debugfs_remove(dev_pasid->debugfs_entry);

> diff --git a/drivers/iommu/intel/iommu.c b/drivers/iommu/intel/iommu.c
> index dd8ff358867d..af9c989035a2 100644
> --- a/drivers/iommu/intel/iommu.c
> +++ b/drivers/iommu/intel/iommu.c
> @@ -2488,6 +2488,13 @@ static int dmar_domain_attach_device(struct dmar_domain *domain,
>   
>   	iommu_enable_pci_caps(info);
>   
> +	/*
> +	 * Create a debugfs directory specified by RID_PASID
> +	 * in the debugfs device directory.
> +	 */
> +	intel_iommu_debugfs_create_dev_pasid(&info->domain->domain,
> +			                     dev, IOMMU_NO_PASID);

The function name is self-explained. So no need to add comments. Ditto
to all other places.

> +
>   	return 0;
>   }
>   
> @@ -3997,6 +4004,12 @@ static void device_block_translation(struct device *dev)
>   
>   	domain_detach_iommu(info->domain, iommu);
>   	info->domain = NULL;
> +
> +	/*
> +	 * Remove the debugfs directory specified by RID_PASID
> +	 * in the device debugfs directory.
> +	 */
> +	intel_iommu_debugfs_remove_dev_pasid(dev, IOMMU_NO_PASID);
>   }
>   
>   static int md_domain_init(struct dmar_domain *domain, int guest_width)
> @@ -4424,6 +4437,9 @@ static struct iommu_device *intel_iommu_probe_device(struct device *dev)
>   		}
>   	}
>   
> +	/* Create a debugfs directory per device. */
> +	intel_iommu_debugfs_create_dev(dev);
> +
>   	return &iommu->iommu;
>   }
>   
> @@ -4436,6 +4452,7 @@ static void intel_iommu_release_device(struct device *dev)
>   	dev_iommu_priv_set(dev, NULL);
>   	kfree(info);
>   	set_dma_ops(dev, NULL);
> +	intel_iommu_debugfs_remove_dev(dev);
>   }
>   
>   static void intel_iommu_probe_finalize(struct device *dev)
> @@ -4729,6 +4746,12 @@ static void intel_iommu_remove_dev_pasid(struct device *dev, ioasid_t pasid)
>   out_tear_down:
>   	intel_pasid_tear_down_entry(iommu, dev, pasid, false);
>   	intel_drain_pasid_prq(dev, pasid);
> +
> +	/*
> +	 * Remove the debugfs directory of the specified pasid
> +	 * in the device debugfs directory.
> +	 */
> +	intel_iommu_debugfs_remove_dev_pasid(dev, pasid);
>   }
>   
>   static int intel_iommu_set_dev_pasid(struct iommu_domain *domain,
> @@ -4777,6 +4800,12 @@ static int intel_iommu_set_dev_pasid(struct iommu_domain *domain,
>   	list_add(&dev_pasid->link_domain, &dmar_domain->dev_pasids);
>   	spin_unlock_irqrestore(&dmar_domain->lock, flags);
>   
> +	/*
> +	 * Create a debugfs directory with specified pasid
> +	 * in the debugfs device directory.
> +	 */
> +	intel_iommu_debugfs_create_dev_pasid(domain, dev, pasid);
> +
>   	return 0;
>   out_detach_iommu:
>   	domain_detach_iommu(dmar_domain, iommu);
> diff --git a/drivers/iommu/intel/iommu.h b/drivers/iommu/intel/iommu.h
> index c18fb699c87a..f1c27e928886 100644
> --- a/drivers/iommu/intel/iommu.h
> +++ b/drivers/iommu/intel/iommu.h
> @@ -883,8 +883,20 @@ static inline void intel_svm_remove_dev_pasid(struct device *dev, ioasid_t pasid
>   
>   #ifdef CONFIG_INTEL_IOMMU_DEBUGFS
>   void intel_iommu_debugfs_init(void);
> +void intel_iommu_debugfs_create_dev(struct device *dev);
> +void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
> +					  struct device *dev, u32 pasid);
> +void intel_iommu_debugfs_remove_dev(struct device *dev);
> +void intel_iommu_debugfs_remove_dev_pasid(struct device *dev, u32 pasid);
>   #else
>   static inline void intel_iommu_debugfs_init(void) {}
> +static inline void intel_iommu_debugfs_create_dev(struct device *dev) {}
> +static inline
> +void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
> +					  struct device *dev, u32 pasid) {}
> +static inline void intel_iommu_debugfs_remove_dev(struct device *dev) {}
> +static inline
> +void intel_iommu_debugfs_remove_dev_pasid(struct device *dev, u32 pasid) {}
>   #endif /* CONFIG_INTEL_IOMMU_DEBUGFS */
>   
>   extern const struct attribute_group *intel_iommu_groups[];

Best regards,
baolu

Re: [PATH v3 2/3] iommu/vt-d: debugfs: Create/remove debugfs file per {device, pasid}

[Liu, Jingqi]

On 9/28/2023 9:58 AM, Baolu Lu wrote:
> On 9/27/23 11:15 PM, Jingqi Liu wrote:
......
>> +
>> +/* Create a debugfs directory for each device. */
>> +void intel_iommu_debugfs_create_dev(struct device *dev)
>> +{
>> +    struct dentry *dev_dir;
>> +
>> +    dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
>> +    if (!dev_dir) {
>> +        dev_dir = debugfs_create_dir(dev_name(dev), intel_iommu_debug);
>> +        if (IS_ERR(dev_dir))
>> +            pr_info("%s: Failed to create debugfs directory.\n",
>> +                dev_name(dev));
>> +    } else
>> +        dput(dev_dir);
>> +}
>
> Above could simply be like this:
>
> void intel_iommu_debugfs_create_dev(struct device *dev)
> {
>     struct device_domain_info *info = dev_iommu_priv_get(dev);
>
>     info->debugfs_entry = debugfs_create_dir(dev_name(dev),
>             intel_iommu_debug);
> }
>
> Isn't it?
Thanks. Good point.
If add an "info->debugfs_dentry" to the device "info" to save the dentry 
of device
debugfs directory, there's no need to lookup the dentry by debugfs_lookup().
Just simply get it from the device "info".

>
>> +
>> +void intel_iommu_debugfs_remove_dev(struct device *dev)
>> +{
>> +    struct dentry *dev_dir, *sub_dir, *dentry;
>> +    struct list_head *plist;
>> +
>> +    dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
>> +    if (!dev_dir)
>> +        return;
>> +
>> +    list_for_each(plist, &(dev_dir->d_subdirs)) {
>> +        sub_dir = list_entry(plist, struct dentry, d_child);
>> +        if(sub_dir) {
>> +            dentry = debugfs_lookup("domain_translation_struct",
>> +                        sub_dir);
>> +            if (!dentry)
>> +                continue;
>> +
>> +            if (dentry->d_inode->i_private)
>> +                kfree(dentry->d_inode->i_private);
>> +
>> +            dput(dentry);
>> +        }
>> +    }
>> +
>> +    debugfs_remove_recursive(dev_dir);
>> +    dput(dev_dir);
>> +}
>
> And this could simply be like this:
>
> void intel_iommu_debugfs_remove_dev(struct device *dev)
> {
>     struct device_domain_info *info = dev_iommu_priv_get(dev);
>
>     debugfs_remove(info->debugfs_entry);
> }
>
Yes.
Just get the debugfs dentry of device simply for removing.
This helper should be called before the "info" is freed
in intel_iommu_release_device(). Like this:

+      intel_iommu_debugfs_remove_dev(dev);
         kfree(info);

>> +
>> +/*
>> + * Create a debugfs directory per pair of {device, pasid},
>> + * then create the corresponding debugfs file in this directory
>> + * for user to dump its page table. e.g.
>> + * 
>> /sys/kernel/debug/iommu/intel/0000:00:01.0/0/domain_translation_struct
>> + */
>> +void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
>> +                      struct device *dev, u32 pasid)
>> +{
>> +    struct dentry *dev_dir, *pasid_dir;
>> +    struct show_domain_info *sinfo;
>> +    char dir_name[10];
>> +
>> +    /*
>> +     * The debugfs only dumps the page tables whose mappings are 
>> created
>> +     * and destroyed by the iommu_map/unmap() interfaces. Check the
>> +     * mapping type of the domain before creating debugfs directory.
>> +     */
>> +    if (!domain || !(domain->type & __IOMMU_DOMAIN_PAGING))
>> +        return;
>> +
>> +    dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
>> +    if (!dev_dir)
>> +        return;
>> +
>> +    sprintf(dir_name, "%x", pasid);
>> +    pasid_dir = debugfs_create_dir(dir_name, dev_dir);
>> +    if (IS_ERR(pasid_dir))
>> +        goto dput_out;
>> +
>> +    sinfo = kzalloc(sizeof(*sinfo), GFP_KERNEL);
>> +    if (!sinfo)
>> +        goto dput_out;
>> +
>> +    sinfo->dev = dev;
>> +    sinfo->pasid = pasid;
>> +    debugfs_create_file("domain_translation_struct", 0444,
>> +                pasid_dir, sinfo,
>> +                &domain_translation_struct_fops);
>> +dput_out:
>> +    dput(dev_dir);
>> +}
>
> And here,
>
> void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
>                       struct device *dev, u32 pasid)
> {
>     sprintf(dir_name, "%x", pasid);
>     dev_pasid->debugfs_entry = debugfs_create_dir(dir_name,
>             info->debugfs_entry);
>
>     debugfs_create_file("domain_translation_struct", 0444,
>                 dev_pasid->debugfs_entry, dev_pasid,
>                 &domain_translation_struct_fops);
> }
>
Thanks.
You mean to add 'debugfs_entry' in below structure.
     struct dev_pasid_info *dev_pasid;
This structure is also allocated per pair of {dev, pasid}.
The debugfs dentry of  {dev, pasid} can be simply obtained from 
'dev_pasid_info'.

So the 'dev_pasid_info' can be passed as a parameter of this helper, right ?
Like this:
void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
                       struct dev_pasid_info *dev_pasid) ;
>> +
>> +/*
>> + * Remove the debugfs directory and file corresponding to each pair of
>> + * {device, pasid}.
>> + */
>> +void intel_iommu_debugfs_remove_dev_pasid(struct device *dev, u32 
>> pasid)
>> +{
>> +    struct dentry *dev_dir, *pasid_dir, *dentry;
>> +    char dir_name[10];
>> +
>> +    dev_dir = debugfs_lookup(dev_name(dev), intel_iommu_debug);
>> +    if (!dev_dir)
>> +        return;
>> +
>> +    sprintf(dir_name, "%x", pasid);
>> +    pasid_dir = debugfs_lookup(dir_name, dev_dir);
>> +    if (!pasid_dir)
>> +        goto dput_dev;
>> +
>> +    dentry = debugfs_lookup("domain_translation_struct", pasid_dir);
>> +    if (!dentry)
>> +        goto dput_pasid;
>> +
>> +    if (dentry->d_inode->i_private)
>> +        kfree(dentry->d_inode->i_private);
>> +
>> +    debugfs_remove_recursive(pasid_dir);
>> +
>> +    dput(dentry);
>> +dput_pasid:
>> +    dput(pasid_dir);
>> +dput_dev:
>> +    dput(dev_dir);
>> +}
>
> The same thing here:
>
>     debugfs_remove(dev_pasid->debugfs_entry);
>
Yes.
Just get the debugfs dentry from "dev_pasid" instead of 'debugfs_lookup()'.
And this helper should be called before the "struct dev_pasid_info" is freed
in intel_iommu_remove_dev_pasid().
Like this:

+        intel_iommu_debugfs_remove_dev_pasid(dev, pasid);
           kfree(dev_pasid);

>> diff --git a/drivers/iommu/intel/iommu.c b/drivers/iommu/intel/iommu.c
>> index dd8ff358867d..af9c989035a2 100644
>> --- a/drivers/iommu/intel/iommu.c
>> +++ b/drivers/iommu/intel/iommu.c
>> @@ -2488,6 +2488,13 @@ static int dmar_domain_attach_device(struct 
>> dmar_domain *domain,
>>         iommu_enable_pci_caps(info);
>>   +    /*
>> +     * Create a debugfs directory specified by RID_PASID
>> +     * in the debugfs device directory.
>> +     */
>> + intel_iommu_debugfs_create_dev_pasid(&info->domain->domain,
>> +                                 dev, IOMMU_NO_PASID);
>
> The function name is self-explained. So no need to add comments. Ditto
> to all other places.
>
Indeed.
I'll delete all related comments.

Thanks,
Jingqi

Re: [PATH v3 2/3] iommu/vt-d: debugfs: Create/remove debugfs file per {device, pasid}

[Baolu Lu]

On 2023/9/28 17:02, Liu, Jingqi wrote:
>> And here,
>>
>> void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
>>                       struct device *dev, u32 pasid)
>> {
>>     sprintf(dir_name, "%x", pasid);
>>     dev_pasid->debugfs_entry = debugfs_create_dir(dir_name,
>>             info->debugfs_entry);
>>
>>     debugfs_create_file("domain_translation_struct", 0444,
>>                 dev_pasid->debugfs_entry, dev_pasid,
>>                 &domain_translation_struct_fops);
>> }
>>
> Thanks.
> You mean to add 'debugfs_entry' in below structure.
>      struct dev_pasid_info *dev_pasid;
> This structure is also allocated per pair of {dev, pasid}.
> The debugfs dentry of  {dev, pasid} can be simply obtained from 
> 'dev_pasid_info'.
> 
> So the 'dev_pasid_info' can be passed as a parameter of this helper, 
> right ?
> Like this:
> void intel_iommu_debugfs_create_dev_pasid(struct iommu_domain *domain,
>                        struct dev_pasid_info *dev_pasid) ;

Yes.

Best regards,
baolu

[PATH v3 3/3] iommu/vt-d: debugfs: Support dumping a specified page table

[Jingqi Liu]

The original debugfs only dumps all page tables without pasid. With
pasid supported, the page table with pasid also needs to be dumped.

This patch supports dumping a specified page table in legacy mode or
scalable mode with or without a specified pasid.

For legacy mode, according to bus number and DEVFN, traverse the root
table and context table to get the pointer of page table in the
context table entry, then dump the specified page table.

For scalable mode, according to bus number, DEVFN and pasid, traverse
the root table, context table, pasid directory and pasid table to get
the pointer of page table in the pasid table entry, then dump the
specified page table..

Examples are as follows:
1) Dump the page table of device "0000:00:1f.0" that only supports
   legacy mode.
   $ sudo cat
   /sys/kernel/debug/iommu/intel/0000:00:1f.0/0/domain_translation_struct

2) Dump the page table of device "0000:00:0a.0" with PASID "1" that
   supports scalable mode.
   $ sudo cat
   /sys/kernel/debug/iommu/intel/0000:00:0a.0/1/domain_translation_struct

Suggested-by: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Jingqi Liu <Jingqi.liu@intel.com>
---
 drivers/iommu/intel/debugfs.c | 163 +++++++++++++++++++++++++---------
 1 file changed, 121 insertions(+), 42 deletions(-)

diff --git a/drivers/iommu/intel/debugfs.c b/drivers/iommu/intel/debugfs.c
index 50dea53ba87c..a0e7f2b4d4d1 100644
--- a/drivers/iommu/intel/debugfs.c
+++ b/drivers/iommu/intel/debugfs.c
@@ -352,58 +352,137 @@ static void pgtable_walk_level(struct seq_file *m, struct dma_pte *pde,
 	}
 }
 
-static int __show_device_domain_translation(struct device *dev, void *data)
+static int domain_translation_struct_show(struct seq_file *m, void *unused)
 {
-	struct dmar_domain *domain;
-	struct seq_file *m = data;
-	u64 path[6] = { 0 };
+	struct device_domain_info *info;
+	struct show_domain_info *sinfo;
+	bool scalable, found = false;
+	struct dmar_drhd_unit *drhd;
+	struct intel_iommu *iommu;
+	u16 devfn, bus, seg;
 
-	domain = to_dmar_domain(iommu_get_domain_for_dev(dev));
-	if (!domain)
-		return 0;
+	if (!m || !m->private) {
+		seq_puts(m, "Invalid device or pasid!\n");
+		return -EINVAL;
+	}
 
-	seq_printf(m, "Device %s @0x%llx\n", dev_name(dev),
-		   (u64)virt_to_phys(domain->pgd));
-	seq_puts(m, "IOVA_PFN\t\tPML5E\t\t\tPML4E\t\t\tPDPE\t\t\tPDE\t\t\tPTE\n");
+	sinfo = (struct show_domain_info*)m->private;
+	if (!sinfo->dev ||
+	    !dev_iommu_priv_get(sinfo->dev) ||
+	    (sinfo->pasid == IOMMU_PASID_INVALID)) {
+		seq_puts(m, "Please specify device or pasid!\n");
+		return -ENODEV;
+	}
 
-	pgtable_walk_level(m, domain->pgd, domain->agaw + 2, 0, path);
-	seq_putc(m, '\n');
+	info = dev_iommu_priv_get(sinfo->dev);
+	bus = info->bus;
+	devfn = info->devfn;
+	seg = info->segment;
 
-	/* Don't iterate */
-	return 1;
-}
+	rcu_read_lock();
+	for_each_active_iommu(iommu, drhd) {
+		struct context_entry *context;
+		u64 pgd, path[6] = { 0 };
+		u32 sts, agaw;
 
-static int show_device_domain_translation(struct device *dev, void *data)
-{
-	struct iommu_group *group;
+		if (seg != iommu->segment)
+			continue;
 
-	device_lock(dev);
-	group = iommu_group_get(dev);
-	device_unlock(dev);
-	if (!group)
-		return 0;
+		sts = dmar_readl(iommu->reg + DMAR_GSTS_REG);
+		if (!(sts & DMA_GSTS_TES)) {
+			seq_printf(m, "DMA Remapping is not enabled on %s\n",
+				   iommu->name);
+			continue;
+		}
+		if (dmar_readq(iommu->reg + DMAR_RTADDR_REG) & DMA_RTADDR_SMT)
+			scalable = true;
+		else
+			scalable = false;
 
-	/*
-	 * The group->mutex is held across the callback, which will
-	 * block calls to iommu_attach/detach_group/device. Hence,
-	 * the domain of the device will not change during traversal.
-	 *
-	 * All devices in an iommu group share a single domain, hence
-	 * we only dump the domain of the first device. Even though,
-	 * this code still possibly races with the iommu_unmap()
-	 * interface. This could be solved by RCU-freeing the page
-	 * table pages in the iommu_unmap() path.
-	 */
-	iommu_group_for_each_dev(group, data, __show_device_domain_translation);
-	iommu_group_put(group);
+		/*
+		 * The iommu->lock is held across the callback, which will
+		 * block calls to domain_attach/domain_detach. Hence,
+		 * the domain of the device will not change during traversal.
+		 *
+		 * Traversing page table possibly races with the iommu_unmap()
+		 * interface. This could be solved by RCU-freeing the page
+		 * table pages in the iommu_unmap() path.
+		 */
+		spin_lock(&iommu->lock);
 
-	return 0;
-}
+		context = iommu_context_addr(iommu, bus, devfn, 0);
+		if (!context || !context_present(context))
+			goto iommu_unlock;
 
-static int domain_translation_struct_show(struct seq_file *m, void *unused)
-{
-	return bus_for_each_dev(&pci_bus_type, NULL, m,
-				show_device_domain_translation);
+		if (scalable) {	/* scalable mode */
+			struct pasid_dir_entry *dir_tbl, *dir_entry;
+			struct pasid_entry *pasid_tbl, *pasid_tbl_entry;
+			u16 pasid_dir_size, dir_idx, tbl_idx, pgtt;
+			u64 pasid_dir_ptr;
+
+			pasid_dir_ptr = context->lo & VTD_PAGE_MASK;
+			pasid_dir_size = get_pasid_dir_size(context);
+
+			/* Dump specified device domain mappings with PASID. */
+			dir_idx = sinfo->pasid >> PASID_PDE_SHIFT;
+			tbl_idx = sinfo->pasid & PASID_PTE_MASK;
+
+			dir_tbl = phys_to_virt(pasid_dir_ptr);
+			dir_entry = &dir_tbl[dir_idx];
+
+			pasid_tbl = get_pasid_table_from_pde(dir_entry);
+			if (!pasid_tbl)
+				goto iommu_unlock;
+
+			pasid_tbl_entry = &pasid_tbl[tbl_idx];
+			if (!pasid_pte_is_present(pasid_tbl_entry))
+				goto iommu_unlock;
+
+			/*
+			 * According to PASID Granular Translation Type(PGTT),
+			 * get the page table pointer.
+			 */
+			pgtt = (u16)(pasid_tbl_entry->val[0] & GENMASK_ULL(8, 6)) >> 6;
+			agaw = (u8)(pasid_tbl_entry->val[0] & GENMASK_ULL(4, 2)) >> 2;
+
+			switch (pgtt) {
+				case PASID_ENTRY_PGTT_FL_ONLY:
+					pgd = pasid_tbl_entry->val[2];
+					break;
+				case PASID_ENTRY_PGTT_SL_ONLY:
+				case PASID_ENTRY_PGTT_NESTED:
+					pgd = pasid_tbl_entry->val[0];
+					break;
+				default:
+					goto iommu_unlock;
+			}
+			pgd &= VTD_PAGE_MASK;
+		} else { /* legacy mode */
+			pgd = context->lo & VTD_PAGE_MASK;
+			agaw = context->hi & 7;
+		}
+
+		seq_printf(m, "Device %04x:%02x:%02x.%x ",
+			   iommu->segment, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
+
+		if (scalable)
+			seq_printf(m, "with pasid %x @0x%llx\n", sinfo->pasid, pgd);
+		else
+			seq_printf(m, "@0x%llx\n", pgd);
+
+		seq_printf(m, "%-17s\t%-18s\t%-18s\t%-18s\t%-18s\t%-s\n",
+			   "IOVA_PFN", "PML5E", "PML4E", "PDPE", "PDE", "PTE");
+		pgtable_walk_level(m, phys_to_virt(pgd), agaw + 2, 0, path);
+
+		found = true;
+iommu_unlock:
+		spin_unlock(&iommu->lock);
+		if (found)
+			break;
+	}
+	rcu_read_unlock();
+
+	return 0;
 }
 DEFINE_SHOW_ATTRIBUTE(domain_translation_struct);
 
-- 
2.21.3