[PATCH] drm/amdkfd: add pasid debugfs entries

[Kasiviswanathan, Harish]

[Public]

Hi Xiaogang,

With support for partitioning, we have multiple pasids. Pasid's are not useful for regular user. It is only for debug purposes and hence it is exposed via debugfs.

Hi Eric,
Please fix checkpatch errors. With that fixed this patch is Reviewed-by: Harish Kasiviswanathan <Harish.Kasiviswanathan at amd.com>

Best Regards,
Harish


-----Original Message-----
From: Chen, Xiaogang <Xiaogang.Chen at amd.com>
Sent: Thursday, April 24, 2025 4:55 PM
To: Huang, JinHuiEric <JinHuiEric.Huang at amd.com>; amd-gfx at lists.freedesktop.org
Cc: Kasiviswanathan, Harish <Harish.Kasiviswanathan at amd.com>
Subject: Re: [PATCH] drm/amdkfd: add pasid debugfs entries


On 4/24/2025 3:04 PM, Eric Huang wrote:
> the entries will be appearing at
> /sys/kernel/debug/kfd/proc/<pid>/pasid_<gpuid>.

why need move pasid from sysfs to debugfs? debugfs is not always
available and does not hide pasid from users either.

Regards

Xiaogang

>
> Signed-off-by: Eric Huang <jinhuieric.huang at amd.com>
> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c | 77 ++++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdkfd/kfd_priv.h    |  5 ++
>   drivers/gpu/drm/amd/amdkfd/kfd_process.c |  3 +
>   3 files changed, 85 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c b/drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c
> index 4a5a0a4e00f2..14ab770fb877 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c
> @@ -27,6 +27,16 @@
>   #include "kfd_priv.h"
>
>   static struct dentry *debugfs_root;
> +static struct dentry *debugfs_proc;
> +static struct list_head procs;
> +
> +struct debugfs_proc_entry {
> +     struct list_head list;
> +     struct dentry *proc_dentry;
> +     pid_t pid;
> +};
> +
> +#define MAX_DEBUGFS_FILENAME_LEN 32
>
>   static int kfd_debugfs_open(struct inode *inode, struct file *file)
>   {
> @@ -92,6 +102,8 @@ static const struct file_operations kfd_debugfs_hang_hws_fops = {
>   void kfd_debugfs_init(void)
>   {
>       debugfs_root = debugfs_create_dir("kfd", NULL);
> +     debugfs_proc = debugfs_create_dir("proc", debugfs_root);
> +     INIT_LIST_HEAD(&procs);
>
>       debugfs_create_file("mqds", S_IFREG | 0444, debugfs_root,
>                           kfd_debugfs_mqds_by_process, &kfd_debugfs_fops);
> @@ -107,5 +119,70 @@ void kfd_debugfs_init(void)
>
>   void kfd_debugfs_fini(void)
>   {
> +     debugfs_remove_recursive(debugfs_proc);
>       debugfs_remove_recursive(debugfs_root);
>   }
> +
> +static ssize_t kfd_debugfs_pasid_read(struct file *file, char __user *buf,
> +                                   size_t count, loff_t *ppos)
> +{
> +    struct kfd_process_device *pdd = file_inode(file)->i_private;
> +    char tmp[32];
> +    int len;
> +
> +    len = snprintf(tmp, sizeof(tmp), "%u\n", pdd->pasid);
> +
> +    return simple_read_from_buffer(buf, count, ppos, tmp, len);
> +}
> +
> +static const struct file_operations kfd_debugfs_pasid_fops = {
> +    .owner = THIS_MODULE,
> +    .read = kfd_debugfs_pasid_read,
> +};
> +
> +void kfd_debugfs_add_process(struct kfd_process *p)
> +{
> +     int i;
> +     char name[MAX_DEBUGFS_FILENAME_LEN];
> +     struct debugfs_proc_entry *entry;
> +
> +     entry = kzalloc(sizeof(*entry), GFP_KERNEL);
> +     if (!entry)
> +             return;
> +
> +     list_add(&entry->list, &procs);
> +     entry->pid = p->lead_thread->pid;
> +     snprintf(name, MAX_DEBUGFS_FILENAME_LEN, "%d",
> +              (int)entry->pid);
> +     entry->proc_dentry = debugfs_create_dir(name, debugfs_proc);
> +
> +     /* Create debugfs files for each GPU:
> +      * - proc/<pid>/pasid_<gpuid>
> +      */
> +     for (i = 0; i < p->n_pdds; i++) {
> +             struct kfd_process_device *pdd = p->pdds[i];
> +
> +             snprintf(name, MAX_DEBUGFS_FILENAME_LEN, "pasid_%u",
> +                      pdd->dev->id);
> +             debugfs_create_file((const char *)name, S_IFREG | 0444,
> +                                 entry->proc_dentry,
> +                                 pdd,
> +                                 &kfd_debugfs_pasid_fops);
> +     }
> +}
> +
> +void kfd_debugfs_remove_process(struct kfd_process *p)
> +{
> +     struct debugfs_proc_entry *entry, *next;
> +
> +     mutex_lock(&kfd_processes_mutex);
> +     list_for_each_entry_safe(entry, next, &procs, list) {
> +             if (entry->pid != p->lead_thread->pid)
> +                     continue;
> +
> +             debugfs_remove_recursive(entry->proc_dentry);
> +             list_del(&entry->list);
> +             kfree(entry);
> +     }
> +     mutex_unlock(&kfd_processes_mutex);
> +}
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> index f6aedf69c644..0ae794539bb0 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> @@ -1581,10 +1581,15 @@ int kfd_debugfs_hang_hws(struct kfd_node *dev);
>   int pm_debugfs_hang_hws(struct packet_manager *pm);
>   int dqm_debugfs_hang_hws(struct device_queue_manager *dqm);
>
> +void kfd_debugfs_add_process(struct kfd_process *p);
> +void kfd_debugfs_remove_process(struct kfd_process *p);
> +
>   #else
>
>   static inline void kfd_debugfs_init(void) {}
>   static inline void kfd_debugfs_fini(void) {}
> +static inline void kfd_debugfs_add_process(struct kfd_process *p) {}
> +static inline void kfd_debugfs_remove_process(struct kfd_process *p) {}
>
>   #endif
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> index 41d7dc8c2850..722ac1662bdc 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> @@ -900,6 +900,8 @@ struct kfd_process *kfd_create_process(struct task_struct *thread)
>               kfd_procfs_add_sysfs_files(process);
>               kfd_procfs_add_sysfs_counters(process);
>
> +             kfd_debugfs_add_process(process);
> +
>               init_waitqueue_head(&process->wait_irq_drain);
>       }
>   out:
> @@ -1176,6 +1178,7 @@ static void kfd_process_wq_release(struct work_struct *work)
>               dma_fence_signal(ef);
>
>       kfd_process_remove_sysfs(p);
> +     kfd_debugfs_remove_process(p);
>
>       kfd_process_kunmap_signal_bo(p);
>       kfd_process_free_outstanding_kfd_bos(p);