[PATCH] drm/amdkfd: add smi events for process start and end

[Eric Huang]

Ping ...

On 2025-04-07 16:52, Eric Huang wrote:
> rocm-smi will be able to show the events for KFD process
> start/end, it is the implementation of this feature.
>
> Signed-off-by: Eric Huang <jinhuieric.huang at amd.com>
> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_process.c    |  4 ++++
>   drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 21 +++++++++++++++++++++
>   drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h |  1 +
>   include/uapi/linux/kfd_ioctl.h              |  5 +++++
>   4 files changed, 31 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> index 7c0c24732481..41d7dc8c2850 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> @@ -1054,6 +1054,8 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)
>   	for (i = 0; i < p->n_pdds; i++) {
>   		struct kfd_process_device *pdd = p->pdds[i];
>   
> +		kfd_smi_event_process(pdd, false);
> +
>   		pr_debug("Releasing pdd (topology id %d, for pid %d)\n",
>   			pdd->dev->id, p->lead_thread->pid);
>   		kfd_process_device_destroy_cwsr_dgpu(pdd);
> @@ -1715,6 +1717,8 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
>   	pdd->pasid = avm->pasid;
>   	pdd->drm_file = drm_file;
>   
> +	kfd_smi_event_process(pdd, true);
> +
>   	return 0;
>   
>   err_get_pasid:
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> index 9b8169761ec5..727a4ce29fe6 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> @@ -345,6 +345,27 @@ void kfd_smi_event_unmap_from_gpu(struct kfd_node *node, pid_t pid,
>   			  pid, address, last - address + 1, node->id, trigger));
>   }
>   
> +void kfd_smi_event_process(struct kfd_process_device *pdd, bool start)
> +{
> +	struct amdgpu_task_info *task_info;
> +	struct amdgpu_vm *avm;
> +
> +	if (pdd->drm_priv)
> +		return;
> +
> +	avm = drm_priv_to_vm(pdd->drm_priv);
> +	task_info = amdgpu_vm_get_task_info_vm(avm);
> +
> +	if (task_info) {
> +		kfd_smi_event_add(0, pdd->dev,
> +				  start ? KFD_SMI_EVENT_PROCESS_START :
> +				  KFD_SMI_EVENT_PROCESS_END,
> +				  KFD_EVENT_FMT_PROCESS(task_info->pid,
> +				  task_info->task_name));
> +		amdgpu_vm_put_task_info(task_info);
> +	}
> +}
> +
>   int kfd_smi_event_open(struct kfd_node *dev, uint32_t *fd)
>   {
>   	struct kfd_smi_client *client;
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
> index 503bff13d815..bb4d72b57387 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
> @@ -53,4 +53,5 @@ void kfd_smi_event_queue_restore_rescheduled(struct mm_struct *mm);
>   void kfd_smi_event_unmap_from_gpu(struct kfd_node *node, pid_t pid,
>   				  unsigned long address, unsigned long last,
>   				  uint32_t trigger);
> +void kfd_smi_event_process(struct kfd_process_device *pdd, bool start);
>   #endif
> diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
> index 1e59344c5673..04c7d283dc7d 100644
> --- a/include/uapi/linux/kfd_ioctl.h
> +++ b/include/uapi/linux/kfd_ioctl.h
> @@ -536,6 +536,8 @@ enum kfd_smi_event {
>   	KFD_SMI_EVENT_QUEUE_EVICTION = 9,
>   	KFD_SMI_EVENT_QUEUE_RESTORE = 10,
>   	KFD_SMI_EVENT_UNMAP_FROM_GPU = 11,
> +	KFD_SMI_EVENT_PROCESS_START = 12,
> +	KFD_SMI_EVENT_PROCESS_END = 13,
>   
>   	/*
>   	 * max event number, as a flag bit to get events from all processes,
> @@ -651,6 +653,9 @@ struct kfd_ioctl_smi_events_args {
>   		"%lld -%d @%lx(%lx) %x %d\n", (ns), (pid), (addr), (size),\
>   		(node), (unmap_trigger)
>   
> +#define KFD_EVENT_FMT_PROCESS(pid, task_name)\
> +		"%x %s\n", (pid), (task_name)
> +
>   /**************************************************************************************************
>    * CRIU IOCTLs (Checkpoint Restore In Userspace)
>    *