Re: [PATCH v2 4/4] drm/amdkfd: SMI report dropped event count

[James Zhu <jamesz@amd.com>]

On 2024-07-30 16:15, Philip Yang wrote:
> Add new SMI event to report the dropped event count when the event kfifo
> is full.
>
> When the kfifo has space for two events, generate a dropped event record
> to report how many events were dropped, together with the next event to
> add to kfifo.
>
> After reading event out from kfifo, if there were events dropped,
> generate a dropped event record and add to kfifo.

[JZ] I am wondering if it is better to use the below method.

can calculate dropped event count during adding event.
Generate/Send dropped event only in event read.
when there are dropped events, If there is space left for dropped 
event,then

send out dropped event with this event read. otherwise, send out
dropped event  with next event read.

>
> Signed-off-by: Philip Yang <Philip.Yang@amd.com>
> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 41 ++++++++++++++++++---
>   include/uapi/linux/kfd_ioctl.h              |  6 +++
>   2 files changed, 41 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> index 9b8169761ec5..9b47657d5160 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> @@ -42,6 +42,7 @@ struct kfd_smi_client {
>   	struct rcu_head rcu;
>   	pid_t pid;
>   	bool suser;
> +	u32 drop_count;
>   };
>   
>   #define KFD_MAX_KFIFO_SIZE	8192
> @@ -103,12 +104,26 @@ static ssize_t kfd_smi_ev_read(struct file *filep, char __user *user,
>   	}
>   	to_copy = min(size, to_copy);
>   	ret = kfifo_out(&client->fifo, buf, to_copy);
> -	spin_unlock(&client->lock);
>   	if (ret <= 0) {
> +		spin_unlock(&client->lock);
>   		ret = -EAGAIN;
>   		goto ret_err;
>   	}
>   
> +	if (client->drop_count) {
> +		char msg[KFD_SMI_EVENT_MSG_SIZE];
> +		int len;
> +
> +		len = snprintf(msg, sizeof(msg), "%x ", KFD_SMI_EVENT_DROPPED_EVENT);
> +		len += snprintf(msg + len, sizeof(msg) - len,
> +				KFD_EVENT_FMT_DROPPED_EVENT(ktime_get_boottime_ns(),
> +				client->pid, client->drop_count));
> +		kfifo_in(&client->fifo, msg, len);
> +		client->drop_count = 0;
> +	}
> +
> +	spin_unlock(&client->lock);
> +
>   	ret = copy_to_user(user, buf, to_copy);
>   	if (ret) {
>   		ret = -EFAULT;
> @@ -173,22 +188,36 @@ static bool kfd_smi_ev_enabled(pid_t pid, struct kfd_smi_client *client,
>   }
>   
>   static void add_event_to_kfifo(pid_t pid, struct kfd_node *dev,
> -			       unsigned int smi_event, char *event_msg, int len)
> +			       unsigned int smi_event, char *event_msg, int event_len)
>   {
>   	struct kfd_smi_client *client;
> +	char msg[KFD_SMI_EVENT_MSG_SIZE];
> +	int len = 0;
>   
>   	rcu_read_lock();
>   
>   	list_for_each_entry_rcu(client, &dev->smi_clients, list) {
>   		if (!kfd_smi_ev_enabled(pid, client, smi_event))
>   			continue;
> +
>   		spin_lock(&client->lock);
> -		if (kfifo_avail(&client->fifo) >= len) {
> -			kfifo_in(&client->fifo, event_msg, len);
> +		if (client->drop_count) {
> +			len = snprintf(msg, sizeof(msg), "%x ", KFD_SMI_EVENT_DROPPED_EVENT);
> +			len += snprintf(msg + len, sizeof(msg) - len,
> +					KFD_EVENT_FMT_DROPPED_EVENT(ktime_get_boottime_ns(), pid,
> +					client->drop_count));
> +		}
> +
> +		if (kfifo_avail(&client->fifo) >= event_len + len) {
> +			if (len)
> +				kfifo_in(&client->fifo, msg, len);
> +			kfifo_in(&client->fifo, event_msg, event_len);
>   			wake_up_all(&client->wait_queue);
> +			client->drop_count = 0;
>   		} else {
> -			pr_debug("smi_event(EventID: %u): no space left\n",
> -					smi_event);
> +			client->drop_count++;
> +			pr_debug("smi_event(EventID: %u): no space left drop_count %d\n",
> +				 smi_event, client->drop_count);
>   		}
>   		spin_unlock(&client->lock);
>   	}
> diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
> index e4ed8fec3294..915d1e7c67fe 100644
> --- a/include/uapi/linux/kfd_ioctl.h
> +++ b/include/uapi/linux/kfd_ioctl.h
> @@ -530,6 +530,7 @@ 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_DROPPED_EVENT = 12,
>   
>   	/*
>   	 * max event number, as a flag bit to get events from all processes,
> @@ -610,6 +611,7 @@ struct kfd_ioctl_smi_events_args {
>    *    rw: 'W' for write page fault, 'R' for read page fault
>    *    rescheduled: 'R' if the queue restore failed and rescheduled to try again
>    *    error_code: migrate failure error code, 0 if no error
> + *    drop_count: how many events dropped when fifo is full
>    */
>   #define KFD_EVENT_FMT_UPDATE_GPU_RESET(reset_seq_num, reset_cause)\
>   		"%x %s\n", (reset_seq_num), (reset_cause)
> @@ -645,6 +647,10 @@ struct kfd_ioctl_smi_events_args {
>   		"%lld -%d @%lx(%lx) %x %d\n", (ns), (pid), (addr), (size),\
>   		(node), (unmap_trigger)
>   
> +#define KFD_EVENT_FMT_DROPPED_EVENT(ns, pid, drop_count)\
> +		"%lld -%d %d\n", (ns), (pid), (drop_count)
> +
> +
>   /**************************************************************************************************
>    * CRIU IOCTLs (Checkpoint Restore In Userspace)
>    *