Re: [PATCH] drm/panfrost: Queue jobs on the hardware

On 8/16/19 11:31 AM, Steven Price wrote:

The hardware has a set of '_NEXT' registers that can hold a second job while the first is executing. Make use of these registers to enqueue a second job per slot.

I like this in principle, but upon some quick testing I found that Mesa is around 10% slower with this patch (when using the performance governor).

There's also the question of how this affects the utilization calculation in the devfreq code.

I will be trying to find time to understand why Mesa is slower and not faster, but TBH performance doesn't have top priority for me yet. Would be great if somebody else could look at it.

Thanks,

Tomeu

Signed-off-by: Steven Price steven.price@arm.com

Note that this is based on top of Rob Herring's "per FD address space" patch[1].

[1] https://marc.info/?i=20190813150115.30338-1-robh%20()%20kernel%20!%20org

drivers/gpu/drm/panfrost/panfrost_device.h | 4 +- drivers/gpu/drm/panfrost/panfrost_job.c | 76 ++++++++++++++++++---- drivers/gpu/drm/panfrost/panfrost_mmu.c | 2 +- 3 files changed, 67 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/panfrost/panfrost_device.h b/drivers/gpu/drm/panfrost/panfrost_device.h index f503c566e99f..0153defd6085 100644 --- a/drivers/gpu/drm/panfrost/panfrost_device.h +++ b/drivers/gpu/drm/panfrost/panfrost_device.h @@ -55,7 +55,7 @@ struct panfrost_devfreq_slot { ktime_t busy_time; ktime_t idle_time; ktime_t time_last_update;

• bool busy;

• int busy; };

  struct panfrost_device {

@@ -80,7 +80,7 @@ struct panfrost_device {

struct panfrost_job_slot *js;

• struct panfrost_job *jobs[NUM_JOB_SLOTS];

• struct panfrost_job *jobs[NUM_JOB_SLOTS][2]; struct list_head scheduled_jobs;

  struct panfrost_perfcnt *perfcnt;

diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c index 05c85f45a0de..b2b5027af976 100644 --- a/drivers/gpu/drm/panfrost/panfrost_job.c +++ b/drivers/gpu/drm/panfrost/panfrost_job.c @@ -138,6 +138,37 @@ static void panfrost_job_write_affinity(struct panfrost_device *pfdev, job_write(pfdev, JS_AFFINITY_NEXT_HI(js), affinity >> 32); }

+static int panfrost_job_count(struct panfrost_device *pfdev, int slot) +{

• if (pfdev->jobs[slot][0] == NULL)

• return 0;
  

• if (pfdev->jobs[slot][1] == NULL)

• return 1;
  

• return 2;

+}

+static struct panfrost_job *panfrost_dequeue_job(

• struct panfrost_device *pfdev, int slot)
  

+{

• struct panfrost_job *job = pfdev->jobs[slot][0];
• pfdev->jobs[slot][0] = pfdev->jobs[slot][1];
• pfdev->jobs[slot][1] = NULL;
• return job;

+}

+static void panfrost_enqueue_job(struct panfrost_device *pfdev, int slot,

• 		 struct panfrost_job *job)
  

+{

• if (pfdev->jobs[slot][0] == NULL) {

• pfdev->jobs[slot][0] = job;
  

• return;
  

• }
• WARN_ON(pfdev->jobs[slot][1] != NULL);
• pfdev->jobs[slot][1] = job;

+}

• static void panfrost_job_hw_submit(struct panfrost_job *job, int js) { struct panfrost_device *pfdev = job->pfdev;

@@ -150,13 +181,16 @@ static void panfrost_job_hw_submit(struct panfrost_job *job, int js) if (ret < 0) return;

• if (WARN_ON(job_read(pfdev, JS_COMMAND_NEXT(js))))

• goto end;
  

• cfg = panfrost_mmu_as_get(pfdev, &job->file_priv->mmu);

• panfrost_devfreq_record_transition(pfdev, js); spin_lock_irqsave(&pfdev->hwaccess_lock, flags);

• panfrost_enqueue_job(pfdev, js, job);

• if (WARN_ON(job_read(pfdev, JS_COMMAND_NEXT(js))))

• goto end;
  

• if (panfrost_job_count(pfdev, js) == 1)

• panfrost_devfreq_record_transition(pfdev, js);
  
  

  job_write(pfdev, JS_HEAD_NEXT_LO(js), jc_head & 0xFFFFFFFF); job_write(pfdev, JS_HEAD_NEXT_HI(js), jc_head >> 32);

@@ -186,9 +220,9 @@ static void panfrost_job_hw_submit(struct panfrost_job *job, int js)

job_write(pfdev, JS_COMMAND_NEXT(js), JS_COMMAND_START);

+end: spin_unlock_irqrestore(&pfdev->hwaccess_lock, flags);

-end: pm_runtime_mark_last_busy(pfdev->dev); pm_runtime_put_autosuspend(pfdev->dev); } @@ -336,8 +370,6 @@ static struct dma_fence *panfrost_job_run(struct drm_sched_job *sched_job) if (unlikely(job->base.s_fence->finished.error)) return NULL;

• pfdev->jobs[slot] = job;
• fence = panfrost_fence_create(pfdev, slot); if (IS_ERR(fence)) return NULL;

@@ -421,21 +453,36 @@ static irqreturn_t panfrost_job_irq_handler(int irq, void *data) struct panfrost_device *pfdev = data; u32 status = job_read(pfdev, JOB_INT_STAT); int j;

• unsigned long flags;

  dev_dbg(pfdev->dev, "jobslot irq status=%x\n", status);

  if (!status) return IRQ_NONE;

• spin_lock_irqsave(&pfdev->hwaccess_lock, flags);

• pm_runtime_mark_last_busy(pfdev->dev);

  for (j = 0; status; j++) { u32 mask = MK_JS_MASK(j);

• int jobs = panfrost_job_count(pfdev, j);
  

• int active;
  
  

  if (!(status & mask)) continue;

  job_write(pfdev, JOB_INT_CLEAR, mask);

• active = (job_read(pfdev, JOB_INT_JS_STATE) &
  

• 	  JOB_INT_MASK_DONE(j)) ? 1 : 0;
  

• if (!(status & JOB_INT_MASK_ERR(j))) {
  

• 	/* Recheck RAWSTAT to check if there's a newly
  

• 	 * failed job (since JOB_INT_STAT was read)
  

• 	 */
  

• 	status |= job_read(pfdev, JOB_INT_RAWSTAT) &
  

• 		JOB_INT_MASK_ERR(j);
  

• }
  
  

  if (status & JOB_INT_MASK_ERR(j)) { job_write(pfdev, JS_COMMAND_NEXT(j), JS_COMMAND_NOP);

@@ -447,20 +494,25 @@ static irqreturn_t panfrost_job_irq_handler(int irq, void *data) job_read(pfdev, JS_TAIL_LO(j)));

	drm_sched_fault(&pfdev->js->queue[j].sched);

• 	jobs --;
  

  }

• if (status & JOB_INT_MASK_DONE(j)) {
  

• 	struct panfrost_job *job = pfdev->jobs[j];
  

• while (jobs -- > active) {
  

• 	struct panfrost_job *job =
  

• 		panfrost_dequeue_job(pfdev, j);
  
  

• 	pfdev->jobs[j] = NULL;
  panfrost_mmu_as_put(pfdev, &job->file_priv->mmu);
  

• 	panfrost_devfreq_record_transition(pfdev, j);
  dma_fence_signal(job->done_fence);
  

  }

• if (!active)
  

• 	panfrost_devfreq_record_transition(pfdev, j);
  

• status &= ~mask; }

• spin_unlock_irqrestore(&pfdev->hwaccess_lock, flags);

• return IRQ_HANDLED; }

@@ -491,7 +543,7 @@ int panfrost_job_init(struct panfrost_device *pfdev)

ret = drm_sched_init(&js->queue[j].sched,
		     &panfrost_sched_ops,

• 		     1, 0, msecs_to_jiffies(500),
  

• 		     2, 0, msecs_to_jiffies(500),
  	     "pan_js");
  

  if (ret) { dev_err(pfdev->dev, "Failed to create scheduler: %d.", ret);

diff --git a/drivers/gpu/drm/panfrost/panfrost_mmu.c b/drivers/gpu/drm/panfrost/panfrost_mmu.c index f22d8f02568d..c25fd88ef437 100644 --- a/drivers/gpu/drm/panfrost/panfrost_mmu.c +++ b/drivers/gpu/drm/panfrost/panfrost_mmu.c @@ -147,7 +147,7 @@ u32 panfrost_mmu_as_get(struct panfrost_device *pfdev, struct panfrost_mmu *mmu) as = mmu->as; if (as >= 0) { int en = atomic_inc_return(&mmu->as_count);

• WARN_ON(en >= NUM_JOB_SLOTS);
  

• WARN_ON(en >= NUM_JOB_SLOTS*2);
  
  

  list_move(&mmu->list, &pfdev->as_lru_list); goto out;