On 19/01/2018 13:45, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin <tvrtko.ursu...@intel.com>

Expose per-client and per-engine busyness under the previously added sysfs
client root.

The new file is named 'busy' and contains a list of, one line for each
engine, monotonically increasing nano-second resolution times each
client's jobs were executing on the GPU.

$ cat /sys/class/drm/card0/clients/5/busy
32516602
0
0
0

This data can serve as an interface to implement a top like utility for
GPU jobs. For instance I have prototyped a tool in IGT which produces
periodic output like:

neverball[  6011]:  rcs0:  41.01%  bcs0:   0.00%  vcs0:   0.00%  vecs0:   0.00%
      Xorg[  5664]:  rcs0:  31.16%  bcs0:   0.00%  vcs0:   0.00%  vecs0:   0.00%
     xfwm4[  5727]:  rcs0:   0.00%  bcs0:   0.00%  vcs0:   0.00%  vecs0:   0.00%

This tools can also be extended to use the i915 PMU and show overall engine
busyness, and engine loads using the queue depth metric.

v2: Use intel_context_engine_get_busy_time.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursu...@intel.com>
---
  drivers/gpu/drm/i915/i915_drv.h |  1 +
  drivers/gpu/drm/i915/i915_gem.c | 66 +++++++++++++++++++++++++++++++++++++++++
  2 files changed, 67 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index a2a883aaa59e..a96cfdfcba03 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -350,6 +350,7 @@ struct drm_i915_file_private {
        struct {
                struct device_attribute pid;
                struct device_attribute name;
+               struct device_attribute busy;
        } attr;
  };
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 7fdd892aa32a..c8753e4d66c4 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -5538,6 +5538,57 @@ show_client_pid(struct device *kdev, struct 
device_attribute *attr, char *buf)
        return snprintf(buf, PAGE_SIZE, "%u", file_priv->client_pid);
  }
+struct busy_ctx {
+       u64 total[I915_NUM_ENGINES];
+};
+
+static int busy_add(int _id, void *p, void *data)
+{
+       struct i915_gem_context *ctx = p;
+       struct busy_ctx *bc = data;
+       struct intel_engine_cs *engine;
+       enum intel_engine_id id;
+
+       for_each_engine(engine, ctx->i915, id)
+               bc->total[id] +=
+                  ktime_to_ns(intel_context_engine_get_busy_time(ctx, engine));
+
+       return 0;
+}
+
+static ssize_t
+show_client_busy(struct device *kdev, struct device_attribute *attr, char *buf)
+{
+       struct drm_i915_file_private *file_priv =
+               container_of(attr, struct drm_i915_file_private, attr.busy);
+       struct drm_i915_private *i915 = file_priv->dev_priv;
+       unsigned int len = PAGE_SIZE;
+       struct busy_ctx bc = { };
+       ssize_t res = 0;
+       struct intel_engine_cs *engine;
+       enum intel_engine_id id;
+       ssize_t ret;
+
+       ret = i915_mutex_lock_interruptible(&i915->drm);
+       if (ret)
+               return ret;
+
+       idr_for_each(&file_priv->context_idr, busy_add, &bc);
+
+       mutex_unlock(&i915->drm.struct_mutex);

This is the part which could be done more efficiently by adding an aggregated busyness to file_priv as a couple atomic_t (busy count and active count I think), and incrementing it from the context_in/context_out hooks as they go. Then the client busyness query would be both lockless and wouldn't need to iterate the file_priv->context_id.

It would probably be preferable to do it like that instead of having to take struct_mutex here. But until there is some idea if people are warming up to this feature in general I did not bother fully polishing it.

Regards,

Tvrtko

+
+       for_each_engine(engine, i915, id) {
+               ret = snprintf(buf, len, "%llu\n", bc.total[id]);
+               if (ret <= 0)
+                       break;
+               res += ret;
+               len -= ret;
+               buf += ret;
+       }
+
+       return res;
+}
+
  int
  i915_gem_add_client(struct drm_i915_private *i915,
                struct drm_i915_file_private *file_priv,
@@ -5578,10 +5629,23 @@ i915_gem_add_client(struct drm_i915_private *i915,
        if (ret)
                goto err_attr_pid;
+ attr = &file_priv->attr.busy;
+       attr->attr.name = "busy";
+       attr->attr.mode = 0444;
+       attr->show = show_client_busy;
+
+       ret = sysfs_create_file(file_priv->client_root,
+                               (struct attribute *)attr);
+       if (ret)
+               goto err_attr_busy;
+
        file_priv->client_pid = pid_nr(get_task_pid(task, PIDTYPE_PID));
return 0; +err_attr_busy:
+       sysfs_remove_file(file_priv->client_root,
+                         (struct attribute *)&file_priv->attr.pid);
  err_attr_pid:
        sysfs_remove_file(file_priv->client_root,
                          (struct attribute *)&file_priv->attr.name);
@@ -5595,6 +5659,8 @@ i915_gem_add_client(struct drm_i915_private *i915,
void i915_gem_remove_client(struct drm_i915_file_private *file_priv)
  {
+       sysfs_remove_file(file_priv->client_root,
+                         (struct attribute *)&file_priv->attr.busy);
        sysfs_remove_file(file_priv->client_root,
                          (struct attribute *)&file_priv->attr.pid);
        sysfs_remove_file(file_priv->client_root,

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reply via email to