With the introduction of asymetric slices in CNL, we cannot rely on
the previous SUBSLICE_MASK getparam. Here we introduce a more detailed
way of querying the Gen's GPU topology that doesn't aggregate numbers.
This is essential for monitoring parts of the GPU with the OA unit,
because signals need to be accounted properly based on whether part of
the GPU has been fused off. The current aggregated numbers like
EU_TOTAL do not gives us sufficient information.
Here is the sysfs layout on a Skylake GT4 :
/sys/devices/pci0000:00/0000:00:02.0/drm/card0/gt/rcs/0/
├── capabilities
├── class
├── enabled_mask
├── id
├── max_eus_per_subslice
├── max_slices
├── max_subslices_per_slice
├── slice0
│ ├── enabled_mask
│ ├── subslice0
│ │ └── enabled_mask
│ ├── subslice1
│ │ └── enabled_mask
│ ├── subslice2
│ │ └── enabled_mask
│ └── subslice3
│ └── enabled_mask
├── slice1
│ ├── enabled_mask
│ ├── subslice0
│ │ └── enabled_mask
│ ├── subslice1
│ │ └── enabled_mask
│ ├── subslice2
│ │ └── enabled_mask
│ └── subslice3
│ └── enabled_mask
└── slice2
├── enabled_mask
├── subslice0
│ └── enabled_mask
├── subslice1
│ └── enabled_mask
├── subslice2
│ └── enabled_mask
└── subslice3
└── enabled_mask
Each enabled_mask file gives us a mask of the enabled units :
$ cat /sys/devices/pci0000\:00/0000\:00\:02.0/drm/card0/gt/rcs/0/enabled_mask
0x7
$ cat /sys/devices/pci0000\:00/0000\:00\:02.0/drm/card0/gt/rcs/0/slice0/subslice2/enabled_mask
0xff
v2: Move topology below rcs engine (Chris)
Add max_eus_per_subslice (Lionel)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@xxxxxxxxx>
---
drivers/gpu/drm/i915/i915_drv.h | 26 ++++++
drivers/gpu/drm/i915/i915_sysfs.c | 175 ++++++++++++++++++++++++++++++++++++++
2 files changed, 201 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 22ac96d2ef42..8591c4f7cca3 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2259,6 +2259,24 @@ struct intel_cdclk_state {
u8 voltage_level;
};
+struct intel_topology_kobject {
+ struct kobject kobj;
+ struct drm_i915_private *dev_priv;
+};
+
+struct intel_slice_kobject {
+ struct kobject kobj;
+ struct drm_i915_private *dev_priv;
+ u8 slice_index;
+};
+
+struct intel_subslice_kobject {
+ struct kobject kobj;
+ struct drm_i915_private *dev_priv;
+ u8 slice_index;
+ u8 subslice_index;
+};
+
struct drm_i915_private {
struct drm_device drm;
@@ -2732,6 +2750,14 @@ struct drm_i915_private {
struct {
struct kobject kobj;
struct kobject classes_kobjs[MAX_ENGINE_CLASS];
+
+ struct sysfs_slice {
+ struct intel_slice_kobject kobj;
+
+ struct sysfs_subslice {
+ struct intel_subslice_kobject kobj;
+ } subslices[GEN_MAX_SUBSLICES];
+ } slices[GEN_MAX_SLICES];
} gt_topology;
/* Abstract the submission mechanism (legacy ringbuffer or execlists) away */
diff --git a/drivers/gpu/drm/i915/i915_sysfs.c b/drivers/gpu/drm/i915/i915_sysfs.c
index fd04d0b93eaf..1ce56666bf19 100644
--- a/drivers/gpu/drm/i915/i915_sysfs.c
+++ b/drivers/gpu/drm/i915/i915_sysfs.c
@@ -559,6 +559,161 @@ static void i915_setup_error_capture(struct device *kdev) {}
static void i915_teardown_error_capture(struct device *kdev) {}
#endif
+static struct attribute mask_attr = {
+ .name = "enabled_mask",
+ .mode = 0444,
+};
+
+static struct attribute max_slices_attr = {
+ .name = "max_slices",
+ .mode = 0444,
+};
+
+static struct attribute max_subslices_per_slice_attr = {
+ .name = "max_subslices_per_slice",
+ .mode = 0444,
+};
+
+static struct attribute max_eus_per_subslice_attr = {
+ .name = "max_eus_per_subslice",
+ .mode = 0444,
+};
+
+static ssize_t
+show_slice_attr(struct kobject *kobj, struct attribute *attr, char *buf)
+{
+ struct intel_slice_kobject *kobj_wrapper =
+ container_of(kobj, struct intel_slice_kobject, kobj);
+ struct drm_i915_private *dev_priv = kobj_wrapper->dev_priv;
+ const struct sseu_dev_info *sseu = &INTEL_INFO(dev_priv)->sseu;
+
+ if (attr == &mask_attr) {
+ return sprintf(buf, "0x%hhx\n",
+ sseu->subslices_mask[kobj_wrapper->slice_index]);
+ }
+
+ return sprintf(buf, "0x0\n");
+}
+
+static const struct sysfs_ops slice_ops = {
+ .show = show_slice_attr,
+};
+
+static struct kobj_type slice_type = {
+ .sysfs_ops = &slice_ops,
+};
+
+static ssize_t
+show_subslice_attr(struct kobject *kobj, struct attribute *attr, char *buf)
+{
+ struct intel_subslice_kobject *kobj_wrapper =
+ container_of(kobj, struct intel_subslice_kobject, kobj);
+ struct drm_i915_private *dev_priv = kobj_wrapper->dev_priv;
+ const struct sseu_dev_info *sseu = &INTEL_INFO(dev_priv)->sseu;
+ int subslice_stride = ALIGN(sseu->max_eus_per_subslice, 8) / 8;
+ int slice_stride = sseu->max_subslices * subslice_stride;
+
+ if (attr == &mask_attr)
+ return sprintf(buf, "0x%hhx\n",
+ sseu->eu_mask[kobj_wrapper->slice_index * slice_stride +
+ kobj_wrapper->subslice_index * subslice_stride]);
+ return sprintf(buf, "0x0\n");
+}
+
+static const struct sysfs_ops subslice_ops = {
+ .show = show_subslice_attr,
+};
+
+static struct kobj_type subslice_type = {
+ .sysfs_ops = &subslice_ops,
+};
+
+static int i915_setup_rcs_topology_sysfs(struct drm_i915_private *dev_priv,
+ struct kobject *engine_kobj)
+{
+ const struct sseu_dev_info *sseu = &INTEL_INFO(dev_priv)->sseu;
+ int ret, s, ss;
+
+ ret = sysfs_create_file(engine_kobj, &mask_attr);
+ if (ret)
+ return ret;
+
+ ret = sysfs_create_file(engine_kobj, &max_slices_attr);
+ if (ret)
+ return ret;
+
+ ret = sysfs_create_file(engine_kobj, &max_subslices_per_slice_attr);
+ if (ret)
+ return ret;
+
+ ret = sysfs_create_file(engine_kobj, &max_eus_per_subslice_attr);
+ if (ret)
+ return ret;
+
+ for (s = 0; s < sseu->max_slices; s++) {
+ struct intel_slice_kobject *slice_kobj =
+ &dev_priv->gt_topology.slices[s].kobj;
+
+ slice_kobj->dev_priv = dev_priv;
+ slice_kobj->slice_index = s;
+ ret = kobject_init_and_add(&slice_kobj->kobj, &slice_type,
+ engine_kobj, "slice%i", s);
+ if (ret)
+ return ret;
+
+ ret = sysfs_create_file(&slice_kobj->kobj, &mask_attr);
+ if (ret)
+ return ret;
+
+ for (ss = 0; ss < sseu->max_subslices; ss++) {
+ struct intel_subslice_kobject *subslice_kobj =
+ &dev_priv->gt_topology.slices[s].subslices[ss].kobj;
+
+ subslice_kobj->dev_priv = dev_priv;
+ subslice_kobj->slice_index = s;
+ subslice_kobj->subslice_index = ss;
+ ret = kobject_init_and_add(&subslice_kobj->kobj,
+ &subslice_type,
+ &slice_kobj->kobj,
+ "subslice%i", ss);
+ if (ret)
+ return ret;
+
+ ret = sysfs_create_file(&subslice_kobj->kobj,
+ &mask_attr);
+ if (ret)
+ return ret;
+ }
+ }
+
+ return 0;
+}
+
+static void i915_teardown_rcs_topology_sysfs(struct drm_i915_private *dev_priv,
+ struct kobject *engine_kobj)
+{
+ const struct sseu_dev_info *sseu = &INTEL_INFO(dev_priv)->sseu;
+ int s, ss;
+
+ for (s = 0; s < sseu->max_slices; s++) {
+ struct intel_slice_kobject *slice_kobj =
+ &dev_priv->gt_topology.slices[s].kobj;
+
+ for (ss = 0; ss < sseu->max_subslices; ss++) {
+ struct intel_subslice_kobject *subslice_kobj =
+ &dev_priv->gt_topology.slices[s].subslices[ss].kobj;
+
+ sysfs_remove_file(&subslice_kobj->kobj, &mask_attr);
+ }
+
+ sysfs_remove_file(&slice_kobj->kobj, &mask_attr);
+ }
+ sysfs_remove_file(engine_kobj, &mask_attr);
+ sysfs_remove_file(engine_kobj, &max_eus_per_subslice_attr);
+ sysfs_remove_file(engine_kobj, &max_subslices_per_slice_attr);
+ sysfs_remove_file(engine_kobj, &max_slices_attr);
+}
+
static struct attribute engine_id_attr = {
.name = "id",
.mode = 0444,
@@ -574,11 +729,20 @@ show_engine_attr(struct kobject *kobj, struct attribute *attr, char *buf)
{
struct intel_engine_cs *engine =
container_of(kobj, struct intel_engine_cs, instance_kobj);
+ const struct sseu_dev_info *sseu = &INTEL_INFO(engine->i915)->sseu;
if (attr == &engine_id_attr)
return sprintf(buf, "%hhu\n", engine->uabi_id);
if (attr == &engine_class_attr)
return sprintf(buf, "%hhu\n", engine->uabi_class);
+ if (attr == &mask_attr)
+ return sprintf(buf, "0x%hhx\n", sseu->slice_mask);
+ if (attr == &max_eus_per_subslice_attr)
+ return sprintf(buf, "%hhd\n", sseu->max_eus_per_subslice);
+ if (attr == &max_subslices_per_slice_attr)
+ return sprintf(buf, "%hhd\n", sseu->max_subslices);
+ if (attr == &max_slices_attr)
+ return sprintf(buf, "%hhd\n", sseu->max_slices);
return sprintf(buf, "\n");
}
@@ -671,6 +835,12 @@ static int i915_setup_engines_sysfs(struct drm_i915_private *dev_priv,
if (ret)
return ret;
}
+ if (engine->id == RCS) {
+ ret = i915_setup_rcs_topology_sysfs(dev_priv,
+ &engine->instance_kobj);
+ if (ret)
+ return ret;
+ }
}
}
@@ -683,6 +853,11 @@ static void i915_teardown_engines_sysfs(struct drm_i915_private *dev_priv)
enum intel_engine_id id;
for_each_engine(engine, dev_priv, id) {
+ if (id == RCS) {
+ i915_teardown_rcs_topology_sysfs(dev_priv,
+ &engine->instance_kobj);
+ }
+
sysfs_remove_file(&engine->instance_kobj, &engine_id_attr);
sysfs_remove_file(&engine->instance_kobj, &engine_class_attr);
sysfs_remove_file(&engine->capabilities_kobj,
--
2.15.0
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
