On 18/04/2023 17:08, Rob Clark wrote:
On Tue, Apr 18, 2023 at 7:58 AM Tvrtko Ursulin
<tvrtko.ursulin@xxxxxxxxxxxxxxx> wrote:
On 18/04/2023 15:39, Rob Clark wrote:
On Mon, Apr 17, 2023 at 8:56 AM Tvrtko Ursulin
<tvrtko.ursulin@xxxxxxxxxxxxxxx> wrote:
From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
Show how more driver specific set of memory stats could be shown,
more specifically where object can reside in multiple regions, showing all
the supported stats, and where there is more to show than just user visible
objects.
WIP...
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
---
drivers/gpu/drm/i915/i915_driver.c | 5 ++
drivers/gpu/drm/i915/i915_drm_client.c | 102 +++++++++++++++++++++++++
drivers/gpu/drm/i915/i915_drm_client.h | 8 ++
drivers/gpu/drm/i915/i915_drv.h | 2 +
4 files changed, 117 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c
index 6493548c69bf..4c70206cbc27 100644
--- a/drivers/gpu/drm/i915/i915_driver.c
+++ b/drivers/gpu/drm/i915/i915_driver.c
@@ -1806,6 +1806,11 @@ static const struct drm_driver i915_drm_driver = {
.dumb_create = i915_gem_dumb_create,
.dumb_map_offset = i915_gem_dumb_mmap_offset,
+#ifdef CONFIG_PROC_FS
+ .query_fdinfo_memory_regions = i915_query_fdinfo_memory_regions,
+ .query_fdinfo_memory_stats = i915_query_fdinfo_memory_stats,
+#endif
+
.ioctls = i915_ioctls,
.num_ioctls = ARRAY_SIZE(i915_ioctls),
.fops = &i915_driver_fops,
diff --git a/drivers/gpu/drm/i915/i915_drm_client.c b/drivers/gpu/drm/i915/i915_drm_client.c
index c654984189f7..65857c68bdb3 100644
--- a/drivers/gpu/drm/i915/i915_drm_client.c
+++ b/drivers/gpu/drm/i915/i915_drm_client.c
@@ -12,6 +12,7 @@
#include <drm/drm_print.h>
#include "gem/i915_gem_context.h"
+#include "intel_memory_region.h"
#include "i915_drm_client.h"
#include "i915_file_private.h"
#include "i915_gem.h"
@@ -112,4 +113,105 @@ void i915_drm_client_fdinfo(struct drm_printer *p, struct drm_file *file)
for (i = 0; i < ARRAY_SIZE(uabi_class_names); i++)
show_client_class(p, i915, file_priv->client, i);
}
+
+char **
+i915_query_fdinfo_memory_regions(struct drm_device *dev, unsigned int *num)
+{
+ struct drm_i915_private *i915 = to_i915(dev);
+ struct intel_memory_region *mr;
+ enum intel_region_id id;
+
+ /* FIXME move to init */
+ for_each_memory_region(mr, i915, id) {
+ if (!i915->mm.region_names[id])
+ i915->mm.region_names[id] = mr->name;
+ }
+
+ *num = id;
+
+ return i915->mm.region_names;
+}
+
+static void
+add_obj(struct drm_i915_gem_object *obj, struct drm_fdinfo_memory_stat *stats)
+{
+ struct intel_memory_region *mr;
+ u64 sz = obj->base.size;
+ enum intel_region_id id;
+ unsigned int i;
+
+ if (!obj)
+ return;
+
+ /* Attribute size and shared to all possible memory regions. */
+ for (i = 0; i < obj->mm.n_placements; i++) {
+ mr = obj->mm.placements[i];
+ id = mr->id;
+
+ stats[id].size += sz;
This implies that summing up all of the categories is not the same as
the toplevel stats that I was proposing
Sorry, I mis-spoke, I meant "summing up all of the regions is not..."
Ah okay. It could be made like that yes.
I wasn't sure what would be more useful for drivers which support memory
regions. To see how much memory file could be using worst case, or
strictly how much it is currently using. So for buffer objects where
userspace allows kernel to choose the region from a supplied list, I
thought it would be useful to show that in total size against all
possible regions.
In a way you see this driver /could/ be using 1G in vram and 1G in
system, but currently it only has resident 1G in vram. Or you see
another file which has 1G vram size and 1G resident size and you can
infer some things.
Perhaps that can be confusing and it would be better to let total size
migrate between regions at runtime as does resident and other
categories. But then the total size per region would change at runtime
influenced by other app activity (as driver is transparently migrating
buffers between regions). Which can also be very confusing, it would
appear as if the app is creating/freeing objects when it isn't.
Correct, my categories are a bit different. You had private and shared as two mutually exclusive buckets, and then resident as subset of either/both. I have size as analogue to VmSize and resident as a subset of that, analogue to VmRss.
I split shared because by definition shared buffers can be counted
against multiple drm_file's, whereas private is only counted against
the single drm_file. Driver or app changes are unlikely to change the
shared size, whereas private footprint is a thing you can optimize to
some degree.
>
Shared is a bit wishy-washy, not sure about that one in either proposals. It can be either imported or exported buffers, but in essence I think it fits better as a subset of total size.
Imported vs exported doesn't really matter.. it is just an
implementation detail of the winsys. But I think it is useful to know
how much of an app's footprint is shared vs private. You could
express it different ways, but my proposal had private and shared,
from which you can calculate total:
total = private + shared
but you could flip the path around and advertise just total and
shared, and calculate private from that.
Yeah I am not sure. My gut feeling was that stable "top level" size is
the best option. Aka "this is how much this file could be using worst case".
If shared for file A can drop once file B closes the object it
previously imported from A, I think that could be confusing. Because A
did nothing - it is not suddenly using more private memory (hasn't
allocated anything) nor has closed any shared memory objects.
And on a tangent, but what about shared vs private stats when we have
userptr object created from shared memory? Core cannot really untangle
those. Or the memory allocated for other than buffer objects as I argue
in the cover letter.
Regards,
Tvrtko
