[PATCH v5 1/2] gpu/trace: make TRACE_GPU_MEM configurable

[PATCH v5 1/2] gpu/trace: make TRACE_GPU_MEM configurable

[Juston Li]

Move the source to a better place in Device Drivers -> Graphics support
now that its configurable.

v4:
 - Move source location (Tvrtko)

v3:
 - Patch introduced to replace per-driver config (Lucas)

Signed-off-by: Juston Li <justonli@chromium.org>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
---
 drivers/Kconfig           |  2 --
 drivers/gpu/trace/Kconfig | 11 ++++++++++-
 drivers/video/Kconfig     |  1 +
 3 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/drivers/Kconfig b/drivers/Kconfig
index 7c556c5ac4fdd..c5edbd2288a19 100644
--- a/drivers/Kconfig
+++ b/drivers/Kconfig
@@ -209,8 +209,6 @@ source "drivers/thunderbolt/Kconfig"
 
 source "drivers/android/Kconfig"
 
-source "drivers/gpu/trace/Kconfig"
-
 source "drivers/nvdimm/Kconfig"
 
 source "drivers/dax/Kconfig"
diff --git a/drivers/gpu/trace/Kconfig b/drivers/gpu/trace/Kconfig
index c24e9edd022e6..cd3d19c4a201c 100644
--- a/drivers/gpu/trace/Kconfig
+++ b/drivers/gpu/trace/Kconfig
@@ -1,4 +1,13 @@
 # SPDX-License-Identifier: GPL-2.0-only
 
 config TRACE_GPU_MEM
-	bool
+	bool "Enable GPU memory usage tracepoints"
+	default n
+	help
+	  Choose this option to enable tracepoints for tracking
+	  global and per-process GPU memory usage. Intended for
+	  performance profiling and required for Android.
+
+	  Tracepoint availability varies by GPU driver.
+
+	  If in doubt, say "N".
diff --git a/drivers/video/Kconfig b/drivers/video/Kconfig
index 5df981920a945..96adceab2df00 100644
--- a/drivers/video/Kconfig
+++ b/drivers/video/Kconfig
@@ -70,5 +70,6 @@ if FB_CORE || SGI_NEWPORT_CONSOLE
 
 endif
 
+source "drivers/gpu/trace/Kconfig"
 
 endmenu
-- 
2.50.0.727.gbf7dc18ff4-goog

[PATCH v5 2/2] drm/xe/bo: add GPU memory trace points

[Juston Li]

Add TRACE_GPU_MEM tracepoints for tracking global GPU memory usage.

These are required by VSR on Android 12+ for reporting GPU driver memory
allocations.

v5:
 - Drop process_mem tracking
 - Set the gpu_id field to dev->primary->index (Lucas, Tvrtko)
 - Formatting cleanup under 80 columns

v3:
 - Use now configurable CONFIG_TRACE_GPU_MEM instead of adding a
   per-driver Kconfig (Lucas)

v2:
 - Use u64 as preferred by checkpatch (Tvrtko)
 - Fix errors in comments/Kconfig description (Tvrtko)
 - drop redundant "CONFIG" in Kconfig

Signed-off-by: Juston Li <justonli@chromium.org>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
---
 drivers/gpu/drm/xe/xe_bo.c           | 22 ++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_device_types.h |  8 ++++++++
 2 files changed, 30 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 4e39188a021ab..950eef514c11c 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -19,6 +19,8 @@
 
 #include <kunit/static_stub.h>
 
+#include <trace/events/gpu_mem.h>
+
 #include "xe_device.h"
 #include "xe_dma_buf.h"
 #include "xe_drm_client.h"
@@ -418,6 +420,24 @@ static void xe_ttm_tt_account_subtract(struct xe_device *xe, struct ttm_tt *tt)
 		xe_shrinker_mod_pages(xe->mem.shrinker, -(long)tt->num_pages, 0);
 }
 
+#if IS_ENABLED(CONFIG_TRACE_GPU_MEM)
+static void update_global_total_pages(struct ttm_device *ttm_dev,
+				      long num_pages)
+{
+	struct xe_device *xe = ttm_to_xe_device(ttm_dev);
+	u64 global_total_pages =
+		atomic64_add_return(num_pages, &xe->global_total_pages);
+
+	trace_gpu_mem_total(xe->drm.primary->index, 0,
+			    global_total_pages << PAGE_SHIFT);
+}
+#else
+static inline void update_global_total_pages(struct ttm_device *ttm_dev,
+					     long num_pages)
+{
+}
+#endif
+
 static struct ttm_tt *xe_ttm_tt_create(struct ttm_buffer_object *ttm_bo,
 				       u32 page_flags)
 {
@@ -525,6 +545,7 @@ static int xe_ttm_tt_populate(struct ttm_device *ttm_dev, struct ttm_tt *tt,
 
 	xe_tt->purgeable = false;
 	xe_ttm_tt_account_add(ttm_to_xe_device(ttm_dev), tt);
+	update_global_total_pages(ttm_dev, tt->num_pages);
 
 	return 0;
 }
@@ -541,6 +562,7 @@ static void xe_ttm_tt_unpopulate(struct ttm_device *ttm_dev, struct ttm_tt *tt)
 
 	ttm_pool_free(&ttm_dev->pool, tt);
 	xe_ttm_tt_account_subtract(xe, tt);
+	update_global_total_pages(ttm_dev, -(long)tt->num_pages);
 }
 
 static void xe_ttm_tt_destroy(struct ttm_device *ttm_dev, struct ttm_tt *tt)
diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
index e5d02a47a5287..6f3698a0bc176 100644
--- a/drivers/gpu/drm/xe/xe_device_types.h
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -641,6 +641,14 @@ struct xe_device {
 		unsigned int fsb_freq, mem_freq, is_ddr3;
 	};
 #endif
+
+#if IS_ENABLED(CONFIG_TRACE_GPU_MEM)
+	/**
+	 * @global_total_pages: global GPU page usage tracked for gpu_mem
+	 * tracepoints
+	 */
+	atomic64_t global_total_pages;
+#endif
 };
 
 /**
-- 
2.50.0.727.gbf7dc18ff4-goog

Re: [PATCH v5 2/2] drm/xe/bo: add GPU memory trace points

[Lucas De Marchi]

On Mon, Jul 07, 2025 at 01:38:23PM -0700, Juston Li wrote:
>Add TRACE_GPU_MEM tracepoints for tracking global GPU memory usage.
>
>These are required by VSR on Android 12+ for reporting GPU driver memory
>allocations.
>
>v5:
> - Drop process_mem tracking
> - Set the gpu_id field to dev->primary->index (Lucas, Tvrtko)
> - Formatting cleanup under 80 columns
>
>v3:
> - Use now configurable CONFIG_TRACE_GPU_MEM instead of adding a
>   per-driver Kconfig (Lucas)
>
>v2:
> - Use u64 as preferred by checkpatch (Tvrtko)
> - Fix errors in comments/Kconfig description (Tvrtko)
> - drop redundant "CONFIG" in Kconfig
>
>Signed-off-by: Juston Li <justonli@chromium.org>
>Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>


Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>

Lucas De Marchi

Re: [PATCH v5 1/2] gpu/trace: make TRACE_GPU_MEM configurable

[Lucas De Marchi]

On Mon, Jul 07, 2025 at 01:38:22PM -0700, Juston Li wrote:
>Move the source to a better place in Device Drivers -> Graphics support
>now that its configurable.
>
>v4:
> - Move source location (Tvrtko)
>
>v3:
> - Patch introduced to replace per-driver config (Lucas)
>
>Signed-off-by: Juston Li <justonli@chromium.org>
>Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>

This first patch is now applied to drm-misc-next.

Thanks
Lucas De Marchi

Re: [PATCH v5 2/2] drm/xe/bo: add GPU memory trace points

[Lucas De Marchi]

On Mon, Jul 07, 2025 at 01:38:23PM -0700, Juston Li wrote:
>Add TRACE_GPU_MEM tracepoints for tracking global GPU memory usage.
>
>These are required by VSR on Android 12+ for reporting GPU driver memory
>allocations.
>
>v5:
> - Drop process_mem tracking
> - Set the gpu_id field to dev->primary->index (Lucas, Tvrtko)
> - Formatting cleanup under 80 columns
>
>v3:
> - Use now configurable CONFIG_TRACE_GPU_MEM instead of adding a
>   per-driver Kconfig (Lucas)
>
>v2:
> - Use u64 as preferred by checkpatch (Tvrtko)
> - Fix errors in comments/Kconfig description (Tvrtko)
> - drop redundant "CONFIG" in Kconfig
>
>Signed-off-by: Juston Li <justonli@chromium.org>
>Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
>---
> drivers/gpu/drm/xe/xe_bo.c           | 22 ++++++++++++++++++++++
> drivers/gpu/drm/xe/xe_device_types.h |  8 ++++++++
> 2 files changed, 30 insertions(+)
>
>diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
>index 4e39188a021ab..950eef514c11c 100644
>--- a/drivers/gpu/drm/xe/xe_bo.c
>+++ b/drivers/gpu/drm/xe/xe_bo.c
>@@ -19,6 +19,8 @@
>
> #include <kunit/static_stub.h>
>
>+#include <trace/events/gpu_mem.h>
>+
> #include "xe_device.h"
> #include "xe_dma_buf.h"
> #include "xe_drm_client.h"
>@@ -418,6 +420,24 @@ static void xe_ttm_tt_account_subtract(struct xe_device *xe, struct ttm_tt *tt)
> 		xe_shrinker_mod_pages(xe->mem.shrinker, -(long)tt->num_pages, 0);
> }
>
>+#if IS_ENABLED(CONFIG_TRACE_GPU_MEM)
>+static void update_global_total_pages(struct ttm_device *ttm_dev,
>+				      long num_pages)
>+{

wouldn't it be better to move the ifdef inside the function?
Any compiler would remove the empty call regardless of the inline
annotation, so I think this would make it less error prone if we change
function signature.

No need to send a new version, I can move that while applying
if agreed.

Lucas De Marchi

Re: [PATCH v5 2/2] drm/xe/bo: add GPU memory trace points

[Juston Li]

On Tue, 2025-07-08 at 15:29 -0500, Lucas De Marchi wrote:
> On Mon, Jul 07, 2025 at 01:38:23PM -0700, Juston Li wrote:
> > Add TRACE_GPU_MEM tracepoints for tracking global GPU memory usage.
> > 
> > These are required by VSR on Android 12+ for reporting GPU driver
> > memory
> > allocations.
> > 
> > v5:
> > - Drop process_mem tracking
> > - Set the gpu_id field to dev->primary->index (Lucas, Tvrtko)
> > - Formatting cleanup under 80 columns
> > 
> > v3:
> > - Use now configurable CONFIG_TRACE_GPU_MEM instead of adding a
> >   per-driver Kconfig (Lucas)
> > 
> > v2:
> > - Use u64 as preferred by checkpatch (Tvrtko)
> > - Fix errors in comments/Kconfig description (Tvrtko)
> > - drop redundant "CONFIG" in Kconfig
> > 
> > Signed-off-by: Juston Li <justonli@chromium.org>
> > Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
> > ---
> > drivers/gpu/drm/xe/xe_bo.c           | 22 ++++++++++++++++++++++
> > drivers/gpu/drm/xe/xe_device_types.h |  8 ++++++++
> > 2 files changed, 30 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c
> > b/drivers/gpu/drm/xe/xe_bo.c
> > index 4e39188a021ab..950eef514c11c 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > @@ -19,6 +19,8 @@
> > 
> > #include <kunit/static_stub.h>
> > 
> > +#include <trace/events/gpu_mem.h>
> > +
> > #include "xe_device.h"
> > #include "xe_dma_buf.h"
> > #include "xe_drm_client.h"
> > @@ -418,6 +420,24 @@ static void xe_ttm_tt_account_subtract(struct
> > xe_device *xe, struct ttm_tt *tt)
> > 		xe_shrinker_mod_pages(xe->mem.shrinker, -(long)tt-
> > >num_pages, 0);
> > }
> > 
> > +#if IS_ENABLED(CONFIG_TRACE_GPU_MEM)
> > +static void update_global_total_pages(struct ttm_device *ttm_dev,
> > +				      long num_pages)
> > +{
> 
> wouldn't it be better to move the ifdef inside the function?
> Any compiler would remove the empty call regardless of the inline
> annotation, so I think this would make it less error prone if we
> change
> function signature.
> 
> No need to send a new version, I can move that while applying
> if agreed.

SGTM, thanks!

Juston

> Lucas De Marchi