Re: [PATCH 2/2] drm/amdgpu: add shared fdinfo stats

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




On 09/01/2024 07:56, Christian König wrote:
Am 07.12.23 um 19:02 schrieb Alex Deucher:
Add shared stats.  Useful for seeing shared memory.

v2: take dma-buf into account as well

Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>
Cc: Rob Clark <robdclark@xxxxxxxxx>
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c |  4 ++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 +++++++++++
  drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  6 ++++++
  3 files changed, 21 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
index 5706b282a0c7..c7df7fa3459f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
@@ -97,6 +97,10 @@ void amdgpu_show_fdinfo(struct drm_printer *p, struct drm_file *file)
             stats.requested_visible_vram/1024UL);
      drm_printf(p, "amd-requested-gtt:\t%llu KiB\n",
             stats.requested_gtt/1024UL);
+    drm_printf(p, "drm-shared-vram:\t%llu KiB\n", stats.vram_shared/1024UL); +    drm_printf(p, "drm-shared-gtt:\t%llu KiB\n", stats.gtt_shared/1024UL); +    drm_printf(p, "drm-shared-cpu:\t%llu KiB\n", stats.cpu_shared/1024UL);
+
      for (hw_ip = 0; hw_ip < AMDGPU_HW_IP_NUM; ++hw_ip) {
          if (!usage[hw_ip])
              continue;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index d79b4ca1ecfc..1b37d95475b8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -1287,25 +1287,36 @@ void amdgpu_bo_get_memory(struct amdgpu_bo *bo,
                struct amdgpu_mem_stats *stats)
  {
      uint64_t size = amdgpu_bo_size(bo);
+    struct drm_gem_object *obj;
      unsigned int domain;
+    bool shared;
      /* Abort if the BO doesn't currently have a backing store */
      if (!bo->tbo.resource)
          return;
+    obj = &bo->tbo.base;
+    shared = (obj->handle_count > 1) || obj->dma_buf;

I still think that looking at handle_count is the completely wrong approach, we should really only look at obj->dma_buf.

Yeah it is all a bit tricky with the handle table walk. I don't think it is even possible to claim it is shared with obj->dma_buf could be the same process creating say via udmabuf and importing into drm. It is a wild scenario yes, but it could be private memory in that case. Not sure where it would leave us if we said this is just a limitation of a BO based tracking.

Would adding a new category "imported" help?

Hmm or we simply change drm-usage-stats.rst:

"""
- drm-shared-<region>: <uint> [KiB|MiB]

The total size of buffers that are shared with another file (ie. have more than than a single handle).
"""

Changing ie into eg coule be get our of jail free card to allow the "(obj->handle_count > 1) || obj->dma_buf;" condition?

Because of the shared with another _file_ wording would cover my wild udmabuf self-import case. Unless there are more such creative private import options.

Regards,

Tvrtko


Regards,
Christian.

+
      domain = amdgpu_mem_type_to_domain(bo->tbo.resource->mem_type);
      switch (domain) {
      case AMDGPU_GEM_DOMAIN_VRAM:
          stats->vram += size;
          if (amdgpu_bo_in_cpu_visible_vram(bo))
              stats->visible_vram += size;
+        if (shared)
+            stats->vram_shared += size;
          break;
      case AMDGPU_GEM_DOMAIN_GTT:
          stats->gtt += size;
+        if (shared)
+            stats->gtt_shared += size;
          break;
      case AMDGPU_GEM_DOMAIN_CPU:
      default:
          stats->cpu += size;
+        if (shared)
+            stats->cpu_shared += size;
          break;
      }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index d28e21baef16..0503af75dc26 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -138,12 +138,18 @@ struct amdgpu_bo_vm {
  struct amdgpu_mem_stats {
      /* current VRAM usage, includes visible VRAM */
      uint64_t vram;
+    /* current shared VRAM usage, includes visible VRAM */
+    uint64_t vram_shared;
      /* current visible VRAM usage */
      uint64_t visible_vram;
      /* current GTT usage */
      uint64_t gtt;
+    /* current shared GTT usage */
+    uint64_t gtt_shared;
      /* current system memory usage */
      uint64_t cpu;
+    /* current shared system memory usage */
+    uint64_t cpu_shared;
      /* sum of evicted buffers, includes visible VRAM */
      uint64_t evicted_vram;
      /* sum of evicted buffers due to CPU access */




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux