drm/xe: Attempt to bring bos back to VRAM after eviction

VRAM+TT bos that are evicted from VRAM to TT may remain in
TT also after a revalidation following eviction or suspend.

This manifests itself as applications becoming sluggish
after buffer objects get evicted or after a resume from
suspend or hibernation.

If the bo supports placement in both VRAM and TT, and
we are on DGFX, mark the TT placement as fallback. This means
that it is tried only after VRAM + eviction.

This flaw has probably been present since the xe module was
upstreamed but use a Fixes: commit below where backporting is
likely to be simple. For earlier versions we need to open-
code the fallback algorithm in the driver.

v2:
- Remove check for dgfx. (Matthew Auld)
- Update the xe_dma_buf kunit test for the new strategy (CI)
- Allow dma-buf to pin in current placement (CI)
- Make xe_bo_validate() for pinned bos a NOP.

Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/5995
Fixes: a78a8da51b36 ("drm/ttm: replace busy placement with flags v6")
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: <stable@vger.kernel.org> # v6.9+
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://lore.kernel.org/r/20250904160715.2613-2-thomas.hellstrom@linux.intel.com
(cherry picked from commit cb3d7b3b46b799c96b54f8e8fe36794a55a77f0b)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

diff --git a/drivers/gpu/drm/xe/tests/xe_bo.c b/drivers/gpu/drm/xe/tests/xe_bo.c
index bb469096d072b509ceb30a6ca5eff4ee7d41bb95..7b40cc8be1c9c2b91e42722ba7f97a4d07845c7a 100644 (file)
--- a/drivers/gpu/drm/xe/tests/xe_bo.c
+++ b/drivers/gpu/drm/xe/tests/xe_bo.c
@@ -236,7 +236,7 @@ static int evict_test_run_tile(struct xe_device *xe, struct xe_tile *tile, struc
                }
 
                xe_bo_lock(external, false);
-               err = xe_bo_pin_external(external);
+               err = xe_bo_pin_external(external, false);
                xe_bo_unlock(external);
                if (err) {
                        KUNIT_FAIL(test, "external bo pin err=%pe\n",

diff --git a/drivers/gpu/drm/xe/tests/xe_dma_buf.c b/drivers/gpu/drm/xe/tests/xe_dma_buf.c
index c53f67ce4b0aa26b9e728bcaa518a92229e3c304..121f17c112ec6a89dba30b4bd8348d0e63963d55 100644 (file)
--- a/drivers/gpu/drm/xe/tests/xe_dma_buf.c
+++ b/drivers/gpu/drm/xe/tests/xe_dma_buf.c
@@ -89,15 +89,7 @@ static void check_residency(struct kunit *test, struct xe_bo *exported,
                return;
        }
 
-       /*
-        * If on different devices, the exporter is kept in system  if
-        * possible, saving a migration step as the transfer is just
-        * likely as fast from system memory.
-        */
-       if (params->mem_mask & XE_BO_FLAG_SYSTEM)
-               KUNIT_EXPECT_TRUE(test, xe_bo_is_mem_type(exported, XE_PL_TT));
-       else
-               KUNIT_EXPECT_TRUE(test, xe_bo_is_mem_type(exported, mem_type));
+       KUNIT_EXPECT_TRUE(test, xe_bo_is_mem_type(exported, mem_type));
 
        if (params->force_different_devices)
                KUNIT_EXPECT_TRUE(test, xe_bo_is_mem_type(imported, XE_PL_TT));

diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 9954bb458ce12d90e59c63b3ab4cc60cab6b95aa..bae7ff2e59276c1d9810cb67632c70a8edb9dc57 100644 (file)
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -186,6 +186,8 @@ static void try_add_system(struct xe_device *xe, struct xe_bo *bo,
 
                bo->placements[*c] = (struct ttm_place) {
                        .mem_type = XE_PL_TT,
+                       .flags = (bo_flags & XE_BO_FLAG_VRAM_MASK) ?
+                       TTM_PL_FLAG_FALLBACK : 0,
                };
                *c += 1;
        }
@@ -2269,6 +2271,7 @@ uint64_t vram_region_gpu_offset(struct ttm_resource *res)
 /**
  * xe_bo_pin_external - pin an external BO
  * @bo: buffer object to be pinned
+ * @in_place: Pin in current placement, don't attempt to migrate.
  *
  * Pin an external (not tied to a VM, can be exported via dma-buf / prime FD)
  * BO. Unique call compared to xe_bo_pin as this function has it own set of
@@ -2276,7 +2279,7 @@ uint64_t vram_region_gpu_offset(struct ttm_resource *res)
  *
  * Returns 0 for success, negative error code otherwise.
  */
-int xe_bo_pin_external(struct xe_bo *bo)
+int xe_bo_pin_external(struct xe_bo *bo, bool in_place)
 {
        struct xe_device *xe = xe_bo_device(bo);
        int err;
@@ -2285,9 +2288,11 @@ int xe_bo_pin_external(struct xe_bo *bo)
        xe_assert(xe, xe_bo_is_user(bo));
 
        if (!xe_bo_is_pinned(bo)) {
-               err = xe_bo_validate(bo, NULL, false);
-               if (err)
-                       return err;
+               if (!in_place) {
+                       err = xe_bo_validate(bo, NULL, false);
+                       if (err)
+                               return err;
+               }
 
                spin_lock(&xe->pinned.lock);
                list_add_tail(&bo->pinned_link, &xe->pinned.late.external);
@@ -2440,6 +2445,9 @@ int xe_bo_validate(struct xe_bo *bo, struct xe_vm *vm, bool allow_res_evict)
        };
        int ret;
 
+       if (xe_bo_is_pinned(bo))
+               return 0;
+
        if (vm) {
                lockdep_assert_held(&vm->lock);
                xe_vm_assert_held(vm);

diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
index 02e8cde4c6b201dcdffed1cf2f9177fe8410a8ee..9ce94d25201562561145923c3d21b1f3ea870e07 100644 (file)
--- a/drivers/gpu/drm/xe/xe_bo.h
+++ b/drivers/gpu/drm/xe/xe_bo.h
@@ -198,7 +198,7 @@ static inline void xe_bo_unlock_vm_held(struct xe_bo *bo)
        }
 }
 
-int xe_bo_pin_external(struct xe_bo *bo);
+int xe_bo_pin_external(struct xe_bo *bo, bool in_place);
 int xe_bo_pin(struct xe_bo *bo);
 void xe_bo_unpin_external(struct xe_bo *bo);
 void xe_bo_unpin(struct xe_bo *bo);

diff --git a/drivers/gpu/drm/xe/xe_dma_buf.c b/drivers/gpu/drm/xe/xe_dma_buf.c
index 346f857f38374f95b337e0fcc0a5e0d603072713..af64baf872ef7b70461e3d5f7abab9bdbf74f70a 100644 (file)
--- a/drivers/gpu/drm/xe/xe_dma_buf.c
+++ b/drivers/gpu/drm/xe/xe_dma_buf.c
@@ -72,7 +72,7 @@ static int xe_dma_buf_pin(struct dma_buf_attachment *attach)
                return ret;
        }
 
-       ret = xe_bo_pin_external(bo);
+       ret = xe_bo_pin_external(bo, true);
        xe_assert(xe, !ret);
 
        return 0;