* [PATCH v3 0/4] drm/xe: More fb pinning optimizations.
@ 2024-04-24 13:02 Maarten Lankhorst
2024-04-24 13:02 ` [PATCH v3 1/4] drm/xe: Remove safety check from __xe_ttm_stolen_io_mem_reserve_stolen Maarten Lankhorst
` (4 more replies)
0 siblings, 5 replies; 6+ messages in thread
From: Maarten Lankhorst @ 2024-04-24 13:02 UTC (permalink / raw
To: intel-xe; +Cc: intel-gfx, Maarten Lankhorst
This reduces the latency of pinning framebuffers by
re-using the previous mapping, if available.
Additionally, DPT is preallocated when creating the FB, instead
of performing a bo allocation on every pin.
Remove the safety check in the first patch again, I didn't realize it was
needed even without the initial FB GGTT pinning removal.
Maarten Lankhorst (4):
drm/xe: Remove safety check from __xe_ttm_stolen_io_mem_reserve_stolen
drm/xe/display: Preparations for preallocating dpt bo
drm/xe: Use simple xchg to cache DPT
drm/xe/display: Re-use display vmas when possible
.../gpu/drm/i915/display/intel_atomic_plane.c | 2 +-
drivers/gpu/drm/i915/display/intel_cursor.c | 2 +-
drivers/gpu/drm/i915/display/intel_fb_pin.c | 3 +-
drivers/gpu/drm/i915/display/intel_fb_pin.h | 3 +-
drivers/gpu/drm/i915/display/intel_fbdev.c | 5 +
drivers/gpu/drm/i915/display/intel_fbdev.h | 9 +
.../gpu/drm/xe/compat-i915-headers/i915_vma.h | 3 +
drivers/gpu/drm/xe/display/xe_fb_pin.c | 220 ++++++++++++++----
drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c | 4 -
9 files changed, 198 insertions(+), 53 deletions(-)
--
2.43.0
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH v3 1/4] drm/xe: Remove safety check from __xe_ttm_stolen_io_mem_reserve_stolen
2024-04-24 13:02 [PATCH v3 0/4] drm/xe: More fb pinning optimizations Maarten Lankhorst
@ 2024-04-24 13:02 ` Maarten Lankhorst
2024-04-24 13:03 ` [PATCH v3 2/4] drm/xe/display: Preparations for preallocating dpt bo Maarten Lankhorst
` (3 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Maarten Lankhorst @ 2024-04-24 13:02 UTC (permalink / raw
To: intel-xe; +Cc: intel-gfx, Maarten Lankhorst
This is invalid with display code when reworking DPT pinning.
The only reason we added it, was because originally all display
allocations also had the bit set.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c | 4 ----
1 file changed, 4 deletions(-)
diff --git a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c
index f77367329760..1613290b9eda 100644
--- a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c
+++ b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c
@@ -298,10 +298,6 @@ static int __xe_ttm_stolen_io_mem_reserve_stolen(struct xe_device *xe,
XE_WARN_ON(IS_DGFX(xe));
- /* XXX: Require BO to be mapped to GGTT? */
- if (drm_WARN_ON(&xe->drm, !(bo->flags & XE_BO_FLAG_GGTT)))
- return -EIO;
-
/* GGTT is always contiguously mapped */
mem->bus.offset = xe_bo_ggtt_addr(bo) + mgr->io_base;
--
2.43.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH v3 2/4] drm/xe/display: Preparations for preallocating dpt bo
2024-04-24 13:02 [PATCH v3 0/4] drm/xe: More fb pinning optimizations Maarten Lankhorst
2024-04-24 13:02 ` [PATCH v3 1/4] drm/xe: Remove safety check from __xe_ttm_stolen_io_mem_reserve_stolen Maarten Lankhorst
@ 2024-04-24 13:03 ` Maarten Lankhorst
2024-04-24 13:03 ` [PATCH v3 3/4] drm/xe: Use simple xchg to cache DPT Maarten Lankhorst
` (2 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Maarten Lankhorst @ 2024-04-24 13:03 UTC (permalink / raw
To: intel-xe; +Cc: intel-gfx, Maarten Lankhorst
The DPT bo should not be allocated when pinning, but in advance when
creating the framebuffer. Split allocation from bo pinning and GGTT
insertion.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/xe/display/xe_fb_pin.c | 159 +++++++++++++++++++------
1 file changed, 123 insertions(+), 36 deletions(-)
diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c
index 3e1ae37c4c8b..5a8d6857fb89 100644
--- a/drivers/gpu/drm/xe/display/xe_fb_pin.c
+++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c
@@ -77,47 +77,130 @@ write_dpt_remapped(struct xe_bo *bo, struct iosys_map *map, u32 *dpt_ofs,
*dpt_ofs = ALIGN(*dpt_ofs, 4096);
}
-static int __xe_pin_fb_vma_dpt(struct intel_framebuffer *fb,
- const struct i915_gtt_view *view,
- struct i915_vma *vma)
+static struct xe_bo *xe_fb_dpt_alloc(struct intel_framebuffer *fb)
{
struct xe_device *xe = to_xe_device(fb->base.dev);
struct xe_tile *tile0 = xe_device_get_root_tile(xe);
- struct xe_ggtt *ggtt = tile0->mem.ggtt;
struct xe_bo *bo = intel_fb_obj(&fb->base), *dpt;
u32 dpt_size, size = bo->ttm.base.size;
- if (view->type == I915_GTT_VIEW_NORMAL)
+ if (!intel_fb_needs_pot_stride_remap(fb))
dpt_size = ALIGN(size / XE_PAGE_SIZE * 8, XE_PAGE_SIZE);
- else if (view->type == I915_GTT_VIEW_REMAPPED)
- dpt_size = ALIGN(intel_remapped_info_size(&fb->remapped_view.gtt.remapped) * 8,
- XE_PAGE_SIZE);
else
- /* display uses 4K tiles instead of bytes here, convert to entries.. */
- dpt_size = ALIGN(intel_rotation_info_size(&view->rotated) * 8,
+ dpt_size = ALIGN(intel_remapped_info_size(&fb->remapped_view.gtt.remapped) * 8,
XE_PAGE_SIZE);
if (IS_DGFX(xe))
- dpt = xe_bo_create_pin_map(xe, tile0, NULL, dpt_size,
- ttm_bo_type_kernel,
- XE_BO_FLAG_VRAM0 |
- XE_BO_FLAG_GGTT |
- XE_BO_FLAG_PAGETABLE);
- else
- dpt = xe_bo_create_pin_map(xe, tile0, NULL, dpt_size,
- ttm_bo_type_kernel,
- XE_BO_FLAG_STOLEN |
- XE_BO_FLAG_GGTT |
- XE_BO_FLAG_PAGETABLE);
+ return xe_bo_create(xe, tile0, NULL, dpt_size,
+ ttm_bo_type_kernel,
+ XE_BO_FLAG_NEEDS_CPU_ACCESS |
+ XE_BO_FLAG_VRAM0 |
+ XE_BO_FLAG_PAGETABLE);
+
+ dpt = xe_bo_create(xe, tile0, NULL, dpt_size,
+ ttm_bo_type_kernel,
+ XE_BO_FLAG_NEEDS_CPU_ACCESS |
+ XE_BO_FLAG_STOLEN |
+ XE_BO_FLAG_PAGETABLE);
if (IS_ERR(dpt))
- dpt = xe_bo_create_pin_map(xe, tile0, NULL, dpt_size,
- ttm_bo_type_kernel,
- XE_BO_FLAG_SYSTEM |
- XE_BO_FLAG_GGTT |
- XE_BO_FLAG_PAGETABLE);
+ dpt = xe_bo_create(xe, tile0, NULL, dpt_size,
+ ttm_bo_type_kernel,
+ XE_BO_FLAG_NEEDS_CPU_ACCESS |
+ XE_BO_FLAG_SYSTEM |
+ XE_BO_FLAG_PAGETABLE);
+
+ return dpt;
+}
+
+static void xe_fb_dpt_free(struct i915_vma *vma)
+{
+ xe_bo_put(vma->dpt);
+ vma->dpt = NULL;
+}
+
+static int xe_fb_dpt_map_ggtt(struct xe_bo *dpt)
+{
+ struct xe_device *xe = xe_bo_device(dpt);
+ struct xe_tile *tile0 = xe_device_get_root_tile(xe);
+ struct xe_ggtt *ggtt = tile0->mem.ggtt;
+ u64 start = 0, end = U64_MAX;
+ u64 alignment = XE_PAGE_SIZE;
+ int err;
+
+ if (dpt->flags & XE_BO_FLAG_INTERNAL_64K)
+ alignment = SZ_64K;
+
+ if (XE_WARN_ON(dpt->ggtt_node.size))
+ return -EINVAL;
+
+ xe_pm_runtime_get_noresume(xe);
+ err = mutex_lock_interruptible(&ggtt->lock);
+ if (err)
+ goto out_put;
+
+ err = drm_mm_insert_node_in_range(&ggtt->mm, &dpt->ggtt_node, dpt->size,
+ alignment, 0, start, end, 0);
+ if (!err)
+ xe_ggtt_map_bo(ggtt, dpt);
+ mutex_unlock(&ggtt->lock);
+
+out_put:
+ xe_pm_runtime_put(xe);
+ return err;
+}
+
+static int
+xe_fb_dpt_alloc_pinned(struct i915_vma *vma, struct intel_framebuffer *fb)
+{
+ struct xe_bo *dpt;
+ int err;
+
+ dpt = xe_fb_dpt_alloc(fb);
if (IS_ERR(dpt))
return PTR_ERR(dpt);
+ vma->dpt = dpt;
+
+ err = ttm_bo_reserve(&dpt->ttm, true, false, NULL);
+ if (!err) {
+ err = xe_bo_validate(dpt, NULL, true);
+ if (!err)
+ err = xe_bo_vmap(dpt);
+ if (!err)
+ ttm_bo_pin(&dpt->ttm);
+ ttm_bo_unreserve(&dpt->ttm);
+ }
+ if (err)
+ xe_fb_dpt_free(vma);
+ return err;
+}
+
+static void xe_fb_dpt_unpin_free(struct i915_vma *vma)
+{
+ ttm_bo_reserve(&vma->dpt->ttm, false, false, NULL);
+ ttm_bo_unpin(&vma->dpt->ttm);
+ ttm_bo_unreserve(&vma->dpt->ttm);
+
+ xe_fb_dpt_free(vma);
+}
+
+static int __xe_pin_fb_vma_dpt(struct intel_framebuffer *fb,
+ const struct i915_gtt_view *view,
+ struct i915_vma *vma)
+{
+ struct xe_device *xe = to_xe_device(fb->base.dev);
+ struct xe_tile *tile0 = xe_device_get_root_tile(xe);
+ struct xe_ggtt *ggtt = tile0->mem.ggtt;
+ struct xe_bo *bo = intel_fb_obj(&fb->base), *dpt;
+ u32 size = bo->ttm.base.size;
+ int ret;
+
+ ret = xe_fb_dpt_alloc_pinned(vma, fb);
+ if (ret)
+ return ret;
+ dpt = vma->dpt;
+
+ /* Create GGTT mapping.. */
if (view->type == I915_GTT_VIEW_NORMAL) {
u32 x;
@@ -152,9 +235,10 @@ static int __xe_pin_fb_vma_dpt(struct intel_framebuffer *fb,
rot_info->plane[i].dst_stride);
}
- vma->dpt = dpt;
- vma->node = dpt->ggtt_node;
- return 0;
+ ret = xe_fb_dpt_map_ggtt(dpt);
+ if (ret)
+ xe_fb_dpt_unpin_free(vma);
+ return ret;
}
static void
@@ -259,7 +343,7 @@ static struct i915_vma *__xe_pin_fb_vma(struct intel_framebuffer *fb,
int ret;
if (!vma)
- return ERR_PTR(-ENODEV);
+ return ERR_PTR(-ENOMEM);
if (IS_DGFX(to_xe_device(bo->ttm.base.dev)) &&
intel_fb_rc_ccs_cc_plane(&fb->base) >= 0 &&
@@ -282,7 +366,7 @@ static struct i915_vma *__xe_pin_fb_vma(struct intel_framebuffer *fb,
* Pin the framebuffer, we can't use xe_bo_(un)pin functions as the
* assumptions are incorrect for framebuffers
*/
- ret = ttm_bo_reserve(&bo->ttm, false, false, NULL);
+ ret = ttm_bo_reserve(&bo->ttm, true, false, NULL);
if (ret)
goto err;
@@ -320,11 +404,14 @@ static void __xe_unpin_fb_vma(struct i915_vma *vma)
struct xe_device *xe = to_xe_device(vma->bo->ttm.base.dev);
struct xe_ggtt *ggtt = xe_device_get_root_tile(xe)->mem.ggtt;
- if (vma->dpt)
- xe_bo_unpin_map_no_vm(vma->dpt);
- else if (!drm_mm_node_allocated(&vma->bo->ggtt_node) ||
- vma->bo->ggtt_node.start != vma->node.start)
- xe_ggtt_remove_node(ggtt, &vma->node, false);
+ if (vma->dpt) {
+ xe_ggtt_remove_bo(ggtt, vma->dpt);
+ xe_fb_dpt_unpin_free(vma);
+ } else {
+ if (!drm_mm_node_allocated(&vma->bo->ggtt_node) ||
+ vma->bo->ggtt_node.start != vma->node.start)
+ xe_ggtt_remove_node(ggtt, &vma->node, false);
+ }
ttm_bo_reserve(&vma->bo->ttm, false, false, NULL);
ttm_bo_unpin(&vma->bo->ttm);
--
2.43.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH v3 3/4] drm/xe: Use simple xchg to cache DPT
2024-04-24 13:02 [PATCH v3 0/4] drm/xe: More fb pinning optimizations Maarten Lankhorst
2024-04-24 13:02 ` [PATCH v3 1/4] drm/xe: Remove safety check from __xe_ttm_stolen_io_mem_reserve_stolen Maarten Lankhorst
2024-04-24 13:03 ` [PATCH v3 2/4] drm/xe/display: Preparations for preallocating dpt bo Maarten Lankhorst
@ 2024-04-24 13:03 ` Maarten Lankhorst
2024-04-24 13:03 ` [PATCH v3 4/4] drm/xe/display: Re-use display vmas when possible Maarten Lankhorst
2024-05-06 6:45 ` ✗ CI.Patch_applied: failure for drm/xe: More fb pinning optimizations. (rev2) Patchwork
4 siblings, 0 replies; 6+ messages in thread
From: Maarten Lankhorst @ 2024-04-24 13:03 UTC (permalink / raw
To: intel-xe; +Cc: intel-gfx, Maarten Lankhorst
Preallocate a DPT when creating the FB, and store it in
i915_address_space. This can be used to prevent an expensive
allocation in the pinning path.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/xe/display/xe_fb_pin.c | 33 +++++++++++++++-----------
1 file changed, 19 insertions(+), 14 deletions(-)
diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c
index 5a8d6857fb89..6ebda3ded8b4 100644
--- a/drivers/gpu/drm/xe/display/xe_fb_pin.c
+++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c
@@ -112,9 +112,11 @@ static struct xe_bo *xe_fb_dpt_alloc(struct intel_framebuffer *fb)
return dpt;
}
-static void xe_fb_dpt_free(struct i915_vma *vma)
+static void xe_fb_dpt_free(struct i915_vma *vma, struct intel_framebuffer *fb)
{
- xe_bo_put(vma->dpt);
+ if (!fb || cmpxchg((struct xe_bo **)&fb->dpt_vm, NULL, vma->dpt))
+ xe_bo_put(vma->dpt);
+
vma->dpt = NULL;
}
@@ -152,10 +154,11 @@ static int xe_fb_dpt_map_ggtt(struct xe_bo *dpt)
static int
xe_fb_dpt_alloc_pinned(struct i915_vma *vma, struct intel_framebuffer *fb)
{
- struct xe_bo *dpt;
+ struct xe_bo *dpt = (struct xe_bo *)xchg(&fb->dpt_vm, NULL);
int err;
- dpt = xe_fb_dpt_alloc(fb);
+ if (!dpt)
+ dpt = xe_fb_dpt_alloc(fb);
if (IS_ERR(dpt))
return PTR_ERR(dpt);
@@ -171,17 +174,17 @@ xe_fb_dpt_alloc_pinned(struct i915_vma *vma, struct intel_framebuffer *fb)
ttm_bo_unreserve(&dpt->ttm);
}
if (err)
- xe_fb_dpt_free(vma);
+ xe_fb_dpt_free(vma, fb);
return err;
}
-static void xe_fb_dpt_unpin_free(struct i915_vma *vma)
+static void xe_fb_dpt_unpin_free(struct i915_vma *vma, struct intel_framebuffer *fb)
{
ttm_bo_reserve(&vma->dpt->ttm, false, false, NULL);
ttm_bo_unpin(&vma->dpt->ttm);
ttm_bo_unreserve(&vma->dpt->ttm);
- xe_fb_dpt_free(vma);
+ xe_fb_dpt_free(vma, fb);
}
static int __xe_pin_fb_vma_dpt(struct intel_framebuffer *fb,
@@ -237,7 +240,7 @@ static int __xe_pin_fb_vma_dpt(struct intel_framebuffer *fb,
ret = xe_fb_dpt_map_ggtt(dpt);
if (ret)
- xe_fb_dpt_unpin_free(vma);
+ xe_fb_dpt_unpin_free(vma, fb);
return ret;
}
@@ -399,14 +402,14 @@ static struct i915_vma *__xe_pin_fb_vma(struct intel_framebuffer *fb,
return ERR_PTR(ret);
}
-static void __xe_unpin_fb_vma(struct i915_vma *vma)
+static void __xe_unpin_fb_vma(struct i915_vma *vma, struct intel_framebuffer *fb)
{
struct xe_device *xe = to_xe_device(vma->bo->ttm.base.dev);
struct xe_ggtt *ggtt = xe_device_get_root_tile(xe)->mem.ggtt;
if (vma->dpt) {
xe_ggtt_remove_bo(ggtt, vma->dpt);
- xe_fb_dpt_unpin_free(vma);
+ xe_fb_dpt_unpin_free(vma, fb);
} else {
if (!drm_mm_node_allocated(&vma->bo->ggtt_node) ||
vma->bo->ggtt_node.start != vma->node.start)
@@ -433,7 +436,7 @@ intel_pin_and_fence_fb_obj(struct drm_framebuffer *fb,
void intel_unpin_fb_vma(struct i915_vma *vma, unsigned long flags)
{
- __xe_unpin_fb_vma(vma);
+ __xe_unpin_fb_vma(vma, NULL);
}
int intel_plane_pin_fb(struct intel_plane_state *plane_state)
@@ -455,7 +458,7 @@ int intel_plane_pin_fb(struct intel_plane_state *plane_state)
void intel_plane_unpin_fb(struct intel_plane_state *old_plane_state)
{
- __xe_unpin_fb_vma(old_plane_state->ggtt_vma);
+ __xe_unpin_fb_vma(old_plane_state->ggtt_vma, to_intel_framebuffer(old_plane_state->hw.fb));
old_plane_state->ggtt_vma = NULL;
}
@@ -465,10 +468,12 @@ void intel_plane_unpin_fb(struct intel_plane_state *old_plane_state)
*/
struct i915_address_space *intel_dpt_create(struct intel_framebuffer *fb)
{
- return NULL;
+ return (struct i915_address_space *)xe_fb_dpt_alloc(fb);
}
void intel_dpt_destroy(struct i915_address_space *vm)
{
- return;
+ struct xe_bo *bo = (struct xe_bo *)vm;
+
+ xe_bo_put(bo);
}
--
2.43.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH v3 4/4] drm/xe/display: Re-use display vmas when possible
2024-04-24 13:02 [PATCH v3 0/4] drm/xe: More fb pinning optimizations Maarten Lankhorst
` (2 preceding siblings ...)
2024-04-24 13:03 ` [PATCH v3 3/4] drm/xe: Use simple xchg to cache DPT Maarten Lankhorst
@ 2024-04-24 13:03 ` Maarten Lankhorst
2024-05-06 6:45 ` ✗ CI.Patch_applied: failure for drm/xe: More fb pinning optimizations. (rev2) Patchwork
4 siblings, 0 replies; 6+ messages in thread
From: Maarten Lankhorst @ 2024-04-24 13:03 UTC (permalink / raw
To: intel-xe; +Cc: intel-gfx, Maarten Lankhorst
i915 has this really nice, infrastructure where everything becomes
complicated, GGTT needs eviction, etc..
Lets not do that, and make the dumbest possible interface instead.
Try to retrieve the VMA from old_plane_state, or intel_fbdev if kernel
fb.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
.../gpu/drm/i915/display/intel_atomic_plane.c | 2 +-
drivers/gpu/drm/i915/display/intel_cursor.c | 2 +-
drivers/gpu/drm/i915/display/intel_fb_pin.c | 3 +-
drivers/gpu/drm/i915/display/intel_fb_pin.h | 3 +-
drivers/gpu/drm/i915/display/intel_fbdev.c | 5 ++
drivers/gpu/drm/i915/display/intel_fbdev.h | 9 ++++
| 3 ++
drivers/gpu/drm/xe/display/xe_fb_pin.c | 46 +++++++++++++++++--
8 files changed, 65 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/i915/display/intel_atomic_plane.c b/drivers/gpu/drm/i915/display/intel_atomic_plane.c
index 76d77d5a0409..24ba55531e8d 100644
--- a/drivers/gpu/drm/i915/display/intel_atomic_plane.c
+++ b/drivers/gpu/drm/i915/display/intel_atomic_plane.c
@@ -1106,7 +1106,7 @@ intel_prepare_plane_fb(struct drm_plane *_plane,
if (!obj)
return 0;
- ret = intel_plane_pin_fb(new_plane_state);
+ ret = intel_plane_pin_fb(new_plane_state, old_plane_state);
if (ret)
return ret;
diff --git a/drivers/gpu/drm/i915/display/intel_cursor.c b/drivers/gpu/drm/i915/display/intel_cursor.c
index 23a122ee20c9..3d3b8e37c0f2 100644
--- a/drivers/gpu/drm/i915/display/intel_cursor.c
+++ b/drivers/gpu/drm/i915/display/intel_cursor.c
@@ -761,7 +761,7 @@ intel_legacy_cursor_update(struct drm_plane *_plane,
if (ret)
goto out_free;
- ret = intel_plane_pin_fb(new_plane_state);
+ ret = intel_plane_pin_fb(new_plane_state, old_plane_state);
if (ret)
goto out_free;
diff --git a/drivers/gpu/drm/i915/display/intel_fb_pin.c b/drivers/gpu/drm/i915/display/intel_fb_pin.c
index b6df9baf481b..7b8a1825ccfc 100644
--- a/drivers/gpu/drm/i915/display/intel_fb_pin.c
+++ b/drivers/gpu/drm/i915/display/intel_fb_pin.c
@@ -236,7 +236,8 @@ void intel_unpin_fb_vma(struct i915_vma *vma, unsigned long flags)
i915_vma_put(vma);
}
-int intel_plane_pin_fb(struct intel_plane_state *plane_state)
+int intel_plane_pin_fb(struct intel_plane_state *plane_state,
+ const struct intel_plane_state *old_plane_state)
{
struct intel_plane *plane = to_intel_plane(plane_state->uapi.plane);
struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
diff --git a/drivers/gpu/drm/i915/display/intel_fb_pin.h b/drivers/gpu/drm/i915/display/intel_fb_pin.h
index de0efaa25905..48675e6233f0 100644
--- a/drivers/gpu/drm/i915/display/intel_fb_pin.h
+++ b/drivers/gpu/drm/i915/display/intel_fb_pin.h
@@ -22,7 +22,8 @@ intel_pin_and_fence_fb_obj(struct drm_framebuffer *fb,
void intel_unpin_fb_vma(struct i915_vma *vma, unsigned long flags);
-int intel_plane_pin_fb(struct intel_plane_state *plane_state);
+int intel_plane_pin_fb(struct intel_plane_state *new_plane_state,
+ const struct intel_plane_state *old_plane_state);
void intel_plane_unpin_fb(struct intel_plane_state *old_plane_state);
#endif
diff --git a/drivers/gpu/drm/i915/display/intel_fbdev.c b/drivers/gpu/drm/i915/display/intel_fbdev.c
index 43855c6c3509..a010b7a8a468 100644
--- a/drivers/gpu/drm/i915/display/intel_fbdev.c
+++ b/drivers/gpu/drm/i915/display/intel_fbdev.c
@@ -668,3 +668,8 @@ struct intel_framebuffer *intel_fbdev_framebuffer(struct intel_fbdev *fbdev)
return to_intel_framebuffer(fbdev->helper.fb);
}
+
+struct i915_vma *intel_fbdev_vma_pointer(struct intel_fbdev *fbdev)
+{
+ return fbdev ? fbdev->vma : NULL;
+}
diff --git a/drivers/gpu/drm/i915/display/intel_fbdev.h b/drivers/gpu/drm/i915/display/intel_fbdev.h
index 04fd523a5023..fa6c0c1ae936 100644
--- a/drivers/gpu/drm/i915/display/intel_fbdev.h
+++ b/drivers/gpu/drm/i915/display/intel_fbdev.h
@@ -22,6 +22,8 @@ void intel_fbdev_set_suspend(struct drm_device *dev, int state, bool synchronous
void intel_fbdev_output_poll_changed(struct drm_device *dev);
void intel_fbdev_restore_mode(struct drm_i915_private *dev_priv);
struct intel_framebuffer *intel_fbdev_framebuffer(struct intel_fbdev *fbdev);
+struct i915_vma *intel_fbdev_vma_pointer(struct intel_fbdev *fbdev);
+
#else
static inline int intel_fbdev_init(struct drm_device *dev)
{
@@ -51,10 +53,17 @@ static inline void intel_fbdev_output_poll_changed(struct drm_device *dev)
static inline void intel_fbdev_restore_mode(struct drm_i915_private *i915)
{
}
+
static inline struct intel_framebuffer *intel_fbdev_framebuffer(struct intel_fbdev *fbdev)
{
return NULL;
}
+
+static inline struct i915_vma *intel_fbdev_vma_pointer(struct intel_fbdev *fbdev)
+{
+ return NULL;
+}
+
#endif
#endif /* __INTEL_FBDEV_H__ */
--git a/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h
index a20d2638ea7a..193382f97823 100644
--- a/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h
@@ -9,6 +9,8 @@
#include <uapi/drm/i915_drm.h>
#include <drm/drm_mm.h>
+#include <linux/refcount.h>
+
/* We don't want these from i915_drm.h in case of Xe */
#undef I915_TILING_X
#undef I915_TILING_Y
@@ -18,6 +20,7 @@
struct xe_bo;
struct i915_vma {
+ refcount_t ref;
struct xe_bo *bo, *dpt;
struct drm_mm_node node;
};
diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c
index 6ebda3ded8b4..75e030f0fb70 100644
--- a/drivers/gpu/drm/xe/display/xe_fb_pin.c
+++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c
@@ -8,6 +8,7 @@
#include "intel_dpt.h"
#include "intel_fb.h"
#include "intel_fb_pin.h"
+#include "intel_fbdev.h"
#include "xe_ggtt.h"
#include "xe_gt.h"
#include "xe_pm.h"
@@ -348,6 +349,7 @@ static struct i915_vma *__xe_pin_fb_vma(struct intel_framebuffer *fb,
if (!vma)
return ERR_PTR(-ENOMEM);
+ refcount_set(&vma->ref, 1);
if (IS_DGFX(to_xe_device(bo->ttm.base.dev)) &&
intel_fb_rc_ccs_cc_plane(&fb->base) >= 0 &&
!(bo->flags & XE_BO_FLAG_NEEDS_CPU_ACCESS)) {
@@ -407,6 +409,9 @@ static void __xe_unpin_fb_vma(struct i915_vma *vma, struct intel_framebuffer *fb
struct xe_device *xe = to_xe_device(vma->bo->ttm.base.dev);
struct xe_ggtt *ggtt = xe_device_get_root_tile(xe)->mem.ggtt;
+ if (!refcount_dec_and_test(&vma->ref))
+ return;
+
if (vma->dpt) {
xe_ggtt_remove_bo(ggtt, vma->dpt);
xe_fb_dpt_unpin_free(vma, fb);
@@ -439,20 +444,53 @@ void intel_unpin_fb_vma(struct i915_vma *vma, unsigned long flags)
__xe_unpin_fb_vma(vma, NULL);
}
-int intel_plane_pin_fb(struct intel_plane_state *plane_state)
+static bool reuse_vma(struct intel_plane_state *new_plane_state,
+ const struct intel_plane_state *old_plane_state)
{
- struct drm_framebuffer *fb = plane_state->hw.fb;
+ struct intel_framebuffer *fb = to_intel_framebuffer(new_plane_state->hw.fb);
+ struct xe_device *xe = to_xe_device(fb->base.dev);
+ struct i915_vma *vma;
+
+ if (old_plane_state->hw.fb == new_plane_state->hw.fb &&
+ !memcmp(&old_plane_state->view.gtt,
+ &new_plane_state->view.gtt,
+ sizeof(new_plane_state->view.gtt))) {
+ vma = old_plane_state->ggtt_vma;
+ goto found;
+ }
+
+ if (fb == intel_fbdev_framebuffer(xe->display.fbdev.fbdev)) {
+ vma = intel_fbdev_vma_pointer(xe->display.fbdev.fbdev);
+ if (vma)
+ goto found;
+ }
+
+ return false;
+
+found:
+ refcount_inc(&vma->ref);
+ new_plane_state->ggtt_vma = vma;
+ return true;
+}
+
+int intel_plane_pin_fb(struct intel_plane_state *new_plane_state,
+ const struct intel_plane_state *old_plane_state)
+{
+ struct drm_framebuffer *fb = new_plane_state->hw.fb;
struct xe_bo *bo = intel_fb_obj(fb);
struct i915_vma *vma;
+ if (reuse_vma(new_plane_state, old_plane_state))
+ return 0;
+
/* We reject creating !SCANOUT fb's, so this is weird.. */
drm_WARN_ON(bo->ttm.base.dev, !(bo->flags & XE_BO_FLAG_SCANOUT));
- vma = __xe_pin_fb_vma(to_intel_framebuffer(fb), &plane_state->view.gtt);
+ vma = __xe_pin_fb_vma(to_intel_framebuffer(fb), &new_plane_state->view.gtt);
if (IS_ERR(vma))
return PTR_ERR(vma);
- plane_state->ggtt_vma = vma;
+ new_plane_state->ggtt_vma = vma;
return 0;
}
--
2.43.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* ✗ CI.Patch_applied: failure for drm/xe: More fb pinning optimizations. (rev2)
2024-04-24 13:02 [PATCH v3 0/4] drm/xe: More fb pinning optimizations Maarten Lankhorst
` (3 preceding siblings ...)
2024-04-24 13:03 ` [PATCH v3 4/4] drm/xe/display: Re-use display vmas when possible Maarten Lankhorst
@ 2024-05-06 6:45 ` Patchwork
4 siblings, 0 replies; 6+ messages in thread
From: Patchwork @ 2024-05-06 6:45 UTC (permalink / raw
To: Maarten Lankhorst; +Cc: intel-xe
== Series Details ==
Series: drm/xe: More fb pinning optimizations. (rev2)
URL : https://patchwork.freedesktop.org/series/132847/
State : failure
== Summary ==
=== Applying kernel patches on branch 'drm-tip' with base: ===
Base commit: ca96a1c62451 drm-tip: 2024y-05m-03d-21h-18m-52s UTC integration manifest
=== git am output follows ===
error: patch failed: drivers/gpu/drm/i915/display/intel_fbdev.h:22
error: drivers/gpu/drm/i915/display/intel_fbdev.h: patch does not apply
hint: Use 'git am --show-current-patch=diff' to see the failed patch
Applying: drm/xe: Remove safety check from __xe_ttm_stolen_io_mem_reserve_stolen
Applying: drm/xe/display: Preparations for preallocating dpt bo
Applying: drm/xe: Use simple xchg to cache DPT
Applying: drm/xe/display: Re-use display vmas when possible
Patch failed at 0004 drm/xe/display: Re-use display vmas when possible
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2024-05-06 6:45 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2024-04-24 13:02 [PATCH v3 0/4] drm/xe: More fb pinning optimizations Maarten Lankhorst
2024-04-24 13:02 ` [PATCH v3 1/4] drm/xe: Remove safety check from __xe_ttm_stolen_io_mem_reserve_stolen Maarten Lankhorst
2024-04-24 13:03 ` [PATCH v3 2/4] drm/xe/display: Preparations for preallocating dpt bo Maarten Lankhorst
2024-04-24 13:03 ` [PATCH v3 3/4] drm/xe: Use simple xchg to cache DPT Maarten Lankhorst
2024-04-24 13:03 ` [PATCH v3 4/4] drm/xe/display: Re-use display vmas when possible Maarten Lankhorst
2024-05-06 6:45 ` ✗ CI.Patch_applied: failure for drm/xe: More fb pinning optimizations. (rev2) Patchwork
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).