* VRAM manager
@ 2016-09-27 9:49 Christian König
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
0 siblings, 1 reply; 18+ messages in thread
From: Christian König @ 2016-09-27 9:49 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Hi guys,
after fixing all those nasty little bugs this seems to be stable now.
Anybody brave enough to give it an review?
Cheers,
Christian.
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* [PATCH 1/6] drm/amdgpu: add AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS flag v3
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
@ 2016-09-27 9:49 ` Christian König
[not found] ` <1474969797-1882-2-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-09-27 9:49 ` [PATCH 2/6] drm/amdgpu: use explicit limit for VRAM_CONTIGUOUS Christian König
` (7 subsequent siblings)
8 siblings, 1 reply; 18+ messages in thread
From: Christian König @ 2016-09-27 9:49 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Christian König <christian.koenig@amd.com>
Add a flag noting that a BO must be created using linear VRAM
and set this flag on all in kernel users where appropriate.
Hopefully I haven't missed anything.
v2: add it in a few more places, fix CPU mapping.
v3: rename to VRAM_CONTIGUOUS, fix typo in CS code.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c | 6 ++++--
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 9 +++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 3 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c | 3 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 3 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 12 ++++++++++--
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 3 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c | 6 ++++--
drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 3 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 6 ++++--
drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 9 ++++++---
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 6 ++++--
include/uapi/drm/amdgpu_drm.h | 2 ++
13 files changed, 53 insertions(+), 18 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c
index 7a8bfa3..a61f418 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c
@@ -146,7 +146,8 @@ static int amdgpu_cgs_alloc_gpu_mem(struct cgs_device *cgs_device,
switch(type) {
case CGS_GPU_MEM_TYPE__VISIBLE_CONTIG_FB:
case CGS_GPU_MEM_TYPE__VISIBLE_FB:
- flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+ flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
domain = AMDGPU_GEM_DOMAIN_VRAM;
if (max_offset > adev->mc.real_vram_size)
return -EINVAL;
@@ -157,7 +158,8 @@ static int amdgpu_cgs_alloc_gpu_mem(struct cgs_device *cgs_device,
break;
case CGS_GPU_MEM_TYPE__INVISIBLE_CONTIG_FB:
case CGS_GPU_MEM_TYPE__INVISIBLE_FB:
- flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
+ flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
domain = AMDGPU_GEM_DOMAIN_VRAM;
if (adev->mc.visible_vram_size < adev->mc.real_vram_size) {
place.fpfn =
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index b0f6e69..187c366 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -1195,6 +1195,15 @@ int amdgpu_cs_sysvm_access_required(struct amdgpu_cs_parser *parser)
r = amdgpu_ttm_bind(&bo->tbo, &bo->tbo.mem);
if (unlikely(r))
return r;
+
+ if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
+ continue;
+
+ bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+ amdgpu_ttm_placement_from_domain(bo, bo->allowed_domains);
+ r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
+ if (unlikely(r))
+ return r;
}
return 0;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index c626434..5ca6a38 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -265,7 +265,8 @@ static int amdgpu_vram_scratch_init(struct amdgpu_device *adev)
if (adev->vram_scratch.robj == NULL) {
r = amdgpu_bo_create(adev, AMDGPU_GPU_PAGE_SIZE,
PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL, &adev->vram_scratch.robj);
if (r) {
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c
index 034e945..f491092 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c
@@ -148,7 +148,8 @@ static int amdgpufb_create_pinned_object(struct amdgpu_fbdev *rfbdev,
aligned_size = ALIGN(size, PAGE_SIZE);
ret = amdgpu_gem_object_create(adev, aligned_size, 0,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
true, &gobj);
if (ret) {
printk(KERN_ERR "failed to allocate framebuffer (%d)\n",
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index 21a1242..964d2a9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -126,7 +126,8 @@ int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
if (adev->gart.robj == NULL) {
r = amdgpu_bo_create(adev, adev->gart.table_size,
PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL, &adev->gart.robj);
if (r) {
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index aa074fa..ba41807 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -245,7 +245,8 @@ int amdgpu_bo_create_kernel(struct amdgpu_device *adev,
int r;
r = amdgpu_bo_create(adev, size, align, true, domain,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL, bo_ptr);
if (r) {
dev_err(adev->dev, "(%d) failed to allocate kernel bo\n", r);
@@ -643,6 +644,8 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
return 0;
}
+
+ bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
amdgpu_ttm_placement_from_domain(bo, domain);
for (i = 0; i < bo->placement.num_placement; i++) {
/* force to pin into visible video ram */
@@ -885,7 +888,9 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
size = bo->mem.num_pages << PAGE_SHIFT;
offset = bo->mem.start << PAGE_SHIFT;
- if ((offset + size) <= adev->mc.visible_vram_size)
+ /* TODO: figure out how to map scattered VRAM to the CPU */
+ if ((offset + size) <= adev->mc.visible_vram_size &&
+ (abo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS))
return 0;
/* Can't move a pinned BO to visible VRAM */
@@ -893,6 +898,7 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
return -EINVAL;
/* hurrah the memory is not visible ! */
+ abo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM);
lpfn = adev->mc.visible_vram_size >> PAGE_SHIFT;
for (i = 0; i < abo->placement.num_placement; i++) {
@@ -954,6 +960,8 @@ u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
WARN_ON_ONCE(!ww_mutex_is_locked(&bo->tbo.resv->lock) &&
!bo->pin_count);
WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
+ WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_VRAM &&
+ !(bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS));
return bo->tbo.offset;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index ce9da20..06ab624 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1116,7 +1116,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
r = amdgpu_bo_create(adev, 256 * 1024, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL, &adev->stollen_vga_memory);
if (r) {
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index e3281ca..4cf3ca7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -1002,7 +1002,8 @@ int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
r = amdgpu_bo_create(adev, 1024, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL, &bo);
if (r)
return r;
@@ -1051,7 +1052,8 @@ int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
r = amdgpu_bo_create(adev, 1024, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL, &bo);
if (r)
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 06b94c1..217cbf2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -157,7 +157,8 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
r = amdgpu_bo_create(adev, size, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL, &adev->vce.vcpu_bo);
if (r) {
dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index bc4b22c..2379eb1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1420,7 +1420,8 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
AMDGPU_GPU_PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
- AMDGPU_GEM_CREATE_SHADOW,
+ AMDGPU_GEM_CREATE_SHADOW |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, resv, &pt);
if (r)
goto error_free;
@@ -1630,7 +1631,8 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
r = amdgpu_bo_create(adev, pd_size, align, true,
AMDGPU_GEM_DOMAIN_VRAM,
AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
- AMDGPU_GEM_CREATE_SHADOW,
+ AMDGPU_GEM_CREATE_SHADOW |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL, &vm->page_directory);
if (r)
goto error_free_sched_entity;
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index 32a6762..2bdcdd0 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -3385,7 +3385,8 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
if (adev->gfx.rlc.save_restore_obj == NULL) {
r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL,
&adev->gfx.rlc.save_restore_obj);
if (r) {
@@ -3429,7 +3430,8 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
if (adev->gfx.rlc.clear_state_obj == NULL) {
r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL,
&adev->gfx.rlc.clear_state_obj);
if (r) {
@@ -3469,7 +3471,8 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
if (adev->gfx.rlc.cp_table_obj == NULL) {
r = amdgpu_bo_create(adev, adev->gfx.rlc.cp_table_size, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL,
&adev->gfx.rlc.cp_table_obj);
if (r) {
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 324764b..140997f 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -1274,7 +1274,8 @@ static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
if (adev->gfx.rlc.clear_state_obj == NULL) {
r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL,
&adev->gfx.rlc.clear_state_obj);
if (r) {
@@ -1316,7 +1317,8 @@ static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
if (adev->gfx.rlc.cp_table_obj == NULL) {
r = amdgpu_bo_create(adev, adev->gfx.rlc.cp_table_size, PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+ AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+ AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
NULL, NULL,
&adev->gfx.rlc.cp_table_obj);
if (r) {
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index c27ccbe..316b111 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -83,6 +83,8 @@ extern "C" {
#define AMDGPU_GEM_CREATE_VRAM_CLEARED (1 << 3)
/* Flag that create shadow bo(GTT) while allocating vram bo */
#define AMDGPU_GEM_CREATE_SHADOW (1 << 4)
+/* Flag that allocating the BO should use linear VRAM */
+#define AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS (1 << 5)
struct drm_amdgpu_gem_create_in {
/** the requested memory size */
--
2.5.0
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 18+ messages in thread
* [PATCH 2/6] drm/amdgpu: use explicit limit for VRAM_CONTIGUOUS
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-09-27 9:49 ` [PATCH 1/6] drm/amdgpu: add AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS flag v3 Christian König
@ 2016-09-27 9:49 ` Christian König
2016-09-27 9:49 ` [PATCH 3/6] drm/amdgpu: set at least the node size in the gtt manager Christian König
` (6 subsequent siblings)
8 siblings, 0 replies; 18+ messages in thread
From: Christian König @ 2016-09-27 9:49 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Christian König <christian.koenig@amd.com>
Split VRAM won't have a valid offset, so just set an explicit limit
when the flag is given to trigger reallocation if necessary.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index ba41807..c6754e7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -121,12 +121,17 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev,
if (domain & AMDGPU_GEM_DOMAIN_VRAM) {
unsigned visible_pfn = adev->mc.visible_vram_size >> PAGE_SHIFT;
+ unsigned lpfn = 0;
+
+ /* This forces a reallocation if the flag wasn't set before */
+ if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
+ lpfn = adev->mc.real_vram_size >> PAGE_SHIFT;
if (flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS &&
!(flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) &&
adev->mc.visible_vram_size < adev->mc.real_vram_size) {
places[c].fpfn = visible_pfn;
- places[c].lpfn = 0;
+ places[c].lpfn = lpfn;
places[c].flags = TTM_PL_FLAG_WC |
TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_VRAM |
TTM_PL_FLAG_TOPDOWN;
@@ -134,7 +139,7 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev,
}
places[c].fpfn = 0;
- places[c].lpfn = 0;
+ places[c].lpfn = lpfn;
places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
TTM_PL_FLAG_VRAM;
if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)
--
2.5.0
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 18+ messages in thread
* [PATCH 3/6] drm/amdgpu: set at least the node size in the gtt manager
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-09-27 9:49 ` [PATCH 1/6] drm/amdgpu: add AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS flag v3 Christian König
2016-09-27 9:49 ` [PATCH 2/6] drm/amdgpu: use explicit limit for VRAM_CONTIGUOUS Christian König
@ 2016-09-27 9:49 ` Christian König
2016-09-27 9:49 ` [PATCH 4/6] drm/amdgpu: handle multiple MM nodes in the VMs v2 Christian König
` (5 subsequent siblings)
8 siblings, 0 replies; 18+ messages in thread
From: Christian König @ 2016-09-27 9:49 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Christian König <christian.koenig@amd.com>
Otherwise the new VM code becomes confused.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index f86c844..3c634f0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -168,6 +168,7 @@ static int amdgpu_gtt_mgr_new(struct ttm_mem_type_manager *man,
return -ENOMEM;
node->start = AMDGPU_BO_INVALID_OFFSET;
+ node->size = mem->num_pages;
mem->mm_node = node;
if (place->fpfn || place->lpfn || place->flags & TTM_PL_FLAG_TOPDOWN) {
--
2.5.0
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 18+ messages in thread
* [PATCH 4/6] drm/amdgpu: handle multiple MM nodes in the VMs v2
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
` (2 preceding siblings ...)
2016-09-27 9:49 ` [PATCH 3/6] drm/amdgpu: set at least the node size in the gtt manager Christian König
@ 2016-09-27 9:49 ` Christian König
2016-09-27 9:49 ` [PATCH 5/6] drm/amdgpu: enable amdgpu_move_blit to handle multiple MM nodes v2 Christian König
` (4 subsequent siblings)
8 siblings, 0 replies; 18+ messages in thread
From: Christian König @ 2016-09-27 9:49 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Christian König <christian.koenig@amd.com>
This allows us to map scattered VRAM BOs to the VMs.
v2: fix offset handling, use pfn instead of offset,
fix PAGE_SIZE != AMDGPU_GPU_PAGE_SIZE case
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 81 +++++++++++++++++++---------------
1 file changed, 46 insertions(+), 35 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 2379eb1..6ed11cc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1069,8 +1069,8 @@ error_free:
* @pages_addr: DMA addresses to use for mapping
* @vm: requested vm
* @mapping: mapped range and flags to use for the update
- * @addr: addr to set the area to
* @flags: HW flags for the mapping
+ * @nodes: array of drm_mm_nodes with the MC addresses
* @fence: optional resulting fence
*
* Split the mapping into smaller chunks so that each update fits
@@ -1083,12 +1083,11 @@ static int amdgpu_vm_bo_split_mapping(struct amdgpu_device *adev,
dma_addr_t *pages_addr,
struct amdgpu_vm *vm,
struct amdgpu_bo_va_mapping *mapping,
- uint32_t flags, uint64_t addr,
+ uint32_t flags,
+ struct drm_mm_node *nodes,
struct fence **fence)
{
- const uint64_t max_size = 64ULL * 1024ULL * 1024ULL / AMDGPU_GPU_PAGE_SIZE;
-
- uint64_t src = 0, start = mapping->it.start;
+ uint64_t pfn, src = 0, start = mapping->it.start;
int r;
/* normally,bo_va->flags only contians READABLE and WIRTEABLE bit go here
@@ -1101,23 +1100,40 @@ static int amdgpu_vm_bo_split_mapping(struct amdgpu_device *adev,
trace_amdgpu_vm_bo_update(mapping);
- if (pages_addr) {
- if (flags == gtt_flags)
- src = adev->gart.table_addr + (addr >> 12) * 8;
- addr = 0;
+ pfn = mapping->offset >> PAGE_SHIFT;
+ if (nodes) {
+ while (pfn >= nodes->size) {
+ pfn -= nodes->size;
+ ++nodes;
+ }
}
- addr += mapping->offset;
- if (!pages_addr || src)
- return amdgpu_vm_bo_update_mapping(adev, exclusive,
- src, pages_addr, vm,
- start, mapping->it.last,
- flags, addr, fence);
+ do {
+ uint64_t max_entries;
+ uint64_t addr, last;
- while (start != mapping->it.last + 1) {
- uint64_t last;
+ if (nodes) {
+ addr = nodes->start << PAGE_SHIFT;
+ max_entries = (nodes->size - pfn) *
+ (PAGE_SIZE / AMDGPU_GPU_PAGE_SIZE);
+ } else {
+ addr = 0;
+ max_entries = S64_MAX;
+ }
- last = min((uint64_t)mapping->it.last, start + max_size - 1);
+ if (pages_addr) {
+ if (flags == gtt_flags)
+ src = adev->gart.table_addr +
+ (addr >> AMDGPU_GPU_PAGE_SHIFT) * 8;
+ else
+ max_entries = min(max_entries, 16ull * 1024ull);
+ addr = 0;
+ } else if (flags & AMDGPU_PTE_VALID) {
+ addr += adev->vm_manager.vram_base_offset;
+ }
+ addr += pfn << PAGE_SHIFT;
+
+ last = min((uint64_t)mapping->it.last, start + max_entries - 1);
r = amdgpu_vm_bo_update_mapping(adev, exclusive,
src, pages_addr, vm,
start, last, flags, addr,
@@ -1125,9 +1141,14 @@ static int amdgpu_vm_bo_split_mapping(struct amdgpu_device *adev,
if (r)
return r;
+ pfn += last - start + 1;
+ if (nodes && nodes->size == pfn) {
+ pfn = 0;
+ ++nodes;
+ }
start = last + 1;
- addr += max_size * AMDGPU_GPU_PAGE_SIZE;
- }
+
+ } while (unlikely(start != mapping->it.last + 1));
return 0;
}
@@ -1151,34 +1172,24 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
dma_addr_t *pages_addr = NULL;
uint32_t gtt_flags, flags;
struct ttm_mem_reg *mem;
+ struct drm_mm_node *nodes;
struct fence *exclusive;
- uint64_t addr;
int r;
if (clear) {
mem = NULL;
- addr = 0;
+ nodes = NULL;
exclusive = NULL;
} else {
struct ttm_dma_tt *ttm;
mem = &bo_va->bo->tbo.mem;
- addr = (u64)mem->start << PAGE_SHIFT;
- switch (mem->mem_type) {
- case TTM_PL_TT:
+ nodes = mem->mm_node;
+ if (mem->mem_type == TTM_PL_TT) {
ttm = container_of(bo_va->bo->tbo.ttm, struct
ttm_dma_tt, ttm);
pages_addr = ttm->dma_address;
- break;
-
- case TTM_PL_VRAM:
- addr += adev->vm_manager.vram_base_offset;
- break;
-
- default:
- break;
}
-
exclusive = reservation_object_get_excl(bo_va->bo->tbo.resv);
}
@@ -1194,7 +1205,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
list_for_each_entry(mapping, &bo_va->invalids, list) {
r = amdgpu_vm_bo_split_mapping(adev, exclusive,
gtt_flags, pages_addr, vm,
- mapping, flags, addr,
+ mapping, flags, nodes,
&bo_va->last_pt_update);
if (r)
return r;
--
2.5.0
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 18+ messages in thread
* [PATCH 5/6] drm/amdgpu: enable amdgpu_move_blit to handle multiple MM nodes v2
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
` (3 preceding siblings ...)
2016-09-27 9:49 ` [PATCH 4/6] drm/amdgpu: handle multiple MM nodes in the VMs v2 Christian König
@ 2016-09-27 9:49 ` Christian König
2016-09-27 9:49 ` [PATCH 6/6] drm/amdgpu: add VRAM manager v2 Christian König
` (3 subsequent siblings)
8 siblings, 0 replies; 18+ messages in thread
From: Christian König @ 2016-09-27 9:49 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Christian König <christian.koenig@amd.com>
This allows us to move scattered buffers around.
v2: fix a couple of typos, handle scattered to scattered moves as well.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 118 +++++++++++++++++++++++---------
1 file changed, 85 insertions(+), 33 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 06ab624..bd377d8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -259,64 +259,116 @@ static void amdgpu_move_null(struct ttm_buffer_object *bo,
new_mem->mm_node = NULL;
}
-static int amdgpu_move_blit(struct ttm_buffer_object *bo,
- bool evict, bool no_wait_gpu,
- struct ttm_mem_reg *new_mem,
- struct ttm_mem_reg *old_mem)
+static int amdgpu_mm_node_addr(struct ttm_buffer_object *bo,
+ struct drm_mm_node *mm_node,
+ struct ttm_mem_reg *mem,
+ uint64_t *addr)
{
- struct amdgpu_device *adev;
- struct amdgpu_ring *ring;
- uint64_t old_start, new_start;
- struct fence *fence;
int r;
- adev = amdgpu_get_adev(bo->bdev);
- ring = adev->mman.buffer_funcs_ring;
-
- switch (old_mem->mem_type) {
+ switch (mem->mem_type) {
case TTM_PL_TT:
- r = amdgpu_ttm_bind(bo, old_mem);
+ r = amdgpu_ttm_bind(bo, mem);
if (r)
return r;
case TTM_PL_VRAM:
- old_start = (u64)old_mem->start << PAGE_SHIFT;
- old_start += bo->bdev->man[old_mem->mem_type].gpu_offset;
+ *addr = mm_node->start << PAGE_SHIFT;
+ *addr += bo->bdev->man[mem->mem_type].gpu_offset;
break;
default:
- DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
+ DRM_ERROR("Unknown placement %d\n", mem->mem_type);
return -EINVAL;
}
- switch (new_mem->mem_type) {
- case TTM_PL_TT:
- r = amdgpu_ttm_bind(bo, new_mem);
- if (r)
- return r;
- case TTM_PL_VRAM:
- new_start = (u64)new_mem->start << PAGE_SHIFT;
- new_start += bo->bdev->man[new_mem->mem_type].gpu_offset;
- break;
- default:
- DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
- return -EINVAL;
- }
+ return 0;
+}
+
+static int amdgpu_move_blit(struct ttm_buffer_object *bo,
+ bool evict, bool no_wait_gpu,
+ struct ttm_mem_reg *new_mem,
+ struct ttm_mem_reg *old_mem)
+{
+ struct amdgpu_device *adev = amdgpu_get_adev(bo->bdev);
+ struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+
+ struct drm_mm_node *old_mm, *new_mm;
+ uint64_t old_start, old_size, new_start, new_size;
+ unsigned long num_pages;
+ struct fence *fence = NULL;
+ int r;
+
+ BUILD_BUG_ON((PAGE_SIZE % AMDGPU_GPU_PAGE_SIZE) != 0);
+
if (!ring->ready) {
DRM_ERROR("Trying to move memory with ring turned off.\n");
return -EINVAL;
}
- BUILD_BUG_ON((PAGE_SIZE % AMDGPU_GPU_PAGE_SIZE) != 0);
+ old_mm = old_mem->mm_node;
+ r = amdgpu_mm_node_addr(bo, old_mm, old_mem, &old_start);
+ if (r)
+ return r;
+ old_size = old_mm->size;
+
- r = amdgpu_copy_buffer(ring, old_start, new_start,
- new_mem->num_pages * PAGE_SIZE, /* bytes */
- bo->resv, &fence, false);
+ new_mm = new_mem->mm_node;
+ r = amdgpu_mm_node_addr(bo, new_mm, new_mem, &new_start);
if (r)
return r;
+ new_size = new_mm->size;
+
+ num_pages = new_mem->num_pages;
+ while (num_pages) {
+ unsigned long cur_pages = min(old_size, new_size);
+ struct fence *next;
+
+ r = amdgpu_copy_buffer(ring, old_start, new_start,
+ cur_pages * PAGE_SIZE,
+ bo->resv, &next, false);
+ if (r)
+ goto error;
+
+ fence_put(fence);
+ fence = next;
+
+ num_pages -= cur_pages;
+ if (!num_pages)
+ break;
+
+ old_size -= cur_pages;
+ if (!old_size) {
+ r = amdgpu_mm_node_addr(bo, ++old_mm, old_mem,
+ &old_start);
+ if (r)
+ goto error;
+ old_size = old_mm->size;
+ } else {
+ old_start += cur_pages * PAGE_SIZE;
+ }
+
+ new_size -= cur_pages;
+ if (!new_size) {
+ r = amdgpu_mm_node_addr(bo, ++new_mm, new_mem,
+ &new_start);
+ if (r)
+ goto error;
+
+ new_size = new_mm->size;
+ } else {
+ new_start += cur_pages * PAGE_SIZE;
+ }
+ }
r = ttm_bo_pipeline_move(bo, fence, evict, new_mem);
fence_put(fence);
return r;
+
+error:
+ if (fence)
+ fence_wait(fence, false);
+ fence_put(fence);
+ return r;
}
static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo,
--
2.5.0
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 18+ messages in thread
* [PATCH 6/6] drm/amdgpu: add VRAM manager v2
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
` (4 preceding siblings ...)
2016-09-27 9:49 ` [PATCH 5/6] drm/amdgpu: enable amdgpu_move_blit to handle multiple MM nodes v2 Christian König
@ 2016-09-27 9:49 ` Christian König
[not found] ` <1474969797-1882-7-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-09-27 10:50 ` VRAM manager Mike Lothian
` (2 subsequent siblings)
8 siblings, 1 reply; 18+ messages in thread
From: Christian König @ 2016-09-27 9:49 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Christian König <christian.koenig@amd.com>
Split VRAM allocations into 4MB blocks.
v2: fix typo in comment, some suggested cleanups
v3: document how to disable the feature, fix rebase issue
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/Makefile | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 222 +++++++++++++++++++++++++++
7 files changed, 237 insertions(+), 2 deletions(-)
create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index 786b28a..2874bfe 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -30,7 +30,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
atombios_encoders.o amdgpu_sa.o atombios_i2c.o \
amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \
amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
- amdgpu_gtt_mgr.o
+ amdgpu_gtt_mgr.o amdgpu_vram_mgr.o
# add asic specific block
amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 73c9a2d..2a95827 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -99,6 +99,7 @@ extern char *amdgpu_disable_cu;
extern int amdgpu_sclk_deep_sleep_en;
extern char *amdgpu_virtual_display;
extern unsigned amdgpu_pp_feature_mask;
+extern int amdgpu_vram_page_split;
#define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000
#define AMDGPU_MAX_USEC_TIMEOUT 100000 /* 100 ms */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 5ca6a38..d66a8df8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -1053,6 +1053,13 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
amdgpu_vm_block_size);
amdgpu_vm_block_size = 9;
}
+
+ if ((amdgpu_vram_page_split != -1 && amdgpu_vram_page_split < 16) ||
+ !amdgpu_check_pot_argument(amdgpu_vram_page_split)) {
+ dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
+ amdgpu_vram_page_split);
+ amdgpu_vram_page_split = 1024;
+ }
}
/**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 08ba441..538d3a7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -84,6 +84,7 @@ int amdgpu_vm_size = 64;
int amdgpu_vm_block_size = -1;
int amdgpu_vm_fault_stop = 0;
int amdgpu_vm_debug = 0;
+int amdgpu_vram_page_split = 1024;
int amdgpu_exp_hw_support = 0;
int amdgpu_dal = -1;
int amdgpu_sched_jobs = 32;
@@ -165,6 +166,9 @@ module_param_named(vm_fault_stop, amdgpu_vm_fault_stop, int, 0444);
MODULE_PARM_DESC(vm_debug, "Debug VM handling (0 = disabled (default), 1 = enabled)");
module_param_named(vm_debug, amdgpu_vm_debug, int, 0644);
+MODULE_PARM_DESC(vram_page_split, "Number of pages after we split VRAM allocations (default 1024, -1 = disable)");
+module_param_named(vram_page_split, amdgpu_vram_page_split, int, 0444);
+
MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index bd377d8..588e242 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -168,7 +168,7 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
break;
case TTM_PL_VRAM:
/* "On-card" video ram */
- man->func = &ttm_bo_manager_func;
+ man->func = &amdgpu_vram_mgr_func;
man->gpu_offset = adev->mc.vram_start;
man->flags = TTM_MEMTYPE_FLAG_FIXED |
TTM_MEMTYPE_FLAG_MAPPABLE;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 9812c80..d1c00c0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -66,6 +66,7 @@ struct amdgpu_mman {
};
extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
+extern const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func;
int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
struct ttm_buffer_object *tbo,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
new file mode 100644
index 0000000..1125f7f
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
@@ -0,0 +1,222 @@
+/*
+ * Copyright 2015 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Christian König
+ */
+
+#include <drm/drmP.h>
+#include "amdgpu.h"
+
+struct amdgpu_vram_mgr {
+ struct drm_mm mm;
+ spinlock_t lock;
+};
+
+/**
+ * amdgpu_vram_mgr_init - init VRAM manager and DRM MM
+ *
+ * @man: TTM memory type manager
+ * @p_size: maximum size of VRAM
+ *
+ * Allocate and initialize the VRAM manager.
+ */
+static int amdgpu_vram_mgr_init(struct ttm_mem_type_manager *man,
+ unsigned long p_size)
+{
+ struct amdgpu_vram_mgr *mgr;
+
+ mgr = kzalloc(sizeof(*mgr), GFP_KERNEL);
+ if (!mgr)
+ return -ENOMEM;
+
+ drm_mm_init(&mgr->mm, 0, p_size);
+ spin_lock_init(&mgr->lock);
+ man->priv = mgr;
+ return 0;
+}
+
+/**
+ * amdgpu_vram_mgr_fini - free and destroy VRAM manager
+ *
+ * @man: TTM memory type manager
+ *
+ * Destroy and free the VRAM manager, returns -EBUSY if ranges are still
+ * allocated inside it.
+ */
+static int amdgpu_vram_mgr_fini(struct ttm_mem_type_manager *man)
+{
+ struct amdgpu_vram_mgr *mgr = man->priv;
+
+ spin_lock(&mgr->lock);
+ if (!drm_mm_clean(&mgr->mm)) {
+ spin_unlock(&mgr->lock);
+ return -EBUSY;
+ }
+
+ drm_mm_takedown(&mgr->mm);
+ spin_unlock(&mgr->lock);
+ kfree(mgr);
+ man->priv = NULL;
+ return 0;
+}
+
+/**
+ * amdgpu_vram_mgr_new - allocate new ranges
+ *
+ * @man: TTM memory type manager
+ * @tbo: TTM BO we need this range for
+ * @place: placement flags and restrictions
+ * @mem: the resulting mem object
+ *
+ * Allocate VRAM for the given BO.
+ */
+static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
+ struct ttm_buffer_object *tbo,
+ const struct ttm_place *place,
+ struct ttm_mem_reg *mem)
+{
+ struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo);
+ struct amdgpu_vram_mgr *mgr = man->priv;
+ struct drm_mm *mm = &mgr->mm;
+ struct drm_mm_node *nodes;
+ enum drm_mm_search_flags sflags = DRM_MM_SEARCH_DEFAULT;
+ enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
+ unsigned long lpfn, num_nodes, pages_per_node, pages_left;
+ unsigned i;
+ int r;
+
+ lpfn = place->lpfn;
+ if (!lpfn)
+ lpfn = man->size;
+
+ if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS ||
+ amdgpu_vram_page_split == -1) {
+ pages_per_node = ~0ul;
+ num_nodes = 1;
+ } else {
+ pages_per_node = max((uint32_t)amdgpu_vram_page_split,
+ mem->page_alignment);
+ num_nodes = DIV_ROUND_UP(mem->num_pages, pages_per_node);
+ }
+
+ nodes = kcalloc(num_nodes, sizeof(*nodes), GFP_KERNEL);
+ if (!nodes)
+ return -ENOMEM;
+
+ if (place->flags & TTM_PL_FLAG_TOPDOWN) {
+ sflags = DRM_MM_SEARCH_BELOW;
+ aflags = DRM_MM_CREATE_TOP;
+ }
+
+ pages_left = mem->num_pages;
+
+ spin_lock(&mgr->lock);
+ for (i = 0; i < num_nodes; ++i) {
+ unsigned long pages = min(pages_left, pages_per_node);
+ uint32_t alignment = mem->page_alignment;
+
+ if (pages == pages_per_node)
+ alignment = pages_per_node;
+ else
+ sflags |= DRM_MM_SEARCH_BEST;
+
+ r = drm_mm_insert_node_in_range_generic(mm, &nodes[i], pages,
+ alignment, 0,
+ place->fpfn, lpfn,
+ sflags, aflags);
+ if (unlikely(r))
+ goto error;
+
+ pages_left -= pages;
+ }
+ spin_unlock(&mgr->lock);
+
+ mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET;
+ mem->mm_node = nodes;
+
+ return 0;
+
+error:
+ while (i--)
+ drm_mm_remove_node(&nodes[i]);
+ spin_unlock(&mgr->lock);
+
+ kfree(nodes);
+ return r == -ENOSPC ? 0 : r;
+}
+
+/**
+ * amdgpu_vram_mgr_del - free ranges
+ *
+ * @man: TTM memory type manager
+ * @tbo: TTM BO we need this range for
+ * @place: placement flags and restrictions
+ * @mem: TTM memory object
+ *
+ * Free the allocated VRAM again.
+ */
+static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
+ struct ttm_mem_reg *mem)
+{
+ struct amdgpu_vram_mgr *mgr = man->priv;
+ struct drm_mm_node *nodes = mem->mm_node;
+ unsigned pages = mem->num_pages;
+
+ if (!mem->mm_node)
+ return;
+
+ spin_lock(&mgr->lock);
+ while (pages) {
+ pages -= nodes->size;
+ drm_mm_remove_node(nodes);
+ ++nodes;
+ }
+ spin_unlock(&mgr->lock);
+
+ kfree(mem->mm_node);
+ mem->mm_node = NULL;
+}
+
+/**
+ * amdgpu_vram_mgr_debug - dump VRAM table
+ *
+ * @man: TTM memory type manager
+ * @prefix: text prefix
+ *
+ * Dump the table content using printk.
+ */
+static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
+ const char *prefix)
+{
+ struct amdgpu_vram_mgr *mgr = man->priv;
+
+ spin_lock(&mgr->lock);
+ drm_mm_debug_table(&mgr->mm, prefix);
+ spin_unlock(&mgr->lock);
+}
+
+const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func = {
+ amdgpu_vram_mgr_init,
+ amdgpu_vram_mgr_fini,
+ amdgpu_vram_mgr_new,
+ amdgpu_vram_mgr_del,
+ amdgpu_vram_mgr_debug
+};
--
2.5.0
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 18+ messages in thread
* Re: [PATCH 6/6] drm/amdgpu: add VRAM manager v2
[not found] ` <1474969797-1882-7-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
@ 2016-09-27 9:57 ` zhoucm1
0 siblings, 0 replies; 18+ messages in thread
From: zhoucm1 @ 2016-09-27 9:57 UTC (permalink / raw)
To: Christian König, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
On 2016年09月27日 17:49, Christian König wrote:
> From: Christian König <christian.koenig@amd.com>
>
> Split VRAM allocations into 4MB blocks.
>
> v2: fix typo in comment, some suggested cleanups
> v3: document how to disable the feature, fix rebase issue
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/Makefile | 2 +-
> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +-
> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 222 +++++++++++++++++++++++++++
> 7 files changed, 237 insertions(+), 2 deletions(-)
> create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
> index 786b28a..2874bfe 100644
> --- a/drivers/gpu/drm/amd/amdgpu/Makefile
> +++ b/drivers/gpu/drm/amd/amdgpu/Makefile
> @@ -30,7 +30,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
> atombios_encoders.o amdgpu_sa.o atombios_i2c.o \
> amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \
> amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
> - amdgpu_gtt_mgr.o
> + amdgpu_gtt_mgr.o amdgpu_vram_mgr.o
>
> # add asic specific block
> amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> index 73c9a2d..2a95827 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> @@ -99,6 +99,7 @@ extern char *amdgpu_disable_cu;
> extern int amdgpu_sclk_deep_sleep_en;
> extern char *amdgpu_virtual_display;
> extern unsigned amdgpu_pp_feature_mask;
> +extern int amdgpu_vram_page_split;
>
> #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000
> #define AMDGPU_MAX_USEC_TIMEOUT 100000 /* 100 ms */
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 5ca6a38..d66a8df8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -1053,6 +1053,13 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
> amdgpu_vm_block_size);
> amdgpu_vm_block_size = 9;
> }
> +
> + if ((amdgpu_vram_page_split != -1 && amdgpu_vram_page_split < 16) ||
> + !amdgpu_check_pot_argument(amdgpu_vram_page_split)) {
> + dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
> + amdgpu_vram_page_split);
> + amdgpu_vram_page_split = 1024;
> + }
> }
>
> /**
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> index 08ba441..538d3a7 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> @@ -84,6 +84,7 @@ int amdgpu_vm_size = 64;
> int amdgpu_vm_block_size = -1;
> int amdgpu_vm_fault_stop = 0;
> int amdgpu_vm_debug = 0;
> +int amdgpu_vram_page_split = 1024;
> int amdgpu_exp_hw_support = 0;
> int amdgpu_dal = -1;
> int amdgpu_sched_jobs = 32;
> @@ -165,6 +166,9 @@ module_param_named(vm_fault_stop, amdgpu_vm_fault_stop, int, 0444);
> MODULE_PARM_DESC(vm_debug, "Debug VM handling (0 = disabled (default), 1 = enabled)");
> module_param_named(vm_debug, amdgpu_vm_debug, int, 0644);
>
> +MODULE_PARM_DESC(vram_page_split, "Number of pages after we split VRAM allocations (default 1024, -1 = disable)");
> +module_param_named(vram_page_split, amdgpu_vram_page_split, int, 0444);
> +
> MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
> module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index bd377d8..588e242 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -168,7 +168,7 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
> break;
> case TTM_PL_VRAM:
> /* "On-card" video ram */
> - man->func = &ttm_bo_manager_func;
> + man->func = &amdgpu_vram_mgr_func;
> man->gpu_offset = adev->mc.vram_start;
> man->flags = TTM_MEMTYPE_FLAG_FIXED |
> TTM_MEMTYPE_FLAG_MAPPABLE;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
> index 9812c80..d1c00c0 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
> @@ -66,6 +66,7 @@ struct amdgpu_mman {
> };
>
> extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
> +extern const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func;
>
> int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
> struct ttm_buffer_object *tbo,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
> new file mode 100644
> index 0000000..1125f7f
> --- /dev/null
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
> @@ -0,0 +1,222 @@
> +/*
> + * Copyright 2015 Advanced Micro Devices, Inc.
2016 is expected.
Regards,
David Zhou
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
> + * OTHER DEALINGS IN THE SOFTWARE.
> + *
> + * Authors: Christian König
> + */
> +
> +#include <drm/drmP.h>
> +#include "amdgpu.h"
> +
> +struct amdgpu_vram_mgr {
> + struct drm_mm mm;
> + spinlock_t lock;
> +};
> +
> +/**
> + * amdgpu_vram_mgr_init - init VRAM manager and DRM MM
> + *
> + * @man: TTM memory type manager
> + * @p_size: maximum size of VRAM
> + *
> + * Allocate and initialize the VRAM manager.
> + */
> +static int amdgpu_vram_mgr_init(struct ttm_mem_type_manager *man,
> + unsigned long p_size)
> +{
> + struct amdgpu_vram_mgr *mgr;
> +
> + mgr = kzalloc(sizeof(*mgr), GFP_KERNEL);
> + if (!mgr)
> + return -ENOMEM;
> +
> + drm_mm_init(&mgr->mm, 0, p_size);
> + spin_lock_init(&mgr->lock);
> + man->priv = mgr;
> + return 0;
> +}
> +
> +/**
> + * amdgpu_vram_mgr_fini - free and destroy VRAM manager
> + *
> + * @man: TTM memory type manager
> + *
> + * Destroy and free the VRAM manager, returns -EBUSY if ranges are still
> + * allocated inside it.
> + */
> +static int amdgpu_vram_mgr_fini(struct ttm_mem_type_manager *man)
> +{
> + struct amdgpu_vram_mgr *mgr = man->priv;
> +
> + spin_lock(&mgr->lock);
> + if (!drm_mm_clean(&mgr->mm)) {
> + spin_unlock(&mgr->lock);
> + return -EBUSY;
> + }
> +
> + drm_mm_takedown(&mgr->mm);
> + spin_unlock(&mgr->lock);
> + kfree(mgr);
> + man->priv = NULL;
> + return 0;
> +}
> +
> +/**
> + * amdgpu_vram_mgr_new - allocate new ranges
> + *
> + * @man: TTM memory type manager
> + * @tbo: TTM BO we need this range for
> + * @place: placement flags and restrictions
> + * @mem: the resulting mem object
> + *
> + * Allocate VRAM for the given BO.
> + */
> +static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
> + struct ttm_buffer_object *tbo,
> + const struct ttm_place *place,
> + struct ttm_mem_reg *mem)
> +{
> + struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo);
> + struct amdgpu_vram_mgr *mgr = man->priv;
> + struct drm_mm *mm = &mgr->mm;
> + struct drm_mm_node *nodes;
> + enum drm_mm_search_flags sflags = DRM_MM_SEARCH_DEFAULT;
> + enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
> + unsigned long lpfn, num_nodes, pages_per_node, pages_left;
> + unsigned i;
> + int r;
> +
> + lpfn = place->lpfn;
> + if (!lpfn)
> + lpfn = man->size;
> +
> + if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS ||
> + amdgpu_vram_page_split == -1) {
> + pages_per_node = ~0ul;
> + num_nodes = 1;
> + } else {
> + pages_per_node = max((uint32_t)amdgpu_vram_page_split,
> + mem->page_alignment);
> + num_nodes = DIV_ROUND_UP(mem->num_pages, pages_per_node);
> + }
> +
> + nodes = kcalloc(num_nodes, sizeof(*nodes), GFP_KERNEL);
> + if (!nodes)
> + return -ENOMEM;
> +
> + if (place->flags & TTM_PL_FLAG_TOPDOWN) {
> + sflags = DRM_MM_SEARCH_BELOW;
> + aflags = DRM_MM_CREATE_TOP;
> + }
> +
> + pages_left = mem->num_pages;
> +
> + spin_lock(&mgr->lock);
> + for (i = 0; i < num_nodes; ++i) {
> + unsigned long pages = min(pages_left, pages_per_node);
> + uint32_t alignment = mem->page_alignment;
> +
> + if (pages == pages_per_node)
> + alignment = pages_per_node;
> + else
> + sflags |= DRM_MM_SEARCH_BEST;
> +
> + r = drm_mm_insert_node_in_range_generic(mm, &nodes[i], pages,
> + alignment, 0,
> + place->fpfn, lpfn,
> + sflags, aflags);
> + if (unlikely(r))
> + goto error;
> +
> + pages_left -= pages;
> + }
> + spin_unlock(&mgr->lock);
> +
> + mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET;
> + mem->mm_node = nodes;
> +
> + return 0;
> +
> +error:
> + while (i--)
> + drm_mm_remove_node(&nodes[i]);
> + spin_unlock(&mgr->lock);
> +
> + kfree(nodes);
> + return r == -ENOSPC ? 0 : r;
> +}
> +
> +/**
> + * amdgpu_vram_mgr_del - free ranges
> + *
> + * @man: TTM memory type manager
> + * @tbo: TTM BO we need this range for
> + * @place: placement flags and restrictions
> + * @mem: TTM memory object
> + *
> + * Free the allocated VRAM again.
> + */
> +static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
> + struct ttm_mem_reg *mem)
> +{
> + struct amdgpu_vram_mgr *mgr = man->priv;
> + struct drm_mm_node *nodes = mem->mm_node;
> + unsigned pages = mem->num_pages;
> +
> + if (!mem->mm_node)
> + return;
> +
> + spin_lock(&mgr->lock);
> + while (pages) {
> + pages -= nodes->size;
> + drm_mm_remove_node(nodes);
> + ++nodes;
> + }
> + spin_unlock(&mgr->lock);
> +
> + kfree(mem->mm_node);
> + mem->mm_node = NULL;
> +}
> +
> +/**
> + * amdgpu_vram_mgr_debug - dump VRAM table
> + *
> + * @man: TTM memory type manager
> + * @prefix: text prefix
> + *
> + * Dump the table content using printk.
> + */
> +static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
> + const char *prefix)
> +{
> + struct amdgpu_vram_mgr *mgr = man->priv;
> +
> + spin_lock(&mgr->lock);
> + drm_mm_debug_table(&mgr->mm, prefix);
> + spin_unlock(&mgr->lock);
> +}
> +
> +const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func = {
> + amdgpu_vram_mgr_init,
> + amdgpu_vram_mgr_fini,
> + amdgpu_vram_mgr_new,
> + amdgpu_vram_mgr_del,
> + amdgpu_vram_mgr_debug
> +};
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: VRAM manager
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
` (5 preceding siblings ...)
2016-09-27 9:49 ` [PATCH 6/6] drm/amdgpu: add VRAM manager v2 Christian König
@ 2016-09-27 10:50 ` Mike Lothian
2016-09-27 11:17 ` Edward O'Callaghan
2016-10-07 22:11 ` Felix Kuehling
8 siblings, 0 replies; 18+ messages in thread
From: Mike Lothian @ 2016-09-27 10:50 UTC (permalink / raw)
To: Christian König; +Cc: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Hi Christian
It's working for me on my Tonga now, thanks for fixing those bugs
Feel free to add a testing by
Cheers
Mike
On 27 September 2016 at 10:49, Christian König <deathsimple@vodafone.de> wrote:
> Hi guys,
>
> after fixing all those nasty little bugs this seems to be stable now.
>
> Anybody brave enough to give it an review?
>
> Cheers,
> Christian.
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: VRAM manager
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
` (6 preceding siblings ...)
2016-09-27 10:50 ` VRAM manager Mike Lothian
@ 2016-09-27 11:17 ` Edward O'Callaghan
2016-10-07 22:11 ` Felix Kuehling
8 siblings, 0 replies; 18+ messages in thread
From: Edward O'Callaghan @ 2016-09-27 11:17 UTC (permalink / raw)
To: Christian König, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
[-- Attachment #1.1.1: Type: text/plain, Size: 553 bytes --]
This series is, to the best of my ability,
Reviewed-by: Edward O'Callaghan <funfunctor-dczkZgxz+BNUPWh3PAxdjQ@public.gmane.org>
On 09/27/2016 07:49 PM, Christian König wrote:
> Hi guys,
>
> after fixing all those nasty little bugs this seems to be stable now.
>
> Anybody brave enough to give it an review?
>
> Cheers,
> Christian.
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>
[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
[-- Attachment #2: Type: text/plain, Size: 154 bytes --]
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH 1/6] drm/amdgpu: add AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS flag v3
[not found] ` <1474969797-1882-2-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
@ 2016-10-07 21:19 ` Felix Kuehling
[not found] ` <627af621-8fb2-ba41-c87e-96bb12e594d8-5C7GfCeVMHo@public.gmane.org>
0 siblings, 1 reply; 18+ messages in thread
From: Felix Kuehling @ 2016-10-07 21:19 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Christian König
On 16-09-27 05:49 AM, Christian König wrote:
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> @@ -1195,6 +1195,15 @@ int amdgpu_cs_sysvm_access_required(struct amdgpu_cs_parser *parser)
> r = amdgpu_ttm_bind(&bo->tbo, &bo->tbo.mem);
> if (unlikely(r))
> return r;
> +
> + if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
> + continue;
Should you also continue if the BO is not VRAM?
Regards,
Felix
> +
> + bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
> + amdgpu_ttm_placement_from_domain(bo, bo->allowed_domains);
> + r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
> + if (unlikely(r))
> + return r;
> }
>
> return 0;
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: VRAM manager
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
` (7 preceding siblings ...)
2016-09-27 11:17 ` Edward O'Callaghan
@ 2016-10-07 22:11 ` Felix Kuehling
[not found] ` <ad665321-d00c-1ded-051f-2ca8eae014b4-5C7GfCeVMHo@public.gmane.org>
8 siblings, 1 reply; 18+ messages in thread
From: Felix Kuehling @ 2016-10-07 22:11 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Christian König
Hi Christian,
I'm back from vacation and finally caught up with my email and
code-review backlog.
I added a comment on patch 1.
I don't understand patch 2. I couldn't find where a non-0 lpfn would
trigger any reallocation. I think this would have to be somewhere in
ttm_bo_validate or ttm_bo_mem_compat. I also grepped for all uses of
lpfn in TTM and AMDGPU. Maybe I'm looking for the wrong thing in the
wrong place.
Patches 3-6 look good to me (other than the copyright date pointed out
by David). Please add my Reviewed-by.
Regards,
Felix
On 16-09-27 05:49 AM, Christian König wrote:
> Hi guys,
>
> after fixing all those nasty little bugs this seems to be stable now.
>
> Anybody brave enough to give it an review?
>
> Cheers,
> Christian.
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: VRAM manager
[not found] ` <ad665321-d00c-1ded-051f-2ca8eae014b4-5C7GfCeVMHo@public.gmane.org>
@ 2016-10-11 9:34 ` Christian König
[not found] ` <6235e143-b44a-0039-b84c-c1be81de4a4c-5C7GfCeVMHo@public.gmane.org>
0 siblings, 1 reply; 18+ messages in thread
From: Christian König @ 2016-10-11 9:34 UTC (permalink / raw)
To: Felix Kuehling, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Hi Felix,
the pfn (mem->start) is set by the address space manager to
AMDGPU_BO_INVALID_OFFSET which is a very high number.
So setting the lpfn to an explicit value for the validation will result
in reallocation of the backing store because AMDGPU_BO_INVALID_OFFSET
will always be larger than lpfn.
Regards,
Christian.
Am 08.10.2016 um 00:11 schrieb Felix Kuehling:
> Hi Christian,
>
> I'm back from vacation and finally caught up with my email and
> code-review backlog.
>
> I added a comment on patch 1.
>
> I don't understand patch 2. I couldn't find where a non-0 lpfn would
> trigger any reallocation. I think this would have to be somewhere in
> ttm_bo_validate or ttm_bo_mem_compat. I also grepped for all uses of
> lpfn in TTM and AMDGPU. Maybe I'm looking for the wrong thing in the
> wrong place.
>
> Patches 3-6 look good to me (other than the copyright date pointed out
> by David). Please add my Reviewed-by.
>
> Regards,
> Felix
>
>
> On 16-09-27 05:49 AM, Christian König wrote:
>> Hi guys,
>>
>> after fixing all those nasty little bugs this seems to be stable now.
>>
>> Anybody brave enough to give it an review?
>>
>> Cheers,
>> Christian.
>>
>> _______________________________________________
>> amd-gfx mailing list
>> amd-gfx@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH 1/6] drm/amdgpu: add AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS flag v3
[not found] ` <627af621-8fb2-ba41-c87e-96bb12e594d8-5C7GfCeVMHo@public.gmane.org>
@ 2016-10-11 9:35 ` Christian König
0 siblings, 0 replies; 18+ messages in thread
From: Christian König @ 2016-10-11 9:35 UTC (permalink / raw)
To: Felix Kuehling, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Am 07.10.2016 um 23:19 schrieb Felix Kuehling:
> On 16-09-27 05:49 AM, Christian König wrote:
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> @@ -1195,6 +1195,15 @@ int amdgpu_cs_sysvm_access_required(struct amdgpu_cs_parser *parser)
>> r = amdgpu_ttm_bind(&bo->tbo, &bo->tbo.mem);
>> if (unlikely(r))
>> return r;
>> +
>> + if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
>> + continue;
> Should you also continue if the BO is not VRAM?
No, the placement of the BO at this point isn't final. For older UVD
blocks this can still be shuffled around to fulfill the placement
restrictions.
So the BO could still move into VRAM even after this.
Regards,
Christian.
>
> Regards,
> Felix
>
>> +
>> + bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
>> + amdgpu_ttm_placement_from_domain(bo, bo->allowed_domains);
>> + r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
>> + if (unlikely(r))
>> + return r;
>> }
>>
>> return 0;
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: VRAM manager
[not found] ` <6235e143-b44a-0039-b84c-c1be81de4a4c-5C7GfCeVMHo@public.gmane.org>
@ 2016-10-11 18:59 ` Felix Kuehling
0 siblings, 0 replies; 18+ messages in thread
From: Felix Kuehling @ 2016-10-11 18:59 UTC (permalink / raw)
To: Christian König, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Thanks for the explanation (and the one on patch 1). The whole series is
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
On 16-10-11 05:34 AM, Christian König wrote:
> Hi Felix,
>
> the pfn (mem->start) is set by the address space manager to
> AMDGPU_BO_INVALID_OFFSET which is a very high number.
>
> So setting the lpfn to an explicit value for the validation will
> result in reallocation of the backing store because
> AMDGPU_BO_INVALID_OFFSET will always be larger than lpfn.
>
> Regards,
> Christian.
>
> Am 08.10.2016 um 00:11 schrieb Felix Kuehling:
>> Hi Christian,
>>
>> I'm back from vacation and finally caught up with my email and
>> code-review backlog.
>>
>> I added a comment on patch 1.
>>
>> I don't understand patch 2. I couldn't find where a non-0 lpfn would
>> trigger any reallocation. I think this would have to be somewhere in
>> ttm_bo_validate or ttm_bo_mem_compat. I also grepped for all uses of
>> lpfn in TTM and AMDGPU. Maybe I'm looking for the wrong thing in the
>> wrong place.
>>
>> Patches 3-6 look good to me (other than the copyright date pointed out
>> by David). Please add my Reviewed-by.
>>
>> Regards,
>> Felix
>>
>>
>> On 16-09-27 05:49 AM, Christian König wrote:
>>> Hi guys,
>>>
>>> after fixing all those nasty little bugs this seems to be stable now.
>>>
>>> Anybody brave enough to give it an review?
>>>
>>> Cheers,
>>> Christian.
>>>
>>> _______________________________________________
>>> amd-gfx mailing list
>>> amd-gfx@lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>
>
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH 6/6] drm/amdgpu: add VRAM manager v2
[not found] ` <CADnq5_NSW3A_7EqUzABsEUddyJYUKyfpUR2fW-Ku4x3VF0Rykw-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
@ 2016-09-15 17:42 ` Christian König
0 siblings, 0 replies; 18+ messages in thread
From: Christian König @ 2016-09-15 17:42 UTC (permalink / raw)
To: Alex Deucher; +Cc: amd-gfx list
Am 15.09.2016 um 16:44 schrieb Alex Deucher:
> On Thu, Sep 15, 2016 at 6:12 AM, Christian König
> <deathsimple@vodafone.de> wrote:
>> From: Christian König <christian.koenig@amd.com>
>>
>> Split VRAM allocations into 4MB blocks.
>>
>> v2: fix typo in comment, some suggested cleanups
>>
>> Signed-off-by: Christian König <christian.koenig@amd.com>
>> ---
>> drivers/gpu/drm/amd/amdgpu/Makefile | 2 +-
>> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +-
>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 222 +++++++++++++++++++++++++++
>> 8 files changed, 238 insertions(+), 2 deletions(-)
>> create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
>> index f2b97cb..236e9df 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/Makefile
>> +++ b/drivers/gpu/drm/amd/amdgpu/Makefile
>> @@ -30,7 +30,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
>> atombios_encoders.o amdgpu_sa.o atombios_i2c.o \
>> amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \
>> amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
>> - amdgpu_gtt_mgr.o
>> + amdgpu_gtt_mgr.o amdgpu_vram_mgr.o
>>
>> # add asic specific block
>> amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> index 588baaf..ee55763 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> @@ -98,6 +98,7 @@ extern char *amdgpu_disable_cu;
>> extern int amdgpu_sclk_deep_sleep_en;
>> extern char *amdgpu_virtual_display;
>> extern unsigned amdgpu_pp_feature_mask;
>> +extern int amdgpu_vram_page_split;
>>
>> #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000
>> #define AMDGPU_MAX_USEC_TIMEOUT 100000 /* 100 ms */
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> index 5686d12..f26b067 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> @@ -1012,6 +1012,13 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
>> amdgpu_vm_block_size);
>> amdgpu_vm_block_size = 9;
>> }
>> +
>> + if ((amdgpu_vram_page_split != -1 && amdgpu_vram_page_split < 16) ||
>> + !amdgpu_check_pot_argument(amdgpu_vram_page_split)) {
>> + dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
>> + amdgpu_vram_page_split);
>> + amdgpu_vram_page_split = 1024;
>> + }
>> }
>>
>> /**
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> index 902da13..44e605d 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> @@ -83,6 +83,7 @@ int amdgpu_vm_size = 64;
>> int amdgpu_vm_block_size = -1;
>> int amdgpu_vm_fault_stop = 0;
>> int amdgpu_vm_debug = 0;
>> +int amdgpu_vram_page_split = 1024;
>> int amdgpu_exp_hw_support = 0;
>> int amdgpu_dal = -1;
>> int amdgpu_sched_jobs = 32;
>> @@ -164,6 +165,9 @@ module_param_named(vm_fault_stop, amdgpu_vm_fault_stop, int, 0444);
>> MODULE_PARM_DESC(vm_debug, "Debug VM handling (0 = disabled (default), 1 = enabled)");
>> module_param_named(vm_debug, amdgpu_vm_debug, int, 0644);
>>
>> +MODULE_PARM_DESC(vram_page_split, "Number of pages after we split VRAM allocations (default 1024)");
>> +module_param_named(vram_page_split, amdgpu_vram_page_split, int, 0444);
> Not a big deal, but maybe make this unsigned instead of int? I don't
> see any use for negative values.
-1 is used to disable the feature. I will update the parameter description.
Christian.
> Alex
>
>> +
>> MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
>> module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> index 03c6bfc..d6e6c93 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> @@ -970,6 +970,7 @@ u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
>> WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
>> WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_VRAM &&
>> !(bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS));
>> + WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
>>
>> return bo->tbo.offset;
>> }
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> index 4b8b39c..428ffb6 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> @@ -168,7 +168,7 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
>> break;
>> case TTM_PL_VRAM:
>> /* "On-card" video ram */
>> - man->func = &ttm_bo_manager_func;
>> + man->func = &amdgpu_vram_mgr_func;
>> man->gpu_offset = adev->mc.vram_start;
>> man->flags = TTM_MEMTYPE_FLAG_FIXED |
>> TTM_MEMTYPE_FLAG_MAPPABLE;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> index 9812c80..d1c00c0 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> @@ -66,6 +66,7 @@ struct amdgpu_mman {
>> };
>>
>> extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
>> +extern const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func;
>>
>> int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
>> struct ttm_buffer_object *tbo,
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>> new file mode 100644
>> index 0000000..1125f7f
>> --- /dev/null
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>> @@ -0,0 +1,222 @@
>> +/*
>> + * Copyright 2015 Advanced Micro Devices, Inc.
>> + *
>> + * Permission is hereby granted, free of charge, to any person obtaining a
>> + * copy of this software and associated documentation files (the "Software"),
>> + * to deal in the Software without restriction, including without limitation
>> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
>> + * and/or sell copies of the Software, and to permit persons to whom the
>> + * Software is furnished to do so, subject to the following conditions:
>> + *
>> + * The above copyright notice and this permission notice shall be included in
>> + * all copies or substantial portions of the Software.
>> + *
>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
>> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
>> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
>> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
>> + * OTHER DEALINGS IN THE SOFTWARE.
>> + *
>> + * Authors: Christian König
>> + */
>> +
>> +#include <drm/drmP.h>
>> +#include "amdgpu.h"
>> +
>> +struct amdgpu_vram_mgr {
>> + struct drm_mm mm;
>> + spinlock_t lock;
>> +};
>> +
>> +/**
>> + * amdgpu_vram_mgr_init - init VRAM manager and DRM MM
>> + *
>> + * @man: TTM memory type manager
>> + * @p_size: maximum size of VRAM
>> + *
>> + * Allocate and initialize the VRAM manager.
>> + */
>> +static int amdgpu_vram_mgr_init(struct ttm_mem_type_manager *man,
>> + unsigned long p_size)
>> +{
>> + struct amdgpu_vram_mgr *mgr;
>> +
>> + mgr = kzalloc(sizeof(*mgr), GFP_KERNEL);
>> + if (!mgr)
>> + return -ENOMEM;
>> +
>> + drm_mm_init(&mgr->mm, 0, p_size);
>> + spin_lock_init(&mgr->lock);
>> + man->priv = mgr;
>> + return 0;
>> +}
>> +
>> +/**
>> + * amdgpu_vram_mgr_fini - free and destroy VRAM manager
>> + *
>> + * @man: TTM memory type manager
>> + *
>> + * Destroy and free the VRAM manager, returns -EBUSY if ranges are still
>> + * allocated inside it.
>> + */
>> +static int amdgpu_vram_mgr_fini(struct ttm_mem_type_manager *man)
>> +{
>> + struct amdgpu_vram_mgr *mgr = man->priv;
>> +
>> + spin_lock(&mgr->lock);
>> + if (!drm_mm_clean(&mgr->mm)) {
>> + spin_unlock(&mgr->lock);
>> + return -EBUSY;
>> + }
>> +
>> + drm_mm_takedown(&mgr->mm);
>> + spin_unlock(&mgr->lock);
>> + kfree(mgr);
>> + man->priv = NULL;
>> + return 0;
>> +}
>> +
>> +/**
>> + * amdgpu_vram_mgr_new - allocate new ranges
>> + *
>> + * @man: TTM memory type manager
>> + * @tbo: TTM BO we need this range for
>> + * @place: placement flags and restrictions
>> + * @mem: the resulting mem object
>> + *
>> + * Allocate VRAM for the given BO.
>> + */
>> +static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
>> + struct ttm_buffer_object *tbo,
>> + const struct ttm_place *place,
>> + struct ttm_mem_reg *mem)
>> +{
>> + struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo);
>> + struct amdgpu_vram_mgr *mgr = man->priv;
>> + struct drm_mm *mm = &mgr->mm;
>> + struct drm_mm_node *nodes;
>> + enum drm_mm_search_flags sflags = DRM_MM_SEARCH_DEFAULT;
>> + enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
>> + unsigned long lpfn, num_nodes, pages_per_node, pages_left;
>> + unsigned i;
>> + int r;
>> +
>> + lpfn = place->lpfn;
>> + if (!lpfn)
>> + lpfn = man->size;
>> +
>> + if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS ||
>> + amdgpu_vram_page_split == -1) {
>> + pages_per_node = ~0ul;
>> + num_nodes = 1;
>> + } else {
>> + pages_per_node = max((uint32_t)amdgpu_vram_page_split,
>> + mem->page_alignment);
>> + num_nodes = DIV_ROUND_UP(mem->num_pages, pages_per_node);
>> + }
>> +
>> + nodes = kcalloc(num_nodes, sizeof(*nodes), GFP_KERNEL);
>> + if (!nodes)
>> + return -ENOMEM;
>> +
>> + if (place->flags & TTM_PL_FLAG_TOPDOWN) {
>> + sflags = DRM_MM_SEARCH_BELOW;
>> + aflags = DRM_MM_CREATE_TOP;
>> + }
>> +
>> + pages_left = mem->num_pages;
>> +
>> + spin_lock(&mgr->lock);
>> + for (i = 0; i < num_nodes; ++i) {
>> + unsigned long pages = min(pages_left, pages_per_node);
>> + uint32_t alignment = mem->page_alignment;
>> +
>> + if (pages == pages_per_node)
>> + alignment = pages_per_node;
>> + else
>> + sflags |= DRM_MM_SEARCH_BEST;
>> +
>> + r = drm_mm_insert_node_in_range_generic(mm, &nodes[i], pages,
>> + alignment, 0,
>> + place->fpfn, lpfn,
>> + sflags, aflags);
>> + if (unlikely(r))
>> + goto error;
>> +
>> + pages_left -= pages;
>> + }
>> + spin_unlock(&mgr->lock);
>> +
>> + mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET;
>> + mem->mm_node = nodes;
>> +
>> + return 0;
>> +
>> +error:
>> + while (i--)
>> + drm_mm_remove_node(&nodes[i]);
>> + spin_unlock(&mgr->lock);
>> +
>> + kfree(nodes);
>> + return r == -ENOSPC ? 0 : r;
>> +}
>> +
>> +/**
>> + * amdgpu_vram_mgr_del - free ranges
>> + *
>> + * @man: TTM memory type manager
>> + * @tbo: TTM BO we need this range for
>> + * @place: placement flags and restrictions
>> + * @mem: TTM memory object
>> + *
>> + * Free the allocated VRAM again.
>> + */
>> +static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
>> + struct ttm_mem_reg *mem)
>> +{
>> + struct amdgpu_vram_mgr *mgr = man->priv;
>> + struct drm_mm_node *nodes = mem->mm_node;
>> + unsigned pages = mem->num_pages;
>> +
>> + if (!mem->mm_node)
>> + return;
>> +
>> + spin_lock(&mgr->lock);
>> + while (pages) {
>> + pages -= nodes->size;
>> + drm_mm_remove_node(nodes);
>> + ++nodes;
>> + }
>> + spin_unlock(&mgr->lock);
>> +
>> + kfree(mem->mm_node);
>> + mem->mm_node = NULL;
>> +}
>> +
>> +/**
>> + * amdgpu_vram_mgr_debug - dump VRAM table
>> + *
>> + * @man: TTM memory type manager
>> + * @prefix: text prefix
>> + *
>> + * Dump the table content using printk.
>> + */
>> +static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
>> + const char *prefix)
>> +{
>> + struct amdgpu_vram_mgr *mgr = man->priv;
>> +
>> + spin_lock(&mgr->lock);
>> + drm_mm_debug_table(&mgr->mm, prefix);
>> + spin_unlock(&mgr->lock);
>> +}
>> +
>> +const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func = {
>> + amdgpu_vram_mgr_init,
>> + amdgpu_vram_mgr_fini,
>> + amdgpu_vram_mgr_new,
>> + amdgpu_vram_mgr_del,
>> + amdgpu_vram_mgr_debug
>> +};
>> --
>> 2.5.0
>>
>> _______________________________________________
>> amd-gfx mailing list
>> amd-gfx@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH 6/6] drm/amdgpu: add VRAM manager v2
[not found] ` <1473934344-2106-6-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
@ 2016-09-15 14:44 ` Alex Deucher
[not found] ` <CADnq5_NSW3A_7EqUzABsEUddyJYUKyfpUR2fW-Ku4x3VF0Rykw-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
0 siblings, 1 reply; 18+ messages in thread
From: Alex Deucher @ 2016-09-15 14:44 UTC (permalink / raw)
To: Christian König; +Cc: amd-gfx list
On Thu, Sep 15, 2016 at 6:12 AM, Christian König
<deathsimple@vodafone.de> wrote:
> From: Christian König <christian.koenig@amd.com>
>
> Split VRAM allocations into 4MB blocks.
>
> v2: fix typo in comment, some suggested cleanups
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/Makefile | 2 +-
> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +-
> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 222 +++++++++++++++++++++++++++
> 8 files changed, 238 insertions(+), 2 deletions(-)
> create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
> index f2b97cb..236e9df 100644
> --- a/drivers/gpu/drm/amd/amdgpu/Makefile
> +++ b/drivers/gpu/drm/amd/amdgpu/Makefile
> @@ -30,7 +30,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
> atombios_encoders.o amdgpu_sa.o atombios_i2c.o \
> amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \
> amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
> - amdgpu_gtt_mgr.o
> + amdgpu_gtt_mgr.o amdgpu_vram_mgr.o
>
> # add asic specific block
> amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> index 588baaf..ee55763 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> @@ -98,6 +98,7 @@ extern char *amdgpu_disable_cu;
> extern int amdgpu_sclk_deep_sleep_en;
> extern char *amdgpu_virtual_display;
> extern unsigned amdgpu_pp_feature_mask;
> +extern int amdgpu_vram_page_split;
>
> #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000
> #define AMDGPU_MAX_USEC_TIMEOUT 100000 /* 100 ms */
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 5686d12..f26b067 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -1012,6 +1012,13 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
> amdgpu_vm_block_size);
> amdgpu_vm_block_size = 9;
> }
> +
> + if ((amdgpu_vram_page_split != -1 && amdgpu_vram_page_split < 16) ||
> + !amdgpu_check_pot_argument(amdgpu_vram_page_split)) {
> + dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
> + amdgpu_vram_page_split);
> + amdgpu_vram_page_split = 1024;
> + }
> }
>
> /**
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> index 902da13..44e605d 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> @@ -83,6 +83,7 @@ int amdgpu_vm_size = 64;
> int amdgpu_vm_block_size = -1;
> int amdgpu_vm_fault_stop = 0;
> int amdgpu_vm_debug = 0;
> +int amdgpu_vram_page_split = 1024;
> int amdgpu_exp_hw_support = 0;
> int amdgpu_dal = -1;
> int amdgpu_sched_jobs = 32;
> @@ -164,6 +165,9 @@ module_param_named(vm_fault_stop, amdgpu_vm_fault_stop, int, 0444);
> MODULE_PARM_DESC(vm_debug, "Debug VM handling (0 = disabled (default), 1 = enabled)");
> module_param_named(vm_debug, amdgpu_vm_debug, int, 0644);
>
> +MODULE_PARM_DESC(vram_page_split, "Number of pages after we split VRAM allocations (default 1024)");
> +module_param_named(vram_page_split, amdgpu_vram_page_split, int, 0444);
Not a big deal, but maybe make this unsigned instead of int? I don't
see any use for negative values.
Alex
> +
> MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
> module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index 03c6bfc..d6e6c93 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -970,6 +970,7 @@ u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
> WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
> WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_VRAM &&
> !(bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS));
> + WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
>
> return bo->tbo.offset;
> }
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index 4b8b39c..428ffb6 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -168,7 +168,7 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
> break;
> case TTM_PL_VRAM:
> /* "On-card" video ram */
> - man->func = &ttm_bo_manager_func;
> + man->func = &amdgpu_vram_mgr_func;
> man->gpu_offset = adev->mc.vram_start;
> man->flags = TTM_MEMTYPE_FLAG_FIXED |
> TTM_MEMTYPE_FLAG_MAPPABLE;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
> index 9812c80..d1c00c0 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
> @@ -66,6 +66,7 @@ struct amdgpu_mman {
> };
>
> extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
> +extern const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func;
>
> int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
> struct ttm_buffer_object *tbo,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
> new file mode 100644
> index 0000000..1125f7f
> --- /dev/null
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
> @@ -0,0 +1,222 @@
> +/*
> + * Copyright 2015 Advanced Micro Devices, Inc.
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
> + * OTHER DEALINGS IN THE SOFTWARE.
> + *
> + * Authors: Christian König
> + */
> +
> +#include <drm/drmP.h>
> +#include "amdgpu.h"
> +
> +struct amdgpu_vram_mgr {
> + struct drm_mm mm;
> + spinlock_t lock;
> +};
> +
> +/**
> + * amdgpu_vram_mgr_init - init VRAM manager and DRM MM
> + *
> + * @man: TTM memory type manager
> + * @p_size: maximum size of VRAM
> + *
> + * Allocate and initialize the VRAM manager.
> + */
> +static int amdgpu_vram_mgr_init(struct ttm_mem_type_manager *man,
> + unsigned long p_size)
> +{
> + struct amdgpu_vram_mgr *mgr;
> +
> + mgr = kzalloc(sizeof(*mgr), GFP_KERNEL);
> + if (!mgr)
> + return -ENOMEM;
> +
> + drm_mm_init(&mgr->mm, 0, p_size);
> + spin_lock_init(&mgr->lock);
> + man->priv = mgr;
> + return 0;
> +}
> +
> +/**
> + * amdgpu_vram_mgr_fini - free and destroy VRAM manager
> + *
> + * @man: TTM memory type manager
> + *
> + * Destroy and free the VRAM manager, returns -EBUSY if ranges are still
> + * allocated inside it.
> + */
> +static int amdgpu_vram_mgr_fini(struct ttm_mem_type_manager *man)
> +{
> + struct amdgpu_vram_mgr *mgr = man->priv;
> +
> + spin_lock(&mgr->lock);
> + if (!drm_mm_clean(&mgr->mm)) {
> + spin_unlock(&mgr->lock);
> + return -EBUSY;
> + }
> +
> + drm_mm_takedown(&mgr->mm);
> + spin_unlock(&mgr->lock);
> + kfree(mgr);
> + man->priv = NULL;
> + return 0;
> +}
> +
> +/**
> + * amdgpu_vram_mgr_new - allocate new ranges
> + *
> + * @man: TTM memory type manager
> + * @tbo: TTM BO we need this range for
> + * @place: placement flags and restrictions
> + * @mem: the resulting mem object
> + *
> + * Allocate VRAM for the given BO.
> + */
> +static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
> + struct ttm_buffer_object *tbo,
> + const struct ttm_place *place,
> + struct ttm_mem_reg *mem)
> +{
> + struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo);
> + struct amdgpu_vram_mgr *mgr = man->priv;
> + struct drm_mm *mm = &mgr->mm;
> + struct drm_mm_node *nodes;
> + enum drm_mm_search_flags sflags = DRM_MM_SEARCH_DEFAULT;
> + enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
> + unsigned long lpfn, num_nodes, pages_per_node, pages_left;
> + unsigned i;
> + int r;
> +
> + lpfn = place->lpfn;
> + if (!lpfn)
> + lpfn = man->size;
> +
> + if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS ||
> + amdgpu_vram_page_split == -1) {
> + pages_per_node = ~0ul;
> + num_nodes = 1;
> + } else {
> + pages_per_node = max((uint32_t)amdgpu_vram_page_split,
> + mem->page_alignment);
> + num_nodes = DIV_ROUND_UP(mem->num_pages, pages_per_node);
> + }
> +
> + nodes = kcalloc(num_nodes, sizeof(*nodes), GFP_KERNEL);
> + if (!nodes)
> + return -ENOMEM;
> +
> + if (place->flags & TTM_PL_FLAG_TOPDOWN) {
> + sflags = DRM_MM_SEARCH_BELOW;
> + aflags = DRM_MM_CREATE_TOP;
> + }
> +
> + pages_left = mem->num_pages;
> +
> + spin_lock(&mgr->lock);
> + for (i = 0; i < num_nodes; ++i) {
> + unsigned long pages = min(pages_left, pages_per_node);
> + uint32_t alignment = mem->page_alignment;
> +
> + if (pages == pages_per_node)
> + alignment = pages_per_node;
> + else
> + sflags |= DRM_MM_SEARCH_BEST;
> +
> + r = drm_mm_insert_node_in_range_generic(mm, &nodes[i], pages,
> + alignment, 0,
> + place->fpfn, lpfn,
> + sflags, aflags);
> + if (unlikely(r))
> + goto error;
> +
> + pages_left -= pages;
> + }
> + spin_unlock(&mgr->lock);
> +
> + mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET;
> + mem->mm_node = nodes;
> +
> + return 0;
> +
> +error:
> + while (i--)
> + drm_mm_remove_node(&nodes[i]);
> + spin_unlock(&mgr->lock);
> +
> + kfree(nodes);
> + return r == -ENOSPC ? 0 : r;
> +}
> +
> +/**
> + * amdgpu_vram_mgr_del - free ranges
> + *
> + * @man: TTM memory type manager
> + * @tbo: TTM BO we need this range for
> + * @place: placement flags and restrictions
> + * @mem: TTM memory object
> + *
> + * Free the allocated VRAM again.
> + */
> +static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
> + struct ttm_mem_reg *mem)
> +{
> + struct amdgpu_vram_mgr *mgr = man->priv;
> + struct drm_mm_node *nodes = mem->mm_node;
> + unsigned pages = mem->num_pages;
> +
> + if (!mem->mm_node)
> + return;
> +
> + spin_lock(&mgr->lock);
> + while (pages) {
> + pages -= nodes->size;
> + drm_mm_remove_node(nodes);
> + ++nodes;
> + }
> + spin_unlock(&mgr->lock);
> +
> + kfree(mem->mm_node);
> + mem->mm_node = NULL;
> +}
> +
> +/**
> + * amdgpu_vram_mgr_debug - dump VRAM table
> + *
> + * @man: TTM memory type manager
> + * @prefix: text prefix
> + *
> + * Dump the table content using printk.
> + */
> +static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
> + const char *prefix)
> +{
> + struct amdgpu_vram_mgr *mgr = man->priv;
> +
> + spin_lock(&mgr->lock);
> + drm_mm_debug_table(&mgr->mm, prefix);
> + spin_unlock(&mgr->lock);
> +}
> +
> +const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func = {
> + amdgpu_vram_mgr_init,
> + amdgpu_vram_mgr_fini,
> + amdgpu_vram_mgr_new,
> + amdgpu_vram_mgr_del,
> + amdgpu_vram_mgr_debug
> +};
> --
> 2.5.0
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 18+ messages in thread
* [PATCH 6/6] drm/amdgpu: add VRAM manager v2
[not found] ` <1473934344-2106-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
@ 2016-09-15 10:12 ` Christian König
[not found] ` <1473934344-2106-6-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
0 siblings, 1 reply; 18+ messages in thread
From: Christian König @ 2016-09-15 10:12 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Christian König <christian.koenig@amd.com>
Split VRAM allocations into 4MB blocks.
v2: fix typo in comment, some suggested cleanups
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/Makefile | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 +
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 222 +++++++++++++++++++++++++++
8 files changed, 238 insertions(+), 2 deletions(-)
create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index f2b97cb..236e9df 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -30,7 +30,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
atombios_encoders.o amdgpu_sa.o atombios_i2c.o \
amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \
amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
- amdgpu_gtt_mgr.o
+ amdgpu_gtt_mgr.o amdgpu_vram_mgr.o
# add asic specific block
amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 588baaf..ee55763 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -98,6 +98,7 @@ extern char *amdgpu_disable_cu;
extern int amdgpu_sclk_deep_sleep_en;
extern char *amdgpu_virtual_display;
extern unsigned amdgpu_pp_feature_mask;
+extern int amdgpu_vram_page_split;
#define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000
#define AMDGPU_MAX_USEC_TIMEOUT 100000 /* 100 ms */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 5686d12..f26b067 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -1012,6 +1012,13 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
amdgpu_vm_block_size);
amdgpu_vm_block_size = 9;
}
+
+ if ((amdgpu_vram_page_split != -1 && amdgpu_vram_page_split < 16) ||
+ !amdgpu_check_pot_argument(amdgpu_vram_page_split)) {
+ dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
+ amdgpu_vram_page_split);
+ amdgpu_vram_page_split = 1024;
+ }
}
/**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 902da13..44e605d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -83,6 +83,7 @@ int amdgpu_vm_size = 64;
int amdgpu_vm_block_size = -1;
int amdgpu_vm_fault_stop = 0;
int amdgpu_vm_debug = 0;
+int amdgpu_vram_page_split = 1024;
int amdgpu_exp_hw_support = 0;
int amdgpu_dal = -1;
int amdgpu_sched_jobs = 32;
@@ -164,6 +165,9 @@ module_param_named(vm_fault_stop, amdgpu_vm_fault_stop, int, 0444);
MODULE_PARM_DESC(vm_debug, "Debug VM handling (0 = disabled (default), 1 = enabled)");
module_param_named(vm_debug, amdgpu_vm_debug, int, 0644);
+MODULE_PARM_DESC(vram_page_split, "Number of pages after we split VRAM allocations (default 1024)");
+module_param_named(vram_page_split, amdgpu_vram_page_split, int, 0444);
+
MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 03c6bfc..d6e6c93 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -970,6 +970,7 @@ u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_VRAM &&
!(bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS));
+ WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
return bo->tbo.offset;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 4b8b39c..428ffb6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -168,7 +168,7 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
break;
case TTM_PL_VRAM:
/* "On-card" video ram */
- man->func = &ttm_bo_manager_func;
+ man->func = &amdgpu_vram_mgr_func;
man->gpu_offset = adev->mc.vram_start;
man->flags = TTM_MEMTYPE_FLAG_FIXED |
TTM_MEMTYPE_FLAG_MAPPABLE;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 9812c80..d1c00c0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -66,6 +66,7 @@ struct amdgpu_mman {
};
extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;
+extern const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func;
int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
struct ttm_buffer_object *tbo,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
new file mode 100644
index 0000000..1125f7f
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
@@ -0,0 +1,222 @@
+/*
+ * Copyright 2015 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Christian König
+ */
+
+#include <drm/drmP.h>
+#include "amdgpu.h"
+
+struct amdgpu_vram_mgr {
+ struct drm_mm mm;
+ spinlock_t lock;
+};
+
+/**
+ * amdgpu_vram_mgr_init - init VRAM manager and DRM MM
+ *
+ * @man: TTM memory type manager
+ * @p_size: maximum size of VRAM
+ *
+ * Allocate and initialize the VRAM manager.
+ */
+static int amdgpu_vram_mgr_init(struct ttm_mem_type_manager *man,
+ unsigned long p_size)
+{
+ struct amdgpu_vram_mgr *mgr;
+
+ mgr = kzalloc(sizeof(*mgr), GFP_KERNEL);
+ if (!mgr)
+ return -ENOMEM;
+
+ drm_mm_init(&mgr->mm, 0, p_size);
+ spin_lock_init(&mgr->lock);
+ man->priv = mgr;
+ return 0;
+}
+
+/**
+ * amdgpu_vram_mgr_fini - free and destroy VRAM manager
+ *
+ * @man: TTM memory type manager
+ *
+ * Destroy and free the VRAM manager, returns -EBUSY if ranges are still
+ * allocated inside it.
+ */
+static int amdgpu_vram_mgr_fini(struct ttm_mem_type_manager *man)
+{
+ struct amdgpu_vram_mgr *mgr = man->priv;
+
+ spin_lock(&mgr->lock);
+ if (!drm_mm_clean(&mgr->mm)) {
+ spin_unlock(&mgr->lock);
+ return -EBUSY;
+ }
+
+ drm_mm_takedown(&mgr->mm);
+ spin_unlock(&mgr->lock);
+ kfree(mgr);
+ man->priv = NULL;
+ return 0;
+}
+
+/**
+ * amdgpu_vram_mgr_new - allocate new ranges
+ *
+ * @man: TTM memory type manager
+ * @tbo: TTM BO we need this range for
+ * @place: placement flags and restrictions
+ * @mem: the resulting mem object
+ *
+ * Allocate VRAM for the given BO.
+ */
+static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
+ struct ttm_buffer_object *tbo,
+ const struct ttm_place *place,
+ struct ttm_mem_reg *mem)
+{
+ struct amdgpu_bo *bo = container_of(tbo, struct amdgpu_bo, tbo);
+ struct amdgpu_vram_mgr *mgr = man->priv;
+ struct drm_mm *mm = &mgr->mm;
+ struct drm_mm_node *nodes;
+ enum drm_mm_search_flags sflags = DRM_MM_SEARCH_DEFAULT;
+ enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
+ unsigned long lpfn, num_nodes, pages_per_node, pages_left;
+ unsigned i;
+ int r;
+
+ lpfn = place->lpfn;
+ if (!lpfn)
+ lpfn = man->size;
+
+ if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS ||
+ amdgpu_vram_page_split == -1) {
+ pages_per_node = ~0ul;
+ num_nodes = 1;
+ } else {
+ pages_per_node = max((uint32_t)amdgpu_vram_page_split,
+ mem->page_alignment);
+ num_nodes = DIV_ROUND_UP(mem->num_pages, pages_per_node);
+ }
+
+ nodes = kcalloc(num_nodes, sizeof(*nodes), GFP_KERNEL);
+ if (!nodes)
+ return -ENOMEM;
+
+ if (place->flags & TTM_PL_FLAG_TOPDOWN) {
+ sflags = DRM_MM_SEARCH_BELOW;
+ aflags = DRM_MM_CREATE_TOP;
+ }
+
+ pages_left = mem->num_pages;
+
+ spin_lock(&mgr->lock);
+ for (i = 0; i < num_nodes; ++i) {
+ unsigned long pages = min(pages_left, pages_per_node);
+ uint32_t alignment = mem->page_alignment;
+
+ if (pages == pages_per_node)
+ alignment = pages_per_node;
+ else
+ sflags |= DRM_MM_SEARCH_BEST;
+
+ r = drm_mm_insert_node_in_range_generic(mm, &nodes[i], pages,
+ alignment, 0,
+ place->fpfn, lpfn,
+ sflags, aflags);
+ if (unlikely(r))
+ goto error;
+
+ pages_left -= pages;
+ }
+ spin_unlock(&mgr->lock);
+
+ mem->start = num_nodes == 1 ? nodes[0].start : AMDGPU_BO_INVALID_OFFSET;
+ mem->mm_node = nodes;
+
+ return 0;
+
+error:
+ while (i--)
+ drm_mm_remove_node(&nodes[i]);
+ spin_unlock(&mgr->lock);
+
+ kfree(nodes);
+ return r == -ENOSPC ? 0 : r;
+}
+
+/**
+ * amdgpu_vram_mgr_del - free ranges
+ *
+ * @man: TTM memory type manager
+ * @tbo: TTM BO we need this range for
+ * @place: placement flags and restrictions
+ * @mem: TTM memory object
+ *
+ * Free the allocated VRAM again.
+ */
+static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
+ struct ttm_mem_reg *mem)
+{
+ struct amdgpu_vram_mgr *mgr = man->priv;
+ struct drm_mm_node *nodes = mem->mm_node;
+ unsigned pages = mem->num_pages;
+
+ if (!mem->mm_node)
+ return;
+
+ spin_lock(&mgr->lock);
+ while (pages) {
+ pages -= nodes->size;
+ drm_mm_remove_node(nodes);
+ ++nodes;
+ }
+ spin_unlock(&mgr->lock);
+
+ kfree(mem->mm_node);
+ mem->mm_node = NULL;
+}
+
+/**
+ * amdgpu_vram_mgr_debug - dump VRAM table
+ *
+ * @man: TTM memory type manager
+ * @prefix: text prefix
+ *
+ * Dump the table content using printk.
+ */
+static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
+ const char *prefix)
+{
+ struct amdgpu_vram_mgr *mgr = man->priv;
+
+ spin_lock(&mgr->lock);
+ drm_mm_debug_table(&mgr->mm, prefix);
+ spin_unlock(&mgr->lock);
+}
+
+const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func = {
+ amdgpu_vram_mgr_init,
+ amdgpu_vram_mgr_fini,
+ amdgpu_vram_mgr_new,
+ amdgpu_vram_mgr_del,
+ amdgpu_vram_mgr_debug
+};
--
2.5.0
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 18+ messages in thread
end of thread, other threads:[~2016-10-11 18:59 UTC | newest]
Thread overview: 18+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-09-27 9:49 VRAM manager Christian König
[not found] ` <1474969797-1882-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-09-27 9:49 ` [PATCH 1/6] drm/amdgpu: add AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS flag v3 Christian König
[not found] ` <1474969797-1882-2-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-10-07 21:19 ` Felix Kuehling
[not found] ` <627af621-8fb2-ba41-c87e-96bb12e594d8-5C7GfCeVMHo@public.gmane.org>
2016-10-11 9:35 ` Christian König
2016-09-27 9:49 ` [PATCH 2/6] drm/amdgpu: use explicit limit for VRAM_CONTIGUOUS Christian König
2016-09-27 9:49 ` [PATCH 3/6] drm/amdgpu: set at least the node size in the gtt manager Christian König
2016-09-27 9:49 ` [PATCH 4/6] drm/amdgpu: handle multiple MM nodes in the VMs v2 Christian König
2016-09-27 9:49 ` [PATCH 5/6] drm/amdgpu: enable amdgpu_move_blit to handle multiple MM nodes v2 Christian König
2016-09-27 9:49 ` [PATCH 6/6] drm/amdgpu: add VRAM manager v2 Christian König
[not found] ` <1474969797-1882-7-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-09-27 9:57 ` zhoucm1
2016-09-27 10:50 ` VRAM manager Mike Lothian
2016-09-27 11:17 ` Edward O'Callaghan
2016-10-07 22:11 ` Felix Kuehling
[not found] ` <ad665321-d00c-1ded-051f-2ca8eae014b4-5C7GfCeVMHo@public.gmane.org>
2016-10-11 9:34 ` Christian König
[not found] ` <6235e143-b44a-0039-b84c-c1be81de4a4c-5C7GfCeVMHo@public.gmane.org>
2016-10-11 18:59 ` Felix Kuehling
-- strict thread matches above, loose matches on Subject: below --
2016-09-15 10:12 [PATCH 1/6] drm/amdgpu: add AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS flag v3 Christian König
[not found] ` <1473934344-2106-1-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-09-15 10:12 ` [PATCH 6/6] drm/amdgpu: add VRAM manager v2 Christian König
[not found] ` <1473934344-2106-6-git-send-email-deathsimple-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org>
2016-09-15 14:44 ` Alex Deucher
[not found] ` <CADnq5_NSW3A_7EqUzABsEUddyJYUKyfpUR2fW-Ku4x3VF0Rykw-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2016-09-15 17:42 ` Christian König
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.