[RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Nirmoy Das]

With this patch series I am trying to remove GPU address dependency in
TTM and moving GPU address calculation to individual drm drivers. This
cleanup will simplify introduction of drm_mem_region/domain work started
by Brian Welty[1].


It would be nice if someone test this for nouveau. Rest of the drivers
are already tested.

v2:
* set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL

v3:
* catch return value of drm_gem_vram_offset() in drm/bochs
* introduce drm_gem_vram_pg_offset() in vram helper
* improve nbo->offset calculation for nouveau

v4:
* minor coding style fixes in amdgpu and radeon
* remove unnecessary kerneldoc for internal function

v5:
* rebase on top of drm-misc-next
* fix return value of drm_gem_vram_pg_offset()
* add a comment in drm_gem_vram_pg_offset() to clearify why we return 0.

v6:
* rebase to drm-misc-next
* removed acked for vmwgfx as there was a small conflict

Nirmoy Das (8):
  drm/amdgpu: move ttm bo->offset to amdgpu_bo
  drm/radeon: don't use ttm bo->offset
  drm/vmwgfx: don't use ttm bo->offset
  drm/nouveau: don't use ttm bo->offset v3
  drm/qxl: don't use ttm bo->offset
  drm/vram-helper: don't use ttm bo->offset v4
  drm/bochs: use drm_gem_vram_offset to get bo offset v2
  drm/ttm: do not keep GPU dependent addresses

 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 ++++++++++++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
 drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
 drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
 drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
 drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
 drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
 drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
 drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
 drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
 drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
 drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
 drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
 drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
 drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
 drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
 drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
 drivers/gpu/drm/radeon/radeon.h             |  1 +
 drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
 drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
 drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
 drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
 include/drm/ttm/ttm_bo_api.h                |  2 --
 include/drm/ttm/ttm_bo_driver.h             |  1 -
 36 files changed, 125 insertions(+), 78 deletions(-)

--
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 1/8] drm/amdgpu: move ttm bo->offset to amdgpu_bo

[Nirmoy Das]

GPU address should belong to driver not in memory management.
This patch moves ttm bo.offset and gpu_offset calculation to amdgpu driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Acked-by: Huang Rui <ray.huang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 ++++++++++++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
 5 files changed, 48 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index c687f5415b3f..a27b6a692119 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -918,7 +918,8 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
 		bo->pin_count++;
 
 		if (max_offset != 0) {
-			u64 domain_start = bo->tbo.bdev->man[mem_type].gpu_offset;
+			u64 domain_start = amdgpu_ttm_domain_start(adev,
+								   mem_type);
 			WARN_ON_ONCE(max_offset <
 				     (amdgpu_bo_gpu_offset(bo) - domain_start));
 		}
@@ -1484,7 +1485,25 @@ u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
 	WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_VRAM &&
 		     !(bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS));
 
-	return amdgpu_gmc_sign_extend(bo->tbo.offset);
+	return amdgpu_bo_gpu_offset_no_check(bo);
+}
+
+/**
+ * amdgpu_bo_gpu_offset_no_check - return GPU offset of bo
+ * @bo:	amdgpu object for which we query the offset
+ *
+ * Returns:
+ * current GPU offset of the object without raising warnings.
+ */
+u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	uint64_t offset;
+
+	offset = (bo->tbo.mem.start << PAGE_SHIFT) +
+		 amdgpu_ttm_domain_start(adev, bo->tbo.mem.mem_type);
+
+	return amdgpu_gmc_sign_extend(offset);
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 5e39ecd8cc28..32edd35d2ccf 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -282,6 +282,7 @@ int amdgpu_bo_sync_wait_resv(struct amdgpu_device *adev, struct dma_resv *resv,
 			     bool intr);
 int amdgpu_bo_sync_wait(struct amdgpu_bo *bo, void *owner, bool intr);
 u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
+u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo);
 int amdgpu_bo_validate(struct amdgpu_bo *bo);
 int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow,
 			     struct dma_fence **fence);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 6309ff72bd78..c2e13e1ec53e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -96,7 +96,6 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	case TTM_PL_TT:
 		/* GTT memory  */
 		man->func = &amdgpu_gtt_mgr_func;
-		man->gpu_offset = adev->gmc.gart_start;
 		man->available_caching = TTM_PL_MASK_CACHING;
 		man->default_caching = TTM_PL_FLAG_CACHED;
 		man->flags = TTM_MEMTYPE_FLAG_MAPPABLE | TTM_MEMTYPE_FLAG_CMA;
@@ -104,7 +103,6 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	case TTM_PL_VRAM:
 		/* "On-card" video ram */
 		man->func = &amdgpu_vram_mgr_func;
-		man->gpu_offset = adev->gmc.vram_start;
 		man->flags = TTM_MEMTYPE_FLAG_FIXED |
 			     TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC;
@@ -115,7 +113,6 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	case AMDGPU_PL_OA:
 		/* On-chip GDS memory*/
 		man->func = &ttm_bo_manager_func;
-		man->gpu_offset = 0;
 		man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_CMA;
 		man->available_caching = TTM_PL_FLAG_UNCACHED;
 		man->default_caching = TTM_PL_FLAG_UNCACHED;
@@ -263,7 +260,8 @@ static uint64_t amdgpu_mm_node_addr(struct ttm_buffer_object *bo,
 
 	if (mm_node->start != AMDGPU_BO_INVALID_OFFSET) {
 		addr = mm_node->start << PAGE_SHIFT;
-		addr += bo->bdev->man[mem->mem_type].gpu_offset;
+		addr += amdgpu_ttm_domain_start(amdgpu_ttm_adev(bo->bdev),
+						mem->mem_type);
 	}
 	return addr;
 }
@@ -750,6 +748,27 @@ static unsigned long amdgpu_ttm_io_mem_pfn(struct ttm_buffer_object *bo,
 		(offset >> PAGE_SHIFT);
 }
 
+/**
+ * amdgpu_ttm_domain_start - Returns GPU start address
+ * @adev: amdgpu device object
+ * @type: type of the memory
+ *
+ * Returns:
+ * GPU start address of a memory domain
+ */
+
+uint64_t amdgpu_ttm_domain_start(struct amdgpu_device *adev, uint32_t type)
+{
+	switch (type) {
+	case TTM_PL_TT:
+		return adev->gmc.gart_start;
+	case TTM_PL_VRAM:
+		return adev->gmc.vram_start;
+	}
+
+	return 0;
+}
+
 /*
  * TTM backend functions.
  */
@@ -1163,9 +1182,6 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
 		bo->mem = tmp;
 	}
 
-	bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		bo->bdev->man[bo->mem.mem_type].gpu_offset;
-
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 6b22dc41ef13..b6086ab21069 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -112,6 +112,7 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
 int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma);
 int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
 int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo);
+uint64_t amdgpu_ttm_domain_start(struct amdgpu_device *adev, uint32_t type);
 
 #if IS_ENABLED(CONFIG_DRM_AMDGPU_USERPTR)
 int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, struct page **pages);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c
index cf96c335b258..2a7a6f62d627 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c
@@ -141,7 +141,7 @@ static void amdgpu_vm_sdma_copy_ptes(struct amdgpu_vm_update_params *p,
 
 	src += p->num_dw_left * 4;
 
-	pe += amdgpu_gmc_sign_extend(bo->tbo.offset);
+	pe += amdgpu_bo_gpu_offset_no_check(bo);
 	trace_amdgpu_vm_copy_ptes(pe, src, count, p->direct);
 
 	amdgpu_vm_copy_pte(p->adev, ib, pe, src, count);
@@ -168,7 +168,7 @@ static void amdgpu_vm_sdma_set_ptes(struct amdgpu_vm_update_params *p,
 {
 	struct amdgpu_ib *ib = p->job->ibs;
 
-	pe += amdgpu_gmc_sign_extend(bo->tbo.offset);
+	pe += amdgpu_bo_gpu_offset_no_check(bo);
 	trace_amdgpu_vm_set_ptes(pe, addr, count, incr, flags, p->direct);
 	if (count < 3) {
 		amdgpu_vm_write_pte(p->adev, ib, pe, addr | flags,
-- 
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 2/8] drm/radeon: don't use ttm bo->offset

[Nirmoy Das]

Calculate GPU offset in radeon_bo_gpu_offset without depending on
bo->offset.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-and-tested-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/radeon/radeon.h        |  1 +
 drivers/gpu/drm/radeon/radeon_object.h | 16 +++++++++++++++-
 drivers/gpu/drm/radeon/radeon_ttm.c    |  4 +---
 3 files changed, 17 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 30e32adc1fc6..b7c3fb2bfb54 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -2828,6 +2828,7 @@ extern void radeon_ttm_set_active_vram_size(struct radeon_device *rdev, u64 size
 extern void radeon_program_register_sequence(struct radeon_device *rdev,
 					     const u32 *registers,
 					     const u32 array_size);
+struct radeon_device *radeon_get_rdev(struct ttm_bo_device *bdev);
 
 /*
  * vm
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index d23f2ed4126e..60275b822f79 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -90,7 +90,21 @@ static inline void radeon_bo_unreserve(struct radeon_bo *bo)
  */
 static inline u64 radeon_bo_gpu_offset(struct radeon_bo *bo)
 {
-	return bo->tbo.offset;
+	struct radeon_device *rdev;
+	u64 start = 0;
+
+	rdev = radeon_get_rdev(bo->tbo.bdev);
+
+	switch (bo->tbo.mem.mem_type) {
+	case TTM_PL_TT:
+		start = rdev->mc.gtt_start;
+		break;
+	case TTM_PL_VRAM:
+		start = rdev->mc.vram_start;
+		break;
+	}
+
+	return (bo->tbo.mem.start << PAGE_SHIFT) + start;
 }
 
 static inline unsigned long radeon_bo_size(struct radeon_bo *bo)
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 5d50c9edbe80..357e8e98cca9 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -56,7 +56,7 @@
 static int radeon_ttm_debugfs_init(struct radeon_device *rdev);
 static void radeon_ttm_debugfs_fini(struct radeon_device *rdev);
 
-static struct radeon_device *radeon_get_rdev(struct ttm_bo_device *bdev)
+struct radeon_device *radeon_get_rdev(struct ttm_bo_device *bdev)
 {
 	struct radeon_mman *mman;
 	struct radeon_device *rdev;
@@ -82,7 +82,6 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		break;
 	case TTM_PL_TT:
 		man->func = &ttm_bo_manager_func;
-		man->gpu_offset = rdev->mc.gtt_start;
 		man->available_caching = TTM_PL_MASK_CACHING;
 		man->default_caching = TTM_PL_FLAG_CACHED;
 		man->flags = TTM_MEMTYPE_FLAG_MAPPABLE | TTM_MEMTYPE_FLAG_CMA;
@@ -104,7 +103,6 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	case TTM_PL_VRAM:
 		/* "On-card" video ram */
 		man->func = &ttm_bo_manager_func;
-		man->gpu_offset = rdev->mc.vram_start;
 		man->flags = TTM_MEMTYPE_FLAG_FIXED |
 			     TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC;
-- 
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 3/8] drm/vmwgfx: don't use ttm bo->offset

[Nirmoy Das]

Calculate GPU offset within vmwgfx driver itself without depending on
bo->offset.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_bo.c         | 4 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c    | 2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c       | 2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c | 2 --
 4 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_bo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
index 8b71bf6b58ef..1e59c019affa 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
@@ -258,7 +258,7 @@ int vmw_bo_pin_in_start_of_vram(struct vmw_private *dev_priv,
 		ret = ttm_bo_validate(bo, &placement, &ctx);

 	/* For some reason we didn't end up at the start of vram */
-	WARN_ON(ret == 0 && bo->offset != 0);
+	WARN_ON(ret == 0 && bo->mem.start != 0);
 	if (!ret)
 		vmw_bo_pin_reserved(buf, true);

@@ -317,7 +317,7 @@ void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *bo,
 {
 	if (bo->mem.mem_type == TTM_PL_VRAM) {
 		ptr->gmrId = SVGA_GMR_FRAMEBUFFER;
-		ptr->offset = bo->offset;
+		ptr->offset = bo->mem.start << PAGE_SHIFT;
 	} else {
 		ptr->gmrId = bo->mem.start;
 		ptr->offset = 0;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 367d5b87ee6a..4284c4bd444d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -3696,7 +3696,7 @@ static void vmw_apply_relocations(struct vmw_sw_context *sw_context)
 		bo = &reloc->vbo->base;
 		switch (bo->mem.mem_type) {
 		case TTM_PL_VRAM:
-			reloc->location->offset += bo->offset;
+			reloc->location->offset += bo->mem.start << PAGE_SHIFT;
 			reloc->location->gmrId = SVGA_GMR_FRAMEBUFFER;
 			break;
 		case VMW_PL_GMR:
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 6941689085ed..a95156fc5db7 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -610,7 +610,7 @@ static int vmw_fifo_emit_dummy_legacy_query(struct vmw_private *dev_priv,

 	if (bo->mem.mem_type == TTM_PL_VRAM) {
 		cmd->body.guestResult.gmrId = SVGA_GMR_FRAMEBUFFER;
-		cmd->body.guestResult.offset = bo->offset;
+		cmd->body.guestResult.offset = bo->mem.start << PAGE_SHIFT;
 	} else {
 		cmd->body.guestResult.gmrId = bo->mem.start;
 		cmd->body.guestResult.offset = 0;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c
index bf0bc4697959..fbcd11a7b215 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c
@@ -750,7 +750,6 @@ static int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	case TTM_PL_VRAM:
 		/* "On-card" video ram */
 		man->func = &vmw_thp_func;
-		man->gpu_offset = 0;
 		man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_FLAG_CACHED;
 		man->default_caching = TTM_PL_FLAG_CACHED;
@@ -763,7 +762,6 @@ static int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		 *  slots as well as the bo size.
 		 */
 		man->func = &vmw_gmrid_manager_func;
-		man->gpu_offset = 0;
 		man->flags = TTM_MEMTYPE_FLAG_CMA | TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_FLAG_CACHED;
 		man->default_caching = TTM_PL_FLAG_CACHED;
--
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 4/8] drm/nouveau: don't use ttm bo->offset v3

[Nirmoy Das]

Store ttm bo->offset in struct nouveau_bo instead.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 +++---
 drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
 drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 +++---
 drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
 drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 ++++----
 drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++++
 drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
 drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
 drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++++-----
 15 files changed, 36 insertions(+), 24 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/dispnv04/crtc.c b/drivers/gpu/drm/nouveau/dispnv04/crtc.c
index 27f511b9987b..cc6ab3c2eec7 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/crtc.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/crtc.c
@@ -845,7 +845,7 @@ nv04_crtc_do_mode_set_base(struct drm_crtc *crtc,
 		fb = nouveau_framebuffer(crtc->primary->fb);
 	}
 
-	nv_crtc->fb.offset = fb->nvbo->bo.offset;
+	nv_crtc->fb.offset = fb->nvbo->offset;
 
 	if (nv_crtc->lut.depth != drm_fb->format->depth) {
 		nv_crtc->lut.depth = drm_fb->format->depth;
@@ -1013,7 +1013,7 @@ nv04_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
 		nv04_cursor_upload(dev, cursor, nv_crtc->cursor.nvbo);
 
 	nouveau_bo_unmap(cursor);
-	nv_crtc->cursor.offset = nv_crtc->cursor.nvbo->bo.offset;
+	nv_crtc->cursor.offset = nv_crtc->cursor.nvbo->offset;
 	nv_crtc->cursor.set_offset(nv_crtc, nv_crtc->cursor.offset);
 	nv_crtc->cursor.show(nv_crtc, true);
 out:
@@ -1191,7 +1191,7 @@ nv04_crtc_page_flip(struct drm_crtc *crtc, struct drm_framebuffer *fb,
 	/* Initialize a page flip struct */
 	*s = (struct nv04_page_flip_state)
 		{ { }, event, crtc, fb->format->cpp[0] * 8, fb->pitches[0],
-		  new_bo->bo.offset };
+		  new_bo->offset };
 
 	/* Keep vblanks on during flip, for the target crtc of this flip */
 	drm_crtc_vblank_get(crtc);
diff --git a/drivers/gpu/drm/nouveau/dispnv04/disp.c b/drivers/gpu/drm/nouveau/dispnv04/disp.c
index 44ee82d0c9b6..9272135998aa 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/disp.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/disp.c
@@ -151,7 +151,8 @@ nv04_display_init(struct drm_device *dev, bool resume, bool runtime)
 			continue;
 
 		if (nv_crtc->cursor.set_offset)
-			nv_crtc->cursor.set_offset(nv_crtc, nv_crtc->cursor.nvbo->bo.offset);
+			nv_crtc->cursor.set_offset(nv_crtc,
+						   nv_crtc->cursor.nvbo->offset);
 		nv_crtc->cursor.set_pos(nv_crtc, nv_crtc->cursor_saved_x,
 						 nv_crtc->cursor_saved_y);
 	}
diff --git a/drivers/gpu/drm/nouveau/dispnv04/overlay.c b/drivers/gpu/drm/nouveau/dispnv04/overlay.c
index a3a0a73ae8ab..9529bd9053e7 100644
--- a/drivers/gpu/drm/nouveau/dispnv04/overlay.c
+++ b/drivers/gpu/drm/nouveau/dispnv04/overlay.c
@@ -150,7 +150,7 @@ nv10_update_plane(struct drm_plane *plane, struct drm_crtc *crtc,
 	nvif_mask(dev, NV_PCRTC_ENGINE_CTRL + soff2, NV_CRTC_FSEL_OVERLAY, 0);
 
 	nvif_wr32(dev, NV_PVIDEO_BASE(flip), 0);
-	nvif_wr32(dev, NV_PVIDEO_OFFSET_BUFF(flip), nv_fb->nvbo->bo.offset);
+	nvif_wr32(dev, NV_PVIDEO_OFFSET_BUFF(flip), nv_fb->nvbo->offset);
 	nvif_wr32(dev, NV_PVIDEO_SIZE_IN(flip), src_h << 16 | src_w);
 	nvif_wr32(dev, NV_PVIDEO_POINT_IN(flip), src_y << 16 | src_x);
 	nvif_wr32(dev, NV_PVIDEO_DS_DX(flip), (src_w << 20) / crtc_w);
@@ -172,7 +172,7 @@ nv10_update_plane(struct drm_plane *plane, struct drm_crtc *crtc,
 	if (format & NV_PVIDEO_FORMAT_PLANAR) {
 		nvif_wr32(dev, NV_PVIDEO_UVPLANE_BASE(flip), 0);
 		nvif_wr32(dev, NV_PVIDEO_UVPLANE_OFFSET_BUFF(flip),
-			nv_fb->nvbo->bo.offset + fb->offsets[1]);
+			nv_fb->nvbo->offset + fb->offsets[1]);
 	}
 	nvif_wr32(dev, NV_PVIDEO_FORMAT(flip), format | fb->pitches[0]);
 	nvif_wr32(dev, NV_PVIDEO_STOP, 0);
@@ -396,7 +396,7 @@ nv04_update_plane(struct drm_plane *plane, struct drm_crtc *crtc,
 
 	for (i = 0; i < 2; i++) {
 		nvif_wr32(dev, NV_PVIDEO_BUFF0_START_ADDRESS + 4 * i,
-			  nv_fb->nvbo->bo.offset);
+			  nv_fb->nvbo->offset);
 		nvif_wr32(dev, NV_PVIDEO_BUFF0_PITCH_LENGTH + 4 * i,
 			  fb->pitches[0]);
 		nvif_wr32(dev, NV_PVIDEO_BUFF0_OFFSET + 4 * i, 0);
diff --git a/drivers/gpu/drm/nouveau/dispnv50/base507c.c b/drivers/gpu/drm/nouveau/dispnv50/base507c.c
index ee782151d332..b60aa987d7b4 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/base507c.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/base507c.c
@@ -275,7 +275,7 @@ base507c_new_(const struct nv50_wndw_func *func, const u32 *format,
 
 	ret = nv50_dmac_create(&drm->client.device, &disp->disp->object,
 			       &oclass, head, &args, sizeof(args),
-			       disp->sync->bo.offset, &wndw->wndw);
+			       disp->sync->offset, &wndw->wndw);
 	if (ret) {
 		NV_ERROR(drm, "base%04x allocation failed: %d\n", oclass, ret);
 		return ret;
diff --git a/drivers/gpu/drm/nouveau/dispnv50/core507d.c b/drivers/gpu/drm/nouveau/dispnv50/core507d.c
index c5152c39c684..9151d0260c8a 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/core507d.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/core507d.c
@@ -100,7 +100,7 @@ core507d_new_(const struct nv50_core_func *func, struct nouveau_drm *drm,
 
 	ret = nv50_dmac_create(&drm->client.device, &disp->disp->object,
 			       &oclass, 0, &args, sizeof(args),
-			       disp->sync->bo.offset, &core->chan);
+			       disp->sync->offset, &core->chan);
 	if (ret) {
 		NV_ERROR(drm, "core%04x allocation failed: %d\n", oclass, ret);
 		return ret;
diff --git a/drivers/gpu/drm/nouveau/dispnv50/ovly507e.c b/drivers/gpu/drm/nouveau/dispnv50/ovly507e.c
index 8ccd96113bad..4cce1078140a 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/ovly507e.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/ovly507e.c
@@ -186,7 +186,7 @@ ovly507e_new_(const struct nv50_wndw_func *func, const u32 *format,
 
 	ret = nv50_dmac_create(&drm->client.device, &disp->disp->object,
 			       &oclass, 0, &args, sizeof(args),
-			       disp->sync->bo.offset, &wndw->wndw);
+			       disp->sync->offset, &wndw->wndw);
 	if (ret) {
 		NV_ERROR(drm, "ovly%04x allocation failed: %d\n", oclass, ret);
 		return ret;
diff --git a/drivers/gpu/drm/nouveau/dispnv50/wndw.c b/drivers/gpu/drm/nouveau/dispnv50/wndw.c
index bb737f9281e6..ee0fd817185e 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/wndw.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/wndw.c
@@ -511,7 +511,7 @@ nv50_wndw_prepare_fb(struct drm_plane *plane, struct drm_plane_state *state)
 	}
 
 	asyw->state.fence = dma_resv_get_excl_rcu(fb->nvbo->bo.base.resv);
-	asyw->image.offset[0] = fb->nvbo->bo.offset;
+	asyw->image.offset[0] = fb->nvbo->offset;
 
 	if (wndw->func->prepare) {
 		asyh = nv50_head_atom_get(asyw->state.state, asyw->state.crtc);
diff --git a/drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c b/drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c
index b92dc3461bbd..bb84e4d54a33 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c
@@ -298,7 +298,7 @@ wndwc37e_new_(const struct nv50_wndw_func *func, struct nouveau_drm *drm,
 
 	ret = nv50_dmac_create(&drm->client.device, &disp->disp->object,
 			       &oclass, 0, &args, sizeof(args),
-			       disp->sync->bo.offset, &wndw->wndw);
+			       disp->sync->offset, &wndw->wndw);
 	if (ret) {
 		NV_ERROR(drm, "qndw%04x allocation failed: %d\n", oclass, ret);
 		return ret;
diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c
index 72c91991b96a..5b2406950e53 100644
--- a/drivers/gpu/drm/nouveau/nouveau_abi16.c
+++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c
@@ -558,13 +558,13 @@ nouveau_abi16_ioctl_notifierobj_alloc(ABI16_IOCTL_ARGS)
 	if (drm->agp.bridge) {
 		args.target = NV_DMA_V0_TARGET_AGP;
 		args.access = NV_DMA_V0_ACCESS_RDWR;
-		args.start += drm->agp.base + chan->ntfy->bo.offset;
-		args.limit += drm->agp.base + chan->ntfy->bo.offset;
+		args.start += drm->agp.base + chan->ntfy->offset;
+		args.limit += drm->agp.base + chan->ntfy->offset;
 	} else {
 		args.target = NV_DMA_V0_TARGET_VM;
 		args.access = NV_DMA_V0_ACCESS_RDWR;
-		args.start += chan->ntfy->bo.offset;
-		args.limit += chan->ntfy->bo.offset;
+		args.start += chan->ntfy->offset;
+		args.limit += chan->ntfy->offset;
 	}
 
 	client->route = NVDRM_OBJECT_ABI16;
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index c40f127de3d0..61355cfb7335 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1317,6 +1317,14 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
 			nouveau_vma_unmap(vma);
 		}
 	}
+
+	if (new_reg) {
+		if (new_reg->mm_node)
+			nvbo->offset = (new_reg->start << PAGE_SHIFT);
+		else
+			nvbo->offset = 0;
+	}
+
 }
 
 static int
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
index 38f9d8350963..e944b4aa5547 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
@@ -24,6 +24,9 @@ struct nouveau_bo {
 	int pbbo_index;
 	bool validate_mapped;
 
+	/* GPU address space is independent of CPU word size */
+	uint64_t offset;
+
 	struct list_head vma_list;
 
 	unsigned contig:1;
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index d9381a053169..3d71dfcb2fde 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -162,7 +162,7 @@ nouveau_channel_prep(struct nouveau_drm *drm, struct nvif_device *device,
 	 * pushbuf lives in, this is because the GEM code requires that
 	 * we be able to call out to other (indirect) push buffers
 	 */
-	chan->push.addr = chan->push.buffer->bo.offset;
+	chan->push.addr = chan->push.buffer->offset;
 
 	if (device->info.family >= NV_DEVICE_INFO_V0_TESLA) {
 		ret = nouveau_vma_new(chan->push.buffer, chan->vmm,
diff --git a/drivers/gpu/drm/nouveau/nouveau_dmem.c b/drivers/gpu/drm/nouveau/nouveau_dmem.c
index ad89e09a0be3..f13086a32f0f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dmem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dmem.c
@@ -90,7 +90,7 @@ static unsigned long nouveau_dmem_page_addr(struct page *page)
 	struct nouveau_dmem_chunk *chunk = page->zone_device_data;
 	unsigned long idx = page_to_pfn(page) - chunk->pfn_first;
 
-	return (idx << PAGE_SHIFT) + chunk->bo->bo.offset;
+	return (idx << PAGE_SHIFT) + chunk->bo->offset;
 }
 
 static void nouveau_dmem_page_free(struct page *page)
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
index 24d543a01f43..1341c6fca3ed 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
@@ -393,7 +393,7 @@ nouveau_fbcon_create(struct drm_fb_helper *helper,
 
 	/* To allow resizeing without swapping buffers */
 	NV_INFO(drm, "allocated %dx%d fb: 0x%llx, bo %p\n",
-		fb->base.width, fb->base.height, fb->nvbo->bo.offset, nvbo);
+		fb->base.width, fb->base.height, fb->nvbo->offset, nvbo);
 
 	vga_switcheroo_client_fb_set(dev->pdev, info);
 	return 0;
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 5631c484d10f..b328dd542dba 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -232,7 +232,7 @@ nouveau_gem_info(struct drm_file *file_priv, struct drm_gem_object *gem,
 		rep->domain = NOUVEAU_GEM_DOMAIN_GART;
 	else
 		rep->domain = NOUVEAU_GEM_DOMAIN_VRAM;
-	rep->offset = nvbo->bo.offset;
+	rep->offset = nvbo->offset;
 	if (vmm->vmm.object.oclass >= NVIF_CLASS_VMM_NV50) {
 		vma = nouveau_vma_find(nvbo, vmm);
 		if (!vma)
@@ -516,7 +516,7 @@ validate_list(struct nouveau_channel *chan, struct nouveau_cli *cli,
 		}
 
 		if (drm->client.device.info.family < NV_DEVICE_INFO_V0_TESLA) {
-			if (nvbo->bo.offset == b->presumed.offset &&
+			if (nvbo->offset == b->presumed.offset &&
 			    ((nvbo->bo.mem.mem_type == TTM_PL_VRAM &&
 			      b->presumed.domain & NOUVEAU_GEM_DOMAIN_VRAM) ||
 			     (nvbo->bo.mem.mem_type == TTM_PL_TT &&
@@ -527,7 +527,7 @@ validate_list(struct nouveau_channel *chan, struct nouveau_cli *cli,
 				b->presumed.domain = NOUVEAU_GEM_DOMAIN_GART;
 			else
 				b->presumed.domain = NOUVEAU_GEM_DOMAIN_VRAM;
-			b->presumed.offset = nvbo->bo.offset;
+			b->presumed.offset = nvbo->offset;
 			b->presumed.valid = 0;
 			relocs++;
 		}
@@ -805,7 +805,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 			struct nouveau_bo *nvbo = (void *)(unsigned long)
 				bo[push[i].bo_index].user_priv;
 
-			OUT_RING(chan, (nvbo->bo.offset + push[i].offset) | 2);
+			OUT_RING(chan, (nvbo->offset + push[i].offset) | 2);
 			OUT_RING(chan, 0);
 		}
 	} else {
@@ -840,7 +840,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 			}
 
 			OUT_RING(chan, 0x20000000 |
-				      (nvbo->bo.offset + push[i].offset));
+				      (nvbo->offset + push[i].offset));
 			OUT_RING(chan, 0);
 			for (j = 0; j < NOUVEAU_DMA_SKIPS; j++)
 				OUT_RING(chan, 0);
-- 
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 5/8] drm/qxl: don't use ttm bo->offset

[Nirmoy Das]

This patch removes slot->gpu_offset which is not required as
VRAM and PRIV slot are in separate PCI bar.

This patch also removes unused qxl_bo_gpu_offset()

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Gerd Hoffmann <kraxel@redhat.com>
---
 drivers/gpu/drm/qxl/qxl_drv.h    | 6 ++----
 drivers/gpu/drm/qxl/qxl_kms.c    | 5 ++---
 drivers/gpu/drm/qxl/qxl_object.h | 5 -----
 drivers/gpu/drm/qxl/qxl_ttm.c    | 9 ---------
 4 files changed, 4 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/qxl/qxl_drv.h b/drivers/gpu/drm/qxl/qxl_drv.h
index 31e35f787df2..9691449aefdb 100644
--- a/drivers/gpu/drm/qxl/qxl_drv.h
+++ b/drivers/gpu/drm/qxl/qxl_drv.h
@@ -134,7 +134,6 @@ struct qxl_memslot {
 	uint64_t	start_phys_addr;
 	uint64_t	size;
 	uint64_t	high_bits;
-	uint64_t        gpu_offset;
 };
 
 enum {
@@ -307,10 +306,9 @@ qxl_bo_physical_address(struct qxl_device *qdev, struct qxl_bo *bo,
 		(bo->tbo.mem.mem_type == TTM_PL_VRAM)
 		? &qdev->main_slot : &qdev->surfaces_slot;
 
-	WARN_ON_ONCE((bo->tbo.offset & slot->gpu_offset) != slot->gpu_offset);
+       /* TODO - need to hold one of the locks to read bo->tbo.mem.start */
 
-	/* TODO - need to hold one of the locks to read tbo.offset */
-	return slot->high_bits | (bo->tbo.offset - slot->gpu_offset + offset);
+	return slot->high_bits | ((bo->tbo.mem.start << PAGE_SHIFT) + offset);
 }
 
 /* qxl_display.c */
diff --git a/drivers/gpu/drm/qxl/qxl_kms.c b/drivers/gpu/drm/qxl/qxl_kms.c
index a6d873052cd4..dc5b3850a4d4 100644
--- a/drivers/gpu/drm/qxl/qxl_kms.c
+++ b/drivers/gpu/drm/qxl/qxl_kms.c
@@ -87,11 +87,10 @@ static void setup_slot(struct qxl_device *qdev,
 	high_bits <<= (64 - (qdev->rom->slot_gen_bits + qdev->rom->slot_id_bits));
 	slot->high_bits = high_bits;
 
-	DRM_INFO("slot %d (%s): base 0x%08lx, size 0x%08lx, gpu_offset 0x%lx\n",
+	DRM_INFO("slot %d (%s): base 0x%08lx, size 0x%08lx\n",
 		 slot->index, slot->name,
 		 (unsigned long)slot->start_phys_addr,
-		 (unsigned long)slot->size,
-		 (unsigned long)slot->gpu_offset);
+		 (unsigned long)slot->size);
 }
 
 void qxl_reinit_memslots(struct qxl_device *qdev)
diff --git a/drivers/gpu/drm/qxl/qxl_object.h b/drivers/gpu/drm/qxl/qxl_object.h
index 8ae54ba7857c..21fa81048f4f 100644
--- a/drivers/gpu/drm/qxl/qxl_object.h
+++ b/drivers/gpu/drm/qxl/qxl_object.h
@@ -48,11 +48,6 @@ static inline void qxl_bo_unreserve(struct qxl_bo *bo)
 	ttm_bo_unreserve(&bo->tbo);
 }
 
-static inline u64 qxl_bo_gpu_offset(struct qxl_bo *bo)
-{
-	return bo->tbo.offset;
-}
-
 static inline unsigned long qxl_bo_size(struct qxl_bo *bo)
 {
 	return bo->tbo.num_pages << PAGE_SHIFT;
diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
index f09a712b1ed2..52eaa2d22745 100644
--- a/drivers/gpu/drm/qxl/qxl_ttm.c
+++ b/drivers/gpu/drm/qxl/qxl_ttm.c
@@ -51,11 +51,6 @@ static struct qxl_device *qxl_get_qdev(struct ttm_bo_device *bdev)
 static int qxl_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 			     struct ttm_mem_type_manager *man)
 {
-	struct qxl_device *qdev = qxl_get_qdev(bdev);
-	unsigned int gpu_offset_shift =
-		64 - (qdev->rom->slot_gen_bits + qdev->rom->slot_id_bits + 8);
-	struct qxl_memslot *slot;
-
 	switch (type) {
 	case TTM_PL_SYSTEM:
 		/* System memory */
@@ -66,11 +61,7 @@ static int qxl_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	case TTM_PL_VRAM:
 	case TTM_PL_PRIV:
 		/* "On-card" video ram */
-		slot = (type == TTM_PL_VRAM) ?
-			&qdev->main_slot : &qdev->surfaces_slot;
-		slot->gpu_offset = (uint64_t)type << gpu_offset_shift;
 		man->func = &ttm_bo_manager_func;
-		man->gpu_offset = slot->gpu_offset;
 		man->flags = TTM_MEMTYPE_FLAG_FIXED |
 			     TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_MASK_CACHING;
-- 
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Nirmoy Das]

Calculate GEM VRAM bo's offset within vram-helper without depending on
bo->offset.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
---
 drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c b/drivers/gpu/drm/drm_gem_vram_helper.c
index 0023ce1d2cf7..ad096600d89f 100644
--- a/drivers/gpu/drm/drm_gem_vram_helper.c
+++ b/drivers/gpu/drm/drm_gem_vram_helper.c
@@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct drm_gem_vram_object *gbo)
 }
 EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
 
+static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
+{
+	/* Keep TTM behavior for now, remove when drivers are audited */
+	if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))
+		return 0;
+
+	return gbo->bo.mem.start;
+}
+
 /**
  * drm_gem_vram_offset() - \
 	Returns a GEM VRAM object's offset in video memory
@@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct drm_gem_vram_object *gbo)
 {
 	if (WARN_ON_ONCE(!gbo->pin_count))
 		return (s64)-ENODEV;
-	return gbo->bo.offset;
+	return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
 }
 EXPORT_SYMBOL(drm_gem_vram_offset);
 
-- 
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 4749 bytes --]

Hi

Am 24.06.20 um 20:26 schrieb Nirmoy Das:
> Calculate GEM VRAM bo's offset within vram-helper without depending on
> bo->offset.
> 
> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
> ---
>  drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
>  1 file changed, 10 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c b/drivers/gpu/drm/drm_gem_vram_helper.c
> index 0023ce1d2cf7..ad096600d89f 100644
> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
> @@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct drm_gem_vram_object *gbo)
>  }
>  EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
>  
> +static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
> +{
> +	/* Keep TTM behavior for now, remove when drivers are audited */
> +	if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))

At this line I got

[  146.133821] ------------[ cut here ]------------
[  146.133872] WARNING: CPU: 6 PID: 7 at
drivers/gpu/drm/drm_gem_vram_helper.c:284 drm_gem_vram_offset+0x59/0x60
[drm_vram_helper]
[  146.133880] Modules linked in: fuse(E) af_packet(E) ebtable_filter(E)
ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E)
ip_tables(E) x_tables(E) bpfilter(E) rfkill(E) dmi_sysfs(E) msr(E)
intel_powerclamp(E) coretemp(E) kv)
[  146.134051]  scsi_dh_emc(E) scsi_dh_alua(E)
[  146.134074] CPU: 6 PID: 7 Comm: kworker/u48:0 Tainted: G            E
    5.9.0-rc4-1-default+ #492
[  146.134083] Hardware name: Sun Microsystems SUN FIRE X2270 M2/SUN
FIRE X2270 M2, BIOS 2.05    07/01/2010
[  146.134099] Workqueue: events_unbound commit_work
[  146.134116] RIP: 0010:drm_gem_vram_offset+0x59/0x60 [drm_vram_helper]
[  146.134128] Code: 02 00 00 00 74 24 48 8d bb 80 02 00 00 e8 ef 27 17
d7 48 8b 83 80 02 00 00 5b 48 c1 e0 0c c3 0f 0b 48 c7 c0 ed ff ff ff 5b
c3 <0f> 0b 31 c0 5b c3 90 66 66 66 66 90 41 56 41 55 49 89 d5 41 54 49
[  146.134137] RSP: 0018:ffffc90000107c38 EFLAGS: 00010246
[  146.134149] RAX: 0000000000000000 RBX: ffff888111155000 RCX:
ffffffffc032323b
[  146.134158] RDX: dffffc0000000000 RSI: ffff88810e236300 RDI:
ffff888111155278
[  146.134168] RBP: ffff888109090000 R08: ffffffffc0323225 R09:
0000000000000002
[  146.134177] R10: ffffed1020675020 R11: 0000000000000001 R12:
ffff888109091050
[  146.134187] R13: ffff88810e236300 R14: ffff888109090000 R15:
0000000000000000
[  146.134197] FS:  0000000000000000(0000) GS:ffff888116000000(0000)
knlGS:0000000000000000
[  146.134206] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  146.134215] CR2: 00007f60547d9100 CR3: 0000000029216002 CR4:
00000000000206e0
[  146.134223] Call Trace:
[  146.134245]  ast_cursor_page_flip+0x3e/0x150 [ast]
[  146.134272]  ast_cursor_plane_helper_atomic_update+0x8a/0xc0 [ast]
[  146.134300]  drm_atomic_helper_commit_planes+0x197/0x4c0
[  146.134341]  drm_atomic_helper_commit_tail_rpm+0x51/0x90
[  146.134357]  commit_tail+0x103/0x1c0
[  146.134390]  process_one_work+0x56a/0xa60
[  146.134431]  ? __lock_acquired+0x1ca/0x3d0
[  146.134447]  ? pwq_dec_nr_in_flight+0x110/0x110
[  146.134460]  ? __lock_contended+0x4a0/0x4a0
[  146.134491]  ? worker_thread+0x150/0x620
[  146.134521]  worker_thread+0x8b/0x620
[  146.134539]  ? _raw_spin_unlock_irqrestore+0x41/0x50
[  146.134583]  ? process_one_work+0xa60/0xa60
[  146.134597]  kthread+0x1e4/0x210
[  146.134612]  ? kthread_create_worker_on_cpu+0xb0/0xb0
[  146.134637]  ret_from_fork+0x22/0x30
[  146.134698] irq event stamp: 74111
[  146.134711] hardirqs last  enabled at (74117): [<ffffffff971c68f9>]
console_unlock+0x539/0x670
[  146.134723] hardirqs last disabled at (74122): [<ffffffff971c68ef>]
console_unlock+0x52f/0x670
[  146.134737] softirqs last  enabled at (73354): [<ffffffff975469d5>]
wb_workfn+0x3f5/0x430
[  146.134749] softirqs last disabled at (73350): [<ffffffff973f81d0>]
wb_wakeup_delayed+0x40/0xa0
[  146.134758] ---[ end trace 74dd5fac6a3a2c0c ]---


Happens with ast when doing

  weston-launch



> +		return 0;
> +
> +	return gbo->bo.mem.start;
> +}
> +
>  /**
>   * drm_gem_vram_offset() - \
>  	Returns a GEM VRAM object's offset in video memory
> @@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct drm_gem_vram_object *gbo)
>  {
>  	if (WARN_ON_ONCE(!gbo->pin_count))
>  		return (s64)-ENODEV;
> -	return gbo->bo.offset;
> +	return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
>  }
>  EXPORT_SYMBOL(drm_gem_vram_offset);
>  
> 

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 516 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Christian König]

Hi Thomas,

Am 17.09.20 um 12:51 schrieb Thomas Zimmermann:
> Hi
>
> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>> Calculate GEM VRAM bo's offset within vram-helper without depending on
>> bo->offset.
>>
>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>> ---
>>   drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
>>   1 file changed, 10 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c b/drivers/gpu/drm/drm_gem_vram_helper.c
>> index 0023ce1d2cf7..ad096600d89f 100644
>> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
>> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
>> @@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct drm_gem_vram_object *gbo)
>>   }
>>   EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
>>   
>> +static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
>> +{
>> +	/* Keep TTM behavior for now, remove when drivers are audited */
>> +	if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))
> At this line I got

Sounds like ast forgot to pin the cursor to VRAM.

If you look at ast_cursor_page_flip(), you see:
>         off = drm_gem_vram_offset(gbo);
>         if (drm_WARN_ON_ONCE(dev, off < 0))
>                 return; /* Bug: we didn't pin the cursor HW BO to VRAM. */

The drm_WARN_ON_ONCE() just never triggered before because it checks for 
the wrong condition (off < 0).

Regards,
Christian.

>
> [  146.133821] ------------[ cut here ]------------
> [  146.133872] WARNING: CPU: 6 PID: 7 at
> drivers/gpu/drm/drm_gem_vram_helper.c:284 drm_gem_vram_offset+0x59/0x60
> [drm_vram_helper]
> [  146.133880] Modules linked in: fuse(E) af_packet(E) ebtable_filter(E)
> ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E)
> ip_tables(E) x_tables(E) bpfilter(E) rfkill(E) dmi_sysfs(E) msr(E)
> intel_powerclamp(E) coretemp(E) kv)
> [  146.134051]  scsi_dh_emc(E) scsi_dh_alua(E)
> [  146.134074] CPU: 6 PID: 7 Comm: kworker/u48:0 Tainted: G            E
>      5.9.0-rc4-1-default+ #492
> [  146.134083] Hardware name: Sun Microsystems SUN FIRE X2270 M2/SUN
> FIRE X2270 M2, BIOS 2.05    07/01/2010
> [  146.134099] Workqueue: events_unbound commit_work
> [  146.134116] RIP: 0010:drm_gem_vram_offset+0x59/0x60 [drm_vram_helper]
> [  146.134128] Code: 02 00 00 00 74 24 48 8d bb 80 02 00 00 e8 ef 27 17
> d7 48 8b 83 80 02 00 00 5b 48 c1 e0 0c c3 0f 0b 48 c7 c0 ed ff ff ff 5b
> c3 <0f> 0b 31 c0 5b c3 90 66 66 66 66 90 41 56 41 55 49 89 d5 41 54 49
> [  146.134137] RSP: 0018:ffffc90000107c38 EFLAGS: 00010246
> [  146.134149] RAX: 0000000000000000 RBX: ffff888111155000 RCX:
> ffffffffc032323b
> [  146.134158] RDX: dffffc0000000000 RSI: ffff88810e236300 RDI:
> ffff888111155278
> [  146.134168] RBP: ffff888109090000 R08: ffffffffc0323225 R09:
> 0000000000000002
> [  146.134177] R10: ffffed1020675020 R11: 0000000000000001 R12:
> ffff888109091050
> [  146.134187] R13: ffff88810e236300 R14: ffff888109090000 R15:
> 0000000000000000
> [  146.134197] FS:  0000000000000000(0000) GS:ffff888116000000(0000)
> knlGS:0000000000000000
> [  146.134206] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [  146.134215] CR2: 00007f60547d9100 CR3: 0000000029216002 CR4:
> 00000000000206e0
> [  146.134223] Call Trace:
> [  146.134245]  ast_cursor_page_flip+0x3e/0x150 [ast]
> [  146.134272]  ast_cursor_plane_helper_atomic_update+0x8a/0xc0 [ast]
> [  146.134300]  drm_atomic_helper_commit_planes+0x197/0x4c0
> [  146.134341]  drm_atomic_helper_commit_tail_rpm+0x51/0x90
> [  146.134357]  commit_tail+0x103/0x1c0
> [  146.134390]  process_one_work+0x56a/0xa60
> [  146.134431]  ? __lock_acquired+0x1ca/0x3d0
> [  146.134447]  ? pwq_dec_nr_in_flight+0x110/0x110
> [  146.134460]  ? __lock_contended+0x4a0/0x4a0
> [  146.134491]  ? worker_thread+0x150/0x620
> [  146.134521]  worker_thread+0x8b/0x620
> [  146.134539]  ? _raw_spin_unlock_irqrestore+0x41/0x50
> [  146.134583]  ? process_one_work+0xa60/0xa60
> [  146.134597]  kthread+0x1e4/0x210
> [  146.134612]  ? kthread_create_worker_on_cpu+0xb0/0xb0
> [  146.134637]  ret_from_fork+0x22/0x30
> [  146.134698] irq event stamp: 74111
> [  146.134711] hardirqs last  enabled at (74117): [<ffffffff971c68f9>]
> console_unlock+0x539/0x670
> [  146.134723] hardirqs last disabled at (74122): [<ffffffff971c68ef>]
> console_unlock+0x52f/0x670
> [  146.134737] softirqs last  enabled at (73354): [<ffffffff975469d5>]
> wb_workfn+0x3f5/0x430
> [  146.134749] softirqs last disabled at (73350): [<ffffffff973f81d0>]
> wb_wakeup_delayed+0x40/0xa0
> [  146.134758] ---[ end trace 74dd5fac6a3a2c0c ]---
>
>
> Happens with ast when doing
>
>    weston-launch
>
>
>
>> +		return 0;
>> +
>> +	return gbo->bo.mem.start;
>> +}
>> +
>>   /**
>>    * drm_gem_vram_offset() - \
>>   	Returns a GEM VRAM object's offset in video memory
>> @@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct drm_gem_vram_object *gbo)
>>   {
>>   	if (WARN_ON_ONCE(!gbo->pin_count))
>>   		return (s64)-ENODEV;
>> -	return gbo->bo.offset;
>> +	return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
>>   }
>>   EXPORT_SYMBOL(drm_gem_vram_offset);
>>   
>>

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 6639 bytes --]

Hi Christian

Am 17.09.20 um 13:12 schrieb Christian König:
> Hi Thomas,
> 
> Am 17.09.20 um 12:51 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>> Calculate GEM VRAM bo's offset within vram-helper without depending on
>>> bo->offset.
>>>
>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>> ---
>>>   drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
>>>   1 file changed, 10 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c
>>> b/drivers/gpu/drm/drm_gem_vram_helper.c
>>> index 0023ce1d2cf7..ad096600d89f 100644
>>> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
>>> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
>>> @@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct
>>> drm_gem_vram_object *gbo)
>>>   }
>>>   EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
>>>   +static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
>>> +{
>>> +    /* Keep TTM behavior for now, remove when drivers are audited */
>>> +    if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))
>> At this line I got
> 
> Sounds like ast forgot to pin the cursor to VRAM.
> 
> If you look at ast_cursor_page_flip(), you see:
>>         off = drm_gem_vram_offset(gbo);
>>         if (drm_WARN_ON_ONCE(dev, off < 0))
>>                 return; /* Bug: we didn't pin the cursor HW BO to
>> VRAM. */
> 
> The drm_WARN_ON_ONCE() just never triggered before because it checks for
> the wrong condition (off < 0).

GEM VRAM BOs have a pin counter. drm_gem_vram_offset() returns -ENODEV
if the BO's pin count is 0 (i.e., the BO has not been pinned anywhere).
That's what we're testing here. Two cursor BOs are permanently pinned to
the top of VRAM memory by ast_cursor_init(). If perma-pinning in
ast_cursor_init() fails, driver initialization should fail entirely.

These cursor BOs do some sort of double buffering, On successful
initialization, the actual cursor image is later blit-ed into one of the
BOs.

All this used to work from what I can tell. Is there any chance that the
recent TTM refactoring broke this?

Best regards
Thomas

> 
> Regards,
> Christian.
> 
>>
>> [  146.133821] ------------[ cut here ]------------
>> [  146.133872] WARNING: CPU: 6 PID: 7 at
>> drivers/gpu/drm/drm_gem_vram_helper.c:284 drm_gem_vram_offset+0x59/0x60
>> [drm_vram_helper]
>> [  146.133880] Modules linked in: fuse(E) af_packet(E) ebtable_filter(E)
>> ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E)
>> ip_tables(E) x_tables(E) bpfilter(E) rfkill(E) dmi_sysfs(E) msr(E)
>> intel_powerclamp(E) coretemp(E) kv)
>> [  146.134051]  scsi_dh_emc(E) scsi_dh_alua(E)
>> [  146.134074] CPU: 6 PID: 7 Comm: kworker/u48:0 Tainted: G            E
>>      5.9.0-rc4-1-default+ #492
>> [  146.134083] Hardware name: Sun Microsystems SUN FIRE X2270 M2/SUN
>> FIRE X2270 M2, BIOS 2.05    07/01/2010
>> [  146.134099] Workqueue: events_unbound commit_work
>> [  146.134116] RIP: 0010:drm_gem_vram_offset+0x59/0x60 [drm_vram_helper]
>> [  146.134128] Code: 02 00 00 00 74 24 48 8d bb 80 02 00 00 e8 ef 27 17
>> d7 48 8b 83 80 02 00 00 5b 48 c1 e0 0c c3 0f 0b 48 c7 c0 ed ff ff ff 5b
>> c3 <0f> 0b 31 c0 5b c3 90 66 66 66 66 90 41 56 41 55 49 89 d5 41 54 49
>> [  146.134137] RSP: 0018:ffffc90000107c38 EFLAGS: 00010246
>> [  146.134149] RAX: 0000000000000000 RBX: ffff888111155000 RCX:
>> ffffffffc032323b
>> [  146.134158] RDX: dffffc0000000000 RSI: ffff88810e236300 RDI:
>> ffff888111155278
>> [  146.134168] RBP: ffff888109090000 R08: ffffffffc0323225 R09:
>> 0000000000000002
>> [  146.134177] R10: ffffed1020675020 R11: 0000000000000001 R12:
>> ffff888109091050
>> [  146.134187] R13: ffff88810e236300 R14: ffff888109090000 R15:
>> 0000000000000000
>> [  146.134197] FS:  0000000000000000(0000) GS:ffff888116000000(0000)
>> knlGS:0000000000000000
>> [  146.134206] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> [  146.134215] CR2: 00007f60547d9100 CR3: 0000000029216002 CR4:
>> 00000000000206e0
>> [  146.134223] Call Trace:
>> [  146.134245]  ast_cursor_page_flip+0x3e/0x150 [ast]
>> [  146.134272]  ast_cursor_plane_helper_atomic_update+0x8a/0xc0 [ast]
>> [  146.134300]  drm_atomic_helper_commit_planes+0x197/0x4c0
>> [  146.134341]  drm_atomic_helper_commit_tail_rpm+0x51/0x90
>> [  146.134357]  commit_tail+0x103/0x1c0
>> [  146.134390]  process_one_work+0x56a/0xa60
>> [  146.134431]  ? __lock_acquired+0x1ca/0x3d0
>> [  146.134447]  ? pwq_dec_nr_in_flight+0x110/0x110
>> [  146.134460]  ? __lock_contended+0x4a0/0x4a0
>> [  146.134491]  ? worker_thread+0x150/0x620
>> [  146.134521]  worker_thread+0x8b/0x620
>> [  146.134539]  ? _raw_spin_unlock_irqrestore+0x41/0x50
>> [  146.134583]  ? process_one_work+0xa60/0xa60
>> [  146.134597]  kthread+0x1e4/0x210
>> [  146.134612]  ? kthread_create_worker_on_cpu+0xb0/0xb0
>> [  146.134637]  ret_from_fork+0x22/0x30
>> [  146.134698] irq event stamp: 74111
>> [  146.134711] hardirqs last  enabled at (74117): [<ffffffff971c68f9>]
>> console_unlock+0x539/0x670
>> [  146.134723] hardirqs last disabled at (74122): [<ffffffff971c68ef>]
>> console_unlock+0x52f/0x670
>> [  146.134737] softirqs last  enabled at (73354): [<ffffffff975469d5>]
>> wb_workfn+0x3f5/0x430
>> [  146.134749] softirqs last disabled at (73350): [<ffffffff973f81d0>]
>> wb_wakeup_delayed+0x40/0xa0
>> [  146.134758] ---[ end trace 74dd5fac6a3a2c0c ]---
>>
>>
>> Happens with ast when doing
>>
>>    weston-launch
>>
>>
>>
>>> +        return 0;
>>> +
>>> +    return gbo->bo.mem.start;
>>> +}
>>> +
>>>   /**
>>>    * drm_gem_vram_offset() - \
>>>       Returns a GEM VRAM object's offset in video memory
>>> @@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct
>>> drm_gem_vram_object *gbo)
>>>   {
>>>       if (WARN_ON_ONCE(!gbo->pin_count))
>>>           return (s64)-ENODEV;
>>> -    return gbo->bo.offset;
>>> +    return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
>>>   }
>>>   EXPORT_SYMBOL(drm_gem_vram_offset);
>>>  
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 516 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Christian König]

Am 17.09.20 um 14:29 schrieb Thomas Zimmermann:
> Hi Christian
>
> Am 17.09.20 um 13:12 schrieb Christian König:
>> Hi Thomas,
>>
>> Am 17.09.20 um 12:51 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>>> Calculate GEM VRAM bo's offset within vram-helper without depending on
>>>> bo->offset.
>>>>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>>> ---
>>>>    drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
>>>>    1 file changed, 10 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>> b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>> index 0023ce1d2cf7..ad096600d89f 100644
>>>> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>> @@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct
>>>> drm_gem_vram_object *gbo)
>>>>    }
>>>>    EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
>>>>    +static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
>>>> +{
>>>> +    /* Keep TTM behavior for now, remove when drivers are audited */
>>>> +    if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))
>>> At this line I got
>> Sounds like ast forgot to pin the cursor to VRAM.
>>
>> If you look at ast_cursor_page_flip(), you see:
>>>          off = drm_gem_vram_offset(gbo);
>>>          if (drm_WARN_ON_ONCE(dev, off < 0))
>>>                  return; /* Bug: we didn't pin the cursor HW BO to
>>> VRAM. */
>> The drm_WARN_ON_ONCE() just never triggered before because it checks for
>> the wrong condition (off < 0).
> GEM VRAM BOs have a pin counter. drm_gem_vram_offset() returns -ENODEV
> if the BO's pin count is 0 (i.e., the BO has not been pinned anywhere).
> That's what we're testing here. Two cursor BOs are permanently pinned to
> the top of VRAM memory by ast_cursor_init(). If perma-pinning in
> ast_cursor_init() fails, driver initialization should fail entirely.
>
> These cursor BOs do some sort of double buffering, On successful
> initialization, the actual cursor image is later blit-ed into one of the
> BOs.
>
> All this used to work from what I can tell. Is there any chance that the
> recent TTM refactoring broke this?

Yeah, certainly possible. If you have time please bisect.

Thanks,
Christian.

>
> Best regards
> Thomas
>
>> Regards,
>> Christian.
>>
>>> [  146.133821] ------------[ cut here ]------------
>>> [  146.133872] WARNING: CPU: 6 PID: 7 at
>>> drivers/gpu/drm/drm_gem_vram_helper.c:284 drm_gem_vram_offset+0x59/0x60
>>> [drm_vram_helper]
>>> [  146.133880] Modules linked in: fuse(E) af_packet(E) ebtable_filter(E)
>>> ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E)
>>> ip_tables(E) x_tables(E) bpfilter(E) rfkill(E) dmi_sysfs(E) msr(E)
>>> intel_powerclamp(E) coretemp(E) kv)
>>> [  146.134051]  scsi_dh_emc(E) scsi_dh_alua(E)
>>> [  146.134074] CPU: 6 PID: 7 Comm: kworker/u48:0 Tainted: G            E
>>>       5.9.0-rc4-1-default+ #492
>>> [  146.134083] Hardware name: Sun Microsystems SUN FIRE X2270 M2/SUN
>>> FIRE X2270 M2, BIOS 2.05    07/01/2010
>>> [  146.134099] Workqueue: events_unbound commit_work
>>> [  146.134116] RIP: 0010:drm_gem_vram_offset+0x59/0x60 [drm_vram_helper]
>>> [  146.134128] Code: 02 00 00 00 74 24 48 8d bb 80 02 00 00 e8 ef 27 17
>>> d7 48 8b 83 80 02 00 00 5b 48 c1 e0 0c c3 0f 0b 48 c7 c0 ed ff ff ff 5b
>>> c3 <0f> 0b 31 c0 5b c3 90 66 66 66 66 90 41 56 41 55 49 89 d5 41 54 49
>>> [  146.134137] RSP: 0018:ffffc90000107c38 EFLAGS: 00010246
>>> [  146.134149] RAX: 0000000000000000 RBX: ffff888111155000 RCX:
>>> ffffffffc032323b
>>> [  146.134158] RDX: dffffc0000000000 RSI: ffff88810e236300 RDI:
>>> ffff888111155278
>>> [  146.134168] RBP: ffff888109090000 R08: ffffffffc0323225 R09:
>>> 0000000000000002
>>> [  146.134177] R10: ffffed1020675020 R11: 0000000000000001 R12:
>>> ffff888109091050
>>> [  146.134187] R13: ffff88810e236300 R14: ffff888109090000 R15:
>>> 0000000000000000
>>> [  146.134197] FS:  0000000000000000(0000) GS:ffff888116000000(0000)
>>> knlGS:0000000000000000
>>> [  146.134206] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> [  146.134215] CR2: 00007f60547d9100 CR3: 0000000029216002 CR4:
>>> 00000000000206e0
>>> [  146.134223] Call Trace:
>>> [  146.134245]  ast_cursor_page_flip+0x3e/0x150 [ast]
>>> [  146.134272]  ast_cursor_plane_helper_atomic_update+0x8a/0xc0 [ast]
>>> [  146.134300]  drm_atomic_helper_commit_planes+0x197/0x4c0
>>> [  146.134341]  drm_atomic_helper_commit_tail_rpm+0x51/0x90
>>> [  146.134357]  commit_tail+0x103/0x1c0
>>> [  146.134390]  process_one_work+0x56a/0xa60
>>> [  146.134431]  ? __lock_acquired+0x1ca/0x3d0
>>> [  146.134447]  ? pwq_dec_nr_in_flight+0x110/0x110
>>> [  146.134460]  ? __lock_contended+0x4a0/0x4a0
>>> [  146.134491]  ? worker_thread+0x150/0x620
>>> [  146.134521]  worker_thread+0x8b/0x620
>>> [  146.134539]  ? _raw_spin_unlock_irqrestore+0x41/0x50
>>> [  146.134583]  ? process_one_work+0xa60/0xa60
>>> [  146.134597]  kthread+0x1e4/0x210
>>> [  146.134612]  ? kthread_create_worker_on_cpu+0xb0/0xb0
>>> [  146.134637]  ret_from_fork+0x22/0x30
>>> [  146.134698] irq event stamp: 74111
>>> [  146.134711] hardirqs last  enabled at (74117): [<ffffffff971c68f9>]
>>> console_unlock+0x539/0x670
>>> [  146.134723] hardirqs last disabled at (74122): [<ffffffff971c68ef>]
>>> console_unlock+0x52f/0x670
>>> [  146.134737] softirqs last  enabled at (73354): [<ffffffff975469d5>]
>>> wb_workfn+0x3f5/0x430
>>> [  146.134749] softirqs last disabled at (73350): [<ffffffff973f81d0>]
>>> wb_wakeup_delayed+0x40/0xa0
>>> [  146.134758] ---[ end trace 74dd5fac6a3a2c0c ]---
>>>
>>>
>>> Happens with ast when doing
>>>
>>>     weston-launch
>>>
>>>
>>>
>>>> +        return 0;
>>>> +
>>>> +    return gbo->bo.mem.start;
>>>> +}
>>>> +
>>>>    /**
>>>>     * drm_gem_vram_offset() - \
>>>>        Returns a GEM VRAM object's offset in video memory
>>>> @@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct
>>>> drm_gem_vram_object *gbo)
>>>>    {
>>>>        if (WARN_ON_ONCE(!gbo->pin_count))
>>>>            return (s64)-ENODEV;
>>>> -    return gbo->bo.offset;
>>>> +    return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
>>>>    }
>>>>    EXPORT_SYMBOL(drm_gem_vram_offset);
>>>>   
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 7497 bytes --]

Hi

Am 17.09.20 um 14:32 schrieb Christian König:
> Am 17.09.20 um 14:29 schrieb Thomas Zimmermann:
>> Hi Christian
>>
>> Am 17.09.20 um 13:12 schrieb Christian König:
>>> Hi Thomas,
>>>
>>> Am 17.09.20 um 12:51 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>>>> Calculate GEM VRAM bo's offset within vram-helper without depending on
>>>>> bo->offset.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>>>> ---
>>>>>    drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
>>>>>    1 file changed, 10 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> index 0023ce1d2cf7..ad096600d89f 100644
>>>>> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> @@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct
>>>>> drm_gem_vram_object *gbo)
>>>>>    }
>>>>>    EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
>>>>>    +static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
>>>>> +{
>>>>> +    /* Keep TTM behavior for now, remove when drivers are audited */
>>>>> +    if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))
>>>> At this line I got
>>> Sounds like ast forgot to pin the cursor to VRAM.
>>>
>>> If you look at ast_cursor_page_flip(), you see:
>>>>          off = drm_gem_vram_offset(gbo);
>>>>          if (drm_WARN_ON_ONCE(dev, off < 0))
>>>>                  return; /* Bug: we didn't pin the cursor HW BO to
>>>> VRAM. */
>>> The drm_WARN_ON_ONCE() just never triggered before because it checks for
>>> the wrong condition (off < 0).
>> GEM VRAM BOs have a pin counter. drm_gem_vram_offset() returns -ENODEV
>> if the BO's pin count is 0 (i.e., the BO has not been pinned anywhere).
>> That's what we're testing here. Two cursor BOs are permanently pinned to
>> the top of VRAM memory by ast_cursor_init(). If perma-pinning in
>> ast_cursor_init() fails, driver initialization should fail entirely.
>>
>> These cursor BOs do some sort of double buffering, On successful
>> initialization, the actual cursor image is later blit-ed into one of the
>> BOs.
>>
>> All this used to work from what I can tell. Is there any chance that the
>> recent TTM refactoring broke this?
> 
> Yeah, certainly possible. If you have time please bisect.

FYI the bug is not in c44264f9f729 ("Merge tag 'v5.8' into drm-next")
from Aug 11. I'll try to bisect.

Best regards
Thomas

> 
> Thanks,
> Christian.
> 
>>
>> Best regards
>> Thomas
>>
>>> Regards,
>>> Christian.
>>>
>>>> [  146.133821] ------------[ cut here ]------------
>>>> [  146.133872] WARNING: CPU: 6 PID: 7 at
>>>> drivers/gpu/drm/drm_gem_vram_helper.c:284 drm_gem_vram_offset+0x59/0x60
>>>> [drm_vram_helper]
>>>> [  146.133880] Modules linked in: fuse(E) af_packet(E)
>>>> ebtable_filter(E)
>>>> ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E)
>>>> ip_tables(E) x_tables(E) bpfilter(E) rfkill(E) dmi_sysfs(E) msr(E)
>>>> intel_powerclamp(E) coretemp(E) kv)
>>>> [  146.134051]  scsi_dh_emc(E) scsi_dh_alua(E)
>>>> [  146.134074] CPU: 6 PID: 7 Comm: kworker/u48:0 Tainted:
>>>> G            E
>>>>       5.9.0-rc4-1-default+ #492
>>>> [  146.134083] Hardware name: Sun Microsystems SUN FIRE X2270 M2/SUN
>>>> FIRE X2270 M2, BIOS 2.05    07/01/2010
>>>> [  146.134099] Workqueue: events_unbound commit_work
>>>> [  146.134116] RIP: 0010:drm_gem_vram_offset+0x59/0x60
>>>> [drm_vram_helper]
>>>> [  146.134128] Code: 02 00 00 00 74 24 48 8d bb 80 02 00 00 e8 ef 27 17
>>>> d7 48 8b 83 80 02 00 00 5b 48 c1 e0 0c c3 0f 0b 48 c7 c0 ed ff ff ff 5b
>>>> c3 <0f> 0b 31 c0 5b c3 90 66 66 66 66 90 41 56 41 55 49 89 d5 41 54 49
>>>> [  146.134137] RSP: 0018:ffffc90000107c38 EFLAGS: 00010246
>>>> [  146.134149] RAX: 0000000000000000 RBX: ffff888111155000 RCX:
>>>> ffffffffc032323b
>>>> [  146.134158] RDX: dffffc0000000000 RSI: ffff88810e236300 RDI:
>>>> ffff888111155278
>>>> [  146.134168] RBP: ffff888109090000 R08: ffffffffc0323225 R09:
>>>> 0000000000000002
>>>> [  146.134177] R10: ffffed1020675020 R11: 0000000000000001 R12:
>>>> ffff888109091050
>>>> [  146.134187] R13: ffff88810e236300 R14: ffff888109090000 R15:
>>>> 0000000000000000
>>>> [  146.134197] FS:  0000000000000000(0000) GS:ffff888116000000(0000)
>>>> knlGS:0000000000000000
>>>> [  146.134206] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>> [  146.134215] CR2: 00007f60547d9100 CR3: 0000000029216002 CR4:
>>>> 00000000000206e0
>>>> [  146.134223] Call Trace:
>>>> [  146.134245]  ast_cursor_page_flip+0x3e/0x150 [ast]
>>>> [  146.134272]  ast_cursor_plane_helper_atomic_update+0x8a/0xc0 [ast]
>>>> [  146.134300]  drm_atomic_helper_commit_planes+0x197/0x4c0
>>>> [  146.134341]  drm_atomic_helper_commit_tail_rpm+0x51/0x90
>>>> [  146.134357]  commit_tail+0x103/0x1c0
>>>> [  146.134390]  process_one_work+0x56a/0xa60
>>>> [  146.134431]  ? __lock_acquired+0x1ca/0x3d0
>>>> [  146.134447]  ? pwq_dec_nr_in_flight+0x110/0x110
>>>> [  146.134460]  ? __lock_contended+0x4a0/0x4a0
>>>> [  146.134491]  ? worker_thread+0x150/0x620
>>>> [  146.134521]  worker_thread+0x8b/0x620
>>>> [  146.134539]  ? _raw_spin_unlock_irqrestore+0x41/0x50
>>>> [  146.134583]  ? process_one_work+0xa60/0xa60
>>>> [  146.134597]  kthread+0x1e4/0x210
>>>> [  146.134612]  ? kthread_create_worker_on_cpu+0xb0/0xb0
>>>> [  146.134637]  ret_from_fork+0x22/0x30
>>>> [  146.134698] irq event stamp: 74111
>>>> [  146.134711] hardirqs last  enabled at (74117): [<ffffffff971c68f9>]
>>>> console_unlock+0x539/0x670
>>>> [  146.134723] hardirqs last disabled at (74122): [<ffffffff971c68ef>]
>>>> console_unlock+0x52f/0x670
>>>> [  146.134737] softirqs last  enabled at (73354): [<ffffffff975469d5>]
>>>> wb_workfn+0x3f5/0x430
>>>> [  146.134749] softirqs last disabled at (73350): [<ffffffff973f81d0>]
>>>> wb_wakeup_delayed+0x40/0xa0
>>>> [  146.134758] ---[ end trace 74dd5fac6a3a2c0c ]---
>>>>
>>>>
>>>> Happens with ast when doing
>>>>
>>>>     weston-launch
>>>>
>>>>
>>>>
>>>>> +        return 0;
>>>>> +
>>>>> +    return gbo->bo.mem.start;
>>>>> +}
>>>>> +
>>>>>    /**
>>>>>     * drm_gem_vram_offset() - \
>>>>>        Returns a GEM VRAM object's offset in video memory
>>>>> @@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct
>>>>> drm_gem_vram_object *gbo)
>>>>>    {
>>>>>        if (WARN_ON_ONCE(!gbo->pin_count))
>>>>>            return (s64)-ENODEV;
>>>>> -    return gbo->bo.offset;
>>>>> +    return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
>>>>>    }
>>>>>    EXPORT_SYMBOL(drm_gem_vram_offset);
>>>>>   
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel@lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 516 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 7454 bytes --]

Hi

Am 17.09.20 um 14:32 schrieb Christian König:
> Am 17.09.20 um 14:29 schrieb Thomas Zimmermann:
>> Hi Christian
>>
>> Am 17.09.20 um 13:12 schrieb Christian König:
>>> Hi Thomas,
>>>
>>> Am 17.09.20 um 12:51 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>>>> Calculate GEM VRAM bo's offset within vram-helper without depending on
>>>>> bo->offset.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>>>> ---
>>>>>    drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
>>>>>    1 file changed, 10 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> index 0023ce1d2cf7..ad096600d89f 100644
>>>>> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> @@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct
>>>>> drm_gem_vram_object *gbo)
>>>>>    }
>>>>>    EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
>>>>>    +static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
>>>>> +{
>>>>> +    /* Keep TTM behavior for now, remove when drivers are audited */
>>>>> +    if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))
>>>> At this line I got
>>> Sounds like ast forgot to pin the cursor to VRAM.
>>>
>>> If you look at ast_cursor_page_flip(), you see:
>>>>          off = drm_gem_vram_offset(gbo);
>>>>          if (drm_WARN_ON_ONCE(dev, off < 0))
>>>>                  return; /* Bug: we didn't pin the cursor HW BO to
>>>> VRAM. */
>>> The drm_WARN_ON_ONCE() just never triggered before because it checks for
>>> the wrong condition (off < 0).
>> GEM VRAM BOs have a pin counter. drm_gem_vram_offset() returns -ENODEV
>> if the BO's pin count is 0 (i.e., the BO has not been pinned anywhere).
>> That's what we're testing here. Two cursor BOs are permanently pinned to
>> the top of VRAM memory by ast_cursor_init(). If perma-pinning in
>> ast_cursor_init() fails, driver initialization should fail entirely.
>>
>> These cursor BOs do some sort of double buffering, On successful
>> initialization, the actual cursor image is later blit-ed into one of the
>> BOs.
>>
>> All this used to work from what I can tell. Is there any chance that the
>> recent TTM refactoring broke this?
> 
> Yeah, certainly possible. If you have time please bisect.

Found it.

https://patchwork.freedesktop.org/patch/391418/

Best regards
Thomas

> 
> Thanks,
> Christian.
> 
>>
>> Best regards
>> Thomas
>>
>>> Regards,
>>> Christian.
>>>
>>>> [  146.133821] ------------[ cut here ]------------
>>>> [  146.133872] WARNING: CPU: 6 PID: 7 at
>>>> drivers/gpu/drm/drm_gem_vram_helper.c:284 drm_gem_vram_offset+0x59/0x60
>>>> [drm_vram_helper]
>>>> [  146.133880] Modules linked in: fuse(E) af_packet(E)
>>>> ebtable_filter(E)
>>>> ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E)
>>>> ip_tables(E) x_tables(E) bpfilter(E) rfkill(E) dmi_sysfs(E) msr(E)
>>>> intel_powerclamp(E) coretemp(E) kv)
>>>> [  146.134051]  scsi_dh_emc(E) scsi_dh_alua(E)
>>>> [  146.134074] CPU: 6 PID: 7 Comm: kworker/u48:0 Tainted:
>>>> G            E
>>>>       5.9.0-rc4-1-default+ #492
>>>> [  146.134083] Hardware name: Sun Microsystems SUN FIRE X2270 M2/SUN
>>>> FIRE X2270 M2, BIOS 2.05    07/01/2010
>>>> [  146.134099] Workqueue: events_unbound commit_work
>>>> [  146.134116] RIP: 0010:drm_gem_vram_offset+0x59/0x60
>>>> [drm_vram_helper]
>>>> [  146.134128] Code: 02 00 00 00 74 24 48 8d bb 80 02 00 00 e8 ef 27 17
>>>> d7 48 8b 83 80 02 00 00 5b 48 c1 e0 0c c3 0f 0b 48 c7 c0 ed ff ff ff 5b
>>>> c3 <0f> 0b 31 c0 5b c3 90 66 66 66 66 90 41 56 41 55 49 89 d5 41 54 49
>>>> [  146.134137] RSP: 0018:ffffc90000107c38 EFLAGS: 00010246
>>>> [  146.134149] RAX: 0000000000000000 RBX: ffff888111155000 RCX:
>>>> ffffffffc032323b
>>>> [  146.134158] RDX: dffffc0000000000 RSI: ffff88810e236300 RDI:
>>>> ffff888111155278
>>>> [  146.134168] RBP: ffff888109090000 R08: ffffffffc0323225 R09:
>>>> 0000000000000002
>>>> [  146.134177] R10: ffffed1020675020 R11: 0000000000000001 R12:
>>>> ffff888109091050
>>>> [  146.134187] R13: ffff88810e236300 R14: ffff888109090000 R15:
>>>> 0000000000000000
>>>> [  146.134197] FS:  0000000000000000(0000) GS:ffff888116000000(0000)
>>>> knlGS:0000000000000000
>>>> [  146.134206] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>> [  146.134215] CR2: 00007f60547d9100 CR3: 0000000029216002 CR4:
>>>> 00000000000206e0
>>>> [  146.134223] Call Trace:
>>>> [  146.134245]  ast_cursor_page_flip+0x3e/0x150 [ast]
>>>> [  146.134272]  ast_cursor_plane_helper_atomic_update+0x8a/0xc0 [ast]
>>>> [  146.134300]  drm_atomic_helper_commit_planes+0x197/0x4c0
>>>> [  146.134341]  drm_atomic_helper_commit_tail_rpm+0x51/0x90
>>>> [  146.134357]  commit_tail+0x103/0x1c0
>>>> [  146.134390]  process_one_work+0x56a/0xa60
>>>> [  146.134431]  ? __lock_acquired+0x1ca/0x3d0
>>>> [  146.134447]  ? pwq_dec_nr_in_flight+0x110/0x110
>>>> [  146.134460]  ? __lock_contended+0x4a0/0x4a0
>>>> [  146.134491]  ? worker_thread+0x150/0x620
>>>> [  146.134521]  worker_thread+0x8b/0x620
>>>> [  146.134539]  ? _raw_spin_unlock_irqrestore+0x41/0x50
>>>> [  146.134583]  ? process_one_work+0xa60/0xa60
>>>> [  146.134597]  kthread+0x1e4/0x210
>>>> [  146.134612]  ? kthread_create_worker_on_cpu+0xb0/0xb0
>>>> [  146.134637]  ret_from_fork+0x22/0x30
>>>> [  146.134698] irq event stamp: 74111
>>>> [  146.134711] hardirqs last  enabled at (74117): [<ffffffff971c68f9>]
>>>> console_unlock+0x539/0x670
>>>> [  146.134723] hardirqs last disabled at (74122): [<ffffffff971c68ef>]
>>>> console_unlock+0x52f/0x670
>>>> [  146.134737] softirqs last  enabled at (73354): [<ffffffff975469d5>]
>>>> wb_workfn+0x3f5/0x430
>>>> [  146.134749] softirqs last disabled at (73350): [<ffffffff973f81d0>]
>>>> wb_wakeup_delayed+0x40/0xa0
>>>> [  146.134758] ---[ end trace 74dd5fac6a3a2c0c ]---
>>>>
>>>>
>>>> Happens with ast when doing
>>>>
>>>>     weston-launch
>>>>
>>>>
>>>>
>>>>> +        return 0;
>>>>> +
>>>>> +    return gbo->bo.mem.start;
>>>>> +}
>>>>> +
>>>>>    /**
>>>>>     * drm_gem_vram_offset() - \
>>>>>        Returns a GEM VRAM object's offset in video memory
>>>>> @@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct
>>>>> drm_gem_vram_object *gbo)
>>>>>    {
>>>>>        if (WARN_ON_ONCE(!gbo->pin_count))
>>>>>            return (s64)-ENODEV;
>>>>> -    return gbo->bo.offset;
>>>>> +    return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
>>>>>    }
>>>>>    EXPORT_SYMBOL(drm_gem_vram_offset);
>>>>>   
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel@lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 516 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Nirmoy]

On 9/17/20 2:29 PM, Thomas Zimmermann wrote:
> Hi Christian
>
> Am 17.09.20 um 13:12 schrieb Christian König:
>> Hi Thomas,
>>
>> Am 17.09.20 um 12:51 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>>> Calculate GEM VRAM bo's offset within vram-helper without depending on
>>>> bo->offset.
>>>>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>>> ---
>>>>    drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
>>>>    1 file changed, 10 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>> b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>> index 0023ce1d2cf7..ad096600d89f 100644
>>>> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>> @@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct
>>>> drm_gem_vram_object *gbo)
>>>>    }
>>>>    EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
>>>>    +static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
>>>> +{
>>>> +    /* Keep TTM behavior for now, remove when drivers are audited */
>>>> +    if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))
>>> At this line I got
>> Sounds like ast forgot to pin the cursor to VRAM.
>>
>> If you look at ast_cursor_page_flip(), you see:
>>>          off = drm_gem_vram_offset(gbo);
>>>          if (drm_WARN_ON_ONCE(dev, off < 0))
>>>                  return; /* Bug: we didn't pin the cursor HW BO to
>>> VRAM. */
>> The drm_WARN_ON_ONCE() just never triggered before because it checks for
>> the wrong condition (off < 0).
> GEM VRAM BOs have a pin counter. drm_gem_vram_offset() returns -ENODEV
> if the BO's pin count is 0 (i.e., the BO has not been pinned anywhere).
> That's what we're testing here. Two cursor BOs are permanently pinned to
> the top of VRAM memory by ast_cursor_init(). If perma-pinning in
> ast_cursor_init() fails, driver initialization should fail entirely.
>
> These cursor BOs do some sort of double buffering, On successful
> initialization, the actual cursor image is later blit-ed into one of the
> BOs.
>
> All this used to work from what I can tell. Is there any chance that the
> recent TTM refactoring broke this?


In ast_cursor_blit(), cursor bo is pinned without any flags --> 
"drm_gem_vram_pin(gbo, 0);"

But in ast_cursor_init() both cursors are pinned with TOPDOWN and VRAM flag.

I wonder if that could cause any issue.


Regards,

Nirmoy



>
> Best regards
> Thomas
>
>> Regards,
>> Christian.
>>
>>> [  146.133821] ------------[ cut here ]------------
>>> [  146.133872] WARNING: CPU: 6 PID: 7 at
>>> drivers/gpu/drm/drm_gem_vram_helper.c:284 drm_gem_vram_offset+0x59/0x60
>>> [drm_vram_helper]
>>> [  146.133880] Modules linked in: fuse(E) af_packet(E) ebtable_filter(E)
>>> ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E)
>>> ip_tables(E) x_tables(E) bpfilter(E) rfkill(E) dmi_sysfs(E) msr(E)
>>> intel_powerclamp(E) coretemp(E) kv)
>>> [  146.134051]  scsi_dh_emc(E) scsi_dh_alua(E)
>>> [  146.134074] CPU: 6 PID: 7 Comm: kworker/u48:0 Tainted: G            E
>>>       5.9.0-rc4-1-default+ #492
>>> [  146.134083] Hardware name: Sun Microsystems SUN FIRE X2270 M2/SUN
>>> FIRE X2270 M2, BIOS 2.05    07/01/2010
>>> [  146.134099] Workqueue: events_unbound commit_work
>>> [  146.134116] RIP: 0010:drm_gem_vram_offset+0x59/0x60 [drm_vram_helper]
>>> [  146.134128] Code: 02 00 00 00 74 24 48 8d bb 80 02 00 00 e8 ef 27 17
>>> d7 48 8b 83 80 02 00 00 5b 48 c1 e0 0c c3 0f 0b 48 c7 c0 ed ff ff ff 5b
>>> c3 <0f> 0b 31 c0 5b c3 90 66 66 66 66 90 41 56 41 55 49 89 d5 41 54 49
>>> [  146.134137] RSP: 0018:ffffc90000107c38 EFLAGS: 00010246
>>> [  146.134149] RAX: 0000000000000000 RBX: ffff888111155000 RCX:
>>> ffffffffc032323b
>>> [  146.134158] RDX: dffffc0000000000 RSI: ffff88810e236300 RDI:
>>> ffff888111155278
>>> [  146.134168] RBP: ffff888109090000 R08: ffffffffc0323225 R09:
>>> 0000000000000002
>>> [  146.134177] R10: ffffed1020675020 R11: 0000000000000001 R12:
>>> ffff888109091050
>>> [  146.134187] R13: ffff88810e236300 R14: ffff888109090000 R15:
>>> 0000000000000000
>>> [  146.134197] FS:  0000000000000000(0000) GS:ffff888116000000(0000)
>>> knlGS:0000000000000000
>>> [  146.134206] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> [  146.134215] CR2: 00007f60547d9100 CR3: 0000000029216002 CR4:
>>> 00000000000206e0
>>> [  146.134223] Call Trace:
>>> [  146.134245]  ast_cursor_page_flip+0x3e/0x150 [ast]
>>> [  146.134272]  ast_cursor_plane_helper_atomic_update+0x8a/0xc0 [ast]
>>> [  146.134300]  drm_atomic_helper_commit_planes+0x197/0x4c0
>>> [  146.134341]  drm_atomic_helper_commit_tail_rpm+0x51/0x90
>>> [  146.134357]  commit_tail+0x103/0x1c0
>>> [  146.134390]  process_one_work+0x56a/0xa60
>>> [  146.134431]  ? __lock_acquired+0x1ca/0x3d0
>>> [  146.134447]  ? pwq_dec_nr_in_flight+0x110/0x110
>>> [  146.134460]  ? __lock_contended+0x4a0/0x4a0
>>> [  146.134491]  ? worker_thread+0x150/0x620
>>> [  146.134521]  worker_thread+0x8b/0x620
>>> [  146.134539]  ? _raw_spin_unlock_irqrestore+0x41/0x50
>>> [  146.134583]  ? process_one_work+0xa60/0xa60
>>> [  146.134597]  kthread+0x1e4/0x210
>>> [  146.134612]  ? kthread_create_worker_on_cpu+0xb0/0xb0
>>> [  146.134637]  ret_from_fork+0x22/0x30
>>> [  146.134698] irq event stamp: 74111
>>> [  146.134711] hardirqs last  enabled at (74117): [<ffffffff971c68f9>]
>>> console_unlock+0x539/0x670
>>> [  146.134723] hardirqs last disabled at (74122): [<ffffffff971c68ef>]
>>> console_unlock+0x52f/0x670
>>> [  146.134737] softirqs last  enabled at (73354): [<ffffffff975469d5>]
>>> wb_workfn+0x3f5/0x430
>>> [  146.134749] softirqs last disabled at (73350): [<ffffffff973f81d0>]
>>> wb_wakeup_delayed+0x40/0xa0
>>> [  146.134758] ---[ end trace 74dd5fac6a3a2c0c ]---
>>>
>>>
>>> Happens with ast when doing
>>>
>>>     weston-launch
>>>
>>>
>>>
>>>> +        return 0;
>>>> +
>>>> +    return gbo->bo.mem.start;
>>>> +}
>>>> +
>>>>    /**
>>>>     * drm_gem_vram_offset() - \
>>>>        Returns a GEM VRAM object's offset in video memory
>>>> @@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct
>>>> drm_gem_vram_object *gbo)
>>>>    {
>>>>        if (WARN_ON_ONCE(!gbo->pin_count))
>>>>            return (s64)-ENODEV;
>>>> -    return gbo->bo.offset;
>>>> +    return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
>>>>    }
>>>>    EXPORT_SYMBOL(drm_gem_vram_offset);
>>>>   
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 6/8] drm/vram-helper: don't use ttm bo->offset v4

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 7677 bytes --]

Hi

Am 17.09.20 um 14:58 schrieb Nirmoy:
> 
> On 9/17/20 2:29 PM, Thomas Zimmermann wrote:
>> Hi Christian
>>
>> Am 17.09.20 um 13:12 schrieb Christian König:
>>> Hi Thomas,
>>>
>>> Am 17.09.20 um 12:51 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>>>> Calculate GEM VRAM bo's offset within vram-helper without depending on
>>>>> bo->offset.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>>>> ---
>>>>>    drivers/gpu/drm/drm_gem_vram_helper.c | 11 ++++++++++-
>>>>>    1 file changed, 10 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> index 0023ce1d2cf7..ad096600d89f 100644
>>>>> --- a/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
>>>>> @@ -281,6 +281,15 @@ u64 drm_gem_vram_mmap_offset(struct
>>>>> drm_gem_vram_object *gbo)
>>>>>    }
>>>>>    EXPORT_SYMBOL(drm_gem_vram_mmap_offset);
>>>>>    +static u64 drm_gem_vram_pg_offset(struct drm_gem_vram_object *gbo)
>>>>> +{
>>>>> +    /* Keep TTM behavior for now, remove when drivers are audited */
>>>>> +    if (WARN_ON_ONCE(!gbo->bo.mem.mm_node))
>>>> At this line I got
>>> Sounds like ast forgot to pin the cursor to VRAM.
>>>
>>> If you look at ast_cursor_page_flip(), you see:
>>>>          off = drm_gem_vram_offset(gbo);
>>>>          if (drm_WARN_ON_ONCE(dev, off < 0))
>>>>                  return; /* Bug: we didn't pin the cursor HW BO to
>>>> VRAM. */
>>> The drm_WARN_ON_ONCE() just never triggered before because it checks for
>>> the wrong condition (off < 0).
>> GEM VRAM BOs have a pin counter. drm_gem_vram_offset() returns -ENODEV
>> if the BO's pin count is 0 (i.e., the BO has not been pinned anywhere).
>> That's what we're testing here. Two cursor BOs are permanently pinned to
>> the top of VRAM memory by ast_cursor_init(). If perma-pinning in
>> ast_cursor_init() fails, driver initialization should fail entirely.
>>
>> These cursor BOs do some sort of double buffering, On successful
>> initialization, the actual cursor image is later blit-ed into one of the
>> BOs.
>>
>> All this used to work from what I can tell. Is there any chance that the
>> recent TTM refactoring broke this?
> 
> 
> In ast_cursor_blit(), cursor bo is pinned without any flags -->
> "drm_gem_vram_pin(gbo, 0);"
> 
> But in ast_cursor_init() both cursors are pinned with TOPDOWN and VRAM
> flag.
> 
> I wonder if that could cause any issue.

It shouldn't. The BO has been pinned already when blit is running. Only
the pin count is being incremented.

Best regards
Thomas

> 
> 
> Regards,
> 
> Nirmoy
> 
> 
> 
>>
>> Best regards
>> Thomas
>>
>>> Regards,
>>> Christian.
>>>
>>>> [  146.133821] ------------[ cut here ]------------
>>>> [  146.133872] WARNING: CPU: 6 PID: 7 at
>>>> drivers/gpu/drm/drm_gem_vram_helper.c:284 drm_gem_vram_offset+0x59/0x60
>>>> [drm_vram_helper]
>>>> [  146.133880] Modules linked in: fuse(E) af_packet(E)
>>>> ebtable_filter(E)
>>>> ebtables(E) ip6table_filter(E) ip6_tables(E) iptable_filter(E)
>>>> ip_tables(E) x_tables(E) bpfilter(E) rfkill(E) dmi_sysfs(E) msr(E)
>>>> intel_powerclamp(E) coretemp(E) kv)
>>>> [  146.134051]  scsi_dh_emc(E) scsi_dh_alua(E)
>>>> [  146.134074] CPU: 6 PID: 7 Comm: kworker/u48:0 Tainted:
>>>> G            E
>>>>       5.9.0-rc4-1-default+ #492
>>>> [  146.134083] Hardware name: Sun Microsystems SUN FIRE X2270 M2/SUN
>>>> FIRE X2270 M2, BIOS 2.05    07/01/2010
>>>> [  146.134099] Workqueue: events_unbound commit_work
>>>> [  146.134116] RIP: 0010:drm_gem_vram_offset+0x59/0x60
>>>> [drm_vram_helper]
>>>> [  146.134128] Code: 02 00 00 00 74 24 48 8d bb 80 02 00 00 e8 ef 27 17
>>>> d7 48 8b 83 80 02 00 00 5b 48 c1 e0 0c c3 0f 0b 48 c7 c0 ed ff ff ff 5b
>>>> c3 <0f> 0b 31 c0 5b c3 90 66 66 66 66 90 41 56 41 55 49 89 d5 41 54 49
>>>> [  146.134137] RSP: 0018:ffffc90000107c38 EFLAGS: 00010246
>>>> [  146.134149] RAX: 0000000000000000 RBX: ffff888111155000 RCX:
>>>> ffffffffc032323b
>>>> [  146.134158] RDX: dffffc0000000000 RSI: ffff88810e236300 RDI:
>>>> ffff888111155278
>>>> [  146.134168] RBP: ffff888109090000 R08: ffffffffc0323225 R09:
>>>> 0000000000000002
>>>> [  146.134177] R10: ffffed1020675020 R11: 0000000000000001 R12:
>>>> ffff888109091050
>>>> [  146.134187] R13: ffff88810e236300 R14: ffff888109090000 R15:
>>>> 0000000000000000
>>>> [  146.134197] FS:  0000000000000000(0000) GS:ffff888116000000(0000)
>>>> knlGS:0000000000000000
>>>> [  146.134206] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>>> [  146.134215] CR2: 00007f60547d9100 CR3: 0000000029216002 CR4:
>>>> 00000000000206e0
>>>> [  146.134223] Call Trace:
>>>> [  146.134245]  ast_cursor_page_flip+0x3e/0x150 [ast]
>>>> [  146.134272]  ast_cursor_plane_helper_atomic_update+0x8a/0xc0 [ast]
>>>> [  146.134300]  drm_atomic_helper_commit_planes+0x197/0x4c0
>>>> [  146.134341]  drm_atomic_helper_commit_tail_rpm+0x51/0x90
>>>> [  146.134357]  commit_tail+0x103/0x1c0
>>>> [  146.134390]  process_one_work+0x56a/0xa60
>>>> [  146.134431]  ? __lock_acquired+0x1ca/0x3d0
>>>> [  146.134447]  ? pwq_dec_nr_in_flight+0x110/0x110
>>>> [  146.134460]  ? __lock_contended+0x4a0/0x4a0
>>>> [  146.134491]  ? worker_thread+0x150/0x620
>>>> [  146.134521]  worker_thread+0x8b/0x620
>>>> [  146.134539]  ? _raw_spin_unlock_irqrestore+0x41/0x50
>>>> [  146.134583]  ? process_one_work+0xa60/0xa60
>>>> [  146.134597]  kthread+0x1e4/0x210
>>>> [  146.134612]  ? kthread_create_worker_on_cpu+0xb0/0xb0
>>>> [  146.134637]  ret_from_fork+0x22/0x30
>>>> [  146.134698] irq event stamp: 74111
>>>> [  146.134711] hardirqs last  enabled at (74117): [<ffffffff971c68f9>]
>>>> console_unlock+0x539/0x670
>>>> [  146.134723] hardirqs last disabled at (74122): [<ffffffff971c68ef>]
>>>> console_unlock+0x52f/0x670
>>>> [  146.134737] softirqs last  enabled at (73354): [<ffffffff975469d5>]
>>>> wb_workfn+0x3f5/0x430
>>>> [  146.134749] softirqs last disabled at (73350): [<ffffffff973f81d0>]
>>>> wb_wakeup_delayed+0x40/0xa0
>>>> [  146.134758] ---[ end trace 74dd5fac6a3a2c0c ]---
>>>>
>>>>
>>>> Happens with ast when doing
>>>>
>>>>     weston-launch
>>>>
>>>>
>>>>
>>>>> +        return 0;
>>>>> +
>>>>> +    return gbo->bo.mem.start;
>>>>> +}
>>>>> +
>>>>>    /**
>>>>>     * drm_gem_vram_offset() - \
>>>>>        Returns a GEM VRAM object's offset in video memory
>>>>> @@ -297,7 +306,7 @@ s64 drm_gem_vram_offset(struct
>>>>> drm_gem_vram_object *gbo)
>>>>>    {
>>>>>        if (WARN_ON_ONCE(!gbo->pin_count))
>>>>>            return (s64)-ENODEV;
>>>>> -    return gbo->bo.offset;
>>>>> +    return drm_gem_vram_pg_offset(gbo) << PAGE_SHIFT;
>>>>>    }
>>>>>    EXPORT_SYMBOL(drm_gem_vram_offset);
>>>>>   
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel@lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 516 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 7/8] drm/bochs: use drm_gem_vram_offset to get bo offset v2

[Nirmoy Das]

Switch over to GEM VRAM's implementation to retrieve bo->offset.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
---
 drivers/gpu/drm/bochs/bochs_kms.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/bochs/bochs_kms.c b/drivers/gpu/drm/bochs/bochs_kms.c
index 05d8373888e8..853081d186d5 100644
--- a/drivers/gpu/drm/bochs/bochs_kms.c
+++ b/drivers/gpu/drm/bochs/bochs_kms.c
@@ -29,16 +29,21 @@ static void bochs_plane_update(struct bochs_device *bochs,
 			       struct drm_plane_state *state)
 {
 	struct drm_gem_vram_object *gbo;
+	s64 gpu_addr;
 
 	if (!state->fb || !bochs->stride)
 		return;
 
 	gbo = drm_gem_vram_of_gem(state->fb->obj[0]);
+	gpu_addr = drm_gem_vram_offset(gbo);
+	if (WARN_ON_ONCE(gpu_addr < 0))
+		return; /* Bug: we didn't pin the BO to VRAM in prepare_fb. */
+
 	bochs_hw_setbase(bochs,
 			 state->crtc_x,
 			 state->crtc_y,
 			 state->fb->pitches[0],
-			 state->fb->offsets[0] + gbo->bo.offset);
+			 state->fb->offsets[0] + gpu_addr);
 	bochs_hw_setformat(bochs, state->fb->format);
 }
 
-- 
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy Das]

GPU address handling is device specific and should be handle by its device
driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
 include/drm/ttm/ttm_bo_api.h    | 2 --
 include/drm/ttm/ttm_bo_driver.h | 1 -
 3 files changed, 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index f73b81c2576e..f78cfc76ad78 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
 	drm_printf(p, "    has_type: %d\n", man->has_type);
 	drm_printf(p, "    use_type: %d\n", man->use_type);
 	drm_printf(p, "    flags: 0x%08X\n", man->flags);
-	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
 	drm_printf(p, "    size: %llu\n", man->size);
 	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
 	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
@@ -343,12 +342,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 moved:
 	bo->evicted = false;
 
-	if (bo->mem.mm_node)
-		bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		    bdev->man[bo->mem.mem_type].gpu_offset;
-	else
-		bo->offset = 0;
-
 	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
 	return 0;
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 0a9d042e075a..a6ec6101d9ec 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -213,8 +213,6 @@ struct ttm_buffer_object {
 	 * either of these locks held.
 	 */
 
-	uint64_t offset; /* GPU address space is independent of CPU word size */
-
 	struct sg_table *sg;
 };
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 54a527aa79cc..aa1f398c2ea7 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
 	bool has_type;
 	bool use_type;
 	uint32_t flags;
-	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
2.27.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Christian König]

I've pushed patches #1, #2 and #5-#8 of this series to drm-misc-next.

Only VMGFX and Nouveau are missing and I'm pretty close to just push 
them with my Acked-by since they should not contain any functional change.

Any objections?

Thanks,
Christian.

Am 24.06.20 um 20:26 schrieb Nirmoy Das:
> With this patch series I am trying to remove GPU address dependency in
> TTM and moving GPU address calculation to individual drm drivers. This
> cleanup will simplify introduction of drm_mem_region/domain work started
> by Brian Welty[1].
>
>
> It would be nice if someone test this for nouveau. Rest of the drivers
> are already tested.
>
> v2:
> * set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL
>
> v3:
> * catch return value of drm_gem_vram_offset() in drm/bochs
> * introduce drm_gem_vram_pg_offset() in vram helper
> * improve nbo->offset calculation for nouveau
>
> v4:
> * minor coding style fixes in amdgpu and radeon
> * remove unnecessary kerneldoc for internal function
>
> v5:
> * rebase on top of drm-misc-next
> * fix return value of drm_gem_vram_pg_offset()
> * add a comment in drm_gem_vram_pg_offset() to clearify why we return 0.
>
> v6:
> * rebase to drm-misc-next
> * removed acked for vmwgfx as there was a small conflict
>
> Nirmoy Das (8):
>    drm/amdgpu: move ttm bo->offset to amdgpu_bo
>    drm/radeon: don't use ttm bo->offset
>    drm/vmwgfx: don't use ttm bo->offset
>    drm/nouveau: don't use ttm bo->offset v3
>    drm/qxl: don't use ttm bo->offset
>    drm/vram-helper: don't use ttm bo->offset v4
>    drm/bochs: use drm_gem_vram_offset to get bo offset v2
>    drm/ttm: do not keep GPU dependent addresses
>
>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 ++++++++++++++++-----
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
>   drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
>   drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
>   drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
>   drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
>   drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
>   drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
>   drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
>   drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
>   drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
>   drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
>   drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
>   drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
>   drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
>   drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
>   drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
>   drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
>   drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
>   drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
>   drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
>   drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
>   drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
>   drivers/gpu/drm/radeon/radeon.h             |  1 +
>   drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
>   drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
>   drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
>   drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
>   drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
>   drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
>   drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
>   include/drm/ttm/ttm_bo_api.h                |  2 --
>   include/drm/ttm/ttm_bo_driver.h             |  1 -
>   36 files changed, 125 insertions(+), 78 deletions(-)
>
> --
> 2.27.0
>

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Daniel Vetter]

On Thu, Jun 25, 2020 at 11:50 AM Christian König
<christian.koenig@amd.com> wrote:
>
> I've pushed patches #1, #2 and #5-#8 of this series to drm-misc-next.
>
> Only VMGFX and Nouveau are missing and I'm pretty close to just push
> them with my Acked-by since they should not contain any functional change.
>
> Any objections?

Sounds good to me, I just suggested to Alex on irc you do just that.
These patches have been floating for long enough that there's stalling
for driver maintainers is imo not needed for longer.
-Daniel

>
> Thanks,
> Christian.
>
> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
> > With this patch series I am trying to remove GPU address dependency in
> > TTM and moving GPU address calculation to individual drm drivers. This
> > cleanup will simplify introduction of drm_mem_region/domain work started
> > by Brian Welty[1].
> >
> >
> > It would be nice if someone test this for nouveau. Rest of the drivers
> > are already tested.
> >
> > v2:
> > * set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL
> >
> > v3:
> > * catch return value of drm_gem_vram_offset() in drm/bochs
> > * introduce drm_gem_vram_pg_offset() in vram helper
> > * improve nbo->offset calculation for nouveau
> >
> > v4:
> > * minor coding style fixes in amdgpu and radeon
> > * remove unnecessary kerneldoc for internal function
> >
> > v5:
> > * rebase on top of drm-misc-next
> > * fix return value of drm_gem_vram_pg_offset()
> > * add a comment in drm_gem_vram_pg_offset() to clearify why we return 0.
> >
> > v6:
> > * rebase to drm-misc-next
> > * removed acked for vmwgfx as there was a small conflict
> >
> > Nirmoy Das (8):
> >    drm/amdgpu: move ttm bo->offset to amdgpu_bo
> >    drm/radeon: don't use ttm bo->offset
> >    drm/vmwgfx: don't use ttm bo->offset
> >    drm/nouveau: don't use ttm bo->offset v3
> >    drm/qxl: don't use ttm bo->offset
> >    drm/vram-helper: don't use ttm bo->offset v4
> >    drm/bochs: use drm_gem_vram_offset to get bo offset v2
> >    drm/ttm: do not keep GPU dependent addresses
> >
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 ++++++++++++++++-----
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
> >   drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
> >   drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
> >   drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
> >   drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
> >   drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
> >   drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
> >   drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
> >   drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
> >   drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
> >   drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
> >   drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
> >   drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
> >   drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
> >   drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
> >   drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
> >   drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
> >   drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
> >   drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
> >   drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
> >   drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
> >   drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
> >   drivers/gpu/drm/radeon/radeon.h             |  1 +
> >   drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
> >   drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
> >   drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
> >   drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
> >   drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
> >   include/drm/ttm/ttm_bo_api.h                |  2 --
> >   include/drm/ttm/ttm_bo_driver.h             |  1 -
> >   36 files changed, 125 insertions(+), 78 deletions(-)
> >
> > --
> > 2.27.0
> >
>


-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Daniel Vetter]

On Thu, Jun 25, 2020 at 11:50 AM Christian König
<christian.koenig@amd.com> wrote:
>
> I've pushed patches #1, #2 and #5-#8 of this series to drm-misc-next.

I think you left an unresolved conflict behind in drm-tip, please resolve per

https://drm.pages.freedesktop.org/maintainer-tools/drm-tip.html#resolving-conflicts-when-rebuilding-drm-tip

The script should have told you that already, so maybe reinvent that
in whatever thing you're using :-) Jani has reported this on
#dri-devel, this also holds up CI testing since we're running on top
of drm-tip.
-Daniel



> Only VMGFX and Nouveau are missing and I'm pretty close to just push
> them with my Acked-by since they should not contain any functional change.
>
> Any objections?
>
> Thanks,
> Christian.
>
> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
> > With this patch series I am trying to remove GPU address dependency in
> > TTM and moving GPU address calculation to individual drm drivers. This
> > cleanup will simplify introduction of drm_mem_region/domain work started
> > by Brian Welty[1].
> >
> >
> > It would be nice if someone test this for nouveau. Rest of the drivers
> > are already tested.
> >
> > v2:
> > * set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL
> >
> > v3:
> > * catch return value of drm_gem_vram_offset() in drm/bochs
> > * introduce drm_gem_vram_pg_offset() in vram helper
> > * improve nbo->offset calculation for nouveau
> >
> > v4:
> > * minor coding style fixes in amdgpu and radeon
> > * remove unnecessary kerneldoc for internal function
> >
> > v5:
> > * rebase on top of drm-misc-next
> > * fix return value of drm_gem_vram_pg_offset()
> > * add a comment in drm_gem_vram_pg_offset() to clearify why we return 0.
> >
> > v6:
> > * rebase to drm-misc-next
> > * removed acked for vmwgfx as there was a small conflict
> >
> > Nirmoy Das (8):
> >    drm/amdgpu: move ttm bo->offset to amdgpu_bo
> >    drm/radeon: don't use ttm bo->offset
> >    drm/vmwgfx: don't use ttm bo->offset
> >    drm/nouveau: don't use ttm bo->offset v3
> >    drm/qxl: don't use ttm bo->offset
> >    drm/vram-helper: don't use ttm bo->offset v4
> >    drm/bochs: use drm_gem_vram_offset to get bo offset v2
> >    drm/ttm: do not keep GPU dependent addresses
> >
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 ++++++++++++++++-----
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
> >   drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
> >   drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
> >   drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
> >   drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
> >   drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
> >   drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
> >   drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
> >   drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
> >   drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
> >   drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
> >   drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
> >   drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
> >   drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
> >   drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
> >   drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
> >   drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
> >   drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
> >   drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
> >   drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
> >   drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
> >   drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
> >   drivers/gpu/drm/radeon/radeon.h             |  1 +
> >   drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
> >   drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
> >   drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
> >   drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
> >   drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
> >   drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
> >   include/drm/ttm/ttm_bo_api.h                |  2 --
> >   include/drm/ttm/ttm_bo_driver.h             |  1 -
> >   36 files changed, 125 insertions(+), 78 deletions(-)
> >
> > --
> > 2.27.0
> >
>


-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Christian König]

Am 25.06.20 um 17:44 schrieb Daniel Vetter:
> On Thu, Jun 25, 2020 at 11:50 AM Christian König
> <christian.koenig@amd.com> wrote:
>> I've pushed patches #1, #2 and #5-#8 of this series to drm-misc-next.
> I think you left an unresolved conflict behind in drm-tip, please resolve per
>
> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdrm.pages.freedesktop.org%2Fmaintainer-tools%2Fdrm-tip.html%23resolving-conflicts-when-rebuilding-drm-tip&amp;data=02%7C01%7Cchristian.koenig%40amd.com%7Cc1441a81e9dc4fa4507208d8191ea0f6%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637286966631595300&amp;sdata=pfWS4VgDvU8IsR2638MDr7fYWE0nefa3b6XxyPCCsOU%3D&amp;reserved=0
>
> The script should have told you that already, so maybe reinvent that
> in whatever thing you're using :-) Jani has reported this on
> #dri-devel, this also holds up CI testing since we're running on top
> of drm-tip.

Well I used dim push-branch. I even had to rebase once more because I 
forgot a signed-of on one of the patches, so I'm pretty sure of that.

Haven't seen anything problematic except for that while doing so.

Regards,
Christian.

> -Daniel
>
>
>
>> Only VMGFX and Nouveau are missing and I'm pretty close to just push
>> them with my Acked-by since they should not contain any functional change.
>>
>> Any objections?
>>
>> Thanks,
>> Christian.
>>
>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>> With this patch series I am trying to remove GPU address dependency in
>>> TTM and moving GPU address calculation to individual drm drivers. This
>>> cleanup will simplify introduction of drm_mem_region/domain work started
>>> by Brian Welty[1].
>>>
>>>
>>> It would be nice if someone test this for nouveau. Rest of the drivers
>>> are already tested.
>>>
>>> v2:
>>> * set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL
>>>
>>> v3:
>>> * catch return value of drm_gem_vram_offset() in drm/bochs
>>> * introduce drm_gem_vram_pg_offset() in vram helper
>>> * improve nbo->offset calculation for nouveau
>>>
>>> v4:
>>> * minor coding style fixes in amdgpu and radeon
>>> * remove unnecessary kerneldoc for internal function
>>>
>>> v5:
>>> * rebase on top of drm-misc-next
>>> * fix return value of drm_gem_vram_pg_offset()
>>> * add a comment in drm_gem_vram_pg_offset() to clearify why we return 0.
>>>
>>> v6:
>>> * rebase to drm-misc-next
>>> * removed acked for vmwgfx as there was a small conflict
>>>
>>> Nirmoy Das (8):
>>>     drm/amdgpu: move ttm bo->offset to amdgpu_bo
>>>     drm/radeon: don't use ttm bo->offset
>>>     drm/vmwgfx: don't use ttm bo->offset
>>>     drm/nouveau: don't use ttm bo->offset v3
>>>     drm/qxl: don't use ttm bo->offset
>>>     drm/vram-helper: don't use ttm bo->offset v4
>>>     drm/bochs: use drm_gem_vram_offset to get bo offset v2
>>>     drm/ttm: do not keep GPU dependent addresses
>>>
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 ++++++++++++++++-----
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
>>>    drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
>>>    drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
>>>    drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
>>>    drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
>>>    drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
>>>    drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
>>>    drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
>>>    drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
>>>    drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
>>>    drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
>>>    drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
>>>    drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
>>>    drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
>>>    drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
>>>    drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
>>>    drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
>>>    drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
>>>    drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
>>>    drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
>>>    drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
>>>    drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
>>>    drivers/gpu/drm/radeon/radeon.h             |  1 +
>>>    drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
>>>    drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
>>>    drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
>>>    drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
>>>    include/drm/ttm/ttm_bo_api.h                |  2 --
>>>    include/drm/ttm/ttm_bo_driver.h             |  1 -
>>>    36 files changed, 125 insertions(+), 78 deletions(-)
>>>
>>> --
>>> 2.27.0
>>>
>

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Christian König]

Am 25.06.20 um 17:52 schrieb Christian König:
> Am 25.06.20 um 17:44 schrieb Daniel Vetter:
>> On Thu, Jun 25, 2020 at 11:50 AM Christian König
>> <christian.koenig@amd.com> wrote:
>>> I've pushed patches #1, #2 and #5-#8 of this series to drm-misc-next.
>> I think you left an unresolved conflict behind in drm-tip, please 
>> resolve per
>>
>> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdrm.pages.freedesktop.org%2Fmaintainer-tools%2Fdrm-tip.html%23resolving-conflicts-when-rebuilding-drm-tip&amp;data=02%7C01%7Cchristian.koenig%40amd.com%7Cc1441a81e9dc4fa4507208d8191ea0f6%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637286966631595300&amp;sdata=pfWS4VgDvU8IsR2638MDr7fYWE0nefa3b6XxyPCCsOU%3D&amp;reserved=0 
>>
>>
>> The script should have told you that already, so maybe reinvent that
>> in whatever thing you're using :-) Jani has reported this on
>> #dri-devel, this also holds up CI testing since we're running on top
>> of drm-tip.
>
> Well I used dim push-branch. I even had to rebase once more because I 
> forgot a signed-of on one of the patches, so I'm pretty sure of that.
>
> Haven't seen anything problematic except for that while doing so.

Ok, anyway that one was trivial to fix.

Thanks,
Christian.

>
> Regards,
> Christian.
>
>> -Daniel
>>
>>
>>
>>> Only VMGFX and Nouveau are missing and I'm pretty close to just push
>>> them with my Acked-by since they should not contain any functional 
>>> change.
>>>
>>> Any objections?
>>>
>>> Thanks,
>>> Christian.
>>>
>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>>> With this patch series I am trying to remove GPU address dependency in
>>>> TTM and moving GPU address calculation to individual drm drivers. This
>>>> cleanup will simplify introduction of drm_mem_region/domain work 
>>>> started
>>>> by Brian Welty[1].
>>>>
>>>>
>>>> It would be nice if someone test this for nouveau. Rest of the drivers
>>>> are already tested.
>>>>
>>>> v2:
>>>> * set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL
>>>>
>>>> v3:
>>>> * catch return value of drm_gem_vram_offset() in drm/bochs
>>>> * introduce drm_gem_vram_pg_offset() in vram helper
>>>> * improve nbo->offset calculation for nouveau
>>>>
>>>> v4:
>>>> * minor coding style fixes in amdgpu and radeon
>>>> * remove unnecessary kerneldoc for internal function
>>>>
>>>> v5:
>>>> * rebase on top of drm-misc-next
>>>> * fix return value of drm_gem_vram_pg_offset()
>>>> * add a comment in drm_gem_vram_pg_offset() to clearify why we 
>>>> return 0.
>>>>
>>>> v6:
>>>> * rebase to drm-misc-next
>>>> * removed acked for vmwgfx as there was a small conflict
>>>>
>>>> Nirmoy Das (8):
>>>>     drm/amdgpu: move ttm bo->offset to amdgpu_bo
>>>>     drm/radeon: don't use ttm bo->offset
>>>>     drm/vmwgfx: don't use ttm bo->offset
>>>>     drm/nouveau: don't use ttm bo->offset v3
>>>>     drm/qxl: don't use ttm bo->offset
>>>>     drm/vram-helper: don't use ttm bo->offset v4
>>>>     drm/bochs: use drm_gem_vram_offset to get bo offset v2
>>>>     drm/ttm: do not keep GPU dependent addresses
>>>>
>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 
>>>> ++++++++++++++++-----
>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
>>>>    drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
>>>>    drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
>>>>    drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
>>>>    drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
>>>>    drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
>>>>    drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
>>>>    drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
>>>>    drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
>>>>    drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
>>>>    drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
>>>>    drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
>>>>    drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
>>>>    drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
>>>>    drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
>>>>    drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
>>>>    drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
>>>>    drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
>>>>    drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
>>>>    drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
>>>>    drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
>>>>    drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
>>>>    drivers/gpu/drm/radeon/radeon.h             |  1 +
>>>>    drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
>>>>    drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
>>>>    drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
>>>>    include/drm/ttm/ttm_bo_api.h                |  2 --
>>>>    include/drm/ttm/ttm_bo_driver.h             |  1 -
>>>>    36 files changed, 125 insertions(+), 78 deletions(-)
>>>>
>>>> -- 
>>>> 2.27.0
>>>>
>>
>

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Christian König]

Am 25.06.20 um 18:02 schrieb Christian König:
> Am 25.06.20 um 17:52 schrieb Christian König:
>> Am 25.06.20 um 17:44 schrieb Daniel Vetter:
>>> On Thu, Jun 25, 2020 at 11:50 AM Christian König
>>> <christian.koenig@amd.com> wrote:
>>>> I've pushed patches #1, #2 and #5-#8 of this series to drm-misc-next.
>>> I think you left an unresolved conflict behind in drm-tip, please 
>>> resolve per
>>>
>>> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdrm.pages.freedesktop.org%2Fmaintainer-tools%2Fdrm-tip.html%23resolving-conflicts-when-rebuilding-drm-tip&amp;data=02%7C01%7Cchristian.koenig%40amd.com%7Cc1441a81e9dc4fa4507208d8191ea0f6%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637286966631595300&amp;sdata=pfWS4VgDvU8IsR2638MDr7fYWE0nefa3b6XxyPCCsOU%3D&amp;reserved=0 
>>>
>>>
>>> The script should have told you that already, so maybe reinvent that
>>> in whatever thing you're using :-) Jani has reported this on
>>> #dri-devel, this also holds up CI testing since we're running on top
>>> of drm-tip.
>>
>> Well I used dim push-branch. I even had to rebase once more because I 
>> forgot a signed-of on one of the patches, so I'm pretty sure of that.
>>
>> Haven't seen anything problematic except for that while doing so.
>
> Ok, anyway that one was trivial to fix.

Ran into the next merge conflict. But no surprise that I missed the 
warning, that needs to be in red or so.

Regards,
Christian.

>
> Thanks,
> Christian.
>
>>
>> Regards,
>> Christian.
>>
>>> -Daniel
>>>
>>>
>>>
>>>> Only VMGFX and Nouveau are missing and I'm pretty close to just push
>>>> them with my Acked-by since they should not contain any functional 
>>>> change.
>>>>
>>>> Any objections?
>>>>
>>>> Thanks,
>>>> Christian.
>>>>
>>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
>>>>> With this patch series I am trying to remove GPU address 
>>>>> dependency in
>>>>> TTM and moving GPU address calculation to individual drm drivers. 
>>>>> This
>>>>> cleanup will simplify introduction of drm_mem_region/domain work 
>>>>> started
>>>>> by Brian Welty[1].
>>>>>
>>>>>
>>>>> It would be nice if someone test this for nouveau. Rest of the 
>>>>> drivers
>>>>> are already tested.
>>>>>
>>>>> v2:
>>>>> * set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL
>>>>>
>>>>> v3:
>>>>> * catch return value of drm_gem_vram_offset() in drm/bochs
>>>>> * introduce drm_gem_vram_pg_offset() in vram helper
>>>>> * improve nbo->offset calculation for nouveau
>>>>>
>>>>> v4:
>>>>> * minor coding style fixes in amdgpu and radeon
>>>>> * remove unnecessary kerneldoc for internal function
>>>>>
>>>>> v5:
>>>>> * rebase on top of drm-misc-next
>>>>> * fix return value of drm_gem_vram_pg_offset()
>>>>> * add a comment in drm_gem_vram_pg_offset() to clearify why we 
>>>>> return 0.
>>>>>
>>>>> v6:
>>>>> * rebase to drm-misc-next
>>>>> * removed acked for vmwgfx as there was a small conflict
>>>>>
>>>>> Nirmoy Das (8):
>>>>>     drm/amdgpu: move ttm bo->offset to amdgpu_bo
>>>>>     drm/radeon: don't use ttm bo->offset
>>>>>     drm/vmwgfx: don't use ttm bo->offset
>>>>>     drm/nouveau: don't use ttm bo->offset v3
>>>>>     drm/qxl: don't use ttm bo->offset
>>>>>     drm/vram-helper: don't use ttm bo->offset v4
>>>>>     drm/bochs: use drm_gem_vram_offset to get bo offset v2
>>>>>     drm/ttm: do not keep GPU dependent addresses
>>>>>
>>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
>>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
>>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 
>>>>> ++++++++++++++++-----
>>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
>>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
>>>>>    drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
>>>>>    drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
>>>>>    drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
>>>>>    drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
>>>>>    drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
>>>>>    drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
>>>>>    drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
>>>>>    drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
>>>>>    drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
>>>>>    drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
>>>>>    drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
>>>>>    drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
>>>>>    drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
>>>>>    drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
>>>>>    drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
>>>>>    drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
>>>>>    drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
>>>>>    drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
>>>>>    drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
>>>>>    drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
>>>>>    drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
>>>>>    drivers/gpu/drm/radeon/radeon.h             |  1 +
>>>>>    drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
>>>>>    drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
>>>>>    drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
>>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
>>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
>>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
>>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
>>>>>    include/drm/ttm/ttm_bo_api.h                |  2 --
>>>>>    include/drm/ttm/ttm_bo_driver.h             |  1 -
>>>>>    36 files changed, 125 insertions(+), 78 deletions(-)
>>>>>
>>>>> -- 
>>>>> 2.27.0
>>>>>
>>>
>>
>

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Daniel Vetter]

On Fri, Jun 26, 2020 at 3:04 PM Christian König
<christian.koenig@amd.com> wrote:
>
> Am 25.06.20 um 18:02 schrieb Christian König:
> > Am 25.06.20 um 17:52 schrieb Christian König:
> >> Am 25.06.20 um 17:44 schrieb Daniel Vetter:
> >>> On Thu, Jun 25, 2020 at 11:50 AM Christian König
> >>> <christian.koenig@amd.com> wrote:
> >>>> I've pushed patches #1, #2 and #5-#8 of this series to drm-misc-next.
> >>> I think you left an unresolved conflict behind in drm-tip, please
> >>> resolve per
> >>>
> >>> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdrm.pages.freedesktop.org%2Fmaintainer-tools%2Fdrm-tip.html%23resolving-conflicts-when-rebuilding-drm-tip&amp;data=02%7C01%7Cchristian.koenig%40amd.com%7Cc1441a81e9dc4fa4507208d8191ea0f6%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637286966631595300&amp;sdata=pfWS4VgDvU8IsR2638MDr7fYWE0nefa3b6XxyPCCsOU%3D&amp;reserved=0
> >>>
> >>>
> >>> The script should have told you that already, so maybe reinvent that
> >>> in whatever thing you're using :-) Jani has reported this on
> >>> #dri-devel, this also holds up CI testing since we're running on top
> >>> of drm-tip.
> >>
> >> Well I used dim push-branch. I even had to rebase once more because I
> >> forgot a signed-of on one of the patches, so I'm pretty sure of that.
> >>
> >> Haven't seen anything problematic except for that while doing so.
> >
> > Ok, anyway that one was trivial to fix.
>
> Ran into the next merge conflict. But no surprise that I missed the
> warning, that needs to be in red or so.

Yeah the trouble is that the merging is done after you pushed, we
don't do a test drm-tip rebuild before you push. Reason is that I've
not gotten around to implementing this since years, it's on the
wishlist.

If we'd do a test-rebuild before pushing (something like use all the
normal branches, excpet for $branch use the local one, or a specific
sha1, with git worktree that's all shared so doable), then it would be
a lot more obvious that people have some work to do before they're
done with their push.

Trouble is only rebuilding in dim push isn't enough, since then you
can't test. So we need the full suite of tools be able to work with a
local branch. And once you have a conflict which needs a fixup patch
that's kinda not possible anymore.

But maybe just doing the test-rebuild and if that fails, pointing it
out and requiring a special --promises-I-fix-it-up-right-afterwards
flag might be good enough.

If you know how to print stuff in red if it's a terminal, then
patching warn_or_fail should catch them all.
-Daniel


>
> Regards,
> Christian.
>
> >
> > Thanks,
> > Christian.
> >
> >>
> >> Regards,
> >> Christian.
> >>
> >>> -Daniel
> >>>
> >>>
> >>>
> >>>> Only VMGFX and Nouveau are missing and I'm pretty close to just push
> >>>> them with my Acked-by since they should not contain any functional
> >>>> change.
> >>>>
> >>>> Any objections?
> >>>>
> >>>> Thanks,
> >>>> Christian.
> >>>>
> >>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
> >>>>> With this patch series I am trying to remove GPU address
> >>>>> dependency in
> >>>>> TTM and moving GPU address calculation to individual drm drivers.
> >>>>> This
> >>>>> cleanup will simplify introduction of drm_mem_region/domain work
> >>>>> started
> >>>>> by Brian Welty[1].
> >>>>>
> >>>>>
> >>>>> It would be nice if someone test this for nouveau. Rest of the
> >>>>> drivers
> >>>>> are already tested.
> >>>>>
> >>>>> v2:
> >>>>> * set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL
> >>>>>
> >>>>> v3:
> >>>>> * catch return value of drm_gem_vram_offset() in drm/bochs
> >>>>> * introduce drm_gem_vram_pg_offset() in vram helper
> >>>>> * improve nbo->offset calculation for nouveau
> >>>>>
> >>>>> v4:
> >>>>> * minor coding style fixes in amdgpu and radeon
> >>>>> * remove unnecessary kerneldoc for internal function
> >>>>>
> >>>>> v5:
> >>>>> * rebase on top of drm-misc-next
> >>>>> * fix return value of drm_gem_vram_pg_offset()
> >>>>> * add a comment in drm_gem_vram_pg_offset() to clearify why we
> >>>>> return 0.
> >>>>>
> >>>>> v6:
> >>>>> * rebase to drm-misc-next
> >>>>> * removed acked for vmwgfx as there was a small conflict
> >>>>>
> >>>>> Nirmoy Das (8):
> >>>>>     drm/amdgpu: move ttm bo->offset to amdgpu_bo
> >>>>>     drm/radeon: don't use ttm bo->offset
> >>>>>     drm/vmwgfx: don't use ttm bo->offset
> >>>>>     drm/nouveau: don't use ttm bo->offset v3
> >>>>>     drm/qxl: don't use ttm bo->offset
> >>>>>     drm/vram-helper: don't use ttm bo->offset v4
> >>>>>     drm/bochs: use drm_gem_vram_offset to get bo offset v2
> >>>>>     drm/ttm: do not keep GPU dependent addresses
> >>>>>
> >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
> >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
> >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30
> >>>>> ++++++++++++++++-----
> >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
> >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
> >>>>>    drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
> >>>>>    drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
> >>>>>    drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
> >>>>>    drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
> >>>>>    drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
> >>>>>    drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
> >>>>>    drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
> >>>>>    drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
> >>>>>    drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
> >>>>>    drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
> >>>>>    drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
> >>>>>    drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
> >>>>>    drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
> >>>>>    drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
> >>>>>    drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
> >>>>>    drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
> >>>>>    drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
> >>>>>    drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
> >>>>>    drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
> >>>>>    drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
> >>>>>    drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
> >>>>>    drivers/gpu/drm/radeon/radeon.h             |  1 +
> >>>>>    drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
> >>>>>    drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
> >>>>>    drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
> >>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
> >>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
> >>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
> >>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
> >>>>>    include/drm/ttm/ttm_bo_api.h                |  2 --
> >>>>>    include/drm/ttm/ttm_bo_driver.h             |  1 -
> >>>>>    36 files changed, 125 insertions(+), 78 deletions(-)
> >>>>>
> >>>>> --
> >>>>> 2.27.0
> >>>>>
> >>>
> >>
> >
>


-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [RESEND] [PATCH v6 0/8] do not store GPU address in TTM

[Daniel Vetter]

On Fri, Jun 26, 2020 at 3:12 PM Daniel Vetter <daniel@ffwll.ch> wrote:
>
> On Fri, Jun 26, 2020 at 3:04 PM Christian König
> <christian.koenig@amd.com> wrote:
> >
> > Am 25.06.20 um 18:02 schrieb Christian König:
> > > Am 25.06.20 um 17:52 schrieb Christian König:
> > >> Am 25.06.20 um 17:44 schrieb Daniel Vetter:
> > >>> On Thu, Jun 25, 2020 at 11:50 AM Christian König
> > >>> <christian.koenig@amd.com> wrote:
> > >>>> I've pushed patches #1, #2 and #5-#8 of this series to drm-misc-next.
> > >>> I think you left an unresolved conflict behind in drm-tip, please
> > >>> resolve per
> > >>>
> > >>> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdrm.pages.freedesktop.org%2Fmaintainer-tools%2Fdrm-tip.html%23resolving-conflicts-when-rebuilding-drm-tip&amp;data=02%7C01%7Cchristian.koenig%40amd.com%7Cc1441a81e9dc4fa4507208d8191ea0f6%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637286966631595300&amp;sdata=pfWS4VgDvU8IsR2638MDr7fYWE0nefa3b6XxyPCCsOU%3D&amp;reserved=0
> > >>>
> > >>>
> > >>> The script should have told you that already, so maybe reinvent that
> > >>> in whatever thing you're using :-) Jani has reported this on
> > >>> #dri-devel, this also holds up CI testing since we're running on top
> > >>> of drm-tip.
> > >>
> > >> Well I used dim push-branch. I even had to rebase once more because I
> > >> forgot a signed-of on one of the patches, so I'm pretty sure of that.
> > >>
> > >> Haven't seen anything problematic except for that while doing so.
> > >
> > > Ok, anyway that one was trivial to fix.
> >
> > Ran into the next merge conflict. But no surprise that I missed the
> > warning, that needs to be in red or so.
>
> Yeah the trouble is that the merging is done after you pushed, we
> don't do a test drm-tip rebuild before you push. Reason is that I've
> not gotten around to implementing this since years, it's on the
> wishlist.
>
> If we'd do a test-rebuild before pushing (something like use all the
> normal branches, excpet for $branch use the local one, or a specific
> sha1, with git worktree that's all shared so doable), then it would be
> a lot more obvious that people have some work to do before they're
> done with their push.
>
> Trouble is only rebuilding in dim push isn't enough, since then you
> can't test. So we need the full suite of tools be able to work with a
> local branch. And once you have a conflict which needs a fixup patch
> that's kinda not possible anymore.
>
> But maybe just doing the test-rebuild and if that fails, pointing it
> out and requiring a special --promises-I-fix-it-up-right-afterwards
> flag might be good enough.
>
> If you know how to print stuff in red if it's a terminal, then
> patching warn_or_fail should catch them all.

Uh I mean patching echoerr.
-Daniel

> -Daniel
>
>
> >
> > Regards,
> > Christian.
> >
> > >
> > > Thanks,
> > > Christian.
> > >
> > >>
> > >> Regards,
> > >> Christian.
> > >>
> > >>> -Daniel
> > >>>
> > >>>
> > >>>
> > >>>> Only VMGFX and Nouveau are missing and I'm pretty close to just push
> > >>>> them with my Acked-by since they should not contain any functional
> > >>>> change.
> > >>>>
> > >>>> Any objections?
> > >>>>
> > >>>> Thanks,
> > >>>> Christian.
> > >>>>
> > >>>> Am 24.06.20 um 20:26 schrieb Nirmoy Das:
> > >>>>> With this patch series I am trying to remove GPU address
> > >>>>> dependency in
> > >>>>> TTM and moving GPU address calculation to individual drm drivers.
> > >>>>> This
> > >>>>> cleanup will simplify introduction of drm_mem_region/domain work
> > >>>>> started
> > >>>>> by Brian Welty[1].
> > >>>>>
> > >>>>>
> > >>>>> It would be nice if someone test this for nouveau. Rest of the
> > >>>>> drivers
> > >>>>> are already tested.
> > >>>>>
> > >>>>> v2:
> > >>>>> * set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL
> > >>>>>
> > >>>>> v3:
> > >>>>> * catch return value of drm_gem_vram_offset() in drm/bochs
> > >>>>> * introduce drm_gem_vram_pg_offset() in vram helper
> > >>>>> * improve nbo->offset calculation for nouveau
> > >>>>>
> > >>>>> v4:
> > >>>>> * minor coding style fixes in amdgpu and radeon
> > >>>>> * remove unnecessary kerneldoc for internal function
> > >>>>>
> > >>>>> v5:
> > >>>>> * rebase on top of drm-misc-next
> > >>>>> * fix return value of drm_gem_vram_pg_offset()
> > >>>>> * add a comment in drm_gem_vram_pg_offset() to clearify why we
> > >>>>> return 0.
> > >>>>>
> > >>>>> v6:
> > >>>>> * rebase to drm-misc-next
> > >>>>> * removed acked for vmwgfx as there was a small conflict
> > >>>>>
> > >>>>> Nirmoy Das (8):
> > >>>>>     drm/amdgpu: move ttm bo->offset to amdgpu_bo
> > >>>>>     drm/radeon: don't use ttm bo->offset
> > >>>>>     drm/vmwgfx: don't use ttm bo->offset
> > >>>>>     drm/nouveau: don't use ttm bo->offset v3
> > >>>>>     drm/qxl: don't use ttm bo->offset
> > >>>>>     drm/vram-helper: don't use ttm bo->offset v4
> > >>>>>     drm/bochs: use drm_gem_vram_offset to get bo offset v2
> > >>>>>     drm/ttm: do not keep GPU dependent addresses
> > >>>>>
> > >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ++++++++++++++--
> > >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
> > >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30
> > >>>>> ++++++++++++++++-----
> > >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
> > >>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  4 +--
> > >>>>>    drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
> > >>>>>    drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
> > >>>>>    drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
> > >>>>>    drivers/gpu/drm/nouveau/dispnv04/disp.c     |  3 ++-
> > >>>>>    drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
> > >>>>>    drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
> > >>>>>    drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
> > >>>>>    drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
> > >>>>>    drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
> > >>>>>    drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
> > >>>>>    drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
> > >>>>>    drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
> > >>>>>    drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
> > >>>>>    drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
> > >>>>>    drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
> > >>>>>    drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
> > >>>>>    drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
> > >>>>>    drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
> > >>>>>    drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
> > >>>>>    drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
> > >>>>>    drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
> > >>>>>    drivers/gpu/drm/radeon/radeon.h             |  1 +
> > >>>>>    drivers/gpu/drm/radeon/radeon_object.h      | 16 ++++++++++-
> > >>>>>    drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
> > >>>>>    drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
> > >>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
> > >>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
> > >>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
> > >>>>>    drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
> > >>>>>    include/drm/ttm/ttm_bo_api.h                |  2 --
> > >>>>>    include/drm/ttm/ttm_bo_driver.h             |  1 -
> > >>>>>    36 files changed, 125 insertions(+), 78 deletions(-)
> > >>>>>
> > >>>>> --
> > >>>>> 2.27.0
> > >>>>>
> > >>>
> > >>
> > >
> >
>
>
> --
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch



-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v5 0/8] do not store GPU address in TTM

[Nirmoy Das]

With this patch series I am trying to remove GPU address dependency in
TTM and moving GPU address calculation to individual drm drivers. This
cleanup will simplify introduction of drm_mem_region/domain work started
by Brian Welty[1].


It would be nice if someone test this for nouveau. Rest of the drivers
are already tested.

v2:
* set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL

v3:
* catch return value of drm_gem_vram_offset() in drm/bochs
* introduce drm_gem_vram_pg_offset() in vram helper
* improve nbo->offset calculation for nouveau

v4:
* minor coding style fixes in amdgpu and radeon
* remove unnecessary kerneldoc for internal function

v5:
* rebase on top of drm-misc-next
* fix return value of drm_gem_vram_pg_offset()
* add a comment in drm_gem_vram_pg_offset() to clearify why we return 0.

Nirmoy Das (8):
  drm/amdgpu: move ttm bo->offset to amdgpu_bo
  drm/radeon: don't use ttm bo->offset
  drm/vmwgfx: don't use ttm bo->offset
  drm/nouveau: don't use ttm bo->offset v3
  drm/qxl: don't use ttm bo->offset
  drm/vram-helper: don't use ttm bo->offset v4
  drm/bochs: use drm_gem_vram_offset to get bo offset v2
  drm/ttm: do not keep GPU dependent addresses

 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 22 ++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 29 ++++++++++++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  2 +-
 drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
 drivers/gpu/drm/drm_gem_vram_helper.c       | 11 +++++++-
 drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
 drivers/gpu/drm/nouveau/dispnv04/disp.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
 drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
 drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
 drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
 drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
 drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
 drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
 drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
 drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
 drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
 drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
 drivers/gpu/drm/radeon/radeon.h             |  1 +
 drivers/gpu/drm/radeon/radeon_object.h      | 16 +++++++++++-
 drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
 drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
 drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
 include/drm/ttm/ttm_bo_api.h                |  2 --
 include/drm/ttm/ttm_bo_driver.h             |  1 -
 36 files changed, 121 insertions(+), 77 deletions(-)

--
2.25.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy Das]

GPU address handling is device specific and should be handle by its device
driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
 include/drm/ttm/ttm_bo_api.h    | 2 --
 include/drm/ttm/ttm_bo_driver.h | 1 -
 3 files changed, 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index f73b81c2576e..f78cfc76ad78 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
 	drm_printf(p, "    has_type: %d\n", man->has_type);
 	drm_printf(p, "    use_type: %d\n", man->use_type);
 	drm_printf(p, "    flags: 0x%08X\n", man->flags);
-	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
 	drm_printf(p, "    size: %llu\n", man->size);
 	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
 	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
@@ -343,12 +342,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 moved:
 	bo->evicted = false;
 
-	if (bo->mem.mm_node)
-		bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		    bdev->man[bo->mem.mem_type].gpu_offset;
-	else
-		bo->offset = 0;
-
 	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
 	return 0;
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index b9bc1b00142e..d6f39ee5bf5d 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -213,8 +213,6 @@ struct ttm_buffer_object {
 	 * either of these locks held.
 	 */
 
-	uint64_t offset; /* GPU address space is independent of CPU word size */
-
 	struct sg_table *sg;
 };
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index c9e0fd09f4b2..c8ce6c181abe 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
 	bool has_type;
 	bool use_type;
 	uint32_t flags;
-	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
2.25.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v4 0/8] do not store GPU address in TTM

[Nirmoy Das]

With this patch series I am trying to remove GPU address dependency in
TTM and moving GPU address calculation to individual drm drivers. This
cleanup will simplify introduction of drm_mem_region/domain work started
by Brian Welty[1].


It would be nice if someone test this for nouveau. Rest of the drivers
are already tested.

v2:
* set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL

v3:
* catch return value of drm_gem_vram_offset() in drm/bochs
* introduce drm_gem_vram_pg_offset() in vram helper
* improve nbo->offset calculation for nouveau

v4:
* minor coding style fixes in amdgpu and radeon
* remove unnecessary kerneldoc for internal function

Nirmoy Das (8):
  drm/amdgpu: move ttm bo->offset to amdgpu_bo
  drm/radeon: don't use ttm bo->offset
  drm/vmwgfx: don't use ttm bo->offset
  drm/nouveau: don't use ttm bo->offset v3
  drm/qxl: don't use ttm bo->offset
  drm/vram-helper: don't use ttm bo->offset v3
  drm/bochs: use drm_gem_vram_offset to get bo offset v2
  drm/ttm: do not keep GPU dependent addresses

 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 22 ++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 29 ++++++++++++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
 drivers/gpu/drm/bochs/bochs_kms.c           |  7 ++++-
 drivers/gpu/drm/drm_gem_vram_helper.c       |  9 ++++++-
 drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
 drivers/gpu/drm/nouveau/dispnv04/disp.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
 drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
 drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
 drivers/gpu/drm/nouveau/nouveau_bo.c        |  8 ++++++
 drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
 drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
 drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
 drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
 drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
 drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
 drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
 drivers/gpu/drm/radeon/radeon.h             |  1 +
 drivers/gpu/drm/radeon/radeon_object.h      | 16 +++++++++++-
 drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
 drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
 drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
 include/drm/ttm/ttm_bo_api.h                |  2 --
 include/drm/ttm/ttm_bo_driver.h             |  1 -
 35 files changed, 118 insertions(+), 76 deletions(-)

--
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy Das]

GPU address handling is device specific and should be handle by its device
driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
 include/drm/ttm/ttm_bo_api.h    | 2 --
 include/drm/ttm/ttm_bo_driver.h | 1 -
 3 files changed, 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 6d1e91be9c78..9f24fb287d71 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
 	drm_printf(p, "    has_type: %d\n", man->has_type);
 	drm_printf(p, "    use_type: %d\n", man->use_type);
 	drm_printf(p, "    flags: 0x%08X\n", man->flags);
-	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
 	drm_printf(p, "    size: %llu\n", man->size);
 	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
 	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
@@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 moved:
 	bo->evicted = false;
 
-	if (bo->mem.mm_node)
-		bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		    bdev->man[bo->mem.mem_type].gpu_offset;
-	else
-		bo->offset = 0;
-
 	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
 	return 0;
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index b9bc1b00142e..d6f39ee5bf5d 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -213,8 +213,6 @@ struct ttm_buffer_object {
 	 * either of these locks held.
 	 */
 
-	uint64_t offset; /* GPU address space is independent of CPU word size */
-
 	struct sg_table *sg;
 };
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index c9e0fd09f4b2..c8ce6c181abe 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
 	bool has_type;
 	bool use_type;
 	uint32_t flags;
-	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy Das]

GPU address handling is device specific and should be handle by its device
driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
 include/drm/ttm/ttm_bo_api.h    | 2 --
 include/drm/ttm/ttm_bo_driver.h | 1 -
 3 files changed, 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 6d1e91be9c78..9f24fb287d71 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
 	drm_printf(p, "    has_type: %d\n", man->has_type);
 	drm_printf(p, "    use_type: %d\n", man->use_type);
 	drm_printf(p, "    flags: 0x%08X\n", man->flags);
-	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
 	drm_printf(p, "    size: %llu\n", man->size);
 	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
 	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
@@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 moved:
 	bo->evicted = false;
 
-	if (bo->mem.mm_node)
-		bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		    bdev->man[bo->mem.mem_type].gpu_offset;
-	else
-		bo->offset = 0;
-
 	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
 	return 0;
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index b9bc1b00142e..d6f39ee5bf5d 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -213,8 +213,6 @@ struct ttm_buffer_object {
 	 * either of these locks held.
 	 */
 
-	uint64_t offset; /* GPU address space is independent of CPU word size */
-
 	struct sg_table *sg;
 };
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index c9e0fd09f4b2..c8ce6c181abe 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
 	bool has_type;
 	bool use_type;
 	uint32_t flags;
-	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
2.25.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 05.03.20 um 14:29 schrieb Nirmoy Das:
> GPU address handling is device specific and should be handle by its device
> driver.
>
> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>

Reviewed-by: Christian König <christian.koenig@amd.com>

> ---
>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>   include/drm/ttm/ttm_bo_api.h    | 2 --
>   include/drm/ttm/ttm_bo_driver.h | 1 -
>   3 files changed, 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 6d1e91be9c78..9f24fb287d71 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
>   	drm_printf(p, "    has_type: %d\n", man->has_type);
>   	drm_printf(p, "    use_type: %d\n", man->use_type);
>   	drm_printf(p, "    flags: 0x%08X\n", man->flags);
> -	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>   	drm_printf(p, "    size: %llu\n", man->size);
>   	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
>   	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
>   moved:
>   	bo->evicted = false;
>   
> -	if (bo->mem.mm_node)
> -		bo->offset = (bo->mem.start << PAGE_SHIFT) +
> -		    bdev->man[bo->mem.mem_type].gpu_offset;
> -	else
> -		bo->offset = 0;
> -
>   	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>   	return 0;
>   
> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
> index b9bc1b00142e..d6f39ee5bf5d 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>   	 * either of these locks held.
>   	 */
>   
> -	uint64_t offset; /* GPU address space is independent of CPU word size */
> -
>   	struct sg_table *sg;
>   };
>   
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index c9e0fd09f4b2..c8ce6c181abe 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>   	bool has_type;
>   	bool use_type;
>   	uint32_t flags;
> -	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
>   	uint64_t size;
>   	uint32_t available_caching;
>   	uint32_t default_caching;

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 05.03.20 um 14:29 schrieb Nirmoy Das:
> GPU address handling is device specific and should be handle by its device
> driver.
>
> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>

Reviewed-by: Christian König <christian.koenig@amd.com>

> ---
>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>   include/drm/ttm/ttm_bo_api.h    | 2 --
>   include/drm/ttm/ttm_bo_driver.h | 1 -
>   3 files changed, 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 6d1e91be9c78..9f24fb287d71 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
>   	drm_printf(p, "    has_type: %d\n", man->has_type);
>   	drm_printf(p, "    use_type: %d\n", man->use_type);
>   	drm_printf(p, "    flags: 0x%08X\n", man->flags);
> -	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>   	drm_printf(p, "    size: %llu\n", man->size);
>   	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
>   	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
>   moved:
>   	bo->evicted = false;
>   
> -	if (bo->mem.mm_node)
> -		bo->offset = (bo->mem.start << PAGE_SHIFT) +
> -		    bdev->man[bo->mem.mem_type].gpu_offset;
> -	else
> -		bo->offset = 0;
> -
>   	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>   	return 0;
>   
> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
> index b9bc1b00142e..d6f39ee5bf5d 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>   	 * either of these locks held.
>   	 */
>   
> -	uint64_t offset; /* GPU address space is independent of CPU word size */
> -
>   	struct sg_table *sg;
>   };
>   
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index c9e0fd09f4b2..c8ce6c181abe 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>   	bool has_type;
>   	bool use_type;
>   	uint32_t flags;
> -	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
>   	uint64_t size;
>   	uint32_t available_caching;
>   	uint32_t default_caching;

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v3 0/8] do not store GPU address in TTM

[Nirmoy Das]

With this patch series I am trying to remove GPU address dependency in
TTM and moving GPU address calculation to individual drm drivers.

I tested this patch series on qxl, bochs and amdgpu. Christian tested it on radeon HW.
It would be nice if someone test this for nouveau and vmgfx.

v2:
* set bo->offset = 0 for drm/nouveau if bo->mem.mm_node == NULL

v3:
* catch return value of drm_gem_vram_offset() in drm/bochs
* introduce drm_gem_vram_pg_offset() in vram helper
* improve nbo->offset calculation for nouveau

Nirmoy Das (8):
  drm/amdgpu: move ttm bo->offset to amdgpu_bo
  drm/radeon: don't use ttm bo->offset
  drm/vmwgfx: don't use ttm bo->offset
  drm/nouveau: don't use ttm bo->offset
  drm/qxl: don't use ttm bo->offset
  drm/vram-helper: don't use ttm bo->offset
  drm/bochs: use drm_gem_vram_offset to get bo offset
  drm/ttm: do not keep GPU dependent addresses

 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 22 ++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 29 ++++++++++++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
 drivers/gpu/drm/bochs/bochs_kms.c           |  2 +-
 drivers/gpu/drm/drm_gem_vram_helper.c       |  2 +-
 drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
 drivers/gpu/drm/nouveau/dispnv04/disp.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
 drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
 drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
 drivers/gpu/drm/nouveau/nouveau_bo.c        |  1 +
 drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
 drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
 drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
 drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
 drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
 drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
 drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
 drivers/gpu/drm/radeon/radeon.h             |  1 +
 drivers/gpu/drm/radeon/radeon_object.h      | 16 +++++++++++-
 drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
 drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
 drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
 include/drm/ttm/ttm_bo_api.h                |  2 --
 include/drm/ttm/ttm_bo_driver.h             |  1 -
 35 files changed, 99 insertions(+), 76 deletions(-)

--
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy Das]

GPU address handling is device specific and should be handle by its device
driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
 include/drm/ttm/ttm_bo_api.h    | 2 --
 include/drm/ttm/ttm_bo_driver.h | 1 -
 3 files changed, 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 151edfd8de77..d5885cd609a3 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
 	drm_printf(p, "    has_type: %d\n", man->has_type);
 	drm_printf(p, "    use_type: %d\n", man->use_type);
 	drm_printf(p, "    flags: 0x%08X\n", man->flags);
-	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
 	drm_printf(p, "    size: %llu\n", man->size);
 	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
 	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
@@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 moved:
 	bo->evicted = false;
 
-	if (bo->mem.mm_node)
-		bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		    bdev->man[bo->mem.mem_type].gpu_offset;
-	else
-		bo->offset = 0;
-
 	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
 	return 0;
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index b9bc1b00142e..d6f39ee5bf5d 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -213,8 +213,6 @@ struct ttm_buffer_object {
 	 * either of these locks held.
 	 */
 
-	uint64_t offset; /* GPU address space is independent of CPU word size */
-
 	struct sg_table *sg;
 };
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index c9e0fd09f4b2..c8ce6c181abe 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
 	bool has_type;
 	bool use_type;
 	uint32_t flags;
-	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy Das]

GPU address handling is device specific and should be handle by its device
driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
 include/drm/ttm/ttm_bo_api.h    | 2 --
 include/drm/ttm/ttm_bo_driver.h | 1 -
 3 files changed, 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 151edfd8de77..d5885cd609a3 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
 	drm_printf(p, "    has_type: %d\n", man->has_type);
 	drm_printf(p, "    use_type: %d\n", man->use_type);
 	drm_printf(p, "    flags: 0x%08X\n", man->flags);
-	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
 	drm_printf(p, "    size: %llu\n", man->size);
 	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
 	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
@@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 moved:
 	bo->evicted = false;
 
-	if (bo->mem.mm_node)
-		bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		    bdev->man[bo->mem.mem_type].gpu_offset;
-	else
-		bo->offset = 0;
-
 	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
 	return 0;
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index b9bc1b00142e..d6f39ee5bf5d 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -213,8 +213,6 @@ struct ttm_buffer_object {
 	 * either of these locks held.
 	 */
 
-	uint64_t offset; /* GPU address space is independent of CPU word size */
-
 	struct sg_table *sg;
 };
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index c9e0fd09f4b2..c8ce6c181abe 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
 	bool has_type;
 	bool use_type;
 	uint32_t flags;
-	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
2.25.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 0/8] do not store GPU address in TTM

[Nirmoy Das]

With this patch series I am trying to remove GPU address dependency in
TTM and moving GPU address calculation to individual drm drivers.

I tested this patch series on qxl, bochs and amdgpu. Christian tested it on radeon HW.
It would be nice if someone test this for nouveau and vmgfx.

Nirmoy Das (8):
  drm/amdgpu: move ttm bo->offset to amdgpu_bo
  drm/radeon: don't use ttm bo->offset
  drm/vmwgfx: don't use ttm bo->offset
  drm/nouveau: don't use ttm bo->offset
  drm/qxl: don't use ttm bo->offset
  drm/vram-helper: don't use ttm bo->offset
  drm/bochs: use drm_gem_vram_offset to get bo offset
  drm/ttm: do not keep GPU dependent addresses

 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 22 ++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 29 ++++++++++++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
 drivers/gpu/drm/bochs/bochs_kms.c           |  2 +-
 drivers/gpu/drm/drm_gem_vram_helper.c       |  2 +-
 drivers/gpu/drm/nouveau/dispnv04/crtc.c     |  6 ++---
 drivers/gpu/drm/nouveau/dispnv04/disp.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv04/overlay.c  |  6 ++---
 drivers/gpu/drm/nouveau/dispnv50/base507c.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/core507d.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/ovly507e.c |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndw.c     |  2 +-
 drivers/gpu/drm/nouveau/dispnv50/wndwc37e.c |  2 +-
 drivers/gpu/drm/nouveau/nouveau_abi16.c     |  8 +++---
 drivers/gpu/drm/nouveau/nouveau_bo.c        |  1 +
 drivers/gpu/drm/nouveau/nouveau_bo.h        |  3 +++
 drivers/gpu/drm/nouveau/nouveau_chan.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_dmem.c      |  2 +-
 drivers/gpu/drm/nouveau/nouveau_fbcon.c     |  2 +-
 drivers/gpu/drm/nouveau/nouveau_gem.c       | 10 +++----
 drivers/gpu/drm/qxl/qxl_drv.h               |  6 ++---
 drivers/gpu/drm/qxl/qxl_kms.c               |  5 ++--
 drivers/gpu/drm/qxl/qxl_object.h            |  5 ----
 drivers/gpu/drm/qxl/qxl_ttm.c               |  9 -------
 drivers/gpu/drm/radeon/radeon.h             |  1 +
 drivers/gpu/drm/radeon/radeon_object.h      | 16 +++++++++++-
 drivers/gpu/drm/radeon/radeon_ttm.c         |  4 +--
 drivers/gpu/drm/ttm/ttm_bo.c                |  7 -----
 drivers/gpu/drm/vmwgfx/vmwgfx_bo.c          |  4 +--
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c     |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c        |  2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c  |  2 --
 include/drm/ttm/ttm_bo_api.h                |  2 --
 include/drm/ttm/ttm_bo_driver.h             |  1 -
 35 files changed, 99 insertions(+), 76 deletions(-)

--
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy Das]

GPU address handling is device specific and should be handle by its device
driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
 include/drm/ttm/ttm_bo_api.h    | 2 --
 include/drm/ttm/ttm_bo_driver.h | 1 -
 3 files changed, 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 151edfd8de77..d5885cd609a3 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
 	drm_printf(p, "    has_type: %d\n", man->has_type);
 	drm_printf(p, "    use_type: %d\n", man->use_type);
 	drm_printf(p, "    flags: 0x%08X\n", man->flags);
-	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
 	drm_printf(p, "    size: %llu\n", man->size);
 	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
 	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
@@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 moved:
 	bo->evicted = false;
 
-	if (bo->mem.mm_node)
-		bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		    bdev->man[bo->mem.mem_type].gpu_offset;
-	else
-		bo->offset = 0;
-
 	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
 	return 0;
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index b9bc1b00142e..d6f39ee5bf5d 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -213,8 +213,6 @@ struct ttm_buffer_object {
 	 * either of these locks held.
 	 */
 
-	uint64_t offset; /* GPU address space is independent of CPU word size */
-
 	struct sg_table *sg;
 };
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index c9e0fd09f4b2..c8ce6c181abe 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
 	bool has_type;
 	bool use_type;
 	uint32_t flags;
-	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy Das]

GPU address handling is device specific and should be handle by its device
driver.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
 include/drm/ttm/ttm_bo_api.h    | 2 --
 include/drm/ttm/ttm_bo_driver.h | 1 -
 3 files changed, 10 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 151edfd8de77..d5885cd609a3 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
 	drm_printf(p, "    has_type: %d\n", man->has_type);
 	drm_printf(p, "    use_type: %d\n", man->use_type);
 	drm_printf(p, "    flags: 0x%08X\n", man->flags);
-	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
 	drm_printf(p, "    size: %llu\n", man->size);
 	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
 	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
@@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 moved:
 	bo->evicted = false;
 
-	if (bo->mem.mm_node)
-		bo->offset = (bo->mem.start << PAGE_SHIFT) +
-		    bdev->man[bo->mem.mem_type].gpu_offset;
-	else
-		bo->offset = 0;
-
 	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
 	return 0;
 
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index b9bc1b00142e..d6f39ee5bf5d 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -213,8 +213,6 @@ struct ttm_buffer_object {
 	 * either of these locks held.
 	 */
 
-	uint64_t offset; /* GPU address space is independent of CPU word size */
-
 	struct sg_table *sg;
 };
 
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index c9e0fd09f4b2..c8ce6c181abe 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
 	bool has_type;
 	bool use_type;
 	uint32_t flags;
-	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
2.25.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 2710 bytes --]

Hi

Am 17.02.20 um 16:04 schrieb Nirmoy Das:
> GPU address handling is device specific and should be handle by its device
> driver.
> 
> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
> ---
>  drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>  include/drm/ttm/ttm_bo_api.h    | 2 --
>  include/drm/ttm/ttm_bo_driver.h | 1 -
>  3 files changed, 10 deletions(-)
> 
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 151edfd8de77..d5885cd609a3 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
>  	drm_printf(p, "    has_type: %d\n", man->has_type);
>  	drm_printf(p, "    use_type: %d\n", man->use_type);
>  	drm_printf(p, "    flags: 0x%08X\n", man->flags);
> -	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>  	drm_printf(p, "    size: %llu\n", man->size);
>  	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
>  	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
>  moved:
>  	bo->evicted = false;
>  
> -	if (bo->mem.mm_node)
> -		bo->offset = (bo->mem.start << PAGE_SHIFT) +
> -		    bdev->man[bo->mem.mem_type].gpu_offset;
> -	else
> -		bo->offset = 0;
> -

After moving this into users, the else branch has been lost. Is
'bo->mem.mm_node' always true?

Best regards
Thomas

>  	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>  	return 0;
>  
> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
> index b9bc1b00142e..d6f39ee5bf5d 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>  	 * either of these locks held.
>  	 */
>  
> -	uint64_t offset; /* GPU address space is independent of CPU word size */
> -
>  	struct sg_table *sg;
>  };
>  
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index c9e0fd09f4b2..c8ce6c181abe 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>  	bool has_type;
>  	bool use_type;
>  	uint32_t flags;
> -	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
>  	uint64_t size;
>  	uint32_t available_caching;
>  	uint32_t default_caching;
> 

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 2710 bytes --]

Hi

Am 17.02.20 um 16:04 schrieb Nirmoy Das:
> GPU address handling is device specific and should be handle by its device
> driver.
> 
> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
> ---
>  drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>  include/drm/ttm/ttm_bo_api.h    | 2 --
>  include/drm/ttm/ttm_bo_driver.h | 1 -
>  3 files changed, 10 deletions(-)
> 
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 151edfd8de77..d5885cd609a3 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
>  	drm_printf(p, "    has_type: %d\n", man->has_type);
>  	drm_printf(p, "    use_type: %d\n", man->use_type);
>  	drm_printf(p, "    flags: 0x%08X\n", man->flags);
> -	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>  	drm_printf(p, "    size: %llu\n", man->size);
>  	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
>  	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
>  moved:
>  	bo->evicted = false;
>  
> -	if (bo->mem.mm_node)
> -		bo->offset = (bo->mem.start << PAGE_SHIFT) +
> -		    bdev->man[bo->mem.mem_type].gpu_offset;
> -	else
> -		bo->offset = 0;
> -

After moving this into users, the else branch has been lost. Is
'bo->mem.mm_node' always true?

Best regards
Thomas

>  	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>  	return 0;
>  
> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
> index b9bc1b00142e..d6f39ee5bf5d 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>  	 * either of these locks held.
>  	 */
>  
> -	uint64_t offset; /* GPU address space is independent of CPU word size */
> -
>  	struct sg_table *sg;
>  };
>  
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index c9e0fd09f4b2..c8ce6c181abe 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>  	bool has_type;
>  	bool use_type;
>  	uint32_t flags;
> -	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
>  	uint64_t size;
>  	uint32_t available_caching;
>  	uint32_t default_caching;
> 

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 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

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
> Hi
>
> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>> GPU address handling is device specific and should be handle by its device
>> driver.
>>
>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>> ---
>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>   3 files changed, 10 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
>> index 151edfd8de77..d5885cd609a3 100644
>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
>>   	drm_printf(p, "    has_type: %d\n", man->has_type);
>>   	drm_printf(p, "    use_type: %d\n", man->use_type);
>>   	drm_printf(p, "    flags: 0x%08X\n", man->flags);
>> -	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>   	drm_printf(p, "    size: %llu\n", man->size);
>>   	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
>>   	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
>>   moved:
>>   	bo->evicted = false;
>>   
>> -	if (bo->mem.mm_node)
>> -		bo->offset = (bo->mem.start << PAGE_SHIFT) +
>> -		    bdev->man[bo->mem.mem_type].gpu_offset;
>> -	else
>> -		bo->offset = 0;
>> -
> After moving this into users, the else branch has been lost. Is
> 'bo->mem.mm_node' always true?

At least for the amdgpu and radeon use cases, yes.

But that is a rather good question I mean for it is illegal to get the 
GPU BO address if it is inaccessible (e.g. in the system domain).

Could be that some driver relied on the behavior to get 0 for the system 
domain here.

Regards,
Christian.

>
> Best regards
> Thomas
>
>>   	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>   	return 0;
>>   
>> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
>> index b9bc1b00142e..d6f39ee5bf5d 100644
>> --- a/include/drm/ttm/ttm_bo_api.h
>> +++ b/include/drm/ttm/ttm_bo_api.h
>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>   	 * either of these locks held.
>>   	 */
>>   
>> -	uint64_t offset; /* GPU address space is independent of CPU word size */
>> -
>>   	struct sg_table *sg;
>>   };
>>   
>> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
>> index c9e0fd09f4b2..c8ce6c181abe 100644
>> --- a/include/drm/ttm/ttm_bo_driver.h
>> +++ b/include/drm/ttm/ttm_bo_driver.h
>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>   	bool has_type;
>>   	bool use_type;
>>   	uint32_t flags;
>> -	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
>>   	uint64_t size;
>>   	uint32_t available_caching;
>>   	uint32_t default_caching;
>>

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
> Hi
>
> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>> GPU address handling is device specific and should be handle by its device
>> driver.
>>
>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>> ---
>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>   3 files changed, 10 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
>> index 151edfd8de77..d5885cd609a3 100644
>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, struct drm_printer *p
>>   	drm_printf(p, "    has_type: %d\n", man->has_type);
>>   	drm_printf(p, "    use_type: %d\n", man->use_type);
>>   	drm_printf(p, "    flags: 0x%08X\n", man->flags);
>> -	drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>   	drm_printf(p, "    size: %llu\n", man->size);
>>   	drm_printf(p, "    available_caching: 0x%08X\n", man->available_caching);
>>   	drm_printf(p, "    default_caching: 0x%08X\n", man->default_caching);
>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
>>   moved:
>>   	bo->evicted = false;
>>   
>> -	if (bo->mem.mm_node)
>> -		bo->offset = (bo->mem.start << PAGE_SHIFT) +
>> -		    bdev->man[bo->mem.mem_type].gpu_offset;
>> -	else
>> -		bo->offset = 0;
>> -
> After moving this into users, the else branch has been lost. Is
> 'bo->mem.mm_node' always true?

At least for the amdgpu and radeon use cases, yes.

But that is a rather good question I mean for it is illegal to get the 
GPU BO address if it is inaccessible (e.g. in the system domain).

Could be that some driver relied on the behavior to get 0 for the system 
domain here.

Regards,
Christian.

>
> Best regards
> Thomas
>
>>   	ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>   	return 0;
>>   
>> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
>> index b9bc1b00142e..d6f39ee5bf5d 100644
>> --- a/include/drm/ttm/ttm_bo_api.h
>> +++ b/include/drm/ttm/ttm_bo_api.h
>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>   	 * either of these locks held.
>>   	 */
>>   
>> -	uint64_t offset; /* GPU address space is independent of CPU word size */
>> -
>>   	struct sg_table *sg;
>>   };
>>   
>> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
>> index c9e0fd09f4b2..c8ce6c181abe 100644
>> --- a/include/drm/ttm/ttm_bo_driver.h
>> +++ b/include/drm/ttm/ttm_bo_driver.h
>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>   	bool has_type;
>>   	bool use_type;
>>   	uint32_t flags;
>> -	uint64_t gpu_offset; /* GPU address space is independent of CPU word size */
>>   	uint64_t size;
>>   	uint32_t available_caching;
>>   	uint32_t default_caching;
>>

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy]

On 2/18/20 1:44 PM, Christian König wrote:
> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>> GPU address handling is device specific and should be handle by its 
>>> device
>>> driver.
>>>
>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>> ---
>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>   3 files changed, 10 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>> index 151edfd8de77..d5885cd609a3 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct 
>>> ttm_bo_device *bdev, struct drm_printer *p
>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>       drm_printf(p, "    available_caching: 0x%08X\n", 
>>> man->available_caching);
>>>       drm_printf(p, "    default_caching: 0x%08X\n", 
>>> man->default_caching);
>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct 
>>> ttm_buffer_object *bo,
>>>   moved:
>>>       bo->evicted = false;
>>>   -    if (bo->mem.mm_node)
>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>> -    else
>>> -        bo->offset = 0;
>>> -
>> After moving this into users, the else branch has been lost. Is
>> 'bo->mem.mm_node' always true?
>
> At least for the amdgpu and radeon use cases, yes.
>
> But that is a rather good question I mean for it is illegal to get the 
> GPU BO address if it is inaccessible (e.g. in the system domain).
>
> Could be that some driver relied on the behavior to get 0 for the 
> system domain here.

I wonder how to verify that ?

If I understand correctly:

1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in  
system domain.

2 unfortunately I can't say the same for bochs but it works with this 
patch series so I think bochs is fine as well.

3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so 
vmwgfx should be fine.

4 amdgpu and radeon runs with 'bo->mem.mm_node' always true

I am not sure about  nouveau as bo->offset is being used in many places.

I could probably mirror the removed logic to nouveau as

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c 
b/drivers/gpu/drm/nouveau/nouveau_bo.c
index f8015e0318d7..5a6a2af91318 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object 
*bo, bool evict,
                 list_for_each_entry(vma, &nvbo->vma_list, head) {
                         nouveau_vma_map(vma, mem);
                 }
+               if (bo->mem.mm_node)
+                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
+               else
+                       nvbo->offset = 0;
         } else {
                 list_for_each_entry(vma, &nvbo->vma_list, head) {
                         WARN_ON(ttm_bo_wait(bo, false, false));

Regards,

Nirmoy


>
> Regards,
> Christian.
>
>>
>> Best regards
>> Thomas
>>
>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>       return 0;
>>>   diff --git a/include/drm/ttm/ttm_bo_api.h 
>>> b/include/drm/ttm/ttm_bo_api.h
>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>> --- a/include/drm/ttm/ttm_bo_api.h
>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>        * either of these locks held.
>>>        */
>>>   -    uint64_t offset; /* GPU address space is independent of CPU 
>>> word size */
>>> -
>>>       struct sg_table *sg;
>>>   };
>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h 
>>> b/include/drm/ttm/ttm_bo_driver.h
>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>       bool has_type;
>>>       bool use_type;
>>>       uint32_t flags;
>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU 
>>> word size */
>>>       uint64_t size;
>>>       uint32_t available_caching;
>>>       uint32_t default_caching;
>>>
>
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy]

On 2/18/20 1:44 PM, Christian König wrote:
> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>> GPU address handling is device specific and should be handle by its 
>>> device
>>> driver.
>>>
>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>> ---
>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>   3 files changed, 10 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>> index 151edfd8de77..d5885cd609a3 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct 
>>> ttm_bo_device *bdev, struct drm_printer *p
>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>       drm_printf(p, "    available_caching: 0x%08X\n", 
>>> man->available_caching);
>>>       drm_printf(p, "    default_caching: 0x%08X\n", 
>>> man->default_caching);
>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct 
>>> ttm_buffer_object *bo,
>>>   moved:
>>>       bo->evicted = false;
>>>   -    if (bo->mem.mm_node)
>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>> -    else
>>> -        bo->offset = 0;
>>> -
>> After moving this into users, the else branch has been lost. Is
>> 'bo->mem.mm_node' always true?
>
> At least for the amdgpu and radeon use cases, yes.
>
> But that is a rather good question I mean for it is illegal to get the 
> GPU BO address if it is inaccessible (e.g. in the system domain).
>
> Could be that some driver relied on the behavior to get 0 for the 
> system domain here.

I wonder how to verify that ?

If I understand correctly:

1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in  
system domain.

2 unfortunately I can't say the same for bochs but it works with this 
patch series so I think bochs is fine as well.

3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so 
vmwgfx should be fine.

4 amdgpu and radeon runs with 'bo->mem.mm_node' always true

I am not sure about  nouveau as bo->offset is being used in many places.

I could probably mirror the removed logic to nouveau as

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c 
b/drivers/gpu/drm/nouveau/nouveau_bo.c
index f8015e0318d7..5a6a2af91318 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object 
*bo, bool evict,
                 list_for_each_entry(vma, &nvbo->vma_list, head) {
                         nouveau_vma_map(vma, mem);
                 }
+               if (bo->mem.mm_node)
+                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
+               else
+                       nvbo->offset = 0;
         } else {
                 list_for_each_entry(vma, &nvbo->vma_list, head) {
                         WARN_ON(ttm_bo_wait(bo, false, false));

Regards,

Nirmoy


>
> Regards,
> Christian.
>
>>
>> Best regards
>> Thomas
>>
>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>       return 0;
>>>   diff --git a/include/drm/ttm/ttm_bo_api.h 
>>> b/include/drm/ttm/ttm_bo_api.h
>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>> --- a/include/drm/ttm/ttm_bo_api.h
>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>        * either of these locks held.
>>>        */
>>>   -    uint64_t offset; /* GPU address space is independent of CPU 
>>> word size */
>>> -
>>>       struct sg_table *sg;
>>>   };
>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h 
>>> b/include/drm/ttm/ttm_bo_driver.h
>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>       bool has_type;
>>>       bool use_type;
>>>       uint32_t flags;
>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU 
>>> word size */
>>>       uint64_t size;
>>>       uint32_t available_caching;
>>>       uint32_t default_caching;
>>>
>
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy]

On 2/18/20 1:44 PM, Christian König wrote:
> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>> GPU address handling is device specific and should be handle by its 
>>> device
>>> driver.
>>>
>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>> ---
>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>   3 files changed, 10 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>> index 151edfd8de77..d5885cd609a3 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct 
>>> ttm_bo_device *bdev, struct drm_printer *p
>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>       drm_printf(p, "    available_caching: 0x%08X\n", 
>>> man->available_caching);
>>>       drm_printf(p, "    default_caching: 0x%08X\n", 
>>> man->default_caching);
>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct 
>>> ttm_buffer_object *bo,
>>>   moved:
>>>       bo->evicted = false;
>>>   -    if (bo->mem.mm_node)
>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>> -    else
>>> -        bo->offset = 0;
>>> -
>> After moving this into users, the else branch has been lost. Is
>> 'bo->mem.mm_node' always true?
>
> At least for the amdgpu and radeon use cases, yes.
>
> But that is a rather good question I mean for it is illegal to get the 
> GPU BO address if it is inaccessible (e.g. in the system domain).
>
> Could be that some driver relied on the behavior to get 0 for the 
> system domain here.

I wonder how to verify that ?

If I understand correctly:

1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in  
system domain.

2 unfortunately I can't say the same for bochs but it works with this 
patch series so I think bochs is fine as well.

3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so 
vmwgfx should be fine.

4 amdgpu and radeon runs with 'bo->mem.mm_node' always true

I am not sure about  nouveau as bo->offset is being used in many places.

I could probably mirror the removed logic to nouveau as

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c 
b/drivers/gpu/drm/nouveau/nouveau_bo.c
index f8015e0318d7..5a6a2af91318 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object 
*bo, bool evict,
                 list_for_each_entry(vma, &nvbo->vma_list, head) {
                         nouveau_vma_map(vma, mem);
                 }
+               if (bo->mem.mm_node)
+                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
+               else
+                       nvbo->offset = 0;
         } else {
                 list_for_each_entry(vma, &nvbo->vma_list, head) {
                         WARN_ON(ttm_bo_wait(bo, false, false));

Regards,

Nirmoy


>
> Regards,
> Christian.
>
>>
>> Best regards
>> Thomas
>>
>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>       return 0;
>>>   diff --git a/include/drm/ttm/ttm_bo_api.h 
>>> b/include/drm/ttm/ttm_bo_api.h
>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>> --- a/include/drm/ttm/ttm_bo_api.h
>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>        * either of these locks held.
>>>        */
>>>   -    uint64_t offset; /* GPU address space is independent of CPU 
>>> word size */
>>> -
>>>       struct sg_table *sg;
>>>   };
>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h 
>>> b/include/drm/ttm/ttm_bo_driver.h
>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>       bool has_type;
>>>       bool use_type;
>>>       uint32_t flags;
>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU 
>>> word size */
>>>       uint64_t size;
>>>       uint32_t available_caching;
>>>       uint32_t default_caching;
>>>
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 18.02.20 um 18:13 schrieb Nirmoy:
>
> On 2/18/20 1:44 PM, Christian König wrote:
>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>> GPU address handling is device specific and should be handle by its 
>>>> device
>>>> driver.
>>>>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>> ---
>>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>   3 files changed, 10 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> index 151edfd8de77..d5885cd609a3 100644
>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct 
>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>>       drm_printf(p, "    available_caching: 0x%08X\n", 
>>>> man->available_caching);
>>>>       drm_printf(p, "    default_caching: 0x%08X\n", 
>>>> man->default_caching);
>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct 
>>>> ttm_buffer_object *bo,
>>>>   moved:
>>>>       bo->evicted = false;
>>>>   -    if (bo->mem.mm_node)
>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>> -    else
>>>> -        bo->offset = 0;
>>>> -
>>> After moving this into users, the else branch has been lost. Is
>>> 'bo->mem.mm_node' always true?
>>
>> At least for the amdgpu and radeon use cases, yes.
>>
>> But that is a rather good question I mean for it is illegal to get 
>> the GPU BO address if it is inaccessible (e.g. in the system domain).
>>
>> Could be that some driver relied on the behavior to get 0 for the 
>> system domain here.
>
> I wonder how to verify that ?

No idea, but I wouldn't worry to much about that.

>
> If I understand correctly:
>
> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not 
> in  system domain.
>
> 2 unfortunately I can't say the same for bochs but it works with this 
> patch series so I think bochs is fine as well.
>
> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so 
> vmwgfx should be fine.
>
> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true

That sounds like those four are ok to me.

>
> I am not sure about  nouveau as bo->offset is being used in many places.
>
> I could probably mirror the removed logic to nouveau as
>
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c 
> b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index f8015e0318d7..5a6a2af91318 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object 
> *bo, bool evict,
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         nouveau_vma_map(vma, mem);
>                 }
> +               if (bo->mem.mm_node)
> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
> +               else
> +                       nvbo->offset = 0;
>         } else {
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         WARN_ON(ttm_bo_wait(bo, false, false));

Yeah, I would go down that route.

Christian.

>
> Regards,
>
> Nirmoy
>
>
>>
>> Regards,
>> Christian.
>>
>>>
>>> Best regards
>>> Thomas
>>>
>>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>       return 0;
>>>>   diff --git a/include/drm/ttm/ttm_bo_api.h 
>>>> b/include/drm/ttm/ttm_bo_api.h
>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>        * either of these locks held.
>>>>        */
>>>>   -    uint64_t offset; /* GPU address space is independent of CPU 
>>>> word size */
>>>> -
>>>>       struct sg_table *sg;
>>>>   };
>>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h 
>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>       bool has_type;
>>>>       bool use_type;
>>>>       uint32_t flags;
>>>> -    uint64_t gpu_offset; /* GPU address space is independent of 
>>>> CPU word size */
>>>>       uint64_t size;
>>>>       uint32_t available_caching;
>>>>       uint32_t default_caching;
>>>>
>>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/nouveau

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 18.02.20 um 18:13 schrieb Nirmoy:
>
> On 2/18/20 1:44 PM, Christian König wrote:
>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>> GPU address handling is device specific and should be handle by its 
>>>> device
>>>> driver.
>>>>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>> ---
>>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>   3 files changed, 10 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> index 151edfd8de77..d5885cd609a3 100644
>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct 
>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>>       drm_printf(p, "    available_caching: 0x%08X\n", 
>>>> man->available_caching);
>>>>       drm_printf(p, "    default_caching: 0x%08X\n", 
>>>> man->default_caching);
>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct 
>>>> ttm_buffer_object *bo,
>>>>   moved:
>>>>       bo->evicted = false;
>>>>   -    if (bo->mem.mm_node)
>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>> -    else
>>>> -        bo->offset = 0;
>>>> -
>>> After moving this into users, the else branch has been lost. Is
>>> 'bo->mem.mm_node' always true?
>>
>> At least for the amdgpu and radeon use cases, yes.
>>
>> But that is a rather good question I mean for it is illegal to get 
>> the GPU BO address if it is inaccessible (e.g. in the system domain).
>>
>> Could be that some driver relied on the behavior to get 0 for the 
>> system domain here.
>
> I wonder how to verify that ?

No idea, but I wouldn't worry to much about that.

>
> If I understand correctly:
>
> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not 
> in  system domain.
>
> 2 unfortunately I can't say the same for bochs but it works with this 
> patch series so I think bochs is fine as well.
>
> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so 
> vmwgfx should be fine.
>
> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true

That sounds like those four are ok to me.

>
> I am not sure about  nouveau as bo->offset is being used in many places.
>
> I could probably mirror the removed logic to nouveau as
>
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c 
> b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index f8015e0318d7..5a6a2af91318 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object 
> *bo, bool evict,
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         nouveau_vma_map(vma, mem);
>                 }
> +               if (bo->mem.mm_node)
> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
> +               else
> +                       nvbo->offset = 0;
>         } else {
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         WARN_ON(ttm_bo_wait(bo, false, false));

Yeah, I would go down that route.

Christian.

>
> Regards,
>
> Nirmoy
>
>
>>
>> Regards,
>> Christian.
>>
>>>
>>> Best regards
>>> Thomas
>>>
>>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>       return 0;
>>>>   diff --git a/include/drm/ttm/ttm_bo_api.h 
>>>> b/include/drm/ttm/ttm_bo_api.h
>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>        * either of these locks held.
>>>>        */
>>>>   -    uint64_t offset; /* GPU address space is independent of CPU 
>>>> word size */
>>>> -
>>>>       struct sg_table *sg;
>>>>   };
>>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h 
>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>       bool has_type;
>>>>       bool use_type;
>>>>       uint32_t flags;
>>>> -    uint64_t gpu_offset; /* GPU address space is independent of 
>>>> CPU word size */
>>>>       uint64_t size;
>>>>       uint32_t available_caching;
>>>>       uint32_t default_caching;
>>>>
>>
> _______________________________________________
> 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

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 18.02.20 um 18:13 schrieb Nirmoy:
>
> On 2/18/20 1:44 PM, Christian König wrote:
>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>> GPU address handling is device specific and should be handle by its 
>>>> device
>>>> driver.
>>>>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>> ---
>>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>   3 files changed, 10 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c 
>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> index 151edfd8de77..d5885cd609a3 100644
>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct 
>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>>       drm_printf(p, "    available_caching: 0x%08X\n", 
>>>> man->available_caching);
>>>>       drm_printf(p, "    default_caching: 0x%08X\n", 
>>>> man->default_caching);
>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct 
>>>> ttm_buffer_object *bo,
>>>>   moved:
>>>>       bo->evicted = false;
>>>>   -    if (bo->mem.mm_node)
>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>> -    else
>>>> -        bo->offset = 0;
>>>> -
>>> After moving this into users, the else branch has been lost. Is
>>> 'bo->mem.mm_node' always true?
>>
>> At least for the amdgpu and radeon use cases, yes.
>>
>> But that is a rather good question I mean for it is illegal to get 
>> the GPU BO address if it is inaccessible (e.g. in the system domain).
>>
>> Could be that some driver relied on the behavior to get 0 for the 
>> system domain here.
>
> I wonder how to verify that ?

No idea, but I wouldn't worry to much about that.

>
> If I understand correctly:
>
> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not 
> in  system domain.
>
> 2 unfortunately I can't say the same for bochs but it works with this 
> patch series so I think bochs is fine as well.
>
> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so 
> vmwgfx should be fine.
>
> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true

That sounds like those four are ok to me.

>
> I am not sure about  nouveau as bo->offset is being used in many places.
>
> I could probably mirror the removed logic to nouveau as
>
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c 
> b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index f8015e0318d7..5a6a2af91318 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object 
> *bo, bool evict,
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         nouveau_vma_map(vma, mem);
>                 }
> +               if (bo->mem.mm_node)
> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
> +               else
> +                       nvbo->offset = 0;
>         } else {
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         WARN_ON(ttm_bo_wait(bo, false, false));

Yeah, I would go down that route.

Christian.

>
> Regards,
>
> Nirmoy
>
>
>>
>> Regards,
>> Christian.
>>
>>>
>>> Best regards
>>> Thomas
>>>
>>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>       return 0;
>>>>   diff --git a/include/drm/ttm/ttm_bo_api.h 
>>>> b/include/drm/ttm/ttm_bo_api.h
>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>        * either of these locks held.
>>>>        */
>>>>   -    uint64_t offset; /* GPU address space is independent of CPU 
>>>> word size */
>>>> -
>>>>       struct sg_table *sg;
>>>>   };
>>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h 
>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>       bool has_type;
>>>>       bool use_type;
>>>>       uint32_t flags;
>>>> -    uint64_t gpu_offset; /* GPU address space is independent of 
>>>> CPU word size */
>>>>       uint64_t size;
>>>>       uint32_t available_caching;
>>>>       uint32_t default_caching;
>>>>
>>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 6014 bytes --]

Hi

Am 18.02.20 um 18:13 schrieb Nirmoy:
> 
> On 2/18/20 1:44 PM, Christian König wrote:
>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>> GPU address handling is device specific and should be handle by its
>>>> device
>>>> driver.
>>>>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das-5C7GfCeVMHo@public.gmane.org>
>>>> ---
>>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>   3 files changed, 10 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> index 151edfd8de77..d5885cd609a3 100644
>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>>       drm_printf(p, "    available_caching: 0x%08X\n",
>>>> man->available_caching);
>>>>       drm_printf(p, "    default_caching: 0x%08X\n",
>>>> man->default_caching);
>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>> ttm_buffer_object *bo,
>>>>   moved:
>>>>       bo->evicted = false;
>>>>   -    if (bo->mem.mm_node)
>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>> -    else
>>>> -        bo->offset = 0;
>>>> -
>>> After moving this into users, the else branch has been lost. Is
>>> 'bo->mem.mm_node' always true?
>>
>> At least for the amdgpu and radeon use cases, yes.
>>
>> But that is a rather good question I mean for it is illegal to get the
>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>
>> Could be that some driver relied on the behavior to get 0 for the
>> system domain here.
> 
> I wonder how to verify that ?
> 
> If I understand correctly:
> 
> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in 
> system domain.
> 
> 2 unfortunately I can't say the same for bochs but it works with this
> patch series so I think bochs is fine as well.
> 
> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
> vmwgfx should be fine.
> 
> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
> 
> I am not sure about  nouveau as bo->offset is being used in many places.
> 
> I could probably mirror the removed logic to nouveau as

I suggest to introduce a ttm helper that contains the original branching
and use it everywhere. Something like

  s64 ttm_bo_offset(struct ttm_buffer_object *bo)
  {
     if (WARN_ON_ONCE(!bo->mem.mm_node))
         return 0;
     return bo->mem.start << PAGE_SHIFT;
  }

Could be static inline. The warning should point to broken drivers. This
also gets rid of the ugly shift in the drivers.

Best regards
Thomas


> 
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
> b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index f8015e0318d7..5a6a2af91318 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
> *bo, bool evict,
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         nouveau_vma_map(vma, mem);
>                 }
> +               if (bo->mem.mm_node)
> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
> +               else
> +                       nvbo->offset = 0;
>         } else {
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         WARN_ON(ttm_bo_wait(bo, false, false));
> 
> Regards,
> 
> Nirmoy
> 
> 
>>
>> Regards,
>> Christian.
>>
>>>
>>> Best regards
>>> Thomas
>>>
>>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>       return 0;
>>>>   diff --git a/include/drm/ttm/ttm_bo_api.h
>>>> b/include/drm/ttm/ttm_bo_api.h
>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>        * either of these locks held.
>>>>        */
>>>>   -    uint64_t offset; /* GPU address space is independent of CPU
>>>> word size */
>>>> -
>>>>       struct sg_table *sg;
>>>>   };
>>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>       bool has_type;
>>>>       bool use_type;
>>>>       uint32_t flags;
>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>> word size */
>>>>       uint64_t size;
>>>>       uint32_t available_caching;
>>>>       uint32_t default_caching;
>>>>
>>
> _______________________________________________
> dri-devel mailing list
> dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[-- Attachment #2: Type: text/plain, Size: 182 bytes --]

_______________________________________________
amd-gfx mailing list
amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 5965 bytes --]

Hi

Am 18.02.20 um 18:13 schrieb Nirmoy:
> 
> On 2/18/20 1:44 PM, Christian König wrote:
>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>> GPU address handling is device specific and should be handle by its
>>>> device
>>>> driver.
>>>>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>> ---
>>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>   3 files changed, 10 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> index 151edfd8de77..d5885cd609a3 100644
>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>>       drm_printf(p, "    available_caching: 0x%08X\n",
>>>> man->available_caching);
>>>>       drm_printf(p, "    default_caching: 0x%08X\n",
>>>> man->default_caching);
>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>> ttm_buffer_object *bo,
>>>>   moved:
>>>>       bo->evicted = false;
>>>>   -    if (bo->mem.mm_node)
>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>> -    else
>>>> -        bo->offset = 0;
>>>> -
>>> After moving this into users, the else branch has been lost. Is
>>> 'bo->mem.mm_node' always true?
>>
>> At least for the amdgpu and radeon use cases, yes.
>>
>> But that is a rather good question I mean for it is illegal to get the
>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>
>> Could be that some driver relied on the behavior to get 0 for the
>> system domain here.
> 
> I wonder how to verify that ?
> 
> If I understand correctly:
> 
> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in 
> system domain.
> 
> 2 unfortunately I can't say the same for bochs but it works with this
> patch series so I think bochs is fine as well.
> 
> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
> vmwgfx should be fine.
> 
> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
> 
> I am not sure about  nouveau as bo->offset is being used in many places.
> 
> I could probably mirror the removed logic to nouveau as

I suggest to introduce a ttm helper that contains the original branching
and use it everywhere. Something like

  s64 ttm_bo_offset(struct ttm_buffer_object *bo)
  {
     if (WARN_ON_ONCE(!bo->mem.mm_node))
         return 0;
     return bo->mem.start << PAGE_SHIFT;
  }

Could be static inline. The warning should point to broken drivers. This
also gets rid of the ugly shift in the drivers.

Best regards
Thomas


> 
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
> b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index f8015e0318d7..5a6a2af91318 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
> *bo, bool evict,
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         nouveau_vma_map(vma, mem);
>                 }
> +               if (bo->mem.mm_node)
> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
> +               else
> +                       nvbo->offset = 0;
>         } else {
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         WARN_ON(ttm_bo_wait(bo, false, false));
> 
> Regards,
> 
> Nirmoy
> 
> 
>>
>> Regards,
>> Christian.
>>
>>>
>>> Best regards
>>> Thomas
>>>
>>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>       return 0;
>>>>   diff --git a/include/drm/ttm/ttm_bo_api.h
>>>> b/include/drm/ttm/ttm_bo_api.h
>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>        * either of these locks held.
>>>>        */
>>>>   -    uint64_t offset; /* GPU address space is independent of CPU
>>>> word size */
>>>> -
>>>>       struct sg_table *sg;
>>>>   };
>>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>       bool has_type;
>>>>       bool use_type;
>>>>       uint32_t flags;
>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>> word size */
>>>>       uint64_t size;
>>>>       uint32_t available_caching;
>>>>       uint32_t default_caching;
>>>>
>>
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 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

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 5965 bytes --]

Hi

Am 18.02.20 um 18:13 schrieb Nirmoy:
> 
> On 2/18/20 1:44 PM, Christian König wrote:
>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>> GPU address handling is device specific and should be handle by its
>>>> device
>>>> driver.
>>>>
>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>> ---
>>>>   drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>   include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>   include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>   3 files changed, 10 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> index 151edfd8de77..d5885cd609a3 100644
>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>       drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>       drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>       drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>       drm_printf(p, "    size: %llu\n", man->size);
>>>>       drm_printf(p, "    available_caching: 0x%08X\n",
>>>> man->available_caching);
>>>>       drm_printf(p, "    default_caching: 0x%08X\n",
>>>> man->default_caching);
>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>> ttm_buffer_object *bo,
>>>>   moved:
>>>>       bo->evicted = false;
>>>>   -    if (bo->mem.mm_node)
>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>> -    else
>>>> -        bo->offset = 0;
>>>> -
>>> After moving this into users, the else branch has been lost. Is
>>> 'bo->mem.mm_node' always true?
>>
>> At least for the amdgpu and radeon use cases, yes.
>>
>> But that is a rather good question I mean for it is illegal to get the
>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>
>> Could be that some driver relied on the behavior to get 0 for the
>> system domain here.
> 
> I wonder how to verify that ?
> 
> If I understand correctly:
> 
> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in 
> system domain.
> 
> 2 unfortunately I can't say the same for bochs but it works with this
> patch series so I think bochs is fine as well.
> 
> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
> vmwgfx should be fine.
> 
> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
> 
> I am not sure about  nouveau as bo->offset is being used in many places.
> 
> I could probably mirror the removed logic to nouveau as

I suggest to introduce a ttm helper that contains the original branching
and use it everywhere. Something like

  s64 ttm_bo_offset(struct ttm_buffer_object *bo)
  {
     if (WARN_ON_ONCE(!bo->mem.mm_node))
         return 0;
     return bo->mem.start << PAGE_SHIFT;
  }

Could be static inline. The warning should point to broken drivers. This
also gets rid of the ugly shift in the drivers.

Best regards
Thomas


> 
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
> b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index f8015e0318d7..5a6a2af91318 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
> *bo, bool evict,
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         nouveau_vma_map(vma, mem);
>                 }
> +               if (bo->mem.mm_node)
> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
> +               else
> +                       nvbo->offset = 0;
>         } else {
>                 list_for_each_entry(vma, &nvbo->vma_list, head) {
>                         WARN_ON(ttm_bo_wait(bo, false, false));
> 
> Regards,
> 
> Nirmoy
> 
> 
>>
>> Regards,
>> Christian.
>>
>>>
>>> Best regards
>>> Thomas
>>>
>>>>       ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>       return 0;
>>>>   diff --git a/include/drm/ttm/ttm_bo_api.h
>>>> b/include/drm/ttm/ttm_bo_api.h
>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>        * either of these locks held.
>>>>        */
>>>>   -    uint64_t offset; /* GPU address space is independent of CPU
>>>> word size */
>>>> -
>>>>       struct sg_table *sg;
>>>>   };
>>>>   diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>       bool has_type;
>>>>       bool use_type;
>>>>       uint32_t flags;
>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>> word size */
>>>>       uint64_t size;
>>>>       uint32_t available_caching;
>>>>       uint32_t default_caching;
>>>>
>>
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
> Hi
>
> Am 18.02.20 um 18:13 schrieb Nirmoy:
>> On 2/18/20 1:44 PM, Christian König wrote:
>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>> GPU address handling is device specific and should be handle by its
>>>>> device
>>>>> driver.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> ---
>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>    3 files changed, 10 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>> man->available_caching);
>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>> man->default_caching);
>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>> ttm_buffer_object *bo,
>>>>>    moved:
>>>>>        bo->evicted = false;
>>>>>    -    if (bo->mem.mm_node)
>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>> -    else
>>>>> -        bo->offset = 0;
>>>>> -
>>>> After moving this into users, the else branch has been lost. Is
>>>> 'bo->mem.mm_node' always true?
>>> At least for the amdgpu and radeon use cases, yes.
>>>
>>> But that is a rather good question I mean for it is illegal to get the
>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>
>>> Could be that some driver relied on the behavior to get 0 for the
>>> system domain here.
>> I wonder how to verify that ?
>>
>> If I understand correctly:
>>
>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>> system domain.
>>
>> 2 unfortunately I can't say the same for bochs but it works with this
>> patch series so I think bochs is fine as well.
>>
>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>> vmwgfx should be fine.
>>
>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>
>> I am not sure about  nouveau as bo->offset is being used in many places.
>>
>> I could probably mirror the removed logic to nouveau as
> I suggest to introduce a ttm helper that contains the original branching
> and use it everywhere. Something like
>
>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>    {
>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>           return 0;
>       return bo->mem.start << PAGE_SHIFT;
>    }
>
> Could be static inline. The warning should point to broken drivers. This
> also gets rid of the ugly shift in the drivers.

Big NAK on this. That is exactly what we should NOT do.

See the shift belongs into the driver, because it is driver dependent if 
we work with page or byte offsets.

For amdgpu we for example want to work with byte offsets and TTM should 
not make any assumption about what bo->mem.start actually contains.

Regards,
Christian.

>
> Best regards
> Thomas
>
>
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index f8015e0318d7..5a6a2af91318 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>> *bo, bool evict,
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          nouveau_vma_map(vma, mem);
>>                  }
>> +               if (bo->mem.mm_node)
>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>> +               else
>> +                       nvbo->offset = 0;
>>          } else {
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>
>> Regards,
>>
>> Nirmoy
>>
>>
>>> Regards,
>>> Christian.
>>>
>>>> Best regards
>>>> Thomas
>>>>
>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>        return 0;
>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>         * either of these locks held.
>>>>>         */
>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>> -
>>>>>        struct sg_table *sg;
>>>>>    };
>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>        bool has_type;
>>>>>        bool use_type;
>>>>>        uint32_t flags;
>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>>        uint64_t size;
>>>>>        uint32_t available_caching;
>>>>>        uint32_t default_caching;
>>>>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/nouveau

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
> Hi
>
> Am 18.02.20 um 18:13 schrieb Nirmoy:
>> On 2/18/20 1:44 PM, Christian König wrote:
>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>> GPU address handling is device specific and should be handle by its
>>>>> device
>>>>> driver.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> ---
>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>    3 files changed, 10 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>> man->available_caching);
>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>> man->default_caching);
>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>> ttm_buffer_object *bo,
>>>>>    moved:
>>>>>        bo->evicted = false;
>>>>>    -    if (bo->mem.mm_node)
>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>> -    else
>>>>> -        bo->offset = 0;
>>>>> -
>>>> After moving this into users, the else branch has been lost. Is
>>>> 'bo->mem.mm_node' always true?
>>> At least for the amdgpu and radeon use cases, yes.
>>>
>>> But that is a rather good question I mean for it is illegal to get the
>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>
>>> Could be that some driver relied on the behavior to get 0 for the
>>> system domain here.
>> I wonder how to verify that ?
>>
>> If I understand correctly:
>>
>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>> system domain.
>>
>> 2 unfortunately I can't say the same for bochs but it works with this
>> patch series so I think bochs is fine as well.
>>
>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>> vmwgfx should be fine.
>>
>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>
>> I am not sure about  nouveau as bo->offset is being used in many places.
>>
>> I could probably mirror the removed logic to nouveau as
> I suggest to introduce a ttm helper that contains the original branching
> and use it everywhere. Something like
>
>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>    {
>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>           return 0;
>       return bo->mem.start << PAGE_SHIFT;
>    }
>
> Could be static inline. The warning should point to broken drivers. This
> also gets rid of the ugly shift in the drivers.

Big NAK on this. That is exactly what we should NOT do.

See the shift belongs into the driver, because it is driver dependent if 
we work with page or byte offsets.

For amdgpu we for example want to work with byte offsets and TTM should 
not make any assumption about what bo->mem.start actually contains.

Regards,
Christian.

>
> Best regards
> Thomas
>
>
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index f8015e0318d7..5a6a2af91318 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>> *bo, bool evict,
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          nouveau_vma_map(vma, mem);
>>                  }
>> +               if (bo->mem.mm_node)
>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>> +               else
>> +                       nvbo->offset = 0;
>>          } else {
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>
>> Regards,
>>
>> Nirmoy
>>
>>
>>> Regards,
>>> Christian.
>>>
>>>> Best regards
>>>> Thomas
>>>>
>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>        return 0;
>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>         * either of these locks held.
>>>>>         */
>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>> -
>>>>>        struct sg_table *sg;
>>>>>    };
>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>        bool has_type;
>>>>>        bool use_type;
>>>>>        uint32_t flags;
>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>>        uint64_t size;
>>>>>        uint32_t available_caching;
>>>>>        uint32_t default_caching;
>>>>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
> Hi
>
> Am 18.02.20 um 18:13 schrieb Nirmoy:
>> On 2/18/20 1:44 PM, Christian König wrote:
>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>> GPU address handling is device specific and should be handle by its
>>>>> device
>>>>> driver.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> ---
>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>    3 files changed, 10 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>> man->available_caching);
>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>> man->default_caching);
>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>> ttm_buffer_object *bo,
>>>>>    moved:
>>>>>        bo->evicted = false;
>>>>>    -    if (bo->mem.mm_node)
>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>> -    else
>>>>> -        bo->offset = 0;
>>>>> -
>>>> After moving this into users, the else branch has been lost. Is
>>>> 'bo->mem.mm_node' always true?
>>> At least for the amdgpu and radeon use cases, yes.
>>>
>>> But that is a rather good question I mean for it is illegal to get the
>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>
>>> Could be that some driver relied on the behavior to get 0 for the
>>> system domain here.
>> I wonder how to verify that ?
>>
>> If I understand correctly:
>>
>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>> system domain.
>>
>> 2 unfortunately I can't say the same for bochs but it works with this
>> patch series so I think bochs is fine as well.
>>
>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>> vmwgfx should be fine.
>>
>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>
>> I am not sure about  nouveau as bo->offset is being used in many places.
>>
>> I could probably mirror the removed logic to nouveau as
> I suggest to introduce a ttm helper that contains the original branching
> and use it everywhere. Something like
>
>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>    {
>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>           return 0;
>       return bo->mem.start << PAGE_SHIFT;
>    }
>
> Could be static inline. The warning should point to broken drivers. This
> also gets rid of the ugly shift in the drivers.

Big NAK on this. That is exactly what we should NOT do.

See the shift belongs into the driver, because it is driver dependent if 
we work with page or byte offsets.

For amdgpu we for example want to work with byte offsets and TTM should 
not make any assumption about what bo->mem.start actually contains.

Regards,
Christian.

>
> Best regards
> Thomas
>
>
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index f8015e0318d7..5a6a2af91318 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>> *bo, bool evict,
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          nouveau_vma_map(vma, mem);
>>                  }
>> +               if (bo->mem.mm_node)
>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>> +               else
>> +                       nvbo->offset = 0;
>>          } else {
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>
>> Regards,
>>
>> Nirmoy
>>
>>
>>> Regards,
>>> Christian.
>>>
>>>> Best regards
>>>> Thomas
>>>>
>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>        return 0;
>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>         * either of these locks held.
>>>>>         */
>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>> -
>>>>>        struct sg_table *sg;
>>>>>    };
>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>        bool has_type;
>>>>>        bool use_type;
>>>>>        uint32_t flags;
>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>>        uint64_t size;
>>>>>        uint32_t available_caching;
>>>>>        uint32_t default_caching;
>>>>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 7220 bytes --]

Hi

Am 18.02.20 um 19:23 schrieb Christian König:
> Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 18.02.20 um 18:13 schrieb Nirmoy:
>>> On 2/18/20 1:44 PM, Christian König wrote:
>>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>>> Hi
>>>>>
>>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>>> GPU address handling is device specific and should be handle by its
>>>>>> device
>>>>>> driver.
>>>>>>
>>>>>> Signed-off-by: Nirmoy Das <nirmoy.das-5C7GfCeVMHo@public.gmane.org>
>>>>>> ---
>>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>>    3 files changed, 10 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>>> man->available_caching);
>>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>>> man->default_caching);
>>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>>> ttm_buffer_object *bo,
>>>>>>    moved:
>>>>>>        bo->evicted = false;
>>>>>>    -    if (bo->mem.mm_node)
>>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>>> -    else
>>>>>> -        bo->offset = 0;
>>>>>> -
>>>>> After moving this into users, the else branch has been lost. Is
>>>>> 'bo->mem.mm_node' always true?
>>>> At least for the amdgpu and radeon use cases, yes.
>>>>
>>>> But that is a rather good question I mean for it is illegal to get the
>>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>>
>>>> Could be that some driver relied on the behavior to get 0 for the
>>>> system domain here.
>>> I wonder how to verify that ?
>>>
>>> If I understand correctly:
>>>
>>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>>> system domain.
>>>
>>> 2 unfortunately I can't say the same for bochs but it works with this
>>> patch series so I think bochs is fine as well.
>>>
>>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>>> vmwgfx should be fine.
>>>
>>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>>
>>> I am not sure about  nouveau as bo->offset is being used in many places.
>>>
>>> I could probably mirror the removed logic to nouveau as
>> I suggest to introduce a ttm helper that contains the original branching
>> and use it everywhere. Something like
>>
>>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>>    {
>>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>>           return 0;
>>       return bo->mem.start << PAGE_SHIFT;
>>    }
>>
>> Could be static inline. The warning should point to broken drivers. This
>> also gets rid of the ugly shift in the drivers.
> 
> Big NAK on this. That is exactly what we should NOT do.
> 
> See the shift belongs into the driver, because it is driver dependent if
> we work with page or byte offsets.
> 
> For amdgpu we for example want to work with byte offsets and TTM should
> not make any assumption about what bo->mem.start actually contains.

OK. What about something like ttm_bo_pg_offset()? Same code without the
shift. Would also make it clear that it's a page offset.

Best regards
Thomas

> 
> Regards,
> Christian.
> 
>>
>> Best regards
>> Thomas
>>
>>
>>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> index f8015e0318d7..5a6a2af91318 100644
>>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>>> *bo, bool evict,
>>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>                          nouveau_vma_map(vma, mem);
>>>                  }
>>> +               if (bo->mem.mm_node)
>>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>>> +               else
>>> +                       nvbo->offset = 0;
>>>          } else {
>>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>>
>>> Regards,
>>>
>>> Nirmoy
>>>
>>>
>>>> Regards,
>>>> Christian.
>>>>
>>>>> Best regards
>>>>> Thomas
>>>>>
>>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>>        return 0;
>>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>>         * either of these locks held.
>>>>>>         */
>>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>>> word size */
>>>>>> -
>>>>>>        struct sg_table *sg;
>>>>>>    };
>>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>>        bool has_type;
>>>>>>        bool use_type;
>>>>>>        uint32_t flags;
>>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>>> word size */
>>>>>>        uint64_t size;
>>>>>>        uint32_t available_caching;
>>>>>>        uint32_t default_caching;
>>>>>>
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[-- Attachment #2: Type: text/plain, Size: 182 bytes --]

_______________________________________________
amd-gfx mailing list
amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 7143 bytes --]

Hi

Am 18.02.20 um 19:23 schrieb Christian König:
> Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 18.02.20 um 18:13 schrieb Nirmoy:
>>> On 2/18/20 1:44 PM, Christian König wrote:
>>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>>> Hi
>>>>>
>>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>>> GPU address handling is device specific and should be handle by its
>>>>>> device
>>>>>> driver.
>>>>>>
>>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>>> ---
>>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>>    3 files changed, 10 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>>> man->available_caching);
>>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>>> man->default_caching);
>>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>>> ttm_buffer_object *bo,
>>>>>>    moved:
>>>>>>        bo->evicted = false;
>>>>>>    -    if (bo->mem.mm_node)
>>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>>> -    else
>>>>>> -        bo->offset = 0;
>>>>>> -
>>>>> After moving this into users, the else branch has been lost. Is
>>>>> 'bo->mem.mm_node' always true?
>>>> At least for the amdgpu and radeon use cases, yes.
>>>>
>>>> But that is a rather good question I mean for it is illegal to get the
>>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>>
>>>> Could be that some driver relied on the behavior to get 0 for the
>>>> system domain here.
>>> I wonder how to verify that ?
>>>
>>> If I understand correctly:
>>>
>>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>>> system domain.
>>>
>>> 2 unfortunately I can't say the same for bochs but it works with this
>>> patch series so I think bochs is fine as well.
>>>
>>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>>> vmwgfx should be fine.
>>>
>>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>>
>>> I am not sure about  nouveau as bo->offset is being used in many places.
>>>
>>> I could probably mirror the removed logic to nouveau as
>> I suggest to introduce a ttm helper that contains the original branching
>> and use it everywhere. Something like
>>
>>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>>    {
>>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>>           return 0;
>>       return bo->mem.start << PAGE_SHIFT;
>>    }
>>
>> Could be static inline. The warning should point to broken drivers. This
>> also gets rid of the ugly shift in the drivers.
> 
> Big NAK on this. That is exactly what we should NOT do.
> 
> See the shift belongs into the driver, because it is driver dependent if
> we work with page or byte offsets.
> 
> For amdgpu we for example want to work with byte offsets and TTM should
> not make any assumption about what bo->mem.start actually contains.

OK. What about something like ttm_bo_pg_offset()? Same code without the
shift. Would also make it clear that it's a page offset.

Best regards
Thomas

> 
> Regards,
> Christian.
> 
>>
>> Best regards
>> Thomas
>>
>>
>>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> index f8015e0318d7..5a6a2af91318 100644
>>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>>> *bo, bool evict,
>>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>                          nouveau_vma_map(vma, mem);
>>>                  }
>>> +               if (bo->mem.mm_node)
>>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>>> +               else
>>> +                       nvbo->offset = 0;
>>>          } else {
>>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>>
>>> Regards,
>>>
>>> Nirmoy
>>>
>>>
>>>> Regards,
>>>> Christian.
>>>>
>>>>> Best regards
>>>>> Thomas
>>>>>
>>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>>        return 0;
>>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>>         * either of these locks held.
>>>>>>         */
>>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>>> word size */
>>>>>> -
>>>>>>        struct sg_table *sg;
>>>>>>    };
>>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>>        bool has_type;
>>>>>>        bool use_type;
>>>>>>        uint32_t flags;
>>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>>> word size */
>>>>>>        uint64_t size;
>>>>>>        uint32_t available_caching;
>>>>>>        uint32_t default_caching;
>>>>>>
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel@lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 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

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 7143 bytes --]

Hi

Am 18.02.20 um 19:23 schrieb Christian König:
> Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 18.02.20 um 18:13 schrieb Nirmoy:
>>> On 2/18/20 1:44 PM, Christian König wrote:
>>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>>> Hi
>>>>>
>>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>>> GPU address handling is device specific and should be handle by its
>>>>>> device
>>>>>> driver.
>>>>>>
>>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>>> ---
>>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>>    3 files changed, 10 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>>> man->available_caching);
>>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>>> man->default_caching);
>>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>>> ttm_buffer_object *bo,
>>>>>>    moved:
>>>>>>        bo->evicted = false;
>>>>>>    -    if (bo->mem.mm_node)
>>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>>> -    else
>>>>>> -        bo->offset = 0;
>>>>>> -
>>>>> After moving this into users, the else branch has been lost. Is
>>>>> 'bo->mem.mm_node' always true?
>>>> At least for the amdgpu and radeon use cases, yes.
>>>>
>>>> But that is a rather good question I mean for it is illegal to get the
>>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>>
>>>> Could be that some driver relied on the behavior to get 0 for the
>>>> system domain here.
>>> I wonder how to verify that ?
>>>
>>> If I understand correctly:
>>>
>>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>>> system domain.
>>>
>>> 2 unfortunately I can't say the same for bochs but it works with this
>>> patch series so I think bochs is fine as well.
>>>
>>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>>> vmwgfx should be fine.
>>>
>>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>>
>>> I am not sure about  nouveau as bo->offset is being used in many places.
>>>
>>> I could probably mirror the removed logic to nouveau as
>> I suggest to introduce a ttm helper that contains the original branching
>> and use it everywhere. Something like
>>
>>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>>    {
>>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>>           return 0;
>>       return bo->mem.start << PAGE_SHIFT;
>>    }
>>
>> Could be static inline. The warning should point to broken drivers. This
>> also gets rid of the ugly shift in the drivers.
> 
> Big NAK on this. That is exactly what we should NOT do.
> 
> See the shift belongs into the driver, because it is driver dependent if
> we work with page or byte offsets.
> 
> For amdgpu we for example want to work with byte offsets and TTM should
> not make any assumption about what bo->mem.start actually contains.

OK. What about something like ttm_bo_pg_offset()? Same code without the
shift. Would also make it clear that it's a page offset.

Best regards
Thomas

> 
> Regards,
> Christian.
> 
>>
>> Best regards
>> Thomas
>>
>>
>>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> index f8015e0318d7..5a6a2af91318 100644
>>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>>> *bo, bool evict,
>>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>                          nouveau_vma_map(vma, mem);
>>>                  }
>>> +               if (bo->mem.mm_node)
>>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>>> +               else
>>> +                       nvbo->offset = 0;
>>>          } else {
>>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>>
>>> Regards,
>>>
>>> Nirmoy
>>>
>>>
>>>> Regards,
>>>> Christian.
>>>>
>>>>> Best regards
>>>>> Thomas
>>>>>
>>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>>        return 0;
>>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>>         * either of these locks held.
>>>>>>         */
>>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>>> word size */
>>>>>> -
>>>>>>        struct sg_table *sg;
>>>>>>    };
>>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>>        bool has_type;
>>>>>>        bool use_type;
>>>>>>        uint32_t flags;
>>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>>> word size */
>>>>>>        uint64_t size;
>>>>>>        uint32_t available_caching;
>>>>>>        uint32_t default_caching;
>>>>>>
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel@lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

[-- Attachment #1.1: Type: text/plain, Size: 7567 bytes --]

Am 18.02.20 um 19:28 schrieb Thomas Zimmermann:
> Hi
>
> Am 18.02.20 um 19:23 schrieb Christian König:
>> Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 18.02.20 um 18:13 schrieb Nirmoy:
>>>> On 2/18/20 1:44 PM, Christian König wrote:
>>>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>>>> Hi
>>>>>>
>>>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>>>> GPU address handling is device specific and should be handle by its
>>>>>>> device
>>>>>>> driver.
>>>>>>>
>>>>>>> Signed-off-by: Nirmoy Das <nirmoy.das-5C7GfCeVMHo@public.gmane.org>
>>>>>>> ---
>>>>>>>     drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>>>     include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>>>     include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>>>     3 files changed, 10 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>>>         drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>>>         drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>>>         drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>>>         drm_printf(p, "    size: %llu\n", man->size);
>>>>>>>         drm_printf(p, "    available_caching: 0x%08X\n",
>>>>>>> man->available_caching);
>>>>>>>         drm_printf(p, "    default_caching: 0x%08X\n",
>>>>>>> man->default_caching);
>>>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>>>> ttm_buffer_object *bo,
>>>>>>>     moved:
>>>>>>>         bo->evicted = false;
>>>>>>>     -    if (bo->mem.mm_node)
>>>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>>>> -    else
>>>>>>> -        bo->offset = 0;
>>>>>>> -
>>>>>> After moving this into users, the else branch has been lost. Is
>>>>>> 'bo->mem.mm_node' always true?
>>>>> At least for the amdgpu and radeon use cases, yes.
>>>>>
>>>>> But that is a rather good question I mean for it is illegal to get the
>>>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>>>
>>>>> Could be that some driver relied on the behavior to get 0 for the
>>>>> system domain here.
>>>> I wonder how to verify that ?
>>>>
>>>> If I understand correctly:
>>>>
>>>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>>>> system domain.
>>>>
>>>> 2 unfortunately I can't say the same for bochs but it works with this
>>>> patch series so I think bochs is fine as well.
>>>>
>>>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>>>> vmwgfx should be fine.
>>>>
>>>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>>>
>>>> I am not sure about  nouveau as bo->offset is being used in many places.
>>>>
>>>> I could probably mirror the removed logic to nouveau as
>>> I suggest to introduce a ttm helper that contains the original branching
>>> and use it everywhere. Something like
>>>
>>>     s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>>>     {
>>>        if (WARN_ON_ONCE(!bo->mem.mm_node))
>>>            return 0;
>>>        return bo->mem.start << PAGE_SHIFT;
>>>     }
>>>
>>> Could be static inline. The warning should point to broken drivers. This
>>> also gets rid of the ugly shift in the drivers.
>> Big NAK on this. That is exactly what we should NOT do.
>>
>> See the shift belongs into the driver, because it is driver dependent if
>> we work with page or byte offsets.
>>
>> For amdgpu we for example want to work with byte offsets and TTM should
>> not make any assumption about what bo->mem.start actually contains.
> OK. What about something like ttm_bo_pg_offset()? Same code without the
> shift. Would also make it clear that it's a page offset.

That is a rather good idea. We could name that ttm_bo_man_offset() and 
put it into ttm_bo_manager.c next to the manager which allocates them.

It's just that this manager is not used by everybody, so amdgpu, radeon 
and nouveau would still need a separate function.

Christian.

>
> Best regards
> Thomas
>
>> Regards,
>> Christian.
>>
>>> Best regards
>>> Thomas
>>>
>>>
>>>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> index f8015e0318d7..5a6a2af91318 100644
>>>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>>>> *bo, bool evict,
>>>>                   list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>>                           nouveau_vma_map(vma, mem);
>>>>                   }
>>>> +               if (bo->mem.mm_node)
>>>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>>>> +               else
>>>> +                       nvbo->offset = 0;
>>>>           } else {
>>>>                   list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>>                           WARN_ON(ttm_bo_wait(bo, false, false));
>>>>
>>>> Regards,
>>>>
>>>> Nirmoy
>>>>
>>>>
>>>>> Regards,
>>>>> Christian.
>>>>>
>>>>>> Best regards
>>>>>> Thomas
>>>>>>
>>>>>>>         ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>>>         return 0;
>>>>>>>     diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>>>          * either of these locks held.
>>>>>>>          */
>>>>>>>     -    uint64_t offset; /* GPU address space is independent of CPU
>>>>>>> word size */
>>>>>>> -
>>>>>>>         struct sg_table *sg;
>>>>>>>     };
>>>>>>>     diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>>>         bool has_type;
>>>>>>>         bool use_type;
>>>>>>>         uint32_t flags;
>>>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>>>> word size */
>>>>>>>         uint64_t size;
>>>>>>>         uint32_t available_caching;
>>>>>>>         uint32_t default_caching;
>>>>>>>
>>>> _______________________________________________
>>>> dri-devel mailing list
>>>> dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
>>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx


[-- Attachment #1.2: Type: text/html, Size: 10009 bytes --]

[-- Attachment #2: Type: text/plain, Size: 182 bytes --]

_______________________________________________
amd-gfx mailing list
amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

[-- Attachment #1.1: Type: text/plain, Size: 7462 bytes --]

Am 18.02.20 um 19:28 schrieb Thomas Zimmermann:
> Hi
>
> Am 18.02.20 um 19:23 schrieb Christian König:
>> Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 18.02.20 um 18:13 schrieb Nirmoy:
>>>> On 2/18/20 1:44 PM, Christian König wrote:
>>>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>>>> Hi
>>>>>>
>>>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>>>> GPU address handling is device specific and should be handle by its
>>>>>>> device
>>>>>>> driver.
>>>>>>>
>>>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>>>> ---
>>>>>>>     drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>>>     include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>>>     include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>>>     3 files changed, 10 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>>>         drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>>>         drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>>>         drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>>>         drm_printf(p, "    size: %llu\n", man->size);
>>>>>>>         drm_printf(p, "    available_caching: 0x%08X\n",
>>>>>>> man->available_caching);
>>>>>>>         drm_printf(p, "    default_caching: 0x%08X\n",
>>>>>>> man->default_caching);
>>>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>>>> ttm_buffer_object *bo,
>>>>>>>     moved:
>>>>>>>         bo->evicted = false;
>>>>>>>     -    if (bo->mem.mm_node)
>>>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>>>> -    else
>>>>>>> -        bo->offset = 0;
>>>>>>> -
>>>>>> After moving this into users, the else branch has been lost. Is
>>>>>> 'bo->mem.mm_node' always true?
>>>>> At least for the amdgpu and radeon use cases, yes.
>>>>>
>>>>> But that is a rather good question I mean for it is illegal to get the
>>>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>>>
>>>>> Could be that some driver relied on the behavior to get 0 for the
>>>>> system domain here.
>>>> I wonder how to verify that ?
>>>>
>>>> If I understand correctly:
>>>>
>>>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>>>> system domain.
>>>>
>>>> 2 unfortunately I can't say the same for bochs but it works with this
>>>> patch series so I think bochs is fine as well.
>>>>
>>>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>>>> vmwgfx should be fine.
>>>>
>>>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>>>
>>>> I am not sure about  nouveau as bo->offset is being used in many places.
>>>>
>>>> I could probably mirror the removed logic to nouveau as
>>> I suggest to introduce a ttm helper that contains the original branching
>>> and use it everywhere. Something like
>>>
>>>     s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>>>     {
>>>        if (WARN_ON_ONCE(!bo->mem.mm_node))
>>>            return 0;
>>>        return bo->mem.start << PAGE_SHIFT;
>>>     }
>>>
>>> Could be static inline. The warning should point to broken drivers. This
>>> also gets rid of the ugly shift in the drivers.
>> Big NAK on this. That is exactly what we should NOT do.
>>
>> See the shift belongs into the driver, because it is driver dependent if
>> we work with page or byte offsets.
>>
>> For amdgpu we for example want to work with byte offsets and TTM should
>> not make any assumption about what bo->mem.start actually contains.
> OK. What about something like ttm_bo_pg_offset()? Same code without the
> shift. Would also make it clear that it's a page offset.

That is a rather good idea. We could name that ttm_bo_man_offset() and 
put it into ttm_bo_manager.c next to the manager which allocates them.

It's just that this manager is not used by everybody, so amdgpu, radeon 
and nouveau would still need a separate function.

Christian.

>
> Best regards
> Thomas
>
>> Regards,
>> Christian.
>>
>>> Best regards
>>> Thomas
>>>
>>>
>>>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> index f8015e0318d7..5a6a2af91318 100644
>>>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>>>> *bo, bool evict,
>>>>                   list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>>                           nouveau_vma_map(vma, mem);
>>>>                   }
>>>> +               if (bo->mem.mm_node)
>>>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>>>> +               else
>>>> +                       nvbo->offset = 0;
>>>>           } else {
>>>>                   list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>>                           WARN_ON(ttm_bo_wait(bo, false, false));
>>>>
>>>> Regards,
>>>>
>>>> Nirmoy
>>>>
>>>>
>>>>> Regards,
>>>>> Christian.
>>>>>
>>>>>> Best regards
>>>>>> Thomas
>>>>>>
>>>>>>>         ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>>>         return 0;
>>>>>>>     diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>>>          * either of these locks held.
>>>>>>>          */
>>>>>>>     -    uint64_t offset; /* GPU address space is independent of CPU
>>>>>>> word size */
>>>>>>> -
>>>>>>>         struct sg_table *sg;
>>>>>>>     };
>>>>>>>     diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>>>         bool has_type;
>>>>>>>         bool use_type;
>>>>>>>         uint32_t flags;
>>>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>>>> word size */
>>>>>>>         uint64_t size;
>>>>>>>         uint32_t available_caching;
>>>>>>>         uint32_t default_caching;
>>>>>>>
>>>> _______________________________________________
>>>> dri-devel mailing list
>>>> dri-devel@lists.freedesktop.org
>>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx


[-- Attachment #1.2: Type: text/html, Size: 9757 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

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Christian König]

[-- Attachment #1.1: Type: text/plain, Size: 7462 bytes --]

Am 18.02.20 um 19:28 schrieb Thomas Zimmermann:
> Hi
>
> Am 18.02.20 um 19:23 schrieb Christian König:
>> Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
>>> Hi
>>>
>>> Am 18.02.20 um 18:13 schrieb Nirmoy:
>>>> On 2/18/20 1:44 PM, Christian König wrote:
>>>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>>>> Hi
>>>>>>
>>>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>>>> GPU address handling is device specific and should be handle by its
>>>>>>> device
>>>>>>> driver.
>>>>>>>
>>>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>>>> ---
>>>>>>>     drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>>>     include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>>>     include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>>>     3 files changed, 10 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>>>         drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>>>         drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>>>         drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>>>         drm_printf(p, "    size: %llu\n", man->size);
>>>>>>>         drm_printf(p, "    available_caching: 0x%08X\n",
>>>>>>> man->available_caching);
>>>>>>>         drm_printf(p, "    default_caching: 0x%08X\n",
>>>>>>> man->default_caching);
>>>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>>>> ttm_buffer_object *bo,
>>>>>>>     moved:
>>>>>>>         bo->evicted = false;
>>>>>>>     -    if (bo->mem.mm_node)
>>>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>>>> -    else
>>>>>>> -        bo->offset = 0;
>>>>>>> -
>>>>>> After moving this into users, the else branch has been lost. Is
>>>>>> 'bo->mem.mm_node' always true?
>>>>> At least for the amdgpu and radeon use cases, yes.
>>>>>
>>>>> But that is a rather good question I mean for it is illegal to get the
>>>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>>>
>>>>> Could be that some driver relied on the behavior to get 0 for the
>>>>> system domain here.
>>>> I wonder how to verify that ?
>>>>
>>>> If I understand correctly:
>>>>
>>>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>>>> system domain.
>>>>
>>>> 2 unfortunately I can't say the same for bochs but it works with this
>>>> patch series so I think bochs is fine as well.
>>>>
>>>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>>>> vmwgfx should be fine.
>>>>
>>>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>>>
>>>> I am not sure about  nouveau as bo->offset is being used in many places.
>>>>
>>>> I could probably mirror the removed logic to nouveau as
>>> I suggest to introduce a ttm helper that contains the original branching
>>> and use it everywhere. Something like
>>>
>>>     s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>>>     {
>>>        if (WARN_ON_ONCE(!bo->mem.mm_node))
>>>            return 0;
>>>        return bo->mem.start << PAGE_SHIFT;
>>>     }
>>>
>>> Could be static inline. The warning should point to broken drivers. This
>>> also gets rid of the ugly shift in the drivers.
>> Big NAK on this. That is exactly what we should NOT do.
>>
>> See the shift belongs into the driver, because it is driver dependent if
>> we work with page or byte offsets.
>>
>> For amdgpu we for example want to work with byte offsets and TTM should
>> not make any assumption about what bo->mem.start actually contains.
> OK. What about something like ttm_bo_pg_offset()? Same code without the
> shift. Would also make it clear that it's a page offset.

That is a rather good idea. We could name that ttm_bo_man_offset() and 
put it into ttm_bo_manager.c next to the manager which allocates them.

It's just that this manager is not used by everybody, so amdgpu, radeon 
and nouveau would still need a separate function.

Christian.

>
> Best regards
> Thomas
>
>> Regards,
>> Christian.
>>
>>> Best regards
>>> Thomas
>>>
>>>
>>>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> index f8015e0318d7..5a6a2af91318 100644
>>>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>>>> *bo, bool evict,
>>>>                   list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>>                           nouveau_vma_map(vma, mem);
>>>>                   }
>>>> +               if (bo->mem.mm_node)
>>>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>>>> +               else
>>>> +                       nvbo->offset = 0;
>>>>           } else {
>>>>                   list_for_each_entry(vma, &nvbo->vma_list, head) {
>>>>                           WARN_ON(ttm_bo_wait(bo, false, false));
>>>>
>>>> Regards,
>>>>
>>>> Nirmoy
>>>>
>>>>
>>>>> Regards,
>>>>> Christian.
>>>>>
>>>>>> Best regards
>>>>>> Thomas
>>>>>>
>>>>>>>         ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>>>         return 0;
>>>>>>>     diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>>>          * either of these locks held.
>>>>>>>          */
>>>>>>>     -    uint64_t offset; /* GPU address space is independent of CPU
>>>>>>> word size */
>>>>>>> -
>>>>>>>         struct sg_table *sg;
>>>>>>>     };
>>>>>>>     diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>>>         bool has_type;
>>>>>>>         bool use_type;
>>>>>>>         uint32_t flags;
>>>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>>>> word size */
>>>>>>>         uint64_t size;
>>>>>>>         uint32_t available_caching;
>>>>>>>         uint32_t default_caching;
>>>>>>>
>>>> _______________________________________________
>>>> dri-devel mailing list
>>>> dri-devel@lists.freedesktop.org
>>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx


[-- Attachment #1.2: Type: text/html, Size: 9757 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Daniel Vetter]

On Tue, Feb 18, 2020 at 07:37:44PM +0100, Christian König wrote:
> Am 18.02.20 um 19:28 schrieb Thomas Zimmermann:
> > Hi
> > 
> > Am 18.02.20 um 19:23 schrieb Christian König:
> > > Am 18.02.20 um 19:16 schrieb Thomas Zimmermann:
> > > > Hi
> > > > 
> > > > Am 18.02.20 um 18:13 schrieb Nirmoy:
> > > > > On 2/18/20 1:44 PM, Christian König wrote:
> > > > > > Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
> > > > > > > Hi
> > > > > > > 
> > > > > > > Am 17.02.20 um 16:04 schrieb Nirmoy Das:
> > > > > > > > GPU address handling is device specific and should be handle by its
> > > > > > > > device
> > > > > > > > driver.
> > > > > > > > 
> > > > > > > > Signed-off-by: Nirmoy Das <nirmoy.das-5C7GfCeVMHo@public.gmane.org>
> > > > > > > > ---
> > > > > > > >     drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
> > > > > > > >     include/drm/ttm/ttm_bo_api.h    | 2 --
> > > > > > > >     include/drm/ttm/ttm_bo_driver.h | 1 -
> > > > > > > >     3 files changed, 10 deletions(-)
> > > > > > > > 
> > > > > > > > diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
> > > > > > > > b/drivers/gpu/drm/ttm/ttm_bo.c
> > > > > > > > index 151edfd8de77..d5885cd609a3 100644
> > > > > > > > --- a/drivers/gpu/drm/ttm/ttm_bo.c
> > > > > > > > +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> > > > > > > > @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
> > > > > > > > ttm_bo_device *bdev, struct drm_printer *p
> > > > > > > >         drm_printf(p, "    has_type: %d\n", man->has_type);
> > > > > > > >         drm_printf(p, "    use_type: %d\n", man->use_type);
> > > > > > > >         drm_printf(p, "    flags: 0x%08X\n", man->flags);
> > > > > > > > -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
> > > > > > > >         drm_printf(p, "    size: %llu\n", man->size);
> > > > > > > >         drm_printf(p, "    available_caching: 0x%08X\n",
> > > > > > > > man->available_caching);
> > > > > > > >         drm_printf(p, "    default_caching: 0x%08X\n",
> > > > > > > > man->default_caching);
> > > > > > > > @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
> > > > > > > > ttm_buffer_object *bo,
> > > > > > > >     moved:
> > > > > > > >         bo->evicted = false;
> > > > > > > >     -    if (bo->mem.mm_node)
> > > > > > > > -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
> > > > > > > > -            bdev->man[bo->mem.mem_type].gpu_offset;
> > > > > > > > -    else
> > > > > > > > -        bo->offset = 0;
> > > > > > > > -
> > > > > > > After moving this into users, the else branch has been lost. Is
> > > > > > > 'bo->mem.mm_node' always true?
> > > > > > At least for the amdgpu and radeon use cases, yes.
> > > > > > 
> > > > > > But that is a rather good question I mean for it is illegal to get the
> > > > > > GPU BO address if it is inaccessible (e.g. in the system domain).
> > > > > > 
> > > > > > Could be that some driver relied on the behavior to get 0 for the
> > > > > > system domain here.
> > > > > I wonder how to verify that ?
> > > > > 
> > > > > If I understand correctly:
> > > > > 
> > > > > 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
> > > > > system domain.
> > > > > 
> > > > > 2 unfortunately I can't say the same for bochs but it works with this
> > > > > patch series so I think bochs is fine as well.
> > > > > 
> > > > > 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
> > > > > vmwgfx should be fine.
> > > > > 
> > > > > 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
> > > > > 
> > > > > I am not sure about  nouveau as bo->offset is being used in many places.
> > > > > 
> > > > > I could probably mirror the removed logic to nouveau as
> > > > I suggest to introduce a ttm helper that contains the original branching
> > > > and use it everywhere. Something like
> > > > 
> > > >     s64 ttm_bo_offset(struct ttm_buffer_object *bo)
> > > >     {
> > > >        if (WARN_ON_ONCE(!bo->mem.mm_node))
> > > >            return 0;
> > > >        return bo->mem.start << PAGE_SHIFT;
> > > >     }
> > > > 
> > > > Could be static inline. The warning should point to broken drivers. This
> > > > also gets rid of the ugly shift in the drivers.
> > > Big NAK on this. That is exactly what we should NOT do.
> > > 
> > > See the shift belongs into the driver, because it is driver dependent if
> > > we work with page or byte offsets.
> > > 
> > > For amdgpu we for example want to work with byte offsets and TTM should
> > > not make any assumption about what bo->mem.start actually contains.
> > OK. What about something like ttm_bo_pg_offset()? Same code without the
> > shift. Would also make it clear that it's a page offset.
> 
> That is a rather good idea. We could name that ttm_bo_man_offset() and put
> it into ttm_bo_manager.c next to the manager which allocates them.
> 
> It's just that this manager is not used by everybody, so amdgpu, radeon and
> nouveau would still need a separate function.

Let me pile on with my bikeshed color choice :-)

I'd do this a wrapper, but for drivers individually. And only for those we
haven't audited yet, or where we think the WARN_ON actually provides
value. So maybe in vram helpers, since that might be used by some drivers
that get stuff wrong. Or maybe noveau isn't audited yet.

That also sidesteps the "what should we call this, drivers are different?"
problem.

Anyway feel free to ignore me, really just a bikeshed at this point.

Cheers, Daniel

> 
> Christian.
> 
> > 
> > Best regards
> > Thomas
> > 
> > > Regards,
> > > Christian.
> > > 
> > > > Best regards
> > > > Thomas
> > > > 
> > > > 
> > > > > diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
> > > > > b/drivers/gpu/drm/nouveau/nouveau_bo.c
> > > > > index f8015e0318d7..5a6a2af91318 100644
> > > > > --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> > > > > +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> > > > > @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
> > > > > *bo, bool evict,
> > > > >                   list_for_each_entry(vma, &nvbo->vma_list, head) {
> > > > >                           nouveau_vma_map(vma, mem);
> > > > >                   }
> > > > > +               if (bo->mem.mm_node)
> > > > > +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
> > > > > +               else
> > > > > +                       nvbo->offset = 0;
> > > > >           } else {
> > > > >                   list_for_each_entry(vma, &nvbo->vma_list, head) {
> > > > >                           WARN_ON(ttm_bo_wait(bo, false, false));
> > > > > 
> > > > > Regards,
> > > > > 
> > > > > Nirmoy
> > > > > 
> > > > > 
> > > > > > Regards,
> > > > > > Christian.
> > > > > > 
> > > > > > > Best regards
> > > > > > > Thomas
> > > > > > > 
> > > > > > > >         ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
> > > > > > > >         return 0;
> > > > > > > >     diff --git a/include/drm/ttm/ttm_bo_api.h
> > > > > > > > b/include/drm/ttm/ttm_bo_api.h
> > > > > > > > index b9bc1b00142e..d6f39ee5bf5d 100644
> > > > > > > > --- a/include/drm/ttm/ttm_bo_api.h
> > > > > > > > +++ b/include/drm/ttm/ttm_bo_api.h
> > > > > > > > @@ -213,8 +213,6 @@ struct ttm_buffer_object {
> > > > > > > >          * either of these locks held.
> > > > > > > >          */
> > > > > > > >     -    uint64_t offset; /* GPU address space is independent of CPU
> > > > > > > > word size */
> > > > > > > > -
> > > > > > > >         struct sg_table *sg;
> > > > > > > >     };
> > > > > > > >     diff --git a/include/drm/ttm/ttm_bo_driver.h
> > > > > > > > b/include/drm/ttm/ttm_bo_driver.h
> > > > > > > > index c9e0fd09f4b2..c8ce6c181abe 100644
> > > > > > > > --- a/include/drm/ttm/ttm_bo_driver.h
> > > > > > > > +++ b/include/drm/ttm/ttm_bo_driver.h
> > > > > > > > @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
> > > > > > > >         bool has_type;
> > > > > > > >         bool use_type;
> > > > > > > >         uint32_t flags;
> > > > > > > > -    uint64_t gpu_offset; /* GPU address space is independent of CPU
> > > > > > > > word size */
> > > > > > > >         uint64_t size;
> > > > > > > >         uint32_t available_caching;
> > > > > > > >         uint32_t default_caching;
> > > > > > > > 
> > > > > _______________________________________________
> > > > > dri-devel mailing list
> > > > > dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
> > > > > https://lists.freedesktop.org/mailman/listinfo/dri-devel
> > > _______________________________________________
> > > dri-devel mailing list
> > > dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
> > > https://lists.freedesktop.org/mailman/listinfo/dri-devel
> > 
> > _______________________________________________
> > amd-gfx mailing list
> > amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
> > https://lists.freedesktop.org/mailman/listinfo/amd-gfx
> 

> _______________________________________________
> Nouveau mailing list
> Nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
> https://lists.freedesktop.org/mailman/listinfo/nouveau


-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy]

On 2/18/20 7:16 PM, Thomas Zimmermann wrote:
> Hi
>
> Am 18.02.20 um 18:13 schrieb Nirmoy:
>> On 2/18/20 1:44 PM, Christian König wrote:
>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>> GPU address handling is device specific and should be handle by its
>>>>> device
>>>>> driver.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> ---
>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>    3 files changed, 10 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>> man->available_caching);
>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>> man->default_caching);
>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>> ttm_buffer_object *bo,
>>>>>    moved:
>>>>>        bo->evicted = false;
>>>>>    -    if (bo->mem.mm_node)
>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>> -    else
>>>>> -        bo->offset = 0;
>>>>> -
>>>> After moving this into users, the else branch has been lost. Is
>>>> 'bo->mem.mm_node' always true?
>>> At least for the amdgpu and radeon use cases, yes.
>>>
>>> But that is a rather good question I mean for it is illegal to get the
>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>
>>> Could be that some driver relied on the behavior to get 0 for the
>>> system domain here.
>> I wonder how to verify that ?
>>
>> If I understand correctly:
>>
>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>> system domain.
>>
>> 2 unfortunately I can't say the same for bochs but it works with this
>> patch series so I think bochs is fine as well.
>>
>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>> vmwgfx should be fine.
>>
>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>
>> I am not sure about  nouveau as bo->offset is being used in many places.
>>
>> I could probably mirror the removed logic to nouveau as
> I suggest to introduce a ttm helper that contains the original branching
> and use it everywhere. Something like
>
>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>    {
>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>           return 0;
>       return bo->mem.start << PAGE_SHIFT;
>    }
>
> Could be static inline. The warning should point to broken drivers. This
> also gets rid of the ugly shift in the drivers.

The idea here is to remove this GPU memory offset calculation and 
reference completely

from ttm and let drivers decide how to handle that. Drivers are full of 
ugly shift anyway imho :)

Regards,

Nirmoy

>
> Best regards
> Thomas
>
>
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index f8015e0318d7..5a6a2af91318 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>> *bo, bool evict,
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          nouveau_vma_map(vma, mem);
>>                  }
>> +               if (bo->mem.mm_node)
>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>> +               else
>> +                       nvbo->offset = 0;
>>          } else {
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>
>> Regards,
>>
>> Nirmoy
>>
>>
>>> Regards,
>>> Christian.
>>>
>>>> Best regards
>>>> Thomas
>>>>
>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>        return 0;
>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>         * either of these locks held.
>>>>>         */
>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>> -
>>>>>        struct sg_table *sg;
>>>>>    };
>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>        bool has_type;
>>>>>        bool use_type;
>>>>>        uint32_t flags;
>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>>        uint64_t size;
>>>>>        uint32_t available_caching;
>>>>>        uint32_t default_caching;
>>>>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy]

On 2/18/20 7:16 PM, Thomas Zimmermann wrote:
> Hi
>
> Am 18.02.20 um 18:13 schrieb Nirmoy:
>> On 2/18/20 1:44 PM, Christian König wrote:
>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>> GPU address handling is device specific and should be handle by its
>>>>> device
>>>>> driver.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> ---
>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>    3 files changed, 10 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>> man->available_caching);
>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>> man->default_caching);
>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>> ttm_buffer_object *bo,
>>>>>    moved:
>>>>>        bo->evicted = false;
>>>>>    -    if (bo->mem.mm_node)
>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>> -    else
>>>>> -        bo->offset = 0;
>>>>> -
>>>> After moving this into users, the else branch has been lost. Is
>>>> 'bo->mem.mm_node' always true?
>>> At least for the amdgpu and radeon use cases, yes.
>>>
>>> But that is a rather good question I mean for it is illegal to get the
>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>
>>> Could be that some driver relied on the behavior to get 0 for the
>>> system domain here.
>> I wonder how to verify that ?
>>
>> If I understand correctly:
>>
>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>> system domain.
>>
>> 2 unfortunately I can't say the same for bochs but it works with this
>> patch series so I think bochs is fine as well.
>>
>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>> vmwgfx should be fine.
>>
>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>
>> I am not sure about  nouveau as bo->offset is being used in many places.
>>
>> I could probably mirror the removed logic to nouveau as
> I suggest to introduce a ttm helper that contains the original branching
> and use it everywhere. Something like
>
>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>    {
>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>           return 0;
>       return bo->mem.start << PAGE_SHIFT;
>    }
>
> Could be static inline. The warning should point to broken drivers. This
> also gets rid of the ugly shift in the drivers.

The idea here is to remove this GPU memory offset calculation and 
reference completely

from ttm and let drivers decide how to handle that. Drivers are full of 
ugly shift anyway imho :)

Regards,

Nirmoy

>
> Best regards
> Thomas
>
>
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index f8015e0318d7..5a6a2af91318 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>> *bo, bool evict,
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          nouveau_vma_map(vma, mem);
>>                  }
>> +               if (bo->mem.mm_node)
>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>> +               else
>> +                       nvbo->offset = 0;
>>          } else {
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>
>> Regards,
>>
>> Nirmoy
>>
>>
>>> Regards,
>>> Christian.
>>>
>>>> Best regards
>>>> Thomas
>>>>
>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>        return 0;
>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>         * either of these locks held.
>>>>>         */
>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>> -
>>>>>        struct sg_table *sg;
>>>>>    };
>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>        bool has_type;
>>>>>        bool use_type;
>>>>>        uint32_t flags;
>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>>        uint64_t size;
>>>>>        uint32_t available_caching;
>>>>>        uint32_t default_caching;
>>>>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy]

On 2/18/20 7:16 PM, Thomas Zimmermann wrote:
> Hi
>
> Am 18.02.20 um 18:13 schrieb Nirmoy:
>> On 2/18/20 1:44 PM, Christian König wrote:
>>> Am 18.02.20 um 13:40 schrieb Thomas Zimmermann:
>>>> Hi
>>>>
>>>> Am 17.02.20 um 16:04 schrieb Nirmoy Das:
>>>>> GPU address handling is device specific and should be handle by its
>>>>> device
>>>>> driver.
>>>>>
>>>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>>>> ---
>>>>>    drivers/gpu/drm/ttm/ttm_bo.c    | 7 -------
>>>>>    include/drm/ttm/ttm_bo_api.h    | 2 --
>>>>>    include/drm/ttm/ttm_bo_driver.h | 1 -
>>>>>    3 files changed, 10 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> index 151edfd8de77..d5885cd609a3 100644
>>>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>>>> @@ -85,7 +85,6 @@ static void ttm_mem_type_debug(struct
>>>>> ttm_bo_device *bdev, struct drm_printer *p
>>>>>        drm_printf(p, "    has_type: %d\n", man->has_type);
>>>>>        drm_printf(p, "    use_type: %d\n", man->use_type);
>>>>>        drm_printf(p, "    flags: 0x%08X\n", man->flags);
>>>>> -    drm_printf(p, "    gpu_offset: 0x%08llX\n", man->gpu_offset);
>>>>>        drm_printf(p, "    size: %llu\n", man->size);
>>>>>        drm_printf(p, "    available_caching: 0x%08X\n",
>>>>> man->available_caching);
>>>>>        drm_printf(p, "    default_caching: 0x%08X\n",
>>>>> man->default_caching);
>>>>> @@ -345,12 +344,6 @@ static int ttm_bo_handle_move_mem(struct
>>>>> ttm_buffer_object *bo,
>>>>>    moved:
>>>>>        bo->evicted = false;
>>>>>    -    if (bo->mem.mm_node)
>>>>> -        bo->offset = (bo->mem.start << PAGE_SHIFT) +
>>>>> -            bdev->man[bo->mem.mem_type].gpu_offset;
>>>>> -    else
>>>>> -        bo->offset = 0;
>>>>> -
>>>> After moving this into users, the else branch has been lost. Is
>>>> 'bo->mem.mm_node' always true?
>>> At least for the amdgpu and radeon use cases, yes.
>>>
>>> But that is a rather good question I mean for it is illegal to get the
>>> GPU BO address if it is inaccessible (e.g. in the system domain).
>>>
>>> Could be that some driver relied on the behavior to get 0 for the
>>> system domain here.
>> I wonder how to verify that ?
>>
>> If I understand correctly:
>>
>> 1 qxl uses bo->offset only in qxl_bo_physical_address() which is not in
>> system domain.
>>
>> 2 unfortunately I can't say the same for bochs but it works with this
>> patch series so I think bochs is fine as well.
>>
>> 3 vmwgfx uses bo->offset only when bo->mem.mem_type == TTM_PL_VRAM so
>> vmwgfx should be fine.
>>
>> 4 amdgpu and radeon runs with 'bo->mem.mm_node' always true
>>
>> I am not sure about  nouveau as bo->offset is being used in many places.
>>
>> I could probably mirror the removed logic to nouveau as
> I suggest to introduce a ttm helper that contains the original branching
> and use it everywhere. Something like
>
>    s64 ttm_bo_offset(struct ttm_buffer_object *bo)
>    {
>       if (WARN_ON_ONCE(!bo->mem.mm_node))
>           return 0;
>       return bo->mem.start << PAGE_SHIFT;
>    }
>
> Could be static inline. The warning should point to broken drivers. This
> also gets rid of the ugly shift in the drivers.

The idea here is to remove this GPU memory offset calculation and 
reference completely

from ttm and let drivers decide how to handle that. Drivers are full of 
ugly shift anyway imho :)

Regards,

Nirmoy

>
> Best regards
> Thomas
>
>
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index f8015e0318d7..5a6a2af91318 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -1317,6 +1317,10 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object
>> *bo, bool evict,
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          nouveau_vma_map(vma, mem);
>>                  }
>> +               if (bo->mem.mm_node)
>> +                       nvbo->offset = (new_reg->start << PAGE_SHIFT);
>> +               else
>> +                       nvbo->offset = 0;
>>          } else {
>>                  list_for_each_entry(vma, &nvbo->vma_list, head) {
>>                          WARN_ON(ttm_bo_wait(bo, false, false));
>>
>> Regards,
>>
>> Nirmoy
>>
>>
>>> Regards,
>>> Christian.
>>>
>>>> Best regards
>>>> Thomas
>>>>
>>>>>        ctx->bytes_moved += bo->num_pages << PAGE_SHIFT;
>>>>>        return 0;
>>>>>    diff --git a/include/drm/ttm/ttm_bo_api.h
>>>>> b/include/drm/ttm/ttm_bo_api.h
>>>>> index b9bc1b00142e..d6f39ee5bf5d 100644
>>>>> --- a/include/drm/ttm/ttm_bo_api.h
>>>>> +++ b/include/drm/ttm/ttm_bo_api.h
>>>>> @@ -213,8 +213,6 @@ struct ttm_buffer_object {
>>>>>         * either of these locks held.
>>>>>         */
>>>>>    -    uint64_t offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>> -
>>>>>        struct sg_table *sg;
>>>>>    };
>>>>>    diff --git a/include/drm/ttm/ttm_bo_driver.h
>>>>> b/include/drm/ttm/ttm_bo_driver.h
>>>>> index c9e0fd09f4b2..c8ce6c181abe 100644
>>>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>>>> @@ -177,7 +177,6 @@ struct ttm_mem_type_manager {
>>>>>        bool has_type;
>>>>>        bool use_type;
>>>>>        uint32_t flags;
>>>>> -    uint64_t gpu_offset; /* GPU address space is independent of CPU
>>>>> word size */
>>>>>        uint64_t size;
>>>>>        uint32_t available_caching;
>>>>>        uint32_t default_caching;
>>>>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Gerd Hoffmann]

  Hi,

> 2 unfortunately I can't say the same for bochs but it works with this patch
> series so I think bochs is fine as well.

bochs needs the offset only to scanout framebuffers, which in turn
requires framebuffers being pinned to vram.  So all green here.

cheers,
  Gerd

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Gerd Hoffmann]

  Hi,

> 2 unfortunately I can't say the same for bochs but it works with this patch
> series so I think bochs is fine as well.

bochs needs the offset only to scanout framebuffers, which in turn
requires framebuffers being pinned to vram.  So all green here.

cheers,
  Gerd

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Gerd Hoffmann]

  Hi,

> 2 unfortunately I can't say the same for bochs but it works with this patch
> series so I think bochs is fine as well.

bochs needs the offset only to scanout framebuffers, which in turn
requires framebuffers being pinned to vram.  So all green here.

cheers,
  Gerd

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 8/8] drm/ttm: do not keep GPU dependent addresses

[Nirmoy]

On 2/24/20 9:04 AM, Gerd Hoffmann wrote:
>    Hi,
>
>> 2 unfortunately I can't say the same for bochs but it works with this patch
>> series so I think bochs is fine as well.
> bochs needs the offset only to scanout framebuffers, which in turn
> requires framebuffers being pinned to vram.  So all green here.
Thanks for confirming!
>
> cheers,
>    Gerd
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel