Support GEM object mappings from I/O memory
Subsystems (24)
Thomas Zimmermann
maintainerv5 cover Support GEM object mappings from I/O memory
DRM's fbdev console uses regular load and store operations to update framebuffer memory. The bochs driver on sparc64 requires the use of I/O-specific load and store operations. We have a workaround, but need a long-term solution to the problem.
This patchset changes GEM's vmap/vunmap interfaces to forward pointers of type struct dma_buf_map and updates the generic fbdev emulation to use them correctly. This enables I/O-memory operations on all framebuffers that require and support them.
Patches #1 to #4 prepare VRAM helpers and drivers.
Next is the update of the GEM vmap functions. Patch #5 adds vmap and vunmap that is usable with TTM-based GEM drivers, and patch #6 updates GEM's vmap/vunmap callback to forward instances of type struct dma_buf_map. While the patch touches many files throughout the DRM modules, the applied changes are mostly trivial interface fixes. Several TTM-based GEM drivers now use the new vmap code. Patch #7 updates GEM's internal vmap/vunmap functions to forward struct dma_buf_map.
With struct dma_buf_map propagated through the layers, patches #8 to #10 convert DRM clients and generic fbdev emulation to use it. Updating the fbdev framebuffer will select the correct functions, either for system or I/O memory.
v5:
* rebase onto latest TTM changes (Chrsitian)
* support TTM premapped memory correctly (Christian)
* implement fb_read/fb_write internally (Sam, Daniel)
* cleanups
v4:
* provide TTM vmap/vunmap plus GEM helpers and convert drivers over (Christian, Daniel)
* remove several empty functions
* more TODOs and documentation (Daniel)
v3:
* recreate the whole patchset on top of struct dma_buf_map
v2:
* RFC patchset
Thomas Zimmermann (10):
drm/vram-helper: Remove invariant parameters from internal kmap
function
drm/cma-helper: Remove empty drm_gem_cma_prime_vunmap() drm/etnaviv: Remove empty etnaviv_gem_prime_vunmap() drm/exynos: Remove empty exynos_drm_gem_prime_{vmap,vunmap}() drm/ttm: Add vmap/vunmap to TTM and TTM GEM helpers drm/gem: Use struct dma_buf_map in GEM vmap ops and convert GEM
backends
drm/gem: Update internal GEM vmap/vunmap interfaces to use struct
dma_buf_map
drm/gem: Store client buffer mappings as struct dma_buf_map
dma-buf-map: Add memcpy and pointer-increment interfaces drm/fb_helper: Support framebuffers in I/O memory
Documentation/gpu/todo.rst | 37 ++- drivers/gpu/drm/Kconfig | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c | 36 --- drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h | 2 - drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 5 +- drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 1 - drivers/gpu/drm/ast/ast_cursor.c | 27 +-- drivers/gpu/drm/ast/ast_drv.h | 7 +- drivers/gpu/drm/bochs/bochs_kms.c | 1 - drivers/gpu/drm/drm_client.c | 38 +-- drivers/gpu/drm/drm_fb_helper.c | 248 ++++++++++++++++++-- drivers/gpu/drm/drm_gem.c | 29 ++- drivers/gpu/drm/drm_gem_cma_helper.c | 27 +-- drivers/gpu/drm/drm_gem_shmem_helper.c | 48 ++-- drivers/gpu/drm/drm_gem_ttm_helper.c | 38 +++ drivers/gpu/drm/drm_gem_vram_helper.c | 117 +++++---- drivers/gpu/drm/drm_internal.h | 5 +- drivers/gpu/drm/drm_prime.c | 14 +- drivers/gpu/drm/etnaviv/etnaviv_drv.h | 3 +- drivers/gpu/drm/etnaviv/etnaviv_gem.c | 1 - drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c | 12 +- drivers/gpu/drm/exynos/exynos_drm_gem.c | 12 - drivers/gpu/drm/exynos/exynos_drm_gem.h | 2 - drivers/gpu/drm/lima/lima_gem.c | 6 +- drivers/gpu/drm/lima/lima_sched.c | 11 +- drivers/gpu/drm/mgag200/mgag200_mode.c | 10 +- drivers/gpu/drm/nouveau/Kconfig | 1 + drivers/gpu/drm/nouveau/nouveau_bo.h | 2 - drivers/gpu/drm/nouveau/nouveau_gem.c | 6 +- drivers/gpu/drm/nouveau/nouveau_gem.h | 2 - drivers/gpu/drm/nouveau/nouveau_prime.c | 20 -- drivers/gpu/drm/panfrost/panfrost_perfcnt.c | 14 +- drivers/gpu/drm/qxl/qxl_display.c | 11 +- drivers/gpu/drm/qxl/qxl_draw.c | 14 +- drivers/gpu/drm/qxl/qxl_drv.h | 11 +- drivers/gpu/drm/qxl/qxl_object.c | 31 ++- drivers/gpu/drm/qxl/qxl_object.h | 2 +- drivers/gpu/drm/qxl/qxl_prime.c | 12 +- drivers/gpu/drm/radeon/radeon.h | 1 - drivers/gpu/drm/radeon/radeon_gem.c | 7 +- drivers/gpu/drm/radeon/radeon_prime.c | 20 -- drivers/gpu/drm/rockchip/rockchip_drm_gem.c | 22 +- drivers/gpu/drm/rockchip/rockchip_drm_gem.h | 4 +- drivers/gpu/drm/tiny/cirrus.c | 10 +- drivers/gpu/drm/tiny/gm12u320.c | 10 +- drivers/gpu/drm/ttm/ttm_bo_util.c | 72 ++++++ drivers/gpu/drm/udl/udl_modeset.c | 8 +- drivers/gpu/drm/vboxvideo/vbox_mode.c | 11 +- drivers/gpu/drm/vc4/vc4_bo.c | 7 +- drivers/gpu/drm/vc4/vc4_drv.h | 2 +- drivers/gpu/drm/vgem/vgem_drv.c | 16 +- drivers/gpu/drm/vkms/vkms_plane.c | 15 +- drivers/gpu/drm/vkms/vkms_writeback.c | 22 +- drivers/gpu/drm/xen/xen_drm_front_gem.c | 18 +- drivers/gpu/drm/xen/xen_drm_front_gem.h | 6 +- include/drm/drm_client.h | 7 +- include/drm/drm_gem.h | 5 +- include/drm/drm_gem_cma_helper.h | 3 +- include/drm/drm_gem_shmem_helper.h | 4 +- include/drm/drm_gem_ttm_helper.h | 6 + include/drm/drm_gem_vram_helper.h | 14 +- include/drm/drm_mode_config.h | 12 - include/drm/ttm/ttm_bo_api.h | 28 +++ include/linux/dma-buf-map.h | 93 +++++++- 64 files changed, 852 insertions(+), 436 deletions(-)
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 1/10 drm/vram-helper: Remove invariant parameters from internal kmap function
The parameters map and is_iomem are always of the same value. Removed them to prepares the function for conversion to struct dma_buf_map.
v4:
* don't check for !kmap->virtual; will always be false
--- drivers/gpu/drm/drm_gem_vram_helper.c | 18 ++++-------------- 1 file changed, 4 insertions(+), 14 deletions(-)
drivers/gpu/drm/drm_gem_vram_helper.c +4 -14
@@ -379,32 +379,22 @@ int drm_gem_vram_unpin(struct drm_gem_vram_object *gbo) }EXPORT_SYMBOL (drm_gem_vram_unpin); -static void *drm_gem_vram_kmap_locked (struct drm_gem_vram_object *gbo, -bool map,bool *is_iomem) +static void *drm_gem_vram_kmap_locked (struct drm_gem_vram_object *gbo) {int ret;struct ttm_bo_kmap_obj *kmap = &gbo->kmap; +bool is_iomem;if (gbo->kmap_use_count >0 ) goto out; -if (kmap->virtual || !map) - goto out; - ret =ttm_bo_kmap (&gbo->bo,0 , gbo->bo.num_pages, kmap);if (ret)return ERR_PTR (ret); out: -if (!kmap->virtual) { -if (is_iomem) - *is_iomem = false; -return NULL ;/* not mapped; don't increment ref */ - } ++gbo->kmap_use_count; -if (is_iomem) -return ttm_kmap_obj_virtual (kmap, is_iomem); -return kmap->virtual; +return ttm_kmap_obj_virtual (kmap, &is_iomem); }static void drm_gem_vram_kunmap_locked (struct drm_gem_vram_object *gbo)
@@ -449,7 +439,7 @@ void *drm_gem_vram_vmap(struct drm_gem_vram_object *gbo) ret =drm_gem_vram_pin_locked (gbo,0 );if (ret) goto err_ttm_bo_unreserve; - base =drm_gem_vram_kmap_locked (gbo, true,NULL ); + base =drm_gem_vram_kmap_locked (gbo);if (IS_ERR (base)) { ret =PTR_ERR (base); goto err_drm_gem_vram_unpin_locked;
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 2/10 drm/cma-helper: Remove empty drm_gem_cma_prime_vunmap()
In reply to Thomas ZimmermannThe function drm_gem_cma_prime_vunmap() is empty. Remove it before changing the interface to use struct drm_buf_map.
--- drivers/gpu/drm/drm_gem_cma_helper.c | 17 ----------------- drivers/gpu/drm/vc4/vc4_bo.c | 1 - include/drm/drm_gem_cma_helper.h | 1 - 3 files changed, 19 deletions(-)
drivers/gpu/drm/drm_gem_cma_helper.c +0 -17
@@ -537,23 +537,6 @@ void *drm_gem_cma_prime_vmap(struct drm_gem_object *obj) }EXPORT_SYMBOL_GPL (drm_gem_cma_prime_vmap); -/** -* drm_gem_cma_prime_vunmap - unmap a CMA GEM object from the kernel's virtual -* address space -* @obj: GEM object -* @vaddr: kernel virtual address where the CMA GEM object was mapped -* -* This function removes a buffer exported via DRM PRIME from the kernel's -* virtual address space. This is a no-op because CMA buffers cannot be -* unmapped from kernel space. Drivers using the CMA helpers should set this -* as their &drm_gem_object_funcs.vunmap callback. -*/ -void drm_gem_cma_prime_vunmap (struct drm_gem_object *obj,void *vaddr) -{ -/* Nothing to do */ -} -EXPORT_SYMBOL_GPL (drm_gem_cma_prime_vunmap); -static const struct drm_gem_object_funcs drm_gem_cma_default_funcs = { .free = drm_gem_cma_free_object, .print_info = drm_gem_cma_print_info,
drivers/gpu/drm/vc4/vc4_bo.c +0 -1
@@ -387,7 +387,6 @@ static const struct drm_gem_object_funcs vc4_gem_object_funcs = { .export = vc4_prime_export, .get_sg_table = drm_gem_cma_prime_get_sg_table, .vmap = vc4_prime_vmap, - .vunmap = drm_gem_cma_prime_vunmap, .vm_ops = &vc4_vm_ops, };
include/drm/drm_gem_cma_helper.h +0 -1
@@ -104,7 +104,6 @@ drm_gem_cma_prime_import_sg_table(struct drm_device *dev,int drm_gem_cma_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma);void *drm_gem_cma_prime_vmap (struct drm_gem_object *obj); -void drm_gem_cma_prime_vunmap (struct drm_gem_object *obj,void *vaddr);struct drm_gem_object *drm_gem_cma_create_object_default_funcs (struct drm_device *dev,size_t size);
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 3/10 drm/etnaviv: Remove empty etnaviv_gem_prime_vunmap()
In reply to Thomas ZimmermannThe function etnaviv_gem_prime_vunmap() is empty. Remove it before changing the interface to use struct drm_buf_map.
--- drivers/gpu/drm/etnaviv/etnaviv_drv.h | 1 - drivers/gpu/drm/etnaviv/etnaviv_gem.c | 1 - drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c | 5 ----- 3 files changed, 7 deletions(-)
drivers/gpu/drm/etnaviv/etnaviv_drv.h +0 -1
@@ -52,7 +52,6 @@ int etnaviv_gem_mmap(struct file *filp, struct vm_area_struct *vma);int etnaviv_gem_mmap_offset (struct drm_gem_object *obj,u64 *offset);struct sg_table *etnaviv_gem_prime_get_sg_table (struct drm_gem_object *obj);void *etnaviv_gem_prime_vmap (struct drm_gem_object *obj); -void etnaviv_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr);int etnaviv_gem_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma); struct drm_gem_object *etnaviv_gem_prime_import_sg_table(struct drm_device *dev,
drivers/gpu/drm/etnaviv/etnaviv_gem.c +0 -1
@@ -571,7 +571,6 @@ static const struct drm_gem_object_funcs etnaviv_gem_object_funcs = { .unpin = etnaviv_gem_prime_unpin, .get_sg_table = etnaviv_gem_prime_get_sg_table, .vmap = etnaviv_gem_prime_vmap, - .vunmap = etnaviv_gem_prime_vunmap, .vm_ops = &vm_ops, };
drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c +0 -5
@@ -27,11 +27,6 @@ void *etnaviv_gem_prime_vmap(struct drm_gem_object *obj)return etnaviv_gem_vmap (obj); } -void etnaviv_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr) -{ -/* TODO msm_gem_vunmap() */ -} -int etnaviv_gem_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma) {
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 4/10 drm/exynos: Remove empty exynos_drm_gem_prime_{vmap, vunmap}()
In reply to Thomas ZimmermannThe functions exynos_drm_gem_prime_{vmap,vunmap}() are empty. Remove them before changing the interface to use struct drm_buf_map. As a side effect of removing drm_gem_prime_vmap(), the error code changes from ENOMEM to EOPNOTSUPP.
--- drivers/gpu/drm/exynos/exynos_drm_gem.c | 12 ------------ drivers/gpu/drm/exynos/exynos_drm_gem.h | 2 -- 2 files changed, 14 deletions(-)
drivers/gpu/drm/exynos/exynos_drm_gem.c +0 -12
@@ -137,8 +137,6 @@ static const struct vm_operations_struct exynos_drm_gem_vm_ops = {static const struct drm_gem_object_funcs exynos_drm_gem_object_funcs = { .free = exynos_drm_gem_free_object, .get_sg_table = exynos_drm_gem_prime_get_sg_table, - .vmap = exynos_drm_gem_prime_vmap, - .vunmap = exynos_drm_gem_prime_vunmap, .vm_ops = &exynos_drm_gem_vm_ops, };
@@ -471,16 +469,6 @@ exynos_drm_gem_prime_import_sg_table(struct drm_device *dev,return &exynos_gem->base; } -void *exynos_drm_gem_prime_vmap (struct drm_gem_object *obj) -{ -return NULL ; -} - -void exynos_drm_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr) -{ -/* Nothing to do */ -} -int exynos_drm_gem_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma) {
drivers/gpu/drm/exynos/exynos_drm_gem.h +0 -2
@@ -107,8 +107,6 @@ struct drm_gem_object * exynos_drm_gem_prime_import_sg_table(struct drm_device *dev, struct dma_buf_attachment *attach, struct sg_table *sgt); -void *exynos_drm_gem_prime_vmap (struct drm_gem_object *obj); -void exynos_drm_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr);int exynos_drm_gem_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma);
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 5/10 drm/ttm: Add vmap/vunmap to TTM and TTM GEM helpers
In reply to Thomas ZimmermannThe new functions ttm_bo_{vmap,vunmap}() map and unmap a TTM BO in kernel address space. The mapping's address is returned as struct dma_buf_map. Each function is a simplified version of TTM's existing kmap code. Both functions respect the memory's location ani/or writecombine flags.
On top TTM's functions, GEM TTM helpers got drm_gem_ttm_{vmap,vunmap}(), two helpers that convert a GEM object into the TTM BO and forward the call to TTM's vmap/vunmap. These helpers can be dropped into the rsp GEM object callbacks.
v5:
* use size_t for storing mapping size (Christian)
* ignore premapped memory areas correctly in ttm_bo_vunmap()
* rebase onto latest TTM interfaces (Christian)
* remove BUG() from ttm_bo_vmap() (Christian)
v4:
* drop ttm_kmap_obj_to_dma_buf() in favor of vmap helpers (Daniel, Christian)
--- drivers/gpu/drm/drm_gem_ttm_helper.c | 38 +++++++++++++++ drivers/gpu/drm/ttm/ttm_bo_util.c | 72 ++++++++++++++++++++++++++++ include/drm/drm_gem_ttm_helper.h | 6 +++ include/drm/ttm/ttm_bo_api.h | 28 +++++++++++ include/linux/dma-buf-map.h | 20 ++++++++ 5 files changed, 164 insertions(+)
drivers/gpu/drm/drm_gem_ttm_helper.c +38 -0
@@ -49,6 +49,44 @@ void drm_gem_ttm_print_info(struct drm_printer *p, unsigned int indent, }EXPORT_SYMBOL (drm_gem_ttm_print_info); +/** +* drm_gem_ttm_vmap() - vmap &ttm_buffer_object +* @gem: GEM object. +* @map: [out] returns the dma-buf mapping. +* +* Maps a GEM object with ttm_bo_vmap(). This function can be used as +* &drm_gem_object_funcs.vmap callback. +* +* Returns: +* 0 on success, or a negative errno code otherwise. +*/ +int drm_gem_ttm_vmap (struct drm_gem_object *gem, +struct dma_buf_map *map) +{ +struct ttm_buffer_object *bo =drm_gem_ttm_of_gem (gem); + +return ttm_bo_vmap (bo, map); + +} +EXPORT_SYMBOL (drm_gem_ttm_vmap); + +/** +* drm_gem_ttm_vunmap() - vunmap &ttm_buffer_object +* @gem: GEM object. +* @map: dma-buf mapping. +* +* Unmaps a GEM object with ttm_bo_vunmap(). This function can be used as +* &drm_gem_object_funcs.vmap callback. +*/ +void drm_gem_ttm_vunmap (struct drm_gem_object *gem, +struct dma_buf_map *map) +{ +struct ttm_buffer_object *bo =drm_gem_ttm_of_gem (gem); + +ttm_bo_vunmap (bo, map); +} +EXPORT_SYMBOL (drm_gem_ttm_vunmap); +/** * drm_gem_ttm_mmap() - mmap &ttm_buffer_object * @gem: GEM object.
drivers/gpu/drm/ttm/ttm_bo_util.c +72 -0
84 lines not on this page · raw message
include/drm/drm_gem_ttm_helper.h +6 -0
@@ -10,11 +10,17 @@#include <drm/ttm/ttm_bo_api.h> #include <drm/ttm/ttm_bo_driver.h> +struct dma_buf_map ; +#define drm_gem_ttm_of_gem (gem_obj) \ container_of(gem_obj, struct ttm_buffer_object, base)void drm_gem_ttm_print_info (struct drm_printer *p,unsigned int indent,const struct drm_gem_object *gem); +int drm_gem_ttm_vmap (struct drm_gem_object *gem, +struct dma_buf_map *map); +void drm_gem_ttm_vunmap (struct drm_gem_object *gem, +struct dma_buf_map *map);int drm_gem_ttm_mmap (struct drm_gem_object *gem,struct vm_area_struct *vma);
include/drm/ttm/ttm_bo_api.h +28 -0
@@ -48,6 +48,8 @@ struct ttm_bo_global;struct ttm_bo_device ; +struct dma_buf_map ; +struct drm_mm_node ;struct ttm_placement ;
@@ -494,6 +496,32 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long start_page,*/ void ttm_bo_kunmap (struct ttm_bo_kmap_obj *map); +/** +* ttm_bo_vmap +* +* @bo: The buffer object. +* @map: pointer to a struct dma_buf_map representing the map. +* +* Sets up a kernel virtual mapping, using ioremap or vmap to the +* data in the buffer object. The parameter @map returns the virtual +* address as struct dma_buf_map. Unmap the buffer with ttm_bo_vunmap(). +* +* Returns +* -ENOMEM: Out of memory. +* -EINVAL: Invalid range. +*/ +int ttm_bo_vmap (struct ttm_buffer_object *bo,struct dma_buf_map *map); + +/** +* ttm_bo_vunmap +* +* @bo: The buffer object. +* @map: Object describing the map to unmap. +* +* Unmaps a kernel map set up by ttm_bo_vmap(). +*/ +void ttm_bo_vunmap (struct ttm_buffer_object *bo,struct dma_buf_map *map); +/** * ttm_bo_mmap_obj - mmap memory backed by a ttm buffer object. *
include/linux/dma-buf-map.h +20 -0
@@ -45,6 +45,12 @@* * dma_buf_map_set_vaddr(&map. 0xdeadbeaf); * +* To set an address in I/O memory, use dma_buf_map_set_vaddr_iomem(). +* +* .. code-block:: c +* +* dma_buf_map_set_vaddr_iomem(&map. 0xdeadbeaf); +* * Test if a mapping is valid with either dma_buf_map_is_set() or * dma_buf_map_is_null(). *
@@ -118,6 +124,20 @@ static inline void dma_buf_map_set_vaddr(struct dma_buf_map *map, void *vaddr) map->is_iomem = false; } +/** +* dma_buf_map_set_vaddr_iomem - Sets a dma-buf mapping structure to an address in I/O memory +* @map: The dma-buf mapping structure +* @vaddr_iomem: An I/O-memory address +* +* Sets the address and the I/O-memory flag. +*/ +static inline void dma_buf_map_set_vaddr_iomem (struct dma_buf_map *map, +void __iomem *vaddr_iomem) +{ + map->vaddr_iomem = vaddr_iomem; + map->is_iomem = true; +} +/** * dma_buf_map_is_equal - Compares two dma-buf mapping structures for equality * @lhs: The dma-buf mapping structure
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 6/10 drm/gem: Use struct dma_buf_map in GEM vmap ops and convert GEM backends
In reply to Thomas ZimmermannThis patch replaces the vmap/vunmap's use of raw pointers in GEM object functions with instances of struct dma_buf_map. GEM backends are converted as well. For most of them, this simply changes the returned type.
TTM-based drivers now return information about the location of the memory, either system or I/O memory. GEM VRAM helpers and qxl now use ttm_bo_vmap() et al. Amdgpu, nouveau and radeon use drm_gem_ttm_vmap() et al instead of implementing their own vmap callbacks.
v5:
* update vkms after switch to shmem
v4:
* use ttm_bo_vmap(), drm_gem_ttm_vmap(), et al. (Daniel, Christian)
* fix a trailing { in drm_gem_vmap()
* remove several empty functions instead of converting them (Daniel)
* comment uses of raw pointers with a TODO (Daniel)
* TODO list: convert more helpers to use struct dma_buf_map
--- Documentation/gpu/todo.rst | 18 ++++ drivers/gpu/drm/Kconfig | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c | 36 ------- drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h | 2 - drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 5 +- drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 1 - drivers/gpu/drm/ast/ast_cursor.c | 27 +++-- drivers/gpu/drm/ast/ast_drv.h | 7 +- drivers/gpu/drm/drm_gem.c | 23 +++-- drivers/gpu/drm/drm_gem_cma_helper.c | 10 +- drivers/gpu/drm/drm_gem_shmem_helper.c | 48 +++++---- drivers/gpu/drm/drm_gem_vram_helper.c | 107 ++++++++++---------- drivers/gpu/drm/etnaviv/etnaviv_drv.h | 2 +- drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c | 9 +- drivers/gpu/drm/lima/lima_gem.c | 6 +- drivers/gpu/drm/lima/lima_sched.c | 11 +- drivers/gpu/drm/mgag200/mgag200_mode.c | 10 +- drivers/gpu/drm/nouveau/Kconfig | 1 + drivers/gpu/drm/nouveau/nouveau_bo.h | 2 - drivers/gpu/drm/nouveau/nouveau_gem.c | 6 +- drivers/gpu/drm/nouveau/nouveau_gem.h | 2 - drivers/gpu/drm/nouveau/nouveau_prime.c | 20 ---- drivers/gpu/drm/panfrost/panfrost_perfcnt.c | 14 +-- drivers/gpu/drm/qxl/qxl_display.c | 11 +- drivers/gpu/drm/qxl/qxl_draw.c | 14 ++- drivers/gpu/drm/qxl/qxl_drv.h | 11 +- drivers/gpu/drm/qxl/qxl_object.c | 31 +++--- drivers/gpu/drm/qxl/qxl_object.h | 2 +- drivers/gpu/drm/qxl/qxl_prime.c | 12 +-- drivers/gpu/drm/radeon/radeon.h | 1 - drivers/gpu/drm/radeon/radeon_gem.c | 7 +- drivers/gpu/drm/radeon/radeon_prime.c | 20 ---- drivers/gpu/drm/rockchip/rockchip_drm_gem.c | 22 ++-- drivers/gpu/drm/rockchip/rockchip_drm_gem.h | 4 +- drivers/gpu/drm/tiny/cirrus.c | 10 +- drivers/gpu/drm/tiny/gm12u320.c | 10 +- drivers/gpu/drm/udl/udl_modeset.c | 8 +- drivers/gpu/drm/vboxvideo/vbox_mode.c | 11 +- drivers/gpu/drm/vc4/vc4_bo.c | 6 +- drivers/gpu/drm/vc4/vc4_drv.h | 2 +- drivers/gpu/drm/vgem/vgem_drv.c | 16 ++- drivers/gpu/drm/vkms/vkms_plane.c | 15 ++- drivers/gpu/drm/vkms/vkms_writeback.c | 22 ++-- drivers/gpu/drm/xen/xen_drm_front_gem.c | 18 ++-- drivers/gpu/drm/xen/xen_drm_front_gem.h | 6 +- include/drm/drm_gem.h | 5 +- include/drm/drm_gem_cma_helper.h | 2 +- include/drm/drm_gem_shmem_helper.h | 4 +- include/drm/drm_gem_vram_helper.h | 14 +-- 49 files changed, 345 insertions(+), 308 deletions(-)
Documentation/gpu/todo.rst +18 -0
@@ -446,6 +446,24 @@ Contact: Ville Syrjälä, Daniel Vetter Level: Intermediate +Use struct dma_buf_map throughout codebase +------------------------------------------ + +Pointers to shared device memory are stored in struct dma_buf_map. Each +instance knows whether it refers to system or I/O memory. Most of the DRM-wide +interface have been converted to use struct dma_buf_map, but implementations +often still use raw pointers. + +The task is to use struct dma_buf_map where it makes sense. + +* Memory managers should use struct dma_buf_map for dma-buf-imported buffers. +* TTM might benefit from using struct dma_buf_map internally. +* Framebuffer copying and blitting helpers should operate on struct dma_buf_map. + +Contact: Thomas Zimmermann <[email protected]>, Christian König, Daniel Vetter + +Level: Intermediate + Core refactorings =================
drivers/gpu/drm/Kconfig +2 -0
@@ -239,6 +239,7 @@ config DRM_RADEON select FW_LOADER select DRM_KMS_HELPER select DRM_TTM + select DRM_TTM_HELPER select POWER_SUPPLY select HWMON select BACKLIGHT_CLASS_DEVICE
@@ -259,6 +260,7 @@ config DRM_AMDGPU select DRM_KMS_HELPER select DRM_SCHED select DRM_TTM + select DRM_TTM_HELPER select POWER_SUPPLY select HWMON select BACKLIGHT_CLASS_DEVICE
drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +0 -36
@@ -41,42 +41,6 @@#include <linux/dma-fence-array.h> #include <linux/pci-p2pdma.h> -/** -* amdgpu_gem_prime_vmap - &dma_buf_ops.vmap implementation -* @obj: GEM BO -* -* Sets up an in-kernel virtual mapping of the BO's memory. -* -* Returns: -* The virtual address of the mapping or an error pointer. -*/ -void *amdgpu_gem_prime_vmap (struct drm_gem_object *obj) -{ -struct amdgpu_bo *bo =gem_to_amdgpu_bo (obj); -int ret; - - ret =ttm_bo_kmap (&bo->tbo,0 , bo->tbo.num_pages, - &bo->dma_buf_vmap); -if (ret) -return ERR_PTR (ret); - -return bo->dma_buf_vmap.virtual; -} - -/** -* amdgpu_gem_prime_vunmap - &dma_buf_ops.vunmap implementation -* @obj: GEM BO -* @vaddr: Virtual address (unused) -* -* Tears down the in-kernel virtual mapping of the BO's memory. -*/ -void amdgpu_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr) -{ -struct amdgpu_bo *bo =gem_to_amdgpu_bo (obj); - -ttm_bo_kunmap (&bo->dma_buf_vmap); -} -/** * amdgpu_gem_prime_mmap - &drm_driver.gem_prime_mmap implementation * @obj: GEM BO
drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h +0 -2
@@ -31,8 +31,6 @@ struct drm_gem_object *amdgpu_gem_prime_import(struct drm_device *dev, struct dma_buf *dma_buf);bool amdgpu_dmabuf_is_xgmi_accessible (struct amdgpu_device *adev,struct amdgpu_bo *bo); -void *amdgpu_gem_prime_vmap (struct drm_gem_object *obj); -void amdgpu_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr);int amdgpu_gem_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma);
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +3 -2
@@ -33,6 +33,7 @@#include <drm/amdgpu_drm.h> #include <drm/drm_debugfs.h> +#include <drm/drm_gem_ttm_helper.h> #include "amdgpu.h" #include "amdgpu_display.h"
@@ -220,8 +221,8 @@ static const struct drm_gem_object_funcs amdgpu_gem_object_funcs = { .open = amdgpu_gem_object_open, .close = amdgpu_gem_object_close, .export = amdgpu_gem_prime_export, - .vmap = amdgpu_gem_prime_vmap, - .vunmap = amdgpu_gem_prime_vunmap, + .vmap = drm_gem_ttm_vmap, + .vunmap = drm_gem_ttm_vunmap, };/*
drivers/gpu/drm/amd/amdgpu/amdgpu_object.h +0 -1
@@ -100,7 +100,6 @@ struct amdgpu_bo {struct amdgpu_bo *parent;struct amdgpu_bo *shadow; -struct ttm_bo_kmap_obj dma_buf_vmap;struct amdgpu_mn *mn;
drivers/gpu/drm/ast/ast_cursor.c +13 -14
@@ -39,7 +39,7 @@ static void ast_cursor_fini(struct ast_private *ast)for (i =0 ; i <ARRAY_SIZE (ast->cursor.gbo); ++i) { gbo = ast->cursor.gbo[i]; -drm_gem_vram_vunmap (gbo, ast->cursor.vaddr[i]); +drm_gem_vram_vunmap (gbo, &ast->cursor.map[i]);drm_gem_vram_unpin (gbo);drm_gem_vram_put (gbo); }
@@ -60,7 +60,7 @@ int ast_cursor_init(struct ast_private *ast)struct drm_device *dev = &ast->base;size_t size, i;struct drm_gem_vram_object *gbo; -void __iomem *vaddr; +struct dma_buf_map map;int ret; size =roundup (AST_HWC_SIZE +AST_HWC_SIGNATURE_SIZE ,PAGE_SIZE );
@@ -77,16 +77,15 @@ int ast_cursor_init(struct ast_private *ast)drm_gem_vram_put (gbo); goto err_drm_gem_vram_put; } - vaddr =drm_gem_vram_vmap (gbo); -if (IS_ERR (vaddr)) { - ret =PTR_ERR (vaddr); + ret =drm_gem_vram_vmap (gbo, &map); +if (ret) {drm_gem_vram_unpin (gbo);drm_gem_vram_put (gbo); goto err_drm_gem_vram_put; } ast->cursor.gbo[i] = gbo; - ast->cursor.vaddr[i] = vaddr; + ast->cursor.map[i] = map; }return drmm_add_action_or_reset (dev, ast_cursor_release,NULL );
@@ -95,7 +94,7 @@ int ast_cursor_init(struct ast_private *ast)while (i) { --i; gbo = ast->cursor.gbo[i]; -drm_gem_vram_vunmap (gbo, ast->cursor.vaddr[i]); +drm_gem_vram_vunmap (gbo, &ast->cursor.map[i]);drm_gem_vram_unpin (gbo);drm_gem_vram_put (gbo); }
@@ -170,6 +169,7 @@ int ast_cursor_blit(struct ast_private *ast, struct drm_framebuffer *fb) {struct drm_device *dev = &ast->base;struct drm_gem_vram_object *gbo; +struct dma_buf_map map;int ret;void *src;void __iomem *dst;
@@ -183,18 +183,17 @@ int ast_cursor_blit(struct ast_private *ast, struct drm_framebuffer *fb) ret =drm_gem_vram_pin (gbo,0 );if (ret)return ret; - src =drm_gem_vram_vmap (gbo); -if (IS_ERR (src)) { - ret =PTR_ERR (src); + ret =drm_gem_vram_vmap (gbo, &map); +if (ret) goto err_drm_gem_vram_unpin; - } + src = map.vaddr;/* TODO: Use mapping abstraction properly */ - dst = ast->cursor.vaddr[ast->cursor.next_index]; + dst = ast->cursor.map[ast->cursor.next_index].vaddr_iomem;/* do data transfer to cursor BO */ update_cursor_image (dst, src, fb->width, fb->height); -drm_gem_vram_vunmap (gbo, src); +drm_gem_vram_vunmap (gbo, &map);drm_gem_vram_unpin (gbo);return 0 ;
@@ -257,7 +256,7 @@ void ast_cursor_show(struct ast_private *ast, int x, int y,u8 __iomem *sig;u8 jreg; - dst = ast->cursor.vaddr[ast->cursor.next_index]; + dst = ast->cursor.map[ast->cursor.next_index].vaddr; sig = dst +AST_HWC_SIZE ;writel (x, sig +AST_HWC_SIGNATURE_X );
drivers/gpu/drm/ast/ast_drv.h +4 -3
@@ -28,10 +28,11 @@#ifndef __AST_DRV_H__#define __AST_DRV_H__ -#include <linux/types.h> -#include <linux/io.h> +#include <linux/dma-buf-map.h> #include <linux/i2c.h> #include <linux/i2c-algo-bit.h> +#include <linux/io.h> +#include <linux/types.h> #include <drm/drm_connector.h> #include <drm/drm_crtc.h>
@@ -131,7 +132,7 @@ struct ast_private {struct {struct drm_gem_vram_object *gbo[AST_DEFAULT_HWC_NUM ]; -void __iomem *vaddr[AST_DEFAULT_HWC_NUM ]; +struct dma_buf_map map[AST_DEFAULT_HWC_NUM ];unsigned int next_index; } cursor;
drivers/gpu/drm/drm_gem.c +14 -9
@@ -36,6 +36,7 @@#include <linux/pagemap.h> #include <linux/shmem_fs.h> #include <linux/dma-buf.h> +#include <linux/dma-buf-map.h> #include <linux/mem_encrypt.h> #include <linux/pagevec.h>
@@ -1207,26 +1208,30 @@ void drm_gem_unpin(struct drm_gem_object *obj)void *drm_gem_vmap (struct drm_gem_object *obj) { -void *vaddr; +struct dma_buf_map map; +int ret; -if (obj->funcs->vmap) - vaddr = obj->funcs->vmap (obj); -else - vaddr =ERR_PTR (-EOPNOTSUPP ); +if (!obj->funcs->vmap) +return ERR_PTR (-EOPNOTSUPP ); -if (!vaddr) - vaddr =ERR_PTR (-ENOMEM ); + ret = obj->funcs->vmap (obj, &map); +if (ret) +return ERR_PTR (ret); +else if (dma_buf_map_is_null (&map)) +return ERR_PTR (-ENOMEM ); -return vaddr; +return map.vaddr; }void drm_gem_vunmap (struct drm_gem_object *obj,void *vaddr) { +struct dma_buf_map map =DMA_BUF_MAP_INIT_VADDR (vaddr); +if (!vaddr)return ;if (obj->funcs->vunmap) - obj->funcs->vunmap (obj, vaddr); + obj->funcs->vunmap (obj, &map); }/**
drivers/gpu/drm/drm_gem_cma_helper.c +7 -3
@@ -519,6 +519,8 @@ EXPORT_SYMBOL_GPL(drm_gem_cma_prime_mmap);* drm_gem_cma_prime_vmap - map a CMA GEM object into the kernel's virtual * address space * @obj: GEM object +* @map: Returns the kernel virtual address of the CMA GEM object's backing +* store. * * This function maps a buffer exported via DRM PRIME into the kernel's * virtual address space. Since the CMA buffers are already mapped into the
@@ -527,13 +529,15 @@ EXPORT_SYMBOL_GPL(drm_gem_cma_prime_mmap);* driver's &drm_gem_object_funcs.vmap callback. * * Returns: -* The kernel virtual address of the CMA GEM object's backing store. +* 0 on success, or a negative error code otherwise. */ -void *drm_gem_cma_prime_vmap (struct drm_gem_object *obj) +int drm_gem_cma_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map) {struct drm_gem_cma_object *cma_obj =to_drm_gem_cma_obj (obj); -return cma_obj->vaddr; +dma_buf_map_set_vaddr (map, cma_obj->vaddr); + +return 0 ; }EXPORT_SYMBOL_GPL (drm_gem_cma_prime_vmap);
drivers/gpu/drm/drm_gem_shmem_helper.c +29 -19
124 lines not on this page · raw message
drivers/gpu/drm/drm_gem_vram_helper.c +54 -53
210 lines not on this page · raw message
drivers/gpu/drm/etnaviv/etnaviv_drv.h +1 -1
@@ -51,7 +51,7 @@ int etnaviv_ioctl_gem_submit(struct drm_device *dev, void *data,int etnaviv_gem_mmap (struct file *filp,struct vm_area_struct *vma);int etnaviv_gem_mmap_offset (struct drm_gem_object *obj,u64 *offset);struct sg_table *etnaviv_gem_prime_get_sg_table (struct drm_gem_object *obj); -void *etnaviv_gem_prime_vmap (struct drm_gem_object *obj); +int etnaviv_gem_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map);int etnaviv_gem_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma); struct drm_gem_object *etnaviv_gem_prime_import_sg_table(struct drm_device *dev,
drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c +7 -2
@@ -22,9 +22,14 @@ struct sg_table *etnaviv_gem_prime_get_sg_table(struct drm_gem_object *obj)return drm_prime_pages_to_sg (obj->dev, etnaviv_obj->pages, npages); } -void *etnaviv_gem_prime_vmap (struct drm_gem_object *obj) +int etnaviv_gem_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map) { -return etnaviv_gem_vmap (obj); +void *vaddr =etnaviv_gem_vmap (obj); +if (!vaddr) +return -ENOMEM ; +dma_buf_map_set_vaddr (map, vaddr); + +return 0 ; } int etnaviv_gem_prime_mmap(struct drm_gem_object *obj,
drivers/gpu/drm/lima/lima_gem.c +3 -3
@@ -182,14 +182,14 @@ static int lima_gem_pin(struct drm_gem_object *obj)return drm_gem_shmem_pin (obj); } -static void *lima_gem_vmap (struct drm_gem_object *obj) +static int lima_gem_vmap (struct drm_gem_object *obj,struct dma_buf_map *map) {struct lima_bo *bo =to_lima_bo (obj);if (bo->heap_size) -return ERR_PTR (-EINVAL ); +return -EINVAL ; -return drm_gem_shmem_vmap (obj); +return drm_gem_shmem_vmap (obj, map); }static int lima_gem_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma)
drivers/gpu/drm/lima/lima_sched.c +7 -4
@@ -1,6 +1,7 @@// SPDX-License-Identifier: GPL-2.0 OR MIT /* Copyright 2017-2019 Qiang Yu <[email protected]> */ +#include <linux/dma-buf-map.h> #include <linux/kthread.h> #include <linux/slab.h> #include <linux/vmalloc.h>
@@ -303,6 +304,8 @@ static void lima_sched_build_error_task_list(struct lima_sched_task *task)struct lima_dump_chunk_buffer *buffer_chunk;u32 size, task_size, mem_size;int i; +struct dma_buf_map map; +int ret;mutex_lock (&dev->error_task_list_lock);
@@ -388,15 +391,15 @@ static void lima_sched_build_error_task_list(struct lima_sched_task *task) }else { buffer_chunk->size =lima_bo_size (bo); - data =drm_gem_shmem_vmap (&bo->base.base); -if (IS_ERR_OR_NULL (data)) { + ret =drm_gem_shmem_vmap (&bo->base.base, &map); +if (ret) {kvfree (et); goto out; } -memcpy (buffer_chunk +1 , data, buffer_chunk->size); +memcpy (buffer_chunk +1 , map.vaddr, buffer_chunk->size); -drm_gem_shmem_vunmap (&bo->base.base, data); +drm_gem_shmem_vunmap (&bo->base.base, &map); } buffer_chunk = (void *)(buffer_chunk +1 ) + buffer_chunk->size;
drivers/gpu/drm/mgag200/mgag200_mode.c +7 -3
@@ -9,6 +9,7 @@*/ #include <linux/delay.h> +#include <linux/dma-buf-map.h> #include <drm/drm_atomic_helper.h> #include <drm/drm_atomic_state_helper.h>
@@ -1556,15 +1557,18 @@ mgag200_handle_damage(struct mga_device *mdev, struct drm_framebuffer *fb, struct drm_rect *clip) {struct drm_device *dev = &mdev->base; +struct dma_buf_map map;void *vmap; +int ret; - vmap =drm_gem_shmem_vmap (fb->obj[0 ]); -if (drm_WARN_ON (dev, !vmap)) + ret =drm_gem_shmem_vmap (fb->obj[0 ], &map); +if (drm_WARN_ON (dev, ret))return ;/* BUG: SHMEM BO should always be vmapped */ + vmap = map.vaddr;/* TODO: Use mapping abstraction properly */ drm_fb_memcpy_dstclip (mdev->vram, vmap, fb, clip); -drm_gem_shmem_vunmap (fb->obj[0 ], vmap); +drm_gem_shmem_vunmap (fb->obj[0 ], &map);/* Always scanout image at VRAM offset 0 */ mgag200_set_startadd (mdev, (u32 )0 );
drivers/gpu/drm/nouveau/Kconfig +1 -0
@@ -6,6 +6,7 @@ config DRM_NOUVEAU select FW_LOADER select DRM_KMS_HELPER select DRM_TTM + select DRM_TTM_HELPER select BACKLIGHT_CLASS_DEVICE if DRM_NOUVEAU_BACKLIGHT select ACPI_VIDEO if ACPI && X86 && BACKLIGHT_CLASS_DEVICE && INPUT select X86_PLATFORM_DEVICES if ACPI && X86
drivers/gpu/drm/nouveau/nouveau_bo.h +0 -2
@@ -39,8 +39,6 @@ struct nouveau_bo {unsigned mode;struct nouveau_drm_tile *tile; - -struct ttm_bo_kmap_obj dma_buf_vmap; }; static inline struct nouveau_bo *
drivers/gpu/drm/nouveau/nouveau_gem.c +4 -2
@@ -24,6 +24,8 @@* */ +#include <drm/drm_gem_ttm_helper.h> +#include "nouveau_drv.h" #include "nouveau_dma.h" #include "nouveau_fence.h"
@@ -176,8 +178,8 @@ const struct drm_gem_object_funcs nouveau_gem_object_funcs = { .pin = nouveau_gem_prime_pin, .unpin = nouveau_gem_prime_unpin, .get_sg_table = nouveau_gem_prime_get_sg_table, - .vmap = nouveau_gem_prime_vmap, - .vunmap = nouveau_gem_prime_vunmap, + .vmap = drm_gem_ttm_vmap, + .vunmap = drm_gem_ttm_vunmap, };int
drivers/gpu/drm/nouveau/nouveau_gem.h +0 -2
@@ -37,7 +37,5 @@ extern void nouveau_gem_prime_unpin(struct drm_gem_object *);extern struct sg_table *nouveau_gem_prime_get_sg_table (struct drm_gem_object *);extern struct drm_gem_object *nouveau_gem_prime_import_sg_table (struct drm_device *,struct dma_buf_attachment *,struct sg_table *); -extern void *nouveau_gem_prime_vmap (struct drm_gem_object *); -extern void nouveau_gem_prime_vunmap (struct drm_gem_object *,void *);#endif
drivers/gpu/drm/nouveau/nouveau_prime.c +0 -20
@@ -35,26 +35,6 @@ struct sg_table *nouveau_gem_prime_get_sg_table(struct drm_gem_object *obj)return drm_prime_pages_to_sg (obj->dev, nvbo->bo.ttm->pages, npages); } -void *nouveau_gem_prime_vmap (struct drm_gem_object *obj) -{ -struct nouveau_bo *nvbo =nouveau_gem_object (obj); -int ret; - - ret =ttm_bo_kmap (&nvbo->bo,0 , nvbo->bo.num_pages, - &nvbo->dma_buf_vmap); -if (ret) -return ERR_PTR (ret); - -return nvbo->dma_buf_vmap.virtual; -} - -void nouveau_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr) -{ -struct nouveau_bo *nvbo =nouveau_gem_object (obj); - -ttm_bo_kunmap (&nvbo->dma_buf_vmap); -} -struct drm_gem_object *nouveau_gem_prime_import_sg_table (struct drm_device *dev,struct dma_buf_attachment *attach,struct sg_table *sg)
drivers/gpu/drm/panfrost/panfrost_perfcnt.c +8 -6
@@ -5,6 +5,7 @@#include <drm/drm_gem_shmem_helper.h> #include <drm/panfrost_drm.h> #include <linux/completion.h> +#include <linux/dma-buf-map.h> #include <linux/iopoll.h> #include <linux/pm_runtime.h> #include <linux/slab.h>
@@ -72,6 +73,7 @@ static int panfrost_perfcnt_enable_locked(struct panfrost_device *pfdev, {struct panfrost_file_priv *user = file_priv->driver_priv;struct panfrost_perfcnt *perfcnt = pfdev->perfcnt; +struct dma_buf_map map;struct drm_gem_shmem_object *bo;u32 cfg, as;int ret;
@@ -103,11 +105,10 @@ static int panfrost_perfcnt_enable_locked(struct panfrost_device *pfdev, goto err_close_bo; } - perfcnt->buf =drm_gem_shmem_vmap (&bo->base); -if (IS_ERR (perfcnt->buf)) { - ret =PTR_ERR (perfcnt->buf); + ret =drm_gem_shmem_vmap (&bo->base, &map); +if (ret) goto err_put_mapping; - } + perfcnt->buf = map.vaddr;/* * Invalidate the cache and clear the counters to start from a fresh
@@ -163,7 +164,7 @@ static int panfrost_perfcnt_enable_locked(struct panfrost_device *pfdev,return 0 ; err_vunmap: -drm_gem_shmem_vunmap (&bo->base, perfcnt->buf); +drm_gem_shmem_vunmap (&bo->base, &map); err_put_mapping:panfrost_gem_mapping_put (perfcnt->mapping); err_close_bo:
@@ -180,6 +181,7 @@ static int panfrost_perfcnt_disable_locked(struct panfrost_device *pfdev, {struct panfrost_file_priv *user = file_priv->driver_priv;struct panfrost_perfcnt *perfcnt = pfdev->perfcnt; +struct dma_buf_map map =DMA_BUF_MAP_INIT_VADDR (perfcnt->buf);if (user != perfcnt->user)return -EINVAL ;
@@ -192,7 +194,7 @@ static int panfrost_perfcnt_disable_locked(struct panfrost_device *pfdev, GPU_PERFCNT_CFG_MODE(GPU_PERFCNT_CFG_MODE_OFF)); perfcnt->user =NULL ; -drm_gem_shmem_vunmap (&perfcnt->mapping->obj->base.base, perfcnt->buf); +drm_gem_shmem_vunmap (&perfcnt->mapping->obj->base.base, &map); perfcnt->buf =NULL ;panfrost_gem_close (&perfcnt->mapping->obj->base.base, file_priv);panfrost_mmu_as_put (pfdev, perfcnt->mapping->mmu);
drivers/gpu/drm/qxl/qxl_display.c +8 -3
@@ -25,6 +25,7 @@#include <linux/crc32.h> #include <linux/delay.h> +#include <linux/dma-buf-map.h> #include <drm/drm_drv.h> #include <drm/drm_atomic.h>
@@ -581,6 +582,8 @@ static void qxl_cursor_atomic_update(struct drm_plane *plane,struct drm_gem_object *obj;struct qxl_bo *cursor_bo =NULL , *user_bo =NULL , *old_cursor_bo =NULL ;int ret; +struct dma_buf_map user_map; +struct dma_buf_map cursor_map;void *user_ptr;int size =64 *64 *4 ;
@@ -595,9 +598,10 @@ static void qxl_cursor_atomic_update(struct drm_plane *plane, user_bo =gem_to_qxl_bo (obj);/* pinning is done in the prepare/cleanup framevbuffer */ - ret =qxl_bo_kmap (user_bo, &user_ptr); + ret =qxl_bo_kmap (user_bo, &user_map);if (ret) goto out_free_release; + user_ptr = user_map.vaddr;/* TODO: Use mapping abstraction properly */ ret = qxl_alloc_bo_reserved(qdev, release, sizeof(struct qxl_cursor) + size,
@@ -613,7 +617,7 @@ static void qxl_cursor_atomic_update(struct drm_plane *plane,if (ret) goto out_unpin; - ret =qxl_bo_kmap (cursor_bo, (void **)&cursor); + ret =qxl_bo_kmap (cursor_bo, &cursor_map);if (ret) goto out_backoff;
@@ -1133,6 +1137,7 @@ int qxl_create_monitors_object(struct qxl_device *qdev) {int ret;struct drm_gem_object *gobj; +struct dma_buf_map map;int monitors_config_size =sizeof (struct qxl_monitors_config ) + qxl_num_crtc *sizeof (struct qxl_head );
@@ -1149,7 +1154,7 @@ int qxl_create_monitors_object(struct qxl_device *qdev)if (ret)return ret; -qxl_bo_kmap (qdev->monitors_config_bo,NULL ); +qxl_bo_kmap (qdev->monitors_config_bo, &map); qdev->monitors_config = qdev->monitors_config_bo->kptr; qdev->ram_header->monitors_config =
drivers/gpu/drm/qxl/qxl_draw.c +10 -4
@@ -20,6 +20,8 @@* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ +#include <linux/dma-buf-map.h> +#include <drm/drm_fourcc.h> #include "qxl_drv.h"
@@ -42,13 +44,15 @@ static struct qxl_rect *drawable_set_clipping(struct qxl_device *qdev,unsigned int num_clips, struct qxl_bo *clips_bo) { + struct dma_buf_map map;struct qxl_clip_rects *dev_clips;int ret; - ret =qxl_bo_kmap (clips_bo, (void **)&dev_clips); -if (ret) { + ret =qxl_bo_kmap (clips_bo, &map); +if (ret)return NULL ; - } + dev_clips = map.vaddr;/* TODO: Use mapping abstraction properly */ + dev_clips->num_rects = num_clips; dev_clips->chunk.next_chunk =0 ; dev_clips->chunk.prev_chunk =0 ;
@@ -142,6 +146,7 @@ void qxl_draw_dirty_fb(struct qxl_device *qdev,int stride = fb->pitches[0 ];/* depth is not actually interesting, we don't mask with it */ int depth = fb->format->cpp[0 ] *8 ; +struct dma_buf_map surface_map;uint8_t *surface_base;struct qxl_release *release;struct qxl_bo *clips_bo;
@@ -197,9 +202,10 @@ void qxl_draw_dirty_fb(struct qxl_device *qdev,if (ret) goto out_release_backoff; - ret =qxl_bo_kmap (bo, (void **)&surface_base); + ret =qxl_bo_kmap (bo, &surface_map);if (ret) goto out_release_backoff; + surface_base = surface_map.vaddr;/* TODO: Use mapping abstraction properly */ ret = qxl_image_init(qdev, release, dimage, surface_base, left - dumb_shadow_offset,
drivers/gpu/drm/qxl/qxl_drv.h +7 -4
@@ -30,6 +30,7 @@* Definitions taken from spice-protocol, plus kernel driver specific bits. */ +#include <linux/dma-buf-map.h> #include <linux/dma-fence.h> #include <linux/firmware.h> #include <linux/platform_device.h>
@@ -50,6 +51,8 @@#include "qxl_dev.h" +struct dma_buf_map ; +#define DRIVER_AUTHOR "Dave Airlie"#define DRIVER_NAME "qxl"
@@ -79,7 +82,7 @@ struct qxl_bo {/* Protected by tbo.reserved */ struct ttm_place placements[3 ];struct ttm_placement placement; -struct ttm_bo_kmap_obj kmap; +struct dma_buf_map map;void *kptr;unsigned int map_count;int type;
@@ -335,7 +338,6 @@ int qxl_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_priv);void qxl_gem_object_close (struct drm_gem_object *obj,struct drm_file *file_priv);void qxl_bo_force_delete (struct qxl_device *qdev); -int qxl_bo_kmap (struct qxl_bo *bo,void **ptr);/* qxl_dumb.c */ int qxl_mode_dumb_create(struct drm_file *file_priv,
@@ -445,8 +447,9 @@ struct sg_table *qxl_gem_prime_get_sg_table(struct drm_gem_object *obj);struct drm_gem_object *qxl_gem_prime_import_sg_table (struct drm_device *dev,struct dma_buf_attachment *attach,struct sg_table *sgt); -void *qxl_gem_prime_vmap (struct drm_gem_object *obj); -void qxl_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr); +int qxl_gem_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map); +void qxl_gem_prime_vunmap (struct drm_gem_object *obj, +struct dma_buf_map *map);int qxl_gem_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma);
drivers/gpu/drm/qxl/qxl_object.c +19 -12
@@ -23,10 +23,12 @@* Alon Levy */ +#include <linux/dma-buf-map.h> +#include <linux/io-mapping.h> +#include "qxl_drv.h" #include "qxl_object.h" -#include <linux/io-mapping.h> static void qxl_ttm_bo_destroy (struct ttm_buffer_object *tbo) {struct qxl_bo *bo;
@@ -152,24 +154,27 @@ int qxl_bo_create(struct qxl_device *qdev,return 0 ; } -int qxl_bo_kmap (struct qxl_bo *bo,void **ptr) +int qxl_bo_kmap (struct qxl_bo *bo,struct dma_buf_map *map) { -bool is_iomem;int r;if (bo->kptr) { -if (ptr) - *ptr = bo->kptr; bo->map_count++; -return 0 ; + goto out; } - r =ttm_bo_kmap (&bo->tbo,0 , bo->tbo.num_pages, &bo->kmap); + r =ttm_bo_vmap (&bo->tbo, &bo->map);if (r)return r; - bo->kptr =ttm_kmap_obj_virtual (&bo->kmap, &is_iomem); -if (ptr) - *ptr = bo->kptr; bo->map_count =1 ; + +/* TODO: Remove kptr in favor of map everywhere. */ +if (bo->map.is_iomem) + bo->kptr = (void *)bo->map.vaddr_iomem; +else + bo->kptr = bo->map.vaddr; + +out: + *map = bo->map;return 0 ; }
@@ -180,6 +185,7 @@ void *qxl_bo_kmap_atomic_page(struct qxl_device *qdev,void *rptr;int ret;struct io_mapping *map; +struct dma_buf_map bo_map;if (bo->tbo.mem.mem_type ==TTM_PL_VRAM ) map = qdev->vram_mapping;
@@ -196,9 +202,10 @@ void *qxl_bo_kmap_atomic_page(struct qxl_device *qdev,return rptr; } - ret =qxl_bo_kmap (bo, &rptr); + ret =qxl_bo_kmap (bo, &bo_map);if (ret)return NULL ; + rptr = bo_map.vaddr;/* TODO: Use mapping abstraction properly */ rptr += page_offset *PAGE_SIZE ;return rptr;
@@ -212,7 +219,7 @@ void qxl_bo_kunmap(struct qxl_bo *bo)if (bo->map_count >0 )return ; bo->kptr =NULL ; -ttm_bo_kunmap (&bo->kmap); +ttm_bo_vunmap (&bo->tbo, &bo->map); } void qxl_bo_kunmap_atomic_page(struct qxl_device *qdev,
drivers/gpu/drm/qxl/qxl_object.h +1 -1
@@ -63,7 +63,7 @@ extern int qxl_bo_create(struct qxl_device *qdev, bool kernel, bool pinned, u32 domain, struct qxl_surface *surf, struct qxl_bo **bo_ptr); -extern int qxl_bo_kmap (struct qxl_bo *bo,void **ptr); +extern int qxl_bo_kmap (struct qxl_bo *bo,struct dma_buf_map *map);extern void qxl_bo_kunmap (struct qxl_bo *bo);void *qxl_bo_kmap_atomic_page (struct qxl_device *qdev,struct qxl_bo *bo,int page_offset);void qxl_bo_kunmap_atomic_page (struct qxl_device *qdev,struct qxl_bo *bo,void *map);
drivers/gpu/drm/qxl/qxl_prime.c +6 -6
@@ -54,20 +54,20 @@ struct drm_gem_object *qxl_gem_prime_import_sg_table(return ERR_PTR (-ENOSYS ); } -void *qxl_gem_prime_vmap (struct drm_gem_object *obj) +int qxl_gem_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map) {struct qxl_bo *bo =gem_to_qxl_bo (obj); -void *ptr;int ret; - ret =qxl_bo_kmap (bo, &ptr); + ret =qxl_bo_kmap (bo, map);if (ret <0 ) -return ERR_PTR (ret); +return ret; -return ptr; +return 0 ; } -void qxl_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr) +void qxl_gem_prime_vunmap (struct drm_gem_object *obj, +struct dma_buf_map *map) {struct qxl_bo *bo =gem_to_qxl_bo (obj);
drivers/gpu/drm/radeon/radeon.h +0 -1
@@ -509,7 +509,6 @@ struct radeon_bo {/* Constant after initialization */ struct radeon_device *rdev; -struct ttm_bo_kmap_obj dma_buf_vmap;pid_t pid;#ifdef CONFIG_MMU_NOTIFIER
drivers/gpu/drm/radeon/radeon_gem.c +3 -4
@@ -31,6 +31,7 @@#include <drm/drm_debugfs.h> #include <drm/drm_device.h> #include <drm/drm_file.h> +#include <drm/drm_gem_ttm_helper.h> #include <drm/radeon_drm.h> #include "radeon.h"
@@ -40,8 +41,6 @@ struct dma_buf *radeon_gem_prime_export(struct drm_gem_object *gobj,struct sg_table *radeon_gem_prime_get_sg_table (struct drm_gem_object *obj);int radeon_gem_prime_pin (struct drm_gem_object *obj);void radeon_gem_prime_unpin (struct drm_gem_object *obj); -void *radeon_gem_prime_vmap (struct drm_gem_object *obj); -void radeon_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr);static const struct drm_gem_object_funcs radeon_gem_object_funcs;
@@ -235,8 +234,8 @@ static const struct drm_gem_object_funcs radeon_gem_object_funcs = { .pin = radeon_gem_prime_pin, .unpin = radeon_gem_prime_unpin, .get_sg_table = radeon_gem_prime_get_sg_table, - .vmap = radeon_gem_prime_vmap, - .vunmap = radeon_gem_prime_vunmap, + .vmap = drm_gem_ttm_vmap, + .vunmap = drm_gem_ttm_vunmap, };/*
drivers/gpu/drm/radeon/radeon_prime.c +0 -20
@@ -39,26 +39,6 @@ struct sg_table *radeon_gem_prime_get_sg_table(struct drm_gem_object *obj)return drm_prime_pages_to_sg (obj->dev, bo->tbo.ttm->pages, npages); } -void *radeon_gem_prime_vmap (struct drm_gem_object *obj) -{ -struct radeon_bo *bo =gem_to_radeon_bo (obj); -int ret; - - ret =ttm_bo_kmap (&bo->tbo,0 , bo->tbo.num_pages, - &bo->dma_buf_vmap); -if (ret) -return ERR_PTR (ret); - -return bo->dma_buf_vmap.virtual; -} - -void radeon_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr) -{ -struct radeon_bo *bo =gem_to_radeon_bo (obj); - -ttm_bo_kunmap (&bo->dma_buf_vmap); -} -struct drm_gem_object *radeon_gem_prime_import_sg_table (struct drm_device *dev,struct dma_buf_attachment *attach,struct sg_table *sg)
drivers/gpu/drm/rockchip/rockchip_drm_gem.c +14 -8
@@ -532,26 +532,32 @@ rockchip_gem_prime_import_sg_table(struct drm_device *drm,return ERR_PTR (ret); } -void *rockchip_gem_prime_vmap (struct drm_gem_object *obj) +int rockchip_gem_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map) {struct rockchip_gem_object *rk_obj =to_rockchip_obj (obj); -if (rk_obj->pages) -return vmap (rk_obj->pages, rk_obj->num_pages,VM_MAP , -pgprot_writecombine (PAGE_KERNEL )); +if (rk_obj->pages) { +void *vaddr =vmap (rk_obj->pages, rk_obj->num_pages,VM_MAP , +pgprot_writecombine (PAGE_KERNEL )); +if (!vaddr) +return -ENOMEM ; +dma_buf_map_set_vaddr (map, vaddr); +return 0 ; + }if (rk_obj->dma_attrs &DMA_ATTR_NO_KERNEL_MAPPING ) -return NULL ; +return -ENOMEM ; +dma_buf_map_set_vaddr (map, rk_obj->kvaddr); -return rk_obj->kvaddr; +return 0 ; } -void rockchip_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr) +void rockchip_gem_prime_vunmap (struct drm_gem_object *obj,struct dma_buf_map *map) {struct rockchip_gem_object *rk_obj =to_rockchip_obj (obj);if (rk_obj->pages) { -vunmap (vaddr); +vunmap (map->vaddr);return ; }
drivers/gpu/drm/rockchip/rockchip_drm_gem.h +2 -2
@@ -31,8 +31,8 @@ struct drm_gem_object * rockchip_gem_prime_import_sg_table(struct drm_device *dev, struct dma_buf_attachment *attach, struct sg_table *sg); -void *rockchip_gem_prime_vmap (struct drm_gem_object *obj); -void rockchip_gem_prime_vunmap (struct drm_gem_object *obj,void *vaddr); +int rockchip_gem_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map); +void rockchip_gem_prime_vunmap (struct drm_gem_object *obj,struct dma_buf_map *map);/* drm driver mmap file operations */ int rockchip_gem_mmap (struct file *filp,struct vm_area_struct *vma);
drivers/gpu/drm/tiny/cirrus.c +6 -4
@@ -17,6 +17,7 @@*/ #include <linux/console.h> +#include <linux/dma-buf-map.h> #include <linux/module.h> #include <linux/pci.h>
@@ -314,6 +315,7 @@ static int cirrus_fb_blit_rect(struct drm_framebuffer *fb, struct drm_rect *rect) {struct cirrus_device *cirrus =to_cirrus (fb->dev); +struct dma_buf_map map;void *vmap;int idx, ret;
@@ -321,10 +323,10 @@ static int cirrus_fb_blit_rect(struct drm_framebuffer *fb,if (!drm_dev_enter (&cirrus->dev, &idx)) goto out; - ret = -ENOMEM ; - vmap =drm_gem_shmem_vmap (fb->obj[0 ]); -if (!vmap) + ret =drm_gem_shmem_vmap (fb->obj[0 ], &map); +if (ret) goto out_dev_exit; + vmap = map.vaddr;/* TODO: Use mapping abstraction properly */ if (cirrus->cpp == fb->format->cpp[0 ]) drm_fb_memcpy_dstclip(cirrus->vram,
@@ -343,7 +345,7 @@ static int cirrus_fb_blit_rect(struct drm_framebuffer *fb,else WARN_ON_ONCE ("cpp mismatch" ); -drm_gem_shmem_vunmap (fb->obj[0 ], vmap); +drm_gem_shmem_vunmap (fb->obj[0 ], &map); ret =0 ; out_dev_exit:
drivers/gpu/drm/tiny/gm12u320.c +6 -4
@@ -248,6 +248,7 @@ static void gm12u320_copy_fb_to_blocks(struct gm12u320_device *gm12u320) {int block, dst_offset, len, remain, ret, x1, x2, y1, y2;struct drm_framebuffer *fb; +struct dma_buf_map map;void *vaddr;u8 *src;
@@ -262,11 +263,12 @@ static void gm12u320_copy_fb_to_blocks(struct gm12u320_device *gm12u320) y1 = gm12u320->fb_update.rect.y1; y2 = gm12u320->fb_update.rect.y2; - vaddr =drm_gem_shmem_vmap (fb->obj[0 ]); -if (IS_ERR (vaddr)) { -GM12U320_ERR ("failed to vmap fb: %ld\n" ,PTR_ERR (vaddr)); + ret =drm_gem_shmem_vmap (fb->obj[0 ], &map); +if (ret) { +GM12U320_ERR ("failed to vmap fb: %d\n" , ret); goto put_fb; } + vaddr = map.vaddr;/* TODO: Use mapping abstraction properly */ if (fb->obj[0 ]->import_attach) { ret = dma_buf_begin_cpu_access(
@@ -318,7 +320,7 @@ static void gm12u320_copy_fb_to_blocks(struct gm12u320_device *gm12u320)GM12U320_ERR ("dma_buf_end_cpu_access err: %d\n" , ret); } vunmap: -drm_gem_shmem_vunmap (fb->obj[0 ], vaddr); +drm_gem_shmem_vunmap (fb->obj[0 ], &map); put_fb:drm_framebuffer_put (fb); gm12u320->fb_update.fb =NULL ;
drivers/gpu/drm/udl/udl_modeset.c +5 -3
@@ -276,6 +276,7 @@ static int udl_handle_damage(struct drm_framebuffer *fb, int x, int y,struct urb *urb;struct drm_rect clip;int log_bpp; +struct dma_buf_map map;void *vaddr; ret =udl_log_cpp (fb->format->cpp[0 ]);
@@ -296,11 +297,12 @@ static int udl_handle_damage(struct drm_framebuffer *fb, int x, int y,return ret; } - vaddr =drm_gem_shmem_vmap (fb->obj[0 ]); -if (IS_ERR (vaddr)) { + ret =drm_gem_shmem_vmap (fb->obj[0 ], &map); +if (ret) {DRM_ERROR ("failed to vmap fb\n" ); goto out_dma_buf_end_cpu_access; } + vaddr = map.vaddr;/* TODO: Use mapping abstraction properly */ urb =udl_get_urb (dev);if (!urb)
@@ -333,7 +335,7 @@ static int udl_handle_damage(struct drm_framebuffer *fb, int x, int y, ret =0 ; out_drm_gem_shmem_vunmap: -drm_gem_shmem_vunmap (fb->obj[0 ], vaddr); +drm_gem_shmem_vunmap (fb->obj[0 ], &map); out_dma_buf_end_cpu_access:if (import_attach) { tmp_ret = dma_buf_end_cpu_access(import_attach->dmabuf,
drivers/gpu/drm/vboxvideo/vbox_mode.c +8 -3
@@ -9,6 +9,8 @@* Michael Thayer <[email protected], * Hans de Goede <[email protected]> */ + +#include <linux/dma-buf-map.h> #include <linux/export.h> #include <drm/drm_atomic.h>
@@ -384,6 +386,8 @@ static void vbox_cursor_atomic_update(struct drm_plane *plane,u32 height = plane->state->crtc_h;size_t data_size, mask_size;u32 flags; +struct dma_buf_map map; +int ret;u8 *src;/*
@@ -397,8 +401,8 @@ static void vbox_cursor_atomic_update(struct drm_plane *plane, vbox_crtc->cursor_enabled = true; - src =drm_gem_vram_vmap (gbo); -if (IS_ERR (src)) { + ret =drm_gem_vram_vmap (gbo, &map); +if (ret) {/* * BUG: we should have pinned the BO in prepare_fb(). */
@@ -406,6 +410,7 @@ static void vbox_cursor_atomic_update(struct drm_plane *plane,DRM_WARN ("Could not map cursor bo, skipping update\n" );return ; } + src = map.vaddr;/* TODO: Use mapping abstraction properly */ /* * The mask must be calculated based on the alpha
@@ -416,7 +421,7 @@ static void vbox_cursor_atomic_update(struct drm_plane *plane, data_size = width * height *4 + mask_size;copy_cursor_image (src, vbox->cursor_data, width, height, mask_size); -drm_gem_vram_vunmap (gbo, src); +drm_gem_vram_vunmap (gbo, &map); flags =VBOX_MOUSE_POINTER_VISIBLE |VBOX_MOUSE_POINTER_SHAPE |VBOX_MOUSE_POINTER_ALPHA ;
drivers/gpu/drm/vc4/vc4_bo.c +3 -3
@@ -785,16 +785,16 @@ int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)return drm_gem_cma_prime_mmap (obj, vma); } -void *vc4_prime_vmap (struct drm_gem_object *obj) +int vc4_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map) {struct vc4_bo *bo =to_vc4_bo (obj);if (bo->validated_shader) {DRM_DEBUG ("mmaping of shader BOs not allowed.\n" ); -return ERR_PTR (-EINVAL ); +return -EINVAL ; } -return drm_gem_cma_prime_vmap (obj); +return drm_gem_cma_prime_vmap (obj, map); } struct drm_gem_object *
drivers/gpu/drm/vc4/vc4_drv.h +1 -1
@@ -806,7 +806,7 @@ int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma);struct drm_gem_object *vc4_prime_import_sg_table (struct drm_device *dev,struct dma_buf_attachment *attach,struct sg_table *sgt); -void *vc4_prime_vmap (struct drm_gem_object *obj); +int vc4_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map);int vc4_bo_cache_init (struct drm_device *dev);void vc4_bo_cache_destroy (struct drm_device *dev);int vc4_bo_inc_usecnt (struct vc4_bo *bo);
drivers/gpu/drm/vgem/vgem_drv.c +11 -5
@@ -361,24 +361,30 @@ static struct drm_gem_object *vgem_prime_import_sg_table(struct drm_device *dev,return &obj->base; } -static void *vgem_prime_vmap (struct drm_gem_object *obj) +static int vgem_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map) {struct drm_vgem_gem_object *bo =to_vgem_bo (obj);long n_pages = obj->size >>PAGE_SHIFT ;struct page **pages; +void *vaddr; pages =vgem_pin_pages (bo);if (IS_ERR (pages)) -return NULL ; +return PTR_ERR (pages); + + vaddr =vmap (pages, n_pages,0 ,pgprot_writecombine (PAGE_KERNEL )); +if (!vaddr) +return -ENOMEM ; +dma_buf_map_set_vaddr (map, vaddr); -return vmap (pages, n_pages,0 ,pgprot_writecombine (PAGE_KERNEL )); +return 0 ; } -static void vgem_prime_vunmap (struct drm_gem_object *obj,void *vaddr) +static void vgem_prime_vunmap (struct drm_gem_object *obj,struct dma_buf_map *map) {struct drm_vgem_gem_object *bo =to_vgem_bo (obj); -vunmap (vaddr); +vunmap (map->vaddr);vgem_unpin_pages (bo); }
drivers/gpu/drm/vkms/vkms_plane.c +10 -5
@@ -1,5 +1,7 @@// SPDX-License-Identifier: GPL-2.0+ +#include <linux/dma-buf-map.h> +#include <drm/drm_atomic.h> #include <drm/drm_atomic_helper.h> #include <drm/drm_fourcc.h>
@@ -146,15 +148,16 @@ static int vkms_prepare_fb(struct drm_plane *plane, struct drm_plane_state *state) {struct drm_gem_object *gem_obj; -void *vaddr; +struct dma_buf_map map; +int ret;if (!state->fb)return 0 ; gem_obj =drm_gem_fb_get_obj (state->fb,0 ); - vaddr =drm_gem_shmem_vmap (gem_obj); -if (IS_ERR (vaddr)) -DRM_ERROR ("vmap failed: %li\n" ,PTR_ERR (vaddr)); + ret =drm_gem_shmem_vmap (gem_obj, &map); +if (ret) +DRM_ERROR ("vmap failed: %d\n" , ret);return drm_gem_fb_prepare_fb (plane, state); }
@@ -164,13 +167,15 @@ static void vkms_cleanup_fb(struct drm_plane *plane, {struct drm_gem_object *gem_obj;struct drm_gem_shmem_object *shmem_obj; +struct dma_buf_map map;if (!old_state->fb)return ; gem_obj =drm_gem_fb_get_obj (old_state->fb,0 ); shmem_obj =to_drm_gem_shmem_obj (drm_gem_fb_get_obj (old_state->fb,0 )); -drm_gem_shmem_vunmap (gem_obj, shmem_obj->vaddr); +dma_buf_map_set_vaddr (&map, shmem_obj->vaddr); +drm_gem_shmem_vunmap (gem_obj, &map); }static const struct drm_plane_helper_funcs vkms_primary_helper_funcs = {
drivers/gpu/drm/vkms/vkms_writeback.c +14 -8
@@ -1,6 +1,7 @@// SPDX-License-Identifier: GPL-2.0+ -#include "vkms_drv.h" +#include <linux/dma-buf-map.h> +#include <drm/drm_fourcc.h> #include <drm/drm_writeback.h> #include <drm/drm_probe_helper.h>
@@ -8,6 +9,8 @@#include <drm/drm_gem_framebuffer_helper.h> #include <drm/drm_gem_shmem_helper.h> +#include "vkms_drv.h" +static const u32 vkms_wb_formats[] = {DRM_FORMAT_XRGB8888 , };
@@ -65,19 +68,20 @@ static int vkms_wb_prepare_job(struct drm_writeback_connector *wb_connector, struct drm_writeback_job *job) {struct drm_gem_object *gem_obj; -void *vaddr; +struct dma_buf_map map; +int ret;if (!job->fb)return 0 ; gem_obj =drm_gem_fb_get_obj (job->fb,0 ); - vaddr =drm_gem_shmem_vmap (gem_obj); -if (IS_ERR (vaddr)) { -DRM_ERROR ("vmap failed: %li\n" ,PTR_ERR (vaddr)); -return PTR_ERR (vaddr); + ret =drm_gem_shmem_vmap (gem_obj, &map); +if (ret) { +DRM_ERROR ("vmap failed: %d\n" , ret); +return ret; } - job->priv = vaddr; + job->priv = map.vaddr;return 0 ; }
@@ -87,12 +91,14 @@ static void vkms_wb_cleanup_job(struct drm_writeback_connector *connector, {struct drm_gem_object *gem_obj;struct vkms_device *vkmsdev; +struct dma_buf_map map;if (!job->fb)return ; gem_obj =drm_gem_fb_get_obj (job->fb,0 ); -drm_gem_shmem_vunmap (gem_obj, job->priv); +dma_buf_map_set_vaddr (&map, job->priv); +drm_gem_shmem_vunmap (gem_obj, &map); vkmsdev =drm_device_to_vkms_device (gem_obj->dev);vkms_set_composer (&vkmsdev->output, false);
drivers/gpu/drm/xen/xen_drm_front_gem.c +12 -6
@@ -290,22 +290,28 @@ int xen_drm_front_gem_mmap(struct file *filp, struct vm_area_struct *vma)return gem_mmap_obj (xen_obj, vma); } -void *xen_drm_front_gem_prime_vmap (struct drm_gem_object *gem_obj) +int xen_drm_front_gem_prime_vmap (struct drm_gem_object *gem_obj,struct dma_buf_map *map) {struct xen_gem_object *xen_obj =to_xen_gem_obj (gem_obj); +void *vaddr;if (!xen_obj->pages) -return NULL ; +return -ENOMEM ;/* Please see comment in gem_mmap_obj on mapping and attributes. */ -return vmap (xen_obj->pages, xen_obj->num_pages, -VM_MAP ,PAGE_KERNEL ); + vaddr =vmap (xen_obj->pages, xen_obj->num_pages, +VM_MAP ,PAGE_KERNEL ); +if (!vaddr) +return -ENOMEM ; +dma_buf_map_set_vaddr (map, vaddr); + +return 0 ; }void xen_drm_front_gem_prime_vunmap (struct drm_gem_object *gem_obj, -void *vaddr) +struct dma_buf_map *map) { -vunmap (vaddr); +vunmap (map->vaddr); } int xen_drm_front_gem_prime_mmap(struct drm_gem_object *gem_obj,
drivers/gpu/drm/xen/xen_drm_front_gem.h +4 -2
@@ -12,6 +12,7 @@#define __XEN_DRM_FRONT_GEM_Hstruct dma_buf_attachment ; +struct dma_buf_map ;struct drm_device ;struct drm_gem_object ;struct file ;
@@ -34,10 +35,11 @@ void xen_drm_front_gem_free_object_unlocked(struct drm_gem_object *gem_obj);int xen_drm_front_gem_mmap (struct file *filp,struct vm_area_struct *vma); -void *xen_drm_front_gem_prime_vmap (struct drm_gem_object *gem_obj); +int xen_drm_front_gem_prime_vmap (struct drm_gem_object *gem_obj, +struct dma_buf_map *map);void xen_drm_front_gem_prime_vunmap (struct drm_gem_object *gem_obj, -void *vaddr); +struct dma_buf_map *map);int xen_drm_front_gem_prime_mmap (struct drm_gem_object *gem_obj,struct vm_area_struct *vma);
include/drm/drm_gem.h +3 -2
@@ -39,6 +39,7 @@#include <drm/drm_vma_manager.h> +struct dma_buf_map ;struct drm_gem_object ;/**
@@ -138,7 +139,7 @@ struct drm_gem_object_funcs {* * This callback is optional. */ -void *(*vmap)(struct drm_gem_object *obj); +int (*vmap)(struct drm_gem_object *obj,struct dma_buf_map *map);/** * @vunmap:
@@ -148,7 +149,7 @@ struct drm_gem_object_funcs {* * This callback is optional. */ -void (*vunmap)(struct drm_gem_object *obj,void *vaddr); +void (*vunmap)(struct drm_gem_object *obj,struct dma_buf_map *map);/** * @mmap:
include/drm/drm_gem_cma_helper.h +1 -1
@@ -103,7 +103,7 @@ drm_gem_cma_prime_import_sg_table(struct drm_device *dev, struct sg_table *sgt);int drm_gem_cma_prime_mmap (struct drm_gem_object *obj,struct vm_area_struct *vma); -void *drm_gem_cma_prime_vmap (struct drm_gem_object *obj); +int drm_gem_cma_prime_vmap (struct drm_gem_object *obj,struct dma_buf_map *map);struct drm_gem_object *drm_gem_cma_create_object_default_funcs (struct drm_device *dev,size_t size);
include/drm/drm_gem_shmem_helper.h +2 -2
@@ -113,8 +113,8 @@ int drm_gem_shmem_get_pages(struct drm_gem_shmem_object *shmem);void drm_gem_shmem_put_pages (struct drm_gem_shmem_object *shmem);int drm_gem_shmem_pin (struct drm_gem_object *obj);void drm_gem_shmem_unpin (struct drm_gem_object *obj); -void *drm_gem_shmem_vmap (struct drm_gem_object *obj); -void drm_gem_shmem_vunmap (struct drm_gem_object *obj,void *vaddr); +int drm_gem_shmem_vmap (struct drm_gem_object *obj,struct dma_buf_map *map); +void drm_gem_shmem_vunmap (struct drm_gem_object *obj,struct dma_buf_map *map);int drm_gem_shmem_madvise (struct drm_gem_object *obj,int madv);
include/drm/drm_gem_vram_helper.h +7 -7
@@ -10,6 +10,7 @@#include <drm/ttm/ttm_bo_api.h> #include <drm/ttm/ttm_bo_driver.h> +#include <linux/dma-buf-map.h> #include <linux/kernel.h> /* for container_of() */ struct drm_mode_create_dumb ;
@@ -29,9 +30,8 @@ struct vm_area_struct;/** * struct drm_gem_vram_object - GEM object backed by VRAM -* @gem: GEM object * @bo: TTM buffer object -* @kmap: Mapping information for @bo +* @map: Mapping information for @bo * @placement: TTM placement information. Supported placements are \ %TTM_PL_VRAM and %TTM_PL_SYSTEM * @placements: TTM placement information.
@@ -50,15 +50,15 @@ struct vm_area_struct;*/ struct drm_gem_vram_object {struct ttm_buffer_object bo; -struct ttm_bo_kmap_obj kmap; +struct dma_buf_map map;/** -* @kmap_use_count: +* @vmap_use_count: * * Reference count on the virtual address. * The address are un-mapped when the count reaches zero. */ -unsigned int kmap_use_count; +unsigned int vmap_use_count;/* Supported placements are %TTM_PL_VRAM and %TTM_PL_SYSTEM */ struct ttm_placement placement;
@@ -97,8 +97,8 @@ u64 drm_gem_vram_mmap_offset(struct drm_gem_vram_object *gbo);s64 drm_gem_vram_offset (struct drm_gem_vram_object *gbo);int drm_gem_vram_pin (struct drm_gem_vram_object *gbo,unsigned long pl_flag);int drm_gem_vram_unpin (struct drm_gem_vram_object *gbo); -void *drm_gem_vram_vmap (struct drm_gem_vram_object *gbo); -void drm_gem_vram_vunmap (struct drm_gem_vram_object *gbo,void *vaddr); +int drm_gem_vram_vmap (struct drm_gem_vram_object *gbo,struct dma_buf_map *map); +void drm_gem_vram_vunmap (struct drm_gem_vram_object *gbo,struct dma_buf_map *map); int drm_gem_vram_fill_create_dumb(struct drm_file *file, struct drm_device *dev,
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 7/10 drm/gem: Update internal GEM vmap/vunmap interfaces to use struct dma_buf_map
In reply to Thomas ZimmermannGEM's vmap and vunmap interfaces now wrap memory pointers in struct dma_buf_map.
--- drivers/gpu/drm/drm_client.c | 18 +++++++++++------- drivers/gpu/drm/drm_gem.c | 26 +++++++++++++------------- drivers/gpu/drm/drm_internal.h | 5 +++-- drivers/gpu/drm/drm_prime.c | 14 ++++---------- 4 files changed, 31 insertions(+), 32 deletions(-)
drivers/gpu/drm/drm_client.c +11 -7
@@ -3,6 +3,7 @@* Copyright 2018 Noralf Trønnes */ +#include <linux/dma-buf-map.h> #include <linux/list.h> #include <linux/module.h> #include <linux/mutex.h>
@@ -304,7 +305,8 @@ drm_client_buffer_create(struct drm_client_dev *client, u32 width, u32 height, u*/ void *drm_client_buffer_vmap (struct drm_client_buffer *buffer) { -void *vaddr; +struct dma_buf_map map; +int ret;if (buffer->vaddr)return buffer->vaddr;
@@ -317,13 +319,13 @@ void *drm_client_buffer_vmap(struct drm_client_buffer *buffer)* fd_install step out of the driver backend hooks, to make that * final step optional for internal users. */ - vaddr =drm_gem_vmap (buffer->gem); -if (IS_ERR (vaddr)) -return vaddr; + ret =drm_gem_vmap (buffer->gem, &map); +if (ret) +return ERR_PTR (ret); - buffer->vaddr = vaddr; + buffer->vaddr = map.vaddr; -return vaddr; +return map.vaddr; }EXPORT_SYMBOL (drm_client_buffer_vmap);
@@ -337,7 +339,9 @@ EXPORT_SYMBOL(drm_client_buffer_vmap);*/ void drm_client_buffer_vunmap (struct drm_client_buffer *buffer) { -drm_gem_vunmap (buffer->gem, buffer->vaddr); +struct dma_buf_map map =DMA_BUF_MAP_INIT_VADDR (buffer->vaddr); + +drm_gem_vunmap (buffer->gem, &map); buffer->vaddr =NULL ; }EXPORT_SYMBOL (drm_client_buffer_vunmap);
drivers/gpu/drm/drm_gem.c +13 -13
@@ -1206,32 +1206,32 @@ void drm_gem_unpin(struct drm_gem_object *obj) obj->funcs->unpin (obj); } -void *drm_gem_vmap (struct drm_gem_object *obj) +int drm_gem_vmap (struct drm_gem_object *obj,struct dma_buf_map *map) { -struct dma_buf_map map;int ret;if (!obj->funcs->vmap) -return ERR_PTR (-EOPNOTSUPP ); +return -EOPNOTSUPP ; - ret = obj->funcs->vmap (obj, &map); + ret = obj->funcs->vmap (obj, map);if (ret) -return ERR_PTR (ret); -else if (dma_buf_map_is_null (&map)) -return ERR_PTR (-ENOMEM ); +return ret; +else if (dma_buf_map_is_null (map)) +return -ENOMEM ; -return map.vaddr; +return 0 ; } -void drm_gem_vunmap (struct drm_gem_object *obj,void *vaddr) +void drm_gem_vunmap (struct drm_gem_object *obj,struct dma_buf_map *map) { -struct dma_buf_map map =DMA_BUF_MAP_INIT_VADDR (vaddr); - -if (!vaddr) +if (dma_buf_map_is_null (map))return ;if (obj->funcs->vunmap) - obj->funcs->vunmap (obj, &map); + obj->funcs->vunmap (obj, map); + +/* Always set the mapping to NULL. Callers may rely on this. */ +dma_buf_map_clear (map); }/**
drivers/gpu/drm/drm_internal.h +3 -2
@@ -33,6 +33,7 @@struct dentry ;struct dma_buf ; +struct dma_buf_map ;struct drm_connector ;struct drm_crtc ;struct drm_framebuffer ;
@@ -187,8 +188,8 @@ void drm_gem_print_info(struct drm_printer *p, unsigned int indent,int drm_gem_pin (struct drm_gem_object *obj);void drm_gem_unpin (struct drm_gem_object *obj); -void *drm_gem_vmap (struct drm_gem_object *obj); -void drm_gem_vunmap (struct drm_gem_object *obj,void *vaddr); +int drm_gem_vmap (struct drm_gem_object *obj,struct dma_buf_map *map); +void drm_gem_vunmap (struct drm_gem_object *obj,struct dma_buf_map *map);/* drm_debugfs.c drm_debugfs_crc.c */ #if defined(CONFIG_DEBUG_FS )
drivers/gpu/drm/drm_prime.c +4 -10
@@ -667,21 +667,15 @@ EXPORT_SYMBOL(drm_gem_unmap_dma_buf);* * Sets up a kernel virtual mapping. This can be used as the &dma_buf_ops.vmap * callback. Calls into &drm_gem_object_funcs.vmap for device specific handling. +* The kernel virtual address is returned in map. * -* Returns the kernel virtual address or NULL on failure. +* Returns 0 on success or a negative errno code otherwise. */ int drm_gem_dmabuf_vmap (struct dma_buf *dma_buf,struct dma_buf_map *map) {struct drm_gem_object *obj = dma_buf->priv; -void *vaddr; - vaddr =drm_gem_vmap (obj); -if (IS_ERR (vaddr)) -return PTR_ERR (vaddr); - -dma_buf_map_set_vaddr (map, vaddr); - -return 0 ; +return drm_gem_vmap (obj, map); }EXPORT_SYMBOL (drm_gem_dmabuf_vmap);
@@ -697,7 +691,7 @@ void drm_gem_dmabuf_vunmap(struct dma_buf *dma_buf, struct dma_buf_map *map) {struct drm_gem_object *obj = dma_buf->priv; -drm_gem_vunmap (obj, map->vaddr); +drm_gem_vunmap (obj, map); }EXPORT_SYMBOL (drm_gem_dmabuf_vunmap);
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 8/10 drm/gem: Store client buffer mappings as struct dma_buf_map
In reply to Thomas ZimmermannKernel DRM clients now store their framebuffer address in an instance of struct dma_buf_map. Depending on the buffer's location, the address refers to system or I/O memory.
Callers of drm_client_buffer_vmap() receive a copy of the value in the call's supplied arguments. It can be accessed and modified with dma_buf_map interfaces.
--- drivers/gpu/drm/drm_client.c | 34 +++++++++++++++++++-------------- drivers/gpu/drm/drm_fb_helper.c | 23 +++++++++++++--------- include/drm/drm_client.h | 7 ++++--- 3 files changed, 38 insertions(+), 26 deletions(-)
drivers/gpu/drm/drm_client.c +20 -14
@@ -235,7 +235,7 @@ static void drm_client_buffer_delete(struct drm_client_buffer *buffer) {struct drm_device *dev = buffer->client->dev; -drm_gem_vunmap (buffer->gem, buffer->vaddr); +drm_gem_vunmap (buffer->gem, &buffer->map);if (buffer->gem)drm_gem_object_put (buffer->gem);
@@ -291,25 +291,31 @@ drm_client_buffer_create(struct drm_client_dev *client, u32 width, u32 height, u/** * drm_client_buffer_vmap - Map DRM client buffer into address space * @buffer: DRM client buffer +* @map_copy: Returns the mapped memory's address * * This function maps a client buffer into kernel address space. If the -* buffer is already mapped, it returns the mapping's address. +* buffer is already mapped, it returns the existing mapping's address. * * Client buffer mappings are not ref'counted. Each call to * drm_client_buffer_vmap() should be followed by a call to * drm_client_buffer_vunmap(); or the client buffer should be mapped * throughout its lifetime. * +* The returned address is a copy of the internal value. In contrast to +* other vmap interfaces, you don't need it for the client's vunmap +* function. So you can modify it at will during blit and draw operations. +* * Returns: -* The mapped memory's address +* 0 on success, or a negative errno code otherwise. */ -void *drm_client_buffer_vmap (struct drm_client_buffer *buffer) +int +drm_client_buffer_vmap (struct drm_client_buffer *buffer,struct dma_buf_map *map_copy) { -struct dma_buf_map map; +struct dma_buf_map *map = &buffer->map;int ret; -if (buffer->vaddr) -return buffer->vaddr; +if (dma_buf_map_is_set (map)) + goto out;/* * FIXME: The dependency on GEM here isn't required, we could
@@ -319,13 +325,14 @@ void *drm_client_buffer_vmap(struct drm_client_buffer *buffer)* fd_install step out of the driver backend hooks, to make that * final step optional for internal users. */ - ret =drm_gem_vmap (buffer->gem, &map); + ret =drm_gem_vmap (buffer->gem, map);if (ret) -return ERR_PTR (ret); +return ret; - buffer->vaddr = map.vaddr; +out: + *map_copy = *map; -return map.vaddr; +return 0 ; }EXPORT_SYMBOL (drm_client_buffer_vmap);
@@ -339,10 +346,9 @@ EXPORT_SYMBOL(drm_client_buffer_vmap);*/ void drm_client_buffer_vunmap (struct drm_client_buffer *buffer) { -struct dma_buf_map map =DMA_BUF_MAP_INIT_VADDR (buffer->vaddr); +struct dma_buf_map *map = &buffer->map; -drm_gem_vunmap (buffer->gem, &map); - buffer->vaddr =NULL ; +drm_gem_vunmap (buffer->gem, map); }EXPORT_SYMBOL (drm_client_buffer_vunmap);
drivers/gpu/drm/drm_fb_helper.c +14 -9
@@ -378,7 +378,7 @@ static void drm_fb_helper_dirty_blit_real(struct drm_fb_helper *fb_helper,unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->vaddr + offset; +void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y;
@@ -400,7 +400,8 @@ static void drm_fb_helper_dirty_work(struct work_struct *work)struct drm_clip_rect *clip = &helper->dirty_clip;struct drm_clip_rect clip_copy;unsigned long flags; -void *vaddr; +struct dma_buf_map map; +int ret;spin_lock_irqsave (&helper->dirty_lock, flags); clip_copy = *clip;
@@ -413,8 +414,8 @@ static void drm_fb_helper_dirty_work(struct work_struct *work)/* Generic fbdev uses a shadow buffer */ if (helper->buffer) { - vaddr =drm_client_buffer_vmap (helper->buffer); -if (IS_ERR (vaddr)) + ret =drm_client_buffer_vmap (helper->buffer, &map); +if (ret)return ;drm_fb_helper_dirty_blit_real (helper, &clip_copy); }
@@ -2060,7 +2061,8 @@ static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper,struct drm_framebuffer *fb;struct fb_info *fbi;u32 format; -void *vaddr; +struct dma_buf_map map; +int ret; drm_dbg_kms(dev, "surface width(%d), height(%d) and bpp(%d)\n", sizes->surface_width, sizes->surface_height,
@@ -2096,11 +2098,14 @@ static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper,fb_deferred_io_init (fbi); }else {/* buffer is mapped for HW framebuffer */ - vaddr =drm_client_buffer_vmap (fb_helper->buffer); -if (IS_ERR (vaddr)) -return PTR_ERR (vaddr); + ret =drm_client_buffer_vmap (fb_helper->buffer, &map); +if (ret) +return ret; +if (map.is_iomem) + fbi->screen_base = map.vaddr_iomem; +else + fbi->screen_buffer = map.vaddr; - fbi->screen_buffer = vaddr;/* Shamelessly leak the physical address to user-space */ #if IS_ENABLED (CONFIG_DRM_FBDEV_LEAK_PHYS_SMEM )if (drm_leak_fbdev_smem && fbi->fix.smem_start ==0 )
include/drm/drm_client.h +4 -3
@@ -3,6 +3,7 @@#ifndef _DRM_CLIENT_H_#define _DRM_CLIENT_H_ +#include <linux/dma-buf-map.h> #include <linux/lockdep.h> #include <linux/mutex.h> #include <linux/types.h>
@@ -141,9 +142,9 @@ struct drm_client_buffer {struct drm_gem_object *gem;/** -* @vaddr: Virtual address for the buffer +* @map: Virtual address for the buffer */ -void *vaddr; +struct dma_buf_map map;/** * @fb: DRM framebuffer
@@ -155,7 +156,7 @@ struct drm_client_buffer * drm_client_framebuffer_create(struct drm_client_dev *client, u32 width, u32 height, u32 format);void drm_client_framebuffer_delete (struct drm_client_buffer *buffer);int drm_client_framebuffer_flush (struct drm_client_buffer *buffer,struct drm_rect *rect); -void *drm_client_buffer_vmap (struct drm_client_buffer *buffer); +int drm_client_buffer_vmap (struct drm_client_buffer *buffer,struct dma_buf_map *map);void drm_client_buffer_vunmap (struct drm_client_buffer *buffer);int drm_client_modeset_create (struct drm_client_dev *client);
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 9/10 dma-buf-map: Add memcpy and pointer-increment interfaces
In reply to Thomas ZimmermannTo do framebuffer updates, one needs memcpy from system memory and a pointer-increment function. Add both interfaces with documentation.
v5:
* include <linux/string.h> to build on sparc64 (Sam)
--- include/linux/dma-buf-map.h | 73 ++++++++++++++++++++++++++++++++----- 1 file changed, 63 insertions(+), 10 deletions(-)
include/linux/dma-buf-map.h +63 -10
@@ -7,6 +7,7 @@#define __DMA_BUF_MAP_H__#include <linux/io.h> +#include <linux/string.h> /** * DOC: overview
@@ -32,6 +33,14 @@* accessing the buffer. Use the returned instance and the helper functions * to access the buffer's memory in the correct way. * +* The type :c:type:`struct dma_buf_map <dma_buf_map>` and its helpers are +* actually independent from the dma-buf infrastructure. When sharing buffers +* among devices, drivers have to know the location of the memory to access +* the buffers in a safe way. :c:type:`struct dma_buf_map <dma_buf_map>` +* solves this problem for dma-buf and its users. If other drivers or +* sub-systems require similar functionality, the type could be generalized +* and moved to a more prominent header file. +* * Open-coding access to :c:type:`struct dma_buf_map <dma_buf_map>` is * considered bad style. Rather then accessing its fields directly, use one * of the provided helper functions, or implement your own. For example,
@@ -51,6 +60,14 @@* * dma_buf_map_set_vaddr_iomem(&map. 0xdeadbeaf); * +* Instances of struct dma_buf_map do not have to be cleaned up, but +* can be cleared to NULL with dma_buf_map_clear(). Cleared mappings +* always refer to system memory. +* +* .. code-block:: c +* +* dma_buf_map_clear(&map); +* * Test if a mapping is valid with either dma_buf_map_is_set() or * dma_buf_map_is_null(). *
@@ -73,17 +90,19 @@* if (dma_buf_map_is_equal(&sys_map, &io_map)) * // always false * -* Instances of struct dma_buf_map do not have to be cleaned up, but -* can be cleared to NULL with dma_buf_map_clear(). Cleared mappings -* always refer to system memory. +* A set up instance of struct dma_buf_map can be used to access or manipulate +* the buffer memory. Depending on the location of the memory, the provided +* helpers will pick the correct operations. Data can be copied into the memory +* with dma_buf_map_memcpy_to(). The address can be manipulated with +* dma_buf_map_incr(). * -* The type :c:type:`struct dma_buf_map <dma_buf_map>` and its helpers are -* actually independent from the dma-buf infrastructure. When sharing buffers -* among devices, drivers have to know the location of the memory to access -* the buffers in a safe way. :c:type:`struct dma_buf_map <dma_buf_map>` -* solves this problem for dma-buf and its users. If other drivers or -* sub-systems require similar functionality, the type could be generalized -* and moved to a more prominent header file. +* .. code-block:: c +* +* const void *src = ...; // source buffer +* size_t len = ...; // length of src +* +* dma_buf_map_memcpy_to(&map, src, len); +* dma_buf_map_incr(&map, len); // go to first byte after the memcpy */ /**
@@ -210,4 +229,38 @@ static inline void dma_buf_map_clear(struct dma_buf_map *map) } } +/** +* dma_buf_map_memcpy_to - Memcpy into dma-buf mapping +* @dst: The dma-buf mapping structure +* @src: The source buffer +* @len: The number of byte in src +* +* Copies data into a dma-buf mapping. The source buffer is in system +* memory. Depending on the buffer's location, the helper picks the correct +* method of accessing the memory. +*/ +static inline void dma_buf_map_memcpy_to (struct dma_buf_map *dst,const void *src,size_t len) +{ +if (dst->is_iomem) +memcpy_toio (dst->vaddr_iomem, src, len); +else +memcpy (dst->vaddr, src, len); +} + +/** +* dma_buf_map_incr - Increments the address stored in a dma-buf mapping +* @map: The dma-buf mapping structure +* @incr: The number of bytes to increment +* +* Increments the address stored in a dma-buf mapping. Depending on the +* buffer's location, the correct value will be updated. +*/ +static inline void dma_buf_map_incr (struct dma_buf_map *map,size_t incr) +{ +if (map->is_iomem) + map->vaddr_iomem += incr; +else + map->vaddr += incr; +} +#endif /* __DMA_BUF_MAP_H__ */
Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerPATCH v5 10/10 drm/fb_helper: Support framebuffers in I/O memory
In reply to Thomas ZimmermannAt least sparc64 requires I/O-specific access to framebuffers. This patch updates the fbdev console accordingly.
For drivers with direct access to the framebuffer memory, the callback functions in struct fb_ops test for the type of memory and call the rsp fb_sys_ of fb_cfb_ functions. Read and write operations are implemented internally by DRM's fbdev helper.
For drivers that employ a shadow buffer, fbdev's blit function retrieves the framebuffer address as struct dma_buf_map, and uses dma_buf_map interfaces to access the buffer.
The bochs driver on sparc64 uses a workaround to flag the framebuffer as I/O memory and avoid a HW exception. With the introduction of struct dma_buf_map, this is not required any longer. The patch removes the rsp code from both, bochs and fbdev.
v5:
* implement fb_read/fb_write internally (Daniel, Sam)
v4:
* move dma_buf_map changes into separate patch (Daniel)
* TODO list: comment on fbdev updates (Daniel)
--- Documentation/gpu/todo.rst | 19 ++- drivers/gpu/drm/bochs/bochs_kms.c | 1 - drivers/gpu/drm/drm_fb_helper.c | 227 ++++++++++++++++++++++++++++-- include/drm/drm_mode_config.h | 12 -- 4 files changed, 230 insertions(+), 29 deletions(-)
Documentation/gpu/todo.rst +17 -2
@@ -197,13 +197,28 @@ Convert drivers to use drm_fbdev_generic_setup() ------------------------------------------------ Most drivers can use drm_fbdev_generic_setup(). Driver have to implement -atomic modesetting and GEM vmap support. Current generic fbdev emulation -expects the framebuffer in system memory (or system-like memory). +atomic modesetting and GEM vmap support. Historically, generic fbdev emulation +expected the framebuffer in system memory or system-like memory. By employing +struct dma_buf_map, drivers with frambuffers in I/O memory can be supported +as well. Contact: Maintainer of the driver you plan to convert Level: Intermediate +Reimplement functions in drm_fbdev_fb_ops without fbdev +------------------------------------------------------- + +A number of callback functions in drm_fbdev_fb_ops could benefit from +being rewritten without dependencies on the fbdev module. Some of the +helpers could further benefit from using struct dma_buf_map instead of +raw pointers. + +Contact: Thomas Zimmermann <[email protected]>, Daniel Vetter + +Level: Advanced + + drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup -----------------------------------------------------------------
drivers/gpu/drm/bochs/bochs_kms.c +0 -1
@@ -151,7 +151,6 @@ int bochs_kms_init(struct bochs_device *bochs) bochs->dev->mode_config.preferred_depth =24 ; bochs->dev->mode_config.prefer_shadow =0 ; bochs->dev->mode_config.prefer_shadow_fbdev =1 ; - bochs->dev->mode_config.fbdev_use_iomem = true; bochs->dev->mode_config.quirk_addfb_prefer_host_byte_order = true; bochs->dev->mode_config.funcs = &bochs_mode_funcs;
drivers/gpu/drm/drm_fb_helper.c +213 -14
262 lines not on this page · raw message
include/drm/drm_mode_config.h +0 -12
@@ -877,18 +877,6 @@ struct drm_mode_config {*/ bool prefer_shadow_fbdev; -/** -* @fbdev_use_iomem: -* -* Set to true if framebuffer reside in iomem. -* When set to true memcpy_toio() is used when copying the framebuffer in -* drm_fb_helper.drm_fb_helper_dirty_blit_real(). -* -* FIXME: This should be replaced with a per-mapping is_iomem -* flag (like ttm does), and then used everywhere in fbdev code. -*/ -bool fbdev_use_iomem; -/** * @quirk_addfb_prefer_xbgr_30bpp: *
Signature · 1 line
2.28.0
Re: [PATCH v5 05/10] drm/ttm: Add vmap/vunmap to TTM and TTM GEM helpers
In reply to Thomas ZimmermannAm 20.10.20 um 14:20 schrieb Thomas Zimmermann:
Quoted 19 lines
The new functions ttm_bo_{vmap,vunmap}() map and unmap a TTM BO in kernel address space. The mapping's address is returned as struct
dma_buf_map. Each function is a simplified version of TTM's existing kmap code. Both functions respect the memory's location ani/or writecombine flags.On top TTM's functions, GEM TTM helpers got drm_gem_ttm_{vmap,vunmap}(), two helpers that convert a GEM object into the TTM BO and forward the call to TTM's vmap/vunmap. These helpers can be dropped into the rsp GEM object callbacks.
v5:
* usesize_tfor storing mapping size (Christian)
* ignore premapped memory areas correctly inttm_bo_vunmap()
* rebase onto latest TTM interfaces (Christian)
* removeBUG()fromttm_bo_vmap()(Christian)
v4:
* dropttm_kmap_obj_to_dma_buf()in favor of vmap helpers (Daniel, Christian)
Quoted 3 lines
Signed-off-by Thomas Zimmermann <[email protected]>Acked-by Daniel Vetter <[email protected]>Tested-by Sam Ravnborg <[email protected]>
Quoted 253 lines
--- drivers/gpu/drm/drm_gem_ttm_helper.c | 38 +++++++++++++++ drivers/gpu/drm/ttm/ttm_bo_util.c | 72 ++++++++++++++++++++++++++++ include/drm/drm_gem_ttm_helper.h | 6 +++ include/drm/ttm/ttm_bo_api.h | 28 +++++++++++ include/linux/dma-buf-map.h | 20 ++++++++ 5 files changed, 164 insertions(+)drivers/gpu/drm/drm_gem_ttm_helper.c +38
-0@@ -49,6 +49,44 @@ void drm_gem_ttm_print_info(struct drm_printer *p, unsigned int indent, }EXPORT_SYMBOL (drm_gem_ttm_print_info); +/** +* drm_gem_ttm_vmap() - vmap &ttm_buffer_object +* @gem: GEM object. +* @map: [out] returns the dma-buf mapping. +* +* Maps a GEM object with ttm_bo_vmap(). This function can be used as +* &drm_gem_object_funcs.vmap callback. +* +* Returns: +* 0 on success, or a negative errno code otherwise. +*/ +int drm_gem_ttm_vmap (struct drm_gem_object *gem, +struct dma_buf_map *map) +{ +struct ttm_buffer_object *bo =drm_gem_ttm_of_gem (gem); + +return ttm_bo_vmap (bo, map); + +} +EXPORT_SYMBOL (drm_gem_ttm_vmap); + +/** +* drm_gem_ttm_vunmap() - vunmap &ttm_buffer_object +* @gem: GEM object. +* @map: dma-buf mapping. +* +* Unmaps a GEM object with ttm_bo_vunmap(). This function can be used as +* &drm_gem_object_funcs.vmap callback. +*/ +void drm_gem_ttm_vunmap (struct drm_gem_object *gem, +struct dma_buf_map *map) +{ +struct ttm_buffer_object *bo =drm_gem_ttm_of_gem (gem); + +ttm_bo_vunmap (bo, map); +} +EXPORT_SYMBOL (drm_gem_ttm_vunmap); +/** * drm_gem_ttm_mmap() - mmap &ttm_buffer_object * @gem: GEM object. drivers/gpu/drm/ttm/ttm_bo_util.c +72
-0@@ -32,6 +32,7 @@#include <drm/ttm/ttm_bo_driver.h> #include <drm/ttm/ttm_placement.h> #include <drm/drm_vma_manager.h> +#include <linux/dma-buf-map.h> #include <linux/io.h> #include <linux/highmem.h> #include <linux/wait.h> @@ -527,6 +528,77 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map) }EXPORT_SYMBOL (ttm_bo_kunmap); +int ttm_bo_vmap (struct ttm_buffer_object *bo,struct dma_buf_map *map) +{ +struct ttm_resource *mem = &bo->mem; +int ret; + + ret =ttm_mem_io_reserve (bo->bdev, mem); +if (ret) +return ret; + +if (mem->bus.is_iomem) { +void __iomem *vaddr_iomem; +size_t size = bo->num_pages <<PAGE_SHIFT ; + +if (mem->bus.addr) + vaddr_iomem = (void __iomem *)mem->bus.addr; +else if (mem->bus.caching == ttm_write_combined) + vaddr_iomem =ioremap_wc (mem->bus.offset, size); +else + vaddr_iomem =ioremap (mem->bus.offset, size); + +if (!vaddr_iomem) +return -ENOMEM ; + +dma_buf_map_set_vaddr_iomem (map, vaddr_iomem); + + }else { +struct ttm_operation_ctx ctx = { + .interruptible = false, + .no_wait_gpu = false + }; +struct ttm_tt *ttm = bo->ttm; +pgprot_t prot; +void *vaddr; + + ret =ttm_tt_populate (bo->bdev, ttm, &ctx); +if (ret) +return ret; + +/* +* We need to use vmap to get the desired page protection +* or to make the buffer object look contiguous. +*/ + prot =ttm_io_prot (bo, mem,PAGE_KERNEL ); + vaddr =vmap (ttm->pages, bo->num_pages,0 , prot); +if (!vaddr) +return -ENOMEM ; + +dma_buf_map_set_vaddr (map, vaddr); + } + +return 0 ; +} +EXPORT_SYMBOL (ttm_bo_vmap); + +void ttm_bo_vunmap (struct ttm_buffer_object *bo,struct dma_buf_map *map) +{ +struct ttm_resource *mem = &bo->mem; + +if (dma_buf_map_is_null (map)) +return ; + +if (!map->is_iomem) +vunmap (map->vaddr); +else if (!mem->bus.addr) +iounmap (map->vaddr_iomem); +dma_buf_map_clear (map); + +ttm_mem_io_free (bo->bdev, &bo->mem); +} +EXPORT_SYMBOL (ttm_bo_vunmap); +static int ttm_bo_wait_free_node (struct ttm_buffer_object *bo,bool dst_use_tt) {include/drm/drm_gem_ttm_helper.h +6
-0@@ -10,11 +10,17 @@#include <drm/ttm/ttm_bo_api.h> #include <drm/ttm/ttm_bo_driver.h> +struct dma_buf_map ; +#define drm_gem_ttm_of_gem (gem_obj) \ container_of(gem_obj, struct ttm_buffer_object, base)void drm_gem_ttm_print_info (struct drm_printer *p,unsigned int indent,const struct drm_gem_object *gem); +int drm_gem_ttm_vmap (struct drm_gem_object *gem, +struct dma_buf_map *map); +void drm_gem_ttm_vunmap (struct drm_gem_object *gem, +struct dma_buf_map *map);int drm_gem_ttm_mmap (struct drm_gem_object *gem,struct vm_area_struct *vma);include/drm/ttm/ttm_bo_api.h +28
-0@@ -48,6 +48,8 @@ struct ttm_bo_global;struct ttm_bo_device ; +struct dma_buf_map ; +struct drm_mm_node ;struct ttm_placement ;@@ -494,6 +496,32 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long start_page,*/ void ttm_bo_kunmap (struct ttm_bo_kmap_obj *map); +/** +* ttm_bo_vmap +* +* @bo: The buffer object. +* @map: pointer to a struct dma_buf_map representing the map. +* +* Sets up a kernel virtual mapping, using ioremap or vmap to the +* data in the buffer object. The parameter @map returns the virtual +* address as struct dma_buf_map. Unmap the buffer with ttm_bo_vunmap(). +* +* Returns +* -ENOMEM: Out of memory. +* -EINVAL: Invalid range. +*/ +int ttm_bo_vmap (struct ttm_buffer_object *bo,struct dma_buf_map *map); + +/** +* ttm_bo_vunmap +* +* @bo: The buffer object. +* @map: Object describing the map to unmap. +* +* Unmaps a kernel map set up by ttm_bo_vmap(). +*/ +void ttm_bo_vunmap (struct ttm_buffer_object *bo,struct dma_buf_map *map); +/** * ttm_bo_mmap_obj - mmap memory backed by a ttm buffer object. * include/linux/dma-buf-map.h +20
-0@@ -45,6 +45,12 @@* * dma_buf_map_set_vaddr(&map. 0xdeadbeaf); * +* To set an address in I/O memory, use dma_buf_map_set_vaddr_iomem(). +* +* .. code-block:: c +* +* dma_buf_map_set_vaddr_iomem(&map. 0xdeadbeaf); +* * Test if a mapping is valid with either dma_buf_map_is_set() or * dma_buf_map_is_null(). * @@ -118,6 +124,20 @@ static inline void dma_buf_map_set_vaddr(struct dma_buf_map *map, void *vaddr) map->is_iomem = false; } +/** +* dma_buf_map_set_vaddr_iomem - Sets a dma-buf mapping structure to an address in I/O memory +* @map: The dma-buf mapping structure +* @vaddr_iomem: An I/O-memory address +* +* Sets the address and the I/O-memory flag. +*/ +static inline void dma_buf_map_set_vaddr_iomem (struct dma_buf_map *map, +void __iomem *vaddr_iomem) +{ + map->vaddr_iomem = vaddr_iomem; + map->is_iomem = true; +} +/** * dma_buf_map_is_equal - Compares two dma-buf mapping structures for equality * @lhs: The dma-buf mapping structure
Re: [PATCH v5 10/10] drm/fb_helper: Support framebuffers in I/O memory
In reply to Thomas ZimmermannQuoted 131 lines · Thomas Zimmermann
At least sparc64 requires I/O-specific access to framebuffers. This patch updates the fbdev console accordingly.
For drivers with direct access to the framebuffer memory, the callback functions in struct
fb_opstest for the type of memory and call the rspfb_sys_offb_cfb_functions. Read and write operations are implemented internally by DRM's fbdev helper.For drivers that employ a shadow buffer, fbdev's blit function retrieves the framebuffer address as struct
dma_buf_map, and usesdma_buf_mapinterfaces to access the buffer.The bochs driver on sparc64 uses a workaround to flag the framebuffer as I/O memory and avoid a HW exception. With the introduction of struct
dma_buf_map, this is not required any longer. The patch removes the rsp code from both, bochs and fbdev.v5:
* implement fb_read/fb_write internally (Daniel, Sam)
v4:
* movedma_buf_mapchanges into separate patch (Daniel)
* TODO list: comment on fbdev updates (Daniel)Signed-off-by Thomas Zimmermann <[email protected]>Tested-by Sam Ravnborg <[email protected]>--- Documentation/gpu/todo.rst | 19 ++- drivers/gpu/drm/bochs/bochs_kms.c | 1 - drivers/gpu/drm/drm_fb_helper.c | 227 ++++++++++++++++++++++++++++-- include/drm/drm_mode_config.h | 12 -- 4 files changed, 230 insertions(+), 29 deletions(-)Documentation/gpu/todo.rst +17
-2@@ -197,13 +197,28 @@ Convert drivers to use drm_fbdev_generic_setup() ------------------------------------------------ Most drivers can use drm_fbdev_generic_setup(). Driver have to implement -atomic modesetting and GEM vmap support. Current generic fbdev emulation -expects the framebuffer in system memory (or system-like memory). +atomic modesetting and GEM vmap support. Historically, generic fbdev emulation +expected the framebuffer in system memory or system-like memory. By employing +struct dma_buf_map, drivers with frambuffers in I/O memory can be supported +as well. Contact: Maintainer of the driver you plan to convert Level: Intermediate +Reimplement functions in drm_fbdev_fb_ops without fbdev +------------------------------------------------------- + +A number of callback functions in drm_fbdev_fb_ops could benefit from +being rewritten without dependencies on the fbdev module. Some of the +helpers could further benefit from using struct dma_buf_map instead of +raw pointers. + +Contact: Thomas Zimmermann <[email protected]>, Daniel Vetter + +Level: Advanced + + drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup -----------------------------------------------------------------drivers/gpu/drm/bochs/bochs_kms.c +0
-1@@ -151,7 +151,6 @@ int bochs_kms_init(struct bochs_device *bochs) bochs->dev->mode_config.preferred_depth =24 ; bochs->dev->mode_config.prefer_shadow =0 ; bochs->dev->mode_config.prefer_shadow_fbdev =1 ; - bochs->dev->mode_config.fbdev_use_iomem = true; bochs->dev->mode_config.quirk_addfb_prefer_host_byte_order = true; bochs->dev->mode_config.funcs = &bochs_mode_funcs;drivers/gpu/drm/drm_fb_helper.c +10
-9@@ -372,24 +372,22 @@ static void drm_fb_helper_resume_worker(struct work_struct *work) }static void drm_fb_helper_dirty_blit_real (struct drm_fb_helper *fb_helper, -struct drm_clip_rect *clip) +struct drm_clip_rect *clip, +struct dma_buf_map *dst) {struct drm_framebuffer *fb = fb_helper->fb;unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y; -for (y = clip->y1; y < clip->y2; y++) { -if (!fb_helper->dev->mode_config.fbdev_use_iomem) -memcpy (dst, src, len); -else -memcpy_toio ((void __iomem *)dst, src, len); +dma_buf_map_incr (dst, offset);/* go to first pixel within clip rect */ +for (y = clip->y1; y < clip->y2; y++) { +dma_buf_map_memcpy_to (dst, src, len); +dma_buf_map_incr (dst, fb->pitches[0 ]); src += fb->pitches[0 ]; - dst += fb->pitches[0 ]; } }@@ -417,8 +415,9 @@ static void drm_fb_helper_dirty_work(struct work_struct *work) ret =drm_client_buffer_vmap (helper->buffer, &map);if (ret)return ; -drm_fb_helper_dirty_blit_real (helper, &clip_copy); +drm_fb_helper_dirty_blit_real (helper, &clip_copy, &map); } +if (helper->fb->funcs->dirty) helper->fb->funcs->dirty (helper->fb,NULL ,0 ,0 , &clip_copy,1 );@@ -2027,6 +2026,206 @@ static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma)return -ENODEV ; } +static bool drm_fbdev_use_iomem (struct fb_info *info) +{
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2040
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+struct drm_fb_helper *fb_helper = info->par; +struct drm_client_buffer *buffer = fb_helper->buffer; + +return !drm_fbdev_use_shadow_fb (fb_helper) && buffer->map.is_iomem; +} + +static ssize_t fb_read_screen_base (struct fb_info *info,char __user *buf,size_t count, +loff_t pos) +{ +const char __iomem *src = info->screen_base + pos;
Maybe a bit much a bikeshed, but I'd write this in terms of drm objects, like the dirty_blit function, using the dma_buf_map (instead of the fb_info parameter). And then instead ofscreen_base and screen_buffer suffixes give them _mem and _iomem suffixes.
Same for write below. Or I'm not quite understanding why we do it like this here - I don't think this code will be used outside of the generic fbdev code, so we can always assume that drm_fb_helper->buffer is set up.
The other thing I think we need is some minimal testcases to make sure. The fbtest tool used way back seems to have disappeared, I couldn't find a copy of the source anywhere anymore.
With all that: Acked-by: Daniel Vetter <daniel.vetter-/[email protected]>
Cheers, Daniel
Quoted 234 lines
drivers/gpu/drm/drm_fb_helper.c +193
-5+size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +char *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; + +while (count) { +size_t c =min (count, alloc_size); + +memcpy_fromio (tmp, src, c); +if (copy_to_user (buf, tmp, c)) { + ret = -EFAULT ; +break ; + } + + src += c; + buf += c; + ret += c; + count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_read_screen_buffer (struct fb_info *info,char __user *buf,size_t count, +loff_t pos) +{ +const char *src = info->screen_buffer + pos; + +if (copy_to_user (buf, src, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_read (struct fb_info *info,char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos >= total_size) +return 0 ; +if (count >= total_size) + count = total_size; +if (total_size - count < pos) + count = total_size - pos; + +if (drm_fbdev_use_iomem (info)) + ret =fb_read_screen_base (info, buf, count, pos); +else + ret =fb_read_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +return ret; +} + +static ssize_t fb_write_screen_base (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos) +{ +char __iomem *dst = info->screen_base + pos; +size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +u8 *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; + +while (count) { +size_t c =min (count, alloc_size); + +if (copy_from_user (tmp, buf, c)) { + ret = -EFAULT ; +break ; + } +memcpy_toio (dst, tmp, c); + + dst += c; + buf += c; + ret += c; + count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_write_screen_buffer (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos) +{ +char *dst = info->screen_buffer + pos; + +if (copy_from_user (dst, buf, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_write (struct fb_info *info,const char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; +int err; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos > total_size) +return -EFBIG ; +if (count > total_size) { + err = -EFBIG ; + count = total_size; + } +if (total_size - count < pos) { +if (!err) + err = -ENOSPC ; + count = total_size - pos; + } + +/* +* Copy to framebuffer even if we already logged an error. Emulates +* the behavior of the original fbdev implementation. +*/ +if (drm_fbdev_use_iomem (info)) + ret =fb_write_screen_base (info, buf, count, pos); +else + ret =fb_write_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +if (err) +return err; + +return ret; +} + +static void drm_fbdev_fb_fillrect (struct fb_info *info, +const struct fb_fillrect *rect) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_fillrect (info, rect); +else +drm_fb_helper_sys_fillrect (info, rect); +} + +static void drm_fbdev_fb_copyarea (struct fb_info *info, +const struct fb_copyarea *area) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_copyarea (info, area); +else +drm_fb_helper_sys_copyarea (info, area); +} + +static void drm_fbdev_fb_imageblit (struct fb_info *info, +const struct fb_image *image) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_imageblit (info, image); +else +drm_fb_helper_sys_imageblit (info, image); +} +static const struct fb_ops drm_fbdev_fb_ops = { .owner =THIS_MODULE ,DRM_FB_HELPER_DEFAULT_OPS ,@@ -2034,11 +2233,11 @@ static const struct fb_ops drm_fbdev_fb_ops = { .fb_release = drm_fbdev_fb_release, .fb_destroy = drm_fbdev_fb_destroy, .fb_mmap = drm_fbdev_fb_mmap, - .fb_read = drm_fb_helper_sys_read, - .fb_write = drm_fb_helper_sys_write, - .fb_fillrect = drm_fb_helper_sys_fillrect, - .fb_copyarea = drm_fb_helper_sys_copyarea, - .fb_imageblit = drm_fb_helper_sys_imageblit, + .fb_read = drm_fbdev_fb_read, + .fb_write = drm_fbdev_fb_write, + .fb_fillrect = drm_fbdev_fb_fillrect, + .fb_copyarea = drm_fbdev_fb_copyarea, + .fb_imageblit = drm_fbdev_fb_imageblit, };static struct fb_deferred_io drm_fbdev_defio = {include/drm/drm_mode_config.h +0
-12@@ -877,18 +877,6 @@ struct drm_mode_config {*/ bool prefer_shadow_fbdev; -/** -* @fbdev_use_iomem: -* -* Set to true if framebuffer reside in iomem. -* When set to true memcpy_toio() is used when copying the framebuffer in -* drm_fb_helper.drm_fb_helper_dirty_blit_real(). -* -* FIXME: This should be replaced with a per-mapping is_iomem -* flag (like ttm does), and then used everywhere in fbdev code. -*/ -bool fbdev_use_iomem; -/** * @quirk_addfb_prefer_xbgr_30bpp: * Signature · 1 line
2.28.0
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Thomas Zimmermann
maintainerRe: [PATCH v5 10/10] drm/fb_helper: Support framebuffers in I/O memory
Hi
Quoted 139 lines · Daniel Vetter
Quoted 138 lines · Thomas Zimmermann
At least sparc64 requires I/O-specific access to framebuffers. This patch updates the fbdev console accordingly.
For drivers with direct access to the framebuffer memory, the callback functions in struct
fb_opstest for the type of memory and call the rspfb_sys_offb_cfb_functions. Read and write operations are implemented internally by DRM's fbdev helper.For drivers that employ a shadow buffer, fbdev's blit function retrieves the framebuffer address as struct
dma_buf_map, and usesdma_buf_mapinterfaces to access the buffer.The bochs driver on sparc64 uses a workaround to flag the framebuffer as I/O memory and avoid a HW exception. With the introduction of struct
dma_buf_map, this is not required any longer. The patch removes the rsp code from both, bochs and fbdev.v5:
* implement fb_read/fb_write internally (Daniel, Sam)
v4:
* movedma_buf_mapchanges into separate patch (Daniel)
* TODO list: comment on fbdev updates (Daniel)Signed-off-by Thomas Zimmermann <[email protected]>Tested-by Sam Ravnborg <[email protected]>--- Documentation/gpu/todo.rst | 19 ++- drivers/gpu/drm/bochs/bochs_kms.c | 1 - drivers/gpu/drm/drm_fb_helper.c | 227 ++++++++++++++++++++++++++++-- include/drm/drm_mode_config.h | 12 -- 4 files changed, 230 insertions(+), 29 deletions(-)Documentation/gpu/todo.rst +17
-2@@ -197,13 +197,28 @@ Convert drivers to use drm_fbdev_generic_setup() ------------------------------------------------ Most drivers can use drm_fbdev_generic_setup(). Driver have to implement -atomic modesetting and GEM vmap support. Current generic fbdev emulation -expects the framebuffer in system memory (or system-like memory). +atomic modesetting and GEM vmap support. Historically, generic fbdev emulation +expected the framebuffer in system memory or system-like memory. By employing +struct dma_buf_map, drivers with frambuffers in I/O memory can be supported +as well. Contact: Maintainer of the driver you plan to convert Level: Intermediate +Reimplement functions in drm_fbdev_fb_ops without fbdev +------------------------------------------------------- + +A number of callback functions in drm_fbdev_fb_ops could benefit from +being rewritten without dependencies on the fbdev module. Some of the +helpers could further benefit from using struct dma_buf_map instead of +raw pointers. + +Contact: Thomas Zimmermann <[email protected]>, Daniel Vetter + +Level: Advanced + + drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup -----------------------------------------------------------------drivers/gpu/drm/bochs/bochs_kms.c +0
-1@@ -151,7 +151,6 @@ int bochs_kms_init(struct bochs_device *bochs) bochs->dev->mode_config.preferred_depth =24 ; bochs->dev->mode_config.prefer_shadow =0 ; bochs->dev->mode_config.prefer_shadow_fbdev =1 ; - bochs->dev->mode_config.fbdev_use_iomem = true; bochs->dev->mode_config.quirk_addfb_prefer_host_byte_order = true; bochs->dev->mode_config.funcs = &bochs_mode_funcs;drivers/gpu/drm/drm_fb_helper.c +17
-9@@ -372,24 +372,22 @@ static void drm_fb_helper_resume_worker(struct work_struct *work) }static void drm_fb_helper_dirty_blit_real (struct drm_fb_helper *fb_helper, -struct drm_clip_rect *clip) +struct drm_clip_rect *clip, +struct dma_buf_map *dst) {struct drm_framebuffer *fb = fb_helper->fb;unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y; -for (y = clip->y1; y < clip->y2; y++) { -if (!fb_helper->dev->mode_config.fbdev_use_iomem) -memcpy (dst, src, len); -else -memcpy_toio ((void __iomem *)dst, src, len); +dma_buf_map_incr (dst, offset);/* go to first pixel within clip rect */ +for (y = clip->y1; y < clip->y2; y++) { +dma_buf_map_memcpy_to (dst, src, len); +dma_buf_map_incr (dst, fb->pitches[0 ]); src += fb->pitches[0 ]; - dst += fb->pitches[0 ]; } }@@ -417,8 +415,9 @@ static void drm_fb_helper_dirty_work(struct work_struct *work) ret =drm_client_buffer_vmap (helper->buffer, &map);if (ret)return ; -drm_fb_helper_dirty_blit_real (helper, &clip_copy); +drm_fb_helper_dirty_blit_real (helper, &clip_copy, &map); } +if (helper->fb->funcs->dirty) helper->fb->funcs->dirty (helper->fb,NULL ,0 ,0 , &clip_copy,1 );@@ -2027,6 +2026,206 @@ static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma)return -ENODEV ; } +static bool drm_fbdev_use_iomem (struct fb_info *info) +{ +struct drm_fb_helper *fb_helper = info->par; +struct drm_client_buffer *buffer = fb_helper->buffer; + +return !drm_fbdev_use_shadow_fb (fb_helper) && buffer->map.is_iomem; +} + +static ssize_t fb_read_screen_base(struct fb_info *info, char __user *buf, size_t count,
Quoted 8 lines
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2040
Quoted 3 lines
drivers/gpu/drm/drm_fb_helper.c +3
-0+ loff_t pos) +{ +const char __iomem *src = info->screen_base + pos;Maybe a bit much a bikeshed, but I'd write this in terms of drm objects, like the
dirty_blitfunction, using thedma_buf_map(instead of thefb_infoparameter). And then instead ofscreen_baseandscreen_buffersuffixes give them_memand_iomemsuffixes.
Screen_buffer can be a shadow buffer. Until the blit worker (see drm_fb_helper_dirty_work() ) completes, it might be more up to date than the real buffer that's stored in the client.
The orignal fbdev code supported an fb_sync callback to synchronize with outstanding screen updates (e.g., HW blit ops), but fb_sync is just overhead here. Copying from screen_buffer or screen_base always returns the most up-to-date image.
Quoted 4 lines
Same for write below. Or I'm not quite understanding why we do it like this here - I don't think this code will be used outside of the generic fbdev code, so we can always assume that
drm_fb_helper->bufferis set up.
It's similar as in the read case. If we write to the client's buffer, an outstanding blit worker could write the now-outdated shadow buffer over the user's newly written framebuffer data.
Thinking about it, we might want to schedule the blit worker at the end of each fb_write, so that the data makes it into the HW buffer in time.
Quoted 4 lines
The other thing I think we need is some minimal testcases to make sure. The fbtest tool used way back seems to have disappeared, I couldn't find a copy of the source anywhere anymore.
As discussed on IRC, I'll add some testcase to the igt test. I'll share the link here when done.
Best regards
Thomas
Quoted 240 lines
With all that: Acked-by: Daniel Vetter <[email protected]>
Cheers, Daniel
Quoted 234 lines
drivers/gpu/drm/drm_fb_helper.c +193
-5+size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +char *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; + +while (count) { +size_t c =min (count, alloc_size); + +memcpy_fromio (tmp, src, c); +if (copy_to_user (buf, tmp, c)) { + ret = -EFAULT ; +break ; + } + + src += c; + buf += c; + ret += c; + count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_read_screen_buffer (struct fb_info *info,char __user *buf,size_t count, +loff_t pos) +{ +const char *src = info->screen_buffer + pos; + +if (copy_to_user (buf, src, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_read (struct fb_info *info,char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos >= total_size) +return 0 ; +if (count >= total_size) + count = total_size; +if (total_size - count < pos) + count = total_size - pos; + +if (drm_fbdev_use_iomem (info)) + ret =fb_read_screen_base (info, buf, count, pos); +else + ret =fb_read_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +return ret; +} + +static ssize_t fb_write_screen_base (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos) +{ +char __iomem *dst = info->screen_base + pos; +size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +u8 *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; + +while (count) { +size_t c =min (count, alloc_size); + +if (copy_from_user (tmp, buf, c)) { + ret = -EFAULT ; +break ; + } +memcpy_toio (dst, tmp, c); + + dst += c; + buf += c; + ret += c; + count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_write_screen_buffer (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos) +{ +char *dst = info->screen_buffer + pos; + +if (copy_from_user (dst, buf, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_write (struct fb_info *info,const char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; +int err; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos > total_size) +return -EFBIG ; +if (count > total_size) { + err = -EFBIG ; + count = total_size; + } +if (total_size - count < pos) { +if (!err) + err = -ENOSPC ; + count = total_size - pos; + } + +/* +* Copy to framebuffer even if we already logged an error. Emulates +* the behavior of the original fbdev implementation. +*/ +if (drm_fbdev_use_iomem (info)) + ret =fb_write_screen_base (info, buf, count, pos); +else + ret =fb_write_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +if (err) +return err; + +return ret; +} + +static void drm_fbdev_fb_fillrect (struct fb_info *info, +const struct fb_fillrect *rect) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_fillrect (info, rect); +else +drm_fb_helper_sys_fillrect (info, rect); +} + +static void drm_fbdev_fb_copyarea (struct fb_info *info, +const struct fb_copyarea *area) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_copyarea (info, area); +else +drm_fb_helper_sys_copyarea (info, area); +} + +static void drm_fbdev_fb_imageblit (struct fb_info *info, +const struct fb_image *image) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_imageblit (info, image); +else +drm_fb_helper_sys_imageblit (info, image); +} +static const struct fb_ops drm_fbdev_fb_ops = { .owner =THIS_MODULE ,DRM_FB_HELPER_DEFAULT_OPS ,@@ -2034,11 +2233,11 @@ static const struct fb_ops drm_fbdev_fb_ops = { .fb_release = drm_fbdev_fb_release, .fb_destroy = drm_fbdev_fb_destroy, .fb_mmap = drm_fbdev_fb_mmap, - .fb_read = drm_fb_helper_sys_read, - .fb_write = drm_fb_helper_sys_write, - .fb_fillrect = drm_fb_helper_sys_fillrect, - .fb_copyarea = drm_fb_helper_sys_copyarea, - .fb_imageblit = drm_fb_helper_sys_imageblit, + .fb_read = drm_fbdev_fb_read, + .fb_write = drm_fbdev_fb_write, + .fb_fillrect = drm_fbdev_fb_fillrect, + .fb_copyarea = drm_fbdev_fb_copyarea, + .fb_imageblit = drm_fbdev_fb_imageblit, };static struct fb_deferred_io drm_fbdev_defio = {include/drm/drm_mode_config.h +0
-12@@ -877,18 +877,6 @@ struct drm_mode_config {*/ bool prefer_shadow_fbdev; -/** -* @fbdev_use_iomem: -* -* Set to true if framebuffer reside in iomem. -* When set to true memcpy_toio() is used when copying the framebuffer in -* drm_fb_helper.drm_fb_helper_dirty_blit_real(). -* -* FIXME: This should be replaced with a per-mapping is_iomem -* flag (like ttm does), and then used everywhere in fbdev code. -*/ -bool fbdev_use_iomem; -/** * @quirk_addfb_prefer_xbgr_30bpp: * Signature · 1 line
2.28.0
Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Re: [PATCH v5 08/10] drm/gem: Store client buffer mappings as struct dma_buf_map
In reply to Thomas ZimmermannQuoted 153 lines · Thomas Zimmermann
Kernel DRM clients now store their framebuffer address in an instance of struct
dma_buf_map. Depending on the buffer's location, the address refers to system or I/O memory.Callers of
drm_client_buffer_vmap()receive a copy of the value in the call's supplied arguments. It can be accessed and modified withdma_buf_mapinterfaces.Signed-off-by Thomas Zimmermann <[email protected]>Reviewed-by Daniel Vetter <daniel.vetter-/[email protected]>Tested-by Sam Ravnborg <[email protected]>--- drivers/gpu/drm/drm_client.c | 34 +++++++++++++++++++-------------- drivers/gpu/drm/drm_fb_helper.c | 23 +++++++++++++--------- include/drm/drm_client.h | 7 ++++--- 3 files changed, 38 insertions(+), 26 deletions(-)drivers/gpu/drm/drm_client.c +20
-14@@ -235,7 +235,7 @@ static void drm_client_buffer_delete(struct drm_client_buffer *buffer) {struct drm_device *dev = buffer->client->dev; -drm_gem_vunmap (buffer->gem, buffer->vaddr); +drm_gem_vunmap (buffer->gem, &buffer->map);if (buffer->gem)drm_gem_object_put (buffer->gem);@@ -291,25 +291,31 @@ drm_client_buffer_create(struct drm_client_dev *client, u32 width, u32 height, u/** * drm_client_buffer_vmap - Map DRM client buffer into address space * @buffer: DRM client buffer +* @map_copy: Returns the mapped memory's address * * This function maps a client buffer into kernel address space. If the -* buffer is already mapped, it returns the mapping's address. +* buffer is already mapped, it returns the existing mapping's address. * * Client buffer mappings are not ref'counted. Each call to * drm_client_buffer_vmap() should be followed by a call to * drm_client_buffer_vunmap(); or the client buffer should be mapped * throughout its lifetime. * +* The returned address is a copy of the internal value. In contrast to +* other vmap interfaces, you don't need it for the client's vunmap +* function. So you can modify it at will during blit and draw operations. +* * Returns: -* The mapped memory's address +* 0 on success, or a negative errno code otherwise. */ -void *drm_client_buffer_vmap (struct drm_client_buffer *buffer) +int +drm_client_buffer_vmap (struct drm_client_buffer *buffer,struct dma_buf_map *map_copy) { -struct dma_buf_map map; +struct dma_buf_map *map = &buffer->map;int ret; -if (buffer->vaddr) -return buffer->vaddr; +if (dma_buf_map_is_set (map)) + goto out;/* * FIXME: The dependency on GEM here isn't required, we could @@ -319,13 +325,14 @@ void *drm_client_buffer_vmap(struct drm_client_buffer *buffer)* fd_install step out of the driver backend hooks, to make that * final step optional for internal users. */ - ret =drm_gem_vmap (buffer->gem, &map); + ret =drm_gem_vmap (buffer->gem, map);if (ret) -return ERR_PTR (ret); +return ret; - buffer->vaddr = map.vaddr; +out: + *map_copy = *map; -return map.vaddr; +return 0 ; }EXPORT_SYMBOL (drm_client_buffer_vmap);@@ -339,10 +346,9 @@ EXPORT_SYMBOL(drm_client_buffer_vmap);*/ void drm_client_buffer_vunmap (struct drm_client_buffer *buffer) { -struct dma_buf_map map =DMA_BUF_MAP_INIT_VADDR (buffer->vaddr); +struct dma_buf_map *map = &buffer->map; -drm_gem_vunmap (buffer->gem, &map); - buffer->vaddr =NULL ; +drm_gem_vunmap (buffer->gem, map); }EXPORT_SYMBOL (drm_client_buffer_vunmap);drivers/gpu/drm/drm_fb_helper.c +9
-8@@ -378,7 +378,7 @@ static void drm_fb_helper_dirty_blit_real(struct drm_fb_helper *fb_helper,unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->vaddr + offset; +void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y;@@ -400,7 +400,8 @@ static void drm_fb_helper_dirty_work(struct work_struct *work)struct drm_clip_rect *clip = &helper->dirty_clip;struct drm_clip_rect clip_copy;unsigned long flags; -void *vaddr; +struct dma_buf_map map; +int ret;spin_lock_irqsave (&helper->dirty_lock, flags); clip_copy = *clip;@@ -413,8 +414,8 @@ static void drm_fb_helper_dirty_work(struct work_struct *work)/* Generic fbdev uses a shadow buffer */ if (helper->buffer) { - vaddr =drm_client_buffer_vmap (helper->buffer); -if (IS_ERR (vaddr)) + ret =drm_client_buffer_vmap (helper->buffer, &map); +if (ret)return ;drm_fb_helper_dirty_blit_real (helper, &clip_copy); }@@ -2060,7 +2061,8 @@ static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper,struct drm_framebuffer *fb;struct fb_info *fbi;u32 format; -void *vaddr; +struct dma_buf_map map; +int ret; drm_dbg_kms(dev, "surface width(%d), height(%d) and bpp(%d)\n", sizes->surface_width, sizes->surface_height,@@ -2096,11 +2098,14 @@ static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper,fb_deferred_io_init (fbi); }else {/* buffer is mapped for HW framebuffer */ - vaddr =drm_client_buffer_vmap (fb_helper->buffer); -if (IS_ERR (vaddr)) -return PTR_ERR (vaddr); + ret =drm_client_buffer_vmap (fb_helper->buffer, &map); +if (ret)
drivers/gpu/drm/drm_fb_helper.c static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper, · line 2111
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +5
-1+return ret; +if (map.is_iomem) + fbi->screen_base = map.vaddr_iomem; +else + fbi->screen_buffer = map.vaddr; - fbi->screen_buffer = vaddr;/* Shamelessly leak the physical address to user-space */ #if IS_ENABLED (CONFIG_DRM_FBDEV_LEAK_PHYS_SMEM )if (drm_leak_fbdev_smem && fbi->fix.smem_start ==0 )
Just noticed a tiny thing here: I think this needs to be patched to only set smem_start when the map is _not_ iomem. Since virt_to_page isn't defined on iomem at all.
I guess it'd be neat if we can set this for iomem too, but I have no idea how to convert an iomem pointer back to a bus_addr_t ...
Cheers, Daniel
Quoted 36 lines
include/drm/drm_client.h +4
-3@@ -3,6 +3,7 @@#ifndef _DRM_CLIENT_H_#define _DRM_CLIENT_H_ +#include <linux/dma-buf-map.h> #include <linux/lockdep.h> #include <linux/mutex.h> #include <linux/types.h> @@ -141,9 +142,9 @@ struct drm_client_buffer {struct drm_gem_object *gem;/** -* @vaddr: Virtual address for the buffer +* @map: Virtual address for the buffer */ -void *vaddr; +struct dma_buf_map map;/** * @fb: DRM framebuffer @@ -155,7 +156,7 @@ struct drm_client_buffer * drm_client_framebuffer_create(struct drm_client_dev *client, u32 width, u32 height, u32 format);void drm_client_framebuffer_delete (struct drm_client_buffer *buffer);int drm_client_framebuffer_flush (struct drm_client_buffer *buffer,struct drm_rect *rect); -void *drm_client_buffer_vmap (struct drm_client_buffer *buffer); +int drm_client_buffer_vmap (struct drm_client_buffer *buffer,struct dma_buf_map *map);void drm_client_buffer_vunmap (struct drm_client_buffer *buffer);int drm_client_modeset_create (struct drm_client_dev *client);Signature · 1 line
2.28.0
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Daniel Vetter
Reviewed-byRe: [PATCH v5 10/10] drm/fb_helper: Support framebuffers in I/O memory
In reply to Thomas ZimmermannQuoted 165 lines · Thomas Zimmermann
Hi
Quoted 139 lines · Daniel Vetter
Quoted 138 lines · Thomas Zimmermann
At least sparc64 requires I/O-specific access to framebuffers. This patch updates the fbdev console accordingly.
For drivers with direct access to the framebuffer memory, the callback functions in struct
fb_opstest for the type of memory and call the rspfb_sys_offb_cfb_functions. Read and write operations are implemented internally by DRM's fbdev helper.For drivers that employ a shadow buffer, fbdev's blit function retrieves the framebuffer address as struct
dma_buf_map, and usesdma_buf_mapinterfaces to access the buffer.The bochs driver on sparc64 uses a workaround to flag the framebuffer as I/O memory and avoid a HW exception. With the introduction of struct
dma_buf_map, this is not required any longer. The patch removes the rsp code from both, bochs and fbdev.v5:
* implement fb_read/fb_write internally (Daniel, Sam)
v4:
* movedma_buf_mapchanges into separate patch (Daniel)
* TODO list: comment on fbdev updates (Daniel)Signed-off-by Thomas Zimmermann <[email protected]>Tested-by Sam Ravnborg <[email protected]>--- Documentation/gpu/todo.rst | 19 ++- drivers/gpu/drm/bochs/bochs_kms.c | 1 - drivers/gpu/drm/drm_fb_helper.c | 227 ++++++++++++++++++++++++++++-- include/drm/drm_mode_config.h | 12 -- 4 files changed, 230 insertions(+), 29 deletions(-)Documentation/gpu/todo.rst +17
-2@@ -197,13 +197,28 @@ Convert drivers to use drm_fbdev_generic_setup() ------------------------------------------------ Most drivers can use drm_fbdev_generic_setup(). Driver have to implement -atomic modesetting and GEM vmap support. Current generic fbdev emulation -expects the framebuffer in system memory (or system-like memory). +atomic modesetting and GEM vmap support. Historically, generic fbdev emulation +expected the framebuffer in system memory or system-like memory. By employing +struct dma_buf_map, drivers with frambuffers in I/O memory can be supported +as well. Contact: Maintainer of the driver you plan to convert Level: Intermediate +Reimplement functions in drm_fbdev_fb_ops without fbdev +------------------------------------------------------- + +A number of callback functions in drm_fbdev_fb_ops could benefit from +being rewritten without dependencies on the fbdev module. Some of the +helpers could further benefit from using struct dma_buf_map instead of +raw pointers. + +Contact: Thomas Zimmermann <[email protected]>, Daniel Vetter + +Level: Advanced + + drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup -----------------------------------------------------------------drivers/gpu/drm/bochs/bochs_kms.c +0
-1@@ -151,7 +151,6 @@ int bochs_kms_init(struct bochs_device *bochs) bochs->dev->mode_config.preferred_depth =24 ; bochs->dev->mode_config.prefer_shadow =0 ; bochs->dev->mode_config.prefer_shadow_fbdev =1 ; - bochs->dev->mode_config.fbdev_use_iomem = true; bochs->dev->mode_config.quirk_addfb_prefer_host_byte_order = true; bochs->dev->mode_config.funcs = &bochs_mode_funcs;drivers/gpu/drm/drm_fb_helper.c +17
-9@@ -372,24 +372,22 @@ static void drm_fb_helper_resume_worker(struct work_struct *work) }static void drm_fb_helper_dirty_blit_real (struct drm_fb_helper *fb_helper, -struct drm_clip_rect *clip) +struct drm_clip_rect *clip, +struct dma_buf_map *dst) {struct drm_framebuffer *fb = fb_helper->fb;unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y; -for (y = clip->y1; y < clip->y2; y++) { -if (!fb_helper->dev->mode_config.fbdev_use_iomem) -memcpy (dst, src, len); -else -memcpy_toio ((void __iomem *)dst, src, len); +dma_buf_map_incr (dst, offset);/* go to first pixel within clip rect */ +for (y = clip->y1; y < clip->y2; y++) { +dma_buf_map_memcpy_to (dst, src, len); +dma_buf_map_incr (dst, fb->pitches[0 ]); src += fb->pitches[0 ]; - dst += fb->pitches[0 ]; } }@@ -417,8 +415,9 @@ static void drm_fb_helper_dirty_work(struct work_struct *work) ret =drm_client_buffer_vmap (helper->buffer, &map);if (ret)return ; -drm_fb_helper_dirty_blit_real (helper, &clip_copy); +drm_fb_helper_dirty_blit_real (helper, &clip_copy, &map); } +if (helper->fb->funcs->dirty) helper->fb->funcs->dirty (helper->fb,NULL ,0 ,0 , &clip_copy,1 );@@ -2027,6 +2026,206 @@ static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma)return -ENODEV ; } +static bool drm_fbdev_use_iomem (struct fb_info *info) +{ +struct drm_fb_helper *fb_helper = info->par; +struct drm_client_buffer *buffer = fb_helper->buffer; + +return !drm_fbdev_use_shadow_fb (fb_helper) && buffer->map.is_iomem; +} + +static ssize_t fb_read_screen_base(struct fb_info *info, char __user *buf, size_t count,Quoted 8 lines
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2040
Quoted 3 lines
drivers/gpu/drm/drm_fb_helper.c +3
-0+ loff_t pos) +{ +const char __iomem *src = info->screen_base + pos;Maybe a bit much a bikeshed, but I'd write this in terms of drm objects, like the
dirty_blitfunction, using thedma_buf_map(instead of thefb_infoparameter). And then instead ofscreen_baseandscreen_buffersuffixes give them_memand_iomemsuffixes.
Screen_buffercan be a shadow buffer. Until the blit worker (seedrm_fb_helper_dirty_work()) completes, it might be more up to date than the real buffer that's stored in the client.The orignal fbdev code supported an
fb_synccallback to synchronize with outstanding screen updates (e.g., HW blit ops), butfb_syncis just overhead here. Copying fromscreen_bufferorscreen_basealways returns the most up-to-date image.Quoted 4 lines
Same for write below. Or I'm not quite understanding why we do it like this here - I don't think this code will be used outside of the generic fbdev code, so we can always assume that
drm_fb_helper->bufferis set up.
Quoted 6 lines
It's similar as in the read case. If we write to the client's buffer, an outstanding blit worker could write the now-outdated shadow buffer over the user's newly written framebuffer data.
Thinking about it, we might want to schedule the blit worker at the end of each
fb_write, so that the data makes it into the HW buffer in time.
Hm ok, makes some sense. I think there's some potential for cleanup if we add a dma_buf_map drm_fb_helper->uapi_map which points at the right thing always. That could then also the drm_fbdev_use_iomem() helper and make this all look really neat.
But maybe a follow up clean up patch, if you're bored. As-is:
While looking at this I also noticed a potential small issue in an earlier patch.
Quoted 258 lines
Quoted 3 lines
The other thing I think we need is some minimal testcases to make sure. The fbtest tool used way back seems to have disappeared, I couldn't find a copy of the source anywhere anymore.
As discussed on IRC, I'll add some testcase to the igt test. I'll share the link here when done.
Best regards
ThomasQuoted 240 lines
With all that: Acked-by: Daniel Vetter <[email protected]>
Cheers, Daniel
Quoted 234 lines
drivers/gpu/drm/drm_fb_helper.c +193
-5+size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +char *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; + +while (count) { +size_t c =min (count, alloc_size); + +memcpy_fromio (tmp, src, c); +if (copy_to_user (buf, tmp, c)) { + ret = -EFAULT ; +break ; + } + + src += c; + buf += c; + ret += c; + count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_read_screen_buffer (struct fb_info *info,char __user *buf,size_t count, +loff_t pos) +{ +const char *src = info->screen_buffer + pos; + +if (copy_to_user (buf, src, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_read (struct fb_info *info,char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos >= total_size) +return 0 ; +if (count >= total_size) + count = total_size; +if (total_size - count < pos) + count = total_size - pos; + +if (drm_fbdev_use_iomem (info)) + ret =fb_read_screen_base (info, buf, count, pos); +else + ret =fb_read_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +return ret; +} + +static ssize_t fb_write_screen_base (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos) +{ +char __iomem *dst = info->screen_base + pos; +size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +u8 *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; + +while (count) { +size_t c =min (count, alloc_size); + +if (copy_from_user (tmp, buf, c)) { + ret = -EFAULT ; +break ; + } +memcpy_toio (dst, tmp, c); + + dst += c; + buf += c; + ret += c; + count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_write_screen_buffer (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos) +{ +char *dst = info->screen_buffer + pos; + +if (copy_from_user (dst, buf, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_write (struct fb_info *info,const char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; +int err; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos > total_size) +return -EFBIG ; +if (count > total_size) { + err = -EFBIG ; + count = total_size; + } +if (total_size - count < pos) { +if (!err) + err = -ENOSPC ; + count = total_size - pos; + } + +/* +* Copy to framebuffer even if we already logged an error. Emulates +* the behavior of the original fbdev implementation. +*/ +if (drm_fbdev_use_iomem (info)) + ret =fb_write_screen_base (info, buf, count, pos); +else + ret =fb_write_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +if (err) +return err; + +return ret; +} + +static void drm_fbdev_fb_fillrect (struct fb_info *info, +const struct fb_fillrect *rect) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_fillrect (info, rect); +else +drm_fb_helper_sys_fillrect (info, rect); +} + +static void drm_fbdev_fb_copyarea (struct fb_info *info, +const struct fb_copyarea *area) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_copyarea (info, area); +else +drm_fb_helper_sys_copyarea (info, area); +} + +static void drm_fbdev_fb_imageblit (struct fb_info *info, +const struct fb_image *image) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_imageblit (info, image); +else +drm_fb_helper_sys_imageblit (info, image); +} +static const struct fb_ops drm_fbdev_fb_ops = { .owner =THIS_MODULE ,DRM_FB_HELPER_DEFAULT_OPS ,@@ -2034,11 +2233,11 @@ static const struct fb_ops drm_fbdev_fb_ops = { .fb_release = drm_fbdev_fb_release, .fb_destroy = drm_fbdev_fb_destroy, .fb_mmap = drm_fbdev_fb_mmap, - .fb_read = drm_fb_helper_sys_read, - .fb_write = drm_fb_helper_sys_write, - .fb_fillrect = drm_fb_helper_sys_fillrect, - .fb_copyarea = drm_fb_helper_sys_copyarea, - .fb_imageblit = drm_fb_helper_sys_imageblit, + .fb_read = drm_fbdev_fb_read, + .fb_write = drm_fbdev_fb_write, + .fb_fillrect = drm_fbdev_fb_fillrect, + .fb_copyarea = drm_fbdev_fb_copyarea, + .fb_imageblit = drm_fbdev_fb_imageblit, };static struct fb_deferred_io drm_fbdev_defio = {include/drm/drm_mode_config.h +0
-12@@ -877,18 +877,6 @@ struct drm_mode_config {*/ bool prefer_shadow_fbdev; -/** -* @fbdev_use_iomem: -* -* Set to true if framebuffer reside in iomem. -* When set to true memcpy_toio() is used when copying the framebuffer in -* drm_fb_helper.drm_fb_helper_dirty_blit_real(). -* -* FIXME: This should be replaced with a per-mapping is_iomem -* flag (like ttm does), and then used everywhere in fbdev code. -*/ -bool fbdev_use_iomem; -/** * @quirk_addfb_prefer_xbgr_30bpp: * Signature · 1 line
2.28.0Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Thomas Zimmermann
maintainerRe: [PATCH v5 08/10] drm/gem: Store client buffer mappings as struct dma_buf_map
In reply to Daniel VetterHi
Quoted 169 lines · Daniel Vetter
Quoted 153 lines · Thomas Zimmermann
Kernel DRM clients now store their framebuffer address in an instance of struct
dma_buf_map. Depending on the buffer's location, the address refers to system or I/O memory.Callers of
drm_client_buffer_vmap()receive a copy of the value in the call's supplied arguments. It can be accessed and modified withdma_buf_mapinterfaces.Signed-off-by Thomas Zimmermann <[email protected]>Reviewed-by Daniel Vetter <[email protected]>Tested-by Sam Ravnborg <[email protected]>--- drivers/gpu/drm/drm_client.c | 34 +++++++++++++++++++-------------- drivers/gpu/drm/drm_fb_helper.c | 23 +++++++++++++--------- include/drm/drm_client.h | 7 ++++--- 3 files changed, 38 insertions(+), 26 deletions(-)drivers/gpu/drm/drm_client.c +20
-14@@ -235,7 +235,7 @@ static void drm_client_buffer_delete(struct drm_client_buffer *buffer) {struct drm_device *dev = buffer->client->dev; -drm_gem_vunmap (buffer->gem, buffer->vaddr); +drm_gem_vunmap (buffer->gem, &buffer->map);if (buffer->gem)drm_gem_object_put (buffer->gem);@@ -291,25 +291,31 @@ drm_client_buffer_create(struct drm_client_dev *client, u32 width, u32 height, u/** * drm_client_buffer_vmap - Map DRM client buffer into address space * @buffer: DRM client buffer +* @map_copy: Returns the mapped memory's address * * This function maps a client buffer into kernel address space. If the -* buffer is already mapped, it returns the mapping's address. +* buffer is already mapped, it returns the existing mapping's address. * * Client buffer mappings are not ref'counted. Each call to * drm_client_buffer_vmap() should be followed by a call to * drm_client_buffer_vunmap(); or the client buffer should be mapped * throughout its lifetime. * +* The returned address is a copy of the internal value. In contrast to +* other vmap interfaces, you don't need it for the client's vunmap +* function. So you can modify it at will during blit and draw operations. +* * Returns: -* The mapped memory's address +* 0 on success, or a negative errno code otherwise. */ -void *drm_client_buffer_vmap (struct drm_client_buffer *buffer) +int +drm_client_buffer_vmap (struct drm_client_buffer *buffer,struct dma_buf_map *map_copy) { -struct dma_buf_map map; +struct dma_buf_map *map = &buffer->map;int ret; -if (buffer->vaddr) -return buffer->vaddr; +if (dma_buf_map_is_set (map)) + goto out;/* * FIXME: The dependency on GEM here isn't required, we could @@ -319,13 +325,14 @@ void *drm_client_buffer_vmap(struct drm_client_buffer *buffer)* fd_install step out of the driver backend hooks, to make that * final step optional for internal users. */ - ret =drm_gem_vmap (buffer->gem, &map); + ret =drm_gem_vmap (buffer->gem, map);if (ret) -return ERR_PTR (ret); +return ret; - buffer->vaddr = map.vaddr; +out: + *map_copy = *map; -return map.vaddr; +return 0 ; }EXPORT_SYMBOL (drm_client_buffer_vmap);@@ -339,10 +346,9 @@ EXPORT_SYMBOL(drm_client_buffer_vmap);*/ void drm_client_buffer_vunmap (struct drm_client_buffer *buffer) { -struct dma_buf_map map =DMA_BUF_MAP_INIT_VADDR (buffer->vaddr); +struct dma_buf_map *map = &buffer->map; -drm_gem_vunmap (buffer->gem, &map); - buffer->vaddr =NULL ; +drm_gem_vunmap (buffer->gem, map); }EXPORT_SYMBOL (drm_client_buffer_vunmap);drivers/gpu/drm/drm_fb_helper.c +9
-8@@ -378,7 +378,7 @@ static void drm_fb_helper_dirty_blit_real(struct drm_fb_helper *fb_helper,unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->vaddr + offset; +void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y;@@ -400,7 +400,8 @@ static void drm_fb_helper_dirty_work(struct work_struct *work)struct drm_clip_rect *clip = &helper->dirty_clip;struct drm_clip_rect clip_copy;unsigned long flags; -void *vaddr; +struct dma_buf_map map; +int ret;spin_lock_irqsave (&helper->dirty_lock, flags); clip_copy = *clip;@@ -413,8 +414,8 @@ static void drm_fb_helper_dirty_work(struct work_struct *work)/* Generic fbdev uses a shadow buffer */ if (helper->buffer) { - vaddr =drm_client_buffer_vmap (helper->buffer); -if (IS_ERR (vaddr)) + ret =drm_client_buffer_vmap (helper->buffer, &map); +if (ret)return ;drm_fb_helper_dirty_blit_real (helper, &clip_copy); }@@ -2060,7 +2061,8 @@ static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper,struct drm_framebuffer *fb;struct fb_info *fbi;u32 format; -void *vaddr; +struct dma_buf_map map; +int ret; drm_dbg_kms(dev, "surface width(%d), height(%d) and bpp(%d)\n", sizes->surface_width, sizes->surface_height,@@ -2096,11 +2098,14 @@ static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper,fb_deferred_io_init (fbi); }else {/* buffer is mapped for HW framebuffer */ - vaddr =drm_client_buffer_vmap (fb_helper->buffer); -if (IS_ERR (vaddr)) -return PTR_ERR (vaddr); + ret =drm_client_buffer_vmap (fb_helper->buffer, &map); +if (ret)drivers/gpu/drm/drm_fb_helper.c static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper, · line 2111
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +5
-1+return ret; +if (map.is_iomem) + fbi->screen_base = map.vaddr_iomem; +else + fbi->screen_buffer = map.vaddr; - fbi->screen_buffer = vaddr;/* Shamelessly leak the physical address to user-space */ #if IS_ENABLED (CONFIG_DRM_FBDEV_LEAK_PHYS_SMEM )if (drm_leak_fbdev_smem && fbi->fix.smem_start ==0 )Just noticed a tiny thing here: I think this needs to be patched to only set
smem_startwhen the map is _not_ iomem. Sincevirt_to_pageisn't defined on iomem at all.
Quoted 2 lines
I guess it'd be neat if we can set this for iomem too, but I have no idea how to convert an iomem pointer back to a
bus_addr_t...
Not that I disagree, but that should be reviewed by the right people. The commit at 4be9bd10e22d ("drm/fb_helper: Allow leaking fbdev smem_start") appears to work around specific userspace drivers.
Best regards
Thomas
Quoted 40 lines
Cheers, Daniel
Quoted 36 lines
include/drm/drm_client.h +4
-3@@ -3,6 +3,7 @@#ifndef _DRM_CLIENT_H_#define _DRM_CLIENT_H_ +#include <linux/dma-buf-map.h> #include <linux/lockdep.h> #include <linux/mutex.h> #include <linux/types.h> @@ -141,9 +142,9 @@ struct drm_client_buffer {struct drm_gem_object *gem;/** -* @vaddr: Virtual address for the buffer +* @map: Virtual address for the buffer */ -void *vaddr; +struct dma_buf_map map;/** * @fb: DRM framebuffer @@ -155,7 +156,7 @@ struct drm_client_buffer * drm_client_framebuffer_create(struct drm_client_dev *client, u32 width, u32 height, u32 format);void drm_client_framebuffer_delete (struct drm_client_buffer *buffer);int drm_client_framebuffer_flush (struct drm_client_buffer *buffer,struct drm_rect *rect); -void *drm_client_buffer_vmap (struct drm_client_buffer *buffer); +int drm_client_buffer_vmap (struct drm_client_buffer *buffer,struct dma_buf_map *map);void drm_client_buffer_vunmap (struct drm_client_buffer *buffer);int drm_client_modeset_create (struct drm_client_dev *client);Signature · 1 line
2.28.0
Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Re: [PATCH v5 08/10] drm/gem: Store client buffer mappings as struct dma_buf_map
Quoted 173 lines · Thomas Zimmermann
Hi
Quoted 169 lines · Daniel Vetter
Quoted 153 lines · Thomas Zimmermann
Kernel DRM clients now store their framebuffer address in an instance of struct
dma_buf_map. Depending on the buffer's location, the address refers to system or I/O memory.Callers of
drm_client_buffer_vmap()receive a copy of the value in the call's supplied arguments. It can be accessed and modified withdma_buf_mapinterfaces.Signed-off-by Thomas Zimmermann <[email protected]>Reviewed-by Daniel Vetter <[email protected]>Tested-by Sam Ravnborg <[email protected]>--- drivers/gpu/drm/drm_client.c | 34 +++++++++++++++++++-------------- drivers/gpu/drm/drm_fb_helper.c | 23 +++++++++++++--------- include/drm/drm_client.h | 7 ++++--- 3 files changed, 38 insertions(+), 26 deletions(-)drivers/gpu/drm/drm_client.c +20
-14@@ -235,7 +235,7 @@ static void drm_client_buffer_delete(struct drm_client_buffer *buffer) {struct drm_device *dev = buffer->client->dev; -drm_gem_vunmap (buffer->gem, buffer->vaddr); +drm_gem_vunmap (buffer->gem, &buffer->map);if (buffer->gem)drm_gem_object_put (buffer->gem);@@ -291,25 +291,31 @@ drm_client_buffer_create(struct drm_client_dev *client, u32 width, u32 height, u/** * drm_client_buffer_vmap - Map DRM client buffer into address space * @buffer: DRM client buffer +* @map_copy: Returns the mapped memory's address * * This function maps a client buffer into kernel address space. If the -* buffer is already mapped, it returns the mapping's address. +* buffer is already mapped, it returns the existing mapping's address. * * Client buffer mappings are not ref'counted. Each call to * drm_client_buffer_vmap() should be followed by a call to * drm_client_buffer_vunmap(); or the client buffer should be mapped * throughout its lifetime. * +* The returned address is a copy of the internal value. In contrast to +* other vmap interfaces, you don't need it for the client's vunmap +* function. So you can modify it at will during blit and draw operations. +* * Returns: -* The mapped memory's address +* 0 on success, or a negative errno code otherwise. */ -void *drm_client_buffer_vmap (struct drm_client_buffer *buffer) +int +drm_client_buffer_vmap (struct drm_client_buffer *buffer,struct dma_buf_map *map_copy) { -struct dma_buf_map map; +struct dma_buf_map *map = &buffer->map;int ret; -if (buffer->vaddr) -return buffer->vaddr; +if (dma_buf_map_is_set (map)) + goto out;/* * FIXME: The dependency on GEM here isn't required, we could @@ -319,13 +325,14 @@ void *drm_client_buffer_vmap(struct drm_client_buffer *buffer)* fd_install step out of the driver backend hooks, to make that * final step optional for internal users. */ - ret =drm_gem_vmap (buffer->gem, &map); + ret =drm_gem_vmap (buffer->gem, map);if (ret) -return ERR_PTR (ret); +return ret; - buffer->vaddr = map.vaddr; +out: + *map_copy = *map; -return map.vaddr; +return 0 ; }EXPORT_SYMBOL (drm_client_buffer_vmap);@@ -339,10 +346,9 @@ EXPORT_SYMBOL(drm_client_buffer_vmap);*/ void drm_client_buffer_vunmap (struct drm_client_buffer *buffer) { -struct dma_buf_map map =DMA_BUF_MAP_INIT_VADDR (buffer->vaddr); +struct dma_buf_map *map = &buffer->map; -drm_gem_vunmap (buffer->gem, &map); - buffer->vaddr =NULL ; +drm_gem_vunmap (buffer->gem, map); }EXPORT_SYMBOL (drm_client_buffer_vunmap);drivers/gpu/drm/drm_fb_helper.c +9
-8@@ -378,7 +378,7 @@ static void drm_fb_helper_dirty_blit_real(struct drm_fb_helper *fb_helper,unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->vaddr + offset; +void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y;@@ -400,7 +400,8 @@ static void drm_fb_helper_dirty_work(struct work_struct *work)struct drm_clip_rect *clip = &helper->dirty_clip;struct drm_clip_rect clip_copy;unsigned long flags; -void *vaddr; +struct dma_buf_map map; +int ret;spin_lock_irqsave (&helper->dirty_lock, flags); clip_copy = *clip;@@ -413,8 +414,8 @@ static void drm_fb_helper_dirty_work(struct work_struct *work)/* Generic fbdev uses a shadow buffer */ if (helper->buffer) { - vaddr =drm_client_buffer_vmap (helper->buffer); -if (IS_ERR (vaddr)) + ret =drm_client_buffer_vmap (helper->buffer, &map); +if (ret)return ;drm_fb_helper_dirty_blit_real (helper, &clip_copy); }@@ -2060,7 +2061,8 @@ static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper,struct drm_framebuffer *fb;struct fb_info *fbi;u32 format; -void *vaddr; +struct dma_buf_map map; +int ret; drm_dbg_kms(dev, "surface width(%d), height(%d) and bpp(%d)\n", sizes->surface_width, sizes->surface_height,@@ -2096,11 +2098,14 @@ static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper,fb_deferred_io_init (fbi); }else {/* buffer is mapped for HW framebuffer */ - vaddr =drm_client_buffer_vmap (fb_helper->buffer); -if (IS_ERR (vaddr)) -return PTR_ERR (vaddr); + ret =drm_client_buffer_vmap (fb_helper->buffer, &map); +if (ret)drivers/gpu/drm/drm_fb_helper.c static int drm_fb_helper_generic_probe(struct drm_fb_helper *fb_helper, · line 2111
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +5
-1+return ret; +if (map.is_iomem) + fbi->screen_base = map.vaddr_iomem; +else + fbi->screen_buffer = map.vaddr; - fbi->screen_buffer = vaddr;/* Shamelessly leak the physical address to user-space */ #if IS_ENABLED (CONFIG_DRM_FBDEV_LEAK_PHYS_SMEM )if (drm_leak_fbdev_smem && fbi->fix.smem_start ==0 )Just noticed a tiny thing here: I think this needs to be patched to only set
smem_startwhen the map is _not_ iomem. Sincevirt_to_pageisn't defined on iomem at all.
Quoted 6 lines
Quoted 2 lines
I guess it'd be neat if we can set this for iomem too, but I have no idea how to convert an iomem pointer back to a
bus_addr_t...Not that I disagree, but that should be reviewed by the right people. The commit at
4be9bd10e22d("drm/fb_helper: Allow leaking fbdevsmem_start") appears to work around specific userspace drivers.
It's for soc drivers, which all use either shmem or cma helpers, so all system memory. Which means your patch here doesn't break anything. But we need to make sure that if someone enables this it doesn't blow up at least when used on a device where we map iomem. -Daniel
Quoted 51 lines
Best regards
ThomasQuoted 35 lines
Cheers, Daniel
Quoted 32 lines
include/drm/drm_client.h +4
-3@@ -3,6 +3,7 @@#ifndef _DRM_CLIENT_H_#define _DRM_CLIENT_H_ +#include <linux/dma-buf-map.h> #include <linux/lockdep.h> #include <linux/mutex.h> #include <linux/types.h> @@ -141,9 +142,9 @@ struct drm_client_buffer {struct drm_gem_object *gem;/** -* @vaddr: Virtual address for the buffer +* @map: Virtual address for the buffer */ -void *vaddr; +struct dma_buf_map map;/** * @fb: DRM framebuffer @@ -155,7 +156,7 @@ struct drm_client_buffer * drm_client_framebuffer_create(struct drm_client_dev *client, u32 width, u32 height, u32 format);void drm_client_framebuffer_delete (struct drm_client_buffer *buffer);int drm_client_framebuffer_flush (struct drm_client_buffer *buffer,struct drm_rect *rect); -void *drm_client_buffer_vmap (struct drm_client_buffer *buffer); +int drm_client_buffer_vmap (struct drm_client_buffer *buffer,struct dma_buf_map *map);void drm_client_buffer_vunmap (struct drm_client_buffer *buffer);Quoted 5 lines
Quoted 4 lines
include/drm/drm_client.h +0
-1int drm_client_modeset_create (struct drm_client_dev *client); --2.28.0
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Re: [PATCH v5 10/10] drm/fb_helper: Support framebuffers in I/O memory
In reply to Thomas ZimmermannHi Thomas.
Quoted 17 lines · Thomas Zimmermann
At least sparc64 requires I/O-specific access to framebuffers. This patch updates the fbdev console accordingly.
For drivers with direct access to the framebuffer memory, the callback functions in struct
fb_opstest for the type of memory and call the rspfb_sys_offb_cfb_functions. Read and write operations are implemented internally by DRM's fbdev helper.For drivers that employ a shadow buffer, fbdev's blit function retrieves the framebuffer address as struct
dma_buf_map, and usesdma_buf_mapinterfaces to access the buffer.The bochs driver on sparc64 uses a workaround to flag the framebuffer as I/O memory and avoid a HW exception. With the introduction of struct
dma_buf_map, this is not required any longer. The patch removes the rsp code from both, bochs and fbdev.
Quoted 8 lines
v5:
* implement fb_read/fb_write internally (Daniel, Sam)
v4:
* movedma_buf_mapchanges into separate patch (Daniel)
* TODO list: comment on fbdev updates (Daniel)Signed-off-by Thomas Zimmermann <[email protected]>Tested-by Sam Ravnborg <[email protected]>
But see a few comments below on naming for you to consider.
Sam
Quoted 104 lines
--- Documentation/gpu/todo.rst | 19 ++- drivers/gpu/drm/bochs/bochs_kms.c | 1 - drivers/gpu/drm/drm_fb_helper.c | 227 ++++++++++++++++++++++++++++-- include/drm/drm_mode_config.h | 12 -- 4 files changed, 230 insertions(+), 29 deletions(-)Documentation/gpu/todo.rst +17
-2@@ -197,13 +197,28 @@ Convert drivers to use drm_fbdev_generic_setup() ------------------------------------------------ Most drivers can use drm_fbdev_generic_setup(). Driver have to implement -atomic modesetting and GEM vmap support. Current generic fbdev emulation -expects the framebuffer in system memory (or system-like memory). +atomic modesetting and GEM vmap support. Historically, generic fbdev emulation +expected the framebuffer in system memory or system-like memory. By employing +struct dma_buf_map, drivers with frambuffers in I/O memory can be supported +as well. Contact: Maintainer of the driver you plan to convert Level: Intermediate +Reimplement functions in drm_fbdev_fb_ops without fbdev +------------------------------------------------------- + +A number of callback functions in drm_fbdev_fb_ops could benefit from +being rewritten without dependencies on the fbdev module. Some of the +helpers could further benefit from using struct dma_buf_map instead of +raw pointers. + +Contact: Thomas Zimmermann <[email protected]>, Daniel Vetter + +Level: Advanced + + drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup -----------------------------------------------------------------drivers/gpu/drm/bochs/bochs_kms.c +0
-1@@ -151,7 +151,6 @@ int bochs_kms_init(struct bochs_device *bochs) bochs->dev->mode_config.preferred_depth =24 ; bochs->dev->mode_config.prefer_shadow =0 ; bochs->dev->mode_config.prefer_shadow_fbdev =1 ; - bochs->dev->mode_config.fbdev_use_iomem = true; bochs->dev->mode_config.quirk_addfb_prefer_host_byte_order = true; bochs->dev->mode_config.funcs = &bochs_mode_funcs;drivers/gpu/drm/drm_fb_helper.c +8
-9@@ -372,24 +372,22 @@ static void drm_fb_helper_resume_worker(struct work_struct *work) }static void drm_fb_helper_dirty_blit_real (struct drm_fb_helper *fb_helper, -struct drm_clip_rect *clip) +struct drm_clip_rect *clip, +struct dma_buf_map *dst) {struct drm_framebuffer *fb = fb_helper->fb;unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y; -for (y = clip->y1; y < clip->y2; y++) { -if (!fb_helper->dev->mode_config.fbdev_use_iomem) -memcpy (dst, src, len); -else -memcpy_toio ((void __iomem *)dst, src, len); +dma_buf_map_incr (dst, offset);/* go to first pixel within clip rect */ +for (y = clip->y1; y < clip->y2; y++) { +dma_buf_map_memcpy_to (dst, src, len); +dma_buf_map_incr (dst, fb->pitches[0 ]); src += fb->pitches[0 ]; - dst += fb->pitches[0 ]; } }@@ -417,8 +415,9 @@ static void drm_fb_helper_dirty_work(struct work_struct *work) ret =drm_client_buffer_vmap (helper->buffer, &map);if (ret)return ; -drm_fb_helper_dirty_blit_real (helper, &clip_copy); +drm_fb_helper_dirty_blit_real (helper, &clip_copy, &map); } +if (helper->fb->funcs->dirty) helper->fb->funcs->dirty (helper->fb,NULL ,0 ,0 , &clip_copy,1 );@@ -2027,6 +2026,206 @@ static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma)return -ENODEV ; }
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2038
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+static bool drm_fbdev_use_iomem (struct fb_info *info) +{ +struct drm_fb_helper *fb_helper = info->par; +struct drm_client_buffer *buffer = fb_helper->buffer; + +return !drm_fbdev_use_shadow_fb (fb_helper) && buffer->map.is_iomem; +} + +static ssize_t fb_read_screen_base (struct fb_info *info,char __user *buf,size_t count, +loff_t pos)
The naming here confused me - a name like:fb_read_iomem() would have helped me more.
With the current naming I shall remember that the screen_base member is the iomem pointer.
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2048
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+{ +const char __iomem *src = info->screen_base + pos; +size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +char *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; +
I looked around and could not find other places where we copy from iomem to mem to usermem in chunks of PAGE_SIZE.
Quoted 12 lines
drivers/gpu/drm/drm_fb_helper.c +12
-0+while (count) { +size_t c =min (count, alloc_size); + +memcpy_fromio (tmp, src, c); +if (copy_to_user (buf, tmp, c)) { + ret = -EFAULT ; +break ; + } + + src += c; + buf += c; + ret += c;
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2070
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+ count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_read_screen_buffer (struct fb_info *info,char __user *buf,size_t count, +loff_t pos)
And fb_read_sysmem() here.
Quoted 34 lines
drivers/gpu/drm/drm_fb_helper.c +34
-0+{ +const char *src = info->screen_buffer + pos; + +if (copy_to_user (buf, src, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_read (struct fb_info *info,char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos >= total_size) +return 0 ; +if (count >= total_size) + count = total_size; +if (total_size - count < pos) + count = total_size - pos; + +if (drm_fbdev_use_iomem (info)) + ret =fb_read_screen_base (info, buf, count, pos); +else
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2114
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+ ret =fb_read_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +return ret; +} + +static ssize_t fb_write_screen_base (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos)
fb_write_iomem()
Quoted 22 lines
drivers/gpu/drm/drm_fb_helper.c +22
-0+{ +char __iomem *dst = info->screen_base + pos; +size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +u8 *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; + +while (count) { +size_t c =min (count, alloc_size); + +if (copy_from_user (tmp, buf, c)) { + ret = -EFAULT ; +break ; + } +memcpy_toio (dst, tmp, c); + + dst += c; + buf += c; + ret += c;
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2146
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+ count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_write_screen_buffer (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos)
fb_write_sysmem()
Quoted 127 lines
drivers/gpu/drm/drm_fb_helper.c +87
-5+{ +char *dst = info->screen_buffer + pos; + +if (copy_from_user (dst, buf, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_write (struct fb_info *info,const char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; +int err; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos > total_size) +return -EFBIG ; +if (count > total_size) { + err = -EFBIG ; + count = total_size; + } +if (total_size - count < pos) { +if (!err) + err = -ENOSPC ; + count = total_size - pos; + } + +/* +* Copy to framebuffer even if we already logged an error. Emulates +* the behavior of the original fbdev implementation. +*/ +if (drm_fbdev_use_iomem (info)) + ret =fb_write_screen_base (info, buf, count, pos); +else + ret =fb_write_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +if (err) +return err; + +return ret; +} + +static void drm_fbdev_fb_fillrect (struct fb_info *info, +const struct fb_fillrect *rect) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_fillrect (info, rect); +else +drm_fb_helper_sys_fillrect (info, rect); +} + +static void drm_fbdev_fb_copyarea (struct fb_info *info, +const struct fb_copyarea *area) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_copyarea (info, area); +else +drm_fb_helper_sys_copyarea (info, area); +} + +static void drm_fbdev_fb_imageblit (struct fb_info *info, +const struct fb_image *image) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_imageblit (info, image); +else +drm_fb_helper_sys_imageblit (info, image); +} +static const struct fb_ops drm_fbdev_fb_ops = { .owner =THIS_MODULE ,DRM_FB_HELPER_DEFAULT_OPS ,@@ -2034,11 +2233,11 @@ static const struct fb_ops drm_fbdev_fb_ops = { .fb_release = drm_fbdev_fb_release, .fb_destroy = drm_fbdev_fb_destroy, .fb_mmap = drm_fbdev_fb_mmap, - .fb_read = drm_fb_helper_sys_read, - .fb_write = drm_fb_helper_sys_write, - .fb_fillrect = drm_fb_helper_sys_fillrect, - .fb_copyarea = drm_fb_helper_sys_copyarea, - .fb_imageblit = drm_fb_helper_sys_imageblit, + .fb_read = drm_fbdev_fb_read, + .fb_write = drm_fbdev_fb_write, + .fb_fillrect = drm_fbdev_fb_fillrect, + .fb_copyarea = drm_fbdev_fb_copyarea, + .fb_imageblit = drm_fbdev_fb_imageblit, };static struct fb_deferred_io drm_fbdev_defio = {include/drm/drm_mode_config.h +0
-12@@ -877,18 +877,6 @@ struct drm_mode_config {*/ bool prefer_shadow_fbdev; -/** -* @fbdev_use_iomem: -* -* Set to true if framebuffer reside in iomem. -* When set to true memcpy_toio() is used when copying the framebuffer in -* drm_fb_helper.drm_fb_helper_dirty_blit_real(). -* -* FIXME: This should be replaced with a per-mapping is_iomem -* flag (like ttm does), and then used everywhere in fbdev code. -*/ -bool fbdev_use_iomem; -/** * @quirk_addfb_prefer_xbgr_30bpp: * Signature · 1 line
2.28.0
Thomas Zimmermann
maintainerRe: [PATCH v5 10/10] drm/fb_helper: Support framebuffers in I/O memory
Hi
Am 24.10.20 um 22:38 schrieb Sam Ravnborg:
Quoted 138 lines
Hi Thomas.
Quoted 17 lines · Thomas Zimmermann
At least sparc64 requires I/O-specific access to framebuffers. This patch updates the fbdev console accordingly.
For drivers with direct access to the framebuffer memory, the callback functions in struct
fb_opstest for the type of memory and call the rspfb_sys_offb_cfb_functions. Read and write operations are implemented internally by DRM's fbdev helper.For drivers that employ a shadow buffer, fbdev's blit function retrieves the framebuffer address as struct
dma_buf_map, and usesdma_buf_mapinterfaces to access the buffer.The bochs driver on sparc64 uses a workaround to flag the framebuffer as I/O memory and avoid a HW exception. With the introduction of struct
dma_buf_map, this is not required any longer. The patch removes the rsp code from both, bochs and fbdev.Quoted 8 lines
v5:
* implement fb_read/fb_write internally (Daniel, Sam)
v4:
* movedma_buf_mapchanges into separate patch (Daniel)
* TODO list: comment on fbdev updates (Daniel)Signed-off-by Thomas Zimmermann <[email protected]>Tested-by Sam Ravnborg <[email protected]>Reviewed-by Sam Ravnborg <[email protected]>But see a few comments below on naming for you to consider.
Sam
Quoted 104 lines
--- Documentation/gpu/todo.rst | 19 ++- drivers/gpu/drm/bochs/bochs_kms.c | 1 - drivers/gpu/drm/drm_fb_helper.c | 227 ++++++++++++++++++++++++++++-- include/drm/drm_mode_config.h | 12 -- 4 files changed, 230 insertions(+), 29 deletions(-)Documentation/gpu/todo.rst +17
-2@@ -197,13 +197,28 @@ Convert drivers to use drm_fbdev_generic_setup() ------------------------------------------------ Most drivers can use drm_fbdev_generic_setup(). Driver have to implement -atomic modesetting and GEM vmap support. Current generic fbdev emulation -expects the framebuffer in system memory (or system-like memory). +atomic modesetting and GEM vmap support. Historically, generic fbdev emulation +expected the framebuffer in system memory or system-like memory. By employing +struct dma_buf_map, drivers with frambuffers in I/O memory can be supported +as well. Contact: Maintainer of the driver you plan to convert Level: Intermediate +Reimplement functions in drm_fbdev_fb_ops without fbdev +------------------------------------------------------- + +A number of callback functions in drm_fbdev_fb_ops could benefit from +being rewritten without dependencies on the fbdev module. Some of the +helpers could further benefit from using struct dma_buf_map instead of +raw pointers. + +Contact: Thomas Zimmermann <[email protected]>, Daniel Vetter + +Level: Advanced + + drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup -----------------------------------------------------------------drivers/gpu/drm/bochs/bochs_kms.c +0
-1@@ -151,7 +151,6 @@ int bochs_kms_init(struct bochs_device *bochs) bochs->dev->mode_config.preferred_depth =24 ; bochs->dev->mode_config.prefer_shadow =0 ; bochs->dev->mode_config.prefer_shadow_fbdev =1 ; - bochs->dev->mode_config.fbdev_use_iomem = true; bochs->dev->mode_config.quirk_addfb_prefer_host_byte_order = true; bochs->dev->mode_config.funcs = &bochs_mode_funcs;drivers/gpu/drm/drm_fb_helper.c +8
-9@@ -372,24 +372,22 @@ static void drm_fb_helper_resume_worker(struct work_struct *work) }static void drm_fb_helper_dirty_blit_real (struct drm_fb_helper *fb_helper, -struct drm_clip_rect *clip) +struct drm_clip_rect *clip, +struct dma_buf_map *dst) {struct drm_framebuffer *fb = fb_helper->fb;unsigned int cpp = fb->format->cpp[0 ];size_t offset = clip->y1 * fb->pitches[0 ] + clip->x1 * cpp;void *src = fb_helper->fbdev->screen_buffer + offset; -void *dst = fb_helper->buffer->map.vaddr + offset;size_t len = (clip->x2 - clip->x1) * cpp;unsigned int y; -for (y = clip->y1; y < clip->y2; y++) { -if (!fb_helper->dev->mode_config.fbdev_use_iomem) -memcpy (dst, src, len); -else -memcpy_toio ((void __iomem *)dst, src, len); +dma_buf_map_incr (dst, offset);/* go to first pixel within clip rect */ +for (y = clip->y1; y < clip->y2; y++) { +dma_buf_map_memcpy_to (dst, src, len); +dma_buf_map_incr (dst, fb->pitches[0 ]); src += fb->pitches[0 ]; - dst += fb->pitches[0 ]; } }@@ -417,8 +415,9 @@ static void drm_fb_helper_dirty_work(struct work_struct *work) ret =drm_client_buffer_vmap (helper->buffer, &map);if (ret)return ; -drm_fb_helper_dirty_blit_real (helper, &clip_copy); +drm_fb_helper_dirty_blit_real (helper, &clip_copy, &map); } +if (helper->fb->funcs->dirty) helper->fb->funcs->dirty (helper->fb,NULL ,0 ,0 , &clip_copy,1 );@@ -2027,6 +2026,206 @@ static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma)return -ENODEV ; }
Quoted 14 lines
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2038
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+static bool drm_fbdev_use_iomem (struct fb_info *info) +{ +struct drm_fb_helper *fb_helper = info->par; +struct drm_client_buffer *buffer = fb_helper->buffer; + +return !drm_fbdev_use_shadow_fb (fb_helper) && buffer->map.is_iomem; +} + +static ssize_t fb_read_screen_base (struct fb_info *info,char __user *buf,size_t count, +loff_t pos)The naming here confused me - a name like:
fb_read_iomem()would have helped me more.
With the current naming I shall remember that thescreen_basemember is the iomem pointer.
Yeah, true. In terms of naming, I was undecided. I was thinking about adopting a naming similar to what you describe, but OTOH we don't use sysmem anywhere in the code. I thought about adopting fbdev's conention of using _sys_ and _cfb_. But that would make sensein the local context.
Quoted 14 lines
drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2048
Quoted 11 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+{ +const char __iomem *src = info->screen_base + pos; +size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +char *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; +I looked around and could not find other places where we copy from iomem to mem to usermem in chunks of
PAGE_SIZE.
I took this pattern from fbdev's original implementation. I think it's done to work nicely with kmalloc.
Best regards
Thomas
Quoted 238 lines
Quoted 13 lines
drivers/gpu/drm/drm_fb_helper.c +12
-0+while (count) { +size_t c =min (count, alloc_size); + +memcpy_fromio (tmp, src, c); +if (copy_to_user (buf, tmp, c)) { + ret = -EFAULT ; +break ; + } + + src += c; + buf += c; + ret += c;drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2070
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+ count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_read_screen_buffer (struct fb_info *info,char __user *buf,size_t count, +loff_t pos)And
fb_read_sysmem()here.Quoted 34 lines
drivers/gpu/drm/drm_fb_helper.c +34
-0+{ +const char *src = info->screen_buffer + pos; + +if (copy_to_user (buf, src, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_read (struct fb_info *info,char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos >= total_size) +return 0 ; +if (count >= total_size) + count = total_size; +if (total_size - count < pos) + count = total_size - pos; + +if (drm_fbdev_use_iomem (info)) + ret =fb_read_screen_base (info, buf, count, pos); +else drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2114
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+ ret =fb_read_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +return ret; +} + +static ssize_t fb_write_screen_base (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos)
fb_write_iomem()Quoted 22 lines
drivers/gpu/drm/drm_fb_helper.c +22
-0+{ +char __iomem *dst = info->screen_base + pos; +size_t alloc_size =min (count,PAGE_SIZE ); +ssize_t ret =0 ; +u8 *tmp; + + tmp =kmalloc (alloc_size,GFP_KERNEL ); +if (!tmp) +return -ENOMEM ; + +while (count) { +size_t c =min (count, alloc_size); + +if (copy_from_user (tmp, buf, c)) { + ret = -EFAULT ; +break ; + } +memcpy_toio (dst, tmp, c); + + dst += c; + buf += c; + ret += c;drivers/gpu/drm/drm_fb_helper.c static int drm_fbdev_fb_mmap(struct fb_info *info, struct vm_area_struct *vma) · line 2146
Quoted 10 lines
drivers/gpu/drm/drm_fb_helper.c +10
-0+ count -= c; + } + +kfree (tmp); + +return ret; +} + +static ssize_t fb_write_screen_buffer (struct fb_info *info,const char __user *buf,size_t count, +loff_t pos)
fb_write_sysmem()Quoted 127 lines
drivers/gpu/drm/drm_fb_helper.c +87
-5+{ +char *dst = info->screen_buffer + pos; + +if (copy_from_user (dst, buf, count)) +return -EFAULT ; + +return count; +} + +static ssize_t drm_fbdev_fb_write (struct fb_info *info,const char __user *buf, +size_t count,loff_t *ppos) +{ +loff_t pos = *ppos; +size_t total_size; +ssize_t ret; +int err; + +if (info->state !=FBINFO_STATE_RUNNING ) +return -EPERM ; + +if (info->screen_size) + total_size = info->screen_size; +else + total_size = info->fix.smem_len; + +if (pos > total_size) +return -EFBIG ; +if (count > total_size) { + err = -EFBIG ; + count = total_size; + } +if (total_size - count < pos) { +if (!err) + err = -ENOSPC ; + count = total_size - pos; + } + +/* +* Copy to framebuffer even if we already logged an error. Emulates +* the behavior of the original fbdev implementation. +*/ +if (drm_fbdev_use_iomem (info)) + ret =fb_write_screen_base (info, buf, count, pos); +else + ret =fb_write_screen_buffer (info, buf, count, pos); + +if (ret >0 ) + *ppos = ret; + +if (err) +return err; + +return ret; +} + +static void drm_fbdev_fb_fillrect (struct fb_info *info, +const struct fb_fillrect *rect) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_fillrect (info, rect); +else +drm_fb_helper_sys_fillrect (info, rect); +} + +static void drm_fbdev_fb_copyarea (struct fb_info *info, +const struct fb_copyarea *area) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_copyarea (info, area); +else +drm_fb_helper_sys_copyarea (info, area); +} + +static void drm_fbdev_fb_imageblit (struct fb_info *info, +const struct fb_image *image) +{ +if (drm_fbdev_use_iomem (info)) +drm_fb_helper_cfb_imageblit (info, image); +else +drm_fb_helper_sys_imageblit (info, image); +} +static const struct fb_ops drm_fbdev_fb_ops = { .owner =THIS_MODULE ,DRM_FB_HELPER_DEFAULT_OPS ,@@ -2034,11 +2233,11 @@ static const struct fb_ops drm_fbdev_fb_ops = { .fb_release = drm_fbdev_fb_release, .fb_destroy = drm_fbdev_fb_destroy, .fb_mmap = drm_fbdev_fb_mmap, - .fb_read = drm_fb_helper_sys_read, - .fb_write = drm_fb_helper_sys_write, - .fb_fillrect = drm_fb_helper_sys_fillrect, - .fb_copyarea = drm_fb_helper_sys_copyarea, - .fb_imageblit = drm_fb_helper_sys_imageblit, + .fb_read = drm_fbdev_fb_read, + .fb_write = drm_fbdev_fb_write, + .fb_fillrect = drm_fbdev_fb_fillrect, + .fb_copyarea = drm_fbdev_fb_copyarea, + .fb_imageblit = drm_fbdev_fb_imageblit, };static struct fb_deferred_io drm_fbdev_defio = {include/drm/drm_mode_config.h +0
-12@@ -877,18 +877,6 @@ struct drm_mode_config {*/ bool prefer_shadow_fbdev; -/** -* @fbdev_use_iomem: -* -* Set to true if framebuffer reside in iomem. -* When set to true memcpy_toio() is used when copying the framebuffer in -* drm_fb_helper.drm_fb_helper_dirty_blit_real(). -* -* FIXME: This should be replaced with a per-mapping is_iomem -* flag (like ttm does), and then used everywhere in fbdev code. -*/ -bool fbdev_use_iomem; -/** * @quirk_addfb_prefer_xbgr_30bpp: * Signature · 1 line
2.28.0_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Re: [PATCH v5 09/10] dma-buf-map: Add memcpy and pointer-increment interfaces
In reply to Thomas ZimmermannOverall I like this, just an inline question:
Quoted 2 lines · Thomas Zimmermann
To do framebuffer updates, one needs memcpy from system memory and a pointer-increment function. Add both interfaces with documentation.
(...)
Quoted 7 lines
include/linux/dma-buf-map.h +7
-0+/** +* dma_buf_map_memcpy_to - Memcpy into dma-buf mapping +* @dst: The dma-buf mapping structure +* @src: The source buffer +* @len: The number of byte in src +* +* Copies data into a dma-buf mapping. The source buffer is in system
include/linux/dma-buf-map.h static inline void dma_buf_map_clear(struct dma_buf_map *map) · line 248
Quoted 10 lines
include/linux/dma-buf-map.h +10
-0+* memory. Depending on the buffer's location, the helper picks the correct +* method of accessing the memory. +*/ +static inline void dma_buf_map_memcpy_to (struct dma_buf_map *dst,const void *src,size_t len) +{ +if (dst->is_iomem) +memcpy_toio (dst->vaddr_iomem, src, len); +else +memcpy (dst->vaddr, src, len); +}
Are these going to be really big memcpy() operations?
Some platforms have DMA offload engines that can perform memcpy(), drivers/dma, include/linux/dmaengine.h
especially if the CPU doesn't really need to touch the contents and flush caches etc.
An example exist in some MTD drivers that move large quantities of data off flash memory like this:drivers/mtd/nand/raw/cadence-nand-controller.c
Notice that DMAengine and DMAbuf does not have much in common, the names can be deceiving.
The value of this varies with the system architecture. It is not just a question about performance but also about power and the CPU being able to do other stuff in parallel for large transfers. So *when* to use this facility to accelerate memcpy() is a delicate question.
What I'm after here is if these can be really big, do we want (in the long run, not now) open up to the idea to slot in hardware-accelerated memcpy() here?
Yours,
Linus Walleij
Thomas Zimmermann
maintainerRe: [PATCH v5 09/10] dma-buf-map: Add memcpy and pointer-increment interfaces
Hi
Am 05.11.20 um 11:07 schrieb Linus Walleij:
Quoted 19 lines
Overall I like this, just an inline question:
Quoted 2 lines · Thomas Zimmermann
To do framebuffer updates, one needs memcpy from system memory and a pointer-increment function. Add both interfaces with documentation.
(...)
Quoted 11 lines
include/linux/dma-buf-map.h +11
-0+/** +* dma_buf_map_memcpy_to - Memcpy into dma-buf mapping +* @dst: The dma-buf mapping structure +* @src: The source buffer +* @len: The number of byte in src +* +* Copies data into a dma-buf mapping. The source buffer is in system +* memory. Depending on the buffer's location, the helper picks the correct +* method of accessing the memory. +*/ +static inline void dma_buf_map_memcpy_to (struct dma_buf_map *dst,const void *src,size_t len)
Quoted 8 lines
include/linux/dma-buf-map.h static inline void dma_buf_map_clear(struct dma_buf_map *map) · line 248
Quoted 6 lines
include/linux/dma-buf-map.h +6
-0+{ +if (dst->is_iomem) +memcpy_toio (dst->vaddr_iomem, src, len); +else +memcpy (dst->vaddr, src, len); +}Are these going to be really big
memcpy()operations?
Individually, each could be a scanline, so a few KiB. (4 bytes * horizontal resolution). Updating a full framebuffer can sum up to several MiB.
Quoted 12 lines
Some platforms have DMA offload engines that can perform memcpy(),They could be
drivers/dma,include/linux/dmaengine.h
especially if the CPU doesn't really need to touch the contents and flush caches etc.
An example exist in some MTD drivers that move large quantities of data off flash memory like this:drivers/mtd/nand/raw/cadence-nand-controller.cNotice that DMAengine and DMAbuf does not have much in common, the names can be deceiving.
Quoted 8 lines
The value of this varies with the system architecture. It is not just a question about performance but also about power and the CPU being able to do other stuff in parallel for large transfers. So *when* to use this facility to accelerate
memcpy()is a delicate question.What I'm after here is if these can be really big, do we want (in the long run, not now) open up to the idea to slot in hardware-accelerated
memcpy()here?
We currently use this functionality for the graphical framebuffer console that most DRM drivers provide. It's non-accelerated and slow, but this has not been much of a problem so far.
Within DRM, we're more interested in removing console code from drivers and going for the generic implementation.
Most of the graphics HW allocates framebuffers from video RAM, system memory or CMA pools and does not really need these memcpys. Only a few systems with small video RAM require a shadow buffer, which we flush into VRAM as needed. Those might benefit.
OTOH, off-loading memcpys to hardware sounds reasonable if we can hide it from the DRM code. I think it all depends on how invasive that change would be.
Best regards
Thomas
Quoted 4 lines
Yours,
Linus Walleij
Signature · 6 lines
Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Re: [PATCH v5 09/10] dma-buf-map: Add memcpy and pointer-increment interfaces
Quoted 63 lines · Thomas Zimmermann
Hi
Am 05.11.20 um 11:07 schrieb Linus Walleij:
Quoted 19 lines
Overall I like this, just an inline question:
Quoted 2 lines · Thomas Zimmermann
To do framebuffer updates, one needs memcpy from system memory and a pointer-increment function. Add both interfaces with documentation.
(...)
Quoted 11 lines
include/linux/dma-buf-map.h +11
-0+/** +* dma_buf_map_memcpy_to - Memcpy into dma-buf mapping +* @dst: The dma-buf mapping structure +* @src: The source buffer +* @len: The number of byte in src +* +* Copies data into a dma-buf mapping. The source buffer is in system +* memory. Depending on the buffer's location, the helper picks the correct +* method of accessing the memory. +*/ +static inline void dma_buf_map_memcpy_to (struct dma_buf_map *dst,const void *src,size_t len)Quoted 8 lines
include/linux/dma-buf-map.h static inline void dma_buf_map_clear(struct dma_buf_map *map) · line 248
Quoted 6 lines
include/linux/dma-buf-map.h +6
-0+{ +if (dst->is_iomem) +memcpy_toio (dst->vaddr_iomem, src, len); +else +memcpy (dst->vaddr, src, len); +}Are these going to be really big
memcpy()operations?Individually, each could be a scanline, so a few KiB. (4 bytes * horizontal resolution). Updating a full framebuffer can sum up to several MiB.
Quoted 12 lines
Some platforms have DMA offload engines that can perform memcpy(),They could be
drivers/dma,include/linux/dmaengine.h
especially if the CPU doesn't really need to touch the contents and flush caches etc.
An example exist in some MTD drivers that move large quantities of data off flash memory like this:drivers/mtd/nand/raw/cadence-nand-controller.cNotice that DMAengine and DMAbuf does not have much in common, the names can be deceiving.
Quoted 8 lines
The value of this varies with the system architecture. It is not just a question about performance but also about power and the CPU being able to do other stuff in parallel for large transfers. So *when* to use this facility to accelerate
memcpy()is a delicate question.What I'm after here is if these can be really big, do we want (in the long run, not now) open up to the idea to slot in hardware-accelerated
memcpy()here?We currently use this functionality for the graphical framebuffer console that most DRM drivers provide. It's non-accelerated and slow, but this has not been much of a problem so far.
Within DRM, we're more interested in removing console code from drivers and going for the generic implementation.
Quoted 8 lines
Most of the graphics HW allocates framebuffers from video RAM, system memory or CMA pools and does not really need these memcpys. Only a few systems with small video RAM require a shadow buffer, which we flush into VRAM as needed. Those might benefit.
OTOH, off-loading memcpys to hardware sounds reasonable if we can hide it from the DRM code. I think it all depends on how invasive that change would be.
I wouldn't, all the additional locks this would pull in sound like nightmare. And when an oops happens, this might be the only thing that manages to get the oops to the user.
Unless someone really starts caring about fbcon acceleration I really wouldn't bother. Ok maybe it also matters for fbdev, but the problem is that the page fault intercepting alone is already expensive, so the only real solution if you care about performance in that case is to use kms natively, and use a dirty rectangle flip (or the DIRTY syscall).
And in there drivers should (and do) use any dma engines they have to upload the frames already.
-Daniel
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch