| Message ID | 20240614133556.11378-4-tzimmermann@suse.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/{ttm,xe}: Improve ttm_bo_vmap() and update xe | expand |
Am 14.06.24 um 15:21 schrieb Thomas Zimmermann: > Add offset and size parameters to ttm_bo_vmap() to allow for partial > mappings of a buffer object. This brings the functionality on par with > ttm_bo_kmap(). Well the long term plan was to remove this functionality from ttm_bo_kmap() and nuke that function sooner or later. What exactly is the use case for partial mappings? Regards, Christian. > > Callers pass the byte offset and size within the buffer object and > receive a page-aligned mapping of the buffer object's memory for the > specified area. > > Also update all callers of ttm_bo_vmap() for the new parameters. As > before, existing callers map the buffer object's complete memory. > > Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> > --- > drivers/gpu/drm/drm_gem_ttm_helper.c | 2 +- > drivers/gpu/drm/drm_gem_vram_helper.c | 2 +- > drivers/gpu/drm/loongson/lsdc_gem.c | 2 +- > drivers/gpu/drm/qxl/qxl_object.c | 2 +- > drivers/gpu/drm/ttm/ttm_bo_util.c | 21 +++++++++++++++------ > drivers/gpu/drm/xe/xe_lrc.c | 2 +- > drivers/gpu/drm/xe/xe_vm.c | 2 +- > include/drm/ttm/ttm_bo.h | 4 +++- > 8 files changed, 24 insertions(+), 13 deletions(-) > > diff --git a/drivers/gpu/drm/drm_gem_ttm_helper.c b/drivers/gpu/drm/drm_gem_ttm_helper.c > index 3734aa2d1c5b5..f26b7c9077a68 100644 > --- a/drivers/gpu/drm/drm_gem_ttm_helper.c > +++ b/drivers/gpu/drm/drm_gem_ttm_helper.c > @@ -67,7 +67,7 @@ int drm_gem_ttm_vmap(struct drm_gem_object *gem, > { > struct ttm_buffer_object *bo = drm_gem_ttm_of_gem(gem); > > - return ttm_bo_vmap(bo, map); > + return ttm_bo_vmap(bo, 0, gem->size, map); > } > EXPORT_SYMBOL(drm_gem_ttm_vmap); > > diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c b/drivers/gpu/drm/drm_gem_vram_helper.c > index 6027584406af6..1670f9a459a9d 100644 > --- a/drivers/gpu/drm/drm_gem_vram_helper.c > +++ b/drivers/gpu/drm/drm_gem_vram_helper.c > @@ -398,7 +398,7 @@ int drm_gem_vram_vmap(struct drm_gem_vram_object *gbo, struct iosys_map *map) > * no mapping present. > */ > if (iosys_map_is_null(&gbo->map)) { > - ret = ttm_bo_vmap(&gbo->bo, &gbo->map); > + ret = ttm_bo_vmap(&gbo->bo, 0, gbo->bo.base.size, &gbo->map); > if (ret) > return ret; > } > diff --git a/drivers/gpu/drm/loongson/lsdc_gem.c b/drivers/gpu/drm/loongson/lsdc_gem.c > index a720d8f532093..f709960c781b9 100644 > --- a/drivers/gpu/drm/loongson/lsdc_gem.c > +++ b/drivers/gpu/drm/loongson/lsdc_gem.c > @@ -77,7 +77,7 @@ static int lsdc_gem_object_vmap(struct drm_gem_object *obj, struct iosys_map *ma > return ret; > } > > - ret = ttm_bo_vmap(tbo, &lbo->map); > + ret = ttm_bo_vmap(tbo, 0, tbo->base.size, &lbo->map); > if (ret) { > drm_err(obj->dev, "ttm bo vmap failed\n"); > lsdc_bo_unpin(lbo); > diff --git a/drivers/gpu/drm/qxl/qxl_object.c b/drivers/gpu/drm/qxl/qxl_object.c > index 5893e27a7ae50..9f06d5e26a32c 100644 > --- a/drivers/gpu/drm/qxl/qxl_object.c > +++ b/drivers/gpu/drm/qxl/qxl_object.c > @@ -164,7 +164,7 @@ int qxl_bo_vmap_locked(struct qxl_bo *bo, struct iosys_map *map) > goto out; > } > > - r = ttm_bo_vmap(&bo->tbo, &bo->map); > + r = ttm_bo_vmap(&bo->tbo, 0, bo->tbo.base.size, &bo->map); > if (r) { > qxl_bo_unpin_locked(bo); > return r; > diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c > index a9df0deff2deb..31f9772f05dac 100644 > --- a/drivers/gpu/drm/ttm/ttm_bo_util.c > +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c > @@ -457,17 +457,23 @@ EXPORT_SYMBOL(ttm_bo_kunmap); > * ttm_bo_vmap > * > * @bo: The buffer object. > + * @offset: Byte offset into the buffer. > + * @size: Number of bytes to map. > * @map: pointer to a struct iosys_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 iosys_map. Unmap the buffer with ttm_bo_vunmap(). > + * The address stored in @map will be aligned to the next lower page > + * boundaries. > * > * Returns > * -ENOMEM: Out of memory. > * -EINVAL: Invalid range. > */ > -int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) > +int ttm_bo_vmap(struct ttm_buffer_object *bo, > + unsigned long offset, unsigned long size, > + struct iosys_map *map) > { > struct ttm_resource *mem = bo->resource; > int ret; > @@ -483,18 +489,18 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) > u16 alloc_flags; > > if (mem->bus.addr) { > - vaddr_iomem = (void __iomem *)mem->bus.addr; > + vaddr_iomem = (u8 __iomem *)mem->bus.addr + offset; > alloc_flags = ttm_bo_map_premapped; > } else if (mem->bus.caching == ttm_write_combined) { > - vaddr_iomem = ioremap_wc(mem->bus.offset, bo->base.size); > + vaddr_iomem = ioremap_wc(mem->bus.offset + offset, size); > alloc_flags = ttm_bo_map_iomap; > #ifdef CONFIG_X86 > } else if (mem->bus.caching == ttm_cached) { > - vaddr_iomem = ioremap_cache(mem->bus.offset, bo->base.size); > + vaddr_iomem = ioremap_cache(mem->bus.offset + offset, size); > alloc_flags = ttm_bo_map_iomap; > #endif > } else { > - vaddr_iomem = ioremap(mem->bus.offset, bo->base.size); > + vaddr_iomem = ioremap(mem->bus.offset + offset, size); > alloc_flags = ttm_bo_map_iomap; > } > > @@ -510,6 +516,9 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) > .no_wait_gpu = false > }; > struct ttm_tt *ttm = bo->ttm; > + unsigned long start_page = offset >> PAGE_SHIFT; > + unsigned long aligned_size = size + (offset - (start_page << PAGE_SHIFT)); > + unsigned long num_pages = DIV_ROUND_UP(aligned_size, PAGE_SIZE); > pgprot_t prot; > void *vaddr; > u16 alloc_flags; > @@ -523,7 +532,7 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) > * or to make the buffer object look contiguous. > */ > prot = ttm_io_prot(bo, mem, PAGE_KERNEL); > - vaddr = vmap(ttm->pages, ttm->num_pages, 0, prot); > + vaddr = vmap(ttm->pages + start_page, num_pages, 0, prot); > if (!vaddr) > return -ENOMEM; > alloc_flags = ttm_bo_map_vmap; > diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c > index c1bb85d2e243f..3a68fe6d592ed 100644 > --- a/drivers/gpu/drm/xe/xe_lrc.c > +++ b/drivers/gpu/drm/xe/xe_lrc.c > @@ -1595,7 +1595,7 @@ void xe_lrc_snapshot_capture_delayed(struct xe_lrc_snapshot *snapshot) > goto put_bo; > > xe_bo_lock(bo, false); > - if (!ttm_bo_vmap(&bo->ttm, &src)) { > + if (!ttm_bo_vmap(&bo->ttm, 0, bo->ttm.base.size, &src)) { > xe_map_memcpy_from(xe_bo_device(bo), > snapshot->lrc_snapshot, &src, snapshot->lrc_offset, > snapshot->lrc_size); > diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c > index 99bf7412475c0..81306c32f5d09 100644 > --- a/drivers/gpu/drm/xe/xe_vm.c > +++ b/drivers/gpu/drm/xe/xe_vm.c > @@ -3485,7 +3485,7 @@ void xe_vm_snapshot_capture_delayed(struct xe_vm_snapshot *snap) > > if (bo) { > xe_bo_lock(bo, false); > - err = ttm_bo_vmap(&bo->ttm, &src); > + err = ttm_bo_vmap(&bo->ttm, 0, bo->ttm.base.size, &src); > if (!err) { > xe_map_memcpy_from(xe_bo_device(bo), > snap->snap[i].data, > diff --git a/include/drm/ttm/ttm_bo.h b/include/drm/ttm/ttm_bo.h > index 6ccf96c91f3ae..c421ffe3563b1 100644 > --- a/include/drm/ttm/ttm_bo.h > +++ b/include/drm/ttm/ttm_bo.h > @@ -375,7 +375,9 @@ int ttm_bo_init_validate(struct ttm_device *bdev, struct ttm_buffer_object *bo, > int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long start_page, > unsigned long num_pages, struct ttm_bo_kmap_obj *map); > void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map); > -int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map); > +int ttm_bo_vmap(struct ttm_buffer_object *bo, > + unsigned long offset, unsigned long size, > + struct iosys_map *map); > void ttm_bo_vunmap(struct ttm_buffer_object *bo, struct iosys_map *map); > int ttm_bo_mmap_obj(struct vm_area_struct *vma, struct ttm_buffer_object *bo); > int ttm_bo_swapout(struct ttm_buffer_object *bo, struct ttm_operation_ctx *ctx,
Hi Am 14.06.24 um 16:33 schrieb Christian König: > Am 14.06.24 um 15:21 schrieb Thomas Zimmermann: >> Add offset and size parameters to ttm_bo_vmap() to allow for partial >> mappings of a buffer object. This brings the functionality on par with >> ttm_bo_kmap(). > > Well the long term plan was to remove this functionality from > ttm_bo_kmap() and nuke that function sooner or later. > > What exactly is the use case for partial mappings? I don't have a use case; except for streamlining the DRM internals. For the drivers, xe uses partial mappings in i915_gem_object_read_from_page(). But struct iosys_map to map the whole buffer and do memcpys within the range. See patch 6 for that. I think it could do without partial mappings. Vmwgfx uses a local mappings at [1], but as the nearby comment says, this could be a TTM helper. Or maybe TTM could provide a dedicated ttm_bo_kmap_local_page() for temporary mappings like this one. There's also [2], which looks like a candidate for a possible ttm_bo_kmap_local_page(). And in vmwgfx, there's the memcpy at [3], which could be replaced with ttm_bo_vmap() plus iosys_map helpers. I think that's it. Best regards Thomas [1] https://elixir.bootlin.com/linux/v6.9/source/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c#L457 [2] https://elixir.bootlin.com/linux/v6.9/source/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c#L418 [3] https://elixir.bootlin.com/linux/v6.9/source/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c#L430 > > Regards, > Christian. > >> >> Callers pass the byte offset and size within the buffer object and >> receive a page-aligned mapping of the buffer object's memory for the >> specified area. >> >> Also update all callers of ttm_bo_vmap() for the new parameters. As >> before, existing callers map the buffer object's complete memory. >> >> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> >> --- >> drivers/gpu/drm/drm_gem_ttm_helper.c | 2 +- >> drivers/gpu/drm/drm_gem_vram_helper.c | 2 +- >> drivers/gpu/drm/loongson/lsdc_gem.c | 2 +- >> drivers/gpu/drm/qxl/qxl_object.c | 2 +- >> drivers/gpu/drm/ttm/ttm_bo_util.c | 21 +++++++++++++++------ >> drivers/gpu/drm/xe/xe_lrc.c | 2 +- >> drivers/gpu/drm/xe/xe_vm.c | 2 +- >> include/drm/ttm/ttm_bo.h | 4 +++- >> 8 files changed, 24 insertions(+), 13 deletions(-) >> >> diff --git a/drivers/gpu/drm/drm_gem_ttm_helper.c >> b/drivers/gpu/drm/drm_gem_ttm_helper.c >> index 3734aa2d1c5b5..f26b7c9077a68 100644 >> --- a/drivers/gpu/drm/drm_gem_ttm_helper.c >> +++ b/drivers/gpu/drm/drm_gem_ttm_helper.c >> @@ -67,7 +67,7 @@ int drm_gem_ttm_vmap(struct drm_gem_object *gem, >> { >> struct ttm_buffer_object *bo = drm_gem_ttm_of_gem(gem); >> - return ttm_bo_vmap(bo, map); >> + return ttm_bo_vmap(bo, 0, gem->size, map); >> } >> EXPORT_SYMBOL(drm_gem_ttm_vmap); >> diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c >> b/drivers/gpu/drm/drm_gem_vram_helper.c >> index 6027584406af6..1670f9a459a9d 100644 >> --- a/drivers/gpu/drm/drm_gem_vram_helper.c >> +++ b/drivers/gpu/drm/drm_gem_vram_helper.c >> @@ -398,7 +398,7 @@ int drm_gem_vram_vmap(struct drm_gem_vram_object >> *gbo, struct iosys_map *map) >> * no mapping present. >> */ >> if (iosys_map_is_null(&gbo->map)) { >> - ret = ttm_bo_vmap(&gbo->bo, &gbo->map); >> + ret = ttm_bo_vmap(&gbo->bo, 0, gbo->bo.base.size, &gbo->map); >> if (ret) >> return ret; >> } >> diff --git a/drivers/gpu/drm/loongson/lsdc_gem.c >> b/drivers/gpu/drm/loongson/lsdc_gem.c >> index a720d8f532093..f709960c781b9 100644 >> --- a/drivers/gpu/drm/loongson/lsdc_gem.c >> +++ b/drivers/gpu/drm/loongson/lsdc_gem.c >> @@ -77,7 +77,7 @@ static int lsdc_gem_object_vmap(struct >> drm_gem_object *obj, struct iosys_map *ma >> return ret; >> } >> - ret = ttm_bo_vmap(tbo, &lbo->map); >> + ret = ttm_bo_vmap(tbo, 0, tbo->base.size, &lbo->map); >> if (ret) { >> drm_err(obj->dev, "ttm bo vmap failed\n"); >> lsdc_bo_unpin(lbo); >> diff --git a/drivers/gpu/drm/qxl/qxl_object.c >> b/drivers/gpu/drm/qxl/qxl_object.c >> index 5893e27a7ae50..9f06d5e26a32c 100644 >> --- a/drivers/gpu/drm/qxl/qxl_object.c >> +++ b/drivers/gpu/drm/qxl/qxl_object.c >> @@ -164,7 +164,7 @@ int qxl_bo_vmap_locked(struct qxl_bo *bo, struct >> iosys_map *map) >> goto out; >> } >> - r = ttm_bo_vmap(&bo->tbo, &bo->map); >> + r = ttm_bo_vmap(&bo->tbo, 0, bo->tbo.base.size, &bo->map); >> if (r) { >> qxl_bo_unpin_locked(bo); >> return r; >> diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c >> b/drivers/gpu/drm/ttm/ttm_bo_util.c >> index a9df0deff2deb..31f9772f05dac 100644 >> --- a/drivers/gpu/drm/ttm/ttm_bo_util.c >> +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c >> @@ -457,17 +457,23 @@ EXPORT_SYMBOL(ttm_bo_kunmap); >> * ttm_bo_vmap >> * >> * @bo: The buffer object. >> + * @offset: Byte offset into the buffer. >> + * @size: Number of bytes to map. >> * @map: pointer to a struct iosys_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 iosys_map. Unmap the buffer with ttm_bo_vunmap(). >> + * The address stored in @map will be aligned to the next lower page >> + * boundaries. >> * >> * Returns >> * -ENOMEM: Out of memory. >> * -EINVAL: Invalid range. >> */ >> -int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) >> +int ttm_bo_vmap(struct ttm_buffer_object *bo, >> + unsigned long offset, unsigned long size, >> + struct iosys_map *map) >> { >> struct ttm_resource *mem = bo->resource; >> int ret; >> @@ -483,18 +489,18 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, >> struct iosys_map *map) >> u16 alloc_flags; >> if (mem->bus.addr) { >> - vaddr_iomem = (void __iomem *)mem->bus.addr; >> + vaddr_iomem = (u8 __iomem *)mem->bus.addr + offset; >> alloc_flags = ttm_bo_map_premapped; >> } else if (mem->bus.caching == ttm_write_combined) { >> - vaddr_iomem = ioremap_wc(mem->bus.offset, bo->base.size); >> + vaddr_iomem = ioremap_wc(mem->bus.offset + offset, size); >> alloc_flags = ttm_bo_map_iomap; >> #ifdef CONFIG_X86 >> } else if (mem->bus.caching == ttm_cached) { >> - vaddr_iomem = ioremap_cache(mem->bus.offset, >> bo->base.size); >> + vaddr_iomem = ioremap_cache(mem->bus.offset + offset, >> size); >> alloc_flags = ttm_bo_map_iomap; >> #endif >> } else { >> - vaddr_iomem = ioremap(mem->bus.offset, bo->base.size); >> + vaddr_iomem = ioremap(mem->bus.offset + offset, size); >> alloc_flags = ttm_bo_map_iomap; >> } >> @@ -510,6 +516,9 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, >> struct iosys_map *map) >> .no_wait_gpu = false >> }; >> struct ttm_tt *ttm = bo->ttm; >> + unsigned long start_page = offset >> PAGE_SHIFT; >> + unsigned long aligned_size = size + (offset - (start_page << >> PAGE_SHIFT)); >> + unsigned long num_pages = DIV_ROUND_UP(aligned_size, >> PAGE_SIZE); >> pgprot_t prot; >> void *vaddr; >> u16 alloc_flags; >> @@ -523,7 +532,7 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, >> struct iosys_map *map) >> * or to make the buffer object look contiguous. >> */ >> prot = ttm_io_prot(bo, mem, PAGE_KERNEL); >> - vaddr = vmap(ttm->pages, ttm->num_pages, 0, prot); >> + vaddr = vmap(ttm->pages + start_page, num_pages, 0, prot); >> if (!vaddr) >> return -ENOMEM; >> alloc_flags = ttm_bo_map_vmap; >> diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c >> index c1bb85d2e243f..3a68fe6d592ed 100644 >> --- a/drivers/gpu/drm/xe/xe_lrc.c >> +++ b/drivers/gpu/drm/xe/xe_lrc.c >> @@ -1595,7 +1595,7 @@ void xe_lrc_snapshot_capture_delayed(struct >> xe_lrc_snapshot *snapshot) >> goto put_bo; >> xe_bo_lock(bo, false); >> - if (!ttm_bo_vmap(&bo->ttm, &src)) { >> + if (!ttm_bo_vmap(&bo->ttm, 0, bo->ttm.base.size, &src)) { >> xe_map_memcpy_from(xe_bo_device(bo), >> snapshot->lrc_snapshot, &src, snapshot->lrc_offset, >> snapshot->lrc_size); >> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c >> index 99bf7412475c0..81306c32f5d09 100644 >> --- a/drivers/gpu/drm/xe/xe_vm.c >> +++ b/drivers/gpu/drm/xe/xe_vm.c >> @@ -3485,7 +3485,7 @@ void xe_vm_snapshot_capture_delayed(struct >> xe_vm_snapshot *snap) >> if (bo) { >> xe_bo_lock(bo, false); >> - err = ttm_bo_vmap(&bo->ttm, &src); >> + err = ttm_bo_vmap(&bo->ttm, 0, bo->ttm.base.size, &src); >> if (!err) { >> xe_map_memcpy_from(xe_bo_device(bo), >> snap->snap[i].data, >> diff --git a/include/drm/ttm/ttm_bo.h b/include/drm/ttm/ttm_bo.h >> index 6ccf96c91f3ae..c421ffe3563b1 100644 >> --- a/include/drm/ttm/ttm_bo.h >> +++ b/include/drm/ttm/ttm_bo.h >> @@ -375,7 +375,9 @@ int ttm_bo_init_validate(struct ttm_device *bdev, >> struct ttm_buffer_object *bo, >> int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long >> start_page, >> unsigned long num_pages, struct ttm_bo_kmap_obj *map); >> void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map); >> -int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map); >> +int ttm_bo_vmap(struct ttm_buffer_object *bo, >> + unsigned long offset, unsigned long size, >> + struct iosys_map *map); >> void ttm_bo_vunmap(struct ttm_buffer_object *bo, struct iosys_map >> *map); >> int ttm_bo_mmap_obj(struct vm_area_struct *vma, struct >> ttm_buffer_object *bo); >> int ttm_bo_swapout(struct ttm_buffer_object *bo, struct >> ttm_operation_ctx *ctx, >
diff --git a/drivers/gpu/drm/drm_gem_ttm_helper.c b/drivers/gpu/drm/drm_gem_ttm_helper.c index 3734aa2d1c5b5..f26b7c9077a68 100644 --- a/drivers/gpu/drm/drm_gem_ttm_helper.c +++ b/drivers/gpu/drm/drm_gem_ttm_helper.c @@ -67,7 +67,7 @@ int drm_gem_ttm_vmap(struct drm_gem_object *gem, { struct ttm_buffer_object *bo = drm_gem_ttm_of_gem(gem); - return ttm_bo_vmap(bo, map); + return ttm_bo_vmap(bo, 0, gem->size, map); } EXPORT_SYMBOL(drm_gem_ttm_vmap); diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c b/drivers/gpu/drm/drm_gem_vram_helper.c index 6027584406af6..1670f9a459a9d 100644 --- a/drivers/gpu/drm/drm_gem_vram_helper.c +++ b/drivers/gpu/drm/drm_gem_vram_helper.c @@ -398,7 +398,7 @@ int drm_gem_vram_vmap(struct drm_gem_vram_object *gbo, struct iosys_map *map) * no mapping present. */ if (iosys_map_is_null(&gbo->map)) { - ret = ttm_bo_vmap(&gbo->bo, &gbo->map); + ret = ttm_bo_vmap(&gbo->bo, 0, gbo->bo.base.size, &gbo->map); if (ret) return ret; } diff --git a/drivers/gpu/drm/loongson/lsdc_gem.c b/drivers/gpu/drm/loongson/lsdc_gem.c index a720d8f532093..f709960c781b9 100644 --- a/drivers/gpu/drm/loongson/lsdc_gem.c +++ b/drivers/gpu/drm/loongson/lsdc_gem.c @@ -77,7 +77,7 @@ static int lsdc_gem_object_vmap(struct drm_gem_object *obj, struct iosys_map *ma return ret; } - ret = ttm_bo_vmap(tbo, &lbo->map); + ret = ttm_bo_vmap(tbo, 0, tbo->base.size, &lbo->map); if (ret) { drm_err(obj->dev, "ttm bo vmap failed\n"); lsdc_bo_unpin(lbo); diff --git a/drivers/gpu/drm/qxl/qxl_object.c b/drivers/gpu/drm/qxl/qxl_object.c index 5893e27a7ae50..9f06d5e26a32c 100644 --- a/drivers/gpu/drm/qxl/qxl_object.c +++ b/drivers/gpu/drm/qxl/qxl_object.c @@ -164,7 +164,7 @@ int qxl_bo_vmap_locked(struct qxl_bo *bo, struct iosys_map *map) goto out; } - r = ttm_bo_vmap(&bo->tbo, &bo->map); + r = ttm_bo_vmap(&bo->tbo, 0, bo->tbo.base.size, &bo->map); if (r) { qxl_bo_unpin_locked(bo); return r; diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c index a9df0deff2deb..31f9772f05dac 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_util.c +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c @@ -457,17 +457,23 @@ EXPORT_SYMBOL(ttm_bo_kunmap); * ttm_bo_vmap * * @bo: The buffer object. + * @offset: Byte offset into the buffer. + * @size: Number of bytes to map. * @map: pointer to a struct iosys_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 iosys_map. Unmap the buffer with ttm_bo_vunmap(). + * The address stored in @map will be aligned to the next lower page + * boundaries. * * Returns * -ENOMEM: Out of memory. * -EINVAL: Invalid range. */ -int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) +int ttm_bo_vmap(struct ttm_buffer_object *bo, + unsigned long offset, unsigned long size, + struct iosys_map *map) { struct ttm_resource *mem = bo->resource; int ret; @@ -483,18 +489,18 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) u16 alloc_flags; if (mem->bus.addr) { - vaddr_iomem = (void __iomem *)mem->bus.addr; + vaddr_iomem = (u8 __iomem *)mem->bus.addr + offset; alloc_flags = ttm_bo_map_premapped; } else if (mem->bus.caching == ttm_write_combined) { - vaddr_iomem = ioremap_wc(mem->bus.offset, bo->base.size); + vaddr_iomem = ioremap_wc(mem->bus.offset + offset, size); alloc_flags = ttm_bo_map_iomap; #ifdef CONFIG_X86 } else if (mem->bus.caching == ttm_cached) { - vaddr_iomem = ioremap_cache(mem->bus.offset, bo->base.size); + vaddr_iomem = ioremap_cache(mem->bus.offset + offset, size); alloc_flags = ttm_bo_map_iomap; #endif } else { - vaddr_iomem = ioremap(mem->bus.offset, bo->base.size); + vaddr_iomem = ioremap(mem->bus.offset + offset, size); alloc_flags = ttm_bo_map_iomap; } @@ -510,6 +516,9 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) .no_wait_gpu = false }; struct ttm_tt *ttm = bo->ttm; + unsigned long start_page = offset >> PAGE_SHIFT; + unsigned long aligned_size = size + (offset - (start_page << PAGE_SHIFT)); + unsigned long num_pages = DIV_ROUND_UP(aligned_size, PAGE_SIZE); pgprot_t prot; void *vaddr; u16 alloc_flags; @@ -523,7 +532,7 @@ int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map) * or to make the buffer object look contiguous. */ prot = ttm_io_prot(bo, mem, PAGE_KERNEL); - vaddr = vmap(ttm->pages, ttm->num_pages, 0, prot); + vaddr = vmap(ttm->pages + start_page, num_pages, 0, prot); if (!vaddr) return -ENOMEM; alloc_flags = ttm_bo_map_vmap; diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c index c1bb85d2e243f..3a68fe6d592ed 100644 --- a/drivers/gpu/drm/xe/xe_lrc.c +++ b/drivers/gpu/drm/xe/xe_lrc.c @@ -1595,7 +1595,7 @@ void xe_lrc_snapshot_capture_delayed(struct xe_lrc_snapshot *snapshot) goto put_bo; xe_bo_lock(bo, false); - if (!ttm_bo_vmap(&bo->ttm, &src)) { + if (!ttm_bo_vmap(&bo->ttm, 0, bo->ttm.base.size, &src)) { xe_map_memcpy_from(xe_bo_device(bo), snapshot->lrc_snapshot, &src, snapshot->lrc_offset, snapshot->lrc_size); diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index 99bf7412475c0..81306c32f5d09 100644 --- a/drivers/gpu/drm/xe/xe_vm.c +++ b/drivers/gpu/drm/xe/xe_vm.c @@ -3485,7 +3485,7 @@ void xe_vm_snapshot_capture_delayed(struct xe_vm_snapshot *snap) if (bo) { xe_bo_lock(bo, false); - err = ttm_bo_vmap(&bo->ttm, &src); + err = ttm_bo_vmap(&bo->ttm, 0, bo->ttm.base.size, &src); if (!err) { xe_map_memcpy_from(xe_bo_device(bo), snap->snap[i].data, diff --git a/include/drm/ttm/ttm_bo.h b/include/drm/ttm/ttm_bo.h index 6ccf96c91f3ae..c421ffe3563b1 100644 --- a/include/drm/ttm/ttm_bo.h +++ b/include/drm/ttm/ttm_bo.h @@ -375,7 +375,9 @@ int ttm_bo_init_validate(struct ttm_device *bdev, struct ttm_buffer_object *bo, int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long start_page, unsigned long num_pages, struct ttm_bo_kmap_obj *map); void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map); -int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map); +int ttm_bo_vmap(struct ttm_buffer_object *bo, + unsigned long offset, unsigned long size, + struct iosys_map *map); void ttm_bo_vunmap(struct ttm_buffer_object *bo, struct iosys_map *map); int ttm_bo_mmap_obj(struct vm_area_struct *vma, struct ttm_buffer_object *bo); int ttm_bo_swapout(struct ttm_buffer_object *bo, struct ttm_operation_ctx *ctx,
Add offset and size parameters to ttm_bo_vmap() to allow for partial mappings of a buffer object. This brings the functionality on par with ttm_bo_kmap(). Callers pass the byte offset and size within the buffer object and receive a page-aligned mapping of the buffer object's memory for the specified area. Also update all callers of ttm_bo_vmap() for the new parameters. As before, existing callers map the buffer object's complete memory. Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> --- drivers/gpu/drm/drm_gem_ttm_helper.c | 2 +- drivers/gpu/drm/drm_gem_vram_helper.c | 2 +- drivers/gpu/drm/loongson/lsdc_gem.c | 2 +- drivers/gpu/drm/qxl/qxl_object.c | 2 +- drivers/gpu/drm/ttm/ttm_bo_util.c | 21 +++++++++++++++------ drivers/gpu/drm/xe/xe_lrc.c | 2 +- drivers/gpu/drm/xe/xe_vm.c | 2 +- include/drm/ttm/ttm_bo.h | 4 +++- 8 files changed, 24 insertions(+), 13 deletions(-)