| Message ID | 20210515145359.64802-3-paul@crapouillou.net (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Add option to mmap GEM buffers cached | expand |
Hi Am 15.05.21 um 16:53 schrieb Paul Cercueil: > This function can be used by drivers that use damage clips and have > CMA GEM objects backed by non-coherent memory. Calling this function > in a plane's .atomic_update ensures that all the data in the backing > memory have been written to RAM. > > v3: - Only sync data if using GEM objects backed by non-coherent memory. > - Use a drm_device pointer instead of device pointer in prototype > > Signed-off-by: Paul Cercueil <paul@crapouillou.net> > --- > drivers/gpu/drm/drm_gem_cma_helper.c | 55 ++++++++++++++++++++++++++++ > include/drm/drm_gem_cma_helper.h | 5 +++ > 2 files changed, 60 insertions(+) > > diff --git a/drivers/gpu/drm/drm_gem_cma_helper.c b/drivers/gpu/drm/drm_gem_cma_helper.c > index 235c7a63da2b..41f309e0e049 100644 > --- a/drivers/gpu/drm/drm_gem_cma_helper.c > +++ b/drivers/gpu/drm/drm_gem_cma_helper.c > @@ -17,9 +17,14 @@ > #include <linux/slab.h> > > #include <drm/drm.h> > +#include <drm/drm_damage_helper.h> > #include <drm/drm_device.h> > #include <drm/drm_drv.h> > +#include <drm/drm_fourcc.h> > +#include <drm/drm_fb_cma_helper.h> Alphabetical order: fb < fourcc > +#include <drm/drm_framebuffer.h> > #include <drm/drm_gem_cma_helper.h> > +#include <drm/drm_plane.h> > #include <drm/drm_vma_manager.h> > > /** > @@ -576,3 +581,53 @@ drm_gem_cma_prime_import_sg_table_vmap(struct drm_device *dev, > return obj; > } > EXPORT_SYMBOL(drm_gem_cma_prime_import_sg_table_vmap); > + > +/** > + * drm_gem_cma_sync_data - Sync GEM object to non-coherent backing memory > + * @drm: DRM device > + * @old_state: Old plane state > + * @state: New plane state > + * > + * This function can be used by drivers that use damage clips and have > + * CMA GEM objects backed by non-coherent memory. Calling this function > + * in a plane's .atomic_update ensures that all the data in the backing > + * memory have been written to RAM. > + */ > +void drm_gem_cma_sync_data(struct drm_device *drm, > + struct drm_plane_state *old_state, > + struct drm_plane_state *state) > +{ > + const struct drm_format_info *finfo = state->fb->format; > + struct drm_atomic_helper_damage_iter iter; > + const struct drm_gem_cma_object *cma_obj; > + unsigned int offset, i; > + struct drm_rect clip; > + dma_addr_t daddr; > + > + for (i = 0; i < finfo->num_planes; i++) { > + cma_obj = drm_fb_cma_get_gem_obj(state->fb, i); > + > + if (cma_obj->map_noncoherent) > + break; > + } > + > + /* No non-coherent buffers - no need to sync anything. */ > + if (i == finfo->num_planes) > + return; > + > + drm_atomic_helper_damage_iter_init(&iter, old_state, state); > + > + drm_atomic_for_each_plane_damage(&iter, &clip) { > + for (i = 0; i < finfo->num_planes; i++) { > + daddr = drm_fb_cma_get_gem_addr(state->fb, state, i); > + > + /* Ignore x1/x2 values, invalidate complete lines */ > + offset = clip.y1 * state->fb->pitches[i]; > + > + dma_sync_single_for_device(drm->dev, daddr + offset, > + (clip.y2 - clip.y1) * state->fb->pitches[i], > + DMA_TO_DEVICE); A framebuffer can have multiple BOs with different coherency. The current loop syncs every BO, but you only have to sync non-coherent memory. I suggest to merge the above test loop into this sync loop, such that only non-coherent BOs get synced damage_iter_init(iter) for_each_damage_plane(iter) { for (i < finfo->num_planes) { cma_obj = drm_fb_cma_get_gem_obj(i) if (!cma_obj->non_coherent) continue; dma_sync_single_for_device() } } For cache locality, it might be better to exchange the loops: for (i < finfo->num_planes) { damage_iter_init(iter) for_each_damage_plane(iter) { } } This way, you operate on the BOs one by one. > + } > + } > +} > +EXPORT_SYMBOL_GPL(drm_gem_cma_sync_data); > diff --git a/include/drm/drm_gem_cma_helper.h b/include/drm/drm_gem_cma_helper.h > index cd13508acbc1..76af066ae3a7 100644 > --- a/include/drm/drm_gem_cma_helper.h > +++ b/include/drm/drm_gem_cma_helper.h > @@ -7,6 +7,7 @@ > #include <drm/drm_gem.h> > > struct drm_mode_create_dumb; > +struct drm_plane_state; > > /** > * struct drm_gem_cma_object - GEM object backed by CMA memory allocations > @@ -185,4 +186,8 @@ drm_gem_cma_prime_import_sg_table_vmap(struct drm_device *drm, > struct dma_buf_attachment *attach, > struct sg_table *sgt); > > +void drm_gem_cma_sync_data(struct drm_device *drm, > + struct drm_plane_state *old_state, > + struct drm_plane_state *state); > + Maybe call this function drm_gem_cma_sync_non_coherent() so that it's clear what the sync is about. Best regards Thomas > #endif /* __DRM_GEM_CMA_HELPER_H__ */ >
diff --git a/drivers/gpu/drm/drm_gem_cma_helper.c b/drivers/gpu/drm/drm_gem_cma_helper.c index 235c7a63da2b..41f309e0e049 100644 --- a/drivers/gpu/drm/drm_gem_cma_helper.c +++ b/drivers/gpu/drm/drm_gem_cma_helper.c @@ -17,9 +17,14 @@ #include <linux/slab.h> #include <drm/drm.h> +#include <drm/drm_damage_helper.h> #include <drm/drm_device.h> #include <drm/drm_drv.h> +#include <drm/drm_fourcc.h> +#include <drm/drm_fb_cma_helper.h> +#include <drm/drm_framebuffer.h> #include <drm/drm_gem_cma_helper.h> +#include <drm/drm_plane.h> #include <drm/drm_vma_manager.h> /** @@ -576,3 +581,53 @@ drm_gem_cma_prime_import_sg_table_vmap(struct drm_device *dev, return obj; } EXPORT_SYMBOL(drm_gem_cma_prime_import_sg_table_vmap); + +/** + * drm_gem_cma_sync_data - Sync GEM object to non-coherent backing memory + * @drm: DRM device + * @old_state: Old plane state + * @state: New plane state + * + * This function can be used by drivers that use damage clips and have + * CMA GEM objects backed by non-coherent memory. Calling this function + * in a plane's .atomic_update ensures that all the data in the backing + * memory have been written to RAM. + */ +void drm_gem_cma_sync_data(struct drm_device *drm, + struct drm_plane_state *old_state, + struct drm_plane_state *state) +{ + const struct drm_format_info *finfo = state->fb->format; + struct drm_atomic_helper_damage_iter iter; + const struct drm_gem_cma_object *cma_obj; + unsigned int offset, i; + struct drm_rect clip; + dma_addr_t daddr; + + for (i = 0; i < finfo->num_planes; i++) { + cma_obj = drm_fb_cma_get_gem_obj(state->fb, i); + + if (cma_obj->map_noncoherent) + break; + } + + /* No non-coherent buffers - no need to sync anything. */ + if (i == finfo->num_planes) + return; + + drm_atomic_helper_damage_iter_init(&iter, old_state, state); + + drm_atomic_for_each_plane_damage(&iter, &clip) { + for (i = 0; i < finfo->num_planes; i++) { + daddr = drm_fb_cma_get_gem_addr(state->fb, state, i); + + /* Ignore x1/x2 values, invalidate complete lines */ + offset = clip.y1 * state->fb->pitches[i]; + + dma_sync_single_for_device(drm->dev, daddr + offset, + (clip.y2 - clip.y1) * state->fb->pitches[i], + DMA_TO_DEVICE); + } + } +} +EXPORT_SYMBOL_GPL(drm_gem_cma_sync_data); diff --git a/include/drm/drm_gem_cma_helper.h b/include/drm/drm_gem_cma_helper.h index cd13508acbc1..76af066ae3a7 100644 --- a/include/drm/drm_gem_cma_helper.h +++ b/include/drm/drm_gem_cma_helper.h @@ -7,6 +7,7 @@ #include <drm/drm_gem.h> struct drm_mode_create_dumb; +struct drm_plane_state; /** * struct drm_gem_cma_object - GEM object backed by CMA memory allocations @@ -185,4 +186,8 @@ drm_gem_cma_prime_import_sg_table_vmap(struct drm_device *drm, struct dma_buf_attachment *attach, struct sg_table *sgt); +void drm_gem_cma_sync_data(struct drm_device *drm, + struct drm_plane_state *old_state, + struct drm_plane_state *state); + #endif /* __DRM_GEM_CMA_HELPER_H__ */
This function can be used by drivers that use damage clips and have CMA GEM objects backed by non-coherent memory. Calling this function in a plane's .atomic_update ensures that all the data in the backing memory have been written to RAM. v3: - Only sync data if using GEM objects backed by non-coherent memory. - Use a drm_device pointer instead of device pointer in prototype Signed-off-by: Paul Cercueil <paul@crapouillou.net> --- drivers/gpu/drm/drm_gem_cma_helper.c | 55 ++++++++++++++++++++++++++++ include/drm/drm_gem_cma_helper.h | 5 +++ 2 files changed, 60 insertions(+)