| Message ID | 1473450533-20869-1-git-send-email-david.s.gordon@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, On 09/09/16 20:48, Dave Gordon wrote: > This just hides the existing obj->dirty flag inside a trivial inline > setter, to discourage non-GEM code from looking too closely. The > flag is renamed to emphasise that it is private to the GEM memory- > management code and ensure that no legacy code continues to use it > directly. > > v2: > Use Chris Wilson's preferred names for flag-related functions > > Inspired-by: http://www.spinics.net/lists/intel-gfx/msg92390.html > Cc: Chris Wilson <chris@chris-wilson.co.uk> > Signed-off-by: Dave Gordon <david.s.gordon@intel.com> > --- > drivers/gpu/drm/i915/i915_debugfs.c | 2 +- > drivers/gpu/drm/i915/i915_drv.h | 22 +++++++++++++++++++++- > drivers/gpu/drm/i915/i915_gem.c | 25 ++++++++++++++----------- > drivers/gpu/drm/i915/i915_gem_context.c | 7 +++++-- > drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- > drivers/gpu/drm/i915/i915_gem_userptr.c | 12 +++++++----- > drivers/gpu/drm/i915/i915_gpu_error.c | 2 +- > drivers/gpu/drm/i915/intel_lrc.c | 29 ++++++++++++++++------------- > 8 files changed, 66 insertions(+), 35 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c > index 02b627e..b77fc27 100644 > --- a/drivers/gpu/drm/i915/i915_debugfs.c > +++ b/drivers/gpu/drm/i915/i915_debugfs.c > @@ -160,7 +160,7 @@ static u64 i915_gem_obj_total_ggtt_size(struct drm_i915_gem_object *obj) > i915_gem_active_get_seqno(&obj->last_write, > &obj->base.dev->struct_mutex), > i915_cache_level_str(dev_priv, obj->cache_level), > - obj->dirty ? " dirty" : "", > + i915_gem_object_is_dirty(obj) ? " dirty" : "", > obj->madv == I915_MADV_DONTNEED ? " purgeable" : ""); > if (obj->base.name) > seq_printf(m, " (name: %d)", obj->base.name); > diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h > index f39bede..333e21b 100644 > --- a/drivers/gpu/drm/i915/i915_drv.h > +++ b/drivers/gpu/drm/i915/i915_drv.h > @@ -2209,7 +2209,7 @@ struct drm_i915_gem_object { > * This is set if the object has been written to since last bound > * to the GTT > */ > - unsigned int dirty:1; > + unsigned int __dirty:1; > > /** > * Advice: are the backing pages purgeable? > @@ -3156,6 +3156,26 @@ static inline void i915_gem_object_pin_pages(struct drm_i915_gem_object *obj) > obj->pages_pin_count++; > } > > +/* > + * Flag the object content as having changed since the last call to > + * i915_gem_object_pin_pages() above, so that the new content is not > + * lost after the next call to i915_gem_object_unpin_pages() below > + */ > +static inline void i915_gem_object_set_dirty(struct drm_i915_gem_object *obj) > +{ > + obj->__dirty = true; > +} > + > +static inline void i915_gem_object_clear_dirty(struct drm_i915_gem_object *obj) > +{ > + obj->__dirty = false; > +} > + > +static inline bool i915_gem_object_is_dirty(struct drm_i915_gem_object *obj) > +{ > + return obj->__dirty; > +} > + > static inline void i915_gem_object_unpin_pages(struct drm_i915_gem_object *obj) > { > BUG_ON(obj->pages_pin_count == 0); > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c > index 2401818..f571a02 100644 > --- a/drivers/gpu/drm/i915/i915_gem.c > +++ b/drivers/gpu/drm/i915/i915_gem.c > @@ -234,9 +234,8 @@ int i915_mutex_lock_interruptible(struct drm_device *dev) > } > > if (obj->madv == I915_MADV_DONTNEED) > - obj->dirty = 0; > - > - if (obj->dirty) { > + i915_gem_object_clear_dirty(obj); > + else if (i915_gem_object_is_dirty(obj)) { > struct address_space *mapping = obj->base.filp->f_mapping; > char *vaddr = obj->phys_handle->vaddr; > int i; > @@ -260,7 +259,7 @@ int i915_mutex_lock_interruptible(struct drm_device *dev) > put_page(page); > vaddr += PAGE_SIZE; > } > - obj->dirty = 0; > + i915_gem_object_clear_dirty(obj); > } > > sg_free_table(obj->pages); > @@ -703,7 +702,7 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, > obj->cache_dirty = true; > > intel_fb_obj_invalidate(obj, ORIGIN_CPU); > - obj->dirty = 1; > + i915_gem_object_set_dirty(obj); > /* return with the pages pinned */ > return 0; > > @@ -1156,7 +1155,7 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, > goto out_unpin; I wonder why diff got so confused with this one, because this isn't i915_gem_obj_prepare_shmem_write any longer. > > intel_fb_obj_invalidate(obj, ORIGIN_CPU); > - obj->dirty = true; > + i915_gem_object_set_dirty(obj); > > user_data = u64_to_user_ptr(args->data_ptr); > offset = args->offset; > @@ -1327,6 +1326,8 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, > offset = args->offset; > remain = args->size; > > + i915_gem_object_set_dirty(obj); > + This is in i915_gem_shmem_pwrite by the look of it. It will have dirtied the object already via i915_gem_obj_prepare_shmem_write. You added it here for some specific reason? > for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, > offset >> PAGE_SHIFT) { > struct page *page = sg_page_iter_page(&sg_iter); > @@ -2133,6 +2134,7 @@ static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) > { > struct sgt_iter sgt_iter; > struct page *page; > + bool dirty; > int ret; > > BUG_ON(obj->madv == __I915_MADV_PURGED); > @@ -2152,10 +2154,11 @@ static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) > i915_gem_object_save_bit_17_swizzle(obj); > > if (obj->madv == I915_MADV_DONTNEED) > - obj->dirty = 0; > + i915_gem_object_clear_dirty(obj); > > + dirty = i915_gem_object_is_dirty(obj); > for_each_sgt_page(page, sgt_iter, obj->pages) { > - if (obj->dirty) > + if (dirty) > set_page_dirty(page); > > if (obj->madv == I915_MADV_WILLNEED) > @@ -2163,7 +2166,7 @@ static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) > > put_page(page); > } > - obj->dirty = 0; > + i915_gem_object_clear_dirty(obj); > > sg_free_table(obj->pages); > kfree(obj->pages); > @@ -3321,7 +3324,7 @@ static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj) > if (write) { > obj->base.read_domains = I915_GEM_DOMAIN_GTT; > obj->base.write_domain = I915_GEM_DOMAIN_GTT; > - obj->dirty = 1; > + i915_gem_object_set_dirty(obj); > } > > trace_i915_gem_object_change_domain(obj, > @@ -4789,7 +4792,7 @@ struct drm_i915_gem_object * > i915_gem_object_pin_pages(obj); > sg = obj->pages; > bytes = sg_copy_from_buffer(sg->sgl, sg->nents, (void *)data, size); > - obj->dirty = 1; /* Backing store is now out of date */ > + i915_gem_object_set_dirty(obj); /* Backing store is now out of date */ > i915_gem_object_unpin_pages(obj); > > if (WARN_ON(bytes != size)) { > diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c > index 35950ee..aa99bc3 100644 > --- a/drivers/gpu/drm/i915/i915_gem_context.c > +++ b/drivers/gpu/drm/i915/i915_gem_context.c > @@ -834,6 +834,8 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) > * MI_SET_CONTEXT instead of when the next seqno has completed. > */ > if (from != NULL) { > + struct i915_vma *from_vma = from->engine[RCS].state; > + > /* As long as MI_SET_CONTEXT is serializing, ie. it flushes the > * whole damn pipeline, we don't need to explicitly mark the > * object dirty. The only exception is that the context must be > @@ -841,9 +843,10 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) > * able to defer doing this until we know the object would be > * swapped, but there is no way to do that yet. > */ > - i915_vma_move_to_active(from->engine[RCS].state, req, 0); > + i915_vma_move_to_active(from_vma, req, 0); > + i915_gem_object_set_dirty(from_vma->obj); Commit message should mention this. > /* state is kept alive until the next request */ > - i915_vma_unpin(from->engine[RCS].state); > + i915_vma_unpin(from_vma); > i915_gem_context_put(from); > } > engine->last_context = i915_gem_context_get(to); > diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c > index 9432d4c..5ca7ba3 100644 > --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c > +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c > @@ -1278,7 +1278,7 @@ void i915_vma_move_to_active(struct i915_vma *vma, > > GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); > > - obj->dirty = 1; /* be paranoid */ > + i915_gem_object_set_dirty(obj); /* be paranoid */ > > /* Add a reference if we're newly entering the active list. > * The order in which we add operations to the retirement queue is > diff --git a/drivers/gpu/drm/i915/i915_gem_userptr.c b/drivers/gpu/drm/i915/i915_gem_userptr.c > index be54825..a78abe2 100644 > --- a/drivers/gpu/drm/i915/i915_gem_userptr.c > +++ b/drivers/gpu/drm/i915/i915_gem_userptr.c > @@ -674,23 +674,25 @@ struct get_pages_work { > { > struct sgt_iter sgt_iter; > struct page *page; > + bool dirty; > > BUG_ON(obj->userptr.work != NULL); > __i915_gem_userptr_set_active(obj, false); > > - if (obj->madv != I915_MADV_WILLNEED) > - obj->dirty = 0; > - > i915_gem_gtt_finish_object(obj); > > + if (obj->madv != I915_MADV_WILLNEED) > + i915_gem_object_clear_dirty(obj); > + Looks like harmless reordering. > + dirty = i915_gem_object_is_dirty(obj); > for_each_sgt_page(page, sgt_iter, obj->pages) { > - if (obj->dirty) > + if (dirty) > set_page_dirty(page); > > mark_page_accessed(page); > put_page(page); > } > - obj->dirty = 0; > + i915_gem_object_clear_dirty(obj); > > sg_free_table(obj->pages); > kfree(obj->pages); > diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c > index 334f15d..257dde1 100644 > --- a/drivers/gpu/drm/i915/i915_gpu_error.c > +++ b/drivers/gpu/drm/i915/i915_gpu_error.c > @@ -804,7 +804,7 @@ static void capture_bo(struct drm_i915_error_buffer *err, > err->write_domain = obj->base.write_domain; > err->fence_reg = vma->fence ? vma->fence->id : -1; > err->tiling = i915_gem_object_get_tiling(obj); > - err->dirty = obj->dirty; > + err->dirty = i915_gem_object_is_dirty(obj); > err->purgeable = obj->madv != I915_MADV_WILLNEED; > err->userptr = obj->userptr.mm != NULL; > err->cache_level = obj->cache_level; > diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c > index 92bfe47..1f48291 100644 > --- a/drivers/gpu/drm/i915/intel_lrc.c > +++ b/drivers/gpu/drm/i915/intel_lrc.c > @@ -764,6 +764,8 @@ static int intel_lr_context_pin(struct i915_gem_context *ctx, > struct intel_engine_cs *engine) > { > struct intel_context *ce = &ctx->engine[engine->id]; > + struct intel_ring *ring = ce->ring; > + struct drm_i915_gem_object *ctx_obj; > void *vaddr; > u32 *lrc_reg_state; > int ret; > @@ -778,24 +780,24 @@ static int intel_lr_context_pin(struct i915_gem_context *ctx, > if (ret) > goto err; > > - vaddr = i915_gem_object_pin_map(ce->state->obj, I915_MAP_WB); > + ctx_obj = ce->state->obj; > + vaddr = i915_gem_object_pin_map(ctx_obj, I915_MAP_WB); > if (IS_ERR(vaddr)) { > ret = PTR_ERR(vaddr); > goto unpin_vma; > } > > - lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; > - > - ret = intel_ring_pin(ce->ring); > + ret = intel_ring_pin(ring); > if (ret) > goto unpin_map; > > - intel_lr_context_descriptor_update(ctx, engine); > + i915_gem_object_set_dirty(ctx_obj); > > - lrc_reg_state[CTX_RING_BUFFER_START+1] = > - i915_ggtt_offset(ce->ring->vma); > + lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; > + lrc_reg_state[CTX_RING_BUFFER_START+1] = i915_ggtt_offset(ring->vma); > ce->lrc_reg_state = lrc_reg_state; > - ce->state->obj->dirty = true; > + > + intel_lr_context_descriptor_update(ctx, engine); > > /* Invalidate GuC TLB. */ > if (i915.enable_guc_submission) { > @@ -1969,7 +1971,7 @@ static u32 intel_lr_indirect_ctx_offset(struct intel_engine_cs *engine) > DRM_DEBUG_DRIVER("Could not map object pages! (%d)\n", ret); > return ret; > } > - ctx_obj->dirty = true; > + i915_gem_object_set_dirty(ctx_obj); > > /* The second page of the context object contains some fields which must > * be set up prior to the first execution. */ > @@ -2182,23 +2184,24 @@ void intel_lr_context_reset(struct drm_i915_private *dev_priv, > > for_each_engine(engine, dev_priv) { > struct intel_context *ce = &ctx->engine[engine->id]; > + struct drm_i915_gem_object *ctx_obj; > void *vaddr; > uint32_t *reg_state; > > if (!ce->state) > continue; > > - vaddr = i915_gem_object_pin_map(ce->state->obj, I915_MAP_WB); > + ctx_obj = ce->state->obj; > + vaddr = i915_gem_object_pin_map(ctx_obj, I915_MAP_WB); > if (WARN_ON(IS_ERR(vaddr))) > continue; > > reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; > - > reg_state[CTX_RING_HEAD+1] = 0; > reg_state[CTX_RING_TAIL+1] = 0; > > - ce->state->obj->dirty = true; > - i915_gem_object_unpin_map(ce->state->obj); > + i915_gem_object_set_dirty(ctx_obj); > + i915_gem_object_unpin_map(ctx_obj); > > ce->ring->head = 0; > ce->ring->tail = 0; > Looks good to me. Just a small commit message extension required and clarification on added extra dirtying. Regards, Tvrtko
On Mon, Sep 12, 2016 at 04:48:51PM +0100, Tvrtko Ursulin wrote: > > * able to defer doing this until we know the object would be > > * swapped, but there is no way to do that yet. > > */ > >- i915_vma_move_to_active(from->engine[RCS].state, req, 0); > >+ i915_vma_move_to_active(from_vma, req, 0); > >+ i915_gem_object_set_dirty(from_vma->obj); > > Commit message should mention this. No. This should not be here as we removed it because we always mark the object as dirty when active on the GPU. I don't see any improvement in this patch over mine, so what's the reason for it? Certainly the lack of debug checks is odd. -Chris
On 12/09/16 16:48, Tvrtko Ursulin wrote: > > Hi, > > On 09/09/16 20:48, Dave Gordon wrote: >> This just hides the existing obj->dirty flag inside a trivial inline >> setter, to discourage non-GEM code from looking too closely. The >> flag is renamed to emphasise that it is private to the GEM memory- >> management code and ensure that no legacy code continues to use it >> directly. >> >> v2: >> Use Chris Wilson's preferred names for flag-related functions >> >> Inspired-by: http://www.spinics.net/lists/intel-gfx/msg92390.html >> Cc: Chris Wilson <chris@chris-wilson.co.uk> >> Signed-off-by: Dave Gordon <david.s.gordon@intel.com> >> --- >> drivers/gpu/drm/i915/i915_debugfs.c | 2 +- >> drivers/gpu/drm/i915/i915_drv.h | 22 +++++++++++++++++++++- >> drivers/gpu/drm/i915/i915_gem.c | 25 >> ++++++++++++++----------- >> drivers/gpu/drm/i915/i915_gem_context.c | 7 +++++-- >> drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- >> drivers/gpu/drm/i915/i915_gem_userptr.c | 12 +++++++----- >> drivers/gpu/drm/i915/i915_gpu_error.c | 2 +- >> drivers/gpu/drm/i915/intel_lrc.c | 29 >> ++++++++++++++++------------- >> 8 files changed, 66 insertions(+), 35 deletions(-) >> >> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c >> b/drivers/gpu/drm/i915/i915_debugfs.c >> index 02b627e..b77fc27 100644 >> --- a/drivers/gpu/drm/i915/i915_debugfs.c >> +++ b/drivers/gpu/drm/i915/i915_debugfs.c >> @@ -160,7 +160,7 @@ static u64 i915_gem_obj_total_ggtt_size(struct >> drm_i915_gem_object *obj) >> i915_gem_active_get_seqno(&obj->last_write, >> &obj->base.dev->struct_mutex), >> i915_cache_level_str(dev_priv, obj->cache_level), >> - obj->dirty ? " dirty" : "", >> + i915_gem_object_is_dirty(obj) ? " dirty" : "", >> obj->madv == I915_MADV_DONTNEED ? " purgeable" : ""); >> if (obj->base.name) >> seq_printf(m, " (name: %d)", obj->base.name); >> diff --git a/drivers/gpu/drm/i915/i915_drv.h >> b/drivers/gpu/drm/i915/i915_drv.h >> index f39bede..333e21b 100644 >> --- a/drivers/gpu/drm/i915/i915_drv.h >> +++ b/drivers/gpu/drm/i915/i915_drv.h >> @@ -2209,7 +2209,7 @@ struct drm_i915_gem_object { >> * This is set if the object has been written to since last bound >> * to the GTT >> */ >> - unsigned int dirty:1; >> + unsigned int __dirty:1; >> >> /** >> * Advice: are the backing pages purgeable? >> @@ -3156,6 +3156,26 @@ static inline void >> i915_gem_object_pin_pages(struct drm_i915_gem_object *obj) >> obj->pages_pin_count++; >> } >> >> +/* >> + * Flag the object content as having changed since the last call to >> + * i915_gem_object_pin_pages() above, so that the new content is not >> + * lost after the next call to i915_gem_object_unpin_pages() below >> + */ >> +static inline void i915_gem_object_set_dirty(struct >> drm_i915_gem_object *obj) >> +{ >> + obj->__dirty = true; >> +} >> + >> +static inline void i915_gem_object_clear_dirty(struct >> drm_i915_gem_object *obj) >> +{ >> + obj->__dirty = false; >> +} >> + >> +static inline bool i915_gem_object_is_dirty(struct >> drm_i915_gem_object *obj) >> +{ >> + return obj->__dirty; >> +} >> + >> static inline void i915_gem_object_unpin_pages(struct >> drm_i915_gem_object *obj) >> { >> BUG_ON(obj->pages_pin_count == 0); >> diff --git a/drivers/gpu/drm/i915/i915_gem.c >> b/drivers/gpu/drm/i915/i915_gem.c >> index 2401818..f571a02 100644 >> --- a/drivers/gpu/drm/i915/i915_gem.c >> +++ b/drivers/gpu/drm/i915/i915_gem.c >> @@ -234,9 +234,8 @@ int i915_mutex_lock_interruptible(struct >> drm_device *dev) >> } >> >> if (obj->madv == I915_MADV_DONTNEED) >> - obj->dirty = 0; >> - >> - if (obj->dirty) { >> + i915_gem_object_clear_dirty(obj); >> + else if (i915_gem_object_is_dirty(obj)) { >> struct address_space *mapping = obj->base.filp->f_mapping; >> char *vaddr = obj->phys_handle->vaddr; >> int i; >> @@ -260,7 +259,7 @@ int i915_mutex_lock_interruptible(struct >> drm_device *dev) >> put_page(page); >> vaddr += PAGE_SIZE; >> } >> - obj->dirty = 0; >> + i915_gem_object_clear_dirty(obj); >> } >> >> sg_free_table(obj->pages); >> @@ -703,7 +702,7 @@ int i915_gem_obj_prepare_shmem_write(struct >> drm_i915_gem_object *obj, >> obj->cache_dirty = true; >> >> intel_fb_obj_invalidate(obj, ORIGIN_CPU); >> - obj->dirty = 1; >> + i915_gem_object_set_dirty(obj); >> /* return with the pages pinned */ >> return 0; >> >> @@ -1156,7 +1155,7 @@ int i915_gem_obj_prepare_shmem_write(struct >> drm_i915_gem_object *obj, >> goto out_unpin; > > I wonder why diff got so confused with this one, because this isn't > i915_gem_obj_prepare_shmem_write any longer. It has to do with functions containing labels. A workaround that sometimes works is to tell git-diff that it's C++ code rather than C, as it then handles labels slightly differently, in a way that usually happens to fix the misidentification of which function the code is in. >> intel_fb_obj_invalidate(obj, ORIGIN_CPU); >> - obj->dirty = true; >> + i915_gem_object_set_dirty(obj); >> >> user_data = u64_to_user_ptr(args->data_ptr); >> offset = args->offset; >> @@ -1327,6 +1326,8 @@ int i915_gem_obj_prepare_shmem_write(struct >> drm_i915_gem_object *obj, >> offset = args->offset; >> remain = args->size; >> >> + i915_gem_object_set_dirty(obj); >> + > > This is in i915_gem_shmem_pwrite by the look of it. It will have dirtied > the object already via i915_gem_obj_prepare_shmem_write. You added it > here for some specific reason? Nope, I think it must have been left over from an earlier version in which not every path to here had already set it. I'll get rid of it. >> for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, >> offset >> PAGE_SHIFT) { >> struct page *page = sg_page_iter_page(&sg_iter); >> @@ -2133,6 +2134,7 @@ static void >> i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) >> { >> struct sgt_iter sgt_iter; >> struct page *page; >> + bool dirty; >> int ret; >> >> BUG_ON(obj->madv == __I915_MADV_PURGED); >> @@ -2152,10 +2154,11 @@ static void >> i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) >> i915_gem_object_save_bit_17_swizzle(obj); >> >> if (obj->madv == I915_MADV_DONTNEED) >> - obj->dirty = 0; >> + i915_gem_object_clear_dirty(obj); >> >> + dirty = i915_gem_object_is_dirty(obj); >> for_each_sgt_page(page, sgt_iter, obj->pages) { >> - if (obj->dirty) >> + if (dirty) >> set_page_dirty(page); >> >> if (obj->madv == I915_MADV_WILLNEED) >> @@ -2163,7 +2166,7 @@ static void >> i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) >> >> put_page(page); >> } >> - obj->dirty = 0; >> + i915_gem_object_clear_dirty(obj); >> >> sg_free_table(obj->pages); >> kfree(obj->pages); >> @@ -3321,7 +3324,7 @@ static void >> i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj) >> if (write) { >> obj->base.read_domains = I915_GEM_DOMAIN_GTT; >> obj->base.write_domain = I915_GEM_DOMAIN_GTT; >> - obj->dirty = 1; >> + i915_gem_object_set_dirty(obj); >> } >> >> trace_i915_gem_object_change_domain(obj, >> @@ -4789,7 +4792,7 @@ struct drm_i915_gem_object * >> i915_gem_object_pin_pages(obj); >> sg = obj->pages; >> bytes = sg_copy_from_buffer(sg->sgl, sg->nents, (void *)data, >> size); >> - obj->dirty = 1; /* Backing store is now out of date */ >> + i915_gem_object_set_dirty(obj); /* Backing store is now out of >> date */ >> i915_gem_object_unpin_pages(obj); >> >> if (WARN_ON(bytes != size)) { >> diff --git a/drivers/gpu/drm/i915/i915_gem_context.c >> b/drivers/gpu/drm/i915/i915_gem_context.c >> index 35950ee..aa99bc3 100644 >> --- a/drivers/gpu/drm/i915/i915_gem_context.c >> +++ b/drivers/gpu/drm/i915/i915_gem_context.c >> @@ -834,6 +834,8 @@ static int do_rcs_switch(struct >> drm_i915_gem_request *req) >> * MI_SET_CONTEXT instead of when the next seqno has completed. >> */ >> if (from != NULL) { >> + struct i915_vma *from_vma = from->engine[RCS].state; >> + >> /* As long as MI_SET_CONTEXT is serializing, ie. it flushes the >> * whole damn pipeline, we don't need to explicitly mark the >> * object dirty. The only exception is that the context must be >> @@ -841,9 +843,10 @@ static int do_rcs_switch(struct >> drm_i915_gem_request *req) >> * able to defer doing this until we know the object would be >> * swapped, but there is no way to do that yet. >> */ >> - i915_vma_move_to_active(from->engine[RCS].state, req, 0); >> + i915_vma_move_to_active(from_vma, req, 0); >> + i915_gem_object_set_dirty(from_vma->obj); > > Commit message should mention this. Hmm .. that wasn't supposed to be there, I think that's also left from a previous iteration. *This* patch was supposed to be just (nearly-) mechanical transformations. >> /* state is kept alive until the next request */ >> - i915_vma_unpin(from->engine[RCS].state); >> + i915_vma_unpin(from_vma); >> i915_gem_context_put(from); >> } >> engine->last_context = i915_gem_context_get(to); >> diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c >> b/drivers/gpu/drm/i915/i915_gem_execbuffer.c >> index 9432d4c..5ca7ba3 100644 >> --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c >> +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c >> @@ -1278,7 +1278,7 @@ void i915_vma_move_to_active(struct i915_vma *vma, >> >> GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); >> >> - obj->dirty = 1; /* be paranoid */ >> + i915_gem_object_set_dirty(obj); /* be paranoid */ >> >> /* Add a reference if we're newly entering the active list. >> * The order in which we add operations to the retirement queue is >> diff --git a/drivers/gpu/drm/i915/i915_gem_userptr.c >> b/drivers/gpu/drm/i915/i915_gem_userptr.c >> index be54825..a78abe2 100644 >> --- a/drivers/gpu/drm/i915/i915_gem_userptr.c >> +++ b/drivers/gpu/drm/i915/i915_gem_userptr.c >> @@ -674,23 +674,25 @@ struct get_pages_work { >> { >> struct sgt_iter sgt_iter; >> struct page *page; >> + bool dirty; >> >> BUG_ON(obj->userptr.work != NULL); >> __i915_gem_userptr_set_active(obj, false); >> >> - if (obj->madv != I915_MADV_WILLNEED) >> - obj->dirty = 0; >> - >> i915_gem_gtt_finish_object(obj); >> >> + if (obj->madv != I915_MADV_WILLNEED) >> + i915_gem_object_clear_dirty(obj); >> + > > Looks like harmless reordering. Yes, it might help the compiler if we keep all the dirty-related code close together. Besides, I don't want to make any assumption about what i915_gem_gtt_finish_object(obj) will do e.g. it could change obj->madv! .Dave. >> + dirty = i915_gem_object_is_dirty(obj); >> for_each_sgt_page(page, sgt_iter, obj->pages) { >> - if (obj->dirty) >> + if (dirty) >> set_page_dirty(page); >> >> mark_page_accessed(page); >> put_page(page); >> } >> - obj->dirty = 0; >> + i915_gem_object_clear_dirty(obj); >> >> sg_free_table(obj->pages); >> kfree(obj->pages); >> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c >> b/drivers/gpu/drm/i915/i915_gpu_error.c >> index 334f15d..257dde1 100644 >> --- a/drivers/gpu/drm/i915/i915_gpu_error.c >> +++ b/drivers/gpu/drm/i915/i915_gpu_error.c >> @@ -804,7 +804,7 @@ static void capture_bo(struct >> drm_i915_error_buffer *err, >> err->write_domain = obj->base.write_domain; >> err->fence_reg = vma->fence ? vma->fence->id : -1; >> err->tiling = i915_gem_object_get_tiling(obj); >> - err->dirty = obj->dirty; >> + err->dirty = i915_gem_object_is_dirty(obj); >> err->purgeable = obj->madv != I915_MADV_WILLNEED; >> err->userptr = obj->userptr.mm != NULL; >> err->cache_level = obj->cache_level; >> diff --git a/drivers/gpu/drm/i915/intel_lrc.c >> b/drivers/gpu/drm/i915/intel_lrc.c >> index 92bfe47..1f48291 100644 >> --- a/drivers/gpu/drm/i915/intel_lrc.c >> +++ b/drivers/gpu/drm/i915/intel_lrc.c >> @@ -764,6 +764,8 @@ static int intel_lr_context_pin(struct >> i915_gem_context *ctx, >> struct intel_engine_cs *engine) >> { >> struct intel_context *ce = &ctx->engine[engine->id]; >> + struct intel_ring *ring = ce->ring; >> + struct drm_i915_gem_object *ctx_obj; >> void *vaddr; >> u32 *lrc_reg_state; >> int ret; >> @@ -778,24 +780,24 @@ static int intel_lr_context_pin(struct >> i915_gem_context *ctx, >> if (ret) >> goto err; >> >> - vaddr = i915_gem_object_pin_map(ce->state->obj, I915_MAP_WB); >> + ctx_obj = ce->state->obj; >> + vaddr = i915_gem_object_pin_map(ctx_obj, I915_MAP_WB); >> if (IS_ERR(vaddr)) { >> ret = PTR_ERR(vaddr); >> goto unpin_vma; >> } >> >> - lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; >> - >> - ret = intel_ring_pin(ce->ring); >> + ret = intel_ring_pin(ring); >> if (ret) >> goto unpin_map; >> >> - intel_lr_context_descriptor_update(ctx, engine); >> + i915_gem_object_set_dirty(ctx_obj); >> >> - lrc_reg_state[CTX_RING_BUFFER_START+1] = >> - i915_ggtt_offset(ce->ring->vma); >> + lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; >> + lrc_reg_state[CTX_RING_BUFFER_START+1] = >> i915_ggtt_offset(ring->vma); >> ce->lrc_reg_state = lrc_reg_state; >> - ce->state->obj->dirty = true; >> + >> + intel_lr_context_descriptor_update(ctx, engine); >> >> /* Invalidate GuC TLB. */ >> if (i915.enable_guc_submission) { >> @@ -1969,7 +1971,7 @@ static u32 intel_lr_indirect_ctx_offset(struct >> intel_engine_cs *engine) >> DRM_DEBUG_DRIVER("Could not map object pages! (%d)\n", ret); >> return ret; >> } >> - ctx_obj->dirty = true; >> + i915_gem_object_set_dirty(ctx_obj); >> >> /* The second page of the context object contains some fields >> which must >> * be set up prior to the first execution. */ >> @@ -2182,23 +2184,24 @@ void intel_lr_context_reset(struct >> drm_i915_private *dev_priv, >> >> for_each_engine(engine, dev_priv) { >> struct intel_context *ce = &ctx->engine[engine->id]; >> + struct drm_i915_gem_object *ctx_obj; >> void *vaddr; >> uint32_t *reg_state; >> >> if (!ce->state) >> continue; >> >> - vaddr = i915_gem_object_pin_map(ce->state->obj, I915_MAP_WB); >> + ctx_obj = ce->state->obj; >> + vaddr = i915_gem_object_pin_map(ctx_obj, I915_MAP_WB); >> if (WARN_ON(IS_ERR(vaddr))) >> continue; >> >> reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; >> - >> reg_state[CTX_RING_HEAD+1] = 0; >> reg_state[CTX_RING_TAIL+1] = 0; >> >> - ce->state->obj->dirty = true; >> - i915_gem_object_unpin_map(ce->state->obj); >> + i915_gem_object_set_dirty(ctx_obj); >> + i915_gem_object_unpin_map(ctx_obj); >> >> ce->ring->head = 0; >> ce->ring->tail = 0; >> > > Looks good to me. Just a small commit message extension required and > clarification on added extra dirtying. > > Regards, > Tvrtko
diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c index 02b627e..b77fc27 100644 --- a/drivers/gpu/drm/i915/i915_debugfs.c +++ b/drivers/gpu/drm/i915/i915_debugfs.c @@ -160,7 +160,7 @@ static u64 i915_gem_obj_total_ggtt_size(struct drm_i915_gem_object *obj) i915_gem_active_get_seqno(&obj->last_write, &obj->base.dev->struct_mutex), i915_cache_level_str(dev_priv, obj->cache_level), - obj->dirty ? " dirty" : "", + i915_gem_object_is_dirty(obj) ? " dirty" : "", obj->madv == I915_MADV_DONTNEED ? " purgeable" : ""); if (obj->base.name) seq_printf(m, " (name: %d)", obj->base.name); diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index f39bede..333e21b 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -2209,7 +2209,7 @@ struct drm_i915_gem_object { * This is set if the object has been written to since last bound * to the GTT */ - unsigned int dirty:1; + unsigned int __dirty:1; /** * Advice: are the backing pages purgeable? @@ -3156,6 +3156,26 @@ static inline void i915_gem_object_pin_pages(struct drm_i915_gem_object *obj) obj->pages_pin_count++; } +/* + * Flag the object content as having changed since the last call to + * i915_gem_object_pin_pages() above, so that the new content is not + * lost after the next call to i915_gem_object_unpin_pages() below + */ +static inline void i915_gem_object_set_dirty(struct drm_i915_gem_object *obj) +{ + obj->__dirty = true; +} + +static inline void i915_gem_object_clear_dirty(struct drm_i915_gem_object *obj) +{ + obj->__dirty = false; +} + +static inline bool i915_gem_object_is_dirty(struct drm_i915_gem_object *obj) +{ + return obj->__dirty; +} + static inline void i915_gem_object_unpin_pages(struct drm_i915_gem_object *obj) { BUG_ON(obj->pages_pin_count == 0); diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 2401818..f571a02 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -234,9 +234,8 @@ int i915_mutex_lock_interruptible(struct drm_device *dev) } if (obj->madv == I915_MADV_DONTNEED) - obj->dirty = 0; - - if (obj->dirty) { + i915_gem_object_clear_dirty(obj); + else if (i915_gem_object_is_dirty(obj)) { struct address_space *mapping = obj->base.filp->f_mapping; char *vaddr = obj->phys_handle->vaddr; int i; @@ -260,7 +259,7 @@ int i915_mutex_lock_interruptible(struct drm_device *dev) put_page(page); vaddr += PAGE_SIZE; } - obj->dirty = 0; + i915_gem_object_clear_dirty(obj); } sg_free_table(obj->pages); @@ -703,7 +702,7 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, obj->cache_dirty = true; intel_fb_obj_invalidate(obj, ORIGIN_CPU); - obj->dirty = 1; + i915_gem_object_set_dirty(obj); /* return with the pages pinned */ return 0; @@ -1156,7 +1155,7 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, goto out_unpin; intel_fb_obj_invalidate(obj, ORIGIN_CPU); - obj->dirty = true; + i915_gem_object_set_dirty(obj); user_data = u64_to_user_ptr(args->data_ptr); offset = args->offset; @@ -1327,6 +1326,8 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, offset = args->offset; remain = args->size; + i915_gem_object_set_dirty(obj); + for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, offset >> PAGE_SHIFT) { struct page *page = sg_page_iter_page(&sg_iter); @@ -2133,6 +2134,7 @@ static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) { struct sgt_iter sgt_iter; struct page *page; + bool dirty; int ret; BUG_ON(obj->madv == __I915_MADV_PURGED); @@ -2152,10 +2154,11 @@ static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) i915_gem_object_save_bit_17_swizzle(obj); if (obj->madv == I915_MADV_DONTNEED) - obj->dirty = 0; + i915_gem_object_clear_dirty(obj); + dirty = i915_gem_object_is_dirty(obj); for_each_sgt_page(page, sgt_iter, obj->pages) { - if (obj->dirty) + if (dirty) set_page_dirty(page); if (obj->madv == I915_MADV_WILLNEED) @@ -2163,7 +2166,7 @@ static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) put_page(page); } - obj->dirty = 0; + i915_gem_object_clear_dirty(obj); sg_free_table(obj->pages); kfree(obj->pages); @@ -3321,7 +3324,7 @@ static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj) if (write) { obj->base.read_domains = I915_GEM_DOMAIN_GTT; obj->base.write_domain = I915_GEM_DOMAIN_GTT; - obj->dirty = 1; + i915_gem_object_set_dirty(obj); } trace_i915_gem_object_change_domain(obj, @@ -4789,7 +4792,7 @@ struct drm_i915_gem_object * i915_gem_object_pin_pages(obj); sg = obj->pages; bytes = sg_copy_from_buffer(sg->sgl, sg->nents, (void *)data, size); - obj->dirty = 1; /* Backing store is now out of date */ + i915_gem_object_set_dirty(obj); /* Backing store is now out of date */ i915_gem_object_unpin_pages(obj); if (WARN_ON(bytes != size)) { diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c index 35950ee..aa99bc3 100644 --- a/drivers/gpu/drm/i915/i915_gem_context.c +++ b/drivers/gpu/drm/i915/i915_gem_context.c @@ -834,6 +834,8 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) * MI_SET_CONTEXT instead of when the next seqno has completed. */ if (from != NULL) { + struct i915_vma *from_vma = from->engine[RCS].state; + /* As long as MI_SET_CONTEXT is serializing, ie. it flushes the * whole damn pipeline, we don't need to explicitly mark the * object dirty. The only exception is that the context must be @@ -841,9 +843,10 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) * able to defer doing this until we know the object would be * swapped, but there is no way to do that yet. */ - i915_vma_move_to_active(from->engine[RCS].state, req, 0); + i915_vma_move_to_active(from_vma, req, 0); + i915_gem_object_set_dirty(from_vma->obj); /* state is kept alive until the next request */ - i915_vma_unpin(from->engine[RCS].state); + i915_vma_unpin(from_vma); i915_gem_context_put(from); } engine->last_context = i915_gem_context_get(to); diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c index 9432d4c..5ca7ba3 100644 --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c @@ -1278,7 +1278,7 @@ void i915_vma_move_to_active(struct i915_vma *vma, GEM_BUG_ON(!drm_mm_node_allocated(&vma->node)); - obj->dirty = 1; /* be paranoid */ + i915_gem_object_set_dirty(obj); /* be paranoid */ /* Add a reference if we're newly entering the active list. * The order in which we add operations to the retirement queue is diff --git a/drivers/gpu/drm/i915/i915_gem_userptr.c b/drivers/gpu/drm/i915/i915_gem_userptr.c index be54825..a78abe2 100644 --- a/drivers/gpu/drm/i915/i915_gem_userptr.c +++ b/drivers/gpu/drm/i915/i915_gem_userptr.c @@ -674,23 +674,25 @@ struct get_pages_work { { struct sgt_iter sgt_iter; struct page *page; + bool dirty; BUG_ON(obj->userptr.work != NULL); __i915_gem_userptr_set_active(obj, false); - if (obj->madv != I915_MADV_WILLNEED) - obj->dirty = 0; - i915_gem_gtt_finish_object(obj); + if (obj->madv != I915_MADV_WILLNEED) + i915_gem_object_clear_dirty(obj); + + dirty = i915_gem_object_is_dirty(obj); for_each_sgt_page(page, sgt_iter, obj->pages) { - if (obj->dirty) + if (dirty) set_page_dirty(page); mark_page_accessed(page); put_page(page); } - obj->dirty = 0; + i915_gem_object_clear_dirty(obj); sg_free_table(obj->pages); kfree(obj->pages); diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c index 334f15d..257dde1 100644 --- a/drivers/gpu/drm/i915/i915_gpu_error.c +++ b/drivers/gpu/drm/i915/i915_gpu_error.c @@ -804,7 +804,7 @@ static void capture_bo(struct drm_i915_error_buffer *err, err->write_domain = obj->base.write_domain; err->fence_reg = vma->fence ? vma->fence->id : -1; err->tiling = i915_gem_object_get_tiling(obj); - err->dirty = obj->dirty; + err->dirty = i915_gem_object_is_dirty(obj); err->purgeable = obj->madv != I915_MADV_WILLNEED; err->userptr = obj->userptr.mm != NULL; err->cache_level = obj->cache_level; diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c index 92bfe47..1f48291 100644 --- a/drivers/gpu/drm/i915/intel_lrc.c +++ b/drivers/gpu/drm/i915/intel_lrc.c @@ -764,6 +764,8 @@ static int intel_lr_context_pin(struct i915_gem_context *ctx, struct intel_engine_cs *engine) { struct intel_context *ce = &ctx->engine[engine->id]; + struct intel_ring *ring = ce->ring; + struct drm_i915_gem_object *ctx_obj; void *vaddr; u32 *lrc_reg_state; int ret; @@ -778,24 +780,24 @@ static int intel_lr_context_pin(struct i915_gem_context *ctx, if (ret) goto err; - vaddr = i915_gem_object_pin_map(ce->state->obj, I915_MAP_WB); + ctx_obj = ce->state->obj; + vaddr = i915_gem_object_pin_map(ctx_obj, I915_MAP_WB); if (IS_ERR(vaddr)) { ret = PTR_ERR(vaddr); goto unpin_vma; } - lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; - - ret = intel_ring_pin(ce->ring); + ret = intel_ring_pin(ring); if (ret) goto unpin_map; - intel_lr_context_descriptor_update(ctx, engine); + i915_gem_object_set_dirty(ctx_obj); - lrc_reg_state[CTX_RING_BUFFER_START+1] = - i915_ggtt_offset(ce->ring->vma); + lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; + lrc_reg_state[CTX_RING_BUFFER_START+1] = i915_ggtt_offset(ring->vma); ce->lrc_reg_state = lrc_reg_state; - ce->state->obj->dirty = true; + + intel_lr_context_descriptor_update(ctx, engine); /* Invalidate GuC TLB. */ if (i915.enable_guc_submission) { @@ -1969,7 +1971,7 @@ static u32 intel_lr_indirect_ctx_offset(struct intel_engine_cs *engine) DRM_DEBUG_DRIVER("Could not map object pages! (%d)\n", ret); return ret; } - ctx_obj->dirty = true; + i915_gem_object_set_dirty(ctx_obj); /* The second page of the context object contains some fields which must * be set up prior to the first execution. */ @@ -2182,23 +2184,24 @@ void intel_lr_context_reset(struct drm_i915_private *dev_priv, for_each_engine(engine, dev_priv) { struct intel_context *ce = &ctx->engine[engine->id]; + struct drm_i915_gem_object *ctx_obj; void *vaddr; uint32_t *reg_state; if (!ce->state) continue; - vaddr = i915_gem_object_pin_map(ce->state->obj, I915_MAP_WB); + ctx_obj = ce->state->obj; + vaddr = i915_gem_object_pin_map(ctx_obj, I915_MAP_WB); if (WARN_ON(IS_ERR(vaddr))) continue; reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; - reg_state[CTX_RING_HEAD+1] = 0; reg_state[CTX_RING_TAIL+1] = 0; - ce->state->obj->dirty = true; - i915_gem_object_unpin_map(ce->state->obj); + i915_gem_object_set_dirty(ctx_obj); + i915_gem_object_unpin_map(ctx_obj); ce->ring->head = 0; ce->ring->tail = 0;
This just hides the existing obj->dirty flag inside a trivial inline setter, to discourage non-GEM code from looking too closely. The flag is renamed to emphasise that it is private to the GEM memory- management code and ensure that no legacy code continues to use it directly. v2: Use Chris Wilson's preferred names for flag-related functions Inspired-by: http://www.spinics.net/lists/intel-gfx/msg92390.html Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Dave Gordon <david.s.gordon@intel.com> --- drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_drv.h | 22 +++++++++++++++++++++- drivers/gpu/drm/i915/i915_gem.c | 25 ++++++++++++++----------- drivers/gpu/drm/i915/i915_gem_context.c | 7 +++++-- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- drivers/gpu/drm/i915/i915_gem_userptr.c | 12 +++++++----- drivers/gpu/drm/i915/i915_gpu_error.c | 2 +- drivers/gpu/drm/i915/intel_lrc.c | 29 ++++++++++++++++------------- 8 files changed, 66 insertions(+), 35 deletions(-)