From fc2f07566a2cb198eb5902684163e7c4f5309619 Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Mon, 20 Mar 2023 07:43:25 -0700 Subject: drm/msm/gem: Tidy up VMA API Stop open coding VMA construction, which will be needed in the next commit. And since the VMA already has a ptr to the adress space, stop passing that around everywhere. (Also, an aspace always has an mmu so we can drop a couple pointless NULL checks.) Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/527833/ Link: https://lore.kernel.org/r/20230320144356.803762-4-robdclark@gmail.com --- drivers/gpu/drm/msm/msm_gem.c | 18 ++++++++---------- 1 file changed, 8 insertions(+), 10 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gem.c') diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index c2fb98a94bc3e..47c05824c5566 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -309,12 +309,10 @@ static struct msm_gem_vma *add_vma(struct drm_gem_object *obj, msm_gem_assert_locked(obj); - vma = kzalloc(sizeof(*vma), GFP_KERNEL); + vma = msm_gem_vma_new(aspace); if (!vma) return ERR_PTR(-ENOMEM); - vma->aspace = aspace; - list_add_tail(&vma->list, &msm_obj->vmas); return vma; @@ -361,9 +359,9 @@ put_iova_spaces(struct drm_gem_object *obj, bool close) list_for_each_entry(vma, &msm_obj->vmas, list) { if (vma->aspace) { - msm_gem_purge_vma(vma->aspace, vma); + msm_gem_vma_purge(vma); if (close) - msm_gem_close_vma(vma->aspace, vma); + msm_gem_vma_close(vma); } } } @@ -399,7 +397,7 @@ static struct msm_gem_vma *get_vma_locked(struct drm_gem_object *obj, if (IS_ERR(vma)) return vma; - ret = msm_gem_init_vma(aspace, vma, obj->size, + ret = msm_gem_vma_init(vma, obj->size, range_start, range_end); if (ret) { del_vma(vma); @@ -437,7 +435,7 @@ int msm_gem_pin_vma_locked(struct drm_gem_object *obj, struct msm_gem_vma *vma) if (IS_ERR(pages)) return PTR_ERR(pages); - ret = msm_gem_map_vma(vma->aspace, vma, prot, msm_obj->sgt, obj->size); + ret = msm_gem_vma_map(vma, prot, msm_obj->sgt, obj->size); if (ret) msm_gem_unpin_locked(obj); @@ -539,8 +537,8 @@ static int clear_iova(struct drm_gem_object *obj, if (msm_gem_vma_inuse(vma)) return -EBUSY; - msm_gem_purge_vma(vma->aspace, vma); - msm_gem_close_vma(vma->aspace, vma); + msm_gem_vma_purge(vma); + msm_gem_vma_close(vma); del_vma(vma); return 0; @@ -589,7 +587,7 @@ void msm_gem_unpin_iova(struct drm_gem_object *obj, msm_gem_lock(obj); vma = lookup_vma(obj, aspace); if (!GEM_WARN_ON(!vma)) { - msm_gem_unpin_vma(vma); + msm_gem_vma_unpin(vma); msm_gem_unpin_locked(obj); } msm_gem_unlock(obj); -- cgit v1.2.3 From d6ae7d1cd58e2c606af005f4606310633ff4859e Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Mon, 20 Mar 2023 07:43:27 -0700 Subject: drm/msm/gem: Simplify vmap vs LRU tracking vmap'ing is just pinning in disguise. So treat it as such and simplify the LRU tracking. Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/527837/ Link: https://lore.kernel.org/r/20230320144356.803762-6-robdclark@gmail.com --- drivers/gpu/drm/msm/msm_gem.c | 17 ++++++++--------- 1 file changed, 8 insertions(+), 9 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gem.c') diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 47c05824c5566..c95cacbcf0bd6 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -626,6 +626,7 @@ fail: static void *get_vaddr(struct drm_gem_object *obj, unsigned madv) { struct msm_gem_object *msm_obj = to_msm_bo(obj); + struct page **pages; int ret = 0; msm_gem_assert_locked(obj); @@ -639,6 +640,10 @@ static void *get_vaddr(struct drm_gem_object *obj, unsigned madv) return ERR_PTR(-EBUSY); } + pages = msm_gem_pin_pages_locked(obj); + if (IS_ERR(pages)) + return ERR_CAST(pages); + /* increment vmap_count *before* vmap() call, so shrinker can * check vmap_count (is_vunmapable()) outside of msm_obj lock. * This guarantees that we won't try to msm_gem_vunmap() this @@ -648,25 +653,19 @@ static void *get_vaddr(struct drm_gem_object *obj, unsigned madv) msm_obj->vmap_count++; if (!msm_obj->vaddr) { - struct page **pages = get_pages(obj); - if (IS_ERR(pages)) { - ret = PTR_ERR(pages); - goto fail; - } msm_obj->vaddr = vmap(pages, obj->size >> PAGE_SHIFT, VM_MAP, msm_gem_pgprot(msm_obj, PAGE_KERNEL)); if (msm_obj->vaddr == NULL) { ret = -ENOMEM; goto fail; } - - update_lru(obj); } return msm_obj->vaddr; fail: msm_obj->vmap_count--; + msm_gem_unpin_locked(obj); return ERR_PTR(ret); } @@ -705,6 +704,7 @@ void msm_gem_put_vaddr_locked(struct drm_gem_object *obj) GEM_WARN_ON(msm_obj->vmap_count < 1); msm_obj->vmap_count--; + msm_gem_unpin_locked(obj); } void msm_gem_put_vaddr(struct drm_gem_object *obj) @@ -813,10 +813,9 @@ static void update_lru(struct drm_gem_object *obj) if (!msm_obj->pages) { GEM_WARN_ON(msm_obj->pin_count); - GEM_WARN_ON(msm_obj->vmap_count); drm_gem_lru_move_tail(&priv->lru.unbacked, obj); - } else if (msm_obj->pin_count || msm_obj->vmap_count) { + } else if (msm_obj->pin_count) { drm_gem_lru_move_tail(&priv->lru.pinned, obj); } else if (msm_obj->madv == MSM_MADV_WILLNEED) { drm_gem_lru_move_tail(&priv->lru.willneed, obj); -- cgit v1.2.3 From 4a02a376cbc839b89bbd1109f98b461e26337236 Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Mon, 20 Mar 2023 07:43:29 -0700 Subject: drm/msm/gem: Move update_lru() Just code-motion. Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/527841/ Link: https://lore.kernel.org/r/20230320144356.803762-8-robdclark@gmail.com --- drivers/gpu/drm/msm/msm_gem.c | 46 +++++++++++++++++++++---------------------- 1 file changed, 22 insertions(+), 24 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gem.c') diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index c95cacbcf0bd6..58ac94507ff12 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -19,8 +19,6 @@ #include "msm_gpu.h" #include "msm_mmu.h" -static void update_lru(struct drm_gem_object *obj); - static dma_addr_t physaddr(struct drm_gem_object *obj) { struct msm_gem_object *msm_obj = to_msm_bo(obj); @@ -63,6 +61,28 @@ static void sync_for_cpu(struct msm_gem_object *msm_obj) dma_unmap_sgtable(dev, msm_obj->sgt, DMA_BIDIRECTIONAL, 0); } +static void update_lru(struct drm_gem_object *obj) +{ + struct msm_drm_private *priv = obj->dev->dev_private; + struct msm_gem_object *msm_obj = to_msm_bo(obj); + + msm_gem_assert_locked(&msm_obj->base); + + if (!msm_obj->pages) { + GEM_WARN_ON(msm_obj->pin_count); + + drm_gem_lru_move_tail(&priv->lru.unbacked, obj); + } else if (msm_obj->pin_count) { + drm_gem_lru_move_tail(&priv->lru.pinned, obj); + } else if (msm_obj->madv == MSM_MADV_WILLNEED) { + drm_gem_lru_move_tail(&priv->lru.willneed, obj); + } else { + GEM_WARN_ON(msm_obj->madv != MSM_MADV_DONTNEED); + + drm_gem_lru_move_tail(&priv->lru.dontneed, obj); + } +} + /* allocate pages from VRAM carveout, used when no IOMMU: */ static struct page **get_pages_vram(struct drm_gem_object *obj, int npages) { @@ -804,28 +824,6 @@ void msm_gem_vunmap(struct drm_gem_object *obj) msm_obj->vaddr = NULL; } -static void update_lru(struct drm_gem_object *obj) -{ - struct msm_drm_private *priv = obj->dev->dev_private; - struct msm_gem_object *msm_obj = to_msm_bo(obj); - - msm_gem_assert_locked(&msm_obj->base); - - if (!msm_obj->pages) { - GEM_WARN_ON(msm_obj->pin_count); - - drm_gem_lru_move_tail(&priv->lru.unbacked, obj); - } else if (msm_obj->pin_count) { - drm_gem_lru_move_tail(&priv->lru.pinned, obj); - } else if (msm_obj->madv == MSM_MADV_WILLNEED) { - drm_gem_lru_move_tail(&priv->lru.willneed, obj); - } else { - GEM_WARN_ON(msm_obj->madv != MSM_MADV_DONTNEED); - - drm_gem_lru_move_tail(&priv->lru.dontneed, obj); - } -} - bool msm_gem_active(struct drm_gem_object *obj) { msm_gem_assert_locked(obj); -- cgit v1.2.3 From 6c7c8fb863f7c3137ce5fc4b435af3891083abeb Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Mon, 20 Mar 2023 07:43:30 -0700 Subject: drm/msm/gem: Protect pin_count/madv by LRU lock Since the LRU lock is already acquired when moving an obj between LRUs, we can use it to protect pin_count and madv, without any significant change in locking (ie. it just expands the scope of the lock by a hand- ful of instructions). This prepares the way to decrement the pin_count in the job_run() path without needing to hold the obj lock, to avoid a potential deadlock (or rather stall) caused by the fence-signaling path (job_run()) blocking on shrinker/reclaim. (Only a stall because the wait for fence signaling wait_for_idle() is not infinite.) Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/527843/ Link: https://lore.kernel.org/r/20230320144356.803762-9-robdclark@gmail.com --- drivers/gpu/drm/msm/msm_gem.c | 48 ++++++++++++++++++++++++++++++++----------- drivers/gpu/drm/msm/msm_gem.h | 9 +++++++- 2 files changed, 44 insertions(+), 13 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gem.c') diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 58ac94507ff12..b9171ac1a79a7 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -61,7 +61,7 @@ static void sync_for_cpu(struct msm_gem_object *msm_obj) dma_unmap_sgtable(dev, msm_obj->sgt, DMA_BIDIRECTIONAL, 0); } -static void update_lru(struct drm_gem_object *obj) +static void update_lru_locked(struct drm_gem_object *obj) { struct msm_drm_private *priv = obj->dev->dev_private; struct msm_gem_object *msm_obj = to_msm_bo(obj); @@ -71,18 +71,27 @@ static void update_lru(struct drm_gem_object *obj) if (!msm_obj->pages) { GEM_WARN_ON(msm_obj->pin_count); - drm_gem_lru_move_tail(&priv->lru.unbacked, obj); + drm_gem_lru_move_tail_locked(&priv->lru.unbacked, obj); } else if (msm_obj->pin_count) { - drm_gem_lru_move_tail(&priv->lru.pinned, obj); + drm_gem_lru_move_tail_locked(&priv->lru.pinned, obj); } else if (msm_obj->madv == MSM_MADV_WILLNEED) { - drm_gem_lru_move_tail(&priv->lru.willneed, obj); + drm_gem_lru_move_tail_locked(&priv->lru.willneed, obj); } else { GEM_WARN_ON(msm_obj->madv != MSM_MADV_DONTNEED); - drm_gem_lru_move_tail(&priv->lru.dontneed, obj); + drm_gem_lru_move_tail_locked(&priv->lru.dontneed, obj); } } +static void update_lru(struct drm_gem_object *obj) +{ + struct msm_drm_private *priv = obj->dev->dev_private; + + mutex_lock(&priv->lru.lock); + update_lru_locked(obj); + mutex_unlock(&priv->lru.lock); +} + /* allocate pages from VRAM carveout, used when no IOMMU: */ static struct page **get_pages_vram(struct drm_gem_object *obj, int npages) { @@ -200,6 +209,7 @@ static void put_pages(struct drm_gem_object *obj) static struct page **msm_gem_pin_pages_locked(struct drm_gem_object *obj) { + struct msm_drm_private *priv = obj->dev->dev_private; struct msm_gem_object *msm_obj = to_msm_bo(obj); struct page **p; @@ -210,10 +220,13 @@ static struct page **msm_gem_pin_pages_locked(struct drm_gem_object *obj) } p = get_pages(obj); - if (!IS_ERR(p)) { - to_msm_bo(obj)->pin_count++; - update_lru(obj); - } + if (IS_ERR(p)) + return p; + + mutex_lock(&priv->lru.lock); + msm_obj->pin_count++; + update_lru_locked(obj); + mutex_unlock(&priv->lru.lock); return p; } @@ -464,14 +477,16 @@ int msm_gem_pin_vma_locked(struct drm_gem_object *obj, struct msm_gem_vma *vma) void msm_gem_unpin_locked(struct drm_gem_object *obj) { + struct msm_drm_private *priv = obj->dev->dev_private; struct msm_gem_object *msm_obj = to_msm_bo(obj); msm_gem_assert_locked(obj); + mutex_lock(&priv->lru.lock); msm_obj->pin_count--; GEM_WARN_ON(msm_obj->pin_count < 0); - - update_lru(obj); + update_lru_locked(obj); + mutex_unlock(&priv->lru.lock); } struct msm_gem_vma *msm_gem_get_vma_locked(struct drm_gem_object *obj, @@ -739,10 +754,13 @@ void msm_gem_put_vaddr(struct drm_gem_object *obj) */ int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv) { + struct msm_drm_private *priv = obj->dev->dev_private; struct msm_gem_object *msm_obj = to_msm_bo(obj); msm_gem_lock(obj); + mutex_lock(&priv->lru.lock); + if (msm_obj->madv != __MSM_MADV_PURGED) msm_obj->madv = madv; @@ -751,7 +769,9 @@ int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv) /* If the obj is inactive, we might need to move it * between inactive lists */ - update_lru(obj); + update_lru_locked(obj); + + mutex_unlock(&priv->lru.lock); msm_gem_unlock(obj); @@ -761,6 +781,7 @@ int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv) void msm_gem_purge(struct drm_gem_object *obj) { struct drm_device *dev = obj->dev; + struct msm_drm_private *priv = obj->dev->dev_private; struct msm_gem_object *msm_obj = to_msm_bo(obj); msm_gem_assert_locked(obj); @@ -777,7 +798,10 @@ void msm_gem_purge(struct drm_gem_object *obj) put_iova_vmas(obj); + mutex_lock(&priv->lru.lock); + /* A one-way transition: */ msm_obj->madv = __MSM_MADV_PURGED; + mutex_unlock(&priv->lru.lock); drm_gem_free_mmap_offset(obj); diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h index 1929f09c5b0dc..0057e8e8fa135 100644 --- a/drivers/gpu/drm/msm/msm_gem.h +++ b/drivers/gpu/drm/msm/msm_gem.h @@ -86,7 +86,9 @@ struct msm_gem_object { uint32_t flags; /** - * Advice: are the backing pages purgeable? + * madv: are the backing pages purgeable? + * + * Protected by obj lock and LRU lock */ uint8_t madv; @@ -114,6 +116,11 @@ struct msm_gem_object { char name[32]; /* Identifier to print for the debugfs files */ + /** + * pin_count: Number of times the pages are pinned + * + * Protected by LRU lock. + */ int pin_count; }; #define to_msm_bo(x) container_of(x, struct msm_gem_object, base) -- cgit v1.2.3 From 17b704f1c0fb9150551567cb7a5414fb761b57ea Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Mon, 20 Mar 2023 07:43:31 -0700 Subject: drm/msm/gem: Avoid obj lock in job_run() Now that everything that controls which LRU an obj lives in *except* the backing pages is protected by the LRU lock, add a special path to unpin in the job_run() path, where we are assured that we already have backing pages and will not be racing against eviction (because the GEM object's dma_resv contains the fence that will be signaled when the submit/job completes). Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/527845/ Link: https://lore.kernel.org/r/20230320144356.803762-10-robdclark@gmail.com --- drivers/gpu/drm/msm/msm_gem.c | 44 ++++++++++++++++++++++++++++++------ drivers/gpu/drm/msm/msm_gem.h | 1 + drivers/gpu/drm/msm/msm_ringbuffer.c | 4 +--- 3 files changed, 39 insertions(+), 10 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gem.c') diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index b9171ac1a79a7..9008f967637a3 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -61,18 +61,14 @@ static void sync_for_cpu(struct msm_gem_object *msm_obj) dma_unmap_sgtable(dev, msm_obj->sgt, DMA_BIDIRECTIONAL, 0); } -static void update_lru_locked(struct drm_gem_object *obj) +static void update_lru_active(struct drm_gem_object *obj) { struct msm_drm_private *priv = obj->dev->dev_private; struct msm_gem_object *msm_obj = to_msm_bo(obj); - msm_gem_assert_locked(&msm_obj->base); - - if (!msm_obj->pages) { - GEM_WARN_ON(msm_obj->pin_count); + GEM_WARN_ON(!msm_obj->pages); - drm_gem_lru_move_tail_locked(&priv->lru.unbacked, obj); - } else if (msm_obj->pin_count) { + if (msm_obj->pin_count) { drm_gem_lru_move_tail_locked(&priv->lru.pinned, obj); } else if (msm_obj->madv == MSM_MADV_WILLNEED) { drm_gem_lru_move_tail_locked(&priv->lru.willneed, obj); @@ -83,6 +79,22 @@ static void update_lru_locked(struct drm_gem_object *obj) } } +static void update_lru_locked(struct drm_gem_object *obj) +{ + struct msm_drm_private *priv = obj->dev->dev_private; + struct msm_gem_object *msm_obj = to_msm_bo(obj); + + msm_gem_assert_locked(&msm_obj->base); + + if (!msm_obj->pages) { + GEM_WARN_ON(msm_obj->pin_count); + + drm_gem_lru_move_tail_locked(&priv->lru.unbacked, obj); + } else { + update_lru_active(obj); + } +} + static void update_lru(struct drm_gem_object *obj) { struct msm_drm_private *priv = obj->dev->dev_private; @@ -489,6 +501,24 @@ void msm_gem_unpin_locked(struct drm_gem_object *obj) mutex_unlock(&priv->lru.lock); } +/* Special unpin path for use in fence-signaling path, avoiding the need + * to hold the obj lock by only depending on things that a protected by + * the LRU lock. In particular we know that that we already have backing + * and and that the object's dma_resv has the fence for the current + * submit/job which will prevent us racing against page eviction. + */ +void msm_gem_unpin_active(struct drm_gem_object *obj) +{ + struct msm_drm_private *priv = obj->dev->dev_private; + struct msm_gem_object *msm_obj = to_msm_bo(obj); + + mutex_lock(&priv->lru.lock); + msm_obj->pin_count--; + GEM_WARN_ON(msm_obj->pin_count < 0); + update_lru_active(obj); + mutex_unlock(&priv->lru.lock); +} + struct msm_gem_vma *msm_gem_get_vma_locked(struct drm_gem_object *obj, struct msm_gem_address_space *aspace) { diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h index 0057e8e8fa135..2bd6846c83a98 100644 --- a/drivers/gpu/drm/msm/msm_gem.h +++ b/drivers/gpu/drm/msm/msm_gem.h @@ -128,6 +128,7 @@ struct msm_gem_object { uint64_t msm_gem_mmap_offset(struct drm_gem_object *obj); int msm_gem_pin_vma_locked(struct drm_gem_object *obj, struct msm_gem_vma *vma); void msm_gem_unpin_locked(struct drm_gem_object *obj); +void msm_gem_unpin_active(struct drm_gem_object *obj); struct msm_gem_vma *msm_gem_get_vma_locked(struct drm_gem_object *obj, struct msm_gem_address_space *aspace); int msm_gem_get_iova(struct drm_gem_object *obj, diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.c b/drivers/gpu/drm/msm/msm_ringbuffer.c index 31b4fbf96c360..b601991844093 100644 --- a/drivers/gpu/drm/msm/msm_ringbuffer.c +++ b/drivers/gpu/drm/msm/msm_ringbuffer.c @@ -24,9 +24,7 @@ static struct dma_fence *msm_job_run(struct drm_sched_job *job) struct drm_gem_object *obj = &submit->bos[i].obj->base; msm_gem_vma_unpin_fenced(submit->bos[i].vma, fctx); - msm_gem_lock(obj); - msm_gem_unpin_locked(obj); - msm_gem_unlock(obj); + msm_gem_unpin_active(obj); submit->bos[i].flags &= ~(BO_VMA_PINNED | BO_OBJ_PINNED); } -- cgit v1.2.3 From b5a24e13c8c8b2c98d114b16da40712b80d5cfc1 Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Wed, 8 Mar 2023 07:53:04 -0800 Subject: drm/msm: Add wait-boost support Add a way for various userspace waits to signal urgency. Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/525817/ Link: https://lore.kernel.org/r/20230308155322.344664-14-robdclark@gmail.com --- drivers/gpu/drm/msm/msm_drv.c | 12 ++++++++---- drivers/gpu/drm/msm/msm_gem.c | 5 +++++ include/uapi/drm/msm_drm.h | 14 ++++++++++++-- 3 files changed, 25 insertions(+), 6 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gem.c') diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index ce1a77b607d1b..9b6f17b1261f4 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -46,6 +46,7 @@ * - 1.8.0 - Add MSM_BO_CACHED_COHERENT for supported GPUs (a6xx) * - 1.9.0 - Add MSM_SUBMIT_FENCE_SN_IN * - 1.10.0 - Add MSM_SUBMIT_BO_NO_IMPLICIT + * - 1.11.0 - Add wait boost (MSM_WAIT_FENCE_BOOST, MSM_PREP_BOOST) */ #define MSM_VERSION_MAJOR 1 #define MSM_VERSION_MINOR 10 @@ -899,7 +900,7 @@ static int msm_ioctl_gem_info(struct drm_device *dev, void *data, } static int wait_fence(struct msm_gpu_submitqueue *queue, uint32_t fence_id, - ktime_t timeout) + ktime_t timeout, uint32_t flags) { struct dma_fence *fence; int ret; @@ -927,6 +928,9 @@ static int wait_fence(struct msm_gpu_submitqueue *queue, uint32_t fence_id, if (!fence) return 0; + if (flags & MSM_WAIT_FENCE_BOOST) + dma_fence_set_deadline(fence, ktime_get()); + ret = dma_fence_wait_timeout(fence, true, timeout_to_jiffies(&timeout)); if (ret == 0) { ret = -ETIMEDOUT; @@ -947,8 +951,8 @@ static int msm_ioctl_wait_fence(struct drm_device *dev, void *data, struct msm_gpu_submitqueue *queue; int ret; - if (args->pad) { - DRM_ERROR("invalid pad: %08x\n", args->pad); + if (args->flags & ~MSM_WAIT_FENCE_FLAGS) { + DRM_ERROR("invalid flags: %08x\n", args->flags); return -EINVAL; } @@ -959,7 +963,7 @@ static int msm_ioctl_wait_fence(struct drm_device *dev, void *data, if (!queue) return -ENOENT; - ret = wait_fence(queue, args->fence, to_ktime(args->timeout)); + ret = wait_fence(queue, args->fence, to_ktime(args->timeout), args->flags); msm_submitqueue_put(queue); diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 9008f967637a3..db6c4e281d75d 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -895,6 +895,11 @@ int msm_gem_cpu_prep(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout) op & MSM_PREP_NOSYNC ? 0 : timeout_to_jiffies(timeout); long ret; + if (op & MSM_PREP_BOOST) { + dma_resv_set_deadline(obj->resv, dma_resv_usage_rw(write), + ktime_get()); + } + ret = dma_resv_wait_timeout(obj->resv, dma_resv_usage_rw(write), true, remain); if (ret == 0) diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index 329100016e7cc..dbf0d6f43fa98 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -151,8 +151,13 @@ struct drm_msm_gem_info { #define MSM_PREP_READ 0x01 #define MSM_PREP_WRITE 0x02 #define MSM_PREP_NOSYNC 0x04 +#define MSM_PREP_BOOST 0x08 -#define MSM_PREP_FLAGS (MSM_PREP_READ | MSM_PREP_WRITE | MSM_PREP_NOSYNC) +#define MSM_PREP_FLAGS (MSM_PREP_READ | \ + MSM_PREP_WRITE | \ + MSM_PREP_NOSYNC | \ + MSM_PREP_BOOST | \ + 0) struct drm_msm_gem_cpu_prep { __u32 handle; /* in */ @@ -286,6 +291,11 @@ struct drm_msm_gem_submit { }; +#define MSM_WAIT_FENCE_BOOST 0x00000001 +#define MSM_WAIT_FENCE_FLAGS ( \ + MSM_WAIT_FENCE_BOOST | \ + 0) + /* The normal way to synchronize with the GPU is just to CPU_PREP on * a buffer if you need to access it from the CPU (other cmdstream * submission from same or other contexts, PAGE_FLIP ioctl, etc, all @@ -295,7 +305,7 @@ struct drm_msm_gem_submit { */ struct drm_msm_wait_fence { __u32 fence; /* in */ - __u32 pad; + __u32 flags; /* in, bitmask of MSM_WAIT_FENCE_x */ struct drm_msm_timespec timeout; /* in */ __u32 queueid; /* in, submitqueue id */ }; -- cgit v1.2.3 From 16eb51aba0a72b8f60c6b98d1493844230293450 Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Mon, 17 Apr 2023 15:55:04 -0700 Subject: drm/msm: Fix vmap madv warning Commit d6ae7d1cd58e ("drm/msm/gem: Simplify vmap vs LRU tracking") introduced a splat in the pin_pages_locked() path for buffers that had been MADV_DONTNEED. ------------[ cut here ]------------ msm_obj->madv != 0 WARNING: CPU: 1 PID: 144 at drivers/gpu/drm/msm/msm_gem.c:230 msm_gem_pin_pages_locked+0x9c/0xd4 Modules linked in: lzo_rle cros_ec_lid_angle cros_ec_sensors cros_ec_sensors_core venus_dec venus_enc videobuf2_dma_contig cdc_ether usbnet mii uvcvideo videobuf2_vmalloc hci_uart btqca qcom_spmi_adc5 uvc qcom_spmi_temp_alarm qcom_vadc_common cros_ec_sensorhub videobuf2_memops cros_ec_typec sx9324 sx_common typec joydev bluetooth industrialio_triggered_buffer ecdh_generic kfifo_buf ecc venus_core qcom_stats v4l2_mem2mem videobuf2_v4l2 videobuf2_common ath11k_ahb ath11k mac80211 cfg80211 fuse zram zsmalloc CPU: 1 PID: 144 Comm: ring0 Tainted: G W 6.3.0-rc2-debug+ #622 Hardware name: Google Villager (rev1+) with LTE (DT) pstate: 60400009 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : msm_gem_pin_pages_locked+0x9c/0xd4 lr : msm_gem_pin_pages_locked+0x9c/0xd4 sp : ffffffc009ffbab0 x29: ffffffc009ffbab0 x28: ffffffee8da75008 x27: ffffff80a10274d0 x26: ffffff8087fe3bf8 x25: ffffff8087fe3c08 x24: 0000000000000001 x23: ffffff80891d5800 x22: ffffff809d0de480 x21: ffffff8081e5a080 x20: 0000000000000002 x19: ffffff80a3564c00 x18: 0000000000000000 x17: 0000000000000000 x16: 0000000000000000 x15: 00000000000a9620 x14: 0000000000000000 x13: 2d2d2d2d2d2d2d2d x12: 2d2d2d2d5d206572 x11: 656820747563205b x10: 2d2d2d2d2d2d2d2d x9 : ffffffee8c705dfc x8 : ffffffee8da75000 x7 : ffffffee8d34e6d0 x6 : 0000000000000000 x5 : 00000000000affa8 x4 : 000000000000000d x3 : ffffffee8da75008 x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffffff8088048040 Call trace: msm_gem_pin_pages_locked+0x9c/0xd4 get_vaddr+0xb0/0x150 msm_gem_get_vaddr_active+0x1c/0x28 snapshot_buf+0x90/0x10c msm_rd_dump_submit+0x30c/0x380 msm_gpu_submit+0x88/0x174 msm_job_run+0x68/0x118 drm_sched_main+0x2b8/0x3a0 kthread+0xf0/0x100 ret_from_fork+0x10/0x20 irq event stamp: 3358 hardirqs last enabled at (3357): [] __up_console_sem+0x7c/0x80 hardirqs last disabled at (3358): [] el1_dbg+0x24/0x80 softirqs last enabled at (3330): [] __do_softirq+0x21c/0x4bc softirqs last disabled at (3325): [] ____do_softirq+0x18/0x24 ---[ end trace 0000000000000000 ]--- But, as with msm_gem_get_vaddr_active(), this is a special case because we know that the buffer won't be purged evicted until it's fence is signaled. We just forgot to propagate the logic get_vaddr() to pin_pages_locked(). Fixes: d6ae7d1cd58e ("drm/msm/gem: Simplify vmap vs LRU tracking") Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/532616/ Link: https://lore.kernel.org/r/20230417225504.494934-1-robdclark@gmail.com --- drivers/gpu/drm/msm/msm_gem.c | 22 ++++++++-------------- 1 file changed, 8 insertions(+), 14 deletions(-) (limited to 'drivers/gpu/drm/msm/msm_gem.c') diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index db6c4e281d75d..cd39b9d8abdbe 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -219,7 +219,8 @@ static void put_pages(struct drm_gem_object *obj) } } -static struct page **msm_gem_pin_pages_locked(struct drm_gem_object *obj) +static struct page **msm_gem_pin_pages_locked(struct drm_gem_object *obj, + unsigned madv) { struct msm_drm_private *priv = obj->dev->dev_private; struct msm_gem_object *msm_obj = to_msm_bo(obj); @@ -227,7 +228,9 @@ static struct page **msm_gem_pin_pages_locked(struct drm_gem_object *obj) msm_gem_assert_locked(obj); - if (GEM_WARN_ON(msm_obj->madv != MSM_MADV_WILLNEED)) { + if (GEM_WARN_ON(msm_obj->madv > madv)) { + DRM_DEV_ERROR(obj->dev->dev, "Invalid madv state: %u vs %u\n", + msm_obj->madv, madv); return ERR_PTR(-EBUSY); } @@ -248,7 +251,7 @@ struct page **msm_gem_pin_pages(struct drm_gem_object *obj) struct page **p; msm_gem_lock(obj); - p = msm_gem_pin_pages_locked(obj); + p = msm_gem_pin_pages_locked(obj, MSM_MADV_WILLNEED); msm_gem_unlock(obj); return p; @@ -473,10 +476,7 @@ int msm_gem_pin_vma_locked(struct drm_gem_object *obj, struct msm_gem_vma *vma) msm_gem_assert_locked(obj); - if (GEM_WARN_ON(msm_obj->madv != MSM_MADV_WILLNEED)) - return -EBUSY; - - pages = msm_gem_pin_pages_locked(obj); + pages = msm_gem_pin_pages_locked(obj, MSM_MADV_WILLNEED); if (IS_ERR(pages)) return PTR_ERR(pages); @@ -699,13 +699,7 @@ static void *get_vaddr(struct drm_gem_object *obj, unsigned madv) if (obj->import_attach) return ERR_PTR(-ENODEV); - if (GEM_WARN_ON(msm_obj->madv > madv)) { - DRM_DEV_ERROR(obj->dev->dev, "Invalid madv state: %u vs %u\n", - msm_obj->madv, madv); - return ERR_PTR(-EBUSY); - } - - pages = msm_gem_pin_pages_locked(obj); + pages = msm_gem_pin_pages_locked(obj, madv); if (IS_ERR(pages)) return ERR_CAST(pages); -- cgit v1.2.3