Christian König
2019-Sep-30 13:12 UTC
[Nouveau] [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver
While working on TTM cleanups I've found that the io_reserve_lru used by Nouveau is actually not working at all. In general we should remove driver specific handling from the memory management, so this patch moves the io_reserve_lru handling into Nouveau instead. The patch should be functional correct, but is only compile tested! Signed-off-by: Christian König <christian.koenig at amd.com> --- drivers/gpu/drm/nouveau/nouveau_bo.c | 106 ++++++++++++++++++++++++++-------- drivers/gpu/drm/nouveau/nouveau_bo.h | 3 + drivers/gpu/drm/nouveau/nouveau_drv.h | 2 + drivers/gpu/drm/nouveau/nouveau_ttm.c | 43 +++++++++++++- 4 files changed, 130 insertions(+), 24 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c index e918b437af17..9532ad1bd474 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.c +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c @@ -139,6 +139,7 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo) if (unlikely(nvbo->bo.base.filp)) DRM_ERROR("bo %p still attached to GEM object\n", bo); WARN_ON(nvbo->pin_refcnt > 0); + nouveau_bo_del_io_reserve_lru(bo); nv10_bo_put_tile_region(dev, nvbo->tile, NULL); kfree(nvbo); } @@ -297,6 +298,7 @@ nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 flags, nouveau_bo_fixup_align(nvbo, flags, &align, &size); nvbo->bo.mem.num_pages = size >> PAGE_SHIFT; nouveau_bo_placement_set(nvbo, flags, 0); + INIT_LIST_HEAD(&nvbo->io_reserve_lru); ret = ttm_bo_init(nvbo->bo.bdev, &nvbo->bo, size, type, &nvbo->placement, align >> PAGE_SHIFT, false, @@ -566,6 +568,26 @@ nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo) PAGE_SIZE, DMA_FROM_DEVICE); } +void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo) +{ + struct nouveau_drm *drm = nouveau_bdev(bo->bdev); + struct nouveau_bo *nvbo = nouveau_bo(bo); + + mutex_lock(&drm->ttm.io_reserve_mutex); + list_move_tail(&nvbo->io_reserve_lru, &drm->ttm.io_reserve_lru); + mutex_unlock(&drm->ttm.io_reserve_mutex); +} + +void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo) +{ + struct nouveau_drm *drm = nouveau_bdev(bo->bdev); + struct nouveau_bo *nvbo = nouveau_bo(bo); + + mutex_lock(&drm->ttm.io_reserve_mutex); + list_del_init(&nvbo->io_reserve_lru); + mutex_unlock(&drm->ttm.io_reserve_mutex); +} + int nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible, bool no_wait_gpu) @@ -674,8 +696,6 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, } man->func = &nouveau_vram_manager; - man->io_reserve_fastpath = false; - man->use_io_reserve_lru = true; } else { man->func = &ttm_bo_manager_func; } @@ -1304,6 +1324,8 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict, if (bo->destroy != nouveau_bo_del_ttm) return; + nouveau_bo_del_io_reserve_lru(bo); + if (mem && new_reg->mem_type != TTM_PL_SYSTEM && mem->mem.page == nvbo->page) { list_for_each_entry(vma, &nvbo->vma_list, head) { @@ -1426,6 +1448,30 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp) filp->private_data); } +static void +nouveau_ttm_io_mem_free_locked(struct nouveau_drm *drm, struct ttm_mem_reg *reg) +{ + struct nouveau_mem *mem = nouveau_mem(reg); + + if (!reg->bus.base) + return; /* already freed */ + + if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) { + switch (reg->mem_type) { + case TTM_PL_TT: + if (mem->kind) + nvif_object_unmap_handle(&mem->mem.object); + break; + case TTM_PL_VRAM: + nvif_object_unmap_handle(&mem->mem.object); + break; + default: + break; + } + } + reg->bus.base = 0; +} + static int nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) { @@ -1433,18 +1479,26 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) struct nouveau_drm *drm = nouveau_bdev(bdev); struct nvkm_device *device = nvxx_device(&drm->client.device); struct nouveau_mem *mem = nouveau_mem(reg); + struct nouveau_bo *nvbo; + int ret; + + if (reg->bus.base) + return 0; /* already mapped */ reg->bus.addr = NULL; reg->bus.offset = 0; reg->bus.size = reg->num_pages << PAGE_SHIFT; - reg->bus.base = 0; reg->bus.is_iomem = false; if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE)) return -EINVAL; + + mutex_lock(&drm->ttm.io_reserve_mutex); +retry: switch (reg->mem_type) { case TTM_PL_SYSTEM: /* System memory */ - return 0; + ret = 0; + goto out; case TTM_PL_TT: #if IS_ENABLED(CONFIG_AGP) if (drm->agp.bridge) { @@ -1468,7 +1522,6 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) } args; u64 handle, length; u32 argc = 0; - int ret; switch (mem->mem.object.oclass) { case NVIF_CLASS_MEM_NV50: @@ -1492,38 +1545,45 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) ret = nvif_object_map_handle(&mem->mem.object, &args, argc, &handle, &length); - if (ret != 1) - return ret ? ret : -EINVAL; + if (ret != 1) { + ret = ret ? ret : -EINVAL; + goto out; + } + ret = 0; reg->bus.base = 0; reg->bus.offset = handle; } break; default: - return -EINVAL; + ret = -EINVAL; } - return 0; + +out: + if (ret == -EAGAIN) { + nvbo = list_first_entry_or_null(&drm->ttm.io_reserve_lru, + typeof(*nvbo), + io_reserve_lru); + if (nvbo) { + list_del_init(&nvbo->io_reserve_lru); + ttm_bo_unmap_virtual(&nvbo->bo); + nouveau_ttm_io_mem_free_locked(drm, &nvbo->bo.mem); + goto retry; + } + + } + mutex_unlock(&drm->ttm.io_reserve_mutex); + return ret; } static void nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) { struct nouveau_drm *drm = nouveau_bdev(bdev); - struct nouveau_mem *mem = nouveau_mem(reg); - if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) { - switch (reg->mem_type) { - case TTM_PL_TT: - if (mem->kind) - nvif_object_unmap_handle(&mem->mem.object); - break; - case TTM_PL_VRAM: - nvif_object_unmap_handle(&mem->mem.object); - break; - default: - break; - } - } + mutex_lock(&drm->ttm.io_reserve_mutex); + nouveau_ttm_io_mem_free_locked(drm, reg); + mutex_unlock(&drm->ttm.io_reserve_mutex); } static int diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h index 62930d834fba..b8a17bc19d7b 100644 --- a/drivers/gpu/drm/nouveau/nouveau_bo.h +++ b/drivers/gpu/drm/nouveau/nouveau_bo.h @@ -17,6 +17,7 @@ struct nouveau_bo { bool force_coherent; struct ttm_bo_kmap_obj kmap; struct list_head head; + struct list_head io_reserve_lru; /* protected by ttm_bo_reserve() */ struct drm_file *reserved_by; @@ -92,6 +93,8 @@ int nouveau_bo_validate(struct nouveau_bo *, bool interruptible, bool no_wait_gpu); void nouveau_bo_sync_for_device(struct nouveau_bo *nvbo); void nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo); +void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo); +void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo); /* TODO: submit equivalent to TTM generic API upstream? */ static inline void __iomem * diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h index 70f34cacc552..060e2ab1fd95 100644 --- a/drivers/gpu/drm/nouveau/nouveau_drv.h +++ b/drivers/gpu/drm/nouveau/nouveau_drv.h @@ -158,6 +158,8 @@ struct nouveau_drm { int type_vram; int type_host[2]; int type_ncoh[2]; + struct mutex io_reserve_mutex; + struct list_head io_reserve_lru; } ttm; /* GEM interface support */ diff --git a/drivers/gpu/drm/nouveau/nouveau_ttm.c b/drivers/gpu/drm/nouveau/nouveau_ttm.c index f0daf958e03a..7f8495066e8b 100644 --- a/drivers/gpu/drm/nouveau/nouveau_ttm.c +++ b/drivers/gpu/drm/nouveau/nouveau_ttm.c @@ -162,13 +162,51 @@ const struct ttm_mem_type_manager_func nv04_gart_manager = { .debug = nouveau_manager_debug }; +static vm_fault_t nouveau_ttm_fault(struct vm_fault *vmf) +{ + struct vm_area_struct *vma = vmf->vma; + struct ttm_buffer_object *bo = vma->vm_private_data; + pgprot_t prot; + vm_fault_t ret; + + ret = ttm_bo_vm_reserve(bo, vmf); + if (ret) + return ret; + + nouveau_bo_del_io_reserve_lru(bo); + + prot = vm_get_page_prot(vma->vm_flags); + ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT); + if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) + return ret; + + nouveau_bo_add_io_reserve_lru(bo); + + dma_resv_unlock(bo->base.resv); + + return ret; +} + +static struct vm_operations_struct nouveau_ttm_vm_ops = { + .fault = nouveau_ttm_fault, + .open = ttm_bo_vm_open, + .close = ttm_bo_vm_close, + .access = ttm_bo_vm_access +}; + int nouveau_ttm_mmap(struct file *filp, struct vm_area_struct *vma) { struct drm_file *file_priv = filp->private_data; struct nouveau_drm *drm = nouveau_drm(file_priv->minor->dev); + int ret; - return ttm_bo_mmap(filp, vma, &drm->ttm.bdev); + ret = ttm_bo_mmap(filp, vma, &drm->ttm.bdev); + if (ret) + return ret; + + vma->vm_ops = &nouveau_ttm_vm_ops; + return 0; } static int @@ -272,6 +310,9 @@ nouveau_ttm_init(struct nouveau_drm *drm) return ret; } + mutex_init(&drm->ttm.io_reserve_mutex); + INIT_LIST_HEAD(&drm->ttm.io_reserve_lru); + NV_INFO(drm, "VRAM: %d MiB\n", (u32)(drm->gem.vram_available >> 20)); NV_INFO(drm, "GART: %d MiB\n", (u32)(drm->gem.gart_available >> 20)); return 0; -- 2.14.1
Christian König
2019-Sep-30 13:12 UTC
[Nouveau] [PATCH 2/2] drm/ttm: remove io_reserve_lru handling
That is not used any more. Signed-off-by: Christian König <christian.koenig at amd.com> --- drivers/gpu/drm/ttm/ttm_bo.c | 37 +++---------- drivers/gpu/drm/ttm/ttm_bo_util.c | 114 +------------------------------------- drivers/gpu/drm/ttm/ttm_bo_vm.c | 33 +++-------- include/drm/ttm/ttm_bo_api.h | 5 -- include/drm/ttm/ttm_bo_driver.h | 14 ----- 5 files changed, 20 insertions(+), 183 deletions(-) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 6394e0c5cc02..324468d0870f 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -328,13 +328,8 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo, int ret = 0; if (old_is_pci || new_is_pci || - ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0)) { - ret = ttm_mem_io_lock(old_man, true); - if (unlikely(ret != 0)) - goto out_err; - ttm_bo_unmap_virtual_locked(bo); - ttm_mem_io_unlock(old_man); - } + ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0)) + ttm_bo_unmap_virtual(bo); /* * Create and bind a ttm if required. @@ -670,15 +665,12 @@ static void ttm_bo_release(struct kref *kref) struct ttm_buffer_object *bo container_of(kref, struct ttm_buffer_object, kref); struct ttm_bo_device *bdev = bo->bdev; - struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type]; if (bo->bdev->driver->release_notify) bo->bdev->driver->release_notify(bo); drm_vma_offset_remove(&bdev->vma_manager, &bo->base.vma_node); - ttm_mem_io_lock(man, false); - ttm_mem_io_free_vm(bo); - ttm_mem_io_unlock(man); + ttm_mem_io_free(bdev, &bo->mem); ttm_bo_cleanup_refs_or_queue(bo); kref_put(&bo->list_kref, ttm_bo_release_list); } @@ -727,8 +719,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, evict_mem = bo->mem; evict_mem.mm_node = NULL; - evict_mem.bus.io_reserved_vm = false; - evict_mem.bus.io_reserved_count = 0; + evict_mem.bus.base = 0; ret = ttm_bo_mem_space(bo, &placement, &evict_mem, ctx); if (ret) { @@ -1188,8 +1179,7 @@ static int ttm_bo_move_buffer(struct ttm_buffer_object *bo, mem.num_pages = bo->num_pages; mem.size = mem.num_pages << PAGE_SHIFT; mem.page_alignment = bo->mem.page_alignment; - mem.bus.io_reserved_vm = false; - mem.bus.io_reserved_count = 0; + mem.bus.base = 0; /* * Determine where to move the buffer. */ @@ -1336,8 +1326,7 @@ int ttm_bo_init_reserved(struct ttm_bo_device *bdev, bo->mem.num_pages = bo->num_pages; bo->mem.mm_node = NULL; bo->mem.page_alignment = page_alignment; - bo->mem.bus.io_reserved_vm = false; - bo->mem.bus.io_reserved_count = 0; + bo->mem.bus.base = 0; bo->moving = NULL; bo->mem.placement = (TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED); bo->acc_size = acc_size; @@ -1788,22 +1777,12 @@ bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) return true; } -void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo) -{ - struct ttm_bo_device *bdev = bo->bdev; - - drm_vma_node_unmap(&bo->base.vma_node, bdev->dev_mapping); - ttm_mem_io_free_vm(bo); -} - void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo) { struct ttm_bo_device *bdev = bo->bdev; - struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type]; - ttm_mem_io_lock(man, false); - ttm_bo_unmap_virtual_locked(bo); - ttm_mem_io_unlock(man); + drm_vma_node_unmap(&bo->base.vma_node, bdev->dev_mapping); + ttm_mem_io_free(bdev, &bo->mem); } diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c index 2eca752c39e9..7df89134b50c 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_util.c +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c @@ -91,124 +91,30 @@ int ttm_bo_move_ttm(struct ttm_buffer_object *bo, } EXPORT_SYMBOL(ttm_bo_move_ttm); -int ttm_mem_io_lock(struct ttm_mem_type_manager *man, bool interruptible) -{ - if (likely(man->io_reserve_fastpath)) - return 0; - - if (interruptible) - return mutex_lock_interruptible(&man->io_reserve_mutex); - - mutex_lock(&man->io_reserve_mutex); - return 0; -} - -void ttm_mem_io_unlock(struct ttm_mem_type_manager *man) -{ - if (likely(man->io_reserve_fastpath)) - return; - - mutex_unlock(&man->io_reserve_mutex); -} - -static int ttm_mem_io_evict(struct ttm_mem_type_manager *man) -{ - struct ttm_buffer_object *bo; - - if (!man->use_io_reserve_lru || list_empty(&man->io_reserve_lru)) - return -EAGAIN; - - bo = list_first_entry(&man->io_reserve_lru, - struct ttm_buffer_object, - io_reserve_lru); - list_del_init(&bo->io_reserve_lru); - ttm_bo_unmap_virtual_locked(bo); - - return 0; -} - - int ttm_mem_io_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) { - struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; - int ret = 0; - if (!bdev->driver->io_mem_reserve) return 0; - if (likely(man->io_reserve_fastpath)) - return bdev->driver->io_mem_reserve(bdev, mem); - - if (bdev->driver->io_mem_reserve && - mem->bus.io_reserved_count++ == 0) { -retry: - ret = bdev->driver->io_mem_reserve(bdev, mem); - if (ret == -EAGAIN) { - ret = ttm_mem_io_evict(man); - if (ret == 0) - goto retry; - } - } - return ret; + + return bdev->driver->io_mem_reserve(bdev, mem); } void ttm_mem_io_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) { - struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; - - if (likely(man->io_reserve_fastpath)) - return; - - if (bdev->driver->io_mem_reserve && - --mem->bus.io_reserved_count == 0 && - bdev->driver->io_mem_free) + if (bdev->driver->io_mem_free) bdev->driver->io_mem_free(bdev, mem); - -} - -int ttm_mem_io_reserve_vm(struct ttm_buffer_object *bo) -{ - struct ttm_mem_reg *mem = &bo->mem; - int ret; - - if (!mem->bus.io_reserved_vm) { - struct ttm_mem_type_manager *man - &bo->bdev->man[mem->mem_type]; - - ret = ttm_mem_io_reserve(bo->bdev, mem); - if (unlikely(ret != 0)) - return ret; - mem->bus.io_reserved_vm = true; - if (man->use_io_reserve_lru) - list_add_tail(&bo->io_reserve_lru, - &man->io_reserve_lru); - } - return 0; -} - -void ttm_mem_io_free_vm(struct ttm_buffer_object *bo) -{ - struct ttm_mem_reg *mem = &bo->mem; - - if (mem->bus.io_reserved_vm) { - mem->bus.io_reserved_vm = false; - list_del_init(&bo->io_reserve_lru); - ttm_mem_io_free(bo->bdev, mem); - } } static int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem, void **virtual) { - struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; int ret; void *addr; *virtual = NULL; - (void) ttm_mem_io_lock(man, false); ret = ttm_mem_io_reserve(bdev, mem); - ttm_mem_io_unlock(man); if (ret || !mem->bus.is_iomem) return ret; @@ -220,9 +126,7 @@ static int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *m else addr = ioremap_nocache(mem->bus.base + mem->bus.offset, mem->bus.size); if (!addr) { - (void) ttm_mem_io_lock(man, false); ttm_mem_io_free(bdev, mem); - ttm_mem_io_unlock(man); return -ENOMEM; } } @@ -239,9 +143,7 @@ static void ttm_mem_reg_iounmap(struct ttm_bo_device *bdev, struct ttm_mem_reg * if (virtual && mem->bus.addr == NULL) iounmap(virtual); - (void) ttm_mem_io_lock(man, false); ttm_mem_io_free(bdev, mem); - ttm_mem_io_unlock(man); } static int ttm_copy_io_page(void *dst, void *src, unsigned long page) @@ -616,8 +518,6 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long start_page, unsigned long num_pages, struct ttm_bo_kmap_obj *map) { - struct ttm_mem_type_manager *man - &bo->bdev->man[bo->mem.mem_type]; unsigned long offset, size; int ret; @@ -628,9 +528,7 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo, if (start_page > bo->num_pages) return -EINVAL; - (void) ttm_mem_io_lock(man, false); ret = ttm_mem_io_reserve(bo->bdev, &bo->mem); - ttm_mem_io_unlock(man); if (ret) return ret; if (!bo->mem.bus.is_iomem) { @@ -645,10 +543,6 @@ EXPORT_SYMBOL(ttm_bo_kmap); void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map) { - struct ttm_buffer_object *bo = map->bo; - struct ttm_mem_type_manager *man - &bo->bdev->man[bo->mem.mem_type]; - if (!map->virtual) return; switch (map->bo_kmap_type) { @@ -666,9 +560,7 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map) default: BUG(); } - (void) ttm_mem_io_lock(man, false); ttm_mem_io_free(map->bo->bdev, &map->bo->mem); - ttm_mem_io_unlock(man); map->virtual = NULL; map->page = NULL; } diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c index ad485d49e19c..1fec54c5fd60 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -193,8 +193,6 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, pgoff_t i; vm_fault_t ret = VM_FAULT_NOPAGE; unsigned long address = vmf->address; - struct ttm_mem_type_manager *man - &bdev->man[bo->mem.mem_type]; /* * Refuse to fault imported pages. This should be handled @@ -233,24 +231,17 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, if (unlikely(ret != 0)) return ret; - err = ttm_mem_io_lock(man, true); + err = ttm_mem_io_reserve(bdev, &bo->mem); if (unlikely(err != 0)) - return VM_FAULT_NOPAGE; - err = ttm_mem_io_reserve_vm(bo); - if (unlikely(err != 0)) { - ret = VM_FAULT_SIGBUS; - goto out_io_unlock; - } + return VM_FAULT_SIGBUS; page_offset = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff - drm_vma_node_start(&bo->base.vma_node); page_last = vma_pages(vma) + vma->vm_pgoff - drm_vma_node_start(&bo->base.vma_node); - if (unlikely(page_offset >= bo->num_pages)) { - ret = VM_FAULT_SIGBUS; - goto out_io_unlock; - } + if (unlikely(page_offset >= bo->num_pages)) + return VM_FAULT_SIGBUS; cvma.vm_page_prot = ttm_io_prot(bo->mem.placement, prot); if (!bo->mem.bus.is_iomem) { @@ -262,10 +253,8 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, }; ttm = bo->ttm; - if (ttm_tt_populate(bo->ttm, &ctx)) { - ret = VM_FAULT_OOM; - goto out_io_unlock; - } + if (ttm_tt_populate(bo->ttm, &ctx)) + return VM_FAULT_OOM; } else { /* Iomem should not be marked encrypted */ cvma.vm_page_prot = pgprot_decrypted(cvma.vm_page_prot); @@ -281,8 +270,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, } else { page = ttm->pages[page_offset]; if (unlikely(!page && i == 0)) { - ret = VM_FAULT_OOM; - goto out_io_unlock; + return VM_FAULT_OOM; } else if (unlikely(!page)) { break; } @@ -300,7 +288,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, /* Never error on prefaulted PTEs */ if (unlikely((ret & VM_FAULT_ERROR))) { if (i == 0) - goto out_io_unlock; + return VM_FAULT_NOPAGE; else break; } @@ -309,10 +297,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, if (unlikely(++page_offset >= page_last)) break; } - ret = VM_FAULT_NOPAGE; -out_io_unlock: - ttm_mem_io_unlock(man); - return ret; + return VM_FAULT_NOPAGE; } EXPORT_SYMBOL(ttm_bo_vm_fault_reserved); diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index 2e308b896d2f..c79ef02450f7 100644 --- a/include/drm/ttm/ttm_bo_api.h +++ b/include/drm/ttm/ttm_bo_api.h @@ -62,8 +62,6 @@ struct ttm_lru_bulk_move; * @is_iomem: is this io memory ? * @size: size in byte * @offset: offset from the base address - * @io_reserved_vm: The VM system has a refcount in @io_reserved_count - * @io_reserved_count: Refcounting the numbers of callers to ttm_mem_io_reserve * * Structure indicating the bus placement of an object. */ @@ -73,11 +71,8 @@ struct ttm_bus_placement { unsigned long size; unsigned long offset; bool is_iomem; - bool io_reserved_vm; - uint64_t io_reserved_count; }; - /** * struct ttm_mem_reg * diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index 6f536caea368..ecf0ceee205b 100644 --- a/include/drm/ttm/ttm_bo_driver.h +++ b/include/drm/ttm/ttm_bo_driver.h @@ -615,20 +615,6 @@ int ttm_bo_device_init(struct ttm_bo_device *bdev, */ void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo); -/** - * ttm_bo_unmap_virtual - * - * @bo: tear down the virtual mappings for this BO - * - * The caller must take ttm_mem_io_lock before calling this function. - */ -void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo); - -int ttm_mem_io_reserve_vm(struct ttm_buffer_object *bo); -void ttm_mem_io_free_vm(struct ttm_buffer_object *bo); -int ttm_mem_io_lock(struct ttm_mem_type_manager *man, bool interruptible); -void ttm_mem_io_unlock(struct ttm_mem_type_manager *man); - void ttm_bo_del_sub_from_lru(struct ttm_buffer_object *bo); void ttm_bo_add_to_lru(struct ttm_buffer_object *bo); -- 2.14.1
Daniel Vetter
2019-Oct-09 15:39 UTC
[Nouveau] [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver
On Mon, Sep 30, 2019 at 03:12:53PM +0200, Christian König wrote:> While working on TTM cleanups I've found that the io_reserve_lru used by > Nouveau is actually not working at all. > > In general we should remove driver specific handling from the memory > management, so this patch moves the io_reserve_lru handling into Nouveau > instead. > > The patch should be functional correct, but is only compile tested! > > Signed-off-by: Christian König <christian.koenig at amd.com>> --- > drivers/gpu/drm/nouveau/nouveau_bo.c | 106 ++++++++++++++++++++++++++-------- > drivers/gpu/drm/nouveau/nouveau_bo.h | 3 + > drivers/gpu/drm/nouveau/nouveau_drv.h | 2 + > drivers/gpu/drm/nouveau/nouveau_ttm.c | 43 +++++++++++++- > 4 files changed, 130 insertions(+), 24 deletions(-) > > diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c > index e918b437af17..9532ad1bd474 100644 > --- a/drivers/gpu/drm/nouveau/nouveau_bo.c > +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c > @@ -139,6 +139,7 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo) > if (unlikely(nvbo->bo.base.filp)) > DRM_ERROR("bo %p still attached to GEM object\n", bo); > WARN_ON(nvbo->pin_refcnt > 0); > + nouveau_bo_del_io_reserve_lru(bo); > nv10_bo_put_tile_region(dev, nvbo->tile, NULL); > kfree(nvbo); > } > @@ -297,6 +298,7 @@ nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 flags, > nouveau_bo_fixup_align(nvbo, flags, &align, &size); > nvbo->bo.mem.num_pages = size >> PAGE_SHIFT; > nouveau_bo_placement_set(nvbo, flags, 0); > + INIT_LIST_HEAD(&nvbo->io_reserve_lru); > > ret = ttm_bo_init(nvbo->bo.bdev, &nvbo->bo, size, type, > &nvbo->placement, align >> PAGE_SHIFT, false, > @@ -566,6 +568,26 @@ nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo) > PAGE_SIZE, DMA_FROM_DEVICE); > } > > +void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo) > +{ > + struct nouveau_drm *drm = nouveau_bdev(bo->bdev); > + struct nouveau_bo *nvbo = nouveau_bo(bo); > + > + mutex_lock(&drm->ttm.io_reserve_mutex); > + list_move_tail(&nvbo->io_reserve_lru, &drm->ttm.io_reserve_lru); > + mutex_unlock(&drm->ttm.io_reserve_mutex); > +} > + > +void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo) > +{ > + struct nouveau_drm *drm = nouveau_bdev(bo->bdev); > + struct nouveau_bo *nvbo = nouveau_bo(bo); > + > + mutex_lock(&drm->ttm.io_reserve_mutex); > + list_del_init(&nvbo->io_reserve_lru); > + mutex_unlock(&drm->ttm.io_reserve_mutex); > +} > + > int > nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible, > bool no_wait_gpu) > @@ -674,8 +696,6 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, > } > > man->func = &nouveau_vram_manager; > - man->io_reserve_fastpath = false; > - man->use_io_reserve_lru = true; > } else { > man->func = &ttm_bo_manager_func; > } > @@ -1304,6 +1324,8 @@ nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict, > if (bo->destroy != nouveau_bo_del_ttm) > return; > > + nouveau_bo_del_io_reserve_lru(bo); > + > if (mem && new_reg->mem_type != TTM_PL_SYSTEM && > mem->mem.page == nvbo->page) { > list_for_each_entry(vma, &nvbo->vma_list, head) { > @@ -1426,6 +1448,30 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp) > filp->private_data); > } > > +static void > +nouveau_ttm_io_mem_free_locked(struct nouveau_drm *drm, struct ttm_mem_reg *reg) > +{ > + struct nouveau_mem *mem = nouveau_mem(reg); > + > + if (!reg->bus.base) > + return; /* already freed */ > + > + if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) { > + switch (reg->mem_type) { > + case TTM_PL_TT: > + if (mem->kind) > + nvif_object_unmap_handle(&mem->mem.object); > + break; > + case TTM_PL_VRAM: > + nvif_object_unmap_handle(&mem->mem.object); > + break; > + default: > + break; > + } > + } > + reg->bus.base = 0; > +} > + > static int > nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) > { > @@ -1433,18 +1479,26 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) > struct nouveau_drm *drm = nouveau_bdev(bdev); > struct nvkm_device *device = nvxx_device(&drm->client.device); > struct nouveau_mem *mem = nouveau_mem(reg); > + struct nouveau_bo *nvbo; > + int ret; > + > + if (reg->bus.base) > + return 0; /* already mapped */ > > reg->bus.addr = NULL; > reg->bus.offset = 0; > reg->bus.size = reg->num_pages << PAGE_SHIFT; > - reg->bus.base = 0; > reg->bus.is_iomem = false; > if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE)) > return -EINVAL; > + > + mutex_lock(&drm->ttm.io_reserve_mutex); > +retry: > switch (reg->mem_type) { > case TTM_PL_SYSTEM: > /* System memory */ > - return 0; > + ret = 0; > + goto out; > case TTM_PL_TT: > #if IS_ENABLED(CONFIG_AGP) > if (drm->agp.bridge) { > @@ -1468,7 +1522,6 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) > } args; > u64 handle, length; > u32 argc = 0; > - int ret; > > switch (mem->mem.object.oclass) { > case NVIF_CLASS_MEM_NV50: > @@ -1492,38 +1545,45 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) > ret = nvif_object_map_handle(&mem->mem.object, > &args, argc, > &handle, &length); > - if (ret != 1) > - return ret ? ret : -EINVAL; > + if (ret != 1) { > + ret = ret ? ret : -EINVAL; > + goto out; > + } > + ret = 0; > > reg->bus.base = 0; > reg->bus.offset = handle; > } > break; > default: > - return -EINVAL; > + ret = -EINVAL; > } > - return 0; > + > +out: > + if (ret == -EAGAIN) { > + nvbo = list_first_entry_or_null(&drm->ttm.io_reserve_lru, > + typeof(*nvbo), > + io_reserve_lru); > + if (nvbo) { > + list_del_init(&nvbo->io_reserve_lru); > + ttm_bo_unmap_virtual(&nvbo->bo); > + nouveau_ttm_io_mem_free_locked(drm, &nvbo->bo.mem); > + goto retry; > + } > + > + } > + mutex_unlock(&drm->ttm.io_reserve_mutex); > + return ret; > } > > static void > nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *reg) > { > struct nouveau_drm *drm = nouveau_bdev(bdev); > - struct nouveau_mem *mem = nouveau_mem(reg); > > - if (drm->client.mem->oclass >= NVIF_CLASS_MEM_NV50) { > - switch (reg->mem_type) { > - case TTM_PL_TT: > - if (mem->kind) > - nvif_object_unmap_handle(&mem->mem.object); > - break; > - case TTM_PL_VRAM: > - nvif_object_unmap_handle(&mem->mem.object); > - break; > - default: > - break; > - } > - } > + mutex_lock(&drm->ttm.io_reserve_mutex); > + nouveau_ttm_io_mem_free_locked(drm, reg); > + mutex_unlock(&drm->ttm.io_reserve_mutex); > } > > static int > diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h > index 62930d834fba..b8a17bc19d7b 100644 > --- a/drivers/gpu/drm/nouveau/nouveau_bo.h > +++ b/drivers/gpu/drm/nouveau/nouveau_bo.h > @@ -17,6 +17,7 @@ struct nouveau_bo { > bool force_coherent; > struct ttm_bo_kmap_obj kmap; > struct list_head head; > + struct list_head io_reserve_lru; > > /* protected by ttm_bo_reserve() */ > struct drm_file *reserved_by; > @@ -92,6 +93,8 @@ int nouveau_bo_validate(struct nouveau_bo *, bool interruptible, > bool no_wait_gpu); > void nouveau_bo_sync_for_device(struct nouveau_bo *nvbo); > void nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo); > +void nouveau_bo_add_io_reserve_lru(struct ttm_buffer_object *bo); > +void nouveau_bo_del_io_reserve_lru(struct ttm_buffer_object *bo); > > /* TODO: submit equivalent to TTM generic API upstream? */ > static inline void __iomem * > diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h > index 70f34cacc552..060e2ab1fd95 100644 > --- a/drivers/gpu/drm/nouveau/nouveau_drv.h > +++ b/drivers/gpu/drm/nouveau/nouveau_drv.h > @@ -158,6 +158,8 @@ struct nouveau_drm { > int type_vram; > int type_host[2]; > int type_ncoh[2]; > + struct mutex io_reserve_mutex; > + struct list_head io_reserve_lru; > } ttm; > > /* GEM interface support */ > diff --git a/drivers/gpu/drm/nouveau/nouveau_ttm.c b/drivers/gpu/drm/nouveau/nouveau_ttm.c > index f0daf958e03a..7f8495066e8b 100644 > --- a/drivers/gpu/drm/nouveau/nouveau_ttm.c > +++ b/drivers/gpu/drm/nouveau/nouveau_ttm.c > @@ -162,13 +162,51 @@ const struct ttm_mem_type_manager_func nv04_gart_manager = { > .debug = nouveau_manager_debug > }; > > +static vm_fault_t nouveau_ttm_fault(struct vm_fault *vmf) > +{ > + struct vm_area_struct *vma = vmf->vma; > + struct ttm_buffer_object *bo = vma->vm_private_data; > + pgprot_t prot; > + vm_fault_t ret; > + > + ret = ttm_bo_vm_reserve(bo, vmf); > + if (ret) > + return ret; > + > + nouveau_bo_del_io_reserve_lru(bo);Isn't this opening up a can of worms (in theory) where a lot of concurrent faults will all fail because they're all removed themselves from the lru, so can't see anyone else to throw out? Same problem really that ttm had (well still has I think for !amdgpu) with its other lru ... Or am I getting totally lost again here? -Daniel> + > + prot = vm_get_page_prot(vma->vm_flags); > + ret = ttm_bo_vm_fault_reserved(vmf, prot, TTM_BO_VM_NUM_PREFAULT); > + if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) > + return ret; > + > + nouveau_bo_add_io_reserve_lru(bo); > + > + dma_resv_unlock(bo->base.resv); > + > + return ret; > +} > + > +static struct vm_operations_struct nouveau_ttm_vm_ops = { > + .fault = nouveau_ttm_fault, > + .open = ttm_bo_vm_open, > + .close = ttm_bo_vm_close, > + .access = ttm_bo_vm_access > +}; > + > int > nouveau_ttm_mmap(struct file *filp, struct vm_area_struct *vma) > { > struct drm_file *file_priv = filp->private_data; > struct nouveau_drm *drm = nouveau_drm(file_priv->minor->dev); > + int ret; > > - return ttm_bo_mmap(filp, vma, &drm->ttm.bdev); > + ret = ttm_bo_mmap(filp, vma, &drm->ttm.bdev); > + if (ret) > + return ret; > + > + vma->vm_ops = &nouveau_ttm_vm_ops; > + return 0; > } > > static int > @@ -272,6 +310,9 @@ nouveau_ttm_init(struct nouveau_drm *drm) > return ret; > } > > + mutex_init(&drm->ttm.io_reserve_mutex); > + INIT_LIST_HEAD(&drm->ttm.io_reserve_lru); > + > NV_INFO(drm, "VRAM: %d MiB\n", (u32)(drm->gem.vram_available >> 20)); > NV_INFO(drm, "GART: %d MiB\n", (u32)(drm->gem.gart_available >> 20)); > return 0; > -- > 2.14.1 >-- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
Christian König
2019-Oct-10 07:54 UTC
[Nouveau] [PATCH 1/2] drm/nouveau: move io_reserve_lru handling into the driver
Am 09.10.19 um 17:39 schrieb Daniel Vetter:> On Mon, Sep 30, 2019 at 03:12:53PM +0200, Christian König wrote: > [SNIP] >> +static vm_fault_t nouveau_ttm_fault(struct vm_fault *vmf) >> +{ >> + struct vm_area_struct *vma = vmf->vma; >> + struct ttm_buffer_object *bo = vma->vm_private_data; >> + pgprot_t prot; >> + vm_fault_t ret; >> + >> + ret = ttm_bo_vm_reserve(bo, vmf); >> + if (ret) >> + return ret; >> + >> + nouveau_bo_del_io_reserve_lru(bo); > Isn't this opening up a can of worms (in theory) where a lot of concurrent > faults will all fail because they're all removed themselves from the lru, > so can't see anyone else to throw out? > > Same problem really that ttm had (well still has I think for !amdgpu) with > its other lru ... > > Or am I getting totally lost again here?No, your are pretty much correct. But we are not opening up that can of worms, it has been here for like always. You just need to imagine that you have two processes kicking out the page tables of each other before somebody had a chance to make progress. This will just loop forever :) The correct solution is to not use a BO based approach at all, but rather a window based approach where you have an LRU of the window you gave out last to access something. This way you have a much larger number of windows you can use (4k if you use 64kB window size on a 256MB BAR). This way you can safely assume that the number of windows you have available is always larger than the number of faults you can process at the same time. But that would require quite a bunch of changes for TTM as well as nouveau. BTW: This is one of the reasons (additional to that it is horrible slow) why we never used the remapping feature on AMD hardware for Linux. Regards, Christian.> -Daniel >