diff options
Diffstat (limited to 'drivers/gpu/drm/xe/xe_bo.c')
| -rw-r--r-- | drivers/gpu/drm/xe/xe_bo.c | 1524 |
1 files changed, 1150 insertions, 374 deletions
diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c index 6603a0ea79c5..7aa2c17825da 100644 --- a/drivers/gpu/drm/xe/xe_bo.c +++ b/drivers/gpu/drm/xe/xe_bo.c @@ -6,14 +6,18 @@ #include "xe_bo.h" #include <linux/dma-buf.h> +#include <linux/nospec.h> #include <drm/drm_drv.h> #include <drm/drm_gem_ttm_helper.h> #include <drm/drm_managed.h> +#include <drm/ttm/ttm_backup.h> #include <drm/ttm/ttm_device.h> #include <drm/ttm/ttm_placement.h> #include <drm/ttm/ttm_tt.h> -#include <drm/xe_drm.h> +#include <uapi/drm/xe_drm.h> + +#include <kunit/static_stub.h> #include "xe_device.h" #include "xe_dma_buf.h" @@ -22,9 +26,12 @@ #include "xe_gt.h" #include "xe_map.h" #include "xe_migrate.h" +#include "xe_pm.h" #include "xe_preempt_fence.h" +#include "xe_pxp.h" #include "xe_res_cursor.h" -#include "xe_trace.h" +#include "xe_shrinker.h" +#include "xe_trace_bo.h" #include "xe_ttm_stolen_mgr.h" #include "xe_vm.h" @@ -48,6 +55,8 @@ static struct ttm_placement sys_placement = { .placement = &sys_placement_flags, }; +static struct ttm_placement purge_placement; + static const struct ttm_place tt_placement_flags[] = { { .fpfn = 0, @@ -95,6 +104,20 @@ bool xe_bo_is_stolen(struct xe_bo *bo) } /** + * xe_bo_has_single_placement - check if BO is placed only in one memory location + * @bo: The BO + * + * This function checks whether a given BO is placed in only one memory location. + * + * Returns: true if the BO is placed in a single memory location, false otherwise. + * + */ +bool xe_bo_has_single_placement(struct xe_bo *bo) +{ + return bo->placement.num_placement == 1; +} + +/** * xe_bo_is_stolen_devmem - check if BO is of stolen type accessed via PCI BAR * @bo: The BO * @@ -109,9 +132,25 @@ bool xe_bo_is_stolen_devmem(struct xe_bo *bo) GRAPHICS_VERx100(xe_bo_device(bo)) >= 1270; } +/** + * xe_bo_is_vm_bound - check if BO has any mappings through VM_BIND + * @bo: The BO + * + * Check if a given bo is bound through VM_BIND. This requires the + * reservation lock for the BO to be held. + * + * Returns: boolean + */ +bool xe_bo_is_vm_bound(struct xe_bo *bo) +{ + xe_bo_assert_held(bo); + + return !list_empty(&bo->ttm.base.gpuva.list); +} + static bool xe_bo_is_user(struct xe_bo *bo) { - return bo->flags & XE_BO_CREATE_USER_BIT; + return bo->flags & XE_BO_FLAG_USER; } static struct xe_migrate * @@ -124,55 +163,69 @@ mem_type_to_migrate(struct xe_device *xe, u32 mem_type) return tile->migrate; } -static struct xe_mem_region *res_to_mem_region(struct ttm_resource *res) +static struct xe_vram_region *res_to_mem_region(struct ttm_resource *res) { struct xe_device *xe = ttm_to_xe_device(res->bo->bdev); struct ttm_resource_manager *mgr; + struct xe_ttm_vram_mgr *vram_mgr; xe_assert(xe, resource_is_vram(res)); mgr = ttm_manager_type(&xe->ttm, res->mem_type); - return to_xe_ttm_vram_mgr(mgr)->vram; + vram_mgr = to_xe_ttm_vram_mgr(mgr); + + return container_of(vram_mgr, struct xe_vram_region, ttm); } static void try_add_system(struct xe_device *xe, struct xe_bo *bo, u32 bo_flags, u32 *c) { - if (bo_flags & XE_BO_CREATE_SYSTEM_BIT) { + if (bo_flags & XE_BO_FLAG_SYSTEM) { xe_assert(xe, *c < ARRAY_SIZE(bo->placements)); bo->placements[*c] = (struct ttm_place) { .mem_type = XE_PL_TT, }; *c += 1; - - if (bo->props.preferred_mem_type == XE_BO_PROPS_INVALID) - bo->props.preferred_mem_type = XE_PL_TT; } } +static bool force_contiguous(u32 bo_flags) +{ + if (bo_flags & XE_BO_FLAG_STOLEN) + return true; /* users expect this */ + else if (bo_flags & XE_BO_FLAG_PINNED && + !(bo_flags & XE_BO_FLAG_PINNED_LATE_RESTORE)) + return true; /* needs vmap */ + + /* + * For eviction / restore on suspend / resume objects pinned in VRAM + * must be contiguous, also only contiguous BOs support xe_bo_vmap. + */ + return bo_flags & XE_BO_FLAG_NEEDS_CPU_ACCESS && + bo_flags & XE_BO_FLAG_PINNED; +} + static void add_vram(struct xe_device *xe, struct xe_bo *bo, struct ttm_place *places, u32 bo_flags, u32 mem_type, u32 *c) { struct ttm_place place = { .mem_type = mem_type }; - struct xe_mem_region *vram; + struct ttm_resource_manager *mgr = ttm_manager_type(&xe->ttm, mem_type); + struct xe_ttm_vram_mgr *vram_mgr = to_xe_ttm_vram_mgr(mgr); + + struct xe_vram_region *vram; u64 io_size; xe_assert(xe, *c < ARRAY_SIZE(bo->placements)); - vram = to_xe_ttm_vram_mgr(ttm_manager_type(&xe->ttm, mem_type))->vram; + vram = container_of(vram_mgr, struct xe_vram_region, ttm); xe_assert(xe, vram && vram->usable_size); io_size = vram->io_size; - /* - * For eviction / restore on suspend / resume objects - * pinned in VRAM must be contiguous - */ - if (bo_flags & (XE_BO_CREATE_PINNED_BIT | - XE_BO_CREATE_GGTT_BIT)) + if (force_contiguous(bo_flags)) place.flags |= TTM_PL_FLAG_CONTIGUOUS; if (io_size < vram->usable_size) { - if (bo_flags & XE_BO_NEEDS_CPU_ACCESS) { + if (bo_flags & XE_BO_FLAG_NEEDS_CPU_ACCESS) { place.fpfn = 0; place.lpfn = io_size >> PAGE_SHIFT; } else { @@ -181,37 +234,26 @@ static void add_vram(struct xe_device *xe, struct xe_bo *bo, } places[*c] = place; *c += 1; - - if (bo->props.preferred_mem_type == XE_BO_PROPS_INVALID) - bo->props.preferred_mem_type = mem_type; } static void try_add_vram(struct xe_device *xe, struct xe_bo *bo, u32 bo_flags, u32 *c) { - if (bo->props.preferred_gt == XE_GT1) { - if (bo_flags & XE_BO_CREATE_VRAM1_BIT) - add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM1, c); - if (bo_flags & XE_BO_CREATE_VRAM0_BIT) - add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM0, c); - } else { - if (bo_flags & XE_BO_CREATE_VRAM0_BIT) - add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM0, c); - if (bo_flags & XE_BO_CREATE_VRAM1_BIT) - add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM1, c); - } + if (bo_flags & XE_BO_FLAG_VRAM0) + add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM0, c); + if (bo_flags & XE_BO_FLAG_VRAM1) + add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM1, c); } static void try_add_stolen(struct xe_device *xe, struct xe_bo *bo, u32 bo_flags, u32 *c) { - if (bo_flags & XE_BO_CREATE_STOLEN_BIT) { + if (bo_flags & XE_BO_FLAG_STOLEN) { xe_assert(xe, *c < ARRAY_SIZE(bo->placements)); bo->placements[*c] = (struct ttm_place) { .mem_type = XE_PL_STOLEN, - .flags = bo_flags & (XE_BO_CREATE_PINNED_BIT | - XE_BO_CREATE_GGTT_BIT) ? + .flags = force_contiguous(bo_flags) ? TTM_PL_FLAG_CONTIGUOUS : 0, }; *c += 1; @@ -223,17 +265,8 @@ static int __xe_bo_placement_for_flags(struct xe_device *xe, struct xe_bo *bo, { u32 c = 0; - bo->props.preferred_mem_type = XE_BO_PROPS_INVALID; - - /* The order of placements should indicate preferred location */ - - if (bo->props.preferred_mem_class == DRM_XE_MEM_REGION_CLASS_SYSMEM) { - try_add_system(xe, bo, bo_flags, &c); - try_add_vram(xe, bo, bo_flags, &c); - } else { - try_add_vram(xe, bo, bo_flags, &c); - try_add_system(xe, bo, bo_flags, &c); - } + try_add_vram(xe, bo, bo_flags, &c); + try_add_system(xe, bo, bo_flags, &c); try_add_stolen(xe, bo, bo_flags, &c); if (!c) @@ -257,6 +290,10 @@ int xe_bo_placement_for_flags(struct xe_device *xe, struct xe_bo *bo, static void xe_evict_flags(struct ttm_buffer_object *tbo, struct ttm_placement *placement) { + struct xe_device *xe = container_of(tbo->bdev, typeof(*xe), ttm); + bool device_unplugged = drm_dev_is_unplugged(&xe->drm); + struct xe_bo *bo; + if (!xe_bo_is_xe_bo(tbo)) { /* Don't handle scatter gather BOs */ if (tbo->type == ttm_bo_type_sg) { @@ -264,10 +301,21 @@ static void xe_evict_flags(struct ttm_buffer_object *tbo, return; } + *placement = device_unplugged ? purge_placement : sys_placement; + return; + } + + bo = ttm_to_xe_bo(tbo); + if (bo->flags & XE_BO_FLAG_CPU_ADDR_MIRROR) { *placement = sys_placement; return; } + if (device_unplugged && !tbo->base.dma_buf) { + *placement = purge_placement; + return; + } + /* * For xe, sg bos that are evicted to system just triggers a * rebind of the sg list upon subsequent validation to XE_PL_TT. @@ -285,11 +333,15 @@ static void xe_evict_flags(struct ttm_buffer_object *tbo, } } +/* struct xe_ttm_tt - Subclassed ttm_tt for xe */ struct xe_ttm_tt { struct ttm_tt ttm; - struct device *dev; + /** @xe - The xe device */ + struct xe_device *xe; struct sg_table sgt; struct sg_table *sg; + /** @purgeable: Whether the content of the pages of @ttm is purgeable. */ + bool purgeable; }; static int xe_tt_map_sg(struct ttm_tt *tt) @@ -298,7 +350,8 @@ static int xe_tt_map_sg(struct ttm_tt *tt) unsigned long num_pages = tt->num_pages; int ret; - XE_WARN_ON(tt->page_flags & TTM_TT_FLAG_EXTERNAL); + XE_WARN_ON((tt->page_flags & TTM_TT_FLAG_EXTERNAL) && + !(tt->page_flags & TTM_TT_FLAG_EXTERNAL_MAPPABLE)); if (xe_tt->sg) return 0; @@ -306,13 +359,13 @@ static int xe_tt_map_sg(struct ttm_tt *tt) ret = sg_alloc_table_from_pages_segment(&xe_tt->sgt, tt->pages, num_pages, 0, (u64)num_pages << PAGE_SHIFT, - xe_sg_segment_size(xe_tt->dev), + xe_sg_segment_size(xe_tt->xe->drm.dev), GFP_KERNEL); if (ret) return ret; xe_tt->sg = &xe_tt->sgt; - ret = dma_map_sgtable(xe_tt->dev, xe_tt->sg, DMA_BIDIRECTIONAL, + ret = dma_map_sgtable(xe_tt->xe->drm.dev, xe_tt->sg, DMA_BIDIRECTIONAL, DMA_ATTR_SKIP_CPU_SYNC); if (ret) { sg_free_table(xe_tt->sg); @@ -323,6 +376,18 @@ static int xe_tt_map_sg(struct ttm_tt *tt) return 0; } +static void xe_tt_unmap_sg(struct ttm_tt *tt) +{ + struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm); + + if (xe_tt->sg) { + dma_unmap_sgtable(xe_tt->xe->drm.dev, xe_tt->sg, + DMA_BIDIRECTIONAL, 0); + sg_free_table(xe_tt->sg); + xe_tt->sg = NULL; + } +} + struct sg_table *xe_bo_sg(struct xe_bo *bo) { struct ttm_tt *tt = bo->ttm.ttm; @@ -331,96 +396,152 @@ struct sg_table *xe_bo_sg(struct xe_bo *bo) return xe_tt->sg; } +/* + * Account ttm pages against the device shrinker's shrinkable and + * purgeable counts. + */ +static void xe_ttm_tt_account_add(struct ttm_tt *tt) +{ + struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm); + + if (xe_tt->purgeable) + xe_shrinker_mod_pages(xe_tt->xe->mem.shrinker, 0, tt->num_pages); + else + xe_shrinker_mod_pages(xe_tt->xe->mem.shrinker, tt->num_pages, 0); +} + +static void xe_ttm_tt_account_subtract(struct ttm_tt *tt) +{ + struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm); + + if (xe_tt->purgeable) + xe_shrinker_mod_pages(xe_tt->xe->mem.shrinker, 0, -(long)tt->num_pages); + else + xe_shrinker_mod_pages(xe_tt->xe->mem.shrinker, -(long)tt->num_pages, 0); +} + static struct ttm_tt *xe_ttm_tt_create(struct ttm_buffer_object *ttm_bo, u32 page_flags) { struct xe_bo *bo = ttm_to_xe_bo(ttm_bo); struct xe_device *xe = xe_bo_device(bo); - struct xe_ttm_tt *tt; + struct xe_ttm_tt *xe_tt; + struct ttm_tt *tt; unsigned long extra_pages; - enum ttm_caching caching; + enum ttm_caching caching = ttm_cached; int err; - tt = kzalloc(sizeof(*tt), GFP_KERNEL); - if (!tt) + xe_tt = kzalloc(sizeof(*xe_tt), GFP_KERNEL); + if (!xe_tt) return NULL; - tt->dev = xe->drm.dev; + tt = &xe_tt->ttm; + xe_tt->xe = xe; extra_pages = 0; if (xe_bo_needs_ccs_pages(bo)) extra_pages = DIV_ROUND_UP(xe_device_ccs_bytes(xe, bo->size), PAGE_SIZE); - switch (bo->cpu_caching) { - case DRM_XE_GEM_CPU_CACHING_WC: - caching = ttm_write_combined; - break; - default: - caching = ttm_cached; - break; + /* + * DGFX system memory is always WB / ttm_cached, since + * other caching modes are only supported on x86. DGFX + * GPU system memory accesses are always coherent with the + * CPU. + */ + if (!IS_DGFX(xe)) { + switch (bo->cpu_caching) { + case DRM_XE_GEM_CPU_CACHING_WC: + caching = ttm_write_combined; + break; + default: + caching = ttm_cached; + break; + } + + WARN_ON((bo->flags & XE_BO_FLAG_USER) && !bo->cpu_caching); + + /* + * Display scanout is always non-coherent with the CPU cache. + * + * For Xe_LPG and beyond, PPGTT PTE lookups are also + * non-coherent and require a CPU:WC mapping. + */ + if ((!bo->cpu_caching && bo->flags & XE_BO_FLAG_SCANOUT) || + (xe->info.graphics_verx100 >= 1270 && + bo->flags & XE_BO_FLAG_PAGETABLE)) + caching = ttm_write_combined; } - WARN_ON((bo->flags & XE_BO_CREATE_USER_BIT) && !bo->cpu_caching); + if (bo->flags & XE_BO_FLAG_NEEDS_UC) { + /* + * Valid only for internally-created buffers only, for + * which cpu_caching is never initialized. + */ + xe_assert(xe, bo->cpu_caching == 0); + caching = ttm_uncached; + } - /* - * Display scanout is always non-coherent with the CPU cache. - * - * For Xe_LPG and beyond, PPGTT PTE lookups are also non-coherent and - * require a CPU:WC mapping. - */ - if ((!bo->cpu_caching && bo->flags & XE_BO_SCANOUT_BIT) || - (xe->info.graphics_verx100 >= 1270 && bo->flags & XE_BO_PAGETABLE)) - caching = ttm_write_combined; + if (ttm_bo->type != ttm_bo_type_sg) + page_flags |= TTM_TT_FLAG_EXTERNAL | TTM_TT_FLAG_EXTERNAL_MAPPABLE; - err = ttm_tt_init(&tt->ttm, &bo->ttm, page_flags, caching, extra_pages); + err = ttm_tt_init(tt, &bo->ttm, page_flags, caching, extra_pages); if (err) { - kfree(tt); + kfree(xe_tt); return NULL; } - return &tt->ttm; + if (ttm_bo->type != ttm_bo_type_sg) { + err = ttm_tt_setup_backup(tt); + if (err) { + ttm_tt_fini(tt); + kfree(xe_tt); + return NULL; + } + } + + return tt; } static int xe_ttm_tt_populate(struct ttm_device *ttm_dev, struct ttm_tt *tt, struct ttm_operation_ctx *ctx) { + struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm); int err; /* * dma-bufs are not populated with pages, and the dma- * addresses are set up when moved to XE_PL_TT. */ - if (tt->page_flags & TTM_TT_FLAG_EXTERNAL) + if ((tt->page_flags & TTM_TT_FLAG_EXTERNAL) && + !(tt->page_flags & TTM_TT_FLAG_EXTERNAL_MAPPABLE)) return 0; - err = ttm_pool_alloc(&ttm_dev->pool, tt, ctx); + if (ttm_tt_is_backed_up(tt) && !xe_tt->purgeable) { + err = ttm_tt_restore(ttm_dev, tt, ctx); + } else { + ttm_tt_clear_backed_up(tt); + err = ttm_pool_alloc(&ttm_dev->pool, tt, ctx); + } if (err) return err; - /* A follow up may move this xe_bo_move when BO is moved to XE_PL_TT */ - err = xe_tt_map_sg(tt); - if (err) - ttm_pool_free(&ttm_dev->pool, tt); + xe_tt->purgeable = false; + xe_ttm_tt_account_add(tt); - return err; + return 0; } static void xe_ttm_tt_unpopulate(struct ttm_device *ttm_dev, struct ttm_tt *tt) { - struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm); - - if (tt->page_flags & TTM_TT_FLAG_EXTERNAL) + if ((tt->page_flags & TTM_TT_FLAG_EXTERNAL) && + !(tt->page_flags & TTM_TT_FLAG_EXTERNAL_MAPPABLE)) return; - if (xe_tt->sg) { - dma_unmap_sgtable(xe_tt->dev, xe_tt->sg, - DMA_BIDIRECTIONAL, 0); - sg_free_table(xe_tt->sg); - xe_tt->sg = NULL; - } + xe_tt_unmap_sg(tt); - return ttm_pool_free(&ttm_dev->pool, tt); + ttm_pool_free(&ttm_dev->pool, tt); + xe_ttm_tt_account_subtract(tt); } static void xe_ttm_tt_destroy(struct ttm_device *ttm_dev, struct ttm_tt *tt) @@ -429,6 +550,14 @@ static void xe_ttm_tt_destroy(struct ttm_device *ttm_dev, struct ttm_tt *tt) kfree(tt); } +static bool xe_ttm_resource_visible(struct ttm_resource *mem) +{ + struct xe_ttm_vram_mgr_resource *vres = + to_xe_ttm_vram_mgr_resource(mem); + + return vres->used_visible_size == mem->size; +} + static int xe_ttm_io_mem_reserve(struct ttm_device *bdev, struct ttm_resource *mem) { @@ -440,11 +569,9 @@ static int xe_ttm_io_mem_reserve(struct ttm_device *bdev, return 0; case XE_PL_VRAM0: case XE_PL_VRAM1: { - struct xe_ttm_vram_mgr_resource *vres = - to_xe_ttm_vram_mgr_resource(mem); - struct xe_mem_region *vram = res_to_mem_region(mem); + struct xe_vram_region *vram = res_to_mem_region(mem); - if (vres->used_visible_size < mem->size) + if (!xe_ttm_resource_visible(mem)) return -EINVAL; mem->bus.offset = mem->start << PAGE_SHIFT; @@ -457,7 +584,7 @@ static int xe_ttm_io_mem_reserve(struct ttm_device *bdev, mem->bus.offset += vram->io_start; mem->bus.is_iomem = true; -#if !defined(CONFIG_X86) +#if !IS_ENABLED(CONFIG_X86) mem->bus.caching = ttm_write_combined; #endif return 0; @@ -546,11 +673,20 @@ static int xe_bo_move_dmabuf(struct ttm_buffer_object *ttm_bo, struct xe_ttm_tt *xe_tt = container_of(ttm_bo->ttm, struct xe_ttm_tt, ttm); struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev); + bool device_unplugged = drm_dev_is_unplugged(&xe->drm); struct sg_table *sg; xe_assert(xe, attach); xe_assert(xe, ttm_bo->ttm); + if (device_unplugged && new_res->mem_type == XE_PL_SYSTEM && + ttm_bo->sg) { + dma_resv_wait_timeout(ttm_bo->base.resv, DMA_RESV_USAGE_BOOKKEEP, + false, MAX_SCHEDULE_TIMEOUT); + dma_buf_unmap_attachment(attach, ttm_bo->sg, DMA_BIDIRECTIONAL); + ttm_bo->sg = NULL; + } + if (new_res->mem_type == XE_PL_SYSTEM) goto out; @@ -649,33 +785,57 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict, bool handle_system_ccs = (!IS_DGFX(xe) && xe_bo_needs_ccs_pages(bo) && ttm && ttm_tt_is_populated(ttm)) ? true : false; int ret = 0; + /* Bo creation path, moving to system or TT. */ if ((!old_mem && ttm) && !handle_system_ccs) { - ttm_bo_move_null(ttm_bo, new_mem); - return 0; + if (new_mem->mem_type == XE_PL_TT) + ret = xe_tt_map_sg(ttm); + if (!ret) + ttm_bo_move_null(ttm_bo, new_mem); + goto out; } if (ttm_bo->type == ttm_bo_type_sg) { ret = xe_bo_move_notify(bo, ctx); if (!ret) ret = xe_bo_move_dmabuf(ttm_bo, new_mem); - goto out; + return ret; } tt_has_data = ttm && (ttm_tt_is_populated(ttm) || (ttm->page_flags & TTM_TT_FLAG_SWAPPED)); - move_lacks_source = handle_system_ccs ? (!bo->ccs_cleared) : - (!mem_type_is_vram(old_mem_type) && !tt_has_data); + move_lacks_source = !old_mem || (handle_system_ccs ? (!bo->ccs_cleared) : + (!mem_type_is_vram(old_mem_type) && !tt_has_data)); needs_clear = (ttm && ttm->page_flags & TTM_TT_FLAG_ZERO_ALLOC) || (!ttm && ttm_bo->type == ttm_bo_type_device); + if (new_mem->mem_type == XE_PL_TT) { + ret = xe_tt_map_sg(ttm); + if (ret) + goto out; + } + if ((move_lacks_source && !needs_clear)) { ttm_bo_move_null(ttm_bo, new_mem); goto out; } + if (!move_lacks_source && (bo->flags & XE_BO_FLAG_CPU_ADDR_MIRROR) && + new_mem->mem_type == XE_PL_SYSTEM) { + ret = xe_svm_bo_evict(bo); + if (!ret) { + drm_dbg(&xe->drm, "Evict system allocator BO success\n"); + ttm_bo_move_null(ttm_bo, new_mem); + } else { + drm_dbg(&xe->drm, "Evict system allocator BO failed=%pe\n", + ERR_PTR(ret)); + } + + goto out; + } + if (old_mem_type == XE_PL_SYSTEM && new_mem->mem_type == XE_PL_TT && !handle_system_ccs) { ttm_bo_move_null(ttm_bo, new_mem); goto out; @@ -701,7 +861,7 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict, new_mem->mem_type == XE_PL_SYSTEM) { long timeout = dma_resv_wait_timeout(ttm_bo->base.resv, DMA_RESV_USAGE_BOOKKEEP, - true, + false, MAX_SCHEDULE_TIMEOUT); if (timeout < 0) { ret = timeout; @@ -737,86 +897,276 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict, xe_assert(xe, migrate); trace_xe_bo_move(bo, new_mem->mem_type, old_mem_type, move_lacks_source); - xe_device_mem_access_get(xe); - - if (xe_bo_is_pinned(bo) && !xe_bo_is_user(bo)) { + if (xe_rpm_reclaim_safe(xe)) { /* - * Kernel memory that is pinned should only be moved on suspend - * / resume, some of the pinned memory is required for the - * device to resume / use the GPU to move other evicted memory - * (user memory) around. This likely could be optimized a bit - * futher where we find the minimum set of pinned memory - * required for resume but for simplity doing a memcpy for all - * pinned memory. + * We might be called through swapout in the validation path of + * another TTM device, so acquire rpm here. */ - ret = xe_bo_vmap(bo); - if (!ret) { - ret = ttm_bo_move_memcpy(ttm_bo, ctx, new_mem); - - /* Create a new VMAP once kernel BO back in VRAM */ - if (!ret && resource_is_vram(new_mem)) { - struct xe_mem_region *vram = res_to_mem_region(new_mem); - void __iomem *new_addr = vram->mapping + - (new_mem->start << PAGE_SHIFT); + xe_pm_runtime_get(xe); + } else { + drm_WARN_ON(&xe->drm, handle_system_ccs); + xe_pm_runtime_get_noresume(xe); + } - if (XE_WARN_ON(new_mem->start == XE_BO_INVALID_OFFSET)) { - ret = -EINVAL; - xe_device_mem_access_put(xe); - goto out; - } + if (move_lacks_source) { + u32 flags = 0; - xe_assert(xe, new_mem->start == - bo->placements->fpfn); + if (mem_type_is_vram(new_mem->mem_type)) + flags |= XE_MIGRATE_CLEAR_FLAG_FULL; + else if (handle_system_ccs) + flags |= XE_MIGRATE_CLEAR_FLAG_CCS_DATA; - iosys_map_set_vaddr_iomem(&bo->vmap, new_addr); - } - } + fence = xe_migrate_clear(migrate, bo, new_mem, flags); } else { - if (move_lacks_source) - fence = xe_migrate_clear(migrate, bo, new_mem); - else - fence = xe_migrate_copy(migrate, bo, bo, old_mem, - new_mem, handle_system_ccs); - if (IS_ERR(fence)) { - ret = PTR_ERR(fence); - xe_device_mem_access_put(xe); - goto out; - } - if (!move_lacks_source) { - ret = ttm_bo_move_accel_cleanup(ttm_bo, fence, evict, - true, new_mem); - if (ret) { - dma_fence_wait(fence, false); - ttm_bo_move_null(ttm_bo, new_mem); - ret = 0; - } - } else { - /* - * ttm_bo_move_accel_cleanup() may blow up if - * bo->resource == NULL, so just attach the - * fence and set the new resource. - */ - dma_resv_add_fence(ttm_bo->base.resv, fence, - DMA_RESV_USAGE_KERNEL); + fence = xe_migrate_copy(migrate, bo, bo, old_mem, new_mem, + handle_system_ccs); + } + if (IS_ERR(fence)) { + ret = PTR_ERR(fence); + xe_pm_runtime_put(xe); + goto out; + } + if (!move_lacks_source) { + ret = ttm_bo_move_accel_cleanup(ttm_bo, fence, evict, true, + new_mem); + if (ret) { + dma_fence_wait(fence, false); ttm_bo_move_null(ttm_bo, new_mem); + ret = 0; } - - dma_fence_put(fence); + } else { + /* + * ttm_bo_move_accel_cleanup() may blow up if + * bo->resource == NULL, so just attach the + * fence and set the new resource. + */ + dma_resv_add_fence(ttm_bo->base.resv, fence, + DMA_RESV_USAGE_KERNEL); + ttm_bo_move_null(ttm_bo, new_mem); } - xe_device_mem_access_put(xe); + dma_fence_put(fence); + xe_pm_runtime_put(xe); out: + if ((!ttm_bo->resource || ttm_bo->resource->mem_type == XE_PL_SYSTEM) && + ttm_bo->ttm) { + long timeout = dma_resv_wait_timeout(ttm_bo->base.resv, + DMA_RESV_USAGE_KERNEL, + false, + MAX_SCHEDULE_TIMEOUT); + if (timeout < 0) + ret = timeout; + + xe_tt_unmap_sg(ttm_bo->ttm); + } + return ret; +} + +static long xe_bo_shrink_purge(struct ttm_operation_ctx *ctx, + struct ttm_buffer_object *bo, + unsigned long *scanned) +{ + long lret; + /* Fake move to system, without copying data. */ + if (bo->resource->mem_type != XE_PL_SYSTEM) { + struct ttm_resource *new_resource; + + lret = ttm_bo_wait_ctx(bo, ctx); + if (lret) + return lret; + + lret = ttm_bo_mem_space(bo, &sys_placement, &new_resource, ctx); + if (lret) + return lret; + + xe_tt_unmap_sg(bo->ttm); + ttm_bo_move_null(bo, new_resource); + } + + *scanned += bo->ttm->num_pages; + lret = ttm_bo_shrink(ctx, bo, (struct ttm_bo_shrink_flags) + {.purge = true, + .writeback = false, + .allow_move = false}); + + if (lret > 0) + xe_ttm_tt_account_subtract(bo->ttm); + + return lret; +} + +static bool +xe_bo_eviction_valuable(struct ttm_buffer_object *bo, const struct ttm_place *place) +{ + struct drm_gpuvm_bo *vm_bo; + + if (!ttm_bo_eviction_valuable(bo, place)) + return false; + + if (!xe_bo_is_xe_bo(bo)) + return true; + + drm_gem_for_each_gpuvm_bo(vm_bo, &bo->base) { + if (xe_vm_is_validating(gpuvm_to_vm(vm_bo->vm))) + return false; + } + + return true; +} + +/** + * xe_bo_shrink() - Try to shrink an xe bo. + * @ctx: The struct ttm_operation_ctx used for shrinking. + * @bo: The TTM buffer object whose pages to shrink. + * @flags: Flags governing the shrink behaviour. + * @scanned: Pointer to a counter of the number of pages + * attempted to shrink. + * + * Try to shrink- or purge a bo, and if it succeeds, unmap dma. + * Note that we need to be able to handle also non xe bos + * (ghost bos), but only if the struct ttm_tt is embedded in + * a struct xe_ttm_tt. When the function attempts to shrink + * the pages of a buffer object, The value pointed to by @scanned + * is updated. + * + * Return: The number of pages shrunken or purged, or negative error + * code on failure. + */ +long xe_bo_shrink(struct ttm_operation_ctx *ctx, struct ttm_buffer_object *bo, + const struct xe_bo_shrink_flags flags, + unsigned long *scanned) +{ + struct ttm_tt *tt = bo->ttm; + struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm); + struct ttm_place place = {.mem_type = bo->resource->mem_type}; + struct xe_bo *xe_bo = ttm_to_xe_bo(bo); + struct xe_device *xe = xe_tt->xe; + bool needs_rpm; + long lret = 0L; + + if (!(tt->page_flags & TTM_TT_FLAG_EXTERNAL_MAPPABLE) || + (flags.purge && !xe_tt->purgeable)) + return -EBUSY; + + if (!xe_bo_eviction_valuable(bo, &place)) + return -EBUSY; + + if (!xe_bo_is_xe_bo(bo) || !xe_bo_get_unless_zero(xe_bo)) + return xe_bo_shrink_purge(ctx, bo, scanned); + + if (xe_tt->purgeable) { + if (bo->resource->mem_type != XE_PL_SYSTEM) + lret = xe_bo_move_notify(xe_bo, ctx); + if (!lret) + lret = xe_bo_shrink_purge(ctx, bo, scanned); + goto out_unref; + } + + /* System CCS needs gpu copy when moving PL_TT -> PL_SYSTEM */ + needs_rpm = (!IS_DGFX(xe) && bo->resource->mem_type != XE_PL_SYSTEM && + xe_bo_needs_ccs_pages(xe_bo)); + if (needs_rpm && !xe_pm_runtime_get_if_active(xe)) + goto out_unref; + + *scanned += tt->num_pages; + lret = ttm_bo_shrink(ctx, bo, (struct ttm_bo_shrink_flags) + {.purge = false, + .writeback = flags.writeback, + .allow_move = true}); + if (needs_rpm) + xe_pm_runtime_put(xe); + + if (lret > 0) + xe_ttm_tt_account_subtract(tt); + +out_unref: + xe_bo_put(xe_bo); + + return lret; +} + +/** + * xe_bo_notifier_prepare_pinned() - Prepare a pinned VRAM object to be backed + * up in system memory. + * @bo: The buffer object to prepare. + * + * On successful completion, the object backup pages are allocated. Expectation + * is that this is called from the PM notifier, prior to suspend/hibernation. + * + * Return: 0 on success. Negative error code on failure. + */ +int xe_bo_notifier_prepare_pinned(struct xe_bo *bo) +{ + struct xe_device *xe = ttm_to_xe_device(bo->ttm.bdev); + struct xe_bo *backup; + int ret = 0; + + xe_bo_lock(bo, false); + + xe_assert(xe, !bo->backup_obj); + + /* + * Since this is called from the PM notifier we might have raced with + * someone unpinning this after we dropped the pinned list lock and + * grabbing the above bo lock. + */ + if (!xe_bo_is_pinned(bo)) + goto out_unlock_bo; + + if (!xe_bo_is_vram(bo)) + goto out_unlock_bo; + + if (bo->flags & XE_BO_FLAG_PINNED_NORESTORE) + goto out_unlock_bo; + + backup = ___xe_bo_create_locked(xe, NULL, NULL, bo->ttm.base.resv, NULL, bo->size, + DRM_XE_GEM_CPU_CACHING_WB, ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | XE_BO_FLAG_NEEDS_CPU_ACCESS | + XE_BO_FLAG_PINNED); + if (IS_ERR(backup)) { + ret = PTR_ERR(backup); + goto out_unlock_bo; + } + + backup->parent_obj = xe_bo_get(bo); /* Released by bo_destroy */ + ttm_bo_pin(&backup->ttm); + bo->backup_obj = backup; + +out_unlock_bo: + xe_bo_unlock(bo); + return ret; +} + +/** + * xe_bo_notifier_unprepare_pinned() - Undo the previous prepare operation. + * @bo: The buffer object to undo the prepare for. + * + * Always returns 0. The backup object is removed, if still present. Expectation + * it that this called from the PM notifier when undoing the prepare step. + * + * Return: Always returns 0. + */ +int xe_bo_notifier_unprepare_pinned(struct xe_bo *bo) +{ + xe_bo_lock(bo, false); + if (bo->backup_obj) { + ttm_bo_unpin(&bo->backup_obj->ttm); + xe_bo_put(bo->backup_obj); + bo->backup_obj = NULL; + } + xe_bo_unlock(bo); + + return 0; } /** * xe_bo_evict_pinned() - Evict a pinned VRAM object to system memory * @bo: The buffer object to move. * - * On successful completion, the object memory will be moved to sytem memory. - * This function blocks until the object has been fully moved. + * On successful completion, the object memory will be moved to system memory. * * This is needed to for special handling of pinned VRAM object during * suspend-resume. @@ -825,61 +1175,99 @@ out: */ int xe_bo_evict_pinned(struct xe_bo *bo) { - struct ttm_place place = { - .mem_type = XE_PL_TT, - }; - struct ttm_placement placement = { - .placement = &place, - .num_placement = 1, - }; - struct ttm_operation_ctx ctx = { - .interruptible = false, - }; - struct ttm_resource *new_mem; - int ret; + struct xe_device *xe = ttm_to_xe_device(bo->ttm.bdev); + struct xe_bo *backup = bo->backup_obj; + bool backup_created = false; + bool unmap = false; + int ret = 0; - xe_bo_assert_held(bo); + xe_bo_lock(bo, false); - if (WARN_ON(!bo->ttm.resource)) - return -EINVAL; + if (WARN_ON(!bo->ttm.resource)) { + ret = -EINVAL; + goto out_unlock_bo; + } - if (WARN_ON(!xe_bo_is_pinned(bo))) - return -EINVAL; + if (WARN_ON(!xe_bo_is_pinned(bo))) { + ret = -EINVAL; + goto out_unlock_bo; + } - if (WARN_ON(!xe_bo_is_vram(bo))) - return -EINVAL; + if (!xe_bo_is_vram(bo)) + goto out_unlock_bo; - ret = ttm_bo_mem_space(&bo->ttm, &placement, &new_mem, &ctx); - if (ret) - return ret; + if (bo->flags & XE_BO_FLAG_PINNED_NORESTORE) + goto out_unlock_bo; - if (!bo->ttm.ttm) { - bo->ttm.ttm = xe_ttm_tt_create(&bo->ttm, 0); - if (!bo->ttm.ttm) { - ret = -ENOMEM; - goto err_res_free; + if (!backup) { + backup = ___xe_bo_create_locked(xe, NULL, NULL, bo->ttm.base.resv, NULL, bo->size, + DRM_XE_GEM_CPU_CACHING_WB, ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | XE_BO_FLAG_NEEDS_CPU_ACCESS | + XE_BO_FLAG_PINNED); + if (IS_ERR(backup)) { + ret = PTR_ERR(backup); + goto out_unlock_bo; } + backup->parent_obj = xe_bo_get(bo); /* Released by bo_destroy */ + backup_created = true; } - ret = ttm_tt_populate(bo->ttm.bdev, bo->ttm.ttm, &ctx); - if (ret) - goto err_res_free; + if (xe_bo_is_user(bo) || (bo->flags & XE_BO_FLAG_PINNED_LATE_RESTORE)) { + struct xe_migrate *migrate; + struct dma_fence *fence; - ret = dma_resv_reserve_fences(bo->ttm.base.resv, 1); - if (ret) - goto err_res_free; + if (bo->tile) + migrate = bo->tile->migrate; + else + migrate = mem_type_to_migrate(xe, bo->ttm.resource->mem_type); - ret = xe_bo_move(&bo->ttm, false, &ctx, new_mem, NULL); - if (ret) - goto err_res_free; + ret = dma_resv_reserve_fences(bo->ttm.base.resv, 1); + if (ret) + goto out_backup; - dma_resv_wait_timeout(bo->ttm.base.resv, DMA_RESV_USAGE_KERNEL, - false, MAX_SCHEDULE_TIMEOUT); + ret = dma_resv_reserve_fences(backup->ttm.base.resv, 1); + if (ret) + goto out_backup; - return 0; + fence = xe_migrate_copy(migrate, bo, backup, bo->ttm.resource, + backup->ttm.resource, false); + if (IS_ERR(fence)) { + ret = PTR_ERR(fence); + goto out_backup; + } + + dma_resv_add_fence(bo->ttm.base.resv, fence, + DMA_RESV_USAGE_KERNEL); + dma_resv_add_fence(backup->ttm.base.resv, fence, + DMA_RESV_USAGE_KERNEL); + dma_fence_put(fence); + } else { + ret = xe_bo_vmap(backup); + if (ret) + goto out_backup; -err_res_free: - ttm_resource_free(&bo->ttm, &new_mem); + if (iosys_map_is_null(&bo->vmap)) { + ret = xe_bo_vmap(bo); + if (ret) + goto out_backup; + unmap = true; + } + + xe_map_memcpy_from(xe, backup->vmap.vaddr, &bo->vmap, 0, + bo->size); + } + + if (!bo->backup_obj) + bo->backup_obj = backup; + +out_backup: + xe_bo_vunmap(backup); + if (ret && backup_created) + xe_bo_put(backup); +out_unlock_bo: + if (unmap) + xe_bo_vunmap(bo); + xe_bo_unlock(bo); return ret; } @@ -888,7 +1276,6 @@ err_res_free: * @bo: The buffer object to move. * * On successful completion, the object memory will be moved back to VRAM. - * This function blocks until the object has been fully moved. * * This is needed to for special handling of pinned VRAM object during * suspend-resume. @@ -899,53 +1286,117 @@ int xe_bo_restore_pinned(struct xe_bo *bo) { struct ttm_operation_ctx ctx = { .interruptible = false, + .gfp_retry_mayfail = false, }; - struct ttm_resource *new_mem; + struct xe_device *xe = ttm_to_xe_device(bo->ttm.bdev); + struct xe_bo *backup = bo->backup_obj; + bool unmap = false; int ret; - xe_bo_assert_held(bo); + if (!backup) + return 0; - if (WARN_ON(!bo->ttm.resource)) - return -EINVAL; + xe_bo_lock(bo, false); - if (WARN_ON(!xe_bo_is_pinned(bo))) - return -EINVAL; + if (!xe_bo_is_pinned(backup)) { + ret = ttm_bo_validate(&backup->ttm, &backup->placement, &ctx); + if (ret) + goto out_unlock_bo; + } - if (WARN_ON(xe_bo_is_vram(bo) || !bo->ttm.ttm)) - return -EINVAL; + if (xe_bo_is_user(bo) || (bo->flags & XE_BO_FLAG_PINNED_LATE_RESTORE)) { + struct xe_migrate *migrate; + struct dma_fence *fence; - ret = ttm_bo_mem_space(&bo->ttm, &bo->placement, &new_mem, &ctx); - if (ret) - return ret; + if (bo->tile) + migrate = bo->tile->migrate; + else + migrate = mem_type_to_migrate(xe, bo->ttm.resource->mem_type); - ret = ttm_tt_populate(bo->ttm.bdev, bo->ttm.ttm, &ctx); - if (ret) - goto err_res_free; + ret = dma_resv_reserve_fences(bo->ttm.base.resv, 1); + if (ret) + goto out_unlock_bo; - ret = dma_resv_reserve_fences(bo->ttm.base.resv, 1); - if (ret) - goto err_res_free; + ret = dma_resv_reserve_fences(backup->ttm.base.resv, 1); + if (ret) + goto out_unlock_bo; - ret = xe_bo_move(&bo->ttm, false, &ctx, new_mem, NULL); - if (ret) - goto err_res_free; + fence = xe_migrate_copy(migrate, backup, bo, + backup->ttm.resource, bo->ttm.resource, + false); + if (IS_ERR(fence)) { + ret = PTR_ERR(fence); + goto out_unlock_bo; + } - dma_resv_wait_timeout(bo->ttm.base.resv, DMA_RESV_USAGE_KERNEL, - false, MAX_SCHEDULE_TIMEOUT); + dma_resv_add_fence(bo->ttm.base.resv, fence, + DMA_RESV_USAGE_KERNEL); + dma_resv_add_fence(backup->ttm.base.resv, fence, + DMA_RESV_USAGE_KERNEL); + dma_fence_put(fence); + } else { + ret = xe_bo_vmap(backup); + if (ret) + goto out_unlock_bo; - return 0; + if (iosys_map_is_null(&bo->vmap)) { + ret = xe_bo_vmap(bo); + if (ret) + goto out_backup; + unmap = true; + } + + xe_map_memcpy_to(xe, &bo->vmap, 0, backup->vmap.vaddr, + bo->size); + } -err_res_free: - ttm_resource_free(&bo->ttm, &new_mem); + bo->backup_obj = NULL; + +out_backup: + xe_bo_vunmap(backup); + if (!bo->backup_obj) { + if (xe_bo_is_pinned(backup)) + ttm_bo_unpin(&backup->ttm); + xe_bo_put(backup); + } +out_unlock_bo: + if (unmap) + xe_bo_vunmap(bo); + xe_bo_unlock(bo); return ret; } +int xe_bo_dma_unmap_pinned(struct xe_bo *bo) +{ + struct ttm_buffer_object *ttm_bo = &bo->ttm; + struct ttm_tt *tt = ttm_bo->ttm; + + if (tt) { + struct xe_ttm_tt *xe_tt = container_of(tt, typeof(*xe_tt), ttm); + + if (ttm_bo->type == ttm_bo_type_sg && ttm_bo->sg) { + dma_buf_unmap_attachment(ttm_bo->base.import_attach, + ttm_bo->sg, + DMA_BIDIRECTIONAL); + ttm_bo->sg = NULL; + xe_tt->sg = NULL; + } else if (xe_tt->sg) { + dma_unmap_sgtable(xe_tt->xe->drm.dev, xe_tt->sg, + DMA_BIDIRECTIONAL, 0); + sg_free_table(xe_tt->sg); + xe_tt->sg = NULL; + } + } + + return 0; +} + static unsigned long xe_ttm_io_mem_pfn(struct ttm_buffer_object *ttm_bo, unsigned long page_offset) { struct xe_bo *bo = ttm_to_xe_bo(ttm_bo); struct xe_res_cursor cursor; - struct xe_mem_region *vram; + struct xe_vram_region *vram; if (ttm_bo->resource->mem_type == XE_PL_STOLEN) return xe_ttm_stolen_io_offset(bo, page_offset << PAGE_SHIFT) >> PAGE_SHIFT; @@ -1049,6 +1500,87 @@ static void xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo) } } +static void xe_ttm_bo_purge(struct ttm_buffer_object *ttm_bo, struct ttm_operation_ctx *ctx) +{ + struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev); + + if (ttm_bo->ttm) { + struct ttm_placement place = {}; + int ret = ttm_bo_validate(ttm_bo, &place, ctx); + + drm_WARN_ON(&xe->drm, ret); + } +} + +static void xe_ttm_bo_swap_notify(struct ttm_buffer_object *ttm_bo) +{ + struct ttm_operation_ctx ctx = { + .interruptible = false, + .gfp_retry_mayfail = false, + }; + + if (ttm_bo->ttm) { + struct xe_ttm_tt *xe_tt = + container_of(ttm_bo->ttm, struct xe_ttm_tt, ttm); + + if (xe_tt->purgeable) + xe_ttm_bo_purge(ttm_bo, &ctx); + } +} + +static int xe_ttm_access_memory(struct ttm_buffer_object *ttm_bo, + unsigned long offset, void *buf, int len, + int write) +{ + struct xe_bo *bo = ttm_to_xe_bo(ttm_bo); + struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev); + struct iosys_map vmap; + struct xe_res_cursor cursor; + struct xe_vram_region *vram; + int bytes_left = len; + int err = 0; + + xe_bo_assert_held(bo); + xe_device_assert_mem_access(xe); + + if (!mem_type_is_vram(ttm_bo->resource->mem_type)) + return -EIO; + + if (!xe_ttm_resource_visible(ttm_bo->resource) || len >= SZ_16K) { + struct xe_migrate *migrate = + mem_type_to_migrate(xe, ttm_bo->resource->mem_type); + + err = xe_migrate_access_memory(migrate, bo, offset, buf, len, + write); + goto out; + } + + vram = res_to_mem_region(ttm_bo->resource); + xe_res_first(ttm_bo->resource, offset & PAGE_MASK, + bo->size - (offset & PAGE_MASK), &cursor); + + do { + unsigned long page_offset = (offset & ~PAGE_MASK); + int byte_count = min((int)(PAGE_SIZE - page_offset), bytes_left); + + iosys_map_set_vaddr_iomem(&vmap, (u8 __iomem *)vram->mapping + + cursor.start); + if (write) + xe_map_memcpy_to(xe, &vmap, page_offset, buf, byte_count); + else + xe_map_memcpy_from(xe, buf, &vmap, page_offset, byte_count); + + buf += byte_count; + offset += byte_count; + bytes_left -= byte_count; + if (bytes_left) + xe_res_next(&cursor, PAGE_SIZE); + } while (bytes_left); + +out: + return err ?: len; +} + const struct ttm_device_funcs xe_ttm_funcs = { .ttm_tt_create = xe_ttm_tt_create, .ttm_tt_populate = xe_ttm_tt_populate, @@ -1058,15 +1590,19 @@ const struct ttm_device_funcs xe_ttm_funcs = { .move = xe_bo_move, .io_mem_reserve = xe_ttm_io_mem_reserve, .io_mem_pfn = xe_ttm_io_mem_pfn, + .access_memory = xe_ttm_access_memory, .release_notify = xe_ttm_bo_release_notify, - .eviction_valuable = ttm_bo_eviction_valuable, + .eviction_valuable = xe_bo_eviction_valuable, .delete_mem_notify = xe_ttm_bo_delete_mem_notify, + .swap_notify = xe_ttm_bo_swap_notify, }; static void xe_ttm_bo_destroy(struct ttm_buffer_object *ttm_bo) { struct xe_bo *bo = ttm_to_xe_bo(ttm_bo); struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev); + struct xe_tile *tile; + u8 id; if (bo->ttm.base.import_attach) drm_prime_gem_destroy(&bo->ttm.base, NULL); @@ -1074,8 +1610,9 @@ static void xe_ttm_bo_destroy(struct ttm_buffer_object *ttm_bo) xe_assert(xe, list_empty(&ttm_bo->base.gpuva.list)); - if (bo->ggtt_node.size) - xe_ggtt_remove_bo(bo->tile->mem.ggtt, bo); + for_each_tile(tile, xe, id) + if (bo->ggtt_node[id] && bo->ggtt_node[id]->base.size) + xe_ggtt_remove_bo(tile->mem.ggtt, bo); #ifdef CONFIG_PROC_FS if (bo->client) @@ -1085,6 +1622,9 @@ static void xe_ttm_bo_destroy(struct ttm_buffer_object *ttm_bo) if (bo->vm && xe_bo_is_user(bo)) xe_vm_put(bo->vm); + if (bo->parent_obj) + xe_bo_put(bo->parent_obj); + mutex_lock(&xe->mem_access.vram_userfault.lock); if (!list_empty(&bo->vram_userfault_link)) list_del(&bo->vram_userfault_link); @@ -1126,25 +1666,18 @@ static void xe_gem_object_close(struct drm_gem_object *obj, } } -static bool should_migrate_to_system(struct xe_bo *bo) -{ - struct xe_device *xe = xe_bo_device(bo); - - return xe_device_in_fault_mode(xe) && bo->props.cpu_atomic; -} - static vm_fault_t xe_gem_fault(struct vm_fault *vmf) { struct ttm_buffer_object *tbo = vmf->vma->vm_private_data; struct drm_device *ddev = tbo->base.dev; struct xe_device *xe = to_xe_device(ddev); struct xe_bo *bo = ttm_to_xe_bo(tbo); - bool needs_rpm = bo->flags & XE_BO_CREATE_VRAM_MASK; + bool needs_rpm = bo->flags & XE_BO_FLAG_VRAM_MASK; vm_fault_t ret; - int idx, r = 0; + int idx; if (needs_rpm) - xe_device_mem_access_get(xe); + xe_pm_runtime_get(xe); ret = ttm_bo_vm_reserve(tbo, vmf); if (ret) @@ -1153,17 +1686,8 @@ static vm_fault_t xe_gem_fault(struct vm_fault *vmf) if (drm_dev_enter(ddev, &idx)) { trace_xe_bo_cpu_fault(bo); - if (should_migrate_to_system(bo)) { - r = xe_bo_migrate(bo, XE_PL_TT); - if (r == -EBUSY || r == -ERESTARTSYS || r == -EINTR) - ret = VM_FAULT_NOPAGE; - else if (r) - ret = VM_FAULT_SIGBUS; - } - if (!ret) - ret = ttm_bo_vm_fault_reserved(vmf, - vmf->vma->vm_page_prot, - TTM_BO_VM_NUM_PREFAULT); + ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot, + TTM_BO_VM_NUM_PREFAULT); drm_dev_exit(idx); } else { ret = ttm_bo_vm_dummy_page(vmf, vmf->vma->vm_page_prot); @@ -1184,7 +1708,46 @@ static vm_fault_t xe_gem_fault(struct vm_fault *vmf) dma_resv_unlock(tbo->base.resv); out: if (needs_rpm) - xe_device_mem_access_put(xe); + xe_pm_runtime_put(xe); + + return ret; +} + +static int xe_bo_vm_access(struct vm_area_struct *vma, unsigned long addr, + void *buf, int len, int write) +{ + struct ttm_buffer_object *ttm_bo = vma->vm_private_data; + struct xe_bo *bo = ttm_to_xe_bo(ttm_bo); + struct xe_device *xe = xe_bo_device(bo); + int ret; + + xe_pm_runtime_get(xe); + ret = ttm_bo_vm_access(vma, addr, buf, len, write); + xe_pm_runtime_put(xe); + + return ret; +} + +/** + * xe_bo_read() - Read from an xe_bo + * @bo: The buffer object to read from. + * @offset: The byte offset to start reading from. + * @dst: Location to store the read. + * @size: Size in bytes for the read. + * + * Read @size bytes from the @bo, starting from @offset, storing into @dst. + * + * Return: Zero on success, or negative error. + */ +int xe_bo_read(struct xe_bo *bo, u64 offset, void *dst, int size) +{ + int ret; + + ret = ttm_bo_access(&bo->ttm, offset, dst, size, 0); + if (ret >= 0 && ret != size) + ret = -EIO; + else if (ret == size) + ret = 0; return ret; } @@ -1193,7 +1756,7 @@ static const struct vm_operations_struct xe_gem_vm_ops = { .fault = xe_gem_fault, .open = ttm_bo_vm_open, .close = ttm_bo_vm_close, - .access = ttm_bo_vm_access + .access = xe_bo_vm_access, }; static const struct drm_gem_object_funcs xe_gem_object_funcs = { @@ -1207,7 +1770,7 @@ static const struct drm_gem_object_funcs xe_gem_object_funcs = { /** * xe_bo_alloc - Allocate storage for a struct xe_bo * - * This funcition is intended to allocate storage to be used for input + * This function is intended to allocate storage to be used for input * to __xe_bo_create_locked(), in the case a pointer to the bo to be * created is needed before the call to __xe_bo_create_locked(). * If __xe_bo_create_locked ends up never to be called, then the @@ -1247,6 +1810,7 @@ struct xe_bo *___xe_bo_create_locked(struct xe_device *xe, struct xe_bo *bo, struct ttm_operation_ctx ctx = { .interruptible = true, .no_wait_gpu = false, + .gfp_retry_mayfail = true, }; struct ttm_placement *placement; uint32_t alignment; @@ -1261,18 +1825,24 @@ struct xe_bo *___xe_bo_create_locked(struct xe_device *xe, struct xe_bo *bo, return ERR_PTR(-EINVAL); } - if (flags & (XE_BO_CREATE_VRAM_MASK | XE_BO_CREATE_STOLEN_BIT) && - !(flags & XE_BO_CREATE_IGNORE_MIN_PAGE_SIZE_BIT) && - xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) { - aligned_size = ALIGN(size, SZ_64K); - if (type != ttm_bo_type_device) - size = ALIGN(size, SZ_64K); - flags |= XE_BO_INTERNAL_64K; - alignment = SZ_64K >> PAGE_SHIFT; + /* XE_BO_FLAG_GGTTx requires XE_BO_FLAG_GGTT also be set */ + if ((flags & XE_BO_FLAG_GGTT_ALL) && !(flags & XE_BO_FLAG_GGTT)) + return ERR_PTR(-EINVAL); + + if (flags & (XE_BO_FLAG_VRAM_MASK | XE_BO_FLAG_STOLEN) && + !(flags & XE_BO_FLAG_IGNORE_MIN_PAGE_SIZE) && + ((xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) || + (flags & (XE_BO_FLAG_NEEDS_64K | XE_BO_FLAG_NEEDS_2M)))) { + size_t align = flags & XE_BO_FLAG_NEEDS_2M ? SZ_2M : SZ_64K; + aligned_size = ALIGN(size, align); + if (type != ttm_bo_type_device) + size = ALIGN(size, align); + flags |= XE_BO_FLAG_INTERNAL_64K; + alignment = align >> PAGE_SHIFT; } else { aligned_size = ALIGN(size, SZ_4K); - flags &= ~XE_BO_INTERNAL_64K; + flags &= ~XE_BO_FLAG_INTERNAL_64K; alignment = SZ_4K >> PAGE_SHIFT; } @@ -1291,9 +1861,6 @@ struct xe_bo *___xe_bo_create_locked(struct xe_device *xe, struct xe_bo *bo, bo->flags = flags; bo->cpu_caching = cpu_caching; bo->ttm.base.funcs = &xe_gem_object_funcs; - bo->props.preferred_mem_class = XE_BO_PROPS_INVALID; - bo->props.preferred_gt = XE_BO_PROPS_INVALID; - bo->props.preferred_mem_type = XE_BO_PROPS_INVALID; bo->ttm.priority = XE_BO_PRIORITY_NORMAL; INIT_LIST_HEAD(&bo->pinned_link); #ifdef CONFIG_PROC_FS @@ -1304,11 +1871,11 @@ struct xe_bo *___xe_bo_create_locked(struct xe_device *xe, struct xe_bo *bo, drm_gem_private_object_init(&xe->drm, &bo->ttm.base, size); if (resv) { - ctx.allow_res_evict = !(flags & XE_BO_CREATE_NO_RESV_EVICT); + ctx.allow_res_evict = !(flags & XE_BO_FLAG_NO_RESV_EVICT); ctx.resv = resv; } - if (!(flags & XE_BO_FIXED_PLACEMENT_BIT)) { + if (!(flags & XE_BO_FLAG_FIXED_PLACEMENT)) { err = __xe_bo_placement_for_flags(xe, bo, bo->flags); if (WARN_ON(err)) { xe_ttm_bo_destroy(&bo->ttm); @@ -1318,7 +1885,7 @@ struct xe_bo *___xe_bo_create_locked(struct xe_device *xe, struct xe_bo *bo, /* Defer populating type_sg bos */ placement = (type == ttm_bo_type_sg || - bo->flags & XE_BO_DEFER_BACKING) ? &sys_placement : + bo->flags & XE_BO_FLAG_DEFER_BACKING) ? &sys_placement : &bo->placement; err = ttm_bo_init_reserved(&xe->ttm, &bo->ttm, type, placement, alignment, @@ -1373,21 +1940,21 @@ static int __xe_bo_fixed_placement(struct xe_device *xe, { struct ttm_place *place = bo->placements; - if (flags & (XE_BO_CREATE_USER_BIT|XE_BO_CREATE_SYSTEM_BIT)) + if (flags & (XE_BO_FLAG_USER | XE_BO_FLAG_SYSTEM)) return -EINVAL; place->flags = TTM_PL_FLAG_CONTIGUOUS; place->fpfn = start >> PAGE_SHIFT; place->lpfn = end >> PAGE_SHIFT; - switch (flags & (XE_BO_CREATE_STOLEN_BIT | XE_BO_CREATE_VRAM_MASK)) { - case XE_BO_CREATE_VRAM0_BIT: + switch (flags & (XE_BO_FLAG_STOLEN | XE_BO_FLAG_VRAM_MASK)) { + case XE_BO_FLAG_VRAM0: place->mem_type = XE_PL_VRAM0; break; - case XE_BO_CREATE_VRAM1_BIT: + case XE_BO_FLAG_VRAM1: place->mem_type = XE_PL_VRAM1; break; - case XE_BO_CREATE_STOLEN_BIT: + case XE_BO_FLAG_STOLEN: place->mem_type = XE_PL_STOLEN; break; @@ -1408,7 +1975,8 @@ static struct xe_bo * __xe_bo_create_locked(struct xe_device *xe, struct xe_tile *tile, struct xe_vm *vm, size_t size, u64 start, u64 end, - u16 cpu_caching, enum ttm_bo_type type, u32 flags) + u16 cpu_caching, enum ttm_bo_type type, u32 flags, + u64 alignment) { struct xe_bo *bo = NULL; int err; @@ -1421,7 +1989,7 @@ __xe_bo_create_locked(struct xe_device *xe, if (IS_ERR(bo)) return bo; - flags |= XE_BO_FIXED_PLACEMENT_BIT; + flags |= XE_BO_FLAG_FIXED_PLACEMENT; err = __xe_bo_fixed_placement(xe, bo, flags, start, end, size); if (err) { xe_bo_free(bo); @@ -1431,12 +1999,14 @@ __xe_bo_create_locked(struct xe_device *xe, bo = ___xe_bo_create_locked(xe, bo, tile, vm ? xe_vm_resv(vm) : NULL, vm && !xe_vm_in_fault_mode(vm) && - flags & XE_BO_CREATE_USER_BIT ? + flags & XE_BO_FLAG_USER ? &vm->lru_bulk_move : NULL, size, cpu_caching, type, flags); if (IS_ERR(bo)) return bo; + bo->min_align = alignment; + /* * Note that instead of taking a reference no the drm_gpuvm_resv_bo(), * to ensure the shared resv doesn't disappear under the bo, the bo @@ -1448,22 +2018,33 @@ __xe_bo_create_locked(struct xe_device *xe, xe_vm_get(vm); bo->vm = vm; - if (bo->flags & XE_BO_CREATE_GGTT_BIT) { - if (!tile && flags & XE_BO_CREATE_STOLEN_BIT) - tile = xe_device_get_root_tile(xe); + if (bo->flags & XE_BO_FLAG_GGTT) { + struct xe_tile *t; + u8 id; - xe_assert(xe, tile); + if (!(bo->flags & XE_BO_FLAG_GGTT_ALL)) { + if (!tile && flags & XE_BO_FLAG_STOLEN) + tile = xe_device_get_root_tile(xe); - if (flags & XE_BO_FIXED_PLACEMENT_BIT) { - err = xe_ggtt_insert_bo_at(tile->mem.ggtt, bo, - start + bo->size, U64_MAX); - } else { - err = xe_ggtt_insert_bo(tile->mem.ggtt, bo); + xe_assert(xe, tile); + } + + for_each_tile(t, xe, id) { + if (t != tile && !(bo->flags & XE_BO_FLAG_GGTTx(t))) + continue; + + if (flags & XE_BO_FLAG_FIXED_PLACEMENT) { + err = xe_ggtt_insert_bo_at(t->mem.ggtt, bo, + start + bo->size, U64_MAX); + } else { + err = xe_ggtt_insert_bo(t->mem.ggtt, bo); + } + if (err) + goto err_unlock_put_bo; } - if (err) - goto err_unlock_put_bo; } + trace_xe_bo_create(bo); return bo; err_unlock_put_bo: @@ -1477,27 +2058,28 @@ struct xe_bo * xe_bo_create_locked_range(struct xe_device *xe, struct xe_tile *tile, struct xe_vm *vm, size_t size, u64 start, u64 end, - enum ttm_bo_type type, u32 flags) + enum ttm_bo_type type, u32 flags, u64 alignment) { - return __xe_bo_create_locked(xe, tile, vm, size, start, end, 0, type, flags); + return __xe_bo_create_locked(xe, tile, vm, size, start, end, 0, type, + flags, alignment); } struct xe_bo *xe_bo_create_locked(struct xe_device *xe, struct xe_tile *tile, struct xe_vm *vm, size_t size, enum ttm_bo_type type, u32 flags) { - return __xe_bo_create_locked(xe, tile, vm, size, 0, ~0ULL, 0, type, flags); + return __xe_bo_create_locked(xe, tile, vm, size, 0, ~0ULL, 0, type, + flags, 0); } struct xe_bo *xe_bo_create_user(struct xe_device *xe, struct xe_tile *tile, struct xe_vm *vm, size_t size, u16 cpu_caching, - enum ttm_bo_type type, u32 flags) { struct xe_bo *bo = __xe_bo_create_locked(xe, tile, vm, size, 0, ~0ULL, - cpu_caching, type, - flags | XE_BO_CREATE_USER_BIT); + cpu_caching, ttm_bo_type_device, + flags | XE_BO_FLAG_USER, 0); if (!IS_ERR(bo)) xe_bo_unlock_vm_held(bo); @@ -1521,17 +2103,29 @@ struct xe_bo *xe_bo_create_pin_map_at(struct xe_device *xe, struct xe_tile *tile size_t size, u64 offset, enum ttm_bo_type type, u32 flags) { + return xe_bo_create_pin_map_at_aligned(xe, tile, vm, size, offset, + type, flags, 0); +} + +struct xe_bo *xe_bo_create_pin_map_at_aligned(struct xe_device *xe, + struct xe_tile *tile, + struct xe_vm *vm, + size_t size, u64 offset, + enum ttm_bo_type type, u32 flags, + u64 alignment) +{ struct xe_bo *bo; int err; u64 start = offset == ~0ull ? 0 : offset; u64 end = offset == ~0ull ? offset : start + size; - if (flags & XE_BO_CREATE_STOLEN_BIT && + if (flags & XE_BO_FLAG_STOLEN && xe_ttm_stolen_cpu_access_needs_ggtt(xe)) - flags |= XE_BO_CREATE_GGTT_BIT; + flags |= XE_BO_FLAG_GGTT; bo = xe_bo_create_locked_range(xe, tile, vm, size, start, end, type, - flags | XE_BO_NEEDS_CPU_ACCESS); + flags | XE_BO_FLAG_NEEDS_CPU_ACCESS | XE_BO_FLAG_PINNED, + alignment); if (IS_ERR(bo)) return bo; @@ -1577,7 +2171,7 @@ struct xe_bo *xe_bo_create_from_data(struct xe_device *xe, struct xe_tile *tile, return bo; } -static void __xe_bo_unpin_map_no_vm(struct drm_device *drm, void *arg) +static void __xe_bo_unpin_map_no_vm(void *arg) { xe_bo_unpin_map_no_vm(arg); } @@ -1588,11 +2182,13 @@ struct xe_bo *xe_managed_bo_create_pin_map(struct xe_device *xe, struct xe_tile struct xe_bo *bo; int ret; + KUNIT_STATIC_STUB_REDIRECT(xe_managed_bo_create_pin_map, xe, tile, size, flags); + bo = xe_bo_create_pin_map(xe, tile, NULL, size, ttm_bo_type_kernel, flags); if (IS_ERR(bo)) return bo; - ret = drmm_add_action_or_reset(&xe->drm, __xe_bo_unpin_map_no_vm, bo); + ret = devm_add_action_or_reset(xe->drm.dev, __xe_bo_unpin_map_no_vm, bo); if (ret) return ERR_PTR(ret); @@ -1628,17 +2224,20 @@ struct xe_bo *xe_managed_bo_create_from_data(struct xe_device *xe, struct xe_til int xe_managed_bo_reinit_in_vram(struct xe_device *xe, struct xe_tile *tile, struct xe_bo **src) { struct xe_bo *bo; + u32 dst_flags = XE_BO_FLAG_VRAM_IF_DGFX(tile) | XE_BO_FLAG_GGTT; + + dst_flags |= (*src)->flags & (XE_BO_FLAG_GGTT_INVALIDATE | + XE_BO_FLAG_PINNED_NORESTORE); xe_assert(xe, IS_DGFX(xe)); xe_assert(xe, !(*src)->vmap.is_iomem); - bo = xe_managed_bo_create_from_data(xe, tile, (*src)->vmap.vaddr, (*src)->size, - XE_BO_CREATE_VRAM_IF_DGFX(tile) | - XE_BO_CREATE_GGTT_BIT); + bo = xe_managed_bo_create_from_data(xe, tile, (*src)->vmap.vaddr, + (*src)->size, dst_flags); if (IS_ERR(bo)) return PTR_ERR(bo); - drmm_release_action(&xe->drm, __xe_bo_unpin_map_no_vm, *src); + devm_release_action(xe->drm.dev, __xe_bo_unpin_map_no_vm, *src); *src = bo; return 0; @@ -1652,10 +2251,16 @@ uint64_t vram_region_gpu_offset(struct ttm_resource *res) { struct xe_device *xe = ttm_to_xe_device(res->bo->bdev); - if (res->mem_type == XE_PL_STOLEN) + switch (res->mem_type) { + case XE_PL_STOLEN: return xe_ttm_stolen_gpu_offset(xe); - - return res_to_mem_region(res)->dpa_base; + case XE_PL_TT: + case XE_PL_SYSTEM: + return 0; + default: + return res_to_mem_region(res)->dpa_base; + } + return 0; } /** @@ -1681,15 +2286,14 @@ int xe_bo_pin_external(struct xe_bo *bo) if (err) return err; - if (xe_bo_is_vram(bo)) { - spin_lock(&xe->pinned.lock); - list_add_tail(&bo->pinned_link, - &xe->pinned.external_vram); - spin_unlock(&xe->pinned.lock); - } + spin_lock(&xe->pinned.lock); + list_add_tail(&bo->pinned_link, &xe->pinned.late.external); + spin_unlock(&xe->pinned.lock); } ttm_bo_pin(&bo->ttm); + if (bo->ttm.ttm && ttm_tt_is_populated(bo->ttm.ttm)) + xe_ttm_tt_account_subtract(bo->ttm.ttm); /* * FIXME: If we always use the reserve / unreserve functions for locking @@ -1702,6 +2306,7 @@ int xe_bo_pin_external(struct xe_bo *bo) int xe_bo_pin(struct xe_bo *bo) { + struct ttm_place *place = &bo->placements[0]; struct xe_device *xe = xe_bo_device(bo); int err; @@ -1709,8 +2314,8 @@ int xe_bo_pin(struct xe_bo *bo) xe_assert(xe, !xe_bo_is_user(bo)); /* Pinned object must be in GGTT or have pinned flag */ - xe_assert(xe, bo->flags & (XE_BO_CREATE_PINNED_BIT | - XE_BO_CREATE_GGTT_BIT)); + xe_assert(xe, bo->flags & (XE_BO_FLAG_PINNED | + XE_BO_FLAG_GGTT)); /* * No reason we can't support pinning imported dma-bufs we just don't @@ -1725,29 +2330,18 @@ int xe_bo_pin(struct xe_bo *bo) if (err) return err; - /* - * For pinned objects in on DGFX, which are also in vram, we expect - * these to be in contiguous VRAM memory. Required eviction / restore - * during suspend / resume (force restore to same physical address). - */ - if (IS_DGFX(xe) && !(IS_ENABLED(CONFIG_DRM_XE_DEBUG) && - bo->flags & XE_BO_INTERNAL_TEST)) { - struct ttm_place *place = &(bo->placements[0]); - - if (mem_type_is_vram(place->mem_type)) { - xe_assert(xe, place->flags & TTM_PL_FLAG_CONTIGUOUS); - - place->fpfn = (xe_bo_addr(bo, 0, PAGE_SIZE) - - vram_region_gpu_offset(bo->ttm.resource)) >> PAGE_SHIFT; - place->lpfn = place->fpfn + (bo->size >> PAGE_SHIFT); - - spin_lock(&xe->pinned.lock); - list_add_tail(&bo->pinned_link, &xe->pinned.kernel_bo_present); - spin_unlock(&xe->pinned.lock); - } + if (mem_type_is_vram(place->mem_type) || bo->flags & XE_BO_FLAG_GGTT) { + spin_lock(&xe->pinned.lock); + if (bo->flags & XE_BO_FLAG_PINNED_LATE_RESTORE) + list_add_tail(&bo->pinned_link, &xe->pinned.late.kernel_bo_present); + else + list_add_tail(&bo->pinned_link, &xe->pinned.early.kernel_bo_present); + spin_unlock(&xe->pinned.lock); } ttm_bo_pin(&bo->ttm); + if (bo->ttm.ttm && ttm_tt_is_populated(bo->ttm.ttm)) + xe_ttm_tt_account_subtract(bo->ttm.ttm); /* * FIXME: If we always use the reserve / unreserve functions for locking @@ -1776,13 +2370,14 @@ void xe_bo_unpin_external(struct xe_bo *bo) xe_assert(xe, xe_bo_is_pinned(bo)); xe_assert(xe, xe_bo_is_user(bo)); - if (bo->ttm.pin_count == 1 && !list_empty(&bo->pinned_link)) { - spin_lock(&xe->pinned.lock); + spin_lock(&xe->pinned.lock); + if (bo->ttm.pin_count == 1 && !list_empty(&bo->pinned_link)) list_del_init(&bo->pinned_link); - spin_unlock(&xe->pinned.lock); - } + spin_unlock(&xe->pinned.lock); ttm_bo_unpin(&bo->ttm); + if (bo->ttm.ttm && ttm_tt_is_populated(bo->ttm.ttm)) + xe_ttm_tt_account_add(bo->ttm.ttm); /* * FIXME: If we always use the reserve / unreserve functions for locking @@ -1793,25 +2388,28 @@ void xe_bo_unpin_external(struct xe_bo *bo) void xe_bo_unpin(struct xe_bo *bo) { + struct ttm_place *place = &bo->placements[0]; struct xe_device *xe = xe_bo_device(bo); xe_assert(xe, !bo->ttm.base.import_attach); xe_assert(xe, xe_bo_is_pinned(bo)); - if (IS_DGFX(xe) && !(IS_ENABLED(CONFIG_DRM_XE_DEBUG) && - bo->flags & XE_BO_INTERNAL_TEST)) { - struct ttm_place *place = &(bo->placements[0]); - - if (mem_type_is_vram(place->mem_type)) { - xe_assert(xe, !list_empty(&bo->pinned_link)); + if (mem_type_is_vram(place->mem_type) || bo->flags & XE_BO_FLAG_GGTT) { + spin_lock(&xe->pinned.lock); + xe_assert(xe, !list_empty(&bo->pinned_link)); + list_del_init(&bo->pinned_link); + spin_unlock(&xe->pinned.lock); - spin_lock(&xe->pinned.lock); - list_del_init(&bo->pinned_link); - spin_unlock(&xe->pinned.lock); + if (bo->backup_obj) { + if (xe_bo_is_pinned(bo->backup_obj)) + ttm_bo_unpin(&bo->backup_obj->ttm); + xe_bo_put(bo->backup_obj); + bo->backup_obj = NULL; } } - ttm_bo_unpin(&bo->ttm); + if (bo->ttm.ttm && ttm_tt_is_populated(bo->ttm.ttm)) + xe_ttm_tt_account_add(bo->ttm.ttm); } /** @@ -1835,7 +2433,10 @@ int xe_bo_validate(struct xe_bo *bo, struct xe_vm *vm, bool allow_res_evict) struct ttm_operation_ctx ctx = { .interruptible = true, .no_wait_gpu = false, + .gfp_retry_mayfail = true, }; + struct pin_cookie cookie; + int ret; if (vm) { lockdep_assert_held(&vm->lock); @@ -1845,7 +2446,12 @@ int xe_bo_validate(struct xe_bo *bo, struct xe_vm *vm, bool allow_res_evict) ctx.resv = xe_vm_resv(vm); } - return ttm_bo_validate(&bo->ttm, &bo->placement, &ctx); + cookie = xe_vm_set_validating(vm, allow_res_evict); + trace_xe_bo_validate(bo); + ret = ttm_bo_validate(&bo->ttm, &bo->placement, &ctx); + xe_vm_clear_validating(vm, allow_res_evict, cookie); + + return ret; } bool xe_bo_is_xe_bo(struct ttm_buffer_object *bo) @@ -1896,13 +2502,15 @@ dma_addr_t xe_bo_addr(struct xe_bo *bo, u64 offset, size_t page_size) int xe_bo_vmap(struct xe_bo *bo) { + struct xe_device *xe = ttm_to_xe_device(bo->ttm.bdev); void *virtual; bool is_iomem; int ret; xe_bo_assert_held(bo); - if (!(bo->flags & XE_BO_NEEDS_CPU_ACCESS)) + if (drm_WARN_ON(&xe->drm, !(bo->flags & XE_BO_FLAG_NEEDS_CPU_ACCESS) || + !force_contiguous(bo->flags))) return -EINVAL; if (!iosys_map_is_null(&bo->vmap)) @@ -1942,6 +2550,93 @@ void xe_bo_vunmap(struct xe_bo *bo) __xe_bo_vunmap(bo); } +static int gem_create_set_pxp_type(struct xe_device *xe, struct xe_bo *bo, u64 value) +{ + if (value == DRM_XE_PXP_TYPE_NONE) + return 0; + + /* we only support DRM_XE_PXP_TYPE_HWDRM for now */ + if (XE_IOCTL_DBG(xe, value != DRM_XE_PXP_TYPE_HWDRM)) + return -EINVAL; + + return xe_pxp_key_assign(xe->pxp, bo); +} + +typedef int (*xe_gem_create_set_property_fn)(struct xe_device *xe, + struct xe_bo *bo, + u64 value); + +static const xe_gem_create_set_property_fn gem_create_set_property_funcs[] = { + [DRM_XE_GEM_CREATE_SET_PROPERTY_PXP_TYPE] = gem_create_set_pxp_type, +}; + +static int gem_create_user_ext_set_property(struct xe_device *xe, + struct xe_bo *bo, + u64 extension) +{ + u64 __user *address = u64_to_user_ptr(extension); + struct drm_xe_ext_set_property ext; + int err; + u32 idx; + + err = copy_from_user(&ext, address, sizeof(ext)); + if (XE_IOCTL_DBG(xe, err)) + return -EFAULT; + + if (XE_IOCTL_DBG(xe, ext.property >= + ARRAY_SIZE(gem_create_set_property_funcs)) || + XE_IOCTL_DBG(xe, ext.pad) || + XE_IOCTL_DBG(xe, ext.property != DRM_XE_GEM_CREATE_EXTENSION_SET_PROPERTY)) + return -EINVAL; + + idx = array_index_nospec(ext.property, ARRAY_SIZE(gem_create_set_property_funcs)); + if (!gem_create_set_property_funcs[idx]) + return -EINVAL; + + return gem_create_set_property_funcs[idx](xe, bo, ext.value); +} + +typedef int (*xe_gem_create_user_extension_fn)(struct xe_device *xe, + struct xe_bo *bo, + u64 extension); + +static const xe_gem_create_user_extension_fn gem_create_user_extension_funcs[] = { + [DRM_XE_GEM_CREATE_EXTENSION_SET_PROPERTY] = gem_create_user_ext_set_property, +}; + +#define MAX_USER_EXTENSIONS 16 +static int gem_create_user_extensions(struct xe_device *xe, struct xe_bo *bo, + u64 extensions, int ext_number) +{ + u64 __user *address = u64_to_user_ptr(extensions); + struct drm_xe_user_extension ext; + int err; + u32 idx; + + if (XE_IOCTL_DBG(xe, ext_number >= MAX_USER_EXTENSIONS)) + return -E2BIG; + + err = copy_from_user(&ext, address, sizeof(ext)); + if (XE_IOCTL_DBG(xe, err)) + return -EFAULT; + + if (XE_IOCTL_DBG(xe, ext.pad) || + XE_IOCTL_DBG(xe, ext.name >= ARRAY_SIZE(gem_create_user_extension_funcs))) + return -EINVAL; + + idx = array_index_nospec(ext.name, + ARRAY_SIZE(gem_create_user_extension_funcs)); + err = gem_create_user_extension_funcs[idx](xe, bo, extensions); + if (XE_IOCTL_DBG(xe, err)) + return err; + + if (ext.next_extension) + return gem_create_user_extensions(xe, bo, ext.next_extension, + ++ext_number); + + return 0; +} + int xe_gem_create_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { @@ -1949,13 +2644,13 @@ int xe_gem_create_ioctl(struct drm_device *dev, void *data, struct xe_file *xef = to_xe_file(file); struct drm_xe_gem_create *args = data; struct xe_vm *vm = NULL; + ktime_t end = 0; struct xe_bo *bo; unsigned int bo_flags; u32 handle; int err; - if (XE_IOCTL_DBG(xe, args->extensions) || - XE_IOCTL_DBG(xe, args->pad[0] || args->pad[1] || args->pad[2]) || + if (XE_IOCTL_DBG(xe, args->pad[0] || args->pad[1] || args->pad[2]) || XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1])) return -EINVAL; @@ -1984,29 +2679,36 @@ int xe_gem_create_ioctl(struct drm_device *dev, void *data, bo_flags = 0; if (args->flags & DRM_XE_GEM_CREATE_FLAG_DEFER_BACKING) - bo_flags |= XE_BO_DEFER_BACKING; + bo_flags |= XE_BO_FLAG_DEFER_BACKING; if (args->flags & DRM_XE_GEM_CREATE_FLAG_SCANOUT) - bo_flags |= XE_BO_SCANOUT_BIT; + bo_flags |= XE_BO_FLAG_SCANOUT; + + bo_flags |= args->placement << (ffs(XE_BO_FLAG_SYSTEM) - 1); - bo_flags |= args->placement << (ffs(XE_BO_CREATE_SYSTEM_BIT) - 1); + /* CCS formats need physical placement at a 64K alignment in VRAM. */ + if ((bo_flags & XE_BO_FLAG_VRAM_MASK) && + (bo_flags & XE_BO_FLAG_SCANOUT) && + !(xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) && + IS_ALIGNED(args->size, SZ_64K)) + bo_flags |= XE_BO_FLAG_NEEDS_64K; if (args->flags & DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM) { - if (XE_IOCTL_DBG(xe, !(bo_flags & XE_BO_CREATE_VRAM_MASK))) + if (XE_IOCTL_DBG(xe, !(bo_flags & XE_BO_FLAG_VRAM_MASK))) return -EINVAL; - bo_flags |= XE_BO_NEEDS_CPU_ACCESS; + bo_flags |= XE_BO_FLAG_NEEDS_CPU_ACCESS; } if (XE_IOCTL_DBG(xe, !args->cpu_caching || args->cpu_caching > DRM_XE_GEM_CPU_CACHING_WC)) return -EINVAL; - if (XE_IOCTL_DBG(xe, bo_flags & XE_BO_CREATE_VRAM_MASK && + if (XE_IOCTL_DBG(xe, bo_flags & XE_BO_FLAG_VRAM_MASK && args->cpu_caching != DRM_XE_GEM_CPU_CACHING_WC)) return -EINVAL; - if (XE_IOCTL_DBG(xe, bo_flags & XE_BO_SCANOUT_BIT && + if (XE_IOCTL_DBG(xe, bo_flags & XE_BO_FLAG_SCANOUT && args->cpu_caching == DRM_XE_GEM_CPU_CACHING_WB)) return -EINVAL; @@ -2014,22 +2716,34 @@ int xe_gem_create_ioctl(struct drm_device *dev, void *data, vm = xe_vm_lookup(xef, args->vm_id); if (XE_IOCTL_DBG(xe, !vm)) return -ENOENT; + } + +retry: + if (vm) { err = xe_vm_lock(vm, true); if (err) goto out_vm; } bo = xe_bo_create_user(xe, NULL, vm, args->size, args->cpu_caching, - ttm_bo_type_device, bo_flags); + bo_flags); if (vm) xe_vm_unlock(vm); if (IS_ERR(bo)) { err = PTR_ERR(bo); + if (xe_vm_validate_should_retry(NULL, err, &end)) + goto retry; goto out_vm; } + if (args->extensions) { + err = gem_create_user_extensions(xe, bo, args->extensions, 0); + if (err) + goto out_bulk; + } + err = drm_gem_handle_create(file, &bo->ttm.base, &handle); if (err) goto out_bulk; @@ -2063,9 +2777,26 @@ int xe_gem_mmap_offset_ioctl(struct drm_device *dev, void *data, XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1])) return -EINVAL; - if (XE_IOCTL_DBG(xe, args->flags)) + if (XE_IOCTL_DBG(xe, args->flags & + ~DRM_XE_MMAP_OFFSET_FLAG_PCI_BARRIER)) return -EINVAL; + if (args->flags & DRM_XE_MMAP_OFFSET_FLAG_PCI_BARRIER) { + if (XE_IOCTL_DBG(xe, !IS_DGFX(xe))) + return -EINVAL; + + if (XE_IOCTL_DBG(xe, args->handle)) + return -EINVAL; + + if (XE_IOCTL_DBG(xe, PAGE_SIZE > SZ_4K)) + return -EINVAL; + + BUILD_BUG_ON(((XE_PCI_BARRIER_MMAP_OFFSET >> XE_PTE_SHIFT) + + SZ_4K) >= DRM_FILE_PAGE_OFFSET_START); + args->offset = XE_PCI_BARRIER_MMAP_OFFSET; + return 0; + } + gem_obj = drm_gem_object_lookup(file, args->handle); if (XE_IOCTL_DBG(xe, !gem_obj)) return -ENOENT; @@ -2171,6 +2902,7 @@ int xe_bo_migrate(struct xe_bo *bo, u32 mem_type) struct ttm_operation_ctx ctx = { .interruptible = true, .no_wait_gpu = false, + .gfp_retry_mayfail = true, }; struct ttm_placement placement; struct ttm_place requested; @@ -2208,19 +2940,18 @@ int xe_bo_migrate(struct xe_bo *bo, u32 mem_type) /** * xe_bo_evict - Evict an object to evict placement * @bo: The buffer object to migrate. - * @force_alloc: Set force_alloc in ttm_operation_ctx * * On successful completion, the object memory will be moved to evict - * placement. Ths function blocks until the object has been fully moved. + * placement. This function blocks until the object has been fully moved. * * Return: 0 on success. Negative error code on failure. */ -int xe_bo_evict(struct xe_bo *bo, bool force_alloc) +int xe_bo_evict(struct xe_bo *bo) { struct ttm_operation_ctx ctx = { .interruptible = false, .no_wait_gpu = false, - .force_alloc = force_alloc, + .gfp_retry_mayfail = true, }; struct ttm_placement placement; int ret; @@ -2247,6 +2978,9 @@ bool xe_bo_needs_ccs_pages(struct xe_bo *bo) { struct xe_device *xe = xe_bo_device(bo); + if (GRAPHICS_VER(xe) >= 20 && IS_DGFX(xe)) + return false; + if (!xe_device_has_flat_ccs(xe) || bo->ttm.type != ttm_bo_type_device) return false; @@ -2255,7 +2989,7 @@ bool xe_bo_needs_ccs_pages(struct xe_bo *bo) * can't be used since there's no CCS storage associated with * non-VRAM addresses. */ - if (IS_DGFX(xe) && (bo->flags & XE_BO_CREATE_SYSTEM_BIT)) + if (IS_DGFX(xe) && (bo->flags & XE_BO_FLAG_SYSTEM)) return false; return true; @@ -2295,6 +3029,49 @@ void xe_bo_put_commit(struct llist_head *deferred) drm_gem_object_free(&bo->ttm.base.refcount); } +static void xe_bo_dev_work_func(struct work_struct *work) +{ + struct xe_bo_dev *bo_dev = container_of(work, typeof(*bo_dev), async_free); + + xe_bo_put_commit(&bo_dev->async_list); +} + +/** + * xe_bo_dev_init() - Initialize BO dev to manage async BO freeing + * @bo_dev: The BO dev structure + */ +void xe_bo_dev_init(struct xe_bo_dev *bo_dev) +{ + INIT_WORK(&bo_dev->async_free, xe_bo_dev_work_func); +} + +/** + * xe_bo_dev_fini() - Finalize BO dev managing async BO freeing + * @bo_dev: The BO dev structure + */ +void xe_bo_dev_fini(struct xe_bo_dev *bo_dev) +{ + flush_work(&bo_dev->async_free); +} + +void xe_bo_put(struct xe_bo *bo) +{ + struct xe_tile *tile; + u8 id; + + might_sleep(); + if (bo) { +#ifdef CONFIG_PROC_FS + if (bo->client) + might_lock(&bo->client->bos_lock); +#endif + for_each_tile(tile, xe_bo_device(bo), id) + if (bo->ggtt_node[id] && bo->ggtt_node[id]->ggtt) + might_lock(&bo->ggtt_node[id]->ggtt->lock); + drm_gem_object_put(&bo->ttm.base); + } +} + /** * xe_bo_dumb_create - Create a dumb bo as backing for a fb * @file_priv: ... @@ -2323,10 +3100,9 @@ int xe_bo_dumb_create(struct drm_file *file_priv, bo = xe_bo_create_user(xe, NULL, NULL, args->size, DRM_XE_GEM_CPU_CACHING_WC, - ttm_bo_type_device, - XE_BO_CREATE_VRAM_IF_DGFX(xe_device_get_root_tile(xe)) | - XE_BO_CREATE_USER_BIT | XE_BO_SCANOUT_BIT | - XE_BO_NEEDS_CPU_ACCESS); + XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(xe)) | + XE_BO_FLAG_SCANOUT | + XE_BO_FLAG_NEEDS_CPU_ACCESS); if (IS_ERR(bo)) return PTR_ERR(bo); |