diff options
Diffstat (limited to 'drivers/gpu/drm/xe')
40 files changed, 436 insertions, 331 deletions
diff --git a/drivers/gpu/drm/xe/Kconfig b/drivers/gpu/drm/xe/Kconfig index 1cced50d8d8c..e36ae1f0d885 100644 --- a/drivers/gpu/drm/xe/Kconfig +++ b/drivers/gpu/drm/xe/Kconfig @@ -47,7 +47,7 @@ config DRM_XE config DRM_XE_DISPLAY bool "Enable display support" - depends on DRM_XE && EXPERT && DRM_XE=m + depends on DRM_XE && DRM_XE=m select FB_IOMEM_HELPERS select I2C select I2C_ALGOBIT diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile index 53bd2a8ba1ae..efcf0ab7a1a6 100644 --- a/drivers/gpu/drm/xe/Makefile +++ b/drivers/gpu/drm/xe/Makefile @@ -17,7 +17,6 @@ subdir-ccflags-y += $(call cc-option, -Wunused-const-variable) subdir-ccflags-y += $(call cc-option, -Wpacked-not-aligned) subdir-ccflags-y += $(call cc-option, -Wformat-overflow) subdir-ccflags-y += $(call cc-option, -Wformat-truncation) -subdir-ccflags-y += $(call cc-option, -Wstringop-overflow) subdir-ccflags-y += $(call cc-option, -Wstringop-truncation) # The following turn off the warnings enabled by -Wextra ifeq ($(findstring 2, $(KBUILD_EXTRA_WARN)),) diff --git a/drivers/gpu/drm/xe/abi/guc_actions_abi.h b/drivers/gpu/drm/xe/abi/guc_actions_abi.h index 3062e0e0d467..79ba98a169f9 100644 --- a/drivers/gpu/drm/xe/abi/guc_actions_abi.h +++ b/drivers/gpu/drm/xe/abi/guc_actions_abi.h @@ -50,8 +50,8 @@ #define HOST2GUC_SELF_CFG_REQUEST_MSG_LEN (GUC_HXG_REQUEST_MSG_MIN_LEN + 3u) #define HOST2GUC_SELF_CFG_REQUEST_MSG_0_MBZ GUC_HXG_REQUEST_MSG_0_DATA0 -#define HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_KEY (0xffff << 16) -#define HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_LEN (0xffff << 0) +#define HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_KEY (0xffffu << 16) +#define HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_LEN (0xffffu << 0) #define HOST2GUC_SELF_CFG_REQUEST_MSG_2_VALUE32 GUC_HXG_REQUEST_MSG_n_DATAn #define HOST2GUC_SELF_CFG_REQUEST_MSG_3_VALUE64 GUC_HXG_REQUEST_MSG_n_DATAn diff --git a/drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h b/drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h index 811add10c30d..c165e26c0976 100644 --- a/drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h +++ b/drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h @@ -242,8 +242,8 @@ struct slpc_shared_data { (HOST2GUC_PC_SLPC_REQUEST_REQUEST_MSG_MIN_LEN + \ HOST2GUC_PC_SLPC_EVENT_MAX_INPUT_ARGS) #define HOST2GUC_PC_SLPC_REQUEST_MSG_0_MBZ GUC_HXG_REQUEST_MSG_0_DATA0 -#define HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ID (0xff << 8) -#define HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ARGC (0xff << 0) +#define HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ID (0xffu << 8) +#define HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ARGC (0xffu << 0) #define HOST2GUC_PC_SLPC_REQUEST_MSG_N_EVENT_DATA_N GUC_HXG_REQUEST_MSG_n_DATAn #endif diff --git a/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h b/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h index 3b83f907ece4..0b1146d0c997 100644 --- a/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h +++ b/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h @@ -82,11 +82,11 @@ static_assert(sizeof(struct guc_ct_buffer_desc) == 64); #define GUC_CTB_HDR_LEN 1u #define GUC_CTB_MSG_MIN_LEN GUC_CTB_HDR_LEN #define GUC_CTB_MSG_MAX_LEN 256u -#define GUC_CTB_MSG_0_FENCE (0xffff << 16) -#define GUC_CTB_MSG_0_FORMAT (0xf << 12) +#define GUC_CTB_MSG_0_FENCE (0xffffu << 16) +#define GUC_CTB_MSG_0_FORMAT (0xfu << 12) #define GUC_CTB_FORMAT_HXG 0u -#define GUC_CTB_MSG_0_RESERVED (0xf << 8) -#define GUC_CTB_MSG_0_NUM_DWORDS (0xff << 0) +#define GUC_CTB_MSG_0_RESERVED (0xfu << 8) +#define GUC_CTB_MSG_0_NUM_DWORDS (0xffu << 0) /** * DOC: CTB HXG Message diff --git a/drivers/gpu/drm/xe/abi/guc_klvs_abi.h b/drivers/gpu/drm/xe/abi/guc_klvs_abi.h index 47094b9b044c..0400bc0fccdc 100644 --- a/drivers/gpu/drm/xe/abi/guc_klvs_abi.h +++ b/drivers/gpu/drm/xe/abi/guc_klvs_abi.h @@ -31,9 +31,9 @@ */ #define GUC_KLV_LEN_MIN 1u -#define GUC_KLV_0_KEY (0xffff << 16) -#define GUC_KLV_0_LEN (0xffff << 0) -#define GUC_KLV_n_VALUE (0xffffffff << 0) +#define GUC_KLV_0_KEY (0xffffu << 16) +#define GUC_KLV_0_LEN (0xffffu << 0) +#define GUC_KLV_n_VALUE (0xffffffffu << 0) /** * DOC: GuC Self Config KLVs diff --git a/drivers/gpu/drm/xe/abi/guc_messages_abi.h b/drivers/gpu/drm/xe/abi/guc_messages_abi.h index 3d199016cf88..29e414c82d56 100644 --- a/drivers/gpu/drm/xe/abi/guc_messages_abi.h +++ b/drivers/gpu/drm/xe/abi/guc_messages_abi.h @@ -40,18 +40,18 @@ */ #define GUC_HXG_MSG_MIN_LEN 1u -#define GUC_HXG_MSG_0_ORIGIN (0x1 << 31) +#define GUC_HXG_MSG_0_ORIGIN (0x1u << 31) #define GUC_HXG_ORIGIN_HOST 0u #define GUC_HXG_ORIGIN_GUC 1u -#define GUC_HXG_MSG_0_TYPE (0x7 << 28) +#define GUC_HXG_MSG_0_TYPE (0x7u << 28) #define GUC_HXG_TYPE_REQUEST 0u #define GUC_HXG_TYPE_EVENT 1u #define GUC_HXG_TYPE_NO_RESPONSE_BUSY 3u #define GUC_HXG_TYPE_NO_RESPONSE_RETRY 5u #define GUC_HXG_TYPE_RESPONSE_FAILURE 6u #define GUC_HXG_TYPE_RESPONSE_SUCCESS 7u -#define GUC_HXG_MSG_0_AUX (0xfffffff << 0) -#define GUC_HXG_MSG_n_PAYLOAD (0xffffffff << 0) +#define GUC_HXG_MSG_0_AUX (0xfffffffu << 0) +#define GUC_HXG_MSG_n_PAYLOAD (0xffffffffu << 0) /** * DOC: HXG Request @@ -85,8 +85,8 @@ */ #define GUC_HXG_REQUEST_MSG_MIN_LEN GUC_HXG_MSG_MIN_LEN -#define GUC_HXG_REQUEST_MSG_0_DATA0 (0xfff << 16) -#define GUC_HXG_REQUEST_MSG_0_ACTION (0xffff << 0) +#define GUC_HXG_REQUEST_MSG_0_DATA0 (0xfffu << 16) +#define GUC_HXG_REQUEST_MSG_0_ACTION (0xffffu << 0) #define GUC_HXG_REQUEST_MSG_n_DATAn GUC_HXG_MSG_n_PAYLOAD /** @@ -117,8 +117,8 @@ */ #define GUC_HXG_EVENT_MSG_MIN_LEN GUC_HXG_MSG_MIN_LEN -#define GUC_HXG_EVENT_MSG_0_DATA0 (0xfff << 16) -#define GUC_HXG_EVENT_MSG_0_ACTION (0xffff << 0) +#define GUC_HXG_EVENT_MSG_0_DATA0 (0xfffu << 16) +#define GUC_HXG_EVENT_MSG_0_ACTION (0xffffu << 0) #define GUC_HXG_EVENT_MSG_n_DATAn GUC_HXG_MSG_n_PAYLOAD /** @@ -188,8 +188,8 @@ */ #define GUC_HXG_FAILURE_MSG_LEN GUC_HXG_MSG_MIN_LEN -#define GUC_HXG_FAILURE_MSG_0_HINT (0xfff << 16) -#define GUC_HXG_FAILURE_MSG_0_ERROR (0xffff << 0) +#define GUC_HXG_FAILURE_MSG_0_HINT (0xfffu << 16) +#define GUC_HXG_FAILURE_MSG_0_ERROR (0xffffu << 0) /** * DOC: HXG Response diff --git a/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h index 5f19550cc845..68d9f6116bdf 100644 --- a/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h +++ b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h @@ -35,12 +35,10 @@ static inline int i915_gem_object_read_from_page(struct xe_bo *bo, u32 ofs, u64 *ptr, u32 size) { struct ttm_bo_kmap_obj map; - void *virtual; + void *src; bool is_iomem; int ret; - XE_WARN_ON(size != 8); - ret = xe_bo_lock(bo, true); if (ret) return ret; @@ -50,11 +48,12 @@ static inline int i915_gem_object_read_from_page(struct xe_bo *bo, goto out_unlock; ofs &= ~PAGE_MASK; - virtual = ttm_kmap_obj_virtual(&map, &is_iomem); + src = ttm_kmap_obj_virtual(&map, &is_iomem); + src += ofs; if (is_iomem) - *ptr = readq((void __iomem *)(virtual + ofs)); + memcpy_fromio(ptr, (void __iomem *)src, size); else - *ptr = *(u64 *)(virtual + ofs); + memcpy(ptr, src, size); ttm_bo_kunmap(&map); out_unlock: diff --git a/drivers/gpu/drm/xe/tests/xe_bo.c b/drivers/gpu/drm/xe/tests/xe_bo.c index 412b2e7ce40c..3436fd9cf2b2 100644 --- a/drivers/gpu/drm/xe/tests/xe_bo.c +++ b/drivers/gpu/drm/xe/tests/xe_bo.c @@ -125,14 +125,13 @@ static void ccs_test_run_tile(struct xe_device *xe, struct xe_tile *tile, bo = xe_bo_create_user(xe, NULL, NULL, SZ_1M, DRM_XE_GEM_CPU_CACHING_WC, ttm_bo_type_device, bo_flags); - - xe_bo_lock(bo, false); - if (IS_ERR(bo)) { KUNIT_FAIL(test, "Failed to create bo.\n"); return; } + xe_bo_lock(bo, false); + kunit_info(test, "Verifying that CCS data is cleared on creation.\n"); ret = ccs_test_migrate(tile, bo, false, 0ULL, 0xdeadbeefdeadbeefULL, test); diff --git a/drivers/gpu/drm/xe/tests/xe_migrate.c b/drivers/gpu/drm/xe/tests/xe_migrate.c index 7a32faa2f688..a6523df0f1d3 100644 --- a/drivers/gpu/drm/xe/tests/xe_migrate.c +++ b/drivers/gpu/drm/xe/tests/xe_migrate.c @@ -331,7 +331,7 @@ static void xe_migrate_sanity_test(struct xe_migrate *m, struct kunit *test) xe_res_first_sg(xe_bo_sg(pt), 0, pt->size, &src_it); emit_pte(m, bb, NUM_KERNEL_PDE - 1, xe_bo_is_vram(pt), false, - &src_it, XE_PAGE_SIZE, pt); + &src_it, XE_PAGE_SIZE, pt->ttm.resource); run_sanity_job(m, xe, bb, bb->len, "Writing PTE for our fake PT", test); diff --git a/drivers/gpu/drm/xe/tests/xe_wa_test.c b/drivers/gpu/drm/xe/tests/xe_wa_test.c index a53c22a19582..b4715b78ef3b 100644 --- a/drivers/gpu/drm/xe/tests/xe_wa_test.c +++ b/drivers/gpu/drm/xe/tests/xe_wa_test.c @@ -74,9 +74,6 @@ static const struct platform_test_case cases[] = { SUBPLATFORM_CASE(DG2, G11, B1), SUBPLATFORM_CASE(DG2, G12, A0), SUBPLATFORM_CASE(DG2, G12, A1), - PLATFORM_CASE(PVC, B0), - PLATFORM_CASE(PVC, B1), - PLATFORM_CASE(PVC, C0), GMDID_CASE(METEORLAKE, 1270, A0, 1300, A0), GMDID_CASE(METEORLAKE, 1271, A0, 1300, A0), GMDID_CASE(LUNARLAKE, 2004, A0, 2000, A0), diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c index 8e4a3b1f6b93..0b0e262e2166 100644 --- a/drivers/gpu/drm/xe/xe_bo.c +++ b/drivers/gpu/drm/xe/xe_bo.c @@ -125,9 +125,9 @@ static struct xe_mem_region *res_to_mem_region(struct ttm_resource *res) static void try_add_system(struct xe_device *xe, struct xe_bo *bo, u32 bo_flags, u32 *c) { - xe_assert(xe, *c < ARRAY_SIZE(bo->placements)); - if (bo_flags & XE_BO_CREATE_SYSTEM_BIT) { + xe_assert(xe, *c < ARRAY_SIZE(bo->placements)); + bo->placements[*c] = (struct ttm_place) { .mem_type = XE_PL_TT, }; @@ -145,6 +145,8 @@ static void add_vram(struct xe_device *xe, struct xe_bo *bo, struct xe_mem_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; xe_assert(xe, vram && vram->usable_size); io_size = vram->io_size; @@ -175,8 +177,6 @@ static void add_vram(struct xe_device *xe, struct xe_bo *bo, static void try_add_vram(struct xe_device *xe, struct xe_bo *bo, u32 bo_flags, u32 *c) { - xe_assert(xe, *c < ARRAY_SIZE(bo->placements)); - 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); @@ -193,9 +193,9 @@ static void try_add_vram(struct xe_device *xe, struct xe_bo *bo, static void try_add_stolen(struct xe_device *xe, struct xe_bo *bo, u32 bo_flags, u32 *c) { - xe_assert(xe, *c < ARRAY_SIZE(bo->placements)); - if (bo_flags & XE_BO_CREATE_STOLEN_BIT) { + 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 | @@ -442,7 +442,7 @@ static int xe_ttm_io_mem_reserve(struct ttm_device *bdev, if (vram->mapping && mem->placement & TTM_PL_FLAG_CONTIGUOUS) - mem->bus.addr = (u8 *)vram->mapping + + mem->bus.addr = (u8 __force *)vram->mapping + mem->bus.offset; mem->bus.offset += vram->io_start; @@ -734,7 +734,7 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict, /* 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 *new_addr = vram->mapping + + void __iomem *new_addr = vram->mapping + (new_mem->start << PAGE_SHIFT); if (XE_WARN_ON(new_mem->start == XE_BO_INVALID_OFFSET)) { diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c index d9ae77fe7382..1f0b4b9ce84f 100644 --- a/drivers/gpu/drm/xe/xe_device.c +++ b/drivers/gpu/drm/xe/xe_device.c @@ -484,7 +484,7 @@ int xe_device_probe(struct xe_device *xe) err = xe_device_set_has_flat_ccs(xe); if (err) - return err; + goto err_irq_shutdown; err = xe_mmio_probe_vram(xe); if (err) @@ -613,7 +613,7 @@ void xe_device_wmb(struct xe_device *xe) u32 xe_device_ccs_bytes(struct xe_device *xe, u64 size) { return xe_device_has_flat_ccs(xe) ? - DIV_ROUND_UP(size, NUM_BYTES_PER_CCS_BYTE(xe)) : 0; + DIV_ROUND_UP_ULL(size, NUM_BYTES_PER_CCS_BYTE(xe)) : 0; } bool xe_device_mem_access_ongoing(struct xe_device *xe) diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h index c45ef17b3473..5dc9127a2029 100644 --- a/drivers/gpu/drm/xe/xe_device_types.h +++ b/drivers/gpu/drm/xe/xe_device_types.h @@ -97,7 +97,7 @@ struct xe_mem_region { */ resource_size_t actual_physical_size; /** @mapping: pointer to VRAM mappable space */ - void *__iomem mapping; + void __iomem *mapping; }; /** @@ -146,7 +146,7 @@ struct xe_tile { size_t size; /** @regs: pointer to tile's MMIO space (starting with registers) */ - void *regs; + void __iomem *regs; } mmio; /** @@ -159,7 +159,7 @@ struct xe_tile { size_t size; /** @regs: pointer to tile's additional MMIO-extension space */ - void *regs; + void __iomem *regs; } mmio_ext; /** @mem: memory management info for tile */ @@ -301,7 +301,7 @@ struct xe_device { /** @size: size of MMIO space for device */ size_t size; /** @regs: pointer to MMIO space for device */ - void *regs; + void __iomem *regs; } mmio; /** @mem: memory info for device */ diff --git a/drivers/gpu/drm/xe/xe_display.c b/drivers/gpu/drm/xe/xe_display.c index 74391d9b11ae..e4db069f0db3 100644 --- a/drivers/gpu/drm/xe/xe_display.c +++ b/drivers/gpu/drm/xe/xe_display.c @@ -134,8 +134,6 @@ static void xe_display_fini_nommio(struct drm_device *dev, void *dummy) int xe_display_init_nommio(struct xe_device *xe) { - int err; - if (!xe->info.enable_display) return 0; @@ -145,10 +143,6 @@ int xe_display_init_nommio(struct xe_device *xe) /* This must be called before any calls to HAS_PCH_* */ intel_detect_pch(xe); - err = intel_power_domains_init(xe); - if (err) - return err; - return drmm_add_action_or_reset(&xe->drm, xe_display_fini_nommio, xe); } diff --git a/drivers/gpu/drm/xe/xe_dma_buf.c b/drivers/gpu/drm/xe/xe_dma_buf.c index 64ed303728fd..da2627ed6ae7 100644 --- a/drivers/gpu/drm/xe/xe_dma_buf.c +++ b/drivers/gpu/drm/xe/xe_dma_buf.c @@ -175,7 +175,7 @@ static int xe_dma_buf_begin_cpu_access(struct dma_buf *dma_buf, return 0; } -const struct dma_buf_ops xe_dmabuf_ops = { +static const struct dma_buf_ops xe_dmabuf_ops = { .attach = xe_dma_buf_attach, .detach = xe_dma_buf_detach, .pin = xe_dma_buf_pin, diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c index d30c0d0689bc..17f26952e665 100644 --- a/drivers/gpu/drm/xe/xe_exec.c +++ b/drivers/gpu/drm/xe/xe_exec.c @@ -111,11 +111,11 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) u64 addresses[XE_HW_ENGINE_MAX_INSTANCE]; struct drm_gpuvm_exec vm_exec = {.extra.fn = xe_exec_fn}; struct drm_exec *exec = &vm_exec.exec; - u32 i, num_syncs = 0; + u32 i, num_syncs = 0, num_ufence = 0; struct xe_sched_job *job; struct dma_fence *rebind_fence; struct xe_vm *vm; - bool write_locked; + bool write_locked, skip_retry = false; ktime_t end = 0; int err = 0; @@ -157,6 +157,14 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) SYNC_PARSE_FLAG_LR_MODE : 0)); if (err) goto err_syncs; + + if (xe_sync_is_ufence(&syncs[i])) + num_ufence++; + } + + if (XE_IOCTL_DBG(xe, num_ufence > 1)) { + err = -EINVAL; + goto err_syncs; } if (xe_exec_queue_is_parallel(q)) { @@ -227,7 +235,8 @@ retry: } if (xe_exec_queue_is_lr(q) && xe_exec_queue_ring_full(q)) { - err = -EWOULDBLOCK; + err = -EWOULDBLOCK; /* Aliased to -EAGAIN */ + skip_retry = true; goto err_exec; } @@ -337,7 +346,7 @@ err_unlock_list: up_write(&vm->lock); else up_read(&vm->lock); - if (err == -EAGAIN) + if (err == -EAGAIN && !skip_retry) goto retry; err_syncs: for (i = 0; i < num_syncs; i++) diff --git a/drivers/gpu/drm/xe/xe_exec_queue.c b/drivers/gpu/drm/xe/xe_exec_queue.c index 44fe8097b7cd..254b1d3af4cb 100644 --- a/drivers/gpu/drm/xe/xe_exec_queue.c +++ b/drivers/gpu/drm/xe/xe_exec_queue.c @@ -67,6 +67,11 @@ static struct xe_exec_queue *__xe_exec_queue_create(struct xe_device *xe, q->sched_props.timeslice_us = hwe->eclass->sched_props.timeslice_us; q->sched_props.preempt_timeout_us = hwe->eclass->sched_props.preempt_timeout_us; + if (q->flags & EXEC_QUEUE_FLAG_KERNEL && + q->flags & EXEC_QUEUE_FLAG_HIGH_PRIORITY) + q->sched_props.priority = XE_EXEC_QUEUE_PRIORITY_KERNEL; + else + q->sched_props.priority = XE_EXEC_QUEUE_PRIORITY_NORMAL; if (xe_exec_queue_is_parallel(q)) { q->parallel.composite_fence_ctx = dma_fence_context_alloc(1); @@ -921,20 +926,24 @@ void xe_exec_queue_last_fence_put_unlocked(struct xe_exec_queue *q) * @q: The exec queue * @vm: The VM the engine does a bind or exec for * - * Get last fence, does not take a ref + * Get last fence, takes a ref * * Returns: last fence if not signaled, dma fence stub if signaled */ struct dma_fence *xe_exec_queue_last_fence_get(struct xe_exec_queue *q, struct xe_vm *vm) { + struct dma_fence *fence; + xe_exec_queue_last_fence_lockdep_assert(q, vm); if (q->last_fence && test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &q->last_fence->flags)) xe_exec_queue_last_fence_put(q, vm); - return q->last_fence ? q->last_fence : dma_fence_get_stub(); + fence = q->last_fence ? q->last_fence : dma_fence_get_stub(); + dma_fence_get(fence); + return fence; } /** diff --git a/drivers/gpu/drm/xe/xe_exec_queue_types.h b/drivers/gpu/drm/xe/xe_exec_queue_types.h index 3d7e704ec3d9..8d4b7feb8c30 100644 --- a/drivers/gpu/drm/xe/xe_exec_queue_types.h +++ b/drivers/gpu/drm/xe/xe_exec_queue_types.h @@ -52,8 +52,6 @@ struct xe_exec_queue { struct xe_vm *vm; /** @class: class of this exec queue */ enum xe_engine_class class; - /** @priority: priority of this exec queue */ - enum xe_exec_queue_priority priority; /** * @logical_mask: logical mask of where job submitted to exec queue can run */ @@ -84,6 +82,8 @@ struct xe_exec_queue { #define EXEC_QUEUE_FLAG_VM BIT(4) /* child of VM queue for multi-tile VM jobs */ #define EXEC_QUEUE_FLAG_BIND_ENGINE_CHILD BIT(5) +/* kernel exec_queue only, set priority to highest level */ +#define EXEC_QUEUE_FLAG_HIGH_PRIORITY BIT(6) /** * @flags: flags for this exec queue, should statically setup aside from ban @@ -142,6 +142,8 @@ struct xe_exec_queue { u32 timeslice_us; /** @preempt_timeout_us: preemption timeout in micro-seconds */ u32 preempt_timeout_us; + /** @priority: priority of this exec queue */ + enum xe_exec_queue_priority priority; } sched_props; /** @compute: compute exec queue state */ diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c index 3af2adec1295..35474ddbaf97 100644 --- a/drivers/gpu/drm/xe/xe_gt.c +++ b/drivers/gpu/drm/xe/xe_gt.c @@ -437,7 +437,10 @@ static int all_fw_domain_init(struct xe_gt *gt) * USM has its only SA pool to non-block behind user operations */ if (gt_to_xe(gt)->info.has_usm) { - gt->usm.bb_pool = xe_sa_bo_manager_init(gt_to_tile(gt), SZ_1M, 16); + struct xe_device *xe = gt_to_xe(gt); + + gt->usm.bb_pool = xe_sa_bo_manager_init(gt_to_tile(gt), + IS_DGFX(xe) ? SZ_1M : SZ_512K, 16); if (IS_ERR(gt->usm.bb_pool)) { err = PTR_ERR(gt->usm.bb_pool); goto err_force_wake; diff --git a/drivers/gpu/drm/xe/xe_gt_freq.c b/drivers/gpu/drm/xe/xe_gt_freq.c index 3adfa6686e7c..e5b0f4ecdbe8 100644 --- a/drivers/gpu/drm/xe/xe_gt_freq.c +++ b/drivers/gpu/drm/xe/xe_gt_freq.c @@ -196,6 +196,9 @@ void xe_gt_freq_init(struct xe_gt *gt) struct xe_device *xe = gt_to_xe(gt); int err; + if (xe->info.skip_guc_pc) + return; + gt->freq = kobject_create_and_add("freq0", gt->sysfs); if (!gt->freq) { drm_warn(&xe->drm, "failed to add freq0 directory to %s\n", diff --git a/drivers/gpu/drm/xe/xe_gt_mcr.c b/drivers/gpu/drm/xe/xe_gt_mcr.c index 77925b35cf8d..8546cd3cc50d 100644 --- a/drivers/gpu/drm/xe/xe_gt_mcr.c +++ b/drivers/gpu/drm/xe/xe_gt_mcr.c @@ -480,7 +480,7 @@ static bool xe_gt_mcr_get_nonterminated_steering(struct xe_gt *gt, * to synchronize with external clients (e.g., firmware), so a semaphore * register will also need to be taken. */ -static void mcr_lock(struct xe_gt *gt) +static void mcr_lock(struct xe_gt *gt) __acquires(>->mcr_lock) { struct xe_device *xe = gt_to_xe(gt); int ret = 0; @@ -500,7 +500,7 @@ static void mcr_lock(struct xe_gt *gt) drm_WARN_ON_ONCE(&xe->drm, ret == -ETIMEDOUT); } -static void mcr_unlock(struct xe_gt *gt) +static void mcr_unlock(struct xe_gt *gt) __releases(>->mcr_lock) { /* Release hardware semaphore - this is done by writing 1 to the register */ if (GRAPHICS_VERx100(gt_to_xe(gt)) >= 1270) diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.c b/drivers/gpu/drm/xe/xe_gt_pagefault.c index 59a70d2e0a7a..73f08f1924df 100644 --- a/drivers/gpu/drm/xe/xe_gt_pagefault.c +++ b/drivers/gpu/drm/xe/xe_gt_pagefault.c @@ -165,7 +165,8 @@ retry_userptr: goto unlock_vm; } - if (!xe_vma_is_userptr(vma) || !xe_vma_userptr_check_repin(vma)) { + if (!xe_vma_is_userptr(vma) || + !xe_vma_userptr_check_repin(to_userptr_vma(vma))) { downgrade_write(&vm->lock); write_locked = false; } @@ -181,11 +182,13 @@ retry_userptr: /* TODO: Validate fault */ if (xe_vma_is_userptr(vma) && write_locked) { + struct xe_userptr_vma *uvma = to_userptr_vma(vma); + spin_lock(&vm->userptr.invalidated_lock); - list_del_init(&vma->userptr.invalidate_link); + list_del_init(&uvma->userptr.invalidate_link); spin_unlock(&vm->userptr.invalidated_lock); - ret = xe_vma_userptr_pin_pages(vma); + ret = xe_vma_userptr_pin_pages(uvma); if (ret) goto unlock_vm; @@ -220,7 +223,7 @@ retry_userptr: dma_fence_put(fence); if (xe_vma_is_userptr(vma)) - ret = xe_vma_userptr_check_repin(vma); + ret = xe_vma_userptr_check_repin(to_userptr_vma(vma)); vma->usm.tile_invalidated &= ~BIT(tile->id); unlock_dma_resv: @@ -332,7 +335,7 @@ int xe_guc_pagefault_handler(struct xe_guc *guc, u32 *msg, u32 len) return -EPROTO; asid = FIELD_GET(PFD_ASID, msg[1]); - pf_queue = >->usm.pf_queue[asid % NUM_PF_QUEUE]; + pf_queue = gt->usm.pf_queue + (asid % NUM_PF_QUEUE); spin_lock_irqsave(&pf_queue->lock, flags); full = pf_queue_full(pf_queue); diff --git a/drivers/gpu/drm/xe/xe_guc.c b/drivers/gpu/drm/xe/xe_guc.c index 482cb0df9f15..0a61390c64a7 100644 --- a/drivers/gpu/drm/xe/xe_guc.c +++ b/drivers/gpu/drm/xe/xe_guc.c @@ -60,7 +60,12 @@ static u32 guc_ctl_debug_flags(struct xe_guc *guc) static u32 guc_ctl_feature_flags(struct xe_guc *guc) { - return GUC_CTL_ENABLE_SLPC; + u32 flags = 0; + + if (!guc_to_xe(guc)->info.skip_guc_pc) + flags |= GUC_CTL_ENABLE_SLPC; + + return flags; } static u32 guc_ctl_log_params_flags(struct xe_guc *guc) diff --git a/drivers/gpu/drm/xe/xe_guc_pc.c b/drivers/gpu/drm/xe/xe_guc_pc.c index f71085228cb3..d91702592520 100644 --- a/drivers/gpu/drm/xe/xe_guc_pc.c +++ b/drivers/gpu/drm/xe/xe_guc_pc.c @@ -963,7 +963,9 @@ void xe_guc_pc_fini(struct xe_guc_pc *pc) struct xe_device *xe = pc_to_xe(pc); if (xe->info.skip_guc_pc) { + xe_device_mem_access_get(xe); xe_gt_idle_disable_c6(pc_to_gt(pc)); + xe_device_mem_access_put(xe); return; } diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c index 21ac68e3246f..54ffcfcdd41f 100644 --- a/drivers/gpu/drm/xe/xe_guc_submit.c +++ b/drivers/gpu/drm/xe/xe_guc_submit.c @@ -421,7 +421,7 @@ static void init_policies(struct xe_guc *guc, struct xe_exec_queue *q) { struct exec_queue_policy policy; struct xe_device *xe = guc_to_xe(guc); - enum xe_exec_queue_priority prio = q->priority; + enum xe_exec_queue_priority prio = q->sched_props.priority; u32 timeslice_us = q->sched_props.timeslice_us; u32 preempt_timeout_us = q->sched_props.preempt_timeout_us; @@ -1231,7 +1231,6 @@ static int guc_exec_queue_init(struct xe_exec_queue *q) err = xe_sched_entity_init(&ge->entity, sched); if (err) goto err_sched; - q->priority = XE_EXEC_QUEUE_PRIORITY_NORMAL; if (xe_exec_queue_is_lr(q)) INIT_WORK(&q->guc->lr_tdr, xe_guc_exec_queue_lr_cleanup); @@ -1301,15 +1300,15 @@ static int guc_exec_queue_set_priority(struct xe_exec_queue *q, { struct xe_sched_msg *msg; - if (q->priority == priority || exec_queue_killed_or_banned(q)) + if (q->sched_props.priority == priority || exec_queue_killed_or_banned(q)) return 0; msg = kmalloc(sizeof(*msg), GFP_KERNEL); if (!msg) return -ENOMEM; + q->sched_props.priority = priority; guc_exec_queue_add_msg(q, msg, SET_SCHED_PROPS); - q->priority = priority; return 0; } diff --git a/drivers/gpu/drm/xe/xe_hw_fence.c b/drivers/gpu/drm/xe/xe_hw_fence.c index a6094c81f2ad..a5de3e7b0bd6 100644 --- a/drivers/gpu/drm/xe/xe_hw_fence.c +++ b/drivers/gpu/drm/xe/xe_hw_fence.c @@ -217,13 +217,13 @@ struct xe_hw_fence *xe_hw_fence_create(struct xe_hw_fence_ctx *ctx, if (!fence) return ERR_PTR(-ENOMEM); - dma_fence_init(&fence->dma, &xe_hw_fence_ops, &ctx->irq->lock, - ctx->dma_fence_ctx, ctx->next_seqno++); - fence->ctx = ctx; fence->seqno_map = seqno_map; INIT_LIST_HEAD(&fence->irq_link); + dma_fence_init(&fence->dma, &xe_hw_fence_ops, &ctx->irq->lock, + ctx->dma_fence_ctx, ctx->next_seqno++); + trace_xe_hw_fence_create(fence); return fence; diff --git a/drivers/gpu/drm/xe/xe_hwmon.c b/drivers/gpu/drm/xe/xe_hwmon.c index 6ef2aa1eae8b..174ed2185481 100644 --- a/drivers/gpu/drm/xe/xe_hwmon.c +++ b/drivers/gpu/drm/xe/xe_hwmon.c @@ -419,7 +419,7 @@ static int xe_hwmon_pcode_read_i1(struct xe_gt *gt, u32 *uval) return xe_pcode_read(gt, PCODE_MBOX(PCODE_POWER_SETUP, POWER_SETUP_SUBCOMMAND_READ_I1, 0), - uval, 0); + uval, NULL); } static int xe_hwmon_pcode_write_i1(struct xe_gt *gt, u32 uval) diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c index b7fa3831b684..0ec5ad2539f1 100644 --- a/drivers/gpu/drm/xe/xe_lrc.c +++ b/drivers/gpu/drm/xe/xe_lrc.c @@ -21,10 +21,10 @@ #include "xe_map.h" #include "xe_vm.h" -#define CTX_VALID (1 << 0) -#define CTX_PRIVILEGE (1 << 8) -#define CTX_ADDRESSING_MODE_SHIFT 3 -#define LEGACY_64B_CONTEXT 3 +#define LRC_VALID (1 << 0) +#define LRC_PRIVILEGE (1 << 8) +#define LRC_ADDRESSING_MODE_SHIFT 3 +#define LRC_LEGACY_64B_CONTEXT 3 #define ENGINE_CLASS_SHIFT 61 #define ENGINE_INSTANCE_SHIFT 48 @@ -762,15 +762,15 @@ int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe, (q->usm.acc_notify << ACC_NOTIFY_S) | q->usm.acc_trigger); - lrc->desc = CTX_VALID; - lrc->desc |= LEGACY_64B_CONTEXT << CTX_ADDRESSING_MODE_SHIFT; + lrc->desc = LRC_VALID; + lrc->desc |= LRC_LEGACY_64B_CONTEXT << LRC_ADDRESSING_MODE_SHIFT; /* TODO: Priority */ /* While this appears to have something about privileged batches or * some such, it really just means PPGTT mode. */ if (vm) - lrc->desc |= CTX_PRIVILEGE; + lrc->desc |= LRC_PRIVILEGE; if (GRAPHICS_VERx100(xe) < 1250) { lrc->desc |= (u64)hwe->instance << ENGINE_INSTANCE_SHIFT; diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c index adf1dab5eba2..70480c305602 100644 --- a/drivers/gpu/drm/xe/xe_migrate.c +++ b/drivers/gpu/drm/xe/xe_migrate.c @@ -62,6 +62,8 @@ struct xe_migrate { * out of the pt_bo. */ struct drm_suballoc_manager vm_update_sa; + /** @min_chunk_size: For dgfx, Minimum chunk size */ + u64 min_chunk_size; }; #define MAX_PREEMPTDISABLE_TRANSFER SZ_8M /* Around 1ms. */ @@ -168,11 +170,6 @@ static int xe_migrate_prepare_vm(struct xe_tile *tile, struct xe_migrate *m, if (!IS_DGFX(xe)) { /* Write out batch too */ m->batch_base_ofs = NUM_PT_SLOTS * XE_PAGE_SIZE; - if (xe->info.has_usm) { - batch = tile->primary_gt->usm.bb_pool->bo; - m->usm_batch_base_ofs = m->batch_base_ofs; - } - for (i = 0; i < batch->size; i += vm->flags & XE_VM_FLAG_64K ? XE_64K_PAGE_SIZE : XE_PAGE_SIZE) { @@ -183,6 +180,24 @@ static int xe_migrate_prepare_vm(struct xe_tile *tile, struct xe_migrate *m, entry); level++; } + if (xe->info.has_usm) { + xe_tile_assert(tile, batch->size == SZ_1M); + + batch = tile->primary_gt->usm.bb_pool->bo; + m->usm_batch_base_ofs = m->batch_base_ofs + SZ_1M; + xe_tile_assert(tile, batch->size == SZ_512K); + + for (i = 0; i < batch->size; + i += vm->flags & XE_VM_FLAG_64K ? XE_64K_PAGE_SIZE : + XE_PAGE_SIZE) { + entry = vm->pt_ops->pte_encode_bo(batch, i, + pat_index, 0); + + xe_map_wr(xe, &bo->vmap, map_ofs + level * 8, u64, + entry); + level++; + } + } } else { u64 batch_addr = xe_bo_addr(batch, 0, XE_PAGE_SIZE); @@ -344,7 +359,8 @@ struct xe_migrate *xe_migrate_init(struct xe_tile *tile) m->q = xe_exec_queue_create(xe, vm, logical_mask, 1, hwe, EXEC_QUEUE_FLAG_KERNEL | - EXEC_QUEUE_FLAG_PERMANENT); + EXEC_QUEUE_FLAG_PERMANENT | + EXEC_QUEUE_FLAG_HIGH_PRIORITY); } else { m->q = xe_exec_queue_create_class(xe, primary_gt, vm, XE_ENGINE_CLASS_COPY, @@ -355,8 +371,6 @@ struct xe_migrate *xe_migrate_init(struct xe_tile *tile) xe_vm_close_and_put(vm); return ERR_CAST(m->q); } - if (xe->info.has_usm) - m->q->priority = XE_EXEC_QUEUE_PRIORITY_KERNEL; mutex_init(&m->job_mutex); @@ -364,6 +378,19 @@ struct xe_migrate *xe_migrate_init(struct xe_tile *tile) if (err) return ERR_PTR(err); + if (IS_DGFX(xe)) { + if (xe_device_has_flat_ccs(xe)) + /* min chunk size corresponds to 4K of CCS Metadata */ + m->min_chunk_size = SZ_4K * SZ_64K / + xe_device_ccs_bytes(xe, SZ_64K); + else + /* Somewhat arbitrary to avoid a huge amount of blits */ + m->min_chunk_size = SZ_64K; + m->min_chunk_size = roundup_pow_of_two(m->min_chunk_size); + drm_dbg(&xe->drm, "Migrate min chunk size is 0x%08llx\n", + (unsigned long long)m->min_chunk_size); + } + return m; } @@ -375,16 +402,35 @@ static u64 max_mem_transfer_per_pass(struct xe_device *xe) return MAX_PREEMPTDISABLE_TRANSFER; } -static u64 xe_migrate_res_sizes(struct xe_device *xe, struct xe_res_cursor *cur) +static u64 xe_migrate_res_sizes(struct xe_migrate *m, struct xe_res_cursor *cur) { - /* - * For VRAM we use identity mapped pages so we are limited to current - * cursor size. For system we program the pages ourselves so we have no - * such limitation. - */ - return min_t(u64, max_mem_transfer_per_pass(xe), - mem_type_is_vram(cur->mem_type) ? cur->size : - cur->remaining); + struct xe_device *xe = tile_to_xe(m->tile); + u64 size = min_t(u64, max_mem_transfer_per_pass(xe), cur->remaining); + + if (mem_type_is_vram(cur->mem_type)) { + /* + * VRAM we want to blit in chunks with sizes aligned to + * min_chunk_size in order for the offset to CCS metadata to be + * page-aligned. If it's the last chunk it may be smaller. + * + * Another constraint is that we need to limit the blit to + * the VRAM block size, unless size is smaller than + * min_chunk_size. + */ + u64 chunk = max_t(u64, cur->size, m->min_chunk_size); + + size = min_t(u64, size, chunk); + if (size > m->min_chunk_size) + size = round_down(size, m->min_chunk_size); + } + + return size; +} + +static bool xe_migrate_allow_identity(u64 size, const struct xe_res_cursor *cur) +{ + /* If the chunk is not fragmented, allow identity map. */ + return cur->size >= size; } static u32 pte_update_size(struct xe_migrate *m, @@ -397,7 +443,12 @@ static u32 pte_update_size(struct xe_migrate *m, u32 cmds = 0; *L0_pt = pt_ofs; - if (!is_vram) { + if (is_vram && xe_migrate_allow_identity(*L0, cur)) { + /* Offset into identity map. */ + *L0_ofs = xe_migrate_vram_ofs(tile_to_xe(m->tile), + cur->start + vram_region_gpu_offset(res)); + cmds += cmd_size; + } else { /* Clip L0 to available size */ u64 size = min(*L0, (u64)avail_pts * SZ_2M); u64 num_4k_pages = DIV_ROUND_UP(size, XE_PAGE_SIZE); @@ -413,11 +464,6 @@ static u32 pte_update_size(struct xe_migrate *m, /* Each chunk has a single blit command */ cmds += cmd_size; - } else { - /* Offset into identity map. */ - *L0_ofs = xe_migrate_vram_ofs(tile_to_xe(m->tile), - cur->start + vram_region_gpu_offset(res)); - cmds += cmd_size; } return cmds; @@ -427,10 +473,10 @@ static void emit_pte(struct xe_migrate *m, struct xe_bb *bb, u32 at_pt, bool is_vram, bool is_comp_pte, struct xe_res_cursor *cur, - u32 size, struct xe_bo *bo) + u32 size, struct ttm_resource *res) { struct xe_device *xe = tile_to_xe(m->tile); - + struct xe_vm *vm = m->q->vm; u16 pat_index; u32 ptes; u64 ofs = at_pt * XE_PAGE_SIZE; @@ -439,17 +485,10 @@ static void emit_pte(struct xe_migrate *m, /* Indirect access needs compression enabled uncached PAT index */ if (GRAPHICS_VERx100(xe) >= 2000) pat_index = is_comp_pte ? xe->pat.idx[XE_CACHE_NONE_COMPRESSION] : - xe->pat.idx[XE_CACHE_NONE]; + xe->pat.idx[XE_CACHE_WB]; else pat_index = xe->pat.idx[XE_CACHE_WB]; - /* - * FIXME: Emitting VRAM PTEs to L0 PTs is forbidden. Currently - * we're only emitting VRAM PTEs during sanity tests, so when - * that's moved to a Kunit test, we should condition VRAM PTEs - * on running tests. - */ - ptes = DIV_ROUND_UP(size, XE_PAGE_SIZE); while (ptes) { @@ -469,20 +508,22 @@ static void emit_pte(struct xe_migrate *m, addr = xe_res_dma(cur) & PAGE_MASK; if (is_vram) { - /* Is this a 64K PTE entry? */ - if ((m->q->vm->flags & XE_VM_FLAG_64K) && - !(cur_ofs & (16 * 8 - 1))) { - xe_tile_assert(m->tile, IS_ALIGNED(addr, SZ_64K)); + if (vm->flags & XE_VM_FLAG_64K) { + u64 va = cur_ofs * XE_PAGE_SIZE / 8; + + xe_assert(xe, (va & (SZ_64K - 1)) == + (addr & (SZ_64K - 1))); + flags |= XE_PTE_PS64; } - addr += vram_region_gpu_offset(bo->ttm.resource); + addr += vram_region_gpu_offset(res); devmem = true; } - addr = m->q->vm->pt_ops->pte_encode_addr(m->tile->xe, - addr, pat_index, - 0, devmem, flags); + addr = vm->pt_ops->pte_encode_addr(m->tile->xe, + addr, pat_index, + 0, devmem, flags); bb->cs[bb->len++] = lower_32_bits(addr); bb->cs[bb->len++] = upper_32_bits(addr); @@ -694,8 +735,8 @@ struct dma_fence *xe_migrate_copy(struct xe_migrate *m, bool usm = xe->info.has_usm; u32 avail_pts = max_mem_transfer_per_pass(xe) / LEVEL0_PAGE_TABLE_ENCODE_SIZE; - src_L0 = xe_migrate_res_sizes(xe, &src_it); - dst_L0 = xe_migrate_res_sizes(xe, &dst_it); + src_L0 = xe_migrate_res_sizes(m, &src_it); + dst_L0 = xe_migrate_res_sizes(m, &dst_it); drm_dbg(&xe->drm, "Pass %u, sizes: %llu & %llu\n", pass++, src_L0, dst_L0); @@ -716,6 +757,7 @@ struct dma_fence *xe_migrate_copy(struct xe_migrate *m, &ccs_ofs, &ccs_pt, 0, 2 * avail_pts, avail_pts); + xe_assert(xe, IS_ALIGNED(ccs_it.start, PAGE_SIZE)); } /* Add copy commands size here */ @@ -728,20 +770,20 @@ struct dma_fence *xe_migrate_copy(struct xe_migrate *m, goto err_sync; } - if (!src_is_vram) - emit_pte(m, bb, src_L0_pt, src_is_vram, true, &src_it, src_L0, - src_bo); - else + if (src_is_vram && xe_migrate_allow_identity(src_L0, &src_it)) xe_res_next(&src_it, src_L0); - - if (!dst_is_vram) - emit_pte(m, bb, dst_L0_pt, dst_is_vram, true, &dst_it, src_L0, - dst_bo); else + emit_pte(m, bb, src_L0_pt, src_is_vram, copy_system_ccs, + &src_it, src_L0, src); + + if (dst_is_vram && xe_migrate_allow_identity(src_L0, &dst_it)) xe_res_next(&dst_it, src_L0); + else + emit_pte(m, bb, dst_L0_pt, dst_is_vram, copy_system_ccs, + &dst_it, src_L0, dst); if (copy_system_ccs) - emit_pte(m, bb, ccs_pt, false, false, &ccs_it, ccs_size, src_bo); + emit_pte(m, bb, ccs_pt, false, false, &ccs_it, ccs_size, src); bb->cs[bb->len++] = MI_BATCH_BUFFER_END; update_idx = bb->len; @@ -950,7 +992,7 @@ struct dma_fence *xe_migrate_clear(struct xe_migrate *m, bool usm = xe->info.has_usm; u32 avail_pts = max_mem_transfer_per_pass(xe) / LEVEL0_PAGE_TABLE_ENCODE_SIZE; - clear_L0 = xe_migrate_res_sizes(xe, &src_it); + clear_L0 = xe_migrate_res_sizes(m, &src_it); drm_dbg(&xe->drm, "Pass %u, size: %llu\n", pass++, clear_L0); @@ -977,12 +1019,12 @@ struct dma_fence *xe_migrate_clear(struct xe_migrate *m, size -= clear_L0; /* Preemption is enabled again by the ring ops. */ - if (!clear_vram) { - emit_pte(m, bb, clear_L0_pt, clear_vram, true, &src_it, clear_L0, - bo); - } else { + if (clear_vram && xe_migrate_allow_identity(clear_L0, &src_it)) xe_res_next(&src_it, clear_L0); - } + else + emit_pte(m, bb, clear_L0_pt, clear_vram, clear_system_ccs, + &src_it, clear_L0, dst); + bb->cs[bb->len++] = MI_BATCH_BUFFER_END; update_idx = bb->len; @@ -1175,8 +1217,11 @@ static bool no_in_syncs(struct xe_vm *vm, struct xe_exec_queue *q, } if (q) { fence = xe_exec_queue_last_fence_get(q, vm); - if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags)) + if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags)) { + dma_fence_put(fence); return false; + } + dma_fence_put(fence); } return true; diff --git a/drivers/gpu/drm/xe/xe_mmio.c b/drivers/gpu/drm/xe/xe_mmio.c index f660cfb79f50..5f6b53ea5528 100644 --- a/drivers/gpu/drm/xe/xe_mmio.c +++ b/drivers/gpu/drm/xe/xe_mmio.c @@ -272,8 +272,8 @@ int xe_mmio_probe_vram(struct xe_device *xe) drm_info(&xe->drm, "VRAM[%u, %u]: Actual physical size %pa, usable size exclude stolen %pa, CPU accessible size %pa\n", id, tile->id, &tile->mem.vram.actual_physical_size, &tile->mem.vram.usable_size, &tile->mem.vram.io_size); drm_info(&xe->drm, "VRAM[%u, %u]: DPA range: [%pa-%llx], io range: [%pa-%llx]\n", id, tile->id, - &tile->mem.vram.dpa_base, tile->mem.vram.dpa_base + tile->mem.vram.actual_physical_size, - &tile->mem.vram.io_start, tile->mem.vram.io_start + tile->mem.vram.io_size); + &tile->mem.vram.dpa_base, tile->mem.vram.dpa_base + (u64)tile->mem.vram.actual_physical_size, + &tile->mem.vram.io_start, tile->mem.vram.io_start + (u64)tile->mem.vram.io_size); /* calculate total size using tile size to get the correct HW sizing */ total_size += tile_size; @@ -303,7 +303,7 @@ void xe_mmio_probe_tiles(struct xe_device *xe) u8 id, tile_count = xe->info.tile_count; struct xe_gt *gt = xe_root_mmio_gt(xe); struct xe_tile *tile; - void *regs; + void __iomem *regs; u32 mtcfg; if (tile_count == 1) diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c index de1030a47588..e45b37c3f0c2 100644 --- a/drivers/gpu/drm/xe/xe_pt.c +++ b/drivers/gpu/drm/xe/xe_pt.c @@ -618,8 +618,8 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma, if (!xe_vma_is_null(vma)) { if (xe_vma_is_userptr(vma)) - xe_res_first_sg(vma->userptr.sg, 0, xe_vma_size(vma), - &curs); + xe_res_first_sg(to_userptr_vma(vma)->userptr.sg, 0, + xe_vma_size(vma), &curs); else if (xe_bo_is_vram(bo) || xe_bo_is_stolen(bo)) xe_res_first(bo->ttm.resource, xe_vma_bo_offset(vma), xe_vma_size(vma), &curs); @@ -906,17 +906,17 @@ static void xe_vm_dbg_print_entries(struct xe_device *xe, #ifdef CONFIG_DRM_XE_USERPTR_INVAL_INJECT -static int xe_pt_userptr_inject_eagain(struct xe_vma *vma) +static int xe_pt_userptr_inject_eagain(struct xe_userptr_vma *uvma) { - u32 divisor = vma->userptr.divisor ? vma->userptr.divisor : 2; + u32 divisor = uvma->userptr.divisor ? uvma->userptr.divisor : 2; static u32 count; if (count++ % divisor == divisor - 1) { - struct xe_vm *vm = xe_vma_vm(vma); + struct xe_vm *vm = xe_vma_vm(&uvma->vma); - vma->userptr.divisor = divisor << 1; + uvma->userptr.divisor = divisor << 1; spin_lock(&vm->userptr.invalidated_lock); - list_move_tail(&vma->userptr.invalidate_link, + list_move_tail(&uvma->userptr.invalidate_link, &vm->userptr.invalidated); spin_unlock(&vm->userptr.invalidated_lock); return true; @@ -927,7 +927,7 @@ static int xe_pt_userptr_inject_eagain(struct xe_vma *vma) #else -static bool xe_pt_userptr_inject_eagain(struct xe_vma *vma) +static bool xe_pt_userptr_inject_eagain(struct xe_userptr_vma *uvma) { return false; } @@ -1000,9 +1000,9 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) { struct xe_pt_migrate_pt_update *userptr_update = container_of(pt_update, typeof(*userptr_update), base); - struct xe_vma *vma = pt_update->vma; - unsigned long notifier_seq = vma->userptr.notifier_seq; - struct xe_vm *vm = xe_vma_vm(vma); + struct xe_userptr_vma *uvma = to_userptr_vma(pt_update->vma); + unsigned long notifier_seq = uvma->userptr.notifier_seq; + struct xe_vm *vm = xe_vma_vm(&uvma->vma); int err = xe_pt_vm_dependencies(pt_update->job, &vm->rftree[pt_update->tile_id], pt_update->start, @@ -1023,7 +1023,7 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) */ do { down_read(&vm->userptr.notifier_lock); - if (!mmu_interval_read_retry(&vma->userptr.notifier, + if (!mmu_interval_read_retry(&uvma->userptr.notifier, notifier_seq)) break; @@ -1032,11 +1032,11 @@ static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update) if (userptr_update->bind) return -EAGAIN; - notifier_seq = mmu_interval_read_begin(&vma->userptr.notifier); + notifier_seq = mmu_interval_read_begin(&uvma->userptr.notifier); } while (true); /* Inject errors to test_whether they are handled correctly */ - if (userptr_update->bind && xe_pt_userptr_inject_eagain(vma)) { + if (userptr_update->bind && xe_pt_userptr_inject_eagain(uvma)) { up_read(&vm->userptr.notifier_lock); return -EAGAIN; } @@ -1297,7 +1297,7 @@ __xe_pt_bind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue vma->tile_present |= BIT(tile->id); if (bind_pt_update.locked) { - vma->userptr.initial_bind = true; + to_userptr_vma(vma)->userptr.initial_bind = true; up_read(&vm->userptr.notifier_lock); xe_bo_put_commit(&deferred); } @@ -1642,7 +1642,7 @@ __xe_pt_unbind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queu if (!vma->tile_present) { spin_lock(&vm->userptr.invalidated_lock); - list_del_init(&vma->userptr.invalidate_link); + list_del_init(&to_userptr_vma(vma)->userptr.invalidate_link); spin_unlock(&vm->userptr.invalidated_lock); } up_read(&vm->userptr.notifier_lock); diff --git a/drivers/gpu/drm/xe/xe_query.c b/drivers/gpu/drm/xe/xe_query.c index 9b35673b286c..7e924faeeea0 100644 --- a/drivers/gpu/drm/xe/xe_query.c +++ b/drivers/gpu/drm/xe/xe_query.c @@ -459,21 +459,21 @@ static size_t calc_topo_query_size(struct xe_device *xe) sizeof_field(struct xe_gt, fuse_topo.eu_mask_per_dss)); } -static void __user *copy_mask(void __user *ptr, - struct drm_xe_query_topology_mask *topo, - void *mask, size_t mask_size) +static int copy_mask(void __user **ptr, + struct drm_xe_query_topology_mask *topo, + void *mask, size_t mask_size) { topo->num_bytes = mask_size; - if (copy_to_user(ptr, topo, sizeof(*topo))) - return ERR_PTR(-EFAULT); - ptr += sizeof(topo); + if (copy_to_user(*ptr, topo, sizeof(*topo))) + return -EFAULT; + *ptr += sizeof(topo); - if (copy_to_user(ptr, mask, mask_size)) - return ERR_PTR(-EFAULT); - ptr += mask_size; + if (copy_to_user(*ptr, mask, mask_size)) + return -EFAULT; + *ptr += mask_size; - return ptr; + return 0; } static int query_gt_topology(struct xe_device *xe, @@ -493,28 +493,28 @@ static int query_gt_topology(struct xe_device *xe, } for_each_gt(gt, xe, id) { + int err; + topo.gt_id = id; topo.type = DRM_XE_TOPO_DSS_GEOMETRY; - query_ptr = copy_mask(query_ptr, &topo, - gt->fuse_topo.g_dss_mask, - sizeof(gt->fuse_topo.g_dss_mask)); - if (IS_ERR(query_ptr)) - return PTR_ERR(query_ptr); + err = copy_mask(&query_ptr, &topo, gt->fuse_topo.g_dss_mask, + sizeof(gt->fuse_topo.g_dss_mask)); + if (err) + return err; topo.type = DRM_XE_TOPO_DSS_COMPUTE; - query_ptr = copy_mask(query_ptr, &topo, - gt->fuse_topo.c_dss_mask, - sizeof(gt->fuse_topo.c_dss_mask)); - if (IS_ERR(query_ptr)) - return PTR_ERR(query_ptr); + err = copy_mask(&query_ptr, &topo, gt->fuse_topo.c_dss_mask, + sizeof(gt->fuse_topo.c_dss_mask)); + if (err) + return err; topo.type = DRM_XE_TOPO_EU_PER_DSS; - query_ptr = copy_mask(query_ptr, &topo, - gt->fuse_topo.eu_mask_per_dss, - sizeof(gt->fuse_topo.eu_mask_per_dss)); - if (IS_ERR(query_ptr)) - return PTR_ERR(query_ptr); + err = copy_mask(&query_ptr, &topo, + gt->fuse_topo.eu_mask_per_dss, + sizeof(gt->fuse_topo.eu_mask_per_dss)); + if (err) + return err; } return 0; diff --git a/drivers/gpu/drm/xe/xe_sched_job.c b/drivers/gpu/drm/xe/xe_sched_job.c index 01106a1156ad..4e2ccad0e52f 100644 --- a/drivers/gpu/drm/xe/xe_sched_job.c +++ b/drivers/gpu/drm/xe/xe_sched_job.c @@ -274,7 +274,6 @@ int xe_sched_job_last_fence_add_dep(struct xe_sched_job *job, struct xe_vm *vm) struct dma_fence *fence; fence = xe_exec_queue_last_fence_get(job->q, vm); - dma_fence_get(fence); return drm_sched_job_add_dependency(&job->drm, fence); } diff --git a/drivers/gpu/drm/xe/xe_sync.c b/drivers/gpu/drm/xe/xe_sync.c index e4c220cf9115..aab92bee1d7c 100644 --- a/drivers/gpu/drm/xe/xe_sync.c +++ b/drivers/gpu/drm/xe/xe_sync.c @@ -307,7 +307,6 @@ xe_sync_in_fence_get(struct xe_sync_entry *sync, int num_sync, /* Easy case... */ if (!num_in_fence) { fence = xe_exec_queue_last_fence_get(q, vm); - dma_fence_get(fence); return fence; } @@ -322,7 +321,6 @@ xe_sync_in_fence_get(struct xe_sync_entry *sync, int num_sync, } } fences[current_fence++] = xe_exec_queue_last_fence_get(q, vm); - dma_fence_get(fences[current_fence - 1]); cf = dma_fence_array_create(num_in_fence, fences, vm->composite_fence_ctx, vm->composite_fence_seqno++, diff --git a/drivers/gpu/drm/xe/xe_sync.h b/drivers/gpu/drm/xe/xe_sync.h index d284afbe917c..f43cdcaca6c5 100644 --- a/drivers/gpu/drm/xe/xe_sync.h +++ b/drivers/gpu/drm/xe/xe_sync.h @@ -33,4 +33,9 @@ struct dma_fence * xe_sync_in_fence_get(struct xe_sync_entry *sync, int num_sync, struct xe_exec_queue *q, struct xe_vm *vm); +static inline bool xe_sync_is_ufence(struct xe_sync_entry *sync) +{ + return !!sync->ufence; +} + #endif diff --git a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c index d2b00d0bf1e2..e5d7d5e2bec1 100644 --- a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c +++ b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c @@ -31,7 +31,7 @@ struct xe_ttm_stolen_mgr { /* GPU base offset */ resource_size_t stolen_base; - void *__iomem mapping; + void __iomem *mapping; }; static inline struct xe_ttm_stolen_mgr * @@ -275,7 +275,7 @@ static int __xe_ttm_stolen_io_mem_reserve_bar2(struct xe_device *xe, drm_WARN_ON(&xe->drm, !(mem->placement & TTM_PL_FLAG_CONTIGUOUS)); if (mem->placement & TTM_PL_FLAG_CONTIGUOUS && mgr->mapping) - mem->bus.addr = (u8 *)mgr->mapping + mem->bus.offset; + mem->bus.addr = (u8 __force *)mgr->mapping + mem->bus.offset; mem->bus.offset += mgr->io_base; mem->bus.is_iomem = true; diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index 0cfe7289b97e..865e10d0a06a 100644 --- a/drivers/gpu/drm/xe/xe_vm.c +++ b/drivers/gpu/drm/xe/xe_vm.c @@ -37,8 +37,6 @@ #include "generated/xe_wa_oob.h" #include "xe_wa.h" -#define TEST_VM_ASYNC_OPS_ERROR - static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm) { return vm->gpuvm.r_obj; @@ -46,7 +44,7 @@ static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm) /** * xe_vma_userptr_check_repin() - Advisory check for repin needed - * @vma: The userptr vma + * @uvma: The userptr vma * * Check if the userptr vma has been invalidated since last successful * repin. The check is advisory only and can the function can be called @@ -56,15 +54,17 @@ static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm) * * Return: 0 if userptr vma is valid, -EAGAIN otherwise; repin recommended. */ -int xe_vma_userptr_check_repin(struct xe_vma *vma) +int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma) { - return mmu_interval_check_retry(&vma->userptr.notifier, - vma->userptr.notifier_seq) ? + return mmu_interval_check_retry(&uvma->userptr.notifier, + uvma->userptr.notifier_seq) ? -EAGAIN : 0; } -int xe_vma_userptr_pin_pages(struct xe_vma *vma) +int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma) { + struct xe_userptr *userptr = &uvma->userptr; + struct xe_vma *vma = &uvma->vma; struct xe_vm *vm = xe_vma_vm(vma); struct xe_device *xe = vm->xe; const unsigned long num_pages = xe_vma_size(vma) >> PAGE_SHIFT; @@ -80,30 +80,30 @@ retry: if (vma->gpuva.flags & XE_VMA_DESTROYED) return 0; - notifier_seq = mmu_interval_read_begin(&vma->userptr.notifier); - if (notifier_seq == vma->userptr.notifier_seq) + notifier_seq = mmu_interval_read_begin(&userptr->notifier); + if (notifier_seq == userptr->notifier_seq) return 0; pages = kvmalloc_array(num_pages, sizeof(*pages), GFP_KERNEL); if (!pages) return -ENOMEM; - if (vma->userptr.sg) { + if (userptr->sg) { dma_unmap_sgtable(xe->drm.dev, - vma->userptr.sg, + userptr->sg, read_only ? DMA_TO_DEVICE : DMA_BIDIRECTIONAL, 0); - sg_free_table(vma->userptr.sg); - vma->userptr.sg = NULL; + sg_free_table(userptr->sg); + userptr->sg = NULL; } pinned = ret = 0; if (in_kthread) { - if (!mmget_not_zero(vma->userptr.notifier.mm)) { + if (!mmget_not_zero(userptr->notifier.mm)) { ret = -EFAULT; goto mm_closed; } - kthread_use_mm(vma->userptr.notifier.mm); + kthread_use_mm(userptr->notifier.mm); } while (pinned < num_pages) { @@ -112,43 +112,40 @@ retry: num_pages - pinned, read_only ? 0 : FOLL_WRITE, &pages[pinned]); - if (ret < 0) { - if (in_kthread) - ret = 0; + if (ret < 0) break; - } pinned += ret; ret = 0; } if (in_kthread) { - kthread_unuse_mm(vma->userptr.notifier.mm); - mmput(vma->userptr.notifier.mm); + kthread_unuse_mm(userptr->notifier.mm); + mmput(userptr->notifier.mm); } mm_closed: if (ret) goto out; - ret = sg_alloc_table_from_pages_segment(&vma->userptr.sgt, pages, + ret = sg_alloc_table_from_pages_segment(&userptr->sgt, pages, pinned, 0, (u64)pinned << PAGE_SHIFT, xe_sg_segment_size(xe->drm.dev), GFP_KERNEL); if (ret) { - vma->userptr.sg = NULL; + userptr->sg = NULL; goto out; } - vma->userptr.sg = &vma->userptr.sgt; + userptr->sg = &userptr->sgt; - ret = dma_map_sgtable(xe->drm.dev, vma->userptr.sg, + ret = dma_map_sgtable(xe->drm.dev, userptr->sg, read_only ? DMA_TO_DEVICE : DMA_BIDIRECTIONAL, DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_NO_KERNEL_MAPPING); if (ret) { - sg_free_table(vma->userptr.sg); - vma->userptr.sg = NULL; + sg_free_table(userptr->sg); + userptr->sg = NULL; goto out; } @@ -167,8 +164,8 @@ out: kvfree(pages); if (!(ret < 0)) { - vma->userptr.notifier_seq = notifier_seq; - if (xe_vma_userptr_check_repin(vma) == -EAGAIN) + userptr->notifier_seq = notifier_seq; + if (xe_vma_userptr_check_repin(uvma) == -EAGAIN) goto retry; } @@ -335,13 +332,13 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q) down_write(&vm->lock); err = drm_gpuvm_exec_lock(&vm_exec); if (err) - return err; + goto out_up_write; pfence = xe_preempt_fence_create(q, q->compute.context, ++q->compute.seqno); if (!pfence) { err = -ENOMEM; - goto out_unlock; + goto out_fini; } list_add(&q->compute.link, &vm->preempt.exec_queues); @@ -364,8 +361,9 @@ int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q) up_read(&vm->userptr.notifier_lock); -out_unlock: +out_fini: drm_exec_fini(exec); +out_up_write: up_write(&vm->lock); return err; @@ -634,7 +632,9 @@ static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni, const struct mmu_notifier_range *range, unsigned long cur_seq) { - struct xe_vma *vma = container_of(mni, struct xe_vma, userptr.notifier); + struct xe_userptr *userptr = container_of(mni, typeof(*userptr), notifier); + struct xe_userptr_vma *uvma = container_of(userptr, typeof(*uvma), userptr); + struct xe_vma *vma = &uvma->vma; struct xe_vm *vm = xe_vma_vm(vma); struct dma_resv_iter cursor; struct dma_fence *fence; @@ -650,7 +650,7 @@ static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni, mmu_interval_set_seq(mni, cur_seq); /* No need to stop gpu access if the userptr is not yet bound. */ - if (!vma->userptr.initial_bind) { + if (!userptr->initial_bind) { up_write(&vm->userptr.notifier_lock); return true; } @@ -662,7 +662,7 @@ static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni, if (!xe_vm_in_fault_mode(vm) && !(vma->gpuva.flags & XE_VMA_DESTROYED) && vma->tile_present) { spin_lock(&vm->userptr.invalidated_lock); - list_move_tail(&vma->userptr.invalidate_link, + list_move_tail(&userptr->invalidate_link, &vm->userptr.invalidated); spin_unlock(&vm->userptr.invalidated_lock); } @@ -702,7 +702,7 @@ static const struct mmu_interval_notifier_ops vma_userptr_notifier_ops = { int xe_vm_userptr_pin(struct xe_vm *vm) { - struct xe_vma *vma, *next; + struct xe_userptr_vma *uvma, *next; int err = 0; LIST_HEAD(tmp_evict); @@ -710,22 +710,23 @@ int xe_vm_userptr_pin(struct xe_vm *vm) /* Collect invalidated userptrs */ spin_lock(&vm->userptr.invalidated_lock); - list_for_each_entry_safe(vma, next, &vm->userptr.invalidated, + list_for_each_entry_safe(uvma, next, &vm->userptr.invalidated, userptr.invalidate_link) { - list_del_init(&vma->userptr.invalidate_link); - list_move_tail(&vma->combined_links.userptr, + list_del_init(&uvma->userptr.invalidate_link); + list_move_tail(&uvma->userptr.repin_link, &vm->userptr.repin_list); } spin_unlock(&vm->userptr.invalidated_lock); /* Pin and move to temporary list */ - list_for_each_entry_safe(vma, next, &vm->userptr.repin_list, - combined_links.userptr) { - err = xe_vma_userptr_pin_pages(vma); + list_for_each_entry_safe(uvma, next, &vm->userptr.repin_list, + userptr.repin_link) { + err = xe_vma_userptr_pin_pages(uvma); if (err < 0) return err; - list_move_tail(&vma->combined_links.userptr, &vm->rebind_list); + list_del_init(&uvma->userptr.repin_link); + list_move_tail(&uvma->vma.combined_links.rebind, &vm->rebind_list); } return 0; @@ -781,6 +782,14 @@ struct dma_fence *xe_vm_rebind(struct xe_vm *vm, bool rebind_worker) return fence; } +static void xe_vma_free(struct xe_vma *vma) +{ + if (xe_vma_is_userptr(vma)) + kfree(to_userptr_vma(vma)); + else + kfree(vma); +} + #define VMA_CREATE_FLAG_READ_ONLY BIT(0) #define VMA_CREATE_FLAG_IS_NULL BIT(1) @@ -799,14 +808,26 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm, xe_assert(vm->xe, start < end); xe_assert(vm->xe, end < vm->size); - if (!bo && !is_null) /* userptr */ + /* + * Allocate and ensure that the xe_vma_is_userptr() return + * matches what was allocated. + */ + if (!bo && !is_null) { + struct xe_userptr_vma *uvma = kzalloc(sizeof(*uvma), GFP_KERNEL); + + if (!uvma) + return ERR_PTR(-ENOMEM); + + vma = &uvma->vma; + } else { vma = kzalloc(sizeof(*vma), GFP_KERNEL); - else - vma = kzalloc(sizeof(*vma) - sizeof(struct xe_userptr), - GFP_KERNEL); - if (!vma) { - vma = ERR_PTR(-ENOMEM); - return vma; + if (!vma) + return ERR_PTR(-ENOMEM); + + if (is_null) + vma->gpuva.flags |= DRM_GPUVA_SPARSE; + if (bo) + vma->gpuva.gem.obj = &bo->ttm.base; } INIT_LIST_HEAD(&vma->combined_links.rebind); @@ -817,8 +838,6 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm, vma->gpuva.va.range = end - start + 1; if (read_only) vma->gpuva.flags |= XE_VMA_READ_ONLY; - if (is_null) - vma->gpuva.flags |= DRM_GPUVA_SPARSE; for_each_tile(tile, vm->xe, id) vma->tile_mask |= 0x1 << id; @@ -835,35 +854,35 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm, vm_bo = drm_gpuvm_bo_obtain(vma->gpuva.vm, &bo->ttm.base); if (IS_ERR(vm_bo)) { - kfree(vma); + xe_vma_free(vma); return ERR_CAST(vm_bo); } drm_gpuvm_bo_extobj_add(vm_bo); drm_gem_object_get(&bo->ttm.base); - vma->gpuva.gem.obj = &bo->ttm.base; vma->gpuva.gem.offset = bo_offset_or_userptr; drm_gpuva_link(&vma->gpuva, vm_bo); drm_gpuvm_bo_put(vm_bo); } else /* userptr or null */ { if (!is_null) { + struct xe_userptr *userptr = &to_userptr_vma(vma)->userptr; u64 size = end - start + 1; int err; - INIT_LIST_HEAD(&vma->userptr.invalidate_link); + INIT_LIST_HEAD(&userptr->invalidate_link); + INIT_LIST_HEAD(&userptr->repin_link); vma->gpuva.gem.offset = bo_offset_or_userptr; - err = mmu_interval_notifier_insert(&vma->userptr.notifier, + err = mmu_interval_notifier_insert(&userptr->notifier, current->mm, xe_vma_userptr(vma), size, &vma_userptr_notifier_ops); if (err) { - kfree(vma); - vma = ERR_PTR(err); - return vma; + xe_vma_free(vma); + return ERR_PTR(err); } - vma->userptr.notifier_seq = LONG_MAX; + userptr->notifier_seq = LONG_MAX; } xe_vm_get(vm); @@ -879,13 +898,15 @@ static void xe_vma_destroy_late(struct xe_vma *vma) bool read_only = xe_vma_read_only(vma); if (xe_vma_is_userptr(vma)) { - if (vma->userptr.sg) { + struct xe_userptr *userptr = &to_userptr_vma(vma)->userptr; + + if (userptr->sg) { dma_unmap_sgtable(xe->drm.dev, - vma->userptr.sg, + userptr->sg, read_only ? DMA_TO_DEVICE : DMA_BIDIRECTIONAL, 0); - sg_free_table(vma->userptr.sg); - vma->userptr.sg = NULL; + sg_free_table(userptr->sg); + userptr->sg = NULL; } /* @@ -893,7 +914,7 @@ static void xe_vma_destroy_late(struct xe_vma *vma) * the notifer until we're sure the GPU is not accessing * them anymore */ - mmu_interval_notifier_remove(&vma->userptr.notifier); + mmu_interval_notifier_remove(&userptr->notifier); xe_vm_put(vm); } else if (xe_vma_is_null(vma)) { xe_vm_put(vm); @@ -901,7 +922,7 @@ static void xe_vma_destroy_late(struct xe_vma *vma) xe_bo_put(xe_vma_bo(vma)); } - kfree(vma); + xe_vma_free(vma); } static void vma_destroy_work_func(struct work_struct *w) @@ -932,7 +953,7 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence) xe_assert(vm->xe, vma->gpuva.flags & XE_VMA_DESTROYED); spin_lock(&vm->userptr.invalidated_lock); - list_del(&vma->userptr.invalidate_link); + list_del(&to_userptr_vma(vma)->userptr.invalidate_link); spin_unlock(&vm->userptr.invalidated_lock); } else if (!xe_vma_is_null(vma)) { xe_bo_assert_held(xe_vma_bo(vma)); @@ -1854,10 +1875,8 @@ int xe_vm_create_ioctl(struct drm_device *dev, void *data, mutex_lock(&xef->vm.lock); err = xa_alloc(&xef->vm.xa, &id, vm, xa_limit_32b, GFP_KERNEL); mutex_unlock(&xef->vm.lock); - if (err) { - xe_vm_close_and_put(vm); - return err; - } + if (err) + goto err_close_and_put; if (xe->info.has_asid) { mutex_lock(&xe->usm.lock); @@ -1865,11 +1884,9 @@ int xe_vm_create_ioctl(struct drm_device *dev, void *data, XA_LIMIT(1, XE_MAX_ASID - 1), &xe->usm.next_asid, GFP_KERNEL); mutex_unlock(&xe->usm.lock); - if (err < 0) { - xe_vm_close_and_put(vm); - return err; - } - err = 0; + if (err < 0) + goto err_free_id; + vm->usm.asid = asid; } @@ -1887,6 +1904,15 @@ int xe_vm_create_ioctl(struct drm_device *dev, void *data, #endif return 0; + +err_free_id: + mutex_lock(&xef->vm.lock); + xa_erase(&xef->vm.xa, id); + mutex_unlock(&xef->vm.lock); +err_close_and_put: + xe_vm_close_and_put(vm); + + return err; } int xe_vm_destroy_ioctl(struct drm_device *dev, void *data, @@ -1953,6 +1979,7 @@ static int xe_vm_prefetch(struct xe_vm *vm, struct xe_vma *vma, xe_exec_queue_last_fence_get(wait_exec_queue, vm); xe_sync_entry_signal(&syncs[i], NULL, fence); + dma_fence_put(fence); } } @@ -2033,7 +2060,6 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, struct drm_gem_object *obj = bo ? &bo->ttm.base : NULL; struct drm_gpuva_ops *ops; struct drm_gpuva_op *__op; - struct xe_vma_op *op; struct drm_gpuvm_bo *vm_bo; int err; @@ -2063,9 +2089,11 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, if (err) return ERR_PTR(err); - vm_bo = drm_gpuvm_bo_find(&vm->gpuvm, obj); - if (!vm_bo) - break; + vm_bo = drm_gpuvm_bo_obtain(&vm->gpuvm, obj); + if (IS_ERR(vm_bo)) { + xe_bo_unlock(bo); + return ERR_CAST(vm_bo); + } ops = drm_gpuvm_bo_unmap_ops_create(vm_bo); drm_gpuvm_bo_put(vm_bo); @@ -2078,15 +2106,6 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo, if (IS_ERR(ops)) return ops; -#ifdef TEST_VM_ASYNC_OPS_ERROR - if (operation & FORCE_ASYNC_OP_ERROR) { - op = list_first_entry_or_null(&ops->list, struct xe_vma_op, - base.entry); - if (op) - op->inject_error = true; - } -#endif - drm_gpuva_for_each_op(__op, ops) { struct xe_vma_op *op = gpuva_op_to_vma_op(__op); @@ -2142,7 +2161,7 @@ static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op, drm_exec_fini(&exec); if (xe_vma_is_userptr(vma)) { - err = xe_vma_userptr_pin_pages(vma); + err = xe_vma_userptr_pin_pages(to_userptr_vma(vma)); if (err) { prep_vma_destroy(vm, vma, false); xe_vma_destroy_unlocked(vma); @@ -2166,8 +2185,10 @@ static u64 xe_vma_max_pte_size(struct xe_vma *vma) return SZ_1G; else if (vma->gpuva.flags & XE_VMA_PTE_2M) return SZ_2M; + else if (vma->gpuva.flags & XE_VMA_PTE_4K) + return SZ_4K; - return SZ_4K; + return SZ_1G; /* Uninitialized, used max size */ } static u64 xe_vma_set_pte_size(struct xe_vma *vma, u64 size) @@ -2497,13 +2518,25 @@ retry_userptr: } drm_exec_fini(&exec); - if (err == -EAGAIN && xe_vma_is_userptr(vma)) { + if (err == -EAGAIN) { lockdep_assert_held_write(&vm->lock); - err = xe_vma_userptr_pin_pages(vma); - if (!err) - goto retry_userptr; - trace_xe_vma_fail(vma); + if (op->base.op == DRM_GPUVA_OP_REMAP) { + if (!op->remap.unmap_done) + vma = gpuva_to_vma(op->base.remap.unmap->va); + else if (op->remap.prev) + vma = op->remap.prev; + else + vma = op->remap.next; + } + + if (xe_vma_is_userptr(vma)) { + err = xe_vma_userptr_pin_pages(to_userptr_vma(vma)); + if (!err) + goto retry_userptr; + + trace_xe_vma_fail(vma); + } } return err; @@ -2515,13 +2548,6 @@ static int xe_vma_op_execute(struct xe_vm *vm, struct xe_vma_op *op) lockdep_assert_held_write(&vm->lock); -#ifdef TEST_VM_ASYNC_OPS_ERROR - if (op->inject_error) { - op->inject_error = false; - return -ENOMEM; - } -#endif - switch (op->base.op) { case DRM_GPUVA_OP_MAP: ret = __xe_vma_op_execute(vm, op->map.vma, op); @@ -2636,7 +2662,7 @@ static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm, { int i; - for (i = num_ops_list - 1; i; ++i) { + for (i = num_ops_list - 1; i >= 0; --i) { struct drm_gpuva_ops *__ops = ops[i]; struct drm_gpuva_op *__op; @@ -2681,16 +2707,9 @@ static int vm_bind_ioctl_ops_execute(struct xe_vm *vm, return 0; } -#ifdef TEST_VM_ASYNC_OPS_ERROR -#define SUPPORTED_FLAGS \ - (FORCE_ASYNC_OP_ERROR | DRM_XE_VM_BIND_FLAG_READONLY | \ - DRM_XE_VM_BIND_FLAG_IMMEDIATE | DRM_XE_VM_BIND_FLAG_NULL | 0xffff) -#else #define SUPPORTED_FLAGS \ (DRM_XE_VM_BIND_FLAG_READONLY | \ - DRM_XE_VM_BIND_FLAG_IMMEDIATE | DRM_XE_VM_BIND_FLAG_NULL | \ - 0xffff) -#endif + DRM_XE_VM_BIND_FLAG_IMMEDIATE | DRM_XE_VM_BIND_FLAG_NULL) #define XE_64K_PAGE_MASK 0xffffull #define ALL_DRM_XE_SYNCS_FLAGS (DRM_XE_SYNCS_FLAG_WAIT_FOR_OP) @@ -2843,7 +2862,7 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) struct drm_gpuva_ops **ops = NULL; struct xe_vm *vm; struct xe_exec_queue *q = NULL; - u32 num_syncs; + u32 num_syncs, num_ufence = 0; struct xe_sync_entry *syncs = NULL; struct drm_xe_vm_bind_op *bind_ops; LIST_HEAD(ops_list); @@ -2980,6 +2999,14 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) SYNC_PARSE_FLAG_DISALLOW_USER_FENCE : 0)); if (err) goto free_syncs; + + if (xe_sync_is_ufence(&syncs[num_syncs])) + num_ufence++; + } + + if (XE_IOCTL_DBG(xe, num_ufence > 1)) { + err = -EINVAL; + goto free_syncs; } if (!args->num_binds) { @@ -3122,8 +3149,8 @@ int xe_vm_invalidate_vma(struct xe_vma *vma) if (IS_ENABLED(CONFIG_PROVE_LOCKING)) { if (xe_vma_is_userptr(vma)) { WARN_ON_ONCE(!mmu_interval_check_retry - (&vma->userptr.notifier, - vma->userptr.notifier_seq)); + (&to_userptr_vma(vma)->userptr.notifier, + to_userptr_vma(vma)->userptr.notifier_seq)); WARN_ON_ONCE(!dma_resv_test_signaled(xe_vm_resv(xe_vma_vm(vma)), DMA_RESV_USAGE_BOOKKEEP)); @@ -3184,11 +3211,11 @@ int xe_analyze_vm(struct drm_printer *p, struct xe_vm *vm, int gt_id) if (is_null) { addr = 0; } else if (is_userptr) { + struct sg_table *sg = to_userptr_vma(vma)->userptr.sg; struct xe_res_cursor cur; - if (vma->userptr.sg) { - xe_res_first_sg(vma->userptr.sg, 0, XE_PAGE_SIZE, - &cur); + if (sg) { + xe_res_first_sg(sg, 0, XE_PAGE_SIZE, &cur); addr = xe_res_dma(&cur); } else { addr = 0; diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h index cf2f96e8c1ab..9654a0612fc2 100644 --- a/drivers/gpu/drm/xe/xe_vm.h +++ b/drivers/gpu/drm/xe/xe_vm.h @@ -160,6 +160,18 @@ static inline bool xe_vma_is_userptr(struct xe_vma *vma) return xe_vma_has_no_bo(vma) && !xe_vma_is_null(vma); } +/** + * to_userptr_vma() - Return a pointer to an embedding userptr vma + * @vma: Pointer to the embedded struct xe_vma + * + * Return: Pointer to the embedding userptr vma + */ +static inline struct xe_userptr_vma *to_userptr_vma(struct xe_vma *vma) +{ + xe_assert(xe_vma_vm(vma)->xe, xe_vma_is_userptr(vma)); + return container_of(vma, struct xe_userptr_vma, vma); +} + u64 xe_vm_pdp4_descriptor(struct xe_vm *vm, struct xe_tile *tile); int xe_vm_create_ioctl(struct drm_device *dev, void *data, @@ -224,9 +236,9 @@ static inline void xe_vm_reactivate_rebind(struct xe_vm *vm) } } -int xe_vma_userptr_pin_pages(struct xe_vma *vma); +int xe_vma_userptr_pin_pages(struct xe_userptr_vma *uvma); -int xe_vma_userptr_check_repin(struct xe_vma *vma); +int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma); bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t *end); diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h index 63e8a50b88e9..5ac9c5bebabc 100644 --- a/drivers/gpu/drm/xe/xe_vm_types.h +++ b/drivers/gpu/drm/xe/xe_vm_types.h @@ -21,9 +21,6 @@ struct xe_bo; struct xe_sync_entry; struct xe_vm; -#define TEST_VM_ASYNC_OPS_ERROR -#define FORCE_ASYNC_OP_ERROR BIT(31) - #define XE_VMA_READ_ONLY DRM_GPUVA_USERBITS #define XE_VMA_DESTROYED (DRM_GPUVA_USERBITS << 1) #define XE_VMA_ATOMIC_PTE_BIT (DRM_GPUVA_USERBITS << 2) @@ -37,6 +34,8 @@ struct xe_vm; struct xe_userptr { /** @invalidate_link: Link for the vm::userptr.invalidated list */ struct list_head invalidate_link; + /** @userptr: link into VM repin list if userptr. */ + struct list_head repin_link; /** * @notifier: MMU notifier for user pointer (invalidation call back) */ @@ -68,8 +67,6 @@ struct xe_vma { * resv. */ union { - /** @userptr: link into VM repin list if userptr. */ - struct list_head userptr; /** @rebind: link into VM if this VMA needs rebinding. */ struct list_head rebind; /** @destroy: link to contested list when VM is being closed. */ @@ -105,11 +102,15 @@ struct xe_vma { * @pat_index: The pat index to use when encoding the PTEs for this vma. */ u16 pat_index; +}; - /** - * @userptr: user pointer state, only allocated for VMAs that are - * user pointers - */ +/** + * struct xe_userptr_vma - A userptr vma subclass + * @vma: The vma. + * @userptr: Additional userptr information. + */ +struct xe_userptr_vma { + struct xe_vma vma; struct xe_userptr userptr; }; @@ -356,11 +357,6 @@ struct xe_vma_op { /** @flags: operation flags */ enum xe_vma_op_flags flags; -#ifdef TEST_VM_ASYNC_OPS_ERROR - /** @inject_error: inject error to test async op error handling */ - bool inject_error; -#endif - union { /** @map: VMA map operation specific data */ struct xe_vma_op_map map; |