From: Thomas Hellström Date: Mon, 8 Sep 2025 10:12:44 +0000 (+0200) Subject: drm/xe: Convert xe_bo_create_pin_map() for exhaustive eviction X-Git-Url: https://www.infradead.org/git/?a=commitdiff_plain;h=59eabff2a3524d7f3bf6c24890cd197c4c116fcb;p=users%2Fhch%2Fmisc.git drm/xe: Convert xe_bo_create_pin_map() for exhaustive eviction Introduce an xe_bo_create_pin_map_novm() function that does not take the drm_exec paramenter to simplify the conversion of many callsites. For the rest, ensure that the same drm_exec context that was used for locking the vm is passed down to validation. Use xe_validation_guard() where appropriate. v2: - Avoid gotos from within xe_validation_guard(). (Matt Brost) - Break out the change to pf_provision_vf_lmem8 to a separate patch. - Adapt to signature change of xe_validation_guard(). Signed-off-by: Thomas Hellström Reviewed-by: Matthew Brost Link: https://lore.kernel.org/r/20250908101246.65025-12-thomas.hellstrom@linux.intel.com --- diff --git a/drivers/gpu/drm/xe/display/intel_fbdev_fb.c b/drivers/gpu/drm/xe/display/intel_fbdev_fb.c index d96ba2b51065..8ea9a472113c 100644 --- a/drivers/gpu/drm/xe/display/intel_fbdev_fb.c +++ b/drivers/gpu/drm/xe/display/intel_fbdev_fb.c @@ -42,11 +42,11 @@ struct intel_framebuffer *intel_fbdev_fb_alloc(struct drm_fb_helper *helper, obj = ERR_PTR(-ENODEV); if (!IS_DGFX(xe) && !XE_GT_WA(xe_root_mmio_gt(xe), 22019338487_display)) { - obj = xe_bo_create_pin_map(xe, xe_device_get_root_tile(xe), - NULL, size, - ttm_bo_type_kernel, XE_BO_FLAG_SCANOUT | - XE_BO_FLAG_STOLEN | - XE_BO_FLAG_GGTT); + obj = xe_bo_create_pin_map_novm(xe, xe_device_get_root_tile(xe), + size, + ttm_bo_type_kernel, XE_BO_FLAG_SCANOUT | + XE_BO_FLAG_STOLEN | + XE_BO_FLAG_GGTT, false); if (!IS_ERR(obj)) drm_info(&xe->drm, "Allocated fbdev into stolen\n"); else @@ -54,10 +54,10 @@ struct intel_framebuffer *intel_fbdev_fb_alloc(struct drm_fb_helper *helper, } if (IS_ERR(obj)) { - obj = xe_bo_create_pin_map(xe, xe_device_get_root_tile(xe), NULL, size, - ttm_bo_type_kernel, XE_BO_FLAG_SCANOUT | - XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(xe)) | - XE_BO_FLAG_GGTT); + obj = xe_bo_create_pin_map_novm(xe, xe_device_get_root_tile(xe), size, + ttm_bo_type_kernel, XE_BO_FLAG_SCANOUT | + XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(xe)) | + XE_BO_FLAG_GGTT, false); } if (IS_ERR(obj)) { diff --git a/drivers/gpu/drm/xe/display/xe_dsb_buffer.c b/drivers/gpu/drm/xe/display/xe_dsb_buffer.c index 9f941fc2e36b..58581d7aaae6 100644 --- a/drivers/gpu/drm/xe/display/xe_dsb_buffer.c +++ b/drivers/gpu/drm/xe/display/xe_dsb_buffer.c @@ -43,11 +43,11 @@ bool intel_dsb_buffer_create(struct intel_crtc *crtc, struct intel_dsb_buffer *d return false; /* Set scanout flag for WC mapping */ - obj = xe_bo_create_pin_map(xe, xe_device_get_root_tile(xe), - NULL, PAGE_ALIGN(size), - ttm_bo_type_kernel, - XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(xe)) | - XE_BO_FLAG_SCANOUT | XE_BO_FLAG_GGTT); + obj = xe_bo_create_pin_map_novm(xe, xe_device_get_root_tile(xe), + PAGE_ALIGN(size), + ttm_bo_type_kernel, + XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(xe)) | + XE_BO_FLAG_SCANOUT | XE_BO_FLAG_GGTT, false); if (IS_ERR(obj)) { kfree(vma); return false; diff --git a/drivers/gpu/drm/xe/display/xe_hdcp_gsc.c b/drivers/gpu/drm/xe/display/xe_hdcp_gsc.c index 30f1073141fc..4ae847b628e2 100644 --- a/drivers/gpu/drm/xe/display/xe_hdcp_gsc.c +++ b/drivers/gpu/drm/xe/display/xe_hdcp_gsc.c @@ -72,10 +72,10 @@ static int intel_hdcp_gsc_initialize_message(struct xe_device *xe, int ret = 0; /* allocate object of two page for HDCP command memory and store it */ - bo = xe_bo_create_pin_map(xe, xe_device_get_root_tile(xe), NULL, PAGE_SIZE * 2, - ttm_bo_type_kernel, - XE_BO_FLAG_SYSTEM | - XE_BO_FLAG_GGTT); + bo = xe_bo_create_pin_map_novm(xe, xe_device_get_root_tile(xe), PAGE_SIZE * 2, + ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | + XE_BO_FLAG_GGTT, false); if (IS_ERR(bo)) { drm_err(&xe->drm, "Failed to allocate bo for HDCP streaming command!\n"); diff --git a/drivers/gpu/drm/xe/tests/xe_migrate.c b/drivers/gpu/drm/xe/tests/xe_migrate.c index afa794e56065..5904d658d1f2 100644 --- a/drivers/gpu/drm/xe/tests/xe_migrate.c +++ b/drivers/gpu/drm/xe/tests/xe_migrate.c @@ -204,7 +204,8 @@ static void xe_migrate_sanity_test(struct xe_migrate *m, struct kunit *test, big = xe_bo_create_pin_map(xe, tile, m->q->vm, SZ_4M, ttm_bo_type_kernel, - XE_BO_FLAG_VRAM_IF_DGFX(tile)); + XE_BO_FLAG_VRAM_IF_DGFX(tile), + exec); if (IS_ERR(big)) { KUNIT_FAIL(test, "Failed to allocate bo: %li\n", PTR_ERR(big)); goto vunmap; @@ -212,7 +213,8 @@ static void xe_migrate_sanity_test(struct xe_migrate *m, struct kunit *test, pt = xe_bo_create_pin_map(xe, tile, m->q->vm, XE_PAGE_SIZE, ttm_bo_type_kernel, - XE_BO_FLAG_VRAM_IF_DGFX(tile)); + XE_BO_FLAG_VRAM_IF_DGFX(tile), + exec); if (IS_ERR(pt)) { KUNIT_FAIL(test, "Failed to allocate fake pt: %li\n", PTR_ERR(pt)); @@ -222,7 +224,8 @@ static void xe_migrate_sanity_test(struct xe_migrate *m, struct kunit *test, tiny = xe_bo_create_pin_map(xe, tile, m->q->vm, 2 * SZ_4K, ttm_bo_type_kernel, - XE_BO_FLAG_VRAM_IF_DGFX(tile)); + XE_BO_FLAG_VRAM_IF_DGFX(tile), + exec); if (IS_ERR(tiny)) { KUNIT_FAIL(test, "Failed to allocate tiny fake pt: %li\n", PTR_ERR(tiny)); diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c index b7d7d1445790..1012655b65fd 100644 --- a/drivers/gpu/drm/xe/xe_bo.c +++ b/drivers/gpu/drm/xe/xe_bo.c @@ -2513,16 +2513,59 @@ xe_bo_create_pin_map_at_novm(struct xe_device *xe, struct xe_tile *tile, return ret ? ERR_PTR(ret) : bo; } +/** + * xe_bo_create_pin_map() - Create pinned and mapped bo + * @xe: The xe device. + * @tile: The tile to select for migration of this bo, and the tile used for + * @vm: The vm to associate the buffer object with. The vm's resv must be locked + * with the transaction represented by @exec. + * GGTT binding if any. Only to be non-NULL for ttm_bo_type_kernel bos. + * @size: The storage size to use for the bo. + * @type: The TTM buffer object type. + * @flags: XE_BO_FLAG_ flags. + * @exec: The drm_exec transaction to use for exhaustive eviction, and + * previously used for locking @vm's resv. + * + * Create a pinned and mapped bo. The bo will be external and not associated + * with a VM. + * + * Return: The buffer object on success. Negative error pointer on failure. + * In particular, the function may return ERR_PTR(%-EINTR) if @exec was + * configured for interruptible locking. + */ struct xe_bo *xe_bo_create_pin_map(struct xe_device *xe, struct xe_tile *tile, struct xe_vm *vm, size_t size, - enum ttm_bo_type type, u32 flags) + enum ttm_bo_type type, u32 flags, + struct drm_exec *exec) { - struct drm_exec *exec = vm ? xe_vm_validation_exec(vm) : XE_VALIDATION_UNIMPLEMENTED; - return xe_bo_create_pin_map_at_aligned(xe, tile, vm, size, ~0ull, type, flags, 0, exec); } +/** + * xe_bo_create_pin_map_novm() - Create pinned and mapped bo + * @xe: The xe device. + * @tile: The tile to select for migration of this bo, and the tile used for + * GGTT binding if any. Only to be non-NULL for ttm_bo_type_kernel bos. + * @size: The storage size to use for the bo. + * @type: The TTM buffer object type. + * @flags: XE_BO_FLAG_ flags. + * @intr: Whether to execut any waits for backing store interruptible. + * + * Create a pinned and mapped bo. The bo will be external and not associated + * with a VM. + * + * Return: The buffer object on success. Negative error pointer on failure. + * In particular, the function may return ERR_PTR(%-EINTR) if @intr was set + * to true on entry. + */ +struct xe_bo *xe_bo_create_pin_map_novm(struct xe_device *xe, struct xe_tile *tile, + size_t size, enum ttm_bo_type type, u32 flags, + bool intr) +{ + return xe_bo_create_pin_map_at_novm(xe, tile, size, ~0ull, type, flags, 0, intr); +} + static void __xe_bo_unpin_map_no_vm(void *arg) { xe_bo_unpin_map_no_vm(arg); @@ -2535,8 +2578,7 @@ struct xe_bo *xe_managed_bo_create_pin_map(struct xe_device *xe, struct xe_tile 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); + bo = xe_bo_create_pin_map_novm(xe, tile, size, ttm_bo_type_kernel, flags, true); if (IS_ERR(bo)) return bo; diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h index d9aef56ab878..143f59bcfb93 100644 --- a/drivers/gpu/drm/xe/xe_bo.h +++ b/drivers/gpu/drm/xe/xe_bo.h @@ -108,7 +108,11 @@ struct xe_bo *xe_bo_create_user(struct xe_device *xe, struct xe_vm *vm, size_t s u16 cpu_caching, u32 flags, struct drm_exec *exec); struct xe_bo *xe_bo_create_pin_map(struct xe_device *xe, struct xe_tile *tile, struct xe_vm *vm, size_t size, - enum ttm_bo_type type, u32 flags); + enum ttm_bo_type type, u32 flags, + struct drm_exec *exec); +struct xe_bo *xe_bo_create_pin_map_novm(struct xe_device *xe, struct xe_tile *tile, + size_t size, enum ttm_bo_type type, u32 flags, + bool intr); struct xe_bo * xe_bo_create_pin_map_at_novm(struct xe_device *xe, struct xe_tile *tile, size_t size, u64 offset, enum ttm_bo_type type, diff --git a/drivers/gpu/drm/xe/xe_gsc.c b/drivers/gpu/drm/xe/xe_gsc.c index f5ae28af60d4..83d61bf8ec62 100644 --- a/drivers/gpu/drm/xe/xe_gsc.c +++ b/drivers/gpu/drm/xe/xe_gsc.c @@ -136,10 +136,10 @@ static int query_compatibility_version(struct xe_gsc *gsc) u64 ggtt_offset; int err; - bo = xe_bo_create_pin_map(xe, tile, NULL, GSC_VER_PKT_SZ * 2, - ttm_bo_type_kernel, - XE_BO_FLAG_SYSTEM | - XE_BO_FLAG_GGTT); + bo = xe_bo_create_pin_map_novm(xe, tile, GSC_VER_PKT_SZ * 2, + ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | + XE_BO_FLAG_GGTT, false); if (IS_ERR(bo)) { xe_gt_err(gt, "failed to allocate bo for GSC version query\n"); return PTR_ERR(bo); diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c b/drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c index c712111aa30d..44cc612b0a75 100644 --- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c +++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c @@ -55,12 +55,12 @@ static int pf_send_guc_save_vf_state(struct xe_gt *gt, unsigned int vfid, xe_gt_assert(gt, size % sizeof(u32) == 0); xe_gt_assert(gt, size == ndwords * sizeof(u32)); - bo = xe_bo_create_pin_map(xe, tile, NULL, - ALIGN(size, PAGE_SIZE), - ttm_bo_type_kernel, - XE_BO_FLAG_SYSTEM | - XE_BO_FLAG_GGTT | - XE_BO_FLAG_GGTT_INVALIDATE); + bo = xe_bo_create_pin_map_novm(xe, tile, + ALIGN(size, PAGE_SIZE), + ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | + XE_BO_FLAG_GGTT | + XE_BO_FLAG_GGTT_INVALIDATE, false); if (IS_ERR(bo)) return PTR_ERR(bo); @@ -91,12 +91,12 @@ static int pf_send_guc_restore_vf_state(struct xe_gt *gt, unsigned int vfid, xe_gt_assert(gt, size % sizeof(u32) == 0); xe_gt_assert(gt, size == ndwords * sizeof(u32)); - bo = xe_bo_create_pin_map(xe, tile, NULL, - ALIGN(size, PAGE_SIZE), - ttm_bo_type_kernel, - XE_BO_FLAG_SYSTEM | - XE_BO_FLAG_GGTT | - XE_BO_FLAG_GGTT_INVALIDATE); + bo = xe_bo_create_pin_map_novm(xe, tile, + ALIGN(size, PAGE_SIZE), + ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | + XE_BO_FLAG_GGTT | + XE_BO_FLAG_GGTT_INVALIDATE, false); if (IS_ERR(bo)) return PTR_ERR(bo); diff --git a/drivers/gpu/drm/xe/xe_guc_engine_activity.c b/drivers/gpu/drm/xe/xe_guc_engine_activity.c index 92e1f9f41b8c..2b99c1ebdd58 100644 --- a/drivers/gpu/drm/xe/xe_guc_engine_activity.c +++ b/drivers/gpu/drm/xe/xe_guc_engine_activity.c @@ -94,16 +94,17 @@ static int allocate_engine_activity_buffers(struct xe_guc *guc, struct xe_tile *tile = gt_to_tile(gt); struct xe_bo *bo, *metadata_bo; - metadata_bo = xe_bo_create_pin_map(gt_to_xe(gt), tile, NULL, PAGE_ALIGN(metadata_size), - ttm_bo_type_kernel, XE_BO_FLAG_SYSTEM | - XE_BO_FLAG_GGTT | XE_BO_FLAG_GGTT_INVALIDATE); + metadata_bo = xe_bo_create_pin_map_novm(gt_to_xe(gt), tile, PAGE_ALIGN(metadata_size), + ttm_bo_type_kernel, XE_BO_FLAG_SYSTEM | + XE_BO_FLAG_GGTT | XE_BO_FLAG_GGTT_INVALIDATE, + false); if (IS_ERR(metadata_bo)) return PTR_ERR(metadata_bo); - bo = xe_bo_create_pin_map(gt_to_xe(gt), tile, NULL, PAGE_ALIGN(size), - ttm_bo_type_kernel, XE_BO_FLAG_VRAM_IF_DGFX(tile) | - XE_BO_FLAG_GGTT | XE_BO_FLAG_GGTT_INVALIDATE); + bo = xe_bo_create_pin_map_novm(gt_to_xe(gt), tile, PAGE_ALIGN(size), + ttm_bo_type_kernel, XE_BO_FLAG_VRAM_IF_DGFX(tile) | + XE_BO_FLAG_GGTT | XE_BO_FLAG_GGTT_INVALIDATE, false); if (IS_ERR(bo)) { xe_bo_unpin_map_no_vm(metadata_bo); diff --git a/drivers/gpu/drm/xe/xe_lmtt.c b/drivers/gpu/drm/xe/xe_lmtt.c index f2bfbfa3efa1..62fc5a1a332d 100644 --- a/drivers/gpu/drm/xe/xe_lmtt.c +++ b/drivers/gpu/drm/xe/xe_lmtt.c @@ -67,12 +67,12 @@ static struct xe_lmtt_pt *lmtt_pt_alloc(struct xe_lmtt *lmtt, unsigned int level goto out; } - bo = xe_bo_create_pin_map(lmtt_to_xe(lmtt), lmtt_to_tile(lmtt), NULL, - PAGE_ALIGN(lmtt->ops->lmtt_pte_size(level) * - lmtt->ops->lmtt_pte_num(level)), - ttm_bo_type_kernel, - XE_BO_FLAG_VRAM_IF_DGFX(lmtt_to_tile(lmtt)) | - XE_BO_FLAG_NEEDS_64K); + bo = xe_bo_create_pin_map_novm(lmtt_to_xe(lmtt), lmtt_to_tile(lmtt), + PAGE_ALIGN(lmtt->ops->lmtt_pte_size(level) * + lmtt->ops->lmtt_pte_num(level)), + ttm_bo_type_kernel, + XE_BO_FLAG_VRAM_IF_DGFX(lmtt_to_tile(lmtt)) | + XE_BO_FLAG_NEEDS_64K, false); if (IS_ERR(bo)) { err = PTR_ERR(bo); goto out_free_pt; diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c index 8f6c3ba47882..6d52e0eb97f5 100644 --- a/drivers/gpu/drm/xe/xe_lrc.c +++ b/drivers/gpu/drm/xe/xe_lrc.c @@ -1340,9 +1340,10 @@ static int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe, if (vm && vm->xef) /* userspace */ bo_flags |= XE_BO_FLAG_PINNED_LATE_RESTORE; - lrc->bo = xe_bo_create_pin_map(xe, tile, NULL, bo_size, - ttm_bo_type_kernel, - bo_flags); + lrc->bo = xe_bo_create_pin_map_novm(xe, tile, + bo_size, + ttm_bo_type_kernel, + bo_flags, false); if (IS_ERR(lrc->bo)) return PTR_ERR(lrc->bo); diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c index 3f0c8832120f..6fad5d469629 100644 --- a/drivers/gpu/drm/xe/xe_migrate.c +++ b/drivers/gpu/drm/xe/xe_migrate.c @@ -35,6 +35,7 @@ #include "xe_sched_job.h" #include "xe_sync.h" #include "xe_trace_bo.h" +#include "xe_validation.h" #include "xe_vm.h" #include "xe_vram.h" @@ -173,7 +174,7 @@ static void xe_migrate_program_identity(struct xe_device *xe, struct xe_vm *vm, } static int xe_migrate_prepare_vm(struct xe_tile *tile, struct xe_migrate *m, - struct xe_vm *vm) + struct xe_vm *vm, struct drm_exec *exec) { struct xe_device *xe = tile_to_xe(tile); u16 pat_index = xe->pat.idx[XE_CACHE_WB]; @@ -200,7 +201,7 @@ static int xe_migrate_prepare_vm(struct xe_tile *tile, struct xe_migrate *m, num_entries * XE_PAGE_SIZE, ttm_bo_type_kernel, XE_BO_FLAG_VRAM_IF_DGFX(tile) | - XE_BO_FLAG_PAGETABLE); + XE_BO_FLAG_PAGETABLE, exec); if (IS_ERR(bo)) return PTR_ERR(bo); @@ -404,6 +405,8 @@ int xe_migrate_init(struct xe_migrate *m) struct xe_tile *tile = m->tile; struct xe_gt *primary_gt = tile->primary_gt; struct xe_device *xe = tile_to_xe(tile); + struct xe_validation_ctx ctx; + struct drm_exec exec; struct xe_vm *vm; int err; @@ -413,11 +416,16 @@ int xe_migrate_init(struct xe_migrate *m) if (IS_ERR(vm)) return PTR_ERR(vm); - xe_vm_lock(vm, false); - err = xe_migrate_prepare_vm(tile, m, vm); - xe_vm_unlock(vm); + err = 0; + xe_validation_guard(&ctx, &xe->val, &exec, (struct xe_val_flags) {}, err) { + err = xe_vm_drm_exec_lock(vm, &exec); + drm_exec_retry_on_contention(&exec); + err = xe_migrate_prepare_vm(tile, m, vm, &exec); + drm_exec_retry_on_contention(&exec); + xe_validation_retry_on_oom(&ctx, &err); + } if (err) - goto err_out; + return err; if (xe->info.has_usm) { struct xe_hw_engine *hwe = xe_gt_hw_engine(primary_gt, diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c index a188bad172ad..a4894eb0d7f3 100644 --- a/drivers/gpu/drm/xe/xe_oa.c +++ b/drivers/gpu/drm/xe/xe_oa.c @@ -883,9 +883,9 @@ static int xe_oa_alloc_oa_buffer(struct xe_oa_stream *stream, size_t size) { struct xe_bo *bo; - bo = xe_bo_create_pin_map(stream->oa->xe, stream->gt->tile, NULL, - size, ttm_bo_type_kernel, - XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT); + bo = xe_bo_create_pin_map_novm(stream->oa->xe, stream->gt->tile, + size, ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | XE_BO_FLAG_GGTT, false); if (IS_ERR(bo)) return PTR_ERR(bo); diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c index 07ecf4132d41..01eea8eb1779 100644 --- a/drivers/gpu/drm/xe/xe_pt.c +++ b/drivers/gpu/drm/xe/xe_pt.c @@ -90,6 +90,7 @@ static void xe_pt_free(struct xe_pt *pt) * @vm: The vm to create for. * @tile: The tile to create for. * @level: The page-table level. + * @exec: The drm_exec object used to lock the vm. * * Allocate and initialize a single struct xe_pt metadata structure. Also * create the corresponding page-table bo, but don't initialize it. If the @@ -101,7 +102,7 @@ static void xe_pt_free(struct xe_pt *pt) * error. */ struct xe_pt *xe_pt_create(struct xe_vm *vm, struct xe_tile *tile, - unsigned int level) + unsigned int level, struct drm_exec *exec) { struct xe_pt *pt; struct xe_bo *bo; @@ -125,9 +126,11 @@ struct xe_pt *xe_pt_create(struct xe_vm *vm, struct xe_tile *tile, bo_flags |= XE_BO_FLAG_PINNED_LATE_RESTORE; pt->level = level; + + drm_WARN_ON(&vm->xe->drm, IS_ERR_OR_NULL(exec)); bo = xe_bo_create_pin_map(vm->xe, tile, vm, SZ_4K, ttm_bo_type_kernel, - bo_flags); + bo_flags, exec); if (IS_ERR(bo)) { err = PTR_ERR(bo); goto err_kfree; @@ -591,7 +594,8 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset, if (covers || !*child) { u64 flags = 0; - xe_child = xe_pt_create(xe_walk->vm, xe_walk->tile, level - 1); + xe_child = xe_pt_create(xe_walk->vm, xe_walk->tile, level - 1, + xe_vm_validation_exec(vm)); if (IS_ERR(xe_child)) return PTR_ERR(xe_child); diff --git a/drivers/gpu/drm/xe/xe_pt.h b/drivers/gpu/drm/xe/xe_pt.h index 5ecf003d513c..4daeebaab5a1 100644 --- a/drivers/gpu/drm/xe/xe_pt.h +++ b/drivers/gpu/drm/xe/xe_pt.h @@ -10,6 +10,7 @@ #include "xe_pt_types.h" struct dma_fence; +struct drm_exec; struct xe_bo; struct xe_device; struct xe_exec_queue; @@ -29,7 +30,7 @@ struct xe_vma_ops; unsigned int xe_pt_shift(unsigned int level); struct xe_pt *xe_pt_create(struct xe_vm *vm, struct xe_tile *tile, - unsigned int level); + unsigned int level, struct drm_exec *exec); void xe_pt_populate_empty(struct xe_tile *tile, struct xe_vm *vm, struct xe_pt *pt); diff --git a/drivers/gpu/drm/xe/xe_pxp_submit.c b/drivers/gpu/drm/xe/xe_pxp_submit.c index ca95f2a4d4ef..e60526e30030 100644 --- a/drivers/gpu/drm/xe/xe_pxp_submit.c +++ b/drivers/gpu/drm/xe/xe_pxp_submit.c @@ -54,8 +54,9 @@ static int allocate_vcs_execution_resources(struct xe_pxp *pxp) * Each termination is 16 DWORDS, so 4K is enough to contain a * termination for each sessions. */ - bo = xe_bo_create_pin_map(xe, tile, NULL, SZ_4K, ttm_bo_type_kernel, - XE_BO_FLAG_SYSTEM | XE_BO_FLAG_PINNED | XE_BO_FLAG_GGTT); + bo = xe_bo_create_pin_map_novm(xe, tile, SZ_4K, ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | XE_BO_FLAG_PINNED | XE_BO_FLAG_GGTT, + false); if (IS_ERR(bo)) { err = PTR_ERR(bo); goto out_queue; @@ -87,7 +88,9 @@ static int allocate_gsc_client_resources(struct xe_gt *gt, { struct xe_tile *tile = gt_to_tile(gt); struct xe_device *xe = tile_to_xe(tile); + struct xe_validation_ctx ctx; struct xe_hw_engine *hwe; + struct drm_exec exec; struct xe_vm *vm; struct xe_bo *bo; struct xe_exec_queue *q; @@ -106,15 +109,26 @@ static int allocate_gsc_client_resources(struct xe_gt *gt, return PTR_ERR(vm); /* We allocate a single object for the batch and the in/out memory */ - xe_vm_lock(vm, false); - bo = xe_bo_create_pin_map(xe, tile, vm, PXP_BB_SIZE + inout_size * 2, - ttm_bo_type_kernel, - XE_BO_FLAG_SYSTEM | XE_BO_FLAG_PINNED | XE_BO_FLAG_NEEDS_UC); - xe_vm_unlock(vm); - if (IS_ERR(bo)) { - err = PTR_ERR(bo); - goto vm_out; + + xe_validation_guard(&ctx, &xe->val, &exec, (struct xe_val_flags){}, err) { + err = xe_vm_drm_exec_lock(vm, &exec); + drm_exec_retry_on_contention(&exec); + if (err) + break; + + bo = xe_bo_create_pin_map(xe, tile, vm, PXP_BB_SIZE + inout_size * 2, + ttm_bo_type_kernel, + XE_BO_FLAG_SYSTEM | XE_BO_FLAG_PINNED | + XE_BO_FLAG_NEEDS_UC, &exec); + drm_exec_retry_on_contention(&exec); + if (IS_ERR(bo)) { + err = PTR_ERR(bo); + xe_validation_retry_on_oom(&ctx, &err); + break; + } } + if (err) + goto vm_out; fence = xe_vm_bind_kernel_bo(vm, bo, NULL, 0, XE_CACHE_WB); if (IS_ERR(fence)) { diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index 785e81cf023d..7b04b392b729 100644 --- a/drivers/gpu/drm/xe/xe_vm.c +++ b/drivers/gpu/drm/xe/xe_vm.c @@ -1370,6 +1370,7 @@ static void vm_destroy_work_func(struct work_struct *w); * @xe: xe device. * @tile: tile to set up for. * @vm: vm to set up for. + * @exec: The struct drm_exec object used to lock the vm resv. * * Sets up a pagetable tree with one page-table per level and a single * leaf PTE. All pagetable entries point to the single page-table or, @@ -1379,20 +1380,19 @@ static void vm_destroy_work_func(struct work_struct *w); * Return: 0 on success, negative error code on error. */ static int xe_vm_create_scratch(struct xe_device *xe, struct xe_tile *tile, - struct xe_vm *vm) + struct xe_vm *vm, struct drm_exec *exec) { u8 id = tile->id; int i; for (i = MAX_HUGEPTE_LEVEL; i < vm->pt_root[id]->level; i++) { - vm->scratch_pt[id][i] = xe_pt_create(vm, tile, i); + vm->scratch_pt[id][i] = xe_pt_create(vm, tile, i, exec); if (IS_ERR(vm->scratch_pt[id][i])) { int err = PTR_ERR(vm->scratch_pt[id][i]); vm->scratch_pt[id][i] = NULL; return err; } - xe_pt_populate_empty(tile, vm, vm->scratch_pt[id][i]); } @@ -1420,9 +1420,26 @@ static void xe_vm_free_scratch(struct xe_vm *vm) } } +static void xe_vm_pt_destroy(struct xe_vm *vm) +{ + struct xe_tile *tile; + u8 id; + + xe_vm_assert_held(vm); + + for_each_tile(tile, vm->xe, id) { + if (vm->pt_root[id]) { + xe_pt_destroy(vm->pt_root[id], vm->flags, NULL); + vm->pt_root[id] = NULL; + } + } +} + struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef) { struct drm_gem_object *vm_resv_obj; + struct xe_validation_ctx ctx; + struct drm_exec exec; struct xe_vm *vm; int err, number_tiles = 0; struct xe_tile *tile; @@ -1507,49 +1524,68 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef) drm_gem_object_put(vm_resv_obj); - err = xe_vm_lock(vm, true); - if (err) - goto err_close; + err = 0; + xe_validation_guard(&ctx, &xe->val, &exec, (struct xe_val_flags) {.interruptible = true}, + err) { + err = xe_vm_drm_exec_lock(vm, &exec); + drm_exec_retry_on_contention(&exec); - if (IS_DGFX(xe) && xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) - vm->flags |= XE_VM_FLAG_64K; + if (IS_DGFX(xe) && xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) + vm->flags |= XE_VM_FLAG_64K; - for_each_tile(tile, xe, id) { - if (flags & XE_VM_FLAG_MIGRATION && - tile->id != XE_VM_FLAG_TILE_ID(flags)) - continue; + for_each_tile(tile, xe, id) { + if (flags & XE_VM_FLAG_MIGRATION && + tile->id != XE_VM_FLAG_TILE_ID(flags)) + continue; - vm->pt_root[id] = xe_pt_create(vm, tile, xe->info.vm_max_level); - if (IS_ERR(vm->pt_root[id])) { - err = PTR_ERR(vm->pt_root[id]); - vm->pt_root[id] = NULL; - goto err_unlock_close; + vm->pt_root[id] = xe_pt_create(vm, tile, xe->info.vm_max_level, + &exec); + if (IS_ERR(vm->pt_root[id])) { + err = PTR_ERR(vm->pt_root[id]); + vm->pt_root[id] = NULL; + xe_vm_pt_destroy(vm); + drm_exec_retry_on_contention(&exec); + xe_validation_retry_on_oom(&ctx, &err); + break; + } } - } + if (err) + break; - if (xe_vm_has_scratch(vm)) { - for_each_tile(tile, xe, id) { - if (!vm->pt_root[id]) - continue; + if (xe_vm_has_scratch(vm)) { + for_each_tile(tile, xe, id) { + if (!vm->pt_root[id]) + continue; - err = xe_vm_create_scratch(xe, tile, vm); + err = xe_vm_create_scratch(xe, tile, vm, &exec); + if (err) { + xe_vm_free_scratch(vm); + xe_vm_pt_destroy(vm); + drm_exec_retry_on_contention(&exec); + xe_validation_retry_on_oom(&ctx, &err); + break; + } + } if (err) - goto err_unlock_close; + break; + vm->batch_invalidate_tlb = true; } - vm->batch_invalidate_tlb = true; - } - if (vm->flags & XE_VM_FLAG_LR_MODE) - vm->batch_invalidate_tlb = false; + if (vm->flags & XE_VM_FLAG_LR_MODE) { + INIT_WORK(&vm->preempt.rebind_work, preempt_rebind_work_func); + vm->batch_invalidate_tlb = false; + } - /* Fill pt_root after allocating scratch tables */ - for_each_tile(tile, xe, id) { - if (!vm->pt_root[id]) - continue; + /* Fill pt_root after allocating scratch tables */ + for_each_tile(tile, xe, id) { + if (!vm->pt_root[id]) + continue; - xe_pt_populate_empty(tile, vm, vm->pt_root[id]); + xe_pt_populate_empty(tile, vm, vm->pt_root[id]); + } } - xe_vm_unlock(vm); + if (err) + goto err_close; /* Kernel migration VM shouldn't have a circular loop.. */ if (!(flags & XE_VM_FLAG_MIGRATION)) { @@ -1582,7 +1618,7 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef) &xe->usm.next_asid, GFP_KERNEL); up_write(&xe->usm.lock); if (err < 0) - goto err_unlock_close; + goto err_close; vm->usm.asid = asid; } @@ -1591,8 +1627,6 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef) return vm; -err_unlock_close: - xe_vm_unlock(vm); err_close: xe_vm_close_and_put(vm); return ERR_PTR(err); @@ -1725,13 +1759,7 @@ void xe_vm_close_and_put(struct xe_vm *vm) * destroy the pagetables immediately. */ xe_vm_free_scratch(vm); - - for_each_tile(tile, xe, id) { - if (vm->pt_root[id]) { - xe_pt_destroy(vm->pt_root[id], vm->flags, NULL); - vm->pt_root[id] = NULL; - } - } + xe_vm_pt_destroy(vm); xe_vm_unlock(vm); /* @@ -3781,7 +3809,6 @@ release_vm_lock: */ int xe_vm_lock(struct xe_vm *vm, bool intr) { - struct drm_exec *exec = XE_VALIDATION_UNIMPLEMENTED; int ret; if (intr) @@ -3789,9 +3816,6 @@ int xe_vm_lock(struct xe_vm *vm, bool intr) else ret = dma_resv_lock(xe_vm_resv(vm), NULL); - if (!ret) - xe_vm_set_validation_exec(vm, exec); - return ret; } @@ -3803,7 +3827,6 @@ int xe_vm_lock(struct xe_vm *vm, bool intr) */ void xe_vm_unlock(struct xe_vm *vm) { - xe_vm_set_validation_exec(vm, NULL); dma_resv_unlock(xe_vm_resv(vm)); }