Intel-XE Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: Matthew Brost <matthew.brost@intel.com>
To: intel-xe@lists.freedesktop.org, dri-devel@lists.freedesktop.org
Cc: airlied@gmail.com, christian.koenig@amd.com,
	thomas.hellstrom@linux.intel.com, matthew.auld@intel.com,
	daniel@ffwll.ch
Subject: [RFC PATCH 19/28] drm/xe: Update PT layer to understand ranges in VRAM
Date: Tue, 27 Aug 2024 19:48:52 -0700	[thread overview]
Message-ID: <20240828024901.2582335-20-matthew.brost@intel.com> (raw)
In-Reply-To: <20240828024901.2582335-1-matthew.brost@intel.com>

Kinda cheating here using BO directly rather than VRAM pages. Same at
the moment as mixed mappings are not supported. If this changes, then
the arary of pages / dma addresses will need a cursor.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
---
 drivers/gpu/drm/xe/xe_pt.c  | 22 ++++++++++++++++------
 drivers/gpu/drm/xe/xe_svm.h | 10 ++++++++++
 2 files changed, 26 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index fc86adf9f0a6..e9195029ea60 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -607,9 +607,12 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
 		 struct xe_vm_pgtable_update *entries, u32 *num_entries)
 {
 	struct xe_device *xe = tile_to_xe(tile);
-	struct xe_bo *bo = xe_vma_bo(vma);
-	bool is_devmem = !xe_vma_is_userptr(vma) && bo &&
-		(xe_bo_is_vram(bo) || xe_bo_is_stolen_devmem(bo));
+	bool range_devmem = range && xe_svm_range_in_vram(range);
+	struct xe_bo *bo = range_devmem ? range->base.vram_allocation :
+		xe_vma_bo(vma);
+	bool is_devmem = range_devmem ||
+		(!xe_vma_is_userptr(vma) && bo &&
+		(xe_bo_is_vram(bo) || xe_bo_is_stolen_devmem(bo)));
 	struct xe_res_cursor curs;
 	struct xe_pt_stage_bind_walk xe_walk = {
 		.base = {
@@ -675,9 +678,16 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
 	xe_bo_assert_held(bo);
 
 	if (range) {
-		xe_res_first_dma(range->base.dma_addr, 0,
-				 range->base.va.end - range->base.va.start,
-				 range->base.order, &curs);
+		if (is_devmem)
+			xe_res_first(bo->ttm.resource, 0,
+				     range->base.va.end - range->base.va.start,
+				     &curs);
+		else if (xe_svm_range_has_dma_mapping(range))
+			xe_res_first_dma(range->base.dma_addr, 0,
+					 range->base.va.end - range->base.va.start,
+					 range->base.order, &curs);
+		else
+			return -EAGAIN;	/* Invalidation corner case */
 	} else if (!xe_vma_is_null(vma)) {
 		if (xe_vma_is_userptr(vma))
 			xe_res_first_sg(to_userptr_vma(vma)->userptr.sg, 0,
diff --git a/drivers/gpu/drm/xe/xe_svm.h b/drivers/gpu/drm/xe/xe_svm.h
index f15df5c813f1..8b72e91cc37d 100644
--- a/drivers/gpu/drm/xe/xe_svm.h
+++ b/drivers/gpu/drm/xe/xe_svm.h
@@ -35,6 +35,16 @@ static inline bool xe_svm_range_pages_valid(struct xe_svm_range *range)
 	return drm_gpusvm_range_pages_valid(range->base.gpusvm, &range->base);
 }
 
+static inline bool xe_svm_range_in_vram(struct xe_svm_range *range)
+{
+	return range->base.flags.has_vram_pages;
+}
+
+static inline bool xe_svm_range_has_dma_mapping(struct xe_svm_range *range)
+{
+	return range->base.flags.has_dma_mapping;
+}
+
 #define xe_svm_notifier_lock(vm__)	\
 	drm_gpusvm_notifier_lock(&(vm__)->svm.gpusvm)
 
-- 
2.34.1


  parent reply	other threads:[~2024-08-28  2:48 UTC|newest]

Thread overview: 100+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-08-28  2:48 [RFC PATCH 00/28] Introduce GPU SVM and Xe SVM implementation Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 01/28] dma-buf: Split out dma fence array create into alloc and arm functions Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 02/28] drm/xe: Invalidate media_gt TLBs in PT code Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 03/28] drm/xe: Retry BO allocation Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 04/28] mm/migrate: Add migrate_device_vma_range Matthew Brost
2024-08-29  9:03   ` Daniel Vetter
2024-08-29 15:58     ` Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 05/28] drm/gpusvm: Add support for GPU Shared Virtual Memory Matthew Brost
2024-08-28 14:31   ` Daniel Vetter
2024-08-28 14:46     ` Christian König
2024-08-28 15:43       ` Matthew Brost
2024-08-28 16:06         ` Alex Deucher
2024-08-28 16:25         ` Daniel Vetter
2024-08-29 16:40           ` Matthew Brost
2024-09-02 11:29             ` Daniel Vetter
2024-08-30  5:00     ` Matthew Brost
2024-09-02 11:36       ` Daniel Vetter
2024-08-28 18:50   ` Daniel Vetter
2024-08-29 16:49     ` Matthew Brost
2024-09-02 11:40       ` Daniel Vetter
2024-08-29  9:16   ` Thomas Hellström
2024-08-29 17:45     ` Matthew Brost
2024-08-29 18:13       ` Matthew Brost
2024-08-29 19:18       ` Thomas Hellström
2024-08-29 20:56         ` Matthew Brost
2024-08-30  8:18           ` Thomas Hellström
2024-08-30 13:58             ` Matthew Brost
2024-09-02  9:57               ` Thomas Hellström
2024-08-30  9:57           ` Thomas Hellström
2024-08-30 13:47             ` Matthew Brost
2024-09-02  9:45               ` Thomas Hellström
2024-09-02 12:33           ` Daniel Vetter
2024-09-04 12:27             ` Thomas Hellström
2024-09-24  8:41               ` Simona Vetter
2024-08-30  1:35     ` Matthew Brost
2024-08-29  9:45   ` Daniel Vetter
2024-08-29 17:27     ` Matthew Brost
2024-09-02 11:53       ` Daniel Vetter
2024-09-02 17:03         ` Matthew Brost
2024-09-11 16:06           ` Matthew Brost
2024-08-30  9:16   ` Thomas Hellström
2024-09-02 12:20     ` Daniel Vetter
2024-09-06 18:41   ` Zeng, Oak
2024-09-24  9:25     ` Simona Vetter
2024-09-25 16:34       ` Zeng, Oak
2024-09-24 10:42   ` Thomas Hellström
2024-09-24 16:30     ` Matthew Brost
2024-09-25 21:12       ` Matthew Brost
2024-10-09 10:50   ` Thomas Hellström
2024-10-16  3:18     ` Matthew Brost
2024-10-16  6:27       ` Thomas Hellström
2024-10-16  8:24         ` Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 06/28] drm/xe/uapi: Add DRM_XE_VM_BIND_FLAG_SYSTEM_ALLOCATON flag Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 07/28] drm/xe: Add SVM init / fini to faulting VMs Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 08/28] drm/xe: Add dma_addr res cursor Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 09/28] drm/xe: Add SVM range invalidation Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 10/28] drm/gpuvm: Add DRM_GPUVA_OP_USER Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 11/28] drm/xe: Add (re)bind to SVM page fault handler Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 12/28] drm/xe: Add SVM garbage collector Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 13/28] drm/xe: Add unbind to " Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 14/28] drm/xe: Do not allow system allocator VMA unbind if the GPU has bindings Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 15/28] drm/xe: Enable system allocator uAPI Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 16/28] drm/xe: Add migrate layer functions for SVM support Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 17/28] drm/xe: Add SVM device memory mirroring Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 18/28] drm/xe: Add GPUSVM copy SRAM / VRAM vfunc functions Matthew Brost
2024-08-28  2:48 ` Matthew Brost [this message]
2024-08-28  2:48 ` [RFC PATCH 20/28] drm/xe: Add Xe SVM populate_vram_pfn vfunc Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 21/28] drm/xe: Add Xe SVM vram_release vfunc Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 22/28] drm/xe: Add BO flags required for SVM Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 23/28] drm/xe: Add SVM VRAM migration Matthew Brost
2024-08-28 16:06   ` Daniel Vetter
2024-08-28 18:22     ` Daniel Vetter
2024-08-29  9:24     ` Christian König
2024-08-29  9:53       ` Thomas Hellström
2024-08-29 11:02         ` Daniel Vetter
2024-08-29 22:12           ` Matthew Brost
2024-08-29 22:23             ` Matthew Brost
2024-09-02 11:01             ` Christian König
2024-09-02 12:50               ` Daniel Vetter
2024-09-02 12:48             ` Daniel Vetter
2024-09-02 22:20               ` Matthew Brost
2024-09-03  8:07                 ` Simona Vetter
2024-08-29 14:30         ` Christian König
2024-08-29 21:53           ` Matthew Brost
2024-08-29 21:48       ` Matthew Brost
2024-09-02 13:02         ` Daniel Vetter
2024-08-28  2:48 ` [RFC PATCH 24/28] drm/xe: Basic SVM BO eviction Matthew Brost
2024-08-29 10:14   ` Daniel Vetter
2024-08-29 15:55     ` Matthew Brost
2024-09-02 13:05       ` Daniel Vetter
2024-08-28  2:48 ` [RFC PATCH 25/28] drm/xe: Add SVM debug Matthew Brost
2024-08-28  2:48 ` [RFC PATCH 26/28] drm/xe: Add modparam for SVM notifier size Matthew Brost
2024-08-28  2:49 ` [RFC PATCH 27/28] drm/xe: Add modparam for SVM prefault Matthew Brost
2024-08-28  2:49 ` [RFC PATCH 28/28] drm/gpusvm: Ensure all pages migrated upon eviction Matthew Brost
2024-08-28  2:55 ` ✓ CI.Patch_applied: success for Introduce GPU SVM and Xe SVM implementation Patchwork
2024-08-28  2:55 ` ✗ CI.checkpatch: warning " Patchwork
2024-08-28  2:56 ` ✗ CI.KUnit: failure " Patchwork
2024-09-24  9:16 ` [RFC PATCH 00/28] " Simona Vetter
2024-09-24 19:36   ` Matthew Brost
2024-09-25 11:41     ` Simona Vetter

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20240828024901.2582335-20-matthew.brost@intel.com \
    --to=matthew.brost@intel.com \
    --cc=airlied@gmail.com \
    --cc=christian.koenig@amd.com \
    --cc=daniel@ffwll.ch \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=intel-xe@lists.freedesktop.org \
    --cc=matthew.auld@intel.com \
    --cc=thomas.hellstrom@linux.intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox