[CI v3] drm/xe/vram: Add early VRAM health check

[Matthew Auld <matthew.auld@intel.com>]

DO NOT REVIEW.

Use the last page as a canary for BAR sizing, CCS sizing, identity map
setup. If something is wrong the last page is where we will likely find
it. Hit it with everything we have.

Assisted-by: Gemini:gemini-3.1-pro-preview
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
---
 drivers/gpu/drm/xe/display/xe_dsb_buffer.c |   2 +-
 drivers/gpu/drm/xe/display/xe_fb_pin.c     |   2 +-
 drivers/gpu/drm/xe/xe_device.c             |  12 +-
 drivers/gpu/drm/xe/xe_device.h             |   2 +-
 drivers/gpu/drm/xe/xe_migrate.c            |  61 +++++++
 drivers/gpu/drm/xe/xe_migrate.h            |   6 +
 drivers/gpu/drm/xe/xe_tile_types.h         |   4 +
 drivers/gpu/drm/xe/xe_vram.c               | 186 +++++++++++++++++++++
 drivers/gpu/drm/xe/xe_vram.h               |  10 ++
 9 files changed, 279 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/xe/display/xe_dsb_buffer.c b/drivers/gpu/drm/xe/display/xe_dsb_buffer.c
index a7158c73a14c..82974e933e8d 100644
--- a/drivers/gpu/drm/xe/display/xe_dsb_buffer.c
+++ b/drivers/gpu/drm/xe/display/xe_dsb_buffer.c
@@ -88,7 +88,7 @@ static void xe_dsb_buffer_flush_map(struct intel_dsb_buffer *dsb_buf)
 	 * both for weak ordering archs and discrete cards.
 	 */
 	xe_device_wmb(xe);
-	xe_device_l2_flush(xe);
+	xe_device_l2_flush(xe, false);
 }
 
 const struct intel_display_dsb_interface xe_display_dsb_interface = {
diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c
index 73469ea5f333..b46a2c32ac07 100644
--- a/drivers/gpu/drm/xe/display/xe_fb_pin.c
+++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c
@@ -203,7 +203,7 @@ static int __xe_pin_fb_vma_dpt(struct drm_gem_object *obj,
 	vma->node = dpt->ggtt_node[tile0->id];
 
 	/* Ensure DPT writes are flushed */
-	xe_device_l2_flush(xe);
+	xe_device_l2_flush(xe, false);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index 74d566693dfd..c1c4c8f61f37 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -1051,6 +1051,8 @@ int xe_device_probe(struct xe_device *xe)
 	if (err)
 		return err;
 
+	xe_vram_reserve_memtest_bo(xe);
+
 	for_each_tile(tile, xe, id) {
 		err = xe_tile_init(tile);
 		if (err)
@@ -1067,6 +1069,10 @@ int xe_device_probe(struct xe_device *xe)
 			return err;
 	}
 
+	err = xe_vram_memtest(xe);
+	if (err)
+		return err;
+
 	err = xe_pagefault_init(xe);
 	if (err)
 		return err;
@@ -1270,7 +1276,7 @@ bool xe_device_is_l2_flush_optimized(struct xe_device *xe)
 	return false;
 }
 
-void xe_device_l2_flush(struct xe_device *xe)
+void xe_device_l2_flush(struct xe_device *xe, bool force)
 {
 	struct xe_gt *gt;
 
@@ -1278,7 +1284,7 @@ void xe_device_l2_flush(struct xe_device *xe)
 	if (!gt)
 		return;
 
-	if (!XE_GT_WA(gt, 16023588340))
+	if (!force && !XE_GT_WA(gt, 16023588340))
 		return;
 
 	CLASS(xe_force_wake, fw_ref)(gt_to_fw(gt), XE_FW_GT);
@@ -1333,7 +1339,7 @@ void xe_device_td_flush(struct xe_device *xe)
 
 	if (XE_GT_WA(root_gt, 16023588340)) {
 		/* A transient flush is not sufficient: flush the L2 */
-		xe_device_l2_flush(xe);
+		xe_device_l2_flush(xe, false);
 	} else {
 		xe_guc_pc_apply_flush_freq_limit(&root_gt->uc.guc.pc);
 		tdf_request_sync(xe);
diff --git a/drivers/gpu/drm/xe/xe_device.h b/drivers/gpu/drm/xe/xe_device.h
index 6c4cfaebc44a..6d3d6d5eba29 100644
--- a/drivers/gpu/drm/xe/xe_device.h
+++ b/drivers/gpu/drm/xe/xe_device.h
@@ -205,7 +205,7 @@ u64 xe_device_uncanonicalize_addr(struct xe_device *xe, u64 address);
 
 bool xe_device_is_l2_flush_optimized(struct xe_device *xe);
 void xe_device_td_flush(struct xe_device *xe);
-void xe_device_l2_flush(struct xe_device *xe);
+void xe_device_l2_flush(struct xe_device *xe, bool force);
 
 static inline bool xe_device_wedged(struct xe_device *xe)
 {
diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
index 75b83687f1b5..271bcfb0859c 100644
--- a/drivers/gpu/drm/xe/xe_migrate.c
+++ b/drivers/gpu/drm/xe/xe_migrate.c
@@ -2633,3 +2633,64 @@ void xe_migrate_job_lock_assert(struct xe_exec_queue *q)
 #if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
 #include "tests/xe_migrate.c"
 #endif
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_MEM)
+void xe_migrate_debug_ccs_overlap(struct xe_migrate *m,
+				  struct xe_bo *scratch_bo,
+				  bool write_to_ccs)
+{
+	struct xe_device *xe = tile_to_xe(m->tile);
+	struct xe_gt *gt = m->tile->primary_gt;
+	struct dma_fence *fence;
+	struct xe_bb *bb;
+	struct xe_sched_job *job;
+	u64 first_page_dpa, clear_L0_ofs, scratch_dpa, scratch_L0_ofs;
+
+	if (!xe_device_has_flat_ccs(xe))
+		return;
+
+	first_page_dpa = xe_vram_region_dpa_base(m->tile->mem.vram);
+	clear_L0_ofs = xe_migrate_vram_ofs(xe, first_page_dpa, true);
+
+	scratch_dpa = xe_bo_addr(scratch_bo, 0, XE_PAGE_SIZE);
+	scratch_L0_ofs = xe_migrate_vram_ofs(xe, scratch_dpa, false);
+
+	bb = xe_bb_new(gt, 32, xe->info.has_usm);
+	if (IS_ERR(bb)) {
+		drm_warn(&xe->drm, "Failed to create bb for VRAM overlap check\n");
+		return;
+	}
+
+	/* 4MB payload = 16KB CCS metadata. Covers up to 4MB of VRAM holes */
+	if (write_to_ccs) {
+		emit_copy_ccs(gt, bb, clear_L0_ofs, true,
+			      scratch_L0_ofs, false, SZ_4M);
+	} else {
+		emit_copy_ccs(gt, bb, scratch_L0_ofs, false,
+			      clear_L0_ofs, true, SZ_4M);
+	}
+
+	bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
+
+	job = xe_bb_create_migration_job(m->q, bb,
+					 xe_migrate_batch_base(m, xe->info.has_usm),
+					 0);
+	if (!IS_ERR(job)) {
+		xe_sched_job_add_migrate_flush(job, MI_FLUSH_DW_CCS);
+
+		mutex_lock(&m->job_mutex);
+		xe_sched_job_arm(job);
+
+		fence = dma_fence_get(&job->drm.s_fence->finished);
+		xe_sched_job_push(job);
+		mutex_unlock(&m->job_mutex);
+
+		dma_fence_wait(fence, false);
+		dma_fence_put(fence);
+	} else {
+		drm_warn(&xe->drm, "Failed to create job for VRAM overlap check\n");
+	}
+
+	xe_bb_free(bb, NULL);
+}
+#endif
diff --git a/drivers/gpu/drm/xe/xe_migrate.h b/drivers/gpu/drm/xe/xe_migrate.h
index c3a268b01768..a106de23aa1e 100644
--- a/drivers/gpu/drm/xe/xe_migrate.h
+++ b/drivers/gpu/drm/xe/xe_migrate.h
@@ -182,4 +182,10 @@ static inline void xe_migrate_job_lock_assert(struct xe_exec_queue *q)
 void xe_migrate_job_lock(struct xe_migrate *m, struct xe_exec_queue *q);
 void xe_migrate_job_unlock(struct xe_migrate *m, struct xe_exec_queue *q);
 
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_MEM)
+void xe_migrate_debug_ccs_overlap(struct xe_migrate *m,
+				  struct xe_bo *scratch_bo,
+				  bool write_to_ccs);
+#endif
+
 #endif
diff --git a/drivers/gpu/drm/xe/xe_tile_types.h b/drivers/gpu/drm/xe/xe_tile_types.h
index 0048100ccb72..e1368c04846a 100644
--- a/drivers/gpu/drm/xe/xe_tile_types.h
+++ b/drivers/gpu/drm/xe/xe_tile_types.h
@@ -97,6 +97,10 @@ struct xe_tile {
 		 * Only main GT has page reclaim list allocations.
 		 */
 		struct xe_sa_manager *reclaim_pool;
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_MEM)
+		/** @mem.memtest_bo: VRAM overlap check BO */
+		struct xe_bo *memtest_bo;
+#endif
 	} mem;
 
 	/** @sriov: tile level virtualization data */
diff --git a/drivers/gpu/drm/xe/xe_vram.c b/drivers/gpu/drm/xe/xe_vram.c
index 23eb7edbdd57..f3bc38983157 100644
--- a/drivers/gpu/drm/xe/xe_vram.c
+++ b/drivers/gpu/drm/xe/xe_vram.c
@@ -17,8 +17,11 @@
 #include "xe_device.h"
 #include "xe_force_wake.h"
 #include "xe_gt_mcr.h"
+#include "xe_map.h"
+#include "xe_migrate.h"
 #include "xe_mmio.h"
 #include "xe_sriov.h"
+#include "xe_tile.h"
 #include "xe_tile_sriov_vf.h"
 #include "xe_ttm_vram_mgr.h"
 #include "xe_vram.h"
@@ -387,3 +390,186 @@ resource_size_t xe_vram_region_actual_physical_size(const struct xe_vram_region
 	return vram ? vram->actual_physical_size : 0;
 }
 EXPORT_SYMBOL_IF_KUNIT(xe_vram_region_actual_physical_size);
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_MEM)
+void xe_vram_reserve_memtest_bo(struct xe_device *xe)
+{
+	struct xe_tile *tile;
+	u8 id;
+
+	for_each_tile(tile, xe, id) {
+		u64 vram_size;
+
+		if (!tile->mem.vram)
+			continue;
+
+		if (tile->mem.vram->io_size < tile->mem.vram->usable_size) {
+			drm_info(&xe->drm,
+				 "Tile %d: Small-BAR system detected, skipping VRAM memtest\n",
+				 id);
+			continue;
+		}
+
+		vram_size = tile->mem.vram->usable_size;
+
+		tile->mem.memtest_bo = xe_bo_create_pin_map_at_novm(xe, tile, SZ_64K,
+							    vram_size - SZ_64K,
+							    ttm_bo_type_kernel,
+							    XE_BO_FLAG_VRAM_IF_DGFX(tile),
+							    0, false);
+		if (IS_ERR(tile->mem.memtest_bo)) {
+			drm_warn(&xe->drm, "Tile %d: Failed to reserve memtest BO\n", id);
+			tile->mem.memtest_bo = NULL;
+			continue;
+		}
+
+		drm_info(&xe->drm, "Tile %d: Reserved memtest BO at offset 0x%llx\n",
+			 id, vram_size - SZ_64K);
+	}
+}
+
+void xe_vram_free_memtest_bos(struct xe_device *xe)
+{
+	struct xe_tile *tile;
+	u8 id;
+
+	for_each_tile(tile, xe, id) {
+		if (tile->mem.memtest_bo) {
+			xe_bo_unpin_map_no_vm(tile->mem.memtest_bo);
+			tile->mem.memtest_bo = NULL;
+		}
+	}
+}
+
+int xe_vram_memtest(struct xe_device *xe)
+{
+	struct xe_tile *tile;
+	u8 id;
+	int err = 0;
+
+	for_each_tile(tile, xe, id) {
+		struct xe_bo *last_page_bo = tile->mem.memtest_bo;
+		struct dma_fence *fence;
+		bool overlap = false;
+		int i;
+		u8 val;
+
+		if (!last_page_bo || !tile->migrate)
+			continue;
+
+		drm_info(&xe->drm, "Tile %d: Running VRAM memtest...\n", id);
+
+		/* 1. CPU write and readback first and last byte of the last page */
+		xe_map_wr(xe, &last_page_bo->vmap, 0, u8, 0xA5);
+		xe_map_wr(xe, &last_page_bo->vmap, SZ_64K - 1, u8, 0x5A);
+
+		val = xe_map_rd(xe, &last_page_bo->vmap, 0, u8);
+		if (drm_WARN(&xe->drm, val != 0xA5,
+			     "Tile %d: CPU memtest failed at offset 0 (expected 0xA5, got 0x%02x)\n",
+			     id, val)) {
+			err = -EIO;
+			goto unpin;
+		}
+
+		val = xe_map_rd(xe, &last_page_bo->vmap, SZ_64K - 1, u8);
+		if (drm_WARN(&xe->drm, val != 0x5A,
+			     "Tile %d: CPU memtest failed at offset 65535 (expected 0x5A, got 0x%02x)\n",
+			     id, val)) {
+			err = -EIO;
+			goto unpin;
+		}
+
+		/* 2. Non-CCS access via GPU on the last page */
+		xe_bo_lock(last_page_bo, false);
+		fence = xe_migrate_clear(tile->migrate, last_page_bo,
+					 last_page_bo->ttm.resource,
+					 XE_MIGRATE_CLEAR_FLAG_BO_DATA);
+		xe_bo_unlock(last_page_bo);
+
+		if (!IS_ERR(fence)) {
+			dma_fence_wait(fence, false);
+			dma_fence_put(fence);
+		} else {
+			err = PTR_ERR(fence);
+			goto unpin;
+		}
+
+		val = xe_map_rd(xe, &last_page_bo->vmap, 0, u8);
+		if (drm_WARN(&xe->drm, val != 0x00,
+			     "Tile %d: GPU memtest clear failed at offset 0 (expected 0x00, got 0x%02x)\n",
+			     id, val)) {
+			err = -EIO;
+			goto unpin;
+		}
+
+		if (xe_device_has_flat_ccs(xe)) {
+			struct xe_bo *scratch_bo_before;
+			struct xe_bo *scratch_bo_after;
+
+			scratch_bo_before = xe_bo_create_pin_map_novm(xe, tile, SZ_1M,
+							      ttm_bo_type_kernel,
+							      XE_BO_FLAG_VRAM_IF_DGFX(tile),
+							      false);
+			if (IS_ERR(scratch_bo_before)) {
+				err = PTR_ERR(scratch_bo_before);
+				goto unpin;
+			}
+
+			scratch_bo_after = xe_bo_create_pin_map_novm(xe, tile, SZ_1M,
+							     ttm_bo_type_kernel,
+							     XE_BO_FLAG_VRAM_IF_DGFX(tile),
+							     false);
+			if (IS_ERR(scratch_bo_after)) {
+				xe_bo_unpin_map_no_vm(scratch_bo_before);
+				err = PTR_ERR(scratch_bo_after);
+				goto unpin;
+			}
+
+			/* Save original CCS metadata for PA 0 */
+			xe_migrate_debug_ccs_overlap(tile->migrate, scratch_bo_before, false);
+
+			/* Fill last page with CPU */
+			xe_map_memset(xe, &last_page_bo->vmap, 0, 0x5A, SZ_64K);
+			xe_device_l2_flush(xe, true);
+
+			/* Use GPU to clear CCS state for PA 0 */
+			xe_map_memset(xe, &scratch_bo_after->vmap, 0, 0x00, SZ_1M);
+			xe_migrate_debug_ccs_overlap(tile->migrate, scratch_bo_after, true);
+			xe_device_l2_flush(xe, true);
+
+			/* Check if last_page_bo was corrupted by the GPU CCS clear */
+			for (i = 0; i < SZ_64K; i += 8) {
+				u64 payload = xe_map_rd(xe, &last_page_bo->vmap, i, u64);
+
+				if (payload != 0x5A5A5A5A5A5A5A5AULL) {
+					overlap = true;
+					break;
+				}
+			}
+
+			/* Restore original CCS metadata for PA 0 */
+			xe_migrate_debug_ccs_overlap(tile->migrate, scratch_bo_before, true);
+
+			xe_bo_unpin_map_no_vm(scratch_bo_before);
+			xe_bo_unpin_map_no_vm(scratch_bo_after);
+		}
+
+		if (drm_WARN(&xe->drm, overlap,
+			     "Tile %d: VRAM bounds overlap CCS region! VRAM sizing is incorrect.\n",
+			     id)) {
+			err = -EINVAL;
+			goto unpin;
+		}
+
+		drm_info(&xe->drm, "Tile %d: VRAM memtest completed.\n", id);
+
+unpin:
+		if (err)
+			break;
+	}
+
+	xe_vram_free_memtest_bos(xe);
+
+	return err;
+}
+#endif
diff --git a/drivers/gpu/drm/xe/xe_vram.h b/drivers/gpu/drm/xe/xe_vram.h
index dd1c8bf17922..0649b21c859a 100644
--- a/drivers/gpu/drm/xe/xe_vram.h
+++ b/drivers/gpu/drm/xe/xe_vram.h
@@ -23,4 +23,14 @@ resource_size_t xe_vram_region_dpa_base(const struct xe_vram_region *vram);
 resource_size_t xe_vram_region_usable_size(const struct xe_vram_region *vram);
 resource_size_t xe_vram_region_actual_physical_size(const struct xe_vram_region *vram);
 
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_MEM)
+void xe_vram_reserve_memtest_bo(struct xe_device *xe);
+void xe_vram_free_memtest_bos(struct xe_device *xe);
+int xe_vram_memtest(struct xe_device *xe);
+#else
+static inline void xe_vram_reserve_memtest_bo(struct xe_device *xe) {}
+static inline void xe_vram_free_memtest_bos(struct xe_device *xe) {}
+static inline int xe_vram_memtest(struct xe_device *xe) { return 0; }
+#endif
+
 #endif
-- 
2.55.0