[Intel-xe] [PATCH 3/4] drm/xe: Size GT device memory correctly

[Michael J. Ruhl]

The current method of sizing GT device memory is not quite right.

Update the algorithm to use the relevant HW information and offsets
to set up the sizing correctly.

Update the stolen memory sizing to reflect the changes, and to be
GT specific.

Reviewed-by: Matthew Auld <matthew.auld at intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl at intel.com>
---
 drivers/gpu/drm/xe/xe_device_types.h |  2 +
 drivers/gpu/drm/xe/xe_gt_types.h     |  2 +
 drivers/gpu/drm/xe/xe_mmio.c         | 93 ++++++++++++++--------------
 3 files changed, 51 insertions(+), 46 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
index a53b44d14b25..51c16d207780 100644
--- a/drivers/gpu/drm/xe/xe_device_types.h
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -207,6 +207,8 @@ struct xe_device {
 			resource_size_t io_size;
 			/** @size: Total size of VRAM */
 			resource_size_t size;
+			/** @base: Offset to apply for Device Physical Address control */
+			resource_size_t base;
 			/** @mapping: pointer to VRAM mappable space */
 			void *__iomem mapping;
 		} vram;
diff --git a/drivers/gpu/drm/xe/xe_gt_types.h b/drivers/gpu/drm/xe/xe_gt_types.h
index 4b66e059ba9b..47f059bb8c6d 100644
--- a/drivers/gpu/drm/xe/xe_gt_types.h
+++ b/drivers/gpu/drm/xe/xe_gt_types.h
@@ -155,6 +155,8 @@ struct xe_gt {
 			 * the first 256M). This configuration is known as small-bar.
 			 */
 			resource_size_t io_size;
+			/** @base: offset of VRAM starting base */
+			resource_size_t base;
 			/** @size: size of VRAM. */
 			resource_size_t size;
 			/** @mapping: pointer to VRAM mappable space */
diff --git a/drivers/gpu/drm/xe/xe_mmio.c b/drivers/gpu/drm/xe/xe_mmio.c
index d04fb84e84ff..25ac84d72c7e 100644
--- a/drivers/gpu/drm/xe/xe_mmio.c
+++ b/drivers/gpu/drm/xe/xe_mmio.c
@@ -182,6 +182,8 @@ static int xe_determine_lmem_bar_size(struct xe_device *xe)
 	if (!xe->mem.vram.io_size)
 		return -EIO;
 
+	xe->mem.vram.base = 0; /* DPA offset */
+
 	/* set up a map to the total memory area. */
 	xe->mem.vram.mapping = ioremap_wc(xe->mem.vram.io_start, xe->mem.vram.io_size);
 
@@ -243,6 +245,9 @@ int xe_mmio_tile_vram_size(struct xe_gt *gt, u64 *vram_size, u64 *tile_size, u64
 
 int xe_mmio_probe_vram(struct xe_device *xe)
 {
+	resource_size_t io_size;
+	u64 available_size = 0;
+	u64 total_size = 0;
 	struct xe_gt *gt;
 	u64 tile_offset;
 	u64 tile_size;
@@ -268,64 +273,60 @@ int xe_mmio_probe_vram(struct xe_device *xe)
 		drm_warn(&xe->drm, "Restricting VRAM size to PCI resource size (0x%llx->0x%llx)\n",
 			 vram_size, (u64)xe->mem.vram.io_size);
 
-	/* Limit size to available memory to account for the current memory algorithm */
-	xe->mem.vram.io_size = min_t(u64, xe->mem.vram.io_size, vram_size);
-	xe->mem.vram.size = xe->mem.vram.io_size;
-
 	drm_info(&xe->drm, "VISIBLE VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
 		 &xe->mem.vram.io_size);
 
-	/* FIXME: Assuming equally partitioned VRAM, incorrect */
-	if (xe->info.tile_count > 1) {
-		u8 adj_tile_count = xe->info.tile_count;
-		resource_size_t size, io_start, io_size;
+	io_size = xe->mem.vram.io_size;
 
-		for_each_gt(gt, xe, id)
-			if (xe_gt_is_media_type(gt))
-				--adj_tile_count;
+	/* gt specific ranges */
+	for_each_gt(gt, xe, id) {
+		if (xe_gt_is_media_type(gt))
+			continue;
 
-		XE_BUG_ON(!adj_tile_count);
+		err = xe_mmio_tile_vram_size(gt, &vram_size, &tile_size, &tile_offset);
+		if (err)
+			return err;
 
-		size = xe->mem.vram.size / adj_tile_count;
-		io_start = xe->mem.vram.io_start;
-		io_size = xe->mem.vram.io_size;
+		gt->mem.vram.io_start = xe->mem.vram.io_start + tile_offset;
+		gt->mem.vram.io_size = min_t(u64, vram_size, io_size);
 
-		for_each_gt(gt, xe, id) {
-			if (id && !xe_gt_is_media_type(gt)) {
-				io_size -= min(io_size, size);
-				io_start += io_size;
-			}
+		if (!gt->mem.vram.io_size) {
+			drm_err(&xe->drm, "Tile without any CPU visible VRAM. Aborting.\n");
+			return -ENODEV;
+		}
 
-			gt->mem.vram.size = size;
-
-			/*
-			 * XXX: multi-tile small-bar might be wild. Hopefully
-			 * full tile without any mappable vram is not something
-			 * we care about.
-			 */
-
-			gt->mem.vram.io_size = min(size, io_size);
-			if (io_size) {
-				gt->mem.vram.io_start = io_start;
-				gt->mem.vram.mapping = xe->mem.vram.mapping +
-					(io_start - xe->mem.vram.io_start);
-			} else {
-				drm_err(&xe->drm, "Tile without any CPU visible VRAM. Aborting.\n");
-				return -ENODEV;
-			}
+		gt->mem.vram.base = tile_offset;
+
+		/* small bar can limit the visible size.  size accordingly */
+		gt->mem.vram.size = min_t(u64, vram_size, io_size);
+		gt->mem.vram.mapping = xe->mem.vram.mapping + tile_offset;
 
-			drm_info(&xe->drm, "VRAM[%u, %u]: %pa, %pa\n",
-				 id, gt->info.vram_id, &gt->mem.vram.io_start,
-				 &gt->mem.vram.size);
+		drm_info(&xe->drm, "VRAM[%u, %u]: %pa, %pa\n", id, gt->info.vram_id,
+			 &gt->mem.vram.io_start, &gt->mem.vram.size);
+
+		if (gt->mem.vram.io_size < gt->mem.vram.size)
+			drm_info(&xe->drm, "VRAM[%u, %u]: CPU access limited to %pa\n", id,
+				 gt->info.vram_id, &gt->mem.vram.io_size);
+
+		/* calculate total size using tile size to get the correct HW sizing */
+		total_size += tile_size;
+		available_size += vram_size;
+
+		if (total_size > xe->mem.vram.io_size) {
+			drm_warn(&xe->drm, "VRAM: %pa is larger than resource %pa\n",
+				 &total_size, &xe->mem.vram.io_size);
 		}
-	} else {
-		gt->mem.vram.size = xe->mem.vram.size;
-		gt->mem.vram.io_start = xe->mem.vram.io_start;
-		gt->mem.vram.io_size = xe->mem.vram.io_size;
-		gt->mem.vram.mapping = xe->mem.vram.mapping;
 
-		drm_info(&xe->drm, "VRAM: %pa\n", &gt->mem.vram.size);
+		io_size -= min_t(u64, tile_size, io_size);
 	}
+
+	xe->mem.vram.size = total_size;
+
+	drm_info(&xe->drm, "Total VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
+		 &xe->mem.vram.size);
+	drm_info(&xe->drm, "Available VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
+		 &available_size);
+
 	return 0;
 }
 
-- 
2.39.2