[PATCH 7.1 062/101] drm/amdgpu: Allocate coredump ring buffers per ring

7.1-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Lijo Lazar <[email protected]>

[ Upstream commit e40ff9840fa8a633d149f0242df10cae5e518062 ]

Allocate each ring buffer separately. A single allocation summing all
ring sizes can exceed the page allocator's MAX_ORDER limit and fail;
per-ring buffers stay small enough to satisfy. The existing allocation
style doesn't capture any ring data if the huge allocation fails.
Splitting into multiple allocations helps to capture as much data as
possible for the core dump.

A failed ring is left with a NULL buffer and skipped when formatting.

Fixes: eea85914d15b ("drm/amdgpu: save ring content before resetting the device")
Signed-off-by: Lijo Lazar <[email protected]>
Assisted-by: Claude Code
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
(cherry picked from commit 3e8e92b7892a6377bef86106bfff1b98cf586aee)
Cc: [email protected]
Signed-off-by: Sasha Levin <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c |   50 +++++++++++------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h |    3 -
 2 files changed, 27 insertions(+), 26 deletions(-)

--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c
@@ -339,7 +339,7 @@ amdgpu_devcoredump_format(char *buffer,
 	struct amdgpu_ip_block *ip_block;
 	struct amdgpu_ring *ring;
 	int ver, i, j;
-	u32 ring_idx, off;
+	u32 ring_idx;
 	bool sizing_pass;
 
 	sizing_pass = buffer == NULL;
@@ -439,7 +439,6 @@ amdgpu_devcoredump_format(char *buffer,
 		for (i = 0; i < coredump->num_rings; i++) {
 			ring_idx = coredump->rings[i].ring_index;
 			ring = coredump->adev->rings[ring_idx];
-			off = coredump->rings[i].offset;
 
 			drm_printf(&p, "ring name: %s\n", ring->name);
 			drm_printf(&p, "Rptr: 0x%llx Wptr: 0x%llx RB mask: %x\n",
@@ -448,12 +447,18 @@ amdgpu_devcoredump_format(char *buffer,
 				   ring->buf_mask);
 			drm_printf(&p, "Ring size in dwords: %d\n",
 				ring->ring_size / 4);
+
+			if (!coredump->rings[i].ring_dw) {
+				drm_printf(&p, "Ring contents unavailable\n");
+				continue;
+			}
+
 			drm_printf(&p, "Ring contents\n");
 			drm_printf(&p, "Offset \t Value\n");
 
 			for (j = 0; j < ring->ring_size; j += 4)
 				drm_printf(&p, "0x%x \t 0x%x\n", j,
-					   coredump->rings_dw[off + j / 4]);
+					   coredump->rings[i].ring_dw[j / 4]);
 		}
 	}
 
@@ -494,10 +499,12 @@ amdgpu_devcoredump_read(char *buffer, lo
 static void amdgpu_devcoredump_free(void *data)
 {
 	struct amdgpu_coredump_info *coredump = data;
+	u32 i;
 
 	kvfree(coredump->formatted);
+	for (i = 0; i < coredump->num_rings; i++)
+		kvfree(coredump->rings[i].ring_dw);
 	kvfree(coredump->rings);
-	kvfree(coredump->rings_dw);
 	kvfree(data);
 }
 
@@ -539,9 +546,9 @@ void amdgpu_coredump(struct amdgpu_devic
 	struct amdgpu_coredump_info *coredump;
 	size_t size = sizeof(*coredump);
 	struct drm_sched_job *s_job;
-	u64 total_ring_size, ring_count;
+	u64 ring_count;
 	struct amdgpu_ring *ring;
-	int i, off, idx;
+	int i, idx;
 
 	/* No need to generate a new coredump if there's one in progress already. */
 	if (work_busy(&adev->coredump_work))
@@ -581,7 +588,6 @@ void amdgpu_coredump(struct amdgpu_devic
 
 	/* Dump ring content if memory allocation succeeds. */
 	ring_count = 0;
-	total_ring_size = 0;
 	for (i = 0; i < adev->num_rings; i++) {
 		ring = adev->rings[i];
 
@@ -590,38 +596,34 @@ void amdgpu_coredump(struct amdgpu_devic
 		    coredump->ring != ring)
 			continue;
 
-		total_ring_size += ring->ring_size;
 		ring_count++;
 	}
-	if (ring_count) {
-		coredump->rings_dw = kvzalloc(total_ring_size, GFP_NOWAIT);
+	if (ring_count)
 		coredump->rings = kvcalloc(ring_count,
 					   sizeof(struct amdgpu_coredump_ring),
 					   GFP_NOWAIT);
-	}
-	if (coredump->rings && coredump->rings_dw) {
-		for (i = 0, off = 0, idx = 0; i < adev->num_rings && idx < ring_count; i++) {
+	if (coredump->rings) {
+		for (i = 0, idx = 0; i < adev->num_rings && idx < ring_count; i++) {
+			struct amdgpu_coredump_ring *cdump_ring;
+
 			ring = adev->rings[i];
 
 			if (atomic_read(&ring->fence_drv.last_seq) == ring->fence_drv.sync_seq &&
 			    coredump->ring != ring)
 				continue;
 
-			coredump->rings[idx].ring_index = ring->idx;
-			coredump->rings[idx].rptr = amdgpu_ring_get_rptr(ring);
-			coredump->rings[idx].wptr = amdgpu_ring_get_wptr(ring);
-			coredump->rings[idx].offset = off;
+			cdump_ring = &coredump->rings[idx];
 
-			memcpy(&coredump->rings_dw[off], ring->ring, ring->ring_size);
-			off += ring->ring_size / 4;
+			cdump_ring->ring_dw = kvzalloc(ring->ring_size, GFP_NOWAIT);
+			if (cdump_ring->ring_dw)
+				memcpy(cdump_ring->ring_dw, ring->ring, ring->ring_size);
+
+			cdump_ring->ring_index = ring->idx;
+			cdump_ring->rptr = amdgpu_ring_get_rptr(ring);
+			cdump_ring->wptr = amdgpu_ring_get_wptr(ring);
 			idx++;
 		}
 		coredump->num_rings = idx;
-	} else {
-		kvfree(coredump->rings_dw);
-		kvfree(coredump->rings);
-		coredump->rings_dw = NULL;
-		coredump->rings = NULL;
 	}
 
 	coredump->adev = adev;
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h
@@ -34,8 +34,8 @@
 struct amdgpu_coredump_ring {
 	u64				rptr;
 	u64				wptr;
+	u32				*ring_dw;
 	u32				ring_index;
-	u32				offset;
 };
 
 struct amdgpu_coredump_ib_info {
@@ -53,7 +53,6 @@ struct amdgpu_coredump_info {
 	struct amdgpu_ring              *ring;
 
 	struct amdgpu_coredump_ring	*rings;
-	u32				*rings_dw;
 	u32				num_rings;
 
 	/* Readable form of coredevdump, generate once to speed up