Le 17/02/2026 à 17:20, Alex Deucher a écrit :
On Wed, Feb 11, 2026 at 6:07 AM Pierre-Eric Pelloux-Prayer
<[email protected]> wrote:
Now that we have a worker thread, we can try to access the
IBs of the job. The process is:
* get the VM from the PASID
* get the BO from its VA and the VM
* map the BO for CPU access
* copy everything, then add it to the dump
Each step can fail so we have to be cautious.
These operations can be slow so when amdgpu_devcoredump_format
is called only to determine the size of the buffer we skip all
of them and assume they will succeed.
Signed-off-by: Pierre-Eric Pelloux-Prayer <[email protected]>
---
.../gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c | 84 ++++++++++++++++++-
1 file changed, 83 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c
index d0af8a294abf..d576518c212d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c
@@ -200,12 +200,20 @@ static void amdgpu_devcoredump_fw_info(struct
amdgpu_device *adev,
static ssize_t
amdgpu_devcoredump_format(char *buffer, size_t count, struct
amdgpu_coredump_info *coredump)
{
+ struct amdgpu_device *adev = coredump->adev;
struct drm_printer p;
struct drm_print_iterator iter;
struct amdgpu_vm_fault_info *fault_info;
+ struct amdgpu_bo_va_mapping *mapping;
struct amdgpu_ip_block *ip_block;
+ struct amdgpu_res_cursor cursor;
+ struct amdgpu_bo *abo, *root;
+ uint64_t va_start, offset;
struct amdgpu_ring *ring;
- int ver, i, j;
+ struct amdgpu_vm *vm;
+ u32 *ib_content;
+ uint8_t *kptr;
+ int ver, i, j, r;
u32 ring_idx, off;
iter.data = buffer;
@@ -323,6 +331,80 @@ amdgpu_devcoredump_format(char *buffer, size_t count,
struct amdgpu_coredump_inf
else if (coredump->reset_vram_lost)
drm_printf(&p, "VRAM is lost due to GPU reset!\n");
+ if (coredump->num_ibs) {
+ if (buffer)
+ vm = amdgpu_vm_lock_by_pasid(adev, &root,
coredump->pasid);
+ else
+ vm = NULL;
Is there any point in doing the loop if the vm is NULL?
Yes : when doing the first pass to size the final buffer I skip the
buffers mapping + read operations that might be slow and instead just
account for the outputting of ib_size_dw dwords.
+
+ for (int i = 0; i < coredump->num_ibs && (!buffer || vm); i++) {
+ ib_content =
kvmalloc_array(coredump->ibs[i].ib_size_dw, 4,
+ GFP_KERNEL);
Shouldn't this be GFP_NOWAIT?
This is executed by a worker so GFP_KERNEL should be ok?
Pierre-Eric
Alex
+ if (!ib_content)
+ continue;
+
+ if (!vm)
+ goto output_ib_content;
+
+ va_start = coredump->ibs[i].gpu_addr &
AMDGPU_GMC_HOLE_MASK;
+ mapping = amdgpu_vm_bo_lookup_mapping(vm, va_start /
AMDGPU_GPU_PAGE_SIZE);
+ if (!mapping)
+ goto free_ib_content;
+
+ offset = va_start - (mapping->start *
AMDGPU_GPU_PAGE_SIZE);
+ abo = amdgpu_bo_ref(mapping->bo_va->base.bo);
+ r = amdgpu_bo_reserve(abo, false);
+ if (r)
+ goto free_ib_content;
+
+ if (abo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) {
+ off = 0;
+
+ if (abo->tbo.resource->mem_type != TTM_PL_VRAM)
+ goto unreserve_abo;
+
+ amdgpu_res_first(abo->tbo.resource, offset,
+ coredump->ibs[i].ib_size_dw *
4,
+ &cursor);
+ while (cursor.remaining) {
+ amdgpu_device_mm_access(adev,
cursor.start / 4,
+
&ib_content[off], cursor.size / 4,
+ false);
+ off += cursor.size;
+ amdgpu_res_next(&cursor, cursor.size);
+ }
+ } else {
+ r = ttm_bo_kmap(&abo->tbo, 0,
+ PFN_UP(abo->tbo.base.size),
+ &abo->kmap);
+ if (r)
+ goto unreserve_abo;
+
+ kptr = amdgpu_bo_kptr(abo);
+ kptr += offset;
+ memcpy(ib_content, kptr,
+ coredump->ibs[i].ib_size_dw * 4);
+
+ amdgpu_bo_kunmap(abo);
+ }
+
+output_ib_content:
+ drm_printf(&p, "\nIB #%d 0x%llx %d dw\n",
+ i, coredump->ibs[i].gpu_addr,
coredump->ibs[i].ib_size_dw);
+ for (int j = 0; j < coredump->ibs[i].ib_size_dw; j++)
+ drm_printf(&p, "0x%08x\n", ib_content[j]);
+unreserve_abo:
+ if (vm)
+ amdgpu_bo_unreserve(abo);
+free_ib_content:
+ kfree(ib_content);
+ }
+ if (vm) {
+ amdgpu_bo_unreserve(root);
+ amdgpu_bo_unref(&root);
+ }
+ }
+
return count - iter.remain;
}
--
2.43.0
