On 27/12/2024 11:11, [email protected] wrote:
From: Apoorva Singh <[email protected]>

Ensure GEM_BUG_ON verifies at least one engine supports migration.
Additionally, it prevents passing an uninitialized variable to
intel_context_create().

Under what circumstances can this happen ie. does it need a Fixes: tag?

Also consider that if it can happen in production on some real hardware then GEM_BUG_ON does not prevent anything (GEM_BUG_ON is compiled out). On debug builds it also strictly doesn't prevent anything but crashes the machine.

Consider this as an alternative (if the error scenario is real):

diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c
index 6f7af4077135..83e6c20cb750 100644
--- a/drivers/gpu/drm/i915/gt/intel_migrate.c
+++ b/drivers/gpu/drm/i915/gt/intel_migrate.c
@@ -296,7 +296,9 @@ static struct intel_context *__migrate_engines(struct intel_gt *gt)
                        engines[count++] = engine;
        }

-       return intel_context_create(engines[random_index(count)]);
+       return count ?
+              intel_context_create(engines[random_index(count)]) :
+              ERR_PTR(-ENOENT);
 }

struct intel_context *intel_migrate_create_context(struct intel_migrate *m)

But check if the whole error propagation chain is fine to handle it.

Regards,

Tvrtko

Cc: Chris Wilson <[email protected]>
Signed-off-by: Apoorva Singh <[email protected]>
---
  drivers/gpu/drm/i915/gt/intel_migrate.c | 3 +++
  1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c 
b/drivers/gpu/drm/i915/gt/intel_migrate.c
index 6f7af4077135..528a7cba3623 100644
--- a/drivers/gpu/drm/i915/gt/intel_migrate.c
+++ b/drivers/gpu/drm/i915/gt/intel_migrate.c
@@ -296,6 +296,9 @@ static struct intel_context *__migrate_engines(struct 
intel_gt *gt)
                        engines[count++] = engine;
        }
+ /* At least one engine must support migration! */
+       GEM_BUG_ON(!count);
+
        return intel_context_create(engines[random_index(count)]);
  }

Reply via email to