Commit 33a5abd9 authored by Matthew Brost's avatar Matthew Brost
Browse files

drm/xe: Adjust long-running workload timeslices to reasonable values 



A 10ms timeslice for long-running workloads is far too long and causes
significant jitter in benchmarks when the system is shared. Adjust the
value to 5ms for preempt-fencing VMs, as the resume step there is quite
costly as memory is moved around, and set it to zero for pagefault VMs,
since switching back to pagefault mode after dma-fence mode is
relatively fast.

Also change min_run_period_ms to 'unsiged int' type rather than 's64' as
only positive values make sense.

Fixes: dd08ebf6 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: stable@vger.kernel.org
Signed-off-by: default avatarMatthew Brost <matthew.brost@intel.com>
Reviewed-by: default avatarThomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patch.msgid.link/20251212182847.1683222-2-matthew.brost@intel.com
parent e057b2d2

drivers/gpu/drm/xe/xe_vm.c

+4 −1
Original line number Diff line number Diff line
@@ -1508,7 +1508,10 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef) 
	INIT_WORK(&vm->destroy_work, vm_destroy_work_func);



	INIT_LIST_HEAD(&vm->preempt.exec_queues);

	vm->preempt.min_run_period_ms = 10;	/* FIXME: Wire up to uAPI */

	if (flags & XE_VM_FLAG_FAULT_MODE)

		vm->preempt.min_run_period_ms = 0;

	else

		vm->preempt.min_run_period_ms = 5;



	for_each_tile(tile, xe, id)

		xe_range_fence_tree_init(&vm->rftree[id]);

drivers/gpu/drm/xe/xe_vm_types.h

+1 −1
Original line number Diff line number Diff line
@@ -263,7 +263,7 @@ struct xe_vm { 
		 * @min_run_period_ms: The minimum run period before preempting

		 * an engine again

		 */

		s64 min_run_period_ms;

		unsigned int min_run_period_ms;

		/** @exec_queues: list of exec queues attached to this VM */

		struct list_head exec_queues;

		/** @num_exec_queues: number exec queues attached to this VM */