This patch makes the gart's buffer size calculation more accurate. This buffer is needed per GPU.
It takes into account maximum number of MQDs, runlist packets, kernel queues and reserves 512KB for other misc allocations.
The total size is just shy of 4MB, for 32 processes and 128 queues per process, which are the defaults for amdkfd kernel module parameters.
Reviewed-by: Alexey Skidanov Alexey.skidanov@amd.com Signed-off-by: Oded Gabbay oded.gabbay@amd.com --- drivers/gpu/drm/amd/amdkfd/kfd_device.c | 22 +++++++++++++++++++--- 1 file changed, 19 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index c1ec162..163b1a4 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -26,6 +26,7 @@ #include <linux/slab.h> #include "kfd_priv.h" #include "kfd_device_queue_manager.h" +#include "kfd_pm4_headers.h"
#define MQD_SIZE_ALIGNED 768
@@ -177,16 +178,31 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, max_num_of_queues_per_process * kfd->device_info->mqd_size_aligned;
- /* add another 512KB for all other allocations on gart */ + /* + * calculate max size of runlist packet. + * There can be only 2 packets at once + */ + size += (max_num_of_processes * sizeof(struct pm4_map_process) + + max_num_of_processes * max_num_of_queues_per_process * + sizeof(struct pm4_map_queues) + sizeof(struct pm4_runlist)) * 2; + + /* Add size of HIQ & DIQ */ + size += KFD_KERNEL_QUEUE_SIZE * 2; + + /* add another 512KB for all other allocations on gart (HPD, fences) */ size += 512 * 1024;
if (kfd2kgd->init_sa_manager(kfd->kgd, size)) { dev_err(kfd_device, - "Error initializing sa manager for device (%x:%x)\n", - kfd->pdev->vendor, kfd->pdev->device); + "Could not allocate %d bytes for device (%x:%x)\n", + size, kfd->pdev->vendor, kfd->pdev->device); goto out; }
+ dev_info(kfd_device, + "Allocated %d bytes on gart for device(%x:%x)\n", + size, kfd->pdev->vendor, kfd->pdev->device); + kfd_doorbell_init(kfd);
if (kfd_topology_add_device(kfd) != 0) {