[PATCH] drm/amdkfd: change error to warning message for SDMA queues creation
Eric Huang
jinhuieric.huang at amd.com
Fri May 2 19:49:18 UTC 2025
SDMA doesn't support oversubsciption, it is the user matter to create
queues over HW limit, but not supposed to be a KFD error.
Signed-off-by: Eric Huang <jinhuieric.huang at amd.com>
---
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 14 ++++++++------
.../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 11 +++++++++--
2 files changed, 17 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index c610e172a2b8..76359c6a3f3a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -1576,8 +1576,9 @@ static int allocate_sdma_queue(struct device_queue_manager *dqm,
int bit;
if (q->properties.type == KFD_QUEUE_TYPE_SDMA) {
- if (bitmap_empty(dqm->sdma_bitmap, KFD_MAX_SDMA_QUEUES)) {
- dev_err(dev, "No more SDMA queue to allocate\n");
+ if (bitmap_empty(dqm->sdma_bitmap, get_num_sdma_queues(dqm))) {
+ dev_warn(dev, "No more SDMA queue to allocate (%d total queues)\n",
+ get_num_sdma_queues(dqm));
return -ENOMEM;
}
@@ -1602,8 +1603,9 @@ static int allocate_sdma_queue(struct device_queue_manager *dqm,
q->properties.sdma_queue_id = q->sdma_id /
kfd_get_num_sdma_engines(dqm->dev);
} else if (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI) {
- if (bitmap_empty(dqm->xgmi_sdma_bitmap, KFD_MAX_SDMA_QUEUES)) {
- dev_err(dev, "No more XGMI SDMA queue to allocate\n");
+ if (bitmap_empty(dqm->xgmi_sdma_bitmap, get_num_xgmi_sdma_queues(dqm))) {
+ dev_warn(dev, "No more XGMI SDMA queue to allocate (%d total queues)\n",
+ get_num_xgmi_sdma_queues(dqm));
return -ENOMEM;
}
if (restore_sdma_id) {
@@ -1662,8 +1664,8 @@ static int allocate_sdma_queue(struct device_queue_manager *dqm,
}
if (!free_bit_found) {
- dev_err(dev, "No more SDMA queue to allocate for target ID %i\n",
- q->properties.sdma_engine_id);
+ dev_warn(dev, "No more SDMA queue to allocate for target ID %i (%d total queues)\n",
+ q->properties.sdma_engine_id, num_queues);
return -ENOMEM;
}
}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index 7eb370b68159..6d5fa57d4a23 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -451,8 +451,15 @@ int pqm_create_queue(struct process_queue_manager *pqm,
}
if (retval != 0) {
- pr_err("process pid %d DQM create queue type %d failed. ret %d\n",
- pqm->process->lead_thread->pid, type, retval);
+ if ((type == KFD_QUEUE_TYPE_SDMA ||
+ type == KFD_QUEUE_TYPE_SDMA_XGMI ||
+ type == KFD_QUEUE_TYPE_SDMA_BY_ENG_ID) &&
+ retval == -ENOMEM)
+ pr_warn("process pid %d DQM create queue type %d failed. ret %d\n",
+ pqm->process->lead_thread->pid, type, retval);
+ else
+ pr_err("process pid %d DQM create queue type %d failed. ret %d\n",
+ pqm->process->lead_thread->pid, type, retval);
goto err_create_queue;
}
--
2.34.1
More information about the amd-gfx
mailing list