drm/amdkfd: Print a warning when the runlist becomes oversubscribed

Oversubscription of queues or processes results in poor performance
mostly because HWS blinbly schedules busy and idle queues, resulting
in poor occupancy if many queues are idle.

Let users know with a warning message when transitioning from a
non-oversubscribed to an oversubscribed runlist.

Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oak Zeng <Oak.Zeng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
index c72c8f5..ccf6b23 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
@@ -203,11 +203,15 @@ static int pm_create_runlist_ib(struct packet_manager *pm,
 
        pr_debug("Finished map process and queues to runlist\n");
 
-       if (is_over_subscription)
+       if (is_over_subscription) {
+               if (!pm->is_over_subscription)
+                       pr_warn("Runlist is getting oversubscribed. Expect reduced ROCm performance.\n");
                retval = pm->pmf->runlist(pm, &rl_buffer[rl_wptr],
                                        *rl_gpu_addr,
                                        alloc_size_bytes / sizeof(uint32_t),
                                        true);
+       }
+       pm->is_over_subscription = is_over_subscription;
 
        for (i = 0; i < alloc_size_bytes / sizeof(uint32_t); i++)
                pr_debug("0x%2X ", rl_buffer[i]);

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index f88c0d7..08a0feb 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -937,6 +937,7 @@ struct packet_manager {
        bool allocated;
        struct kfd_mem_obj *ib_buffer_obj;
        unsigned int ib_size_bytes;
+       bool is_over_subscription;
 
        const struct packet_manager_funcs *pmf;
 };