Add a kthread to trigger pc sampling trap.

Signed-off-by: James Zhu <james....@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_pc_sampling.c | 68 +++++++++++++++++++-
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h        |  1 +
 2 files changed, 68 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_pc_sampling.c 
b/drivers/gpu/drm/amd/amdkfd/kfd_pc_sampling.c
index 49b5d4c9f7e0..04cc25c79a76 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_pc_sampling.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_pc_sampling.c
@@ -39,6 +39,66 @@ struct supported_pc_sample_info supported_formats[] = {
        { IP_VERSION(9, 4, 2), &sample_info_hosttrap_9_0_0 },
 };
 
+static int kfd_pc_sample_thread(void *param)
+{
+       struct amdgpu_device *adev;
+       struct kfd_node *node = param;
+       uint32_t timeout = 0;
+
+       mutex_lock(&node->pcs_data.mutex);
+       if (node->pcs_data.hosttrap_entry.base.active_count &&
+               node->pcs_data.hosttrap_entry.base.pc_sample_info.interval &&
+               node->kfd2kgd->trigger_pc_sample_trap) {
+               switch (node->pcs_data.hosttrap_entry.base.pc_sample_info.type) 
{
+               case KFD_IOCTL_PCS_TYPE_TIME_US:
+                       timeout = 
(uint32_t)node->pcs_data.hosttrap_entry.base.pc_sample_info.interval;
+                       break;
+               default:
+                       pr_debug("PC Sampling type %d not supported.",
+                                       
node->pcs_data.hosttrap_entry.base.pc_sample_info.type);
+               }
+       }
+       mutex_unlock(&node->pcs_data.mutex);
+       if (!timeout)
+               return -EINVAL;
+
+       adev = node->adev;
+
+       allow_signal(SIGKILL);
+       while (!kthread_should_stop() ||
+                       
!READ_ONCE(node->pcs_data.hosttrap_entry.base.stop_enable)) {
+               node->kfd2kgd->trigger_pc_sample_trap(adev, 
node->vm_info.last_vmid_kfd,
+                               &node->pcs_data.hosttrap_entry.base.target_simd,
+                               
&node->pcs_data.hosttrap_entry.base.target_wave_slot,
+                               
node->pcs_data.hosttrap_entry.base.pc_sample_info.method);
+               pr_debug_ratelimited("triggered a host trap.");
+
+               if 
(signal_pending(node->pcs_data.hosttrap_entry.base.pc_sample_thread))
+                       break;
+               usleep_range(timeout, timeout + 10);
+       }
+       node->pcs_data.hosttrap_entry.base.pc_sample_thread = NULL;
+
+       return 0;
+}
+
+static int kfd_pc_sample_thread_start(struct kfd_node *node)
+{
+       char thread_name[32];
+       int ret = 0;
+
+       snprintf(thread_name, 32, "pc_sampling_%08x", node->id);
+       node->pcs_data.hosttrap_entry.base.pc_sample_thread =
+               kthread_run(kfd_pc_sample_thread, node, thread_name);
+       if (IS_ERR(node->pcs_data.hosttrap_entry.base.pc_sample_thread)) {
+               ret = 
PTR_ERR(node->pcs_data.hosttrap_entry.base.pc_sample_thread);
+               node->pcs_data.hosttrap_entry.base.pc_sample_thread = NULL;
+               pr_debug("Failed to create pc sample thread for %s.\n", 
thread_name);
+       }
+
+       return ret;
+}
+
 static int kfd_pc_sample_query_cap(struct kfd_process_device *pdd,
                                        struct kfd_ioctl_pc_sample_args __user 
*user_args)
 {
@@ -88,6 +148,7 @@ static int kfd_pc_sample_start(struct kfd_process_device 
*pdd,
                                        struct pc_sampling_entry *pcs_entry)
 {
        bool pc_sampling_start = false;
+       int ret = 0;
 
        pcs_entry->enabled = true;
        mutex_lock(&pdd->dev->pcs_data.mutex);
@@ -102,11 +163,13 @@ static int kfd_pc_sample_start(struct kfd_process_device 
*pdd,
                } else {
                        kfd_process_set_trap_pc_sampling_flag(&pdd->qpd,
                                
pdd->dev->pcs_data.hosttrap_entry.base.pc_sample_info.method, true);
+                       if 
(!pdd->dev->pcs_data.hosttrap_entry.base.pc_sample_thread)
+                               ret = kfd_pc_sample_thread_start(pdd->dev);
                        break;
                }
        }
 
-       return 0;
+       return ret;
 }
 
 static int kfd_pc_sample_stop(struct kfd_process_device *pdd,
@@ -124,6 +187,9 @@ static int kfd_pc_sample_stop(struct kfd_process_device 
*pdd,
        mutex_unlock(&pdd->dev->pcs_data.mutex);
 
        if (pc_sampling_stop) {
+               
kthread_stop(pdd->dev->pcs_data.hosttrap_entry.base.pc_sample_thread);
+               while (pdd->dev->pcs_data.hosttrap_entry.base.pc_sample_thread)
+                       usleep_range(1000, 2000);
                kfd_process_set_trap_pc_sampling_flag(&pdd->qpd,
                        
pdd->dev->pcs_data.hosttrap_entry.base.pc_sample_info.method, false);
                remap_queue(pdd->dev->dqm,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h 
b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 0839a0ca3099..f69d76c7a388 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -276,6 +276,7 @@ struct kfd_dev_pc_sampling_data {
        uint32_t target_wave_slot;  /* target wave slot for trap */
        bool stop_enable;           /* pc sampling stop in process */
        struct idr pc_sampling_idr;
+       struct task_struct *pc_sample_thread;
        struct kfd_pc_sample_info pc_sample_info;
 };
 
-- 
2.25.1

Reply via email to