Add support for GPU reset SMI events

Add handling for both pre GPU reset and post GPU reset SMI events. Change-Id: I64d5e006bef58cb28b1c580c75f482a4590427da
ROCm · Oct 22, 2020 · 99a335f · 99a335f
1 parent 0dec51f
commit 99a335f
Show file tree

Hide file tree

Showing 3 changed files with 7 additions and 1 deletion.
diff --git a/include/rocm_smi/kfd_ioctl.h b/include/rocm_smi/kfd_ioctl.h
@@ -545,6 +545,8 @@ enum kfd_smi_event {
  KFD_SMI_EVENT_NONE = 0, /* not used */
  KFD_SMI_EVENT_VMFAULT = 1, /* event start counting at 1 */
  KFD_SMI_EVENT_THERMAL_THROTTLE = 2,
+ KFD_SMI_EVENT_GPU_PRE_RESET = 3,
+ KFD_SMI_EVENT_GPU_POST_RESET = 4,
 };
 
 #define KFD_SMI_EVENT_MASK_FROM_INDEX(i) (1ULL << ((i) - 1))

diff --git a/include/rocm_smi/rocm_smi.h b/include/rocm_smi/rocm_smi.h
@@ -281,8 +281,10 @@ typedef enum {
  RSMI_EVT_NOTIF_VMFAULT = KFD_SMI_EVENT_VMFAULT, //!< VM page fault
  RSMI_EVT_NOTIF_FIRST = RSMI_EVT_NOTIF_VMFAULT,
  RSMI_EVT_NOTIF_THERMAL_THROTTLE = KFD_SMI_EVENT_THERMAL_THROTTLE,
+ RSMI_EVT_NOTIF_GPU_PRE_RESET = KFD_SMI_EVENT_GPU_PRE_RESET,
+ RSMI_EVT_NOTIF_GPU_POST_RESET = KFD_SMI_EVENT_GPU_POST_RESET,
 
- RSMI_EVT_NOTIF_LAST = RSMI_EVT_NOTIF_THERMAL_THROTTLE
+ RSMI_EVT_NOTIF_LAST = RSMI_EVT_NOTIF_GPU_POST_RESET
 } rsmi_evt_notification_type_t;
 
 /*

diff --git a/tests/rocm_smi_test/test_utils.cc b/tests/rocm_smi_test/test_utils.cc
@@ -81,6 +81,8 @@ static const std::map<rsmi_evt_notification_type_t, const char *>
  kEvtNotifEvntNameMap = {
  {RSMI_EVT_NOTIF_VMFAULT, "RSMI_EVT_NOTIF_VMFAULT"},
  {RSMI_EVT_NOTIF_THERMAL_THROTTLE, "RSMI_EVT_NOTIF_THERMAL_THROTTLE"},
+ {RSMI_EVT_NOTIF_GPU_PRE_RESET, "RSMI_EVT_NOTIF_GPU_PRE_RESET"},
+ {RSMI_EVT_NOTIF_GPU_POST_RESET, "RSMI_EVT_NOTIF_GPU_POST_RESET"},
 };
 const char *
 NameFromEvtNotifType(rsmi_evt_notification_type_t evt) {