Update KFD SMI event notification handling

Event bitmask in KFD SMI event is now replaced with event index in
the SMI event message. Sending a event bitmask, which was a 64-bit
field with only 1 bit set, was quite wasteful of memory and also
potentially limiting to 64 events. Instead the kernel would send
event index in the SMI event message. As a result, update the
KFD SMI event handling to expect the event index in the message.

Change-Id: I3e74620788d3c1f7c0bdaa69e9d9ab3d1aba2c92

include/rocm_smi/kfd_ioctl.h

@@ -541,8 +541,12 @@ struct kfd_ioctl_import_dmabuf_args {
 /*
  * KFD SMI(System Management Interface) events
  */
-/* Event type (defined by bitmask) */
+enum kfd_smi_event {
-#define KFD_SMI_EVENT_VMFAULT     0x0000000000000001
+  KFD_SMI_EVENT_NONE = 0, /* not used */
+  KFD_SMI_EVENT_VMFAULT = 1, /* event start counting at 1 */
+};
+
+#define KFD_SMI_EVENT_MASK_FROM_INDEX(i) (1ULL << ((i) - 1))
 
 struct kfd_ioctl_smi_events_args {
   __u32 gpuid;    /* to KFD */

include/rocm_smi/rocm_smi.h

@@ -284,6 +284,11 @@ typedef enum {
   RSMI_EVT_NOTIF_LAST = RSMI_EVT_NOTIF_VMFAULT
 } rsmi_evt_notification_type_t;
 
+/*
+ * Macro to generate event bitmask from event id
+ */
+#define RSMI_EVENT_MASK_FROM_INDEX(i) (1ULL << ((i) - 1))
+
 //! Maximum number of characters an event notification message will be
 #define MAX_EVENT_NOTIFICATION_MSG_SIZE 64
 
@@ -3217,8 +3222,14 @@ rsmi_event_notification_init(uint32_t dv_ind);
  * @param dv_ind a device index corresponding to the device on which to
  * listen for events
  *
- * @param mask 0 or more elements of ::rsmi_evt_notification_type_t OR'd
+ * @param mask Bitmask generated by OR'ing 1 or more elements of
- * together that indicate which event types to listen for.
+ * ::rsmi_evt_notification_type_t indicating which event types to listen for,
+ * where the rsmi_evt_notification_type_t value indicates the bit field, with
+ * bit position starting from 1.
+ * For example, if the mask field is 0x0000000000000003, which means first bit,
+ * bit 1 (bit position start from 1) and bit 2 are set, which indicate interest
+ * in receiving RSMI_EVT_NOTIF_VMFAULT (which has a value of 1) and the next
+ * event in that list which has a value of 2.
  *
  * @retval ::RSMI_STATUS_INIT_ERROR is returned if
  * ::rsmi_event_notification_init() has not been called before a call to this

src/rocm_smi.cc

@@ -3595,8 +3595,8 @@ rsmi_event_notification_get(int timeout_ms,
       data_item =
            reinterpret_cast<rsmi_evt_notification_data_t *>(&data[*num_elem]);
 
-      uint64_t event;
+      uint32_t event;
-      while (fscanf(anon_fp, "%lx %63s\n", &event,
+      while (fscanf(anon_fp, "%x %63s\n", &event,
                           reinterpret_cast<char *>(&data_item->message)) == 2) {
         /* Output is in format as "event information\n"
          * Both event are expressed in hex.

tests/rocm_smi_test/functional/evt_notif_read_write.cc

@@ -101,11 +101,11 @@ void TestEvtNotifReadWrite::Run(void) {
   }
 
   rsmi_evt_notification_type_t evt_type = RSMI_EVT_NOTIF_FIRST;
-  uint64_t mask = evt_type;
+  uint64_t mask = RSMI_EVENT_MASK_FROM_INDEX(evt_type);
   while (evt_type != RSMI_EVT_NOTIF_LAST) {
-    mask |= evt_type;
+    mask |= RSMI_EVENT_MASK_FROM_INDEX(evt_type);
     evt_type = static_cast<rsmi_evt_notification_type_t>(
-                                           static_cast<uint64_t>(evt_type)*2);
+                                           static_cast<uint32_t>(evt_type)+1);
   }
 
   for (dv_ind = 0; dv_ind < num_monitor_devs(); ++dv_ind) {