Completing error notifier

nx/include/switch.h

@@ -22,6 +22,7 @@ extern "C" {
 #include "switch/kernel/tmem.h"
 #include "switch/kernel/shmem.h"
 #include "switch/kernel/mutex.h"
+#include "switch/kernel/event.h"
 #include "switch/kernel/rwlock.h"
 #include "switch/kernel/condvar.h"
 #include "switch/kernel/thread.h"

nx/include/switch/nvidia/gpu/error_notifier.h

@@ -2,6 +2,7 @@ typedef struct NvGpu NvGpu;
 
 typedef struct {
     NvGpu* parent;
+    Event  event;
     bool   has_init;
 } NvErrorNotifier;
 

nx/include/switch/nvidia/ioctl.h

@@ -140,6 +140,13 @@ typedef enum {
     NvErrorType_PbdmaPushbufferCrcMismatch=80
 } NvErrorType;
 
+typedef struct {
+    u64 tickstamp;
+    u32 error_type;
+    u16 unk16;
+    u16 status; // always -1
+} NvError;
+
 Result nvioctlNvhostCtrl_EventSignal(u32 fd, u32 event_id);
 Result nvioctlNvhostCtrl_EventWait(u32 fd, u32 syncpt_id, u32 threshold, s32 timeout, u32 event_id, u32 *out);
 Result nvioctlNvhostCtrl_EventRegister(u32 fd, u32 event_id);
@@ -166,6 +173,7 @@ Result nvioctlChannel_SubmitGpfifo(u32 fd, nvioctl_gpfifo_entry *entries, u32 nu
 Result nvioctlChannel_AllocObjCtx(u32 fd, u32 class_num, u32 flags, u64* id_out);
 Result nvioctlChannel_ZCullBind(u32 fd, u64 gpu_va, u32 mode);
 Result nvioctlChannel_SetErrorNotifier(u32 fd, u32 enable);
+Result nvioctlChannel_GetErrorNotification(u32 fd, NvError* out);
 Result nvioctlChannel_SetPriority(u32 fd, u32 priority);
 Result nvioctlChannel_AllocGpfifoEx2(u32 fd, u32 num_entries, u32 flags, u32 unk0, u32 unk1, u32 unk2, u32 unk3, nvioctl_fence *fence_out);
 Result nvioctlChannel_SetUserData(u32 fd, void* addr);

nx/include/switch/services/nv.h

@@ -10,6 +10,18 @@
 Result nvInitialize(void);
 void nvExit(void);
 
+typedef enum {
+    NvEventId_Gpu_SmException_BptIntReport=1,
+    NvEventId_Gpu_SmException_BptPauseReport=2,
+    NvEventId_Gpu_ErrorNotifier=3,
+
+    NvEventId_CtrlGpu_ErrorEventHandle=1,
+    NvEventId_CtrlGpu_Unknown=2,
+} NvEventId;
+
+#define NV_EVENT_ID_CTRL__SYNCPT(slot, syncpt) \
+    ((1u<<28) | ((syncpt) << 16) | (slot))
+
 Result nvOpen(u32 *fd, const char *devicepath);
 Result nvIoctl(u32 fd, u32 request, void* argp);
 Result nvClose(u32 fd);

nx/source/nvidia/gpu/error_notifier.c

@@ -1,9 +1,16 @@
 #include <switch.h>
 
-Result nverrCreate(NvErrorNotifier* t, NvGpu* parent) {
+Result nverrCreate(NvErrorNotifier* t, NvGpu* parent)
+{
     Result rc;
+    Handle handle;
 
-    rc = nvioctlChannel_SetErrorNotifier(parent->gpu_channel.fd, 1);
+    rc = nvQueryEvent(parent->gpu_channel.fd, NvEventId_Gpu_ErrorNotifier, &handle);
+
+    if (R_SUCCEEDED(rc)) {
+        eventLoadRemote(&t->event, handle);
+        rc = nvioctlChannel_SetErrorNotifier(parent->gpu_channel.fd, 1);
+    }
 
     if (R_SUCCEEDED(rc)) {
         t->parent = parent;
@@ -13,9 +20,19 @@ Result nverrCreate(NvErrorNotifier* t, NvGpu* parent) {
     return rc;
 }
 
-void nverrClose(NvErrorNotifier* t) {
+void nverrClose(NvErrorNotifier* t)
+{
     if (!t->has_init)
         return;
 
     nvioctlChannel_SetErrorNotifier(t->parent->gpu_channel.fd, 0);
+    eventClose(&t->event);
+}
+
+Result nverrWait(NvErrorNotifier* t, u64 timeout) {
+    return eventWait(&t->event, timeout);
+}
+
+Result nverrGetError(NvErrorNotifier* t, NvError* out) {
+    return nvioctlChannel_GetErrorNotification(t->parent->gpu_channel.fd, out);
 }

nx/source/nvidia/ioctl/nvchannel.c

@@ -100,6 +100,10 @@ Result nvioctlChannel_SetErrorNotifier(u32 fd, u32 enable) {
     return nvIoctl(fd, _NV_IOWR(0x48, 0x0C, data), &data);
 }
 
+Result nvioctlChannel_GetErrorNotification(u32 fd, NvError* out) {
+    return nvIoctl(fd, _NV_IOWR(0x48, 0x17, *out), out);
+}
+
 Result nvioctlChannel_SetPriority(u32 fd, u32 priority) {
     struct {
         __nv_in u32 priority; // 0x32 is low, 0x64 is medium and 0x96 is high