drm/nouveau/fault: switch non-replayable faults to nvkm_event_ntfy

v2: fix flush_work() being called uninitialised during init
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>

drivers/gpu/drm/nouveau/include/nvkm/subdev/fault.h

@@ -2,7 +2,6 @@
 #define __NVKM_FAULT_H__
 #include <core/subdev.h>
 #include <core/event.h>
-#include <core/notify.h>
 
 struct nvkm_fault {
 	const struct nvkm_fault_func *func;
@@ -11,9 +10,11 @@ struct nvkm_fault {
 	struct nvkm_fault_buffer *buffer[2];
 	int buffer_nr;
 
+#define NVKM_FAULT_BUFFER_EVENT_PENDING BIT(0)
 	struct nvkm_event event;
 
-	struct nvkm_notify nrpfb;
+	struct nvkm_event_ntfy nrpfb;
+	struct work_struct nrpfb_work;
 
 	struct nvkm_device_oclass user;
 };

drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c

@@ -145,7 +145,7 @@ nvkm_fault_dtor(struct nvkm_subdev *subdev)
 	struct nvkm_fault *fault = nvkm_fault(subdev);
 	int i;
 
-	nvkm_notify_fini(&fault->nrpfb);
+	nvkm_event_ntfy_del(&fault->nrpfb);
 	nvkm_event_fini(&fault->event);
 
 	for (i = 0; i < fault->buffer_nr; i++) {

drivers/gpu/drm/nouveau/nvkm/subdev/fault/gp100.c

@@ -65,7 +65,7 @@ gp100_fault_buffer_info(struct nvkm_fault_buffer *buffer)
 void
 gp100_fault_intr(struct nvkm_fault *fault)
 {
-	nvkm_event_send(&fault->event, 1, 0, NULL, 0);
+	nvkm_event_send(&fault->event, NVKM_FAULT_BUFFER_EVENT_PENDING, 0, NULL, 0);
 }
 
 static const struct nvkm_fault_func

drivers/gpu/drm/nouveau/nvkm/subdev/fault/gv100.c

@@ -27,10 +27,12 @@
 
 #include <nvif/class.h>
 
-static void
-gv100_fault_buffer_process(struct nvkm_fault_buffer *buffer)
+void
+gv100_fault_buffer_process(struct work_struct *work)
 {
-	struct nvkm_device *device = buffer->fault->subdev.device;
+	struct nvkm_fault *fault = container_of(work, typeof(*fault), nrpfb_work);
+	struct nvkm_fault_buffer *buffer = fault->buffer[0];
+	struct nvkm_device *device = fault->subdev.device;
 	struct nvkm_memory *mem = buffer->mem;
 	u32 get = nvkm_rd32(device, buffer->get);
 	u32 put = nvkm_rd32(device, buffer->put);
@@ -115,11 +117,12 @@ gv100_fault_buffer_info(struct nvkm_fault_buffer *buffer)
 }
 
 static int
-gv100_fault_ntfy_nrpfb(struct nvkm_notify *notify)
+gv100_fault_ntfy_nrpfb(struct nvkm_event_ntfy *ntfy, u32 bits)
 {
-	struct nvkm_fault *fault = container_of(notify, typeof(*fault), nrpfb);
-	gv100_fault_buffer_process(fault->buffer[0]);
-	return NVKM_NOTIFY_KEEP;
+	struct nvkm_fault *fault = container_of(ntfy, typeof(*fault), nrpfb);
+
+	schedule_work(&fault->nrpfb_work);
+	return NVKM_EVENT_KEEP;
 }
 
 static void
@@ -163,14 +166,14 @@ gv100_fault_intr(struct nvkm_fault *fault)
 
 	if (stat & 0x20000000) {
 		if (fault->buffer[0]) {
-			nvkm_event_send(&fault->event, 1, 0, NULL, 0);
+			nvkm_event_send(&fault->event, NVKM_FAULT_BUFFER_EVENT_PENDING, 0, NULL, 0);
 			stat &= ~0x20000000;
 		}
 	}
 
 	if (stat & 0x08000000) {
 		if (fault->buffer[1]) {
-			nvkm_event_send(&fault->event, 1, 1, NULL, 0);
+			nvkm_event_send(&fault->event, NVKM_FAULT_BUFFER_EVENT_PENDING, 1, NULL, 0);
 			stat &= ~0x08000000;
 		}
 	}
@@ -183,9 +186,12 @@ gv100_fault_intr(struct nvkm_fault *fault)
 static void
 gv100_fault_fini(struct nvkm_fault *fault)
 {
-	nvkm_notify_put(&fault->nrpfb);
+	nvkm_event_ntfy_block(&fault->nrpfb);
+	flush_work(&fault->nrpfb_work);
+
 	if (fault->buffer[0])
 		fault->func->buffer.fini(fault->buffer[0]);
+
 	nvkm_mask(fault->subdev.device, 0x100a34, 0x80000000, 0x80000000);
 }
 
@@ -194,15 +200,15 @@ gv100_fault_init(struct nvkm_fault *fault)
 {
 	nvkm_mask(fault->subdev.device, 0x100a2c, 0x80000000, 0x80000000);
 	fault->func->buffer.init(fault->buffer[0]);
-	nvkm_notify_get(&fault->nrpfb);
+	nvkm_event_ntfy_allow(&fault->nrpfb);
 }
 
 int
 gv100_fault_oneinit(struct nvkm_fault *fault)
 {
-	return nvkm_notify_init(&fault->buffer[0]->object, &fault->event,
-				gv100_fault_ntfy_nrpfb, true, NULL, 0, 0,
-				&fault->nrpfb);
+	nvkm_event_ntfy_add(&fault->event, 0, NVKM_FAULT_BUFFER_EVENT_PENDING, true,
+			    gv100_fault_ntfy_nrpfb, &fault->nrpfb);
+	return 0;
 }
 
 static const struct nvkm_fault_func
@@ -231,5 +237,10 @@ int
 gv100_fault_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
 		struct nvkm_fault **pfault)
 {
-	return nvkm_fault_new_(&gv100_fault, device, type, inst, pfault);
+	int ret = nvkm_fault_new_(&gv100_fault, device, type, inst, pfault);
+	if (ret)
+		return ret;
+
+	INIT_WORK(&(*pfault)->nrpfb_work, gv100_fault_buffer_process);
+	return 0;
 }

drivers/gpu/drm/nouveau/nvkm/subdev/fault/priv.h

@@ -46,6 +46,7 @@ void gp100_fault_buffer_fini(struct nvkm_fault_buffer *);
 void gp100_fault_buffer_init(struct nvkm_fault_buffer *);
 u64 gp100_fault_buffer_pin(struct nvkm_fault_buffer *);
 void gp100_fault_buffer_info(struct nvkm_fault_buffer *);
+void gv100_fault_buffer_process(struct work_struct *);
 void gp100_fault_intr(struct nvkm_fault *);
 
 u64 gp10b_fault_buffer_pin(struct nvkm_fault_buffer *);

drivers/gpu/drm/nouveau/nvkm/subdev/fault/tu102.c

@@ -126,7 +126,7 @@ tu102_fault_intr(struct nvkm_fault *fault)
 		nvkm_wr32(device, 0xb81010, 0x10);
 
 		if (fault->buffer[0]) {
-			nvkm_event_send(&fault->event, 1, 0, NULL, 0);
+			nvkm_event_send(&fault->event, NVKM_FAULT_BUFFER_EVENT_PENDING, 0, NULL, 0);
 			stat &= ~0x00000200;
 		}
 	}
@@ -137,7 +137,7 @@ tu102_fault_intr(struct nvkm_fault *fault)
 		nvkm_wr32(device, 0xb81008, 0x1);
 
 		if (fault->buffer[1]) {
-			nvkm_event_send(&fault->event, 1, 1, NULL, 0);
+			nvkm_event_send(&fault->event, NVKM_FAULT_BUFFER_EVENT_PENDING, 1, NULL, 0);
 			stat &= ~0x00000100;
 		}
 	}
@@ -150,7 +150,9 @@ tu102_fault_intr(struct nvkm_fault *fault)
 static void
 tu102_fault_fini(struct nvkm_fault *fault)
 {
-	nvkm_notify_put(&fault->nrpfb);
+	nvkm_event_ntfy_block(&fault->nrpfb);
+	flush_work(&fault->nrpfb_work);
+
 	if (fault->buffer[0])
 		fault->func->buffer.fini(fault->buffer[0]);
 	/*XXX: disable priv faults */
@@ -161,7 +163,7 @@ tu102_fault_init(struct nvkm_fault *fault)
 {
 	/*XXX: enable priv faults */
 	fault->func->buffer.init(fault->buffer[0]);
-	nvkm_notify_get(&fault->nrpfb);
+	nvkm_event_ntfy_allow(&fault->nrpfb);
 }
 
 static const struct nvkm_fault_func
@@ -184,5 +186,10 @@ int
 tu102_fault_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
 		struct nvkm_fault **pfault)
 {
-	return nvkm_fault_new_(&tu102_fault, device, type, inst, pfault);
+	int ret = nvkm_fault_new_(&tu102_fault, device, type, inst, pfault);
+	if (ret)
+		return ret;
+
+	INIT_WORK(&(*pfault)->nrpfb_work, gv100_fault_buffer_process);
+	return 0;
 }