drm/nouveau/fault: add explicit control over fault buffer interrupts

The GPU will continually fire interrupts while a fault buffer GET != PUT,
and to stop the spurious interrupts while the handler does its thing, we
were disabling the fault buffer temporarily.

This is not actually a great idea to begin with, and made worse by Volta
resetting GET/PUT when it's reactivated.  So, let's not do that.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

4 files changed, 29 insertions, 8 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c
index 75587cafa0ce..4ba1e21e8fda 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c
@@ -28,14 +28,14 @@ static void
 nvkm_fault_ntfy_fini(struct nvkm_event *event, int type, int index)
 {
 	struct nvkm_fault *fault = container_of(event, typeof(*fault), event);
-	fault->func->buffer.fini(fault->buffer[index]);
+	fault->func->buffer.intr(fault->buffer[index], false);
 }
 
 static void
 nvkm_fault_ntfy_init(struct nvkm_event *event, int type, int index)
 {
 	struct nvkm_fault *fault = container_of(event, typeof(*fault), event);
-	fault->func->buffer.init(fault->buffer[index]);
+	fault->func->buffer.intr(fault->buffer[index], true);
 }
 
 static int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gp100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gp100.c
index 67d587a50f14..8fb96fe614f9 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gp100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gp100.c
@@ -21,6 +21,15 @@
  */
 #include "priv.h"
 
+#include <subdev/mc.h>
+
+static void
+gp100_fault_buffer_intr(struct nvkm_fault_buffer *buffer, bool enable)
+{
+	struct nvkm_device *device = buffer->fault->subdev.device;
+	nvkm_mc_intr_mask(device, NVKM_SUBDEV_FAULT, enable);
+}
+
 static void
 gp100_fault_buffer_fini(struct nvkm_fault_buffer *buffer)
 {
@@ -59,6 +68,7 @@ gp100_fault = {
 	.buffer.info = gp100_fault_buffer_info,
 	.buffer.init = gp100_fault_buffer_init,
 	.buffer.fini = gp100_fault_buffer_fini,
+	.buffer.intr = gp100_fault_buffer_intr,
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gv100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gv100.c
index c5f735cf8eb3..5de916fc5ccd 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gv100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/gv100.c
@@ -69,13 +69,21 @@ gv100_fault_buffer_process(struct nvkm_fault_buffer *buffer)
 }
 
 static void
-gv100_fault_buffer_fini(struct nvkm_fault_buffer *buffer)
+gv100_fault_buffer_intr(struct nvkm_fault_buffer *buffer, bool enable)
 {
 	struct nvkm_device *device = buffer->fault->subdev.device;
 	const u32 intr = buffer->id ? 0x08000000 : 0x20000000;
-	const u32 foff = buffer->id * 0x14;
+	if (enable)
+		nvkm_mask(device, 0x100a2c, intr, intr);
+	else
+		nvkm_mask(device, 0x100a34, intr, intr);
+}
 
-	nvkm_mask(device, 0x100a34, intr, intr);
+static void
+gv100_fault_buffer_fini(struct nvkm_fault_buffer *buffer)
+{
+	struct nvkm_device *device = buffer->fault->subdev.device;
+	const u32 foff = buffer->id * 0x14;
 	nvkm_mask(device, 0x100e34 + foff, 0x80000000, 0x00000000);
 }
 
@@ -83,14 +91,12 @@ static void
 gv100_fault_buffer_init(struct nvkm_fault_buffer *buffer)
 {
 	struct nvkm_device *device = buffer->fault->subdev.device;
-	const u32 intr = buffer->id ? 0x08000000 : 0x20000000;
 	const u32 foff = buffer->id * 0x14;
 
 	nvkm_mask(device, 0x100e34 + foff, 0xc0000000, 0x40000000);
 	nvkm_wr32(device, 0x100e28 + foff, upper_32_bits(buffer->addr));
 	nvkm_wr32(device, 0x100e24 + foff, lower_32_bits(buffer->addr));
 	nvkm_mask(device, 0x100e34 + foff, 0x80000000, 0x80000000);
-	nvkm_mask(device, 0x100a2c, intr, intr);
 }
 
 static void
@@ -169,6 +175,8 @@ static void
 gv100_fault_fini(struct nvkm_fault *fault)
 {
 	nvkm_notify_put(&fault->nrpfb);
+	if (fault->buffer[0])
+		fault->func->buffer.fini(fault->buffer[0]);
 	nvkm_mask(fault->subdev.device, 0x100a34, 0x80000000, 0x80000000);
 }
 
@@ -176,6 +184,7 @@ static void
 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);
 }
 
@@ -183,7 +192,7 @@ static int
 gv100_fault_oneinit(struct nvkm_fault *fault)
 {
 	return nvkm_notify_init(&fault->buffer[0]->object, &fault->event,
-				gv100_fault_ntfy_nrpfb, false, NULL, 0, 0,
+				gv100_fault_ntfy_nrpfb, true, NULL, 0, 0,
 				&fault->nrpfb);
 }
 
@@ -198,6 +207,7 @@ gv100_fault = {
 	.buffer.info = gv100_fault_buffer_info,
 	.buffer.init = gv100_fault_buffer_init,
 	.buffer.fini = gv100_fault_buffer_fini,
+	.buffer.intr = gv100_fault_buffer_intr,
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/priv.h
index b9a30a6f6aaf..ec085dcb5c5c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/priv.h
@@ -32,6 +32,7 @@ struct nvkm_fault_func {
 		void (*info)(struct nvkm_fault_buffer *);
 		void (*init)(struct nvkm_fault_buffer *);
 		void (*fini)(struct nvkm_fault_buffer *);
+		void (*intr)(struct nvkm_fault_buffer *, bool enable);
 	} buffer;
 };
 #endif