drm/nouveau/fifo: kill channel on NV_PPBDMA_INTR_1_CTXNOTVALID

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>

drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c

@@ -164,6 +164,12 @@ gk104_runq_intr_1(struct nvkm_runq *runq)
 	u32 chid = nvkm_rd32(device, 0x040120 + (runq->id * 0x2000)) & 0xfff;
 	char msg[128];
 
+	if (stat & 0x80000000) {
+		if (runq->func->intr_1_ctxnotvalid &&
+		    runq->func->intr_1_ctxnotvalid(runq, chid))
+			stat &= ~0x80000000;
+	}
+
 	if (stat) {
 		nvkm_snprintbf(msg, sizeof(msg), gk104_runq_intr_1_names, stat);
 		nvkm_error(subdev, "PBDMA%d: %08x [%s] ch %d %08x %08x\n",

drivers/gpu/drm/nouveau/nvkm/engine/fifo/gv100.c

@@ -55,11 +55,34 @@ const struct nvkm_engn_func
 gv100_engn_ce = {
 };
 
+static bool
+gv100_runq_intr_1_ctxnotvalid(struct nvkm_runq *runq, int chid)
+{
+	struct nvkm_fifo *fifo = runq->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_chan *chan;
+	unsigned long flags;
+
+	RUNQ_ERROR(runq, "CTXNOTVALID chid:%d", chid);
+
+	chan = nvkm_chan_get_chid(&fifo->engine, chid, &flags);
+	if (WARN_ON_ONCE(!chan))
+		return false;
+
+	nvkm_chan_error(chan, true);
+	nvkm_chan_put(&chan, flags);
+
+	nvkm_mask(device, 0x0400ac + (runq->id * 0x2000), 0x00030000, 0x00030000);
+	nvkm_wr32(device, 0x040148 + (runq->id * 0x2000), 0x80000000);
+	return true;
+}
+
 const struct nvkm_runq_func
 gv100_runq = {
 	.init = gk208_runq_init,
 	.intr = gk104_runq_intr,
 	.intr_0_names = gk104_runq_intr_0_names,
+	.intr_1_ctxnotvalid = gv100_runq_intr_1_ctxnotvalid,
 };
 
 void

drivers/gpu/drm/nouveau/nvkm/engine/fifo/runq.h

@@ -9,6 +9,7 @@ struct nvkm_runq {
 		void (*init)(struct nvkm_runq *);
 		bool (*intr)(struct nvkm_runq *, struct nvkm_runl *);
 		const struct nvkm_bitfield *intr_0_names;
+		bool (*intr_1_ctxnotvalid)(struct nvkm_runq *, int chid);
 	} *func;
 	struct nvkm_fifo *fifo;
 	int id;