drm/nouveau/fifo/gk104-: require explicit runlist selection for channel allocation

We didn't used to be aware that runlist/engine IDs weren't the same thing,
or that there was such variability in configuration between GPUs.

By exposing this information to a client, and giving it explicit control
of which runlist it's allocating a channel on, we're able to make better
choices.

The immediate effect of this is that on GPUs where CE0 is the "GRCE", we
will now be allocating a copy engine running asynchronously to GR for BO
migrations - as intended.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/include/nvif/cla06f.h

@@ -4,25 +4,11 @@
 
 struct kepler_channel_gpfifo_a_v0 {
 	__u8  version;
-	__u8  pad01[5];
+	__u8  pad01[1];
 	__u16 chid;
-#define NVA06F_V0_ENGINE_SW                                          0x00000001
-#define NVA06F_V0_ENGINE_GR                                          0x00000002
-#define NVA06F_V0_ENGINE_SEC                                         0x00000004
-#define NVA06F_V0_ENGINE_MSVLD                                       0x00000010
-#define NVA06F_V0_ENGINE_MSPDEC                                      0x00000020
-#define NVA06F_V0_ENGINE_MSPPP                                       0x00000040
-#define NVA06F_V0_ENGINE_MSENC                                       0x00000080
-#define NVA06F_V0_ENGINE_VIC                                         0x00000100
-#define NVA06F_V0_ENGINE_NVDEC                                       0x00000200
-#define NVA06F_V0_ENGINE_NVENC0                                      0x00000400
-#define NVA06F_V0_ENGINE_NVENC1                                      0x00000800
-#define NVA06F_V0_ENGINE_CE0                                         0x00010000
-#define NVA06F_V0_ENGINE_CE1                                         0x00020000
-#define NVA06F_V0_ENGINE_CE2                                         0x00040000
-	__u32 engines;
 	__u32 ilength;
 	__u64 ioffset;
+	__u64 runlist;
 	__u64 vmm;
 };
 

drivers/gpu/drm/nouveau/nouveau_abi16.c

@@ -23,6 +23,7 @@
 
 #include <nvif/client.h>
 #include <nvif/driver.h>
+#include <nvif/fifo.h>
 #include <nvif/ioctl.h>
 #include <nvif/class.h>
 #include <nvif/cl0002.h>
@@ -256,6 +257,7 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS)
 	struct nouveau_abi16 *abi16 = nouveau_abi16_get(file_priv);
 	struct nouveau_abi16_chan *chan;
 	struct nvif_device *device;
+	u64 engine;
 	int ret;
 
 	if (unlikely(!abi16))
@@ -268,25 +270,26 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS)
 
 	/* hack to allow channel engine type specification on kepler */
 	if (device->info.family >= NV_DEVICE_INFO_V0_KEPLER) {
-		if (init->fb_ctxdma_handle != ~0)
-			init->fb_ctxdma_handle = NVA06F_V0_ENGINE_GR;
-		else {
-			init->fb_ctxdma_handle = 0;
-#define _(A,B) if (init->tt_ctxdma_handle & (A)) init->fb_ctxdma_handle |= (B)
-			_(0x01, NVA06F_V0_ENGINE_GR);
-			_(0x02, NVA06F_V0_ENGINE_MSPDEC);
-			_(0x04, NVA06F_V0_ENGINE_MSPPP);
-			_(0x08, NVA06F_V0_ENGINE_MSVLD);
-			_(0x10, NVA06F_V0_ENGINE_CE0);
-			_(0x20, NVA06F_V0_ENGINE_CE1);
-			_(0x40, NVA06F_V0_ENGINE_MSENC);
-#undef _
+		if (init->fb_ctxdma_handle == ~0) {
+			switch (init->tt_ctxdma_handle) {
+			case 0x01: engine = NV_DEVICE_INFO_ENGINE_GR    ; break;
+			case 0x02: engine = NV_DEVICE_INFO_ENGINE_MSPDEC; break;
+			case 0x04: engine = NV_DEVICE_INFO_ENGINE_MSPPP ; break;
+			case 0x08: engine = NV_DEVICE_INFO_ENGINE_MSVLD ; break;
+			case 0x30: engine = NV_DEVICE_INFO_ENGINE_CE    ; break;
+			default:
+				return nouveau_abi16_put(abi16, -ENOSYS);
+			}
+		} else {
+			engine = NV_DEVICE_INFO_ENGINE_GR;
 		}
 
-		/* allow flips to be executed if this is a graphics channel */
+		if (engine != NV_DEVICE_INFO_ENGINE_CE)
+			engine = nvif_fifo_runlist(device, engine);
+		else
+			engine = nvif_fifo_runlist_ce(device);
+		init->fb_ctxdma_handle = engine;
 		init->tt_ctxdma_handle = 0;
-		if (init->fb_ctxdma_handle == NVA06F_V0_ENGINE_GR)
-			init->tt_ctxdma_handle = 1;
 	}
 
 	if (init->fb_ctxdma_handle == ~0 || init->tt_ctxdma_handle == ~0)

drivers/gpu/drm/nouveau/nouveau_chan.c

@@ -214,7 +214,7 @@ nouveau_channel_prep(struct nouveau_drm *drm, struct nvif_device *device,
 
 static int
 nouveau_channel_ind(struct nouveau_drm *drm, struct nvif_device *device,
-		    u32 engine, struct nouveau_channel **pchan)
+		    u64 runlist, struct nouveau_channel **pchan)
 {
 	struct nouveau_cli *cli = (void *)device->object.client;
 	static const u16 oclasses[] = { PASCAL_CHANNEL_GPFIFO_A,
@@ -245,9 +245,9 @@ nouveau_channel_ind(struct nouveau_drm *drm, struct nvif_device *device,
 	do {
 		if (oclass[0] >= KEPLER_CHANNEL_GPFIFO_A) {
 			args.kepler.version = 0;
-			args.kepler.engines = engine;
 			args.kepler.ilength = 0x02000;
 			args.kepler.ioffset = 0x10000 + chan->push.addr;
+			args.kepler.runlist = runlist;
 			args.kepler.vmm = nvif_handle(&cli->vmm.vmm.object);
 			size = sizeof(args.kepler);
 		} else

drivers/gpu/drm/nouveau/nouveau_drm.c

@@ -38,6 +38,7 @@
 #include <core/tegra.h>
 
 #include <nvif/driver.h>
+#include <nvif/fifo.h>
 
 #include <nvif/class.h>
 #include <nvif/cl0002.h>
@@ -358,13 +359,12 @@ nouveau_accel_init(struct nouveau_drm *drm)
 
 	if (device->info.family >= NV_DEVICE_INFO_V0_KEPLER) {
 		ret = nouveau_channel_new(drm, &drm->client.device,
-					  NVA06F_V0_ENGINE_CE0 |
-					  NVA06F_V0_ENGINE_CE1,
-					  0, &drm->cechan);
+					  nvif_fifo_runlist_ce(device), 0,
+					  &drm->cechan);
 		if (ret)
 			NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
 
-		arg0 = NVA06F_V0_ENGINE_GR;
+		arg0 = nvif_fifo_runlist(device, NV_DEVICE_INFO_ENGINE_GR);
 		arg1 = 1;
 	} else
 	if (device->info.chipset >= 0xa3 &&

drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c

@@ -222,62 +222,30 @@ gk104_fifo_gpfifo_func = {
 	.engine_fini = gk104_fifo_gpfifo_engine_fini,
 };
 
-struct gk104_fifo_chan_func {
-	u32 engine;
-	u64 subdev;
-};
-
 static int
-gk104_fifo_gpfifo_new_(const struct gk104_fifo_chan_func *func,
-		       struct gk104_fifo *fifo, u32 *engmask, u16 *chid,
+gk104_fifo_gpfifo_new_(struct gk104_fifo *fifo, u64 *runlists, u16 *chid,
 		       u64 vmm, u64 ioffset, u64 ilength,
 		       const struct nvkm_oclass *oclass,
 		       struct nvkm_object **pobject)
 {
 	struct gk104_fifo_chan *chan;
-	int runlist = -1, ret = -ENOSYS, i, j;
-	u32 engines = 0, present = 0;
+	int runlist = ffs(*runlists) -1, ret, i;
+	unsigned long engm;
 	u64 subdevs = 0;
 	u64 usermem;
 
-	if (!vmm)
+	if (!vmm || runlist < 0 || runlist >= fifo->runlist_nr)
 		return -EINVAL;
+	*runlists = BIT_ULL(runlist);
 
-	/* Determine which downstream engines are present */
-	for (i = 0; i < fifo->engine_nr; i++) {
-		struct nvkm_engine *engine = fifo->engine[i].engine;
-		if (engine) {
-			u64 submask = BIT_ULL(engine->subdev.index);
-			for (j = 0; func[j].subdev; j++) {
-				if (func[j].subdev & submask) {
-					present |= func[j].engine;
-					break;
-				}
-			}
-
-			if (!func[j].subdev)
-				continue;
-
-			if (runlist < 0 && (*engmask & present))
-				runlist = fifo->engine[i].runl;
-			if (runlist == fifo->engine[i].runl) {
-				engines |= func[j].engine;
-				subdevs |= func[j].subdev;
-			}
-		}
+	engm = fifo->runlist[runlist].engm;
+	for_each_set_bit(i, &engm, fifo->engine_nr) {
+		if (fifo->engine[i].engine)
+			subdevs |= BIT_ULL(fifo->engine[i].engine->subdev.index);
 	}
 
-	/* Just an engine mask query?  All done here! */
-	if (!*engmask) {
-		*engmask = present;
-		return nvkm_object_new(oclass, NULL, 0, pobject);
-	}
-
-	/* No runlist?  No supported engines. */
-	*engmask = present;
-	if (runlist < 0)
-		return -ENODEV;
-	*engmask = engines;
+	if (subdevs & BIT_ULL(NVKM_ENGINE_GR))
+		subdevs |= BIT_ULL(NVKM_ENGINE_SW);
 
 	/* Allocate the channel. */
 	if (!(chan = kzalloc(sizeof(*chan), GFP_KERNEL)))
@@ -327,26 +295,6 @@ gk104_fifo_gpfifo_new_(const struct gk104_fifo_chan_func *func,
 	return 0;
 }
 
-static const struct gk104_fifo_chan_func
-gk104_fifo_gpfifo[] = {
-	{ NVA06F_V0_ENGINE_SW | NVA06F_V0_ENGINE_GR,
-		BIT_ULL(NVKM_ENGINE_SW) | BIT_ULL(NVKM_ENGINE_GR)
-	},
-	{ NVA06F_V0_ENGINE_SEC   , BIT_ULL(NVKM_ENGINE_SEC   ) },
-	{ NVA06F_V0_ENGINE_MSVLD , BIT_ULL(NVKM_ENGINE_MSVLD ) },
-	{ NVA06F_V0_ENGINE_MSPDEC, BIT_ULL(NVKM_ENGINE_MSPDEC) },
-	{ NVA06F_V0_ENGINE_MSPPP , BIT_ULL(NVKM_ENGINE_MSPPP ) },
-	{ NVA06F_V0_ENGINE_MSENC , BIT_ULL(NVKM_ENGINE_MSENC ) },
-	{ NVA06F_V0_ENGINE_VIC   , BIT_ULL(NVKM_ENGINE_VIC   ) },
-	{ NVA06F_V0_ENGINE_NVDEC , BIT_ULL(NVKM_ENGINE_NVDEC ) },
-	{ NVA06F_V0_ENGINE_NVENC0, BIT_ULL(NVKM_ENGINE_NVENC0) },
-	{ NVA06F_V0_ENGINE_NVENC1, BIT_ULL(NVKM_ENGINE_NVENC1) },
-	{ NVA06F_V0_ENGINE_CE0   , BIT_ULL(NVKM_ENGINE_CE0   ) },
-	{ NVA06F_V0_ENGINE_CE1   , BIT_ULL(NVKM_ENGINE_CE1   ) },
-	{ NVA06F_V0_ENGINE_CE2   , BIT_ULL(NVKM_ENGINE_CE2   ) },
-	{}
-};
-
 int
 gk104_fifo_gpfifo_new(struct nvkm_fifo *base, const struct nvkm_oclass *oclass,
 		      void *data, u32 size, struct nvkm_object **pobject)
@@ -361,11 +309,12 @@ gk104_fifo_gpfifo_new(struct nvkm_fifo *base, const struct nvkm_oclass *oclass,
 	nvif_ioctl(parent, "create channel gpfifo size %d\n", size);
 	if (!(ret = nvif_unpack(ret, &data, &size, args->v0, 0, 0, false))) {
 		nvif_ioctl(parent, "create channel gpfifo vers %d vmm %llx "
-				   "ioffset %016llx ilength %08x engine %08x\n",
+				   "ioffset %016llx ilength %08x "
+				   "runlist %016llx\n",
 			   args->v0.version, args->v0.vmm, args->v0.ioffset,
-			   args->v0.ilength, args->v0.engines);
-		return gk104_fifo_gpfifo_new_(gk104_fifo_gpfifo, fifo,
-					      &args->v0.engines,
+			   args->v0.ilength, args->v0.runlist);
+		return gk104_fifo_gpfifo_new_(fifo,
+					      &args->v0.runlist,
 					      &args->v0.chid,
 					       args->v0.vmm,
 					       args->v0.ioffset,