drm/nouveau/fifo: add new channel classes

Exposes a bunch of the new features that became possible as a result
of the earlier commits.  DRM will build on this in the future to add
support for features such as SCG ("async compute") and multi-device
rendering, as part of the work necessary to be able to write a half-
decent vulkan driver - finally.

For the moment, this just crudely ports DRM to the API changes.

- channel class interfaces now the same for all HW classes
- channel group class exposed (SCG)
- channel runqueue selector exposed (SCG)
- channel sub-device id control exposed (multi-device rendering)
- channel names in logging will reflect creating process, not fd owner
- explicit USERD allocation required by VOLTA_CHANNEL_GPFIFO_A and newer
- drm is smarter about determining the appropriate channel class to use

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

63 files changed, 524 insertions, 1763 deletions

diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl006b.h b/drivers/gpu/drm/nouveau/include/nvif/cl006b.h
deleted file mode 100644
index c960c449e430..000000000000
--- a/drivers/gpu/drm/nouveau/include/nvif/cl006b.h
+++ /dev/null
@@ -1,12 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-#ifndef __NVIF_CL006B_H__
-#define __NVIF_CL006B_H__
-
-struct nv03_channel_dma_v0 {
-	__u8  version;
-	__u8  chid;
-	__u8  pad02[2];
-	__u32 offset;
-	__u64 pushbuf;
-};
-#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl506e.h b/drivers/gpu/drm/nouveau/include/nvif/cl506e.h
deleted file mode 100644
index 9df289c7a84f..000000000000
--- a/drivers/gpu/drm/nouveau/include/nvif/cl506e.h
+++ /dev/null
@@ -1,13 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-#ifndef __NVIF_CL506E_H__
-#define __NVIF_CL506E_H__
-
-struct nv50_channel_dma_v0 {
-	__u8  version;
-	__u8  chid;
-	__u8  pad02[6];
-	__u64 vmm;
-	__u64 pushbuf;
-	__u64 offset;
-};
-#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl506f.h b/drivers/gpu/drm/nouveau/include/nvif/cl506f.h
deleted file mode 100644
index 327c96a994bb..000000000000
--- a/drivers/gpu/drm/nouveau/include/nvif/cl506f.h
+++ /dev/null
@@ -1,14 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-#ifndef __NVIF_CL506F_H__
-#define __NVIF_CL506F_H__
-
-struct nv50_channel_gpfifo_v0 {
-	__u8  version;
-	__u8  chid;
-	__u8  pad02[2];
-	__u32 ilength;
-	__u64 ioffset;
-	__u64 pushbuf;
-	__u64 vmm;
-};
-#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl826e.h b/drivers/gpu/drm/nouveau/include/nvif/cl826e.h
deleted file mode 100644
index ef3033b836f0..000000000000
--- a/drivers/gpu/drm/nouveau/include/nvif/cl826e.h
+++ /dev/null
@@ -1,13 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-#ifndef __NVIF_CL826E_H__
-#define __NVIF_CL826E_H__
-
-struct g82_channel_dma_v0 {
-	__u8  version;
-	__u8  chid;
-	__u8  pad02[6];
-	__u64 vmm;
-	__u64 pushbuf;
-	__u64 offset;
-};
-#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl826f.h b/drivers/gpu/drm/nouveau/include/nvif/cl826f.h
deleted file mode 100644
index c5d5e56b04cc..000000000000
--- a/drivers/gpu/drm/nouveau/include/nvif/cl826f.h
+++ /dev/null
@@ -1,14 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-#ifndef __NVIF_CL826F_H__
-#define __NVIF_CL826F_H__
-
-struct g82_channel_gpfifo_v0 {
-	__u8  version;
-	__u8  chid;
-	__u8  pad02[2];
-	__u32 ilength;
-	__u64 ioffset;
-	__u64 pushbuf;
-	__u64 vmm;
-};
-#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl906f.h b/drivers/gpu/drm/nouveau/include/nvif/cl906f.h
deleted file mode 100644
index 5ccc8fd8c458..000000000000
--- a/drivers/gpu/drm/nouveau/include/nvif/cl906f.h
+++ /dev/null
@@ -1,13 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-#ifndef __NVIF_CL906F_H__
-#define __NVIF_CL906F_H__
-
-struct fermi_channel_gpfifo_v0 {
-	__u8  version;
-	__u8  chid;
-	__u8  pad02[2];
-	__u32 ilength;
-	__u64 ioffset;
-	__u64 vmm;
-};
-#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/cla06f.h b/drivers/gpu/drm/nouveau/include/nvif/cla06f.h
deleted file mode 100644
index 10449accd3e8..000000000000
--- a/drivers/gpu/drm/nouveau/include/nvif/cla06f.h
+++ /dev/null
@@ -1,15 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-#ifndef __NVIF_CLA06F_H__
-#define __NVIF_CLA06F_H__
-
-struct kepler_channel_gpfifo_a_v0 {
-	__u8  version;
-	__u8  priv;
-	__u16 chid;
-	__u32 ilength;
-	__u64 ioffset;
-	__u64 runlist;
-	__u64 vmm;
-	__u64 inst;
-};
-#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/class.h b/drivers/gpu/drm/nouveau/include/nvif/class.h
index e9d9f314a744..ceea074b064b 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/class.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/class.h
@@ -70,23 +70,23 @@
 #define MAXWELL_FAULT_BUFFER_A                        /* clb069.h */ 0x0000b069
 #define VOLTA_FAULT_BUFFER_A                          /* clb069.h */ 0x0000c369
 
-#define NV03_CHANNEL_DMA                              /* cl506b.h */ 0x0000006b
-#define NV10_CHANNEL_DMA                              /* cl506b.h */ 0x0000006e
-#define NV17_CHANNEL_DMA                              /* cl506b.h */ 0x0000176e
-#define NV40_CHANNEL_DMA                              /* cl506b.h */ 0x0000406e
+#define NV03_CHANNEL_DMA                              /* if0020.h */ 0x0000006b
+#define NV10_CHANNEL_DMA                              /* if0020.h */ 0x0000006e
+#define NV17_CHANNEL_DMA                              /* if0020.h */ 0x0000176e
+#define NV40_CHANNEL_DMA                              /* if0020.h */ 0x0000406e
 
 #define KEPLER_CHANNEL_GROUP_A                        /* if0021.h */ 0x0000a06c
 
-#define NV50_CHANNEL_GPFIFO                           /* cl506f.h */ 0x0000506f
-#define G82_CHANNEL_GPFIFO                            /* cl826f.h */ 0x0000826f
-#define FERMI_CHANNEL_GPFIFO                          /* cl906f.h */ 0x0000906f
-#define KEPLER_CHANNEL_GPFIFO_A                       /* cla06f.h */ 0x0000a06f
-#define KEPLER_CHANNEL_GPFIFO_B                       /* cla06f.h */ 0x0000a16f
-#define MAXWELL_CHANNEL_GPFIFO_A                      /* cla06f.h */ 0x0000b06f
-#define PASCAL_CHANNEL_GPFIFO_A                       /* cla06f.h */ 0x0000c06f
-#define VOLTA_CHANNEL_GPFIFO_A                        /* clc36f.h */ 0x0000c36f
-#define TURING_CHANNEL_GPFIFO_A                       /* clc36f.h */ 0x0000c46f
-#define AMPERE_CHANNEL_GPFIFO_B                       /* clc36f.h */ 0x0000c76f
+#define NV50_CHANNEL_GPFIFO                           /* if0020.h */ 0x0000506f
+#define G82_CHANNEL_GPFIFO                            /* if0020.h */ 0x0000826f
+#define FERMI_CHANNEL_GPFIFO                          /* if0020.h */ 0x0000906f
+#define KEPLER_CHANNEL_GPFIFO_A                       /* if0020.h */ 0x0000a06f
+#define KEPLER_CHANNEL_GPFIFO_B                       /* if0020.h */ 0x0000a16f
+#define MAXWELL_CHANNEL_GPFIFO_A                      /* if0020.h */ 0x0000b06f
+#define PASCAL_CHANNEL_GPFIFO_A                       /* if0020.h */ 0x0000c06f
+#define VOLTA_CHANNEL_GPFIFO_A                        /* if0020.h */ 0x0000c36f
+#define TURING_CHANNEL_GPFIFO_A                       /* if0020.h */ 0x0000c46f
+#define AMPERE_CHANNEL_GPFIFO_B                       /* if0020.h */ 0x0000c76f
 
 #define NV50_DISP                                     /* if0010.h */ 0x00005070
 #define G82_DISP                                      /* if0010.h */ 0x00008270
diff --git a/drivers/gpu/drm/nouveau/include/nvif/clc36f.h b/drivers/gpu/drm/nouveau/include/nvif/clc36f.h
deleted file mode 100644
index cdf6708e1d9a..000000000000
--- a/drivers/gpu/drm/nouveau/include/nvif/clc36f.h
+++ /dev/null
@@ -1,16 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-#ifndef __NVIF_CLC36F_H__
-#define __NVIF_CLC36F_H__
-
-struct volta_channel_gpfifo_a_v0 {
-	__u8  version;
-	__u8  priv;
-	__u16 chid;
-	__u32 ilength;
-	__u64 ioffset;
-	__u64 runlist;
-	__u64 vmm;
-	__u64 inst;
-	__u32 token;
-};
-#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/if0020.h b/drivers/gpu/drm/nouveau/include/nvif/if0020.h
index 1893b8aa0abb..085e0ae8a450 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/if0020.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/if0020.h
@@ -2,6 +2,38 @@
 #ifndef __NVIF_IF0020_H__
 #define __NVIF_IF0020_H__
 
+union nvif_chan_args {
+	struct nvif_chan_v0 {
+		__u8  version;
+		__u8  namelen;
+		__u8  runlist;
+		__u8  runq;
+		__u8  priv;
+		__u8  pad05;
+		__u16 devm;
+		__u64 vmm;
+
+		__u64 ctxdma;
+		__u64 offset;
+		__u64 length;
+
+		__u64 huserd;
+		__u64 ouserd;
+
+		__u32 token;
+		__u16 chid;
+		__u8  pad3e;
+#define NVIF_CHAN_V0_INST_APER_VRAM 0
+#define NVIF_CHAN_V0_INST_APER_HOST 1
+#define NVIF_CHAN_V0_INST_APER_NCOH 2
+#define NVIF_CHAN_V0_INST_APER_INST 0xff
+		__u8  aper;
+		__u64 inst;
+
+		__u8  name[];
+	} v0;
+};
+
 union nvif_chan_event_args {
 	struct nvif_chan_event_v0 {
 		__u8 version;
diff --git a/drivers/gpu/drm/nouveau/include/nvif/if0021.h b/drivers/gpu/drm/nouveau/include/nvif/if0021.h
new file mode 100644
index 000000000000..5013def90455
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/include/nvif/if0021.h
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: MIT */
+#ifndef __NVIF_IF0021_H__
+#define __NVIF_IF0021_H__
+
+union nvif_cgrp_args {
+	struct nvif_cgrp_v0 {
+		__u8  version;
+		__u8  namelen;
+		__u8  runlist;
+		__u8  pad03[3];
+		__u16 cgid;
+		__u64 vmm;
+		__u8  name[];
+	} v0;
+};
+#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
index da8abcbeb965..c71f412bbb7f 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
@@ -16,7 +16,8 @@ struct nvkm_chan {
 
 	struct nvkm_gpuobj *inst;
 	struct nvkm_vmm *vmm;
-	union { int id; int chid; }; /*FIXME: remove later */
+	struct nvkm_gpuobj *push;
+	int id;
 
 	struct {
 		struct nvkm_memory *mem;
@@ -35,11 +36,7 @@ struct nvkm_chan {
 	atomic_t errored;
 
 	struct list_head cctxs;
-
-	struct nvkm_object object;
-
 	struct list_head head;
-	struct nvkm_gpuobj *push;
 };
 
 struct nvkm_chan *nvkm_chan_get_chid(struct nvkm_engine *, int id, unsigned long *irqflags);
diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c
index 9237ef328b9f..82dab51d8aeb 100644
--- a/drivers/gpu/drm/nouveau/nouveau_abi16.c
+++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c
@@ -27,7 +27,6 @@
 #include <nvif/ioctl.h>
 #include <nvif/class.h>
 #include <nvif/cl0002.h>
-#include <nvif/cla06f.h>
 #include <nvif/unpack.h>
 
 #include "nouveau_drv.h"
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index 5b7042d94e80..efd6cf46921b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -25,13 +25,7 @@
 
 #include <nvif/class.h>
 #include <nvif/cl0002.h>
-#include <nvif/cl006b.h>
-#include <nvif/cl506f.h>
-#include <nvif/cl906f.h>
-#include <nvif/cla06f.h>
-#include <nvif/clc36f.h>
 #include <nvif/if0020.h>
-#include <nvif/ioctl.h>
 
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
@@ -101,6 +95,7 @@ nouveau_channel_del(struct nouveau_channel **pchan)
 		nvif_object_dtor(&chan->vram);
 		nvif_event_dtor(&chan->kill);
 		nvif_object_dtor(&chan->user);
+		nvif_mem_dtor(&chan->mem_userd);
 		nvif_object_dtor(&chan->push.ctxdma);
 		nouveau_vma_del(&chan->push.vma);
 		nouveau_bo_unmap(chan->push.buffer);
@@ -250,134 +245,112 @@ nouveau_channel_prep(struct nouveau_drm *drm, struct nvif_device *device,
 }
 
 static int
-nouveau_channel_ind(struct nouveau_drm *drm, struct nvif_device *device,
-		    u64 runlist, bool priv, struct nouveau_channel **pchan)
+nouveau_channel_ctor(struct nouveau_drm *drm, struct nvif_device *device, bool priv, u64 runm,
+		     struct nouveau_channel **pchan)
 {
-	static const u16 oclasses[] = { AMPERE_CHANNEL_GPFIFO_B,
-					TURING_CHANNEL_GPFIFO_A,
-					VOLTA_CHANNEL_GPFIFO_A,
-					PASCAL_CHANNEL_GPFIFO_A,
-					MAXWELL_CHANNEL_GPFIFO_A,
-					KEPLER_CHANNEL_GPFIFO_B,
-					KEPLER_CHANNEL_GPFIFO_A,
-					FERMI_CHANNEL_GPFIFO,
-					G82_CHANNEL_GPFIFO,
-					NV50_CHANNEL_GPFIFO,
-					0 };
-	const u16 *oclass = oclasses;
-	union {
-		struct nv50_channel_gpfifo_v0 nv50;
-		struct fermi_channel_gpfifo_v0 fermi;
-		struct kepler_channel_gpfifo_a_v0 kepler;
-		struct volta_channel_gpfifo_a_v0 volta;
+	static const struct {
+		s32 oclass;
+		int version;
+	} hosts[] = {
+		{  AMPERE_CHANNEL_GPFIFO_B, 0 },
+		{  TURING_CHANNEL_GPFIFO_A, 0 },
+		{   VOLTA_CHANNEL_GPFIFO_A, 0 },
+		{  PASCAL_CHANNEL_GPFIFO_A, 0 },
+		{ MAXWELL_CHANNEL_GPFIFO_A, 0 },
+		{  KEPLER_CHANNEL_GPFIFO_B, 0 },
+		{  KEPLER_CHANNEL_GPFIFO_A, 0 },
+		{   FERMI_CHANNEL_GPFIFO  , 0 },
+		{     G82_CHANNEL_GPFIFO  , 0 },
+		{    NV50_CHANNEL_GPFIFO  , 0 },
+		{    NV40_CHANNEL_DMA     , 0 },
+		{    NV17_CHANNEL_DMA     , 0 },
+		{    NV10_CHANNEL_DMA     , 0 },
+		{    NV03_CHANNEL_DMA     , 0 },
+		{}
+	};
+	struct {
+		struct nvif_chan_v0 chan;
+		char name[TASK_COMM_LEN+16];
 	} args;
+	struct nouveau_cli *cli = (void *)device->object.client;
 	struct nouveau_channel *chan;
-	u32 size;
-	int ret;
+	const u64 plength = 0x10000;
+	const u64 ioffset = plength;
+	const u64 ilength = 0x02000;
+	char name[TASK_COMM_LEN];
+	int cid, ret;
+	u64 size;
+
+	cid = nvif_mclass(&device->object, hosts);
+	if (cid < 0)
+		return cid;
+
+	if (hosts[cid].oclass < NV50_CHANNEL_GPFIFO)
+		size = plength;
+	else
+		size = ioffset + ilength;
 
 	/* allocate dma push buffer */
-	ret = nouveau_channel_prep(drm, device, 0x12000, &chan);
+	ret = nouveau_channel_prep(drm, device, size, &chan);
 	*pchan = chan;
 	if (ret)
 		return ret;
 
 	/* create channel object */
-	do {
-		if (oclass[0] >= VOLTA_CHANNEL_GPFIFO_A) {
-			args.volta.version = 0;
-			args.volta.ilength = 0x02000;
-			args.volta.ioffset = 0x10000 + chan->push.addr;
-			args.volta.runlist = runlist;
-			args.volta.vmm = nvif_handle(&chan->vmm->vmm.object);
-			args.volta.priv = priv;
-			size = sizeof(args.volta);
-		} else
-		if (oclass[0] >= KEPLER_CHANNEL_GPFIFO_A) {
-			args.kepler.version = 0;
-			args.kepler.ilength = 0x02000;
-			args.kepler.ioffset = 0x10000 + chan->push.addr;
-			args.kepler.runlist = runlist;
-			args.kepler.vmm = nvif_handle(&chan->vmm->vmm.object);
-			args.kepler.priv = priv;
-			size = sizeof(args.kepler);
-		} else
-		if (oclass[0] >= FERMI_CHANNEL_GPFIFO) {
-			args.fermi.version = 0;
-			args.fermi.ilength = 0x02000;
-			args.fermi.ioffset = 0x10000 + chan->push.addr;
-			args.fermi.vmm = nvif_handle(&chan->vmm->vmm.object);
-			size = sizeof(args.fermi);
-		} else {
-			args.nv50.version = 0;
-			args.nv50.ilength = 0x02000;
-			args.nv50.ioffset = 0x10000 + chan->push.addr;
-			args.nv50.pushbuf = nvif_handle(&chan->push.ctxdma);
-			args.nv50.vmm = nvif_handle(&chan->vmm->vmm.object);
-			size = sizeof(args.nv50);
-		}
-
-		ret = nvif_object_ctor(&device->object, "abi16ChanUser", 0,
-				       *oclass++, &args, size, &chan->user);
-		if (ret == 0) {
-			if (chan->user.oclass >= VOLTA_CHANNEL_GPFIFO_A) {
-				chan->chid = args.volta.chid;
-				chan->inst = args.volta.inst;
-				chan->token = args.volta.token;
-			} else
-			if (chan->user.oclass >= KEPLER_CHANNEL_GPFIFO_A) {
-				chan->chid = args.kepler.chid;
-				chan->inst = args.kepler.inst;
-			} else
-			if (chan->user.oclass >= FERMI_CHANNEL_GPFIFO) {
-				chan->chid = args.fermi.chid;
-			} else {
-				chan->chid = args.nv50.chid;
-			}
+	args.chan.version = 0;
+	args.chan.namelen = sizeof(args.name);
+	args.chan.runlist = __ffs64(runm);
+	args.chan.runq = 0;
+	args.chan.priv = priv;
+	args.chan.devm = BIT(0);
+	if (hosts[cid].oclass < NV50_CHANNEL_GPFIFO) {
+		args.chan.vmm = 0;
+		args.chan.ctxdma = nvif_handle(&chan->push.ctxdma);
+		args.chan.offset = chan->push.addr;
+		args.chan.length = 0;
+	} else {
+		args.chan.vmm = nvif_handle(&chan->vmm->vmm.object);
+		if (hosts[cid].oclass < FERMI_CHANNEL_GPFIFO)
+			args.chan.ctxdma = nvif_handle(&chan->push.ctxdma);
+		else
+			args.chan.ctxdma = 0;
+		args.chan.offset = ioffset + chan->push.addr;
+		args.chan.length = ilength;
+	}
+	args.chan.huserd = 0;
+	args.chan.ouserd = 0;
+
+	/* allocate userd */
+	if (hosts[cid].oclass >= VOLTA_CHANNEL_GPFIFO_A) {
+		ret = nvif_mem_ctor(&cli->mmu, "abi16ChanUSERD", NVIF_CLASS_MEM_GF100,
+				    NVIF_MEM_VRAM | NVIF_MEM_COHERENT | NVIF_MEM_MAPPABLE,
+				    0, PAGE_SIZE, NULL, 0, &chan->mem_userd);
+		if (ret)
 			return ret;
-		}
-	} while (*oclass);
 
-	nouveau_channel_del(pchan);
-	return ret;
-}
+		args.chan.huserd = nvif_handle(&chan->mem_userd.object);
+		args.chan.ouserd = 0;
 
-static int
-nouveau_channel_dma(struct nouveau_drm *drm, struct nvif_device *device,
-		    struct nouveau_channel **pchan)
-{
-	static const u16 oclasses[] = { NV40_CHANNEL_DMA,
-					NV17_CHANNEL_DMA,
-					NV10_CHANNEL_DMA,
-					NV03_CHANNEL_DMA,
-					0 };
-	const u16 *oclass = oclasses;
-	struct nv03_channel_dma_v0 args;
-	struct nouveau_channel *chan;
-	int ret;
+		chan->userd = &chan->mem_userd.object;