drm/nouveau/fifo/ga100-: initial support

- replaces the hacked-up version that existed solely to support TTM

v2. remove earlier hack preventing use of non-stall intr for fences

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

13 files changed, 610 insertions, 276 deletions

diff --git a/drivers/gpu/drm/nouveau/include/nvif/class.h b/drivers/gpu/drm/nouveau/include/nvif/class.h
index 36c59c7afd95..15529d2c85f9 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/class.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/class.h
@@ -86,6 +86,7 @@
 #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_A                       /* if0020.h */ 0x0000c56f
 #define AMPERE_CHANNEL_GPFIFO_B                       /* if0020.h */ 0x0000c76f
 
 #define NV50_DISP                                     /* if0010.h */ 0x00005070
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
index c71f412bbb7f..4825403a2688 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
@@ -56,6 +56,7 @@ struct nvkm_fifo {
 	struct {
 #define NVKM_FIFO_NONSTALL_EVENT BIT(0)
 		struct nvkm_event event;
+		struct nvkm_inth intr;
 	} nonstall;
 
 	struct {
@@ -92,5 +93,6 @@ int gm200_fifo_new(struct nvkm_device *, enum nvkm_subdev_type, int inst, struct
 int gp100_fifo_new(struct nvkm_device *, enum nvkm_subdev_type, int inst, struct nvkm_fifo **);
 int gv100_fifo_new(struct nvkm_device *, enum nvkm_subdev_type, int inst, struct nvkm_fifo **);
 int tu102_fifo_new(struct nvkm_device *, enum nvkm_subdev_type, int inst, struct nvkm_fifo **);
+int ga100_fifo_new(struct nvkm_device *, enum nvkm_subdev_type, int inst, struct nvkm_fifo **);
 int ga102_fifo_new(struct nvkm_device *, enum nvkm_subdev_type, int inst, struct nvkm_fifo **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index efd6cf46921b..e648ecd0c1a0 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -253,6 +253,7 @@ nouveau_channel_ctor(struct nouveau_drm *drm, struct nvif_device *device, bool p
 		int version;
 	} hosts[] = {
 		{  AMPERE_CHANNEL_GPFIFO_B, 0 },
+		{  AMPERE_CHANNEL_GPFIFO_A, 0 },
 		{  TURING_CHANNEL_GPFIFO_A, 0 },
 		{   VOLTA_CHANNEL_GPFIFO_A, 0 },
 		{  PASCAL_CHANNEL_GPFIFO_A, 0 },
@@ -365,8 +366,7 @@ nouveau_channel_init(struct nouveau_channel *chan, u32 vram, u32 gart)
 	if (ret)
 		return ret;
 
-	if (chan->user.oclass >= FERMI_CHANNEL_GPFIFO &&
-	    chan->user.oclass < AMPERE_CHANNEL_GPFIFO_B) {
+	if (chan->user.oclass >= FERMI_CHANNEL_GPFIFO) {
 		struct {
 			struct nvif_event_v0 base;
 			struct nvif_chan_event_v0 host;
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index 61be972f8385..a19f18b251f3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -348,9 +348,6 @@ nouveau_accel_gr_init(struct nouveau_drm *drm)
 	u64 runm;
 	int ret;
 
-	if (device->info.family >= NV_DEVICE_INFO_V0_AMPERE)
-		return;
-
 	/* Allocate channel that has access to the graphics engine. */
 	runm = nvif_fifo_runlist(device, NV_DEVICE_HOST_RUNLIST_ENGINES_GR);
 	if (!runm) {
@@ -473,6 +470,7 @@ nouveau_accel_init(struct nouveau_drm *drm)
 		case PASCAL_CHANNEL_GPFIFO_A:
 		case VOLTA_CHANNEL_GPFIFO_A:
 		case TURING_CHANNEL_GPFIFO_A:
+		case AMPERE_CHANNEL_GPFIFO_A:
 		case AMPERE_CHANNEL_GPFIFO_B:
 			ret = nvc0_fence_create(drm);
 			break;
diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c
index a01fd934c85e..812b8c62eeba 100644
--- a/drivers/gpu/drm/nouveau/nv84_fence.c
+++ b/drivers/gpu/drm/nouveau/nv84_fence.c
@@ -210,7 +210,7 @@ nv84_fence_create(struct nouveau_drm *drm)
 	priv->base.context_new = nv84_fence_context_new;
 	priv->base.context_del = nv84_fence_context_del;
 
-	priv->base.uevent = drm->client.device.info.family < NV_DEVICE_INFO_V0_AMPERE;
+	priv->base.uevent = true;
 
 	mutex_init(&priv->mutex);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
index 490fac3a4a3a..852c18aec4cd 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
@@ -2590,6 +2590,7 @@ nv170_chipset = {
 	.top      = { 0x00000001, ga100_top_new },
 	.vfn      = { 0x00000001, ga100_vfn_new },
 	.ce       = { 0x000003ff, ga100_ce_new },
+	.fifo     = { 0x00000001, ga100_fifo_new },
 };
 
 static const struct nvkm_device_chip
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild
index f9aad4cf8271..5a074b9970ab 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild
@@ -23,6 +23,7 @@ nvkm-y += nvkm/engine/fifo/gm200.o
 nvkm-y += nvkm/engine/fifo/gp100.o
 nvkm-y += nvkm/engine/fifo/gv100.o
 nvkm-y += nvkm/engine/fifo/tu102.o
+nvkm-y += nvkm/engine/fifo/ga100.o
 nvkm-y += nvkm/engine/fifo/ga102.o
 
 nvkm-y += nvkm/engine/fifo/ucgrp.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
index 61534a8a33d7..abce1ecfd4d4 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
@@ -124,6 +124,7 @@ nvkm_fifo_init(struct nvkm_engine *engine)
 {
 	struct nvkm_fifo *fifo = nvkm_fifo(engine);
 	struct nvkm_runq *runq;
+	struct nvkm_runl *runl;
 	u32 mask = 0;
 
 	if (fifo->func->init_pbdmas) {
@@ -136,7 +137,13 @@ nvkm_fifo_init(struct nvkm_engine *engine)
 			runq->func->init(runq);
 	}
 
-	fifo->func->init(fifo);
+	nvkm_runl_foreach(runl, fifo) {
+		if (runl->func->init)
+			runl->func->init(runl);
+	}
+
+	if (fifo->func->init)
+		fifo->func->init(fifo);
 
 	nvkm_inth_allow(&fifo->engine.subdev.inth);
 	return 0;
@@ -243,7 +250,7 @@ nvkm_fifo_oneinit(struct nvkm_engine *engine)
 		return ret;
 
 	nvkm_runl_foreach(runl, fifo) {
-		RUNL_DEBUG(runl, "");
+		RUNL_DEBUG(runl, "chan:%06x", runl->chan);
 		nvkm_runl_foreach_engn(engn, runl) {
 			ENGN_DEBUG(engn, "");
 		}
@@ -259,6 +266,14 @@ nvkm_fifo_oneinit(struct nvkm_engine *engine)
 		}
 	}
 
+	/* Initialise non-stall intr handling. */
+	if (fifo->func->nonstall_ctor) {
+		ret = fifo->func->nonstall_ctor(fifo);
+		if (ret) {
+			nvkm_error(subdev, "nonstall %d\n", ret);
+		}
+	}
+
 	/* Allocate USERD + BAR1 polling area. */
 	if (fifo->func->chan.func->userd->bar == 1) {
 		struct nvkm_vmm *bar1 = nvkm_bar_bar1_vmm(device);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c
new file mode 100644
index 000000000000..12a5d99d5e77
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga100.c
@@ -0,0 +1,550 @@
+/*
+ * Copyright 2021 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+#include "cgrp.h"
+#include "chan.h"
+#include "chid.h"
+#include "runl.h"
+#include "runq.h"
+
+#include <core/gpuobj.h>
+#include <subdev/top.h>
+#include <subdev/vfn.h>
+
+#include <nvif/class.h>
+
+/*TODO: allocate? */
+#define GA100_FIFO_NONSTALL_VECTOR 0
+
+static u32
+ga100_chan_doorbell_handle(struct nvkm_chan *chan)
+{
+	return (chan->cgrp->runl->doorbell << 16) | chan->id;
+}
+
+static void
+ga100_chan_stop(struct nvkm_chan *chan)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+
+	nvkm_wr32(runl->fifo->engine.subdev.device, runl->chan + (chan->id * 4), 0x00000003);
+}
+
+static void
+ga100_chan_start(struct nvkm_chan *chan)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	const int gfid = 0;
+
+	nvkm_wr32(device, runl->chan + (chan->id * 4), 0x00000002);
+	nvkm_wr32(device, runl->addr + 0x0090, (gfid << 16) | chan->id); /* INTERNAL_DOORBELL. */
+}
+
+static void
+ga100_chan_unbind(struct nvkm_chan *chan)
+{
+	struct nvkm_runl *runl = chan->cgrp->runl;
+
+	nvkm_wr32(runl->fifo->engine.subdev.device, runl->chan + (chan->id * 4), 0xffffffff);
+}
+
+static int
+ga100_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, bool priv)
+{
+	const u32 limit2 = ilog2(length / 8);
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x010, 0x0000face);
+	nvkm_wo32(chan->inst, 0x030, 0x7ffff902);
+	nvkm_wo32(chan->inst, 0x048, lower_32_bits(offset));
+	nvkm_wo32(chan->inst, 0x04c, upper_32_bits(offset) | (limit2 << 16));
+	nvkm_wo32(chan->inst, 0x084, 0x20400000);
+	nvkm_wo32(chan->inst, 0x094, 0x30000000 | devm);
+	nvkm_wo32(chan->inst, 0x0e4, priv ? 0x00000020 : 0x00000000);
+	nvkm_wo32(chan->inst, 0x0e8, chan->id);
+	nvkm_wo32(chan->inst, 0x0f4, 0x00001000 | (priv ? 0x00000100 : 0x00000000));
+	nvkm_wo32(chan->inst, 0x0f8, 0x80000000 | GA100_FIFO_NONSTALL_VECTOR);
+	nvkm_mo32(chan->inst, 0x218, 0x00000000, 0x00000000);
+	nvkm_done(chan->inst);
+	return 0;
+}
+
+static const struct nvkm_chan_func_ramfc
+ga100_chan_ramfc = {
+	.write = ga100_chan_ramfc_write,
+	.devm = 0xfff,
+	.priv = true,
+};
+
+const struct nvkm_chan_func
+ga100_chan = {
+	.inst = &gf100_chan_inst,
+	.userd = &gv100_chan_userd,
+	.ramfc = &ga100_chan_ramfc,
+	.unbind = ga100_chan_unbind,
+	.start = ga100_chan_start,
+	.stop = ga100_chan_stop,
+	.preempt = gk110_chan_preempt,
+	.doorbell_handle = ga100_chan_doorbell_handle,
+};
+
+static void
+ga100_cgrp_preempt(struct nvkm_cgrp *cgrp)
+{
+	struct nvkm_runl *runl = cgrp->runl;
+
+	nvkm_wr32(runl->fifo->engine.subdev.device, runl->addr + 0x098, 0x01000000 | cgrp->id);
+}
+
+const struct nvkm_cgrp_func
+ga100_cgrp = {
+	.preempt = ga100_cgrp_preempt,
+};
+
+static int
+ga100_engn_cxid(struct nvkm_engn *engn, bool *cgid)
+{
+	struct nvkm_runl *runl = engn->runl;
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u32 stat = nvkm_rd32(device, runl->addr + 0x200 + engn->id * 0x40);
+
+	ENGN_DEBUG(engn, "status %08x", stat);
+	*cgid = true;
+
+	switch ((stat & 0x0000e000) >> 13) {
+	case 0 /* INVALID */: return -ENODEV;
+	case 1 /*   VALID */:
+	case 5 /*    SAVE */: return (stat & 0x00000fff);
+	case 6 /*    LOAD */: return (stat & 0x0fff0000) >> 16;
+	case 7 /*  SWITCH */:
+		if (nvkm_engine_chsw_load(engn->engine))
+			return (stat & 0x0fff0000) >> 16;
+		return (stat & 0x00000fff);
+	default:
+		WARN_ON(1);
+		break;
+	}
+
+	return -ENODEV;
+}
+
+const struct nvkm_engn_func
+ga100_engn = {
+	.cxid = ga100_engn_cxid,
+	.ctor = gk104_ectx_ctor,
+	.bind = gv100_ectx_bind,
+};
+
+const struct nvkm_engn_func
+ga100_engn_ce = {
+	.cxid = ga100_engn_cxid,
+	.ctor = gv100_ectx_ce_ctor,
+	.bind = gv100_ectx_ce_bind,
+};
+
+static bool
+ga100_runq_idle(struct nvkm_runq *runq)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+
+	return !(nvkm_rd32(device, 0x04015c + (runq->id * 0x800)) & 0x0000e000);
+}
+
+static bool
+ga100_runq_intr_1(struct nvkm_runq *runq, struct nvkm_runl *runl)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+	u32 inte = nvkm_rd32(device, 0x040180 + (runq->id * 0x800));
+	u32 intr = nvkm_rd32(device, 0x040148 + (runq->id * 0x800));
+	u32 stat = intr & inte;
+
+	if (!stat) {
+		RUNQ_DEBUG(runq, "inte1 %08x %08x", intr, inte);
+		return false;
+	}
+
+	if (stat & 0x80000000) {
+		u32 chid = nvkm_rd32(device, 0x040120 + (runq->id * 0x0800)) & runl->chid->mask;
+		struct nvkm_chan *chan;
+		unsigned long flags;
+
+		RUNQ_ERROR(runq, "CTXNOTVALID chid:%d", chid);
+		chan = nvkm_runl_chan_get_chid(runl, chid, &flags);
+		if (chan) {
+			nvkm_chan_error(chan, true);
+			nvkm_chan_put(&chan, flags);
+		}
+
+		nvkm_mask(device, 0x0400ac + (runq->id * 0x800), 0x00030000, 0x00030000);
+		stat &= ~0x80000000;
+	}
+
+	if (stat) {
+		RUNQ_ERROR(runq, "intr1 %08x", stat);
+		nvkm_wr32(device, 0x0401a0 + (runq->id * 0x800), stat);
+	}
+
+	nvkm_wr32(device, 0x040148 + (runq->id * 0x800), intr);
+	return true;
+}
+
+static bool
+ga100_runq_intr_0(struct nvkm_runq *runq, struct nvkm_runl *runl)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+	u32 inte = nvkm_rd32(device, 0x040170 + (runq->id * 0x800));
+	u32 intr = nvkm_rd32(device, 0x040108 + (runq->id * 0x800));
+	u32 stat = intr & inte;
+
+	if (!stat) {
+		RUNQ_DEBUG(runq, "inte0 %08x %08x", intr, inte);
+		return false;
+	}
+
+	/*TODO: expand on this when fixing up gf100's version. */
+	if (stat & 0xc6afe000) {
+		u32 chid = nvkm_rd32(device, 0x040120 + (runq->id * 0x0800)) & runl->chid->mask;
+		struct nvkm_chan *chan;
+		unsigned long flags;
+
+		RUNQ_ERROR(runq, "intr0 %08x", stat);
+		chan = nvkm_runl_chan_get_chid(runl, chid, &flags);
+		if (chan) {
+			nvkm_chan_error(chan, true);
+			nvkm_chan_put(&chan, flags);
+		}
+
+		stat &= ~0xc6afe000;
+	}
+
+	if (stat) {
+		RUNQ_ERROR(runq, "intr0 %08x", stat);
+		nvkm_wr32(device, 0x040190 + (runq->id * 0x800), stat);
+	}
+
+	nvkm_wr32(device, 0x040108 + (runq->id * 0x800), intr);
+	return true;
+}
+
+static bool
+ga100_runq_intr(struct nvkm_runq *runq, struct nvkm_runl *runl)
+{
+	bool intr0 = ga100_runq_intr_0(runq, runl);
+	bool intr1 = ga100_runq_intr_1(runq, runl);
+
+	return intr0 || intr1;
+}
+
+static void
+ga100_runq_init(struct nvkm_runq *runq)
+{
+	struct nvkm_device *device = runq->fifo->engine.subdev.device;
+
+	nvkm_wr32(device, 0x040108 + (runq->id * 0x800), 0xffffffff); /* INTR_0 */
+	nvkm_wr32(device, 0x040148 + (runq->id * 0x800), 0xffffffff); /* INTR_1 */
+	nvkm_wr32(device, 0x040170 + (runq->id * 0x800), 0xffffffff); /* INTR_0_EN_SET_TREE */
+	nvkm_wr32(device, 0x040180 + (runq->id * 0x800), 0xffffffff); /* INTR_1_EN_SET_TREE */
+}
+
+const struct nvkm_runq_func
+ga100_runq = {
+	.init = ga100_runq_init,
+	.intr = ga100_runq_intr,
+	.idle = ga100_runq_idle,
+};
+
+static bool
+ga100_runl_preempt_pending(struct nvkm_runl *runl)
+{
+	return nvkm_rd32(runl->fifo->engine.subdev.device, runl->addr + 0x098) & 0x00100000;
+}
+
+static void
+ga100_runl_preempt(struct nvkm_runl *runl)
+{
+	nvkm_wr32(runl->fifo->engine.subdev.device, runl->addr + 0x098, 0x00000000);
+}
+
+static void
+ga100_runl_allow(struct nvkm_runl *runl, u32 engm)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, runl->addr + 0x094, 0x00000001, 0x00000000);
+}
+
+static void
+ga100_runl_block(struct nvkm_runl *runl, u32 engm)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, runl->addr + 0x094, 0x00000001, 0x00000001);
+}
+
+static bool
+ga100_runl_pending(struct nvkm_runl *runl)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+
+	return nvkm_rd32(device, runl->addr + 0x08c) & 0x00008000;
+}
+
+static void
+ga100_runl_commit(struct nvkm_runl *runl, struct nvkm_memory *memory, u32 start, int count)
+{
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u64 addr = nvkm_memory_addr(memory) + start;
+
+	nvkm_wr32(device, runl->addr + 0x080, lower_32_bits(addr));
+	nvkm_wr32(device, runl->addr + 0x084, upper_32_bits(addr));
+	nvkm_wr32(device, runl->addr + 0x088, count);
+}
+
+static irqreturn_t
+ga100_runl_intr(struct nvkm_inth *inth)
+{
+	struct nvkm_runl *runl = container_of(inth, typeof(*runl), inth);
+	struct nvkm_engn *engn;
+	struct nvkm_device *device = runl->fifo->engine.subdev.device;
+	u32 inte = nvkm_rd32(device, runl->addr + 0x120);
+	u32 intr = nvkm_rd32(device, runl->addr + 0x100);
+	u32 stat = intr & inte;
+	u32 info;
+
+	if (!stat) {
+		RUNL_DEBUG(runl, "inte %08x %08x", intr, inte);
+		return IRQ_NONE;
+	}
+
+	if (stat & 0x00000007) {
+		nvkm_runl_foreach_engn_cond(engn, runl, stat & BIT(engn->id)) {
+			info = nvkm_rd32(device, runl->addr + 0x224 + (engn->id * 0x40));
+
+			tu102_fifo_intr_ctxsw_timeout_info(engn, info);
+
+			nvkm_wr32(device, runl->addr + 0x100, BIT(engn->id));
+			stat &= ~BIT(engn->id);
+		}
+	}
+
+	if (stat & 0x00000300) {
+		nvkm_wr32(device, runl->addr + 0x100, stat & 0x00000300);
+		stat &= ~0x00000300;
+	}
+
+	if (stat & 0x00010000) {
+		if (runl->runq[0]) {
+			if (runl->runq[0]->func->intr(runl->runq[0], runl))
+				stat &= ~0x00010000;
+		}
+	}
+
+	if (stat & 0x00020000) {
+		if (runl->runq[1]) {
+			if (runl->runq[1]->func->intr(runl->runq[1], runl))
+				stat &= ~0x00020000;
+		}
+	}
+
+	if (stat) {
+		RUNL_ERROR(runl, "intr %08x", stat);
+		nvkm_wr32(device, runl->addr + 0x140, stat);
+	}
+
+	nvkm_wr32(device, runl->addr + 0x180, 0x00000001);
+	return IRQ_HANDLED;
+}
+
+static void
+ga100_runl_fini(struct nvkm_runl *runl)
+{
+	nvkm_mask(runl->fifo->engine.subdev.device, runl->addr + 0x300, 0x80000000, 0x00000000);
+	nvkm_inth_block(&runl->inth);
+}
+
+static void
+ga100_runl_init(struct nvkm_runl *runl)
+{
+	struct nvkm_fifo *fifo = runl->fifo;
+	struct nvkm_runq *runq;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	int i;
+
+	/* Submit NULL runlist and preempt. */
+	nvkm_wr32(device, runl->addr + 0x088, 0x00000000);
+	runl->func->preempt(runl);
+
+	/* Enable doorbell. */
+	nvkm_mask(device, runl->addr + 0x300, 0x80000000, 0x80000000);
+
+	nvkm_wr32(device, runl->addr + 0x100, 0xffffffff); /* INTR_0 */
+	nvkm_wr32(device, runl->addr + 0x140, 0xffffffff); /* INTR_0_EN_CLEAR_TREE(0) */
+	nvkm_wr32(device, runl->addr + 0x120, 0x000f1307); /* INTR_0_EN_SET_TREE(0) */
+	nvkm_wr32(device, runl->addr + 0x148, 0xffffffff); /* INTR_0_EN_CLEAR_TREE(1) */
+	nvkm_wr32(device, runl->addr + 0x128, 0x00000000); /* INTR_0_EN_SET_TREE(1) */
+
+	/* Init PBDMA(s). */
+	for (i = 0; i < runl->runq_nr; i++) {
+		runq = runl->runq[i];
+		runq->func->init(runq);
+	}
+
+	nvkm_inth_allow(&runl->inth);
+}
+
+const struct nvkm_runl_func
+ga100_runl = {
+	.init = ga100_runl_init,
+	.fini = ga100_runl_fini,
+	.size = 16,
+	.update = nv50_runl_update,
+	.insert_cgrp = gv100_runl_insert_cgrp,
+	.insert_chan = gv100_runl_insert_chan,
+	.commit = ga100_runl_commit,
+	.wait = nv50_runl_wait,
+	.pending = ga100_runl_pending,
+	.block = ga100_runl_block,
+	.allow = ga100_runl_allow,
+	.preempt = ga100_runl_preempt,
+	.preempt_pending = ga100_runl_preempt_pending,
+};
+
+static int
+ga100_runl_new(struct nvkm_fifo *fifo, int id, u32 addr, struct nvkm_runl **prunl)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_runl *runl;
+	u32 chcfg  = nvkm_rd32(device, addr + 0x004);
+	u32 chnum  = 1 << (chcfg & 0x0000000f);
+	u32 chaddr = (chcfg & 0xfffffff0);
+	u32 dbcfg  = nvkm_rd32(device, addr + 0x008);
+	u32 vector = nvkm_rd32(device, addr + 0x160);
+	int i, ret;
+
+	runl = *prunl = nvkm_runl_new(fifo, id, addr, chnum);
+	if (IS_ERR(runl))
+		return PTR_ERR(runl);
+
+	for (i = 0; i < 2; i++) {
+		u32 pbcfg = nvkm_rd32(device, addr + 0x010 + (i * 0x04));
+		if (pbcfg & 0x80000000) {
+			runl->runq[runl->runq_nr] =
+				nvkm_runq_new(fifo, ((pbcfg & 0x03fffc00) - 0x040000) / 0x800);
+			if (!runl->runq[runl->runq_nr])
+				return -ENOMEM;
+
+			runl->runq_nr++;
+		}
+	}
+
+	ret = nvkm_inth_add(&device->vfn->intr, vector & 0x00000fff, NVKM_INTR_PRIO_NORMAL,
+			    &fifo->engine.subdev, ga100_runl_intr, &runl->inth);
+	if (ret)
+		return ret;
+
+	runl->chan = chaddr;
+	runl->doorbell = dbcfg >> 16;
+	return 0;
+}
+
+static irqreturn_t
+ga100_fifo_nonstall_intr(struct nvkm_inth *inth)
+{
+	struct nvkm_fifo *fifo = container_of(inth, typeof(*fifo), nonstall.intr);
+
+	nvkm_event_ntfy(&fifo->nonstall.event, 0, NVKM_FIFO_NONSTALL_EVENT);
+	return IRQ_HANDLED;
+}
+
+static void
+ga100_fifo_nonstall_block(struct nvkm_event *event, int type, int index)
+{
+	struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event);
+
+	nvkm_inth_block(&fifo->nonstall.intr);
+}
+
+static void
+ga100_fifo_nonstall_allow(struct nvkm_event *event, int type, int index)
+{
+	struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), nonstall.event);
+
+	nvkm_inth_allow(&fifo->nonstall.intr);
+}
+
+const struct nvkm_event_func
+ga100_fifo_nonstall = {
+	.init = ga100_fifo_nonstall_allow,
+	.fini = ga100_fifo_nonstall_block,
+};
+
+int
+ga100_fifo_nonstall_ctor(struct nvkm_fifo *fifo)
+{
+	return nvkm_inth_add(&fifo->engine.subdev.device->vfn->intr, GA100_FIFO_NONSTALL_VECTOR,
+			     NVKM_INTR_PRIO_NORMAL, &fifo->engine.subdev, ga100_fifo_nonstall_intr,
+			     &fifo->nonstall.intr);
+}
+
+int
+ga100_fifo_runl_ctor(struct nvkm_fifo *fifo)
+{
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_top_device *tdev;
+	struct nvkm_runl *runl;
+	int id = 0, ret;
+
+	nvkm_list_foreach(tdev, &device->top->device, head, tdev->runlist >= 0) {
+		runl = nvkm_runl_get(fifo, -1, tdev->runlist);
+		if (!runl) {
+			ret = ga100_runl_new(fifo, id++, tdev->runlist, &runl);
+			if (ret)
+				return ret;
+		}
+
+		if (tdev->engine < 0)
+			continue;
+
+		nvkm_runl_add(runl, tdev->engine, (tdev->type == NVKM_ENGINE_CE) ?
+			      fifo->func->engn_ce : fifo->func->engn, tdev->type, tdev->inst);
+	}
+
+	return 0;
+}
+
+static const struct nvkm_fifo_func
+ga100_fifo = {
+	.runl_ctor = ga100_fifo_runl_ctor,
+	.mmu_fault = &tu102_fifo_mmu_fault,
+	.nonstall_ctor = ga100_fifo_nonstall_ctor,
+	.nonstall = &ga100_fifo_nonstall,
+	.runl = &ga100_runl,
+	.runq = &ga100_runq,
+	.engn = &ga100_engn,
+	.engn_ce = &ga100_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &ga100_cgrp, .force = true },
+	.chan = {{ 0, 0, AMPERE_CHANNEL_GPFIFO_A }, &ga100_chan },
+};
+
+int
+ga100_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+	       struct nvkm_fifo **pfifo)
+{
+	return nvkm_fifo_new_(&ga100_fifo, device, type, inst, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c
index b0c7b562bfc8..2cdf5da339b6 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/ga102.c
@@ -19,281 +19,27 @@
  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  * OTHER DEALINGS IN THE SOFTWARE.
  */
-#define ga102_fifo(p) container_of((p), struct ga102_fifo, base.engine)
-#define ga102_chan(p) container_of((p), struct ga102_chan, object)
-#include <engine/fifo.h>
+#include "priv.h"
 
-#include <core/memory.h>
-#include <subdev/mmu.h>
-#include <subdev/timer.h>
-#include <subdev/top.h>
-
-#include <nvif/cl0080.h>
-#include <nvif/if0020.h>
 #include <nvif/class.h>
 
-struct ga102_fifo {
-	struct nvkm_fifo base;
-};
-
-struct ga102_chan {
-	struct nvkm_object object;
-
-	struct {
-		u32 runl;
-		u32 chan;
-	} ctrl;
-
-	struct nvkm_memory *mthd;
-	struct nvkm_memory *inst;
-	struct nvkm_memory *user;
-	struct nvkm_memory *runl;
-
-	struct nvkm_vmm *vmm;
-};
-
-static int
-ga102_chan_sclass(struct nvkm_object *object, int index, struct nvkm_oclass *oclass)
-{
-	if (index == 0) {
-		oclass->ctor = nvkm_object_new;
-		oclass->base = (struct nvkm_sclass) { -1, -1, AMPERE_DMA_COPY_B };
-		return 0;
-	}
-
-	return -EINVAL;
-}
-
-static int
-ga102_chan_map(struct nvkm_object *object, void *argv, u32 argc,
-	       enum nvkm_object_map *type, u64 *addr, u64 *size)
-{
-	struct ga102_chan *chan = ga102_chan(object);
-	struct nvkm_device *device = chan->object.engine->subdev.device;
-	u64 bar2 = nvkm_memory_bar2(chan->user);
-
-	if (bar2 == ~0ULL)
-		return -EFAULT;
-
-	*type = NVKM_OBJECT_MAP_IO;
-	*addr = device->func->resource_addr(device, 3) + bar2;
-	*size = 0x1000;
-	return 0;
-}
-
-static int
-ga102_chan_fini(struct nvkm_object *object, bool suspend)
-{
-	struct ga102_chan *chan = ga102_chan(object);
-	struct nvkm_device *device = chan->object.engine->subdev.device;
-
-	nvkm_wr32(device, chan->ctrl.chan, 0x00000003);
-
-	nvkm_wr32(device, chan->ctrl.runl + 0x098, 0x01000000);
-	nvkm_msec(device, 2000,
-		if (!(nvkm_rd32(device, chan->ctrl.runl + 0x098) & 0x00100000))
-			break;
-	);
-
-	nvkm_wr32(device, chan->ctrl.runl + 0x088, 0);
-
-	nvkm_wr32(device, chan->ctrl.chan, 0xffffffff);
-	return 0;
-}
-
-static int
-ga102_chan_init(struct nvkm_object *object)
-{
-	struct ga102_chan *chan = ga102_chan(object);
-	struct nvkm_device *device = chan->object.engine->subdev.device;
-
-	nvkm_mask(device, chan->ctrl.runl + 0x300, 0x80000000, 0x80000000);
-
-	nvkm_wr32(device, chan->ctrl.runl + 0x080, lower_32_bits(nvkm_memory_addr(chan->runl)));
-	nvkm_wr32(device, chan->ctrl.runl + 0x084, upper_32_bits(nvkm_memory_addr(chan->runl)));
-	nvkm_wr32(device, chan->ctrl.runl + 0x088, 2);
-
-	nvkm_wr32(device, chan->ctrl.chan, 0x00000002);
-	nvkm_wr32(device, chan->ctrl.runl + 0x0090, 0);
-	return 0;
-}
-
-static void *
-ga102_chan_dtor(struct nvkm_object *object)
-{
-	struct ga102_chan *chan = ga102_chan(object);
-
-	if (chan->vmm) {
-		nvkm_vmm_part(chan->vmm, chan->inst);
-		nvkm_vmm_unref(&chan->vmm);
-	}
-
-	nvkm_memory_unref(&chan->runl);
-	nvkm_memory_unref(&chan->user);
-	nvkm_memory_unref(&chan->inst);
-	nvkm_memory_unref(&chan->mthd);
-	return chan;
-}
-
-static const struct nvkm_object_func
-ga102_chan = {
-	.dtor = ga102_chan_dtor,
-	.init = ga102_chan_init,
-	.fini = ga102_chan_fini,
-	.map = ga102_chan_map,
-	.sclass = ga102_chan_sclass,
-};
-
-static int
-ga102_chan_new(struct nvkm_device *device,
-	       const struct nvkm_oclass *oclass, void *argv, u32 argc, struct nvkm_object **pobject)
-{
-	struct nvif_chan_v0 *args = argv;
-	struct nvkm_top_device *tdev;
-	struct nvkm_vmm *vmm;
-	struct nvkm_memory *userd;
-	struct ga102_chan *chan;
-	int ret;
-
-	if (argc != sizeof(*args))
-		return -ENOSYS;
-
-	vmm = nvkm_uvmm_search(oclass->client, args->vmm);
-	if (IS_ERR(vmm))
-		return PTR_ERR(vmm);
-
-	if (!(chan = kzalloc(sizeof(*chan), GFP_KERNEL)))
-		return -ENOMEM;
-
-	nvkm_object_ctor(&ga102_chan, oclass, &chan->object);
-	*pobject = &chan->object;
-
-	list_for_each_entry(tdev, &device->top->device, head) {
-		if (tdev->type == NVKM_ENGINE_CE) {
-			chan->ctrl.runl = tdev->runlist;
-			break;
-		}
-	}
-
-	if (!chan->ctrl.runl)
-		return -ENODEV;
-
-	chan->ctrl.chan = nvkm_rd32(device, chan->ctrl.runl + 0x004) & 0xfffffff0;
-
-	args->chid = 0;
-	args->inst = 0;
-	args->token = nvkm_rd32(device, chan->ctrl.runl + 0x008) & 0xffff0000;
-
-	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x1000, 0x1000, true, &chan->mthd);
-	if (ret)
-		return ret;
-
-	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x1000, 0x1000, true, &chan->inst);
-	if (ret)
-		return ret;
-
-	nvkm_kmap(chan->inst);
-	nvkm_wo32(chan->inst, 0x010, 0x0000face);
-	nvkm_wo32(chan->inst, 0x030, 0x7ffff902);
-	nvkm_wo32(chan->inst, 0x048, lower_32_bits(args->offset));
-	nvkm_wo32(chan->inst, 0x04c, upper_32_bits(args->offset) |
-				     (order_base_2(args->length / 8) << 16));
-	nvkm_wo32(chan->inst, 0x084, 0x20400000);
-	nvkm_wo32(chan->inst, 0x094, 0x30000001);
-	nvkm_wo32(chan->inst, 0x0ac, 0x00020000);
-	nvkm_wo32(chan->inst, 0x0e4, 0x00000000);
-	nvkm_wo32(chan->inst, 0x0e8, 0);
-	nvkm_wo32(chan->inst, 0x0f4, 0x00001000);
-	nvkm_wo32(chan->inst, 0x0f8, 0x10003080);
-	nvkm_mo32(chan->inst, 0x218, 0x00000000, 0x00000000);
-	nvkm_wo32(chan->inst, 0x220, lower_32_bits(nvkm_memory_bar2(chan->mthd)));
-	nvkm_wo32(chan->inst, 0x224, upper_32_bits(nvkm_memory_bar2(chan->mthd)));
-	nvkm_done(chan->inst);
-
-	userd = nvkm_umem_search(oclass->client, args->huserd);
-	if (IS_ERR(userd))
-		return PTR_ERR(userd);
-
-	ret = nvkm_memory_kmap(userd, &chan->user);
-	nvkm_memory_unref(&userd);
-	if (ret)
-		return ret;
-
-	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x1000, 0x1000, true, &chan->runl);
-	if (ret)
-		return ret;
-
-	nvkm_kmap(chan->runl);
-	nvkm_wo32(chan->runl, 0x00, 0x80030001);
-	nvkm_wo32(chan->runl, 0x04, 1);
-	nvkm_wo32(chan->runl, 0x08, 0);
-	nvkm_wo32(chan->runl, 0x0c, 0x00000000);
-	nvkm_wo32(chan->runl, 0x10, lower_32_bits(nvkm_memory_addr(chan->user)));
-	nvkm_wo32(chan->runl, 0x14, upper_32_bits(nvkm_memory_addr(chan->user)));
-	nvkm_wo32(chan->runl, 0x18, lower_32_bits(nvkm_memory_addr(chan->inst)));
-	nvkm_wo32(chan->runl, 0x1c, upper_32_bits(nvkm_memory_addr(chan->inst)));
-	nvkm_done(chan->runl);
-
-	ret = nvkm_vmm_join(vmm, chan->inst);
-	if (ret)
-		return ret;
-
-	chan->vmm = nvkm_vmm_ref(vmm);
-	return 0;
-}
-
-static const struct nvkm_device_oclass
-ga102_chan_oclass = {
-	.ctor = ga102_chan_new,
-};
-
-static int
-ga102_fifo_sclass(struct nvkm_oclass *oclass, int index, const struct nvkm_device_oclass **class)
-{
-	if (index == 0) {
-		oclass->base = (struct nvkm_sclass) { 0, 0, AMPERE_CHANNEL_GPFIFO_B };
-		*class = &ga102_chan_oclass;
-		return 0;
-	}
-
-	return 1;
-}
-
-static int
-ga102_fifo_info(struct nvkm_engine *engine, u64 mthd, u64 *data)
-{
-	switch (mthd) {
-	case NV_DEVICE_HOST_CHANNELS: *data = 1; return 0;
-	default:
-		break;
-	}
-
-	return -ENOSYS;
-}
-
-static void *
-ga102_fifo_dtor(struct nvkm_engine *engine)
-{
-	return ga102_fifo(engine);
-}
-
-static const struct nvkm_engine_func
+static const struct nvkm_fifo_func
 ga102_fifo = {
-	.dtor = ga102_fifo_dtor,
-	.info = ga102_fifo_info,
-	.base.sclass = ga102_fifo_sclass,
+	.runl_ctor = ga100_fifo_runl_ctor,
+	.mmu_fault = &tu102_fifo_mmu_fault,
+	.nonstall_ctor = ga100_fifo_nonstall_ctor,
+	.nonstall = &ga100_fifo_nonstall,
+	.runl = &ga100_runl,
+	.runq = &ga100_runq,
+	.engn = &ga100_engn,
+	.engn_ce = &ga100_engn_ce,
+	.cgrp = {{ 0, 0, KEPLER_CHANNEL_GROUP_A  }, &ga100_cgrp, .force = true },
+	.chan = {{ 0, 0, AMPERE_CHANNEL_GPFIFO_B }, &ga100_chan },
 };
 
 int
 ga102_fifo_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
 	       struct nvkm_fifo **pfifo)
 {
-	struct ga102_fifo *fifo;
-
-	if (!(fifo = kzalloc(sizeof(*fifo), GFP_KERNEL)))
-		return -ENOMEM;
-
-	nvkm_engine_ctor(&ga102_fifo, device, type, inst, true, &fifo->base.engine);
-	*pfifo = &fifo->base;
-	return 0;
+	return nvkm_fifo_new_(&ga102_fifo, device, type, inst, pfifo);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
index 24125e2d5d4f..4d448be19224 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
@@ -37,6 +37,7 @@ struct nvkm_fifo_func {
 	void (*pause)(struct nvkm_fifo *, unsigned long *);
 	void (*start)(struct nvkm_fifo *, unsigned long *);
 
+	int (*nonstall_ctor)(struct nvkm_fifo *);
 	const struct nvkm_event_func *nonstall;
 
 	const struct nvkm_runl_func *runl;
@@ -191,6 +192,16 @@ extern const struct nvkm_chan_func_ramfc gv100_chan_ramfc;
 void tu102_fifo_intr_ctxsw_timeout_info(struct nvkm_engn *, u32 info);
 extern const struct nvkm_fifo_func_mmu_fault tu102_fifo_mmu_fault;
 
+int ga100_fifo_runl_ctor(struct nvkm_fifo *);
+int ga100_fifo_nonstall_ctor(struct nvkm_fifo *);
+extern const struct nvkm_event_func ga100_fifo_nonstall;
+extern const struct nvkm_runl_func ga100_runl;
+extern const struct nvkm_runq_func ga100_runq;
+extern const struct nvkm_engn_func ga100_engn;
+extern const struct nvkm_engn_func ga100_engn_ce;
+extern const struct nvkm_cgrp_func ga100_cgrp;
+extern const struct nvkm_chan_func ga100_chan;
+
 int nvkm_uchan_new(struct nvkm_fifo *, struct nvkm_cgrp *, const struct nvkm_oclass *,
 		   void *argv, u32 argc, struct nvkm_object **);
 int nvkm_ucgrp_new(struct nvkm_fifo *, const struct nvkm_oclass *, void *argv, u32 argc,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c
index e654c1c4f769..fe51c075d6c4 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c
@@ -308,6 +308,9 @@ nvkm_runl_block(struct nvkm_runl *runl)
 void
 nvkm_runl_fini(struct nvkm_runl *runl)
 {
+	if (runl->func->fini)
+		runl->func->fini(runl);
+
 	flush_work(&runl->work);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h
index 5159b64377c5..c93d21bb7bd5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h
@@ -1,6 +1,6 @@
 #ifndef __NVKM_RUNL_H__
 #define __NVKM_RUNL_H__
-#include <core/os.h>
+#include <core/intr.h>
 struct nvkm_cctx;
 struct nvkm_cgrp;
 struct nvkm_chan;
@@ -36,6 +36,8 @@ struct nvkm_engn {
 
 struct nvkm_runl {
 	const struct nvkm_runl_func {
+		void (*init)(struct nvkm_runl *);
+		void (*fini)(struct nvkm_runl *);
 		int runqs;
 		u8 size;
 		int (*update)(struct nvkm_runl *);
@@ -53,6 +55,8 @@ struct nvkm_runl {
 	struct nvkm_fifo *fifo;
 	int id;
 	u32 addr;
+	u32 chan;
+	u16 doorbell;
 
 	struct nvkm_chid *cgid;
 #define NVKM_CHAN_EVENT_ERRORED BIT(0)
@@ -63,6 +67,8 @@ struct nvkm_runl {
 	struct nvkm_runq *runq[2];
 	int runq_nr;
 
+	struct nvkm_inth inth;
+
 	struct list_head cgrps;
 	int cgrp_nr;
 	int chan_nr;