Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / gpu / drm / nouveau / nvkm / engine / fifo / gk104.c
diff --git a/kernel/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c b/kernel/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
new file mode 100644 (file)
index 0000000..e10f964
--- /dev/null
@@ -0,0 +1,1139 @@
+/*
+ * Copyright 2012 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.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "gk104.h"
+
+#include <core/client.h>
+#include <core/engctx.h>
+#include <core/enum.h>
+#include <core/handle.h>
+#include <subdev/bar.h>
+#include <subdev/fb.h>
+#include <subdev/mmu.h>
+#include <subdev/timer.h>
+
+#include <nvif/class.h>
+#include <nvif/unpack.h>
+
+#define _(a,b) { (a), ((1ULL << (a)) | (b)) }
+static const struct {
+       u64 subdev;
+       u64 mask;
+} fifo_engine[] = {
+       _(NVDEV_ENGINE_GR      , (1ULL << NVDEV_ENGINE_SW) |
+                                (1ULL << NVDEV_ENGINE_CE2)),
+       _(NVDEV_ENGINE_MSPDEC  , 0),
+       _(NVDEV_ENGINE_MSPPP   , 0),
+       _(NVDEV_ENGINE_MSVLD   , 0),
+       _(NVDEV_ENGINE_CE0     , 0),
+       _(NVDEV_ENGINE_CE1     , 0),
+       _(NVDEV_ENGINE_MSENC   , 0),
+};
+#undef _
+#define FIFO_ENGINE_NR ARRAY_SIZE(fifo_engine)
+
+struct gk104_fifo_engn {
+       struct nvkm_gpuobj *runlist[2];
+       int cur_runlist;
+       wait_queue_head_t wait;
+};
+
+struct gk104_fifo_priv {
+       struct nvkm_fifo base;
+
+       struct work_struct fault;
+       u64 mask;
+
+       struct gk104_fifo_engn engine[FIFO_ENGINE_NR];
+       struct {
+               struct nvkm_gpuobj *mem;
+               struct nvkm_vma bar;
+       } user;
+       int spoon_nr;
+};
+
+struct gk104_fifo_base {
+       struct nvkm_fifo_base base;
+       struct nvkm_gpuobj *pgd;
+       struct nvkm_vm *vm;
+};
+
+struct gk104_fifo_chan {
+       struct nvkm_fifo_chan base;
+       u32 engine;
+       enum {
+               STOPPED,
+               RUNNING,
+               KILLED
+       } state;
+};
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+static void
+gk104_fifo_runlist_update(struct gk104_fifo_priv *priv, u32 engine)
+{
+       struct nvkm_bar *bar = nvkm_bar(priv);
+       struct gk104_fifo_engn *engn = &priv->engine[engine];
+       struct nvkm_gpuobj *cur;
+       int i, p;
+
+       mutex_lock(&nv_subdev(priv)->mutex);
+       cur = engn->runlist[engn->cur_runlist];
+       engn->cur_runlist = !engn->cur_runlist;
+
+       for (i = 0, p = 0; i < priv->base.max; i++) {
+               struct gk104_fifo_chan *chan = (void *)priv->base.channel[i];
+               if (chan && chan->state == RUNNING && chan->engine == engine) {
+                       nv_wo32(cur, p + 0, i);
+                       nv_wo32(cur, p + 4, 0x00000000);
+                       p += 8;
+               }
+       }
+       bar->flush(bar);
+
+       nv_wr32(priv, 0x002270, cur->addr >> 12);
+       nv_wr32(priv, 0x002274, (engine << 20) | (p >> 3));
+
+       if (wait_event_timeout(engn->wait, !(nv_rd32(priv, 0x002284 +
+                              (engine * 0x08)) & 0x00100000),
+                               msecs_to_jiffies(2000)) == 0)
+               nv_error(priv, "runlist %d update timeout\n", engine);
+       mutex_unlock(&nv_subdev(priv)->mutex);
+}
+
+static int
+gk104_fifo_context_attach(struct nvkm_object *parent,
+                         struct nvkm_object *object)
+{
+       struct nvkm_bar *bar = nvkm_bar(parent);
+       struct gk104_fifo_base *base = (void *)parent->parent;
+       struct nvkm_engctx *ectx = (void *)object;
+       u32 addr;
+       int ret;
+
+       switch (nv_engidx(object->engine)) {
+       case NVDEV_ENGINE_SW   :
+               return 0;
+       case NVDEV_ENGINE_CE0:
+       case NVDEV_ENGINE_CE1:
+       case NVDEV_ENGINE_CE2:
+               nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
+               return 0;
+       case NVDEV_ENGINE_GR    : addr = 0x0210; break;
+       case NVDEV_ENGINE_MSVLD : addr = 0x0270; break;
+       case NVDEV_ENGINE_MSPDEC: addr = 0x0250; break;
+       case NVDEV_ENGINE_MSPPP : addr = 0x0260; break;
+       default:
+               return -EINVAL;
+       }
+
+       if (!ectx->vma.node) {
+               ret = nvkm_gpuobj_map_vm(nv_gpuobj(ectx), base->vm,
+                                        NV_MEM_ACCESS_RW, &ectx->vma);
+               if (ret)
+                       return ret;
+
+               nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
+       }
+
+       nv_wo32(base, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
+       nv_wo32(base, addr + 0x04, upper_32_bits(ectx->vma.offset));
+       bar->flush(bar);
+       return 0;
+}
+
+static int
+gk104_fifo_context_detach(struct nvkm_object *parent, bool suspend,
+                         struct nvkm_object *object)
+{
+       struct nvkm_bar *bar = nvkm_bar(parent);
+       struct gk104_fifo_priv *priv = (void *)parent->engine;
+       struct gk104_fifo_base *base = (void *)parent->parent;
+       struct gk104_fifo_chan *chan = (void *)parent;
+       u32 addr;
+
+       switch (nv_engidx(object->engine)) {
+       case NVDEV_ENGINE_SW    : return 0;
+       case NVDEV_ENGINE_CE0   :
+       case NVDEV_ENGINE_CE1   :
+       case NVDEV_ENGINE_CE2   : addr = 0x0000; break;
+       case NVDEV_ENGINE_GR    : addr = 0x0210; break;
+       case NVDEV_ENGINE_MSVLD : addr = 0x0270; break;
+       case NVDEV_ENGINE_MSPDEC: addr = 0x0250; break;
+       case NVDEV_ENGINE_MSPPP : addr = 0x0260; break;
+       default:
+               return -EINVAL;
+       }
+
+       nv_wr32(priv, 0x002634, chan->base.chid);
+       if (!nv_wait(priv, 0x002634, 0xffffffff, chan->base.chid)) {
+               nv_error(priv, "channel %d [%s] kick timeout\n",
+                        chan->base.chid, nvkm_client_name(chan));
+               if (suspend)
+                       return -EBUSY;
+       }
+
+       if (addr) {
+               nv_wo32(base, addr + 0x00, 0x00000000);
+               nv_wo32(base, addr + 0x04, 0x00000000);
+               bar->flush(bar);
+       }
+
+       return 0;
+}
+
+static int
+gk104_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
+                    struct nvkm_oclass *oclass, void *data, u32 size,
+                    struct nvkm_object **pobject)
+{
+       union {
+               struct kepler_channel_gpfifo_a_v0 v0;
+       } *args = data;
+       struct nvkm_bar *bar = nvkm_bar(parent);
+       struct gk104_fifo_priv *priv = (void *)engine;
+       struct gk104_fifo_base *base = (void *)parent;
+       struct gk104_fifo_chan *chan;
+       u64 usermem, ioffset, ilength;
+       int ret, i;
+
+       nv_ioctl(parent, "create channel gpfifo size %d\n", size);
+       if (nvif_unpack(args->v0, 0, 0, false)) {
+               nv_ioctl(parent, "create channel gpfifo vers %d pushbuf %08x "
+                                "ioffset %016llx ilength %08x engine %08x\n",
+                        args->v0.version, args->v0.pushbuf, args->v0.ioffset,
+                        args->v0.ilength, args->v0.engine);
+       } else
+               return ret;
+
+       for (i = 0; i < FIFO_ENGINE_NR; i++) {
+               if (args->v0.engine & (1 << i)) {
+                       if (nvkm_engine(parent, fifo_engine[i].subdev)) {
+                               args->v0.engine = (1 << i);
+                               break;
+                       }
+               }
+       }
+
+       if (i == FIFO_ENGINE_NR) {
+               nv_error(priv, "unsupported engines 0x%08x\n", args->v0.engine);
+               return -ENODEV;
+       }
+
+       ret = nvkm_fifo_channel_create(parent, engine, oclass, 1,
+                                      priv->user.bar.offset, 0x200,
+                                      args->v0.pushbuf,
+                                      fifo_engine[i].mask, &chan);
+       *pobject = nv_object(chan);
+       if (ret)
+               return ret;
+
+       args->v0.chid = chan->base.chid;
+
+       nv_parent(chan)->context_attach = gk104_fifo_context_attach;
+       nv_parent(chan)->context_detach = gk104_fifo_context_detach;
+       chan->engine = i;
+
+       usermem = chan->base.chid * 0x200;
+       ioffset = args->v0.ioffset;
+       ilength = order_base_2(args->v0.ilength / 8);
+
+       for (i = 0; i < 0x200; i += 4)
+               nv_wo32(priv->user.mem, usermem + i, 0x00000000);
+
+       nv_wo32(base, 0x08, lower_32_bits(priv->user.mem->addr + usermem));
+       nv_wo32(base, 0x0c, upper_32_bits(priv->user.mem->addr + usermem));
+       nv_wo32(base, 0x10, 0x0000face);
+       nv_wo32(base, 0x30, 0xfffff902);
+       nv_wo32(base, 0x48, lower_32_bits(ioffset));
+       nv_wo32(base, 0x4c, upper_32_bits(ioffset) | (ilength << 16));
+       nv_wo32(base, 0x84, 0x20400000);
+       nv_wo32(base, 0x94, 0x30000001);
+       nv_wo32(base, 0x9c, 0x00000100);
+       nv_wo32(base, 0xac, 0x0000001f);
+       nv_wo32(base, 0xe8, chan->base.chid);
+       nv_wo32(base, 0xb8, 0xf8000000);
+       nv_wo32(base, 0xf8, 0x10003080); /* 0x002310 */
+       nv_wo32(base, 0xfc, 0x10000010); /* 0x002350 */
+       bar->flush(bar);
+       return 0;
+}
+
+static int
+gk104_fifo_chan_init(struct nvkm_object *object)
+{
+       struct nvkm_gpuobj *base = nv_gpuobj(object->parent);
+       struct gk104_fifo_priv *priv = (void *)object->engine;
+       struct gk104_fifo_chan *chan = (void *)object;
+       u32 chid = chan->base.chid;
+       int ret;
+
+       ret = nvkm_fifo_channel_init(&chan->base);
+       if (ret)
+               return ret;
+
+       nv_mask(priv, 0x800004 + (chid * 8), 0x000f0000, chan->engine << 16);
+       nv_wr32(priv, 0x800000 + (chid * 8), 0x80000000 | base->addr >> 12);
+
+       if (chan->state == STOPPED && (chan->state = RUNNING) == RUNNING) {
+               nv_mask(priv, 0x800004 + (chid * 8), 0x00000400, 0x00000400);
+               gk104_fifo_runlist_update(priv, chan->engine);
+               nv_mask(priv, 0x800004 + (chid * 8), 0x00000400, 0x00000400);
+       }
+
+       return 0;
+}
+
+static int
+gk104_fifo_chan_fini(struct nvkm_object *object, bool suspend)
+{
+       struct gk104_fifo_priv *priv = (void *)object->engine;
+       struct gk104_fifo_chan *chan = (void *)object;
+       u32 chid = chan->base.chid;
+
+       if (chan->state == RUNNING && (chan->state = STOPPED) == STOPPED) {
+               nv_mask(priv, 0x800004 + (chid * 8), 0x00000800, 0x00000800);
+               gk104_fifo_runlist_update(priv, chan->engine);
+       }
+
+       nv_wr32(priv, 0x800000 + (chid * 8), 0x00000000);
+       return nvkm_fifo_channel_fini(&chan->base, suspend);
+}
+
+struct nvkm_ofuncs
+gk104_fifo_chan_ofuncs = {
+       .ctor = gk104_fifo_chan_ctor,
+       .dtor = _nvkm_fifo_channel_dtor,
+       .init = gk104_fifo_chan_init,
+       .fini = gk104_fifo_chan_fini,
+       .map  = _nvkm_fifo_channel_map,
+       .rd32 = _nvkm_fifo_channel_rd32,
+       .wr32 = _nvkm_fifo_channel_wr32,
+       .ntfy = _nvkm_fifo_channel_ntfy
+};
+
+static struct nvkm_oclass
+gk104_fifo_sclass[] = {
+       { KEPLER_CHANNEL_GPFIFO_A, &gk104_fifo_chan_ofuncs },
+       {}
+};
+
+/*******************************************************************************
+ * FIFO context - instmem heap and vm setup
+ ******************************************************************************/
+
+static int
+gk104_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
+                       struct nvkm_oclass *oclass, void *data, u32 size,
+                       struct nvkm_object **pobject)
+{
+       struct gk104_fifo_base *base;
+       int ret;
+
+       ret = nvkm_fifo_context_create(parent, engine, oclass, NULL, 0x1000,
+                                      0x1000, NVOBJ_FLAG_ZERO_ALLOC, &base);
+       *pobject = nv_object(base);
+       if (ret)
+               return ret;
+
+       ret = nvkm_gpuobj_new(nv_object(base), NULL, 0x10000, 0x1000, 0,
+                             &base->pgd);
+       if (ret)
+               return ret;
+
+       nv_wo32(base, 0x0200, lower_32_bits(base->pgd->addr));
+       nv_wo32(base, 0x0204, upper_32_bits(base->pgd->addr));
+       nv_wo32(base, 0x0208, 0xffffffff);
+       nv_wo32(base, 0x020c, 0x000000ff);
+
+       ret = nvkm_vm_ref(nvkm_client(parent)->vm, &base->vm, base->pgd);
+       if (ret)
+               return ret;
+
+       return 0;
+}
+
+static void
+gk104_fifo_context_dtor(struct nvkm_object *object)
+{
+       struct gk104_fifo_base *base = (void *)object;
+       nvkm_vm_ref(NULL, &base->vm, base->pgd);
+       nvkm_gpuobj_ref(NULL, &base->pgd);
+       nvkm_fifo_context_destroy(&base->base);
+}
+
+static struct nvkm_oclass
+gk104_fifo_cclass = {
+       .handle = NV_ENGCTX(FIFO, 0xe0),
+       .ofuncs = &(struct nvkm_ofuncs) {
+               .ctor = gk104_fifo_context_ctor,
+               .dtor = gk104_fifo_context_dtor,
+               .init = _nvkm_fifo_context_init,
+               .fini = _nvkm_fifo_context_fini,
+               .rd32 = _nvkm_fifo_context_rd32,
+               .wr32 = _nvkm_fifo_context_wr32,
+       },
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+static inline int
+gk104_fifo_engidx(struct gk104_fifo_priv *priv, u32 engn)
+{
+       switch (engn) {
+       case NVDEV_ENGINE_GR    :
+       case NVDEV_ENGINE_CE2   : engn = 0; break;
+       case NVDEV_ENGINE_MSVLD : engn = 1; break;
+       case NVDEV_ENGINE_MSPPP : engn = 2; break;
+       case NVDEV_ENGINE_MSPDEC: engn = 3; break;
+       case NVDEV_ENGINE_CE0   : engn = 4; break;
+       case NVDEV_ENGINE_CE1   : engn = 5; break;
+       case NVDEV_ENGINE_MSENC : engn = 6; break;
+       default:
+               return -1;
+       }
+
+       return engn;
+}
+
+static inline struct nvkm_engine *
+gk104_fifo_engine(struct gk104_fifo_priv *priv, u32 engn)
+{
+       if (engn >= ARRAY_SIZE(fifo_engine))
+               return NULL;
+       return nvkm_engine(priv, fifo_engine[engn].subdev);
+}
+
+static void
+gk104_fifo_recover_work(struct work_struct *work)
+{
+       struct gk104_fifo_priv *priv = container_of(work, typeof(*priv), fault);
+       struct nvkm_object *engine;
+       unsigned long flags;
+       u32 engn, engm = 0;
+       u64 mask, todo;
+
+       spin_lock_irqsave(&priv->base.lock, flags);
+       mask = priv->mask;
+       priv->mask = 0ULL;
+       spin_unlock_irqrestore(&priv->base.lock, flags);
+
+       for (todo = mask; engn = __ffs64(todo), todo; todo &= ~(1 << engn))
+               engm |= 1 << gk104_fifo_engidx(priv, engn);
+       nv_mask(priv, 0x002630, engm, engm);
+
+       for (todo = mask; engn = __ffs64(todo), todo; todo &= ~(1 << engn)) {
+               if ((engine = (void *)nvkm_engine(priv, engn))) {
+                       nv_ofuncs(engine)->fini(engine, false);
+                       WARN_ON(nv_ofuncs(engine)->init(engine));
+               }
+               gk104_fifo_runlist_update(priv, gk104_fifo_engidx(priv, engn));
+       }
+
+       nv_wr32(priv, 0x00262c, engm);
+       nv_mask(priv, 0x002630, engm, 0x00000000);
+}
+
+static void
+gk104_fifo_recover(struct gk104_fifo_priv *priv, struct nvkm_engine *engine,
+                 struct gk104_fifo_chan *chan)
+{
+       u32 chid = chan->base.chid;
+       unsigned long flags;
+
+       nv_error(priv, "%s engine fault on channel %d, recovering...\n",
+                      nv_subdev(engine)->name, chid);
+
+       nv_mask(priv, 0x800004 + (chid * 0x08), 0x00000800, 0x00000800);
+       chan->state = KILLED;
+
+       spin_lock_irqsave(&priv->base.lock, flags);
+       priv->mask |= 1ULL << nv_engidx(engine);
+       spin_unlock_irqrestore(&priv->base.lock, flags);
+       schedule_work(&priv->fault);
+}
+
+static int
+gk104_fifo_swmthd(struct gk104_fifo_priv *priv, u32 chid, u32 mthd, u32 data)
+{
+       struct gk104_fifo_chan *chan = NULL;
+       struct nvkm_handle *bind;
+       unsigned long flags;
+       int ret = -EINVAL;
+
+       spin_lock_irqsave(&priv->base.lock, flags);
+       if (likely(chid >= priv->base.min && chid <= priv->base.max))
+               chan = (void *)priv->base.channel[chid];
+       if (unlikely(!chan))
+               goto out;
+
+       bind = nvkm_namedb_get_class(nv_namedb(chan), 0x906e);
+       if (likely(bind)) {
+               if (!mthd || !nv_call(bind->object, mthd, data))
+                       ret = 0;
+               nvkm_namedb_put(bind);
+       }
+
+out:
+       spin_unlock_irqrestore(&priv->base.lock, flags);
+       return ret;
+}
+
+static const struct nvkm_enum
+gk104_fifo_bind_reason[] = {
+       { 0x01, "BIND_NOT_UNBOUND" },
+       { 0x02, "SNOOP_WITHOUT_BAR1" },
+       { 0x03, "UNBIND_WHILE_RUNNING" },
+       { 0x05, "INVALID_RUNLIST" },
+       { 0x06, "INVALID_CTX_TGT" },
+       { 0x0b, "UNBIND_WHILE_PARKED" },
+       {}
+};
+
+static void
+gk104_fifo_intr_bind(struct gk104_fifo_priv *priv)
+{
+       u32 intr = nv_rd32(priv, 0x00252c);
+       u32 code = intr & 0x000000ff;
+       const struct nvkm_enum *en;
+       char enunk[6] = "";
+
+       en = nvkm_enum_find(gk104_fifo_bind_reason, code);
+       if (!en)
+               snprintf(enunk, sizeof(enunk), "UNK%02x", code);
+
+       nv_error(priv, "BIND_ERROR [ %s ]\n", en ? en->name : enunk);
+}
+
+static const struct nvkm_enum
+gk104_fifo_sched_reason[] = {
+       { 0x0a, "CTXSW_TIMEOUT" },
+       {}
+};
+
+static void
+gk104_fifo_intr_sched_ctxsw(struct gk104_fifo_priv *priv)
+{
+       struct nvkm_engine *engine;
+       struct gk104_fifo_chan *chan;
+       u32 engn;
+
+       for (engn = 0; engn < ARRAY_SIZE(fifo_engine); engn++) {
+               u32 stat = nv_rd32(priv, 0x002640 + (engn * 0x04));
+               u32 busy = (stat & 0x80000000);
+               u32 next = (stat & 0x07ff0000) >> 16;
+               u32 chsw = (stat & 0x00008000);
+               u32 save = (stat & 0x00004000);
+               u32 load = (stat & 0x00002000);
+               u32 prev = (stat & 0x000007ff);
+               u32 chid = load ? next : prev;
+               (void)save;
+
+               if (busy && chsw) {
+                       if (!(chan = (void *)priv->base.channel[chid]))
+                               continue;
+                       if (!(engine = gk104_fifo_engine(priv, engn)))
+                               continue;
+                       gk104_fifo_recover(priv, engine, chan);
+               }
+       }
+}
+
+static void
+gk104_fifo_intr_sched(struct gk104_fifo_priv *priv)
+{
+       u32 intr = nv_rd32(priv, 0x00254c);
+       u32 code = intr & 0x000000ff;
+       const struct nvkm_enum *en;
+       char enunk[6] = "";
+
+       en = nvkm_enum_find(gk104_fifo_sched_reason, code);
+       if (!en)
+               snprintf(enunk, sizeof(enunk), "UNK%02x", code);
+
+       nv_error(priv, "SCHED_ERROR [ %s ]\n", en ? en->name : enunk);
+
+       switch (code) {
+       case 0x0a:
+               gk104_fifo_intr_sched_ctxsw(priv);
+               break;
+       default:
+               break;
+       }
+}
+
+static void
+gk104_fifo_intr_chsw(struct gk104_fifo_priv *priv)
+{
+       u32 stat = nv_rd32(priv, 0x00256c);
+       nv_error(priv, "CHSW_ERROR 0x%08x\n", stat);
+       nv_wr32(priv, 0x00256c, stat);
+}
+
+static void
+gk104_fifo_intr_dropped_fault(struct gk104_fifo_priv *priv)
+{
+       u32 stat = nv_rd32(priv, 0x00259c);
+       nv_error(priv, "DROPPED_MMU_FAULT 0x%08x\n", stat);
+}
+
+static const struct nvkm_enum
+gk104_fifo_fault_engine[] = {
+       { 0x00, "GR", NULL, NVDEV_ENGINE_GR },
+       { 0x03, "IFB", NULL, NVDEV_ENGINE_IFB },
+       { 0x04, "BAR1", NULL, NVDEV_SUBDEV_BAR },
+       { 0x05, "BAR3", NULL, NVDEV_SUBDEV_INSTMEM },
+       { 0x07, "PBDMA0", NULL, NVDEV_ENGINE_FIFO },
+       { 0x08, "PBDMA1", NULL, NVDEV_ENGINE_FIFO },
+       { 0x09, "PBDMA2", NULL, NVDEV_ENGINE_FIFO },
+       { 0x10, "MSVLD", NULL, NVDEV_ENGINE_MSVLD },
+       { 0x11, "MSPPP", NULL, NVDEV_ENGINE_MSPPP },
+       { 0x13, "PERF" },
+       { 0x14, "MSPDEC", NULL, NVDEV_ENGINE_MSPDEC },
+       { 0x15, "CE0", NULL, NVDEV_ENGINE_CE0 },
+       { 0x16, "CE1", NULL, NVDEV_ENGINE_CE1 },
+       { 0x17, "PMU" },
+       { 0x19, "MSENC", NULL, NVDEV_ENGINE_MSENC },
+       { 0x1b, "CE2", NULL, NVDEV_ENGINE_CE2 },
+       {}
+};
+
+static const struct nvkm_enum
+gk104_fifo_fault_reason[] = {
+       { 0x00, "PDE" },
+       { 0x01, "PDE_SIZE" },
+       { 0x02, "PTE" },
+       { 0x03, "VA_LIMIT_VIOLATION" },
+       { 0x04, "UNBOUND_INST_BLOCK" },
+       { 0x05, "PRIV_VIOLATION" },
+       { 0x06, "RO_VIOLATION" },
+       { 0x07, "WO_VIOLATION" },
+       { 0x08, "PITCH_MASK_VIOLATION" },
+       { 0x09, "WORK_CREATION" },
+       { 0x0a, "UNSUPPORTED_APERTURE" },
+       { 0x0b, "COMPRESSION_FAILURE" },
+       { 0x0c, "UNSUPPORTED_KIND" },
+       { 0x0d, "REGION_VIOLATION" },
+       { 0x0e, "BOTH_PTES_VALID" },
+       { 0x0f, "INFO_TYPE_POISONED" },
+       {}
+};
+
+static const struct nvkm_enum
+gk104_fifo_fault_hubclient[] = {
+       { 0x00, "VIP" },
+       { 0x01, "CE0" },
+       { 0x02, "CE1" },
+       { 0x03, "DNISO" },
+       { 0x04, "FE" },
+       { 0x05, "FECS" },
+       { 0x06, "HOST" },
+       { 0x07, "HOST_CPU" },
+       { 0x08, "HOST_CPU_NB" },
+       { 0x09, "ISO" },
+       { 0x0a, "MMU" },
+       { 0x0b, "MSPDEC" },
+       { 0x0c, "MSPPP" },
+       { 0x0d, "MSVLD" },
+       { 0x0e, "NISO" },
+       { 0x0f, "P2P" },
+       { 0x10, "PD" },
+       { 0x11, "PERF" },
+       { 0x12, "PMU" },
+       { 0x13, "RASTERTWOD" },
+       { 0x14, "SCC" },
+       { 0x15, "SCC_NB" },
+       { 0x16, "SEC" },
+       { 0x17, "SSYNC" },
+       { 0x18, "GR_CE" },
+       { 0x19, "CE2" },
+       { 0x1a, "XV" },
+       { 0x1b, "MMU_NB" },
+       { 0x1c, "MSENC" },
+       { 0x1d, "DFALCON" },
+       { 0x1e, "SKED" },
+       { 0x1f, "AFALCON" },
+       {}
+};
+
+static const struct nvkm_enum
+gk104_fifo_fault_gpcclient[] = {
+       { 0x00, "L1_0" }, { 0x01, "T1_0" }, { 0x02, "PE_0" },
+       { 0x03, "L1_1" }, { 0x04, "T1_1" }, { 0x05, "PE_1" },
+       { 0x06, "L1_2" }, { 0x07, "T1_2" }, { 0x08, "PE_2" },
+       { 0x09, "L1_3" }, { 0x0a, "T1_3" }, { 0x0b, "PE_3" },
+       { 0x0c, "RAST" },
+       { 0x0d, "GCC" },
+       { 0x0e, "GPCCS" },
+       { 0x0f, "PROP_0" },
+       { 0x10, "PROP_1" },
+       { 0x11, "PROP_2" },
+       { 0x12, "PROP_3" },
+       { 0x13, "L1_4" }, { 0x14, "T1_4" }, { 0x15, "PE_4" },
+       { 0x16, "L1_5" }, { 0x17, "T1_5" }, { 0x18, "PE_5" },
+       { 0x19, "L1_6" }, { 0x1a, "T1_6" }, { 0x1b, "PE_6" },
+       { 0x1c, "L1_7" }, { 0x1d, "T1_7" }, { 0x1e, "PE_7" },
+       { 0x1f, "GPM" },
+       { 0x20, "LTP_UTLB_0" },
+       { 0x21, "LTP_UTLB_1" },
+       { 0x22, "LTP_UTLB_2" },
+       { 0x23, "LTP_UTLB_3" },
+       { 0x24, "GPC_RGG_UTLB" },
+       {}
+};
+
+static void
+gk104_fifo_intr_fault(struct gk104_fifo_priv *priv, int unit)
+{
+       u32 inst = nv_rd32(priv, 0x002800 + (unit * 0x10));
+       u32 valo = nv_rd32(priv, 0x002804 + (unit * 0x10));
+       u32 vahi = nv_rd32(priv, 0x002808 + (unit * 0x10));
+       u32 stat = nv_rd32(priv, 0x00280c + (unit * 0x10));
+       u32 gpc    = (stat & 0x1f000000) >> 24;
+       u32 client = (stat & 0x00001f00) >> 8;
+       u32 write  = (stat & 0x00000080);
+       u32 hub    = (stat & 0x00000040);
+       u32 reason = (stat & 0x0000000f);
+       struct nvkm_object *engctx = NULL, *object;
+       struct nvkm_engine *engine = NULL;
+       const struct nvkm_enum *er, *eu, *ec;
+       char erunk[6] = "";
+       char euunk[6] = "";
+       char ecunk[6] = "";
+       char gpcid[3] = "";
+
+       er = nvkm_enum_find(gk104_fifo_fault_reason, reason);
+       if (!er)
+               snprintf(erunk, sizeof(erunk), "UNK%02X", reason);
+
+       eu = nvkm_enum_find(gk104_fifo_fault_engine, unit);
+       if (eu) {
+               switch (eu->data2) {
+               case NVDEV_SUBDEV_BAR:
+                       nv_mask(priv, 0x001704, 0x00000000, 0x00000000);
+                       break;
+               case NVDEV_SUBDEV_INSTMEM:
+                       nv_mask(priv, 0x001714, 0x00000000, 0x00000000);
+                       break;
+               case NVDEV_ENGINE_IFB:
+                       nv_mask(priv, 0x001718, 0x00000000, 0x00000000);
+                       break;
+               default:
+                       engine = nvkm_engine(priv, eu->data2);
+                       if (engine)
+                               engctx = nvkm_engctx_get(engine, inst);
+                       break;
+               }
+       } else {
+               snprintf(euunk, sizeof(euunk), "UNK%02x", unit);
+       }
+
+       if (hub) {
+               ec = nvkm_enum_find(gk104_fifo_fault_hubclient, client);
+       } else {
+               ec = nvkm_enum_find(gk104_fifo_fault_gpcclient, client);
+               snprintf(gpcid, sizeof(gpcid), "%d", gpc);
+       }
+
+       if (!ec)
+               snprintf(ecunk, sizeof(ecunk), "UNK%02x", client);
+
+       nv_error(priv, "%s fault at 0x%010llx [%s] from %s/%s%s%s%s on "
+                      "channel 0x%010llx [%s]\n", write ? "write" : "read",
+                (u64)vahi << 32 | valo, er ? er->name : erunk,
+                eu ? eu->name : euunk, hub ? "" : "GPC", gpcid, hub ? "" : "/",
+                ec ? ec->name : ecunk, (u64)inst << 12,
+                nvkm_client_name(engctx));
+
+       object = engctx;
+       while (object) {
+               switch (nv_mclass(object)) {
+               case KEPLER_CHANNEL_GPFIFO_A:
+               case MAXWELL_CHANNEL_GPFIFO_A:
+                       gk104_fifo_recover(priv, engine, (void *)object);
+                       break;
+               }
+               object = object->parent;
+       }
+
+       nvkm_engctx_put(engctx);
+}
+
+static const struct nvkm_bitfield gk104_fifo_pbdma_intr_0[] = {
+       { 0x00000001, "MEMREQ" },
+       { 0x00000002, "MEMACK_TIMEOUT" },
+       { 0x00000004, "MEMACK_EXTRA" },
+       { 0x00000008, "MEMDAT_TIMEOUT" },
+       { 0x00000010, "MEMDAT_EXTRA" },
+       { 0x00000020, "MEMFLUSH" },
+       { 0x00000040, "MEMOP" },
+       { 0x00000080, "LBCONNECT" },
+       { 0x00000100, "LBREQ" },
+       { 0x00000200, "LBACK_TIMEOUT" },
+       { 0x00000400, "LBACK_EXTRA" },
+       { 0x00000800, "LBDAT_TIMEOUT" },
+       { 0x00001000, "LBDAT_EXTRA" },
+       { 0x00002000, "GPFIFO" },
+       { 0x00004000, "GPPTR" },
+       { 0x00008000, "GPENTRY" },
+       { 0x00010000, "GPCRC" },
+       { 0x00020000, "PBPTR" },
+       { 0x00040000, "PBENTRY" },
+       { 0x00080000, "PBCRC" },
+       { 0x00100000, "XBARCONNECT" },
+       { 0x00200000, "METHOD" },
+       { 0x00400000, "METHODCRC" },
+       { 0x00800000, "DEVICE" },
+       { 0x02000000, "SEMAPHORE" },
+       { 0x04000000, "ACQUIRE" },
+       { 0x08000000, "PRI" },
+       { 0x20000000, "NO_CTXSW_SEG" },
+       { 0x40000000, "PBSEG" },
+       { 0x80000000, "SIGNATURE" },
+       {}
+};
+
+static void
+gk104_fifo_intr_pbdma_0(struct gk104_fifo_priv *priv, int unit)
+{
+       u32 mask = nv_rd32(priv, 0x04010c + (unit * 0x2000));
+       u32 stat = nv_rd32(priv, 0x040108 + (unit * 0x2000)) & mask;
+       u32 addr = nv_rd32(priv, 0x0400c0 + (unit * 0x2000));
+       u32 data = nv_rd32(priv, 0x0400c4 + (unit * 0x2000));
+       u32 chid = nv_rd32(priv, 0x040120 + (unit * 0x2000)) & 0xfff;
+       u32 subc = (addr & 0x00070000) >> 16;
+       u32 mthd = (addr & 0x00003ffc);
+       u32 show = stat;
+
+       if (stat & 0x00800000) {
+               if (!gk104_fifo_swmthd(priv, chid, mthd, data))
+                       show &= ~0x00800000;
+               nv_wr32(priv, 0x0400c0 + (unit * 0x2000), 0x80600008);
+       }
+
+       if (show) {
+               nv_error(priv, "PBDMA%d:", unit);
+               nvkm_bitfield_print(gk104_fifo_pbdma_intr_0, show);
+               pr_cont("\n");
+               nv_error(priv,
+                        "PBDMA%d: ch %d [%s] subc %d mthd 0x%04x data 0x%08x\n",
+                        unit, chid,
+                        nvkm_client_name_for_fifo_chid(&priv->base, chid),
+                        subc, mthd, data);
+       }
+
+       nv_wr32(priv, 0x040108 + (unit * 0x2000), stat);
+}
+
+static const struct nvkm_bitfield gk104_fifo_pbdma_intr_1[] = {
+       { 0x00000001, "HCE_RE_ILLEGAL_OP" },
+       { 0x00000002, "HCE_RE_ALIGNB" },
+       { 0x00000004, "HCE_PRIV" },
+       { 0x00000008, "HCE_ILLEGAL_MTHD" },
+       { 0x00000010, "HCE_ILLEGAL_CLASS" },
+       {}
+};
+
+static void
+gk104_fifo_intr_pbdma_1(struct gk104_fifo_priv *priv, int unit)
+{
+       u32 mask = nv_rd32(priv, 0x04014c + (unit * 0x2000));
+       u32 stat = nv_rd32(priv, 0x040148 + (unit * 0x2000)) & mask;
+       u32 chid = nv_rd32(priv, 0x040120 + (unit * 0x2000)) & 0xfff;
+
+       if (stat) {
+               nv_error(priv, "PBDMA%d:", unit);
+               nvkm_bitfield_print(gk104_fifo_pbdma_intr_1, stat);
+               pr_cont("\n");
+               nv_error(priv, "PBDMA%d: ch %d %08x %08x\n", unit, chid,
+                        nv_rd32(priv, 0x040150 + (unit * 0x2000)),
+                        nv_rd32(priv, 0x040154 + (unit * 0x2000)));
+       }
+
+       nv_wr32(priv, 0x040148 + (unit * 0x2000), stat);
+}
+
+static void
+gk104_fifo_intr_runlist(struct gk104_fifo_priv *priv)
+{
+       u32 mask = nv_rd32(priv, 0x002a00);
+       while (mask) {
+               u32 engn = __ffs(mask);
+               wake_up(&priv->engine[engn].wait);
+               nv_wr32(priv, 0x002a00, 1 << engn);
+               mask &= ~(1 << engn);
+       }
+}
+
+static void
+gk104_fifo_intr_engine(struct gk104_fifo_priv *priv)
+{
+       nvkm_fifo_uevent(&priv->base);
+}
+
+static void
+gk104_fifo_intr(struct nvkm_subdev *subdev)
+{
+       struct gk104_fifo_priv *priv = (void *)subdev;
+       u32 mask = nv_rd32(priv, 0x002140);
+       u32 stat = nv_rd32(priv, 0x002100) & mask;
+
+       if (stat & 0x00000001) {
+               gk104_fifo_intr_bind(priv);
+               nv_wr32(priv, 0x002100, 0x00000001);
+               stat &= ~0x00000001;
+       }
+
+       if (stat & 0x00000010) {
+               nv_error(priv, "PIO_ERROR\n");
+               nv_wr32(priv, 0x002100, 0x00000010);
+               stat &= ~0x00000010;
+       }
+
+       if (stat & 0x00000100) {
+               gk104_fifo_intr_sched(priv);
+               nv_wr32(priv, 0x002100, 0x00000100);
+               stat &= ~0x00000100;
+       }
+
+       if (stat & 0x00010000) {
+               gk104_fifo_intr_chsw(priv);
+               nv_wr32(priv, 0x002100, 0x00010000);
+               stat &= ~0x00010000;
+       }
+
+       if (stat & 0x00800000) {
+               nv_error(priv, "FB_FLUSH_TIMEOUT\n");
+               nv_wr32(priv, 0x002100, 0x00800000);
+               stat &= ~0x00800000;
+       }
+
+       if (stat & 0x01000000) {
+               nv_error(priv, "LB_ERROR\n");
+               nv_wr32(priv, 0x002100, 0x01000000);
+               stat &= ~0x01000000;
+       }
+
+       if (stat & 0x08000000) {
+               gk104_fifo_intr_dropped_fault(priv);
+               nv_wr32(priv, 0x002100, 0x08000000);
+               stat &= ~0x08000000;
+       }
+
+       if (stat & 0x10000000) {
+               u32 mask = nv_rd32(priv, 0x00259c);
+               while (mask) {
+                       u32 unit = __ffs(mask);
+                       gk104_fifo_intr_fault(priv, unit);
+                       nv_wr32(priv, 0x00259c, (1 << unit));
+                       mask &= ~(1 << unit);
+               }
+               stat &= ~0x10000000;
+       }
+
+       if (stat & 0x20000000) {
+               u32 mask = nv_rd32(priv, 0x0025a0);
+               while (mask) {
+                       u32 unit = __ffs(mask);
+                       gk104_fifo_intr_pbdma_0(priv, unit);
+                       gk104_fifo_intr_pbdma_1(priv, unit);
+                       nv_wr32(priv, 0x0025a0, (1 << unit));
+                       mask &= ~(1 << unit);
+               }
+               stat &= ~0x20000000;
+       }
+
+       if (stat & 0x40000000) {
+               gk104_fifo_intr_runlist(priv);
+               stat &= ~0x40000000;
+       }
+
+       if (stat & 0x80000000) {
+               nv_wr32(priv, 0x002100, 0x80000000);
+               gk104_fifo_intr_engine(priv);
+               stat &= ~0x80000000;
+       }
+
+       if (stat) {
+               nv_error(priv, "INTR 0x%08x\n", stat);
+               nv_mask(priv, 0x002140, stat, 0x00000000);
+               nv_wr32(priv, 0x002100, stat);
+       }
+}
+
+static void
+gk104_fifo_uevent_init(struct nvkm_event *event, int type, int index)
+{
+       struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), uevent);
+       nv_mask(fifo, 0x002140, 0x80000000, 0x80000000);
+}
+
+static void
+gk104_fifo_uevent_fini(struct nvkm_event *event, int type, int index)
+{
+       struct nvkm_fifo *fifo = container_of(event, typeof(*fifo), uevent);
+       nv_mask(fifo, 0x002140, 0x80000000, 0x00000000);
+}
+
+static const struct nvkm_event_func
+gk104_fifo_uevent_func = {
+       .ctor = nvkm_fifo_uevent_ctor,
+       .init = gk104_fifo_uevent_init,
+       .fini = gk104_fifo_uevent_fini,
+};
+
+int
+gk104_fifo_fini(struct nvkm_object *object, bool suspend)
+{
+       struct gk104_fifo_priv *priv = (void *)object;
+       int ret;
+
+       ret = nvkm_fifo_fini(&priv->base, suspend);
+       if (ret)
+               return ret;
+
+       /* allow mmu fault interrupts, even when we're not using fifo */
+       nv_mask(priv, 0x002140, 0x10000000, 0x10000000);
+       return 0;
+}
+
+int
+gk104_fifo_init(struct nvkm_object *object)
+{
+       struct gk104_fifo_priv *priv = (void *)object;
+       int ret, i;
+
+       ret = nvkm_fifo_init(&priv->base);
+       if (ret)
+               return ret;
+
+       /* enable all available PBDMA units */
+       nv_wr32(priv, 0x000204, 0xffffffff);
+       priv->spoon_nr = hweight32(nv_rd32(priv, 0x000204));
+       nv_debug(priv, "%d PBDMA unit(s)\n", priv->spoon_nr);
+
+       /* PBDMA[n] */
+       for (i = 0; i < priv->spoon_nr; i++) {
+               nv_mask(priv, 0x04013c + (i * 0x2000), 0x10000100, 0x00000000);
+               nv_wr32(priv, 0x040108 + (i * 0x2000), 0xffffffff); /* INTR */
+               nv_wr32(priv, 0x04010c + (i * 0x2000), 0xfffffeff); /* INTREN */
+       }
+
+       /* PBDMA[n].HCE */
+       for (i = 0; i < priv->spoon_nr; i++) {
+               nv_wr32(priv, 0x040148 + (i * 0x2000), 0xffffffff); /* INTR */
+               nv_wr32(priv, 0x04014c + (i * 0x2000), 0xffffffff); /* INTREN */
+       }
+
+       nv_wr32(priv, 0x002254, 0x10000000 | priv->user.bar.offset >> 12);
+
+       nv_wr32(priv, 0x002100, 0xffffffff);
+       nv_wr32(priv, 0x002140, 0x7fffffff);
+       return 0;
+}
+
+void
+gk104_fifo_dtor(struct nvkm_object *object)
+{
+       struct gk104_fifo_priv *priv = (void *)object;
+       int i;
+
+       nvkm_gpuobj_unmap(&priv->user.bar);
+       nvkm_gpuobj_ref(NULL, &priv->user.mem);
+
+       for (i = 0; i < FIFO_ENGINE_NR; i++) {
+               nvkm_gpuobj_ref(NULL, &priv->engine[i].runlist[1]);
+               nvkm_gpuobj_ref(NULL, &priv->engine[i].runlist[0]);
+       }
+
+       nvkm_fifo_destroy(&priv->base);
+}
+
+int
+gk104_fifo_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
+               struct nvkm_oclass *oclass, void *data, u32 size,
+               struct nvkm_object **pobject)
+{
+       struct gk104_fifo_impl *impl = (void *)oclass;
+       struct gk104_fifo_priv *priv;
+       int ret, i;
+
+       ret = nvkm_fifo_create(parent, engine, oclass, 0,
+                              impl->channels - 1, &priv);
+       *pobject = nv_object(priv);
+       if (ret)
+               return ret;
+
+       INIT_WORK(&priv->fault, gk104_fifo_recover_work);
+
+       for (i = 0; i < FIFO_ENGINE_NR; i++) {
+               ret = nvkm_gpuobj_new(nv_object(priv), NULL, 0x8000, 0x1000,
+                                     0, &priv->engine[i].runlist[0]);
+               if (ret)
+                       return ret;
+
+               ret = nvkm_gpuobj_new(nv_object(priv), NULL, 0x8000, 0x1000,
+                                     0, &priv->engine[i].runlist[1]);
+               if (ret)
+                       return ret;
+
+               init_waitqueue_head(&priv->engine[i].wait);
+       }
+
+       ret = nvkm_gpuobj_new(nv_object(priv), NULL, impl->channels * 0x200,
+                             0x1000, NVOBJ_FLAG_ZERO_ALLOC, &priv->user.mem);
+       if (ret)
+               return ret;
+
+       ret = nvkm_gpuobj_map(priv->user.mem, NV_MEM_ACCESS_RW,
+                             &priv->user.bar);
+       if (ret)
+               return ret;
+
+       ret = nvkm_event_init(&gk104_fifo_uevent_func, 1, 1, &priv->base.uevent);
+       if (ret)
+               return ret;
+
+       nv_subdev(priv)->unit = 0x00000100;
+       nv_subdev(priv)->intr = gk104_fifo_intr;
+       nv_engine(priv)->cclass = &gk104_fifo_cclass;
+       nv_engine(priv)->sclass = gk104_fifo_sclass;
+       return 0;
+}
+
+struct nvkm_oclass *
+gk104_fifo_oclass = &(struct gk104_fifo_impl) {
+       .base.handle = NV_ENGINE(FIFO, 0xe0),
+       .base.ofuncs = &(struct nvkm_ofuncs) {
+               .ctor = gk104_fifo_ctor,
+               .dtor = gk104_fifo_dtor,
+               .init = gk104_fifo_init,
+               .fini = gk104_fifo_fini,
+       },
+       .channels = 4096,
+}.base;