static int
nv50_mpeg_context_new(struct nouveau_channel *chan, int engine)
{
struct drm_device *dev = chan->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *ramin = chan->ramin;
struct nouveau_gpuobj *ctx = NULL;
int ret;
NV_DEBUG(dev, "ch%d\n", chan->id);
ret = nouveau_gpuobj_new(dev, chan, 128 * 4, 0, NVOBJ_FLAG_ZERO_ALLOC |
NVOBJ_FLAG_ZERO_FREE, &ctx);
if (ret)
return ret;
nv_wo32(ramin, CTX_PTR(dev, 0x00), 0x80190002);
nv_wo32(ramin, CTX_PTR(dev, 0x04), ctx->vinst + ctx->size - 1);
nv_wo32(ramin, CTX_PTR(dev, 0x08), ctx->vinst);
nv_wo32(ramin, CTX_PTR(dev, 0x0c), 0);
nv_wo32(ramin, CTX_PTR(dev, 0x10), 0);
nv_wo32(ramin, CTX_PTR(dev, 0x14), 0x00010000);
nv_wo32(ctx, 0x70, 0x00801ec1);
nv_wo32(ctx, 0x7c, 0x0000037c);
dev_priv->engine.instmem.flush(dev);
chan->engctx[engine] = ctx;
return 0;
}
static int
nv25_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
struct nouveau_oclass *oclass, void *data, u32 size,
struct nouveau_object **pobject)
{
struct nv20_graph_priv *priv;
int ret;
ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
*pobject = nv_object(priv);
if (ret)
return ret;
ret = nouveau_gpuobj_new(parent, NULL, 32 * 4, 16,
NVOBJ_FLAG_ZERO_ALLOC, &priv->ctxtab);
if (ret)
return ret;
nv_subdev(priv)->unit = 0x00001000;
nv_subdev(priv)->intr = nv20_graph_intr;
nv_engine(priv)->cclass = &nv25_graph_cclass;
nv_engine(priv)->sclass = nv25_graph_sclass;
nv_engine(priv)->tile_prog = nv20_graph_tile_prog;
return 0;
}
static int
nva3_copy_context_new(struct nouveau_channel *chan, int engine)
{
struct drm_device *dev = chan->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *ramin = chan->ramin;
struct nouveau_gpuobj *ctx = NULL;
int ret;
NV_DEBUG(dev, "ch%d\n", chan->id);
ret = nouveau_gpuobj_new(dev, chan, 256, 0, NVOBJ_FLAG_ZERO_ALLOC |
NVOBJ_FLAG_ZERO_FREE, &ctx);
if (ret)
return ret;
nv_wo32(ramin, 0xc0, 0x00190000);
nv_wo32(ramin, 0xc4, ctx->vinst + ctx->size - 1);
nv_wo32(ramin, 0xc8, ctx->vinst);
nv_wo32(ramin, 0xcc, 0x00000000);
nv_wo32(ramin, 0xd0, 0x00000000);
nv_wo32(ramin, 0xd4, 0x00000000);
dev_priv->engine.instmem.flush(dev);
atomic_inc(&chan->vm->engref[engine]);
chan->engctx[engine] = ctx;
return 0;
}
int
nv84_fence_create(struct drm_device *dev)
{
struct nouveau_fifo_priv *pfifo = nv_engine(dev, NVOBJ_ENGINE_FIFO);
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nv84_fence_priv *priv;
int ret;
priv = kzalloc(sizeof(*priv), GFP_KERNEL);
if (!priv)
return -ENOMEM;
priv->base.engine.destroy = nv84_fence_destroy;
priv->base.engine.init = nv84_fence_init;
priv->base.engine.fini = nv84_fence_fini;
priv->base.engine.context_new = nv84_fence_context_new;
priv->base.engine.context_del = nv84_fence_context_del;
priv->base.emit = nv84_fence_emit;
priv->base.sync = nv84_fence_sync;
priv->base.read = nv84_fence_read;
dev_priv->eng[NVOBJ_ENGINE_FENCE] = &priv->base.engine;
ret = nouveau_gpuobj_new(dev, NULL, 16 * pfifo->channels,
0x1000, 0, &priv->mem);
if (ret)
goto out;
out:
if (ret)
nv84_fence_destroy(dev, NVOBJ_ENGINE_FENCE);
return ret;
}
static int
nv50_channel_new(struct drm_device *dev, u32 size, struct nouveau_vm *vm,
struct nouveau_channel **pchan)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
u32 pgd = (dev_priv->chipset == 0x50) ? 0x1400 : 0x0200;
u32 fc = (dev_priv->chipset == 0x50) ? 0x0000 : 0x4200;
struct nouveau_channel *chan;
int ret, i;
chan = kzalloc(sizeof(*chan), GFP_KERNEL);
if (!chan)
return -ENOMEM;
chan->dev = dev;
ret = nouveau_gpuobj_new(dev, NULL, size, 0x1000, 0, &chan->ramin);
if (ret) {
nv50_channel_del(&chan);
return ret;
}
ret = drm_mm_init(&chan->ramin_heap, 0x6000, chan->ramin->size);
if (ret) {
nv50_channel_del(&chan);
return ret;
}
ret = nouveau_gpuobj_new_fake(dev, chan->ramin->pinst == ~0 ? ~0 :
chan->ramin->pinst + pgd,
chan->ramin->vinst + pgd,
0x4000, NVOBJ_FLAG_ZERO_ALLOC,
&chan->vm_pd);
if (ret) {
nv50_channel_del(&chan);
return ret;
}
for (i = 0; i < 0x4000; i += 8) {
nv_wo32(chan->vm_pd, i + 0, 0x00000000);
nv_wo32(chan->vm_pd, i + 4, 0xdeadcafe);
}
ret = nouveau_vm_ref(vm, &chan->vm, chan->vm_pd);
if (ret) {
nv50_channel_del(&chan);
return ret;
}
ret = nouveau_gpuobj_new_fake(dev, chan->ramin->pinst == ~0 ? ~0 :
chan->ramin->pinst + fc,
chan->ramin->vinst + fc, 0x100,
NVOBJ_FLAG_ZERO_ALLOC, &chan->ramfc);
if (ret) {
nv50_channel_del(&chan);
return ret;
}
*pchan = chan;
return 0;
}
static int
nv04_instmem_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
struct nouveau_oclass *oclass, void *data, u32 size,
struct nouveau_object **pobject)
{
struct nv04_instmem_priv *priv;
int ret;
ret = nouveau_instmem_create(parent, engine, oclass, &priv);
*pobject = nv_object(priv);
if (ret)
return ret;
/* PRAMIN aperture maps over the end of VRAM, reserve it */
priv->base.reserved = 512 * 1024;
ret = nouveau_mm_init(&priv->heap, 0, priv->base.reserved, 1);
if (ret)
return ret;
/* 0x00000-0x10000: reserve for probable vbios image */
ret = nouveau_gpuobj_new(nv_object(priv), NULL, 0x10000, 0, 0,
&priv->vbios);
if (ret)
return ret;
/* 0x10000-0x18000: reserve for RAMHT */
ret = nouveau_ramht_new(nv_object(priv), NULL, 0x08000, 0, &priv->ramht);
if (ret)
return ret;
/* 0x18000-0x18800: reserve for RAMFC (enough for 32 nv30 channels) */
ret = nouveau_gpuobj_new(nv_object(priv), NULL, 0x00800, 0,
NVOBJ_FLAG_ZERO_ALLOC, &priv->ramfc);
if (ret)
return ret;
/* 0x18800-0x18a00: reserve for RAMRO */
ret = nouveau_gpuobj_new(nv_object(priv), NULL, 0x00200, 0, 0,
&priv->ramro);
if (ret)
return ret;
return 0;
}
int
nv20_graph_create_context(struct nouveau_channel *chan)
{
struct drm_device *dev = chan->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
void (*ctx_init)(struct drm_device *, struct nouveau_gpuobj *);
unsigned int idoffs = 0x28;
int ret;
switch (dev_priv->chipset) {
case 0x20:
ctx_init = nv20_graph_context_init;
idoffs = 0;
break;
case 0x25:
case 0x28:
ctx_init = nv25_graph_context_init;
break;
case 0x2a:
ctx_init = nv2a_graph_context_init;
idoffs = 0;
break;
case 0x30:
case 0x31:
ctx_init = nv30_31_graph_context_init;
break;
case 0x34:
ctx_init = nv34_graph_context_init;
break;
case 0x35:
case 0x36:
ctx_init = nv35_36_graph_context_init;
break;
default:
BUG_ON(1);
}
ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 16,
NVOBJ_FLAG_ZERO_ALLOC, &chan->ramin_grctx);
if (ret)
return ret;
/* Initialise default context values */
ctx_init(dev, chan->ramin_grctx);
/* nv20: nv_wo32(dev, chan->ramin_grctx->gpuobj, 10, chan->id<<24); */
nv_wo32(chan->ramin_grctx, idoffs,
(chan->id << 24) | 0x1); /* CTX_USER */
nv_wo32(pgraph->ctx_table, chan->id * 4, chan->ramin_grctx->pinst >> 4);
return 0;
}
static int
nv41_vmmgr_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
struct nouveau_oclass *oclass, void *data, u32 size,
struct nouveau_object **pobject)
{
struct nouveau_device *device = nv_device(parent);
struct nv04_vmmgr_priv *priv;
int ret;
if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
!nouveau_boolopt(device->cfgopt, "NvPCIE", true)) {
return nouveau_object_ctor(parent, engine, &nv04_vmmgr_oclass,
data, size, pobject);
}
ret = nouveau_vmmgr_create(parent, engine, oclass, "PCIEGART",
"pciegart", &priv);
*pobject = nv_object(priv);
if (ret)
return ret;
priv->base.create = nv04_vm_create;
priv->base.limit = NV41_GART_SIZE;
priv->base.dma_bits = 39;
priv->base.pgt_bits = 32 - 12;
priv->base.spg_shift = 12;
priv->base.lpg_shift = 12;
priv->base.map_sg = nv41_vm_map_sg;
priv->base.unmap = nv41_vm_unmap;
priv->base.flush = nv41_vm_flush;
ret = nouveau_vm_create(&priv->base, 0, NV41_GART_SIZE, 0, 4096,
&priv->vm);
if (ret)
return ret;
ret = nouveau_gpuobj_new(nv_object(priv), NULL,
(NV41_GART_SIZE / NV41_GART_PAGE) * 4,
16, NVOBJ_FLAG_ZERO_ALLOC,
&priv->vm->pgt[0].obj[0]);
priv->vm->pgt[0].refcount[0] = 1;
if (ret)
return ret;
return 0;
}
static int
nv50_dmaobj_bind(struct nouveau_dmaobj *dmaobj,
struct nouveau_object *parent,
struct nouveau_gpuobj **pgpuobj)
{
struct nv50_dmaobj_priv *priv = (void *)dmaobj;
int ret;
if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
switch (nv_mclass(parent->parent)) {
case NV40_CHANNEL_DMA:
case NV50_CHANNEL_GPFIFO:
case G82_CHANNEL_GPFIFO:
case NV50_DISP_CORE_CHANNEL_DMA:
case G82_DISP_CORE_CHANNEL_DMA:
case GT206_DISP_CORE_CHANNEL_DMA:
case GT200_DISP_CORE_CHANNEL_DMA:
case GT214_DISP_CORE_CHANNEL_DMA:
case NV50_DISP_BASE_CHANNEL_DMA:
case G82_DISP_BASE_CHANNEL_DMA:
case GT200_DISP_BASE_CHANNEL_DMA:
case GT214_DISP_BASE_CHANNEL_DMA:
case NV50_DISP_OVERLAY_CHANNEL_DMA:
case G82_DISP_OVERLAY_CHANNEL_DMA:
case GT200_DISP_OVERLAY_CHANNEL_DMA:
case GT214_DISP_OVERLAY_CHANNEL_DMA:
break;
default:
return -EINVAL;
}
}
ret = nouveau_gpuobj_new(parent, parent, 24, 32, 0, pgpuobj);
if (ret == 0) {
nv_wo32(*pgpuobj, 0x00, priv->flags0 | nv_mclass(dmaobj));
nv_wo32(*pgpuobj, 0x04, lower_32_bits(priv->base.limit));
nv_wo32(*pgpuobj, 0x08, lower_32_bits(priv->base.start));
nv_wo32(*pgpuobj, 0x0c, upper_32_bits(priv->base.limit) << 24 |
upper_32_bits(priv->base.start));
nv_wo32(*pgpuobj, 0x10, 0x00000000);
nv_wo32(*pgpuobj, 0x14, priv->flags5);
}
return ret;
}
int
_nouveau_xtensa_init(struct nouveau_object *object)
{
struct nouveau_device *device = nv_device(object);
struct nouveau_xtensa *xtensa = (void *)object;
const struct firmware *fw;
char name[32];
int i, ret;
u32 tmp;
ret = nouveau_engine_init(&xtensa->base);
if (ret)
return ret;
if (!xtensa->gpu_fw) {
snprintf(name, sizeof(name), "nouveau/nv84_xuc%03x",
xtensa->addr >> 12);
ret = request_firmware(&fw, name, nv_device_base(device));
if (ret) {
nv_warn(xtensa, "unable to load firmware %s\n", name);
return ret;
}
if (fw->size > 0x40000) {
nv_warn(xtensa, "firmware %s too large\n", name);
release_firmware(fw);
return -EINVAL;
}
ret = nouveau_gpuobj_new(object, NULL, 0x40000, 0x1000, 0,
&xtensa->gpu_fw);
if (ret) {
release_firmware(fw);
return ret;
}
nv_debug(xtensa, "Loading firmware to address: 0x%"PRIxMAX"\n",
(uintmax_t)xtensa->gpu_fw->addr);
for (i = 0; i < fw->size / 4; i++)
nv_wo32(xtensa->gpu_fw, i * 4, *((u32 *)fw->data + i));
release_firmware(fw);
}
static int
nv50_graph_context_new(struct nouveau_channel *chan, int engine)
{
struct drm_device *dev = chan->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *ramin = chan->ramin;
struct nouveau_gpuobj *grctx = NULL;
struct nv50_graph_engine *pgraph = nv_engine(dev, engine);
struct nouveau_grctx ctx = {};
int hdr, ret;
NV_DEBUG(dev, "ch%d\n", chan->id);
ret = nouveau_gpuobj_new(dev, NULL, pgraph->grctx_size, 0,
NVOBJ_FLAG_ZERO_ALLOC |
NVOBJ_FLAG_ZERO_FREE, &grctx);
if (ret)
return ret;
hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
nv_wo32(ramin, hdr + 0x00, 0x00190002);
nv_wo32(ramin, hdr + 0x04, grctx->vinst + grctx->size - 1);
nv_wo32(ramin, hdr + 0x08, grctx->vinst);
nv_wo32(ramin, hdr + 0x0c, 0);
nv_wo32(ramin, hdr + 0x10, 0);
nv_wo32(ramin, hdr + 0x14, 0x00010000);
ctx.dev = chan->dev;
ctx.mode = NOUVEAU_GRCTX_VALS;
ctx.data = grctx;
nv50_grctx_init(&ctx);
nv_wo32(grctx, 0x00000, chan->ramin->vinst >> 12);
dev_priv->engine.instmem.flush(dev);
atomic_inc(&chan->vm->engref[NVOBJ_ENGINE_GR]);
chan->engctx[NVOBJ_ENGINE_GR] = grctx;
return 0;
}
static int
nvc0_channel_new(struct drm_device *dev, u32 size, struct nouveau_vm *vm,
struct nouveau_channel **pchan,
struct nouveau_gpuobj *pgd, u64 vm_size)
{
struct nouveau_channel *chan;
int ret;
chan = kzalloc(sizeof(*chan), GFP_KERNEL);
if (!chan)
return -ENOMEM;
chan->dev = dev;
ret = nouveau_gpuobj_new(dev, NULL, size, 0x1000, 0, &chan->ramin);
if (ret) {
nvc0_channel_del(&chan);
return ret;
}
ret = drm_mm_init(&chan->ramin_heap, 0x1000, size - 0x1000);
if (ret) {
nvc0_channel_del(&chan);
return ret;
}
ret = nouveau_vm_ref(vm, &chan->vm, NULL);
if (ret) {
nvc0_channel_del(&chan);
return ret;
}
nv_wo32(chan->ramin, 0x0200, lower_32_bits(pgd->vinst));
nv_wo32(chan->ramin, 0x0204, upper_32_bits(pgd->vinst));
nv_wo32(chan->ramin, 0x0208, lower_32_bits(vm_size - 1));
nv_wo32(chan->ramin, 0x020c, upper_32_bits(vm_size - 1));
*pchan = chan;
return 0;
}
int
nv50_graph_create_context(struct nouveau_channel *chan)
{
struct drm_device *dev = chan->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *ramin = chan->ramin;
struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
struct nouveau_grctx ctx = {};
int hdr, ret;
NV_DEBUG(dev, "ch%d\n", chan->id);
ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 0,
NVOBJ_FLAG_ZERO_ALLOC |
NVOBJ_FLAG_ZERO_FREE, &chan->ramin_grctx);
if (ret)
return ret;
hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
nv_wo32(ramin, hdr + 0x00, 0x00190002);
nv_wo32(ramin, hdr + 0x04, chan->ramin_grctx->vinst +
pgraph->grctx_size - 1);
nv_wo32(ramin, hdr + 0x08, chan->ramin_grctx->vinst);
nv_wo32(ramin, hdr + 0x0c, 0);
nv_wo32(ramin, hdr + 0x10, 0);
nv_wo32(ramin, hdr + 0x14, 0x00010000);
ctx.dev = chan->dev;
ctx.mode = NOUVEAU_GRCTX_VALS;
ctx.data = chan->ramin_grctx;
nv50_grctx_init(&ctx);
nv_wo32(chan->ramin_grctx, 0x00000, chan->ramin->vinst >> 12);
dev_priv->engine.instmem.flush(dev);
atomic_inc(&chan->vm->pgraph_refs);
return 0;
}
static int
nvc0_copy_context_new(struct nouveau_channel *chan, int engine)
{
struct nvc0_copy_engine *pcopy = nv_engine(chan->dev, engine);
struct drm_device *dev = chan->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *ramin = chan->ramin;
struct nouveau_gpuobj *ctx = NULL;
int ret;
ret = nouveau_gpuobj_new(dev, chan, 256, 256,
NVOBJ_FLAG_VM | NVOBJ_FLAG_VM_USER |
NVOBJ_FLAG_ZERO_ALLOC, &ctx);
if (ret)
return ret;
nv_wo32(ramin, pcopy->ctx + 0, lower_32_bits(ctx->linst));
nv_wo32(ramin, pcopy->ctx + 4, upper_32_bits(ctx->linst));
dev_priv->engine.instmem.flush(dev);
chan->engctx[engine] = ctx;
return 0;
}
nouveau_gpuobj_ref(NULL, &grctx);
atomic_dec(&chan->vm->engref[engine]);
chan->engctx[engine] = NULL;
}
static int
nv50_graph_object_new(struct nouveau_channel *chan, int engine,
u32 handle, u16 class)
{
struct drm_device *dev = chan->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *obj = NULL;
int ret;
ret = nouveau_gpuobj_new(dev, chan, 16, 16, NVOBJ_FLAG_ZERO_FREE, &obj);
if (ret)
return ret;
obj->engine = 1;
obj->class = class;
nv_wo32(obj, 0x00, class);
nv_wo32(obj, 0x04, 0x00000000);
nv_wo32(obj, 0x08, 0x00000000);
nv_wo32(obj, 0x0c, 0x00000000);
dev_priv->engine.instmem.flush(dev);
ret = nouveau_ramht_insert(chan, handle, obj);
nouveau_gpuobj_ref(NULL, &obj);
return ret;
}
static int
nvc0_dmaobj_bind(struct nouveau_dmaeng *dmaeng,
struct nouveau_object *parent,
struct nouveau_dmaobj *dmaobj,
struct nouveau_gpuobj **pgpuobj)
{
u32 flags0 = nv_mclass(dmaobj);
u32 flags5 = 0x00000000;
int ret;
if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
switch (nv_mclass(parent->parent)) {
case NVA3_DISP_MAST_CLASS:
case NVA3_DISP_SYNC_CLASS:
case NVA3_DISP_OVLY_CLASS:
break;
default:
return -EINVAL;
}
} else
return 0;
if (!(dmaobj->conf0 & NVC0_DMA_CONF0_ENABLE)) {
if (dmaobj->target == NV_MEM_TARGET_VM) {
dmaobj->conf0 = NVC0_DMA_CONF0_PRIV_VM;
dmaobj->conf0 |= NVC0_DMA_CONF0_TYPE_VM;
} else {
dmaobj->conf0 = NVC0_DMA_CONF0_PRIV_US;
dmaobj->conf0 |= NVC0_DMA_CONF0_TYPE_LINEAR;
dmaobj->conf0 |= 0x00020000;
}
}
flags0 |= (dmaobj->conf0 & NVC0_DMA_CONF0_TYPE) << 22;
flags0 |= (dmaobj->conf0 & NVC0_DMA_CONF0_PRIV);
flags5 |= (dmaobj->conf0 & NVC0_DMA_CONF0_UNKN);
switch (dmaobj->target) {
case NV_MEM_TARGET_VM:
flags0 |= 0x00000000;
break;
case NV_MEM_TARGET_VRAM:
flags0 |= 0x00010000;
break;
case NV_MEM_TARGET_PCI:
flags0 |= 0x00020000;
break;
case NV_MEM_TARGET_PCI_NOSNOOP:
flags0 |= 0x00030000;
break;
default:
return -EINVAL;
}
switch (dmaobj->access) {
case NV_MEM_ACCESS_VM:
break;
case NV_MEM_ACCESS_RO:
flags0 |= 0x00040000;
break;
case NV_MEM_ACCESS_WO:
case NV_MEM_ACCESS_RW:
flags0 |= 0x00080000;
break;
}
ret = nouveau_gpuobj_new(parent, parent, 24, 32, 0, pgpuobj);
if (ret == 0) {
nv_wo32(*pgpuobj, 0x00, flags0);
nv_wo32(*pgpuobj, 0x04, lower_32_bits(dmaobj->limit));
nv_wo32(*pgpuobj, 0x08, lower_32_bits(dmaobj->start));
nv_wo32(*pgpuobj, 0x0c, upper_32_bits(dmaobj->limit) << 24 |
upper_32_bits(dmaobj->start));
nv_wo32(*pgpuobj, 0x10, 0x00000000);
nv_wo32(*pgpuobj, 0x14, flags5);
}
return ret;
}
static void
nouveau_accel_init(struct nouveau_drm *drm)
{
struct nouveau_device *device = nv_device(drm->device);
struct nouveau_object *object;
u32 arg0, arg1;
int ret;
if (nouveau_noaccel || !nouveau_fifo(device) /*XXX*/)
return;
/* initialise synchronisation routines */
if (device->card_type < NV_10) ret = nv04_fence_create(drm);
else if (device->card_type < NV_11 ||
device->chipset < 0x17) ret = nv10_fence_create(drm);
else if (device->card_type < NV_50) ret = nv17_fence_create(drm);
else if (device->chipset < 0x84) ret = nv50_fence_create(drm);
else if (device->card_type < NV_C0) ret = nv84_fence_create(drm);
else ret = nvc0_fence_create(drm);
if (ret) {
NV_ERROR(drm, "failed to initialise sync subsystem, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->card_type >= NV_E0) {
ret = nouveau_channel_new(drm, &drm->client, NVDRM_DEVICE,
NVDRM_CHAN + 1,
NVE0_CHANNEL_IND_ENGINE_CE0 |
NVE0_CHANNEL_IND_ENGINE_CE1, 0,
&drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = NVE0_CHANNEL_IND_ENGINE_GR;
arg1 = 1;
} else
if (device->chipset >= 0xa3 &&
device->chipset != 0xaa &&
device->chipset != 0xac) {
ret = nouveau_channel_new(drm, &drm->client, NVDRM_DEVICE,
NVDRM_CHAN + 1, NvDmaFB, NvDmaTT,
&drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = NvDmaFB;
arg1 = NvDmaTT;
} else {
arg0 = NvDmaFB;
arg1 = NvDmaTT;
}
ret = nouveau_channel_new(drm, &drm->client, NVDRM_DEVICE, NVDRM_CHAN,
arg0, arg1, &drm->channel);
if (ret) {
NV_ERROR(drm, "failed to create kernel channel, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nouveau_object_new(nv_object(drm), NVDRM_CHAN, NVDRM_NVSW,
nouveau_abi16_swclass(drm), NULL, 0, &object);
if (ret == 0) {
struct nouveau_software_chan *swch = (void *)object->parent;
ret = RING_SPACE(drm->channel, 2);
if (ret == 0) {
if (device->card_type < NV_C0) {
BEGIN_NV04(drm->channel, NvSubSw, 0, 1);
OUT_RING (drm->channel, NVDRM_NVSW);
} else
if (device->card_type < NV_E0) {
BEGIN_NVC0(drm->channel, FermiSw, 0, 1);
OUT_RING (drm->channel, 0x001f0000);
}
}
swch = (void *)object->parent;
swch->flip = nouveau_flip_complete;
swch->flip_data = drm->channel;
}
if (ret) {
NV_ERROR(drm, "failed to allocate software object, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->card_type < NV_C0) {
ret = nouveau_gpuobj_new(drm->device, NULL, 32, 0, 0,
&drm->notify);
if (ret) {
NV_ERROR(drm, "failed to allocate notifier, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nouveau_object_new(nv_object(drm),
drm->channel->handle, NvNotify0,
0x003d, &(struct nv_dma_class) {
.flags = NV_DMA_TARGET_VRAM |
NV_DMA_ACCESS_RDWR,
.start = drm->notify->addr,
.limit = drm->notify->addr + 31
}, sizeof(struct nv_dma_class),
static void
nouveau_accel_init(struct nouveau_drm *drm)
{
struct nouveau_device *device = nv_device(drm->device);
struct nouveau_object *object;
u32 arg0, arg1;
int ret;
if (nouveau_noaccel)
return;
/* initialise synchronisation routines */
if (device->card_type < NV_10) ret = nv04_fence_create(drm);
else if (device->card_type < NV_50) ret = nv10_fence_create(drm);
else if (device->chipset < 0x84) ret = nv50_fence_create(drm);
else if (device->card_type < NV_C0) ret = nv84_fence_create(drm);
else ret = nvc0_fence_create(drm);
if (ret) {
NV_ERROR(drm, "failed to initialise sync subsystem, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->card_type >= NV_E0) {
ret = nouveau_channel_new(drm, &drm->client, NVDRM_DEVICE,
NVDRM_CHAN + 1,
NVE0_CHANNEL_IND_ENGINE_CE0 |
NVE0_CHANNEL_IND_ENGINE_CE1, 0,
&drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = NVE0_CHANNEL_IND_ENGINE_GR;
arg1 = 0;
} else {
arg0 = NvDmaFB;
arg1 = NvDmaTT;
}
ret = nouveau_channel_new(drm, &drm->client, NVDRM_DEVICE, NVDRM_CHAN,
arg0, arg1, &drm->channel);
if (ret) {
NV_ERROR(drm, "failed to create kernel channel, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->card_type < NV_C0) {
ret = nouveau_gpuobj_new(drm->device, NULL, 32, 0, 0,
&drm->notify);
if (ret) {
NV_ERROR(drm, "failed to allocate notifier, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nouveau_object_new(nv_object(drm),
drm->channel->handle, NvNotify0,
0x003d, &(struct nv_dma_class) {
.flags = NV_DMA_TARGET_VRAM |
NV_DMA_ACCESS_RDWR,
.start = drm->notify->addr,
.limit = drm->notify->addr + 31
}, sizeof(struct nv_dma_class),
static int
nv50_bar_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
struct nouveau_oclass *oclass, void *data, u32 size,
struct nouveau_object **pobject)
{
struct nouveau_device *device = nv_device(parent);
struct nouveau_object *heap;
struct nouveau_vm *vm;
struct nv50_bar_priv *priv;
u64 start, limit;
int ret;
ret = nouveau_bar_create(parent, engine, oclass, &priv);
*pobject = nv_object(priv);
if (ret)
return ret;
ret = nouveau_gpuobj_new(nv_object(priv), NULL, 0x20000, 0,
NVOBJ_FLAG_HEAP, &priv->mem);
heap = nv_object(priv->mem);
if (ret)
return ret;
ret = nouveau_gpuobj_new(nv_object(priv), heap,
(device->chipset == 0x50) ? 0x1400 : 0x0200,
0, 0, &priv->pad);
if (ret)
return ret;
ret = nouveau_gpuobj_new(nv_object(priv), heap, 0x4000, 0,
0, &priv->pgd);
if (ret)
return ret;
/* BAR3 */
start = 0x0100000000ULL;
limit = start + nv_device_resource_len(device, 3);
ret = nouveau_vm_new(device, start, limit, start, &vm);
if (ret)
return ret;
atomic_inc(&vm->engref[NVDEV_SUBDEV_BAR]);
ret = nouveau_gpuobj_new(nv_object(priv), heap,
((limit-- - start) >> 12) * 8, 0x1000,
NVOBJ_FLAG_ZERO_ALLOC, &vm->pgt[0].obj[0]);
vm->pgt[0].refcount[0] = 1;
if (ret)
return ret;
ret = nouveau_vm_ref(vm, &priv->bar3_vm, priv->pgd);
nouveau_vm_ref(NULL, &vm, NULL);
if (ret)
return ret;
ret = nouveau_gpuobj_new(nv_object(priv), heap, 24, 16, 0, &priv->bar3);
if (ret)
return ret;
nv_wo32(priv->bar3, 0x00, 0x7fc00000);
nv_wo32(priv->bar3, 0x04, lower_32_bits(limit));
nv_wo32(priv->bar3, 0x08, lower_32_bits(start));
nv_wo32(priv->bar3, 0x0c, upper_32_bits(limit) << 24 |
upper_32_bits(start));
nv_wo32(priv->bar3, 0x10, 0x00000000);
nv_wo32(priv->bar3, 0x14, 0x00000000);
/* BAR1 */
start = 0x0000000000ULL;
limit = start + nv_device_resource_len(device, 1);
ret = nouveau_vm_new(device, start, limit--, start, &vm);
if (ret)
return ret;
atomic_inc(&vm->engref[NVDEV_SUBDEV_BAR]);
ret = nouveau_vm_ref(vm, &priv->bar1_vm, priv->pgd);
nouveau_vm_ref(NULL, &vm, NULL);
if (ret)
return ret;
ret = nouveau_gpuobj_new(nv_object(priv), heap, 24, 16, 0, &priv->bar1);
if (ret)
return ret;
nv_wo32(priv->bar1, 0x00, 0x7fc00000);
nv_wo32(priv->bar1, 0x04, lower_32_bits(limit));
nv_wo32(priv->bar1, 0x08, lower_32_bits(start));
nv_wo32(priv->bar1, 0x0c, upper_32_bits(limit) << 24 |
upper_32_bits(start));
nv_wo32(priv->bar1, 0x10, 0x00000000);
nv_wo32(priv->bar1, 0x14, 0x00000000);
priv->base.alloc = nouveau_bar_alloc;
priv->base.kmap = nv50_bar_kmap;
priv->base.umap = nv50_bar_umap;
priv->base.unmap = nv50_bar_unmap;
if (device->chipset == 0x50)
priv->base.flush = nv50_bar_flush;
else
priv->base.flush = nv84_bar_flush;
spin_lock_init(&priv->lock);
return 0;
}
int
nvc0_graph_context_ctor(struct nouveau_object *parent,
struct nouveau_object *engine,
struct nouveau_oclass *oclass, void *args, u32 size,
struct nouveau_object **pobject)
{
struct nouveau_vm *vm = nouveau_client(parent)->vm;
struct nvc0_graph_priv *priv = (void *)engine;
struct nvc0_graph_data *data = priv->mmio_data;
struct nvc0_graph_mmio *mmio = priv->mmio_list;
struct nvc0_graph_chan *chan;
int ret, i;
/* allocate memory for context, and fill with default values */
ret = nouveau_graph_context_create(parent, engine, oclass, NULL,
priv->size, 0x100,
NVOBJ_FLAG_ZERO_ALLOC, &chan);
*pobject = nv_object(chan);
if (ret)
return ret;
/* allocate memory for a "mmio list" buffer that's used by the HUB
* fuc to modify some per-context register settings on first load
* of the context.
*/
ret = nouveau_gpuobj_new(nv_object(chan), NULL, 0x1000, 0x100, 0,
&chan->mmio);
if (ret)
return ret;
ret = nouveau_gpuobj_map_vm(nv_gpuobj(chan->mmio), vm,
NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS,
&chan->mmio_vma);
if (ret)
return ret;
/* allocate buffers referenced by mmio list */
for (i = 0; data->size && i < ARRAY_SIZE(priv->mmio_data); i++) {
ret = nouveau_gpuobj_new(nv_object(chan), NULL, data->size,
data->align, 0, &chan->data[i].mem);
if (ret)
return ret;
ret = nouveau_gpuobj_map_vm(chan->data[i].mem, vm, data->access,
&chan->data[i].vma);
if (ret)
return ret;
data++;
}
/* finally, fill in the mmio list and point the context at it */
for (i = 0; mmio->addr && i < ARRAY_SIZE(priv->mmio_list); i++) {
u32 addr = mmio->addr;
u32 data = mmio->data;
if (mmio->shift) {
u64 info = chan->data[mmio->buffer].vma.offset;
data |= info >> mmio->shift;
}
nv_wo32(chan->mmio, chan->mmio_nr++ * 4, addr);
nv_wo32(chan->mmio, chan->mmio_nr++ * 4, data);
mmio++;
}
static int
nvc0_bar_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
struct nouveau_oclass *oclass, void *data, u32 size,
struct nouveau_object **pobject)
{
struct nouveau_device *device = nv_device(parent);
struct pci_dev *pdev = device->pdev;
struct nvc0_bar_priv *priv;
struct nouveau_gpuobj *mem;
struct nouveau_vm *vm;
int ret;
ret = nouveau_bar_create(parent, engine, oclass, &priv);
*pobject = nv_object(priv);
if (ret)
return ret;
/* BAR3 */
ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 0, 0, &priv->bar[0].mem);
mem = priv->bar[0].mem;
if (ret)
return ret;
ret = nouveau_gpuobj_new(parent, NULL, 0x8000, 0, 0, &priv->bar[0].pgd);
if (ret)
return ret;
ret = nouveau_vm_new(device, 0, pci_resource_len(pdev, 3), 0, &vm);
if (ret)
return ret;
ret = nouveau_gpuobj_new(parent, NULL,
(pci_resource_len(pdev, 3) >> 12) * 8,
0x1000, NVOBJ_FLAG_ZERO_ALLOC,
&vm->pgt[0].obj[0]);
vm->pgt[0].refcount[0] = 1;
if (ret)
return ret;
ret = nouveau_vm_ref(vm, &priv->bar[0].vm, priv->bar[0].pgd);
nouveau_vm_ref(NULL, &vm, NULL);
if (ret)
return ret;
nv_wo32(mem, 0x0200, lower_32_bits(priv->bar[0].pgd->addr));
nv_wo32(mem, 0x0204, upper_32_bits(priv->bar[0].pgd->addr));
nv_wo32(mem, 0x0208, lower_32_bits(pci_resource_len(pdev, 3) - 1));
nv_wo32(mem, 0x020c, upper_32_bits(pci_resource_len(pdev, 3) - 1));
/* BAR1 */
ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 0, 0, &priv->bar[1].mem);
mem = priv->bar[1].mem;
if (ret)
return ret;
ret = nouveau_gpuobj_new(parent, NULL, 0x8000, 0, 0, &priv->bar[1].pgd);
if (ret)
return ret;
ret = nouveau_vm_new(device, 0, pci_resource_len(pdev, 1), 0, &vm);
if (ret)
return ret;
ret = nouveau_vm_ref(vm, &priv->bar[1].vm, priv->bar[1].pgd);
nouveau_vm_ref(NULL, &vm, NULL);
if (ret)
return ret;
nv_wo32(mem, 0x0200, lower_32_bits(priv->bar[1].pgd->addr));
nv_wo32(mem, 0x0204, upper_32_bits(priv->bar[1].pgd->addr));
nv_wo32(mem, 0x0208, lower_32_bits(pci_resource_len(pdev, 1) - 1));
nv_wo32(mem, 0x020c, upper_32_bits(pci_resource_len(pdev, 1) - 1));
priv->base.alloc = nouveau_bar_alloc;
priv->base.kmap = nvc0_bar_kmap;
priv->base.umap = nvc0_bar_umap;
priv->base.unmap = nvc0_bar_unmap;
priv->base.flush = nv84_bar_flush;
spin_lock_init(&priv->lock);
return 0;
}
int
nv20_graph_init(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
uint32_t tmp, vramsz;
int ret, i;
switch (dev_priv->chipset) {
case 0x20:
pgraph->grctx_size = NV20_GRCTX_SIZE;
break;
case 0x25:
case 0x28:
pgraph->grctx_size = NV25_GRCTX_SIZE;
break;
case 0x2a:
pgraph->grctx_size = NV2A_GRCTX_SIZE;
break;
default:
NV_ERROR(dev, "unknown chipset, disabling acceleration\n");
pgraph->accel_blocked = true;
return 0;
}
nv_wr32(dev, NV03_PMC_ENABLE,
nv_rd32(dev, NV03_PMC_ENABLE) & ~NV_PMC_ENABLE_PGRAPH);
nv_wr32(dev, NV03_PMC_ENABLE,
nv_rd32(dev, NV03_PMC_ENABLE) | NV_PMC_ENABLE_PGRAPH);
if (!pgraph->ctx_table) {
/* Create Context Pointer Table */
ret = nouveau_gpuobj_new(dev, NULL, 32 * 4, 16,
NVOBJ_FLAG_ZERO_ALLOC,
&pgraph->ctx_table);
if (ret)
return ret;
}
nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE,
pgraph->ctx_table->pinst >> 4);
nv20_graph_rdi(dev);
nv_wr32(dev, NV03_PGRAPH_INTR , 0xFFFFFFFF);
nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
nv_wr32(dev, NV04_PGRAPH_DEBUG_0, 0xFFFFFFFF);
nv_wr32(dev, NV04_PGRAPH_DEBUG_0, 0x00000000);
nv_wr32(dev, NV04_PGRAPH_DEBUG_1, 0x00118700);
nv_wr32(dev, NV04_PGRAPH_DEBUG_3, 0xF3CE0475); /* 0x4 = auto ctx switch */
nv_wr32(dev, NV10_PGRAPH_DEBUG_4, 0x00000000);
nv_wr32(dev, 0x40009C , 0x00000040);
if (dev_priv->chipset >= 0x25) {
nv_wr32(dev, 0x400890, 0x00080000);
nv_wr32(dev, 0x400610, 0x304B1FB6);
nv_wr32(dev, 0x400B80, 0x18B82880);
nv_wr32(dev, 0x400B84, 0x44000000);
nv_wr32(dev, 0x400098, 0x40000080);
nv_wr32(dev, 0x400B88, 0x000000ff);
} else {
nv_wr32(dev, 0x400880, 0x00080000); /* 0x0008c7df */
nv_wr32(dev, 0x400094, 0x00000005);
nv_wr32(dev, 0x400B80, 0x45CAA208); /* 0x45eae20e */
nv_wr32(dev, 0x400B84, 0x24000000);
nv_wr32(dev, 0x400098, 0x00000040);
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00E00038);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA , 0x00000030);
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00E10038);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA , 0x00000030);
}
/* Turn all the tiling regions off. */
for (i = 0; i < NV10_PFB_TILE__SIZE; i++)
nv20_graph_set_region_tiling(dev, i, 0, 0, 0);
for (i = 0; i < 8; i++) {
nv_wr32(dev, 0x400980 + i * 4, nv_rd32(dev, 0x100300 + i * 4));
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0090 + i * 4);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA,
nv_rd32(dev, 0x100300 + i * 4));
}
nv_wr32(dev, 0x4009a0, nv_rd32(dev, 0x100324));
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA000C);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA, nv_rd32(dev, 0x100324));
nv_wr32(dev, NV10_PGRAPH_CTX_CONTROL, 0x10000100);
nv_wr32(dev, NV10_PGRAPH_STATE , 0xFFFFFFFF);
tmp = nv_rd32(dev, NV10_PGRAPH_SURFACE) & 0x0007ff00;
nv_wr32(dev, NV10_PGRAPH_SURFACE, tmp);
tmp = nv_rd32(dev, NV10_PGRAPH_SURFACE) | 0x00020100;
nv_wr32(dev, NV10_PGRAPH_SURFACE, tmp);
/* begin RAM config */
vramsz = pci_resource_len(dev->pdev, 0) - 1;
nv_wr32(dev, 0x4009A4, nv_rd32(dev, NV04_PFB_CFG0));
nv_wr32(dev, 0x4009A8, nv_rd32(dev, NV04_PFB_CFG1));
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0000);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA , nv_rd32(dev, NV04_PFB_CFG0));
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0004);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA , nv_rd32(dev, NV04_PFB_CFG1));
nv_wr32(dev, 0x400820, 0);
nv_wr32(dev, 0x400824, 0);
nv_wr32(dev, 0x400864, vramsz - 1);
nv_wr32(dev, 0x400868, vramsz - 1);
/* interesting.. the below overwrites some of the tile setup above.. */
nv_wr32(dev, 0x400B20, 0x00000000);
nv_wr32(dev, 0x400B04, 0xFFFFFFFF);
nv_wr32(dev, NV03_PGRAPH_ABS_UCLIP_XMIN, 0);
nv_wr32(dev, NV03_PGRAPH_ABS_UCLIP_YMIN, 0);
nv_wr32(dev, NV03_PGRAPH_ABS_UCLIP_XMAX, 0x7fff);
nv_wr32(dev, NV03_PGRAPH_ABS_UCLIP_YMAX, 0x7fff);
return 0;
}
int
nv30_graph_init(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
int ret, i;
switch (dev_priv->chipset) {
case 0x30:
case 0x31:
pgraph->grctx_size = NV30_31_GRCTX_SIZE;
break;
case 0x34:
pgraph->grctx_size = NV34_GRCTX_SIZE;
break;
case 0x35:
case 0x36:
pgraph->grctx_size = NV35_36_GRCTX_SIZE;
break;
default:
NV_ERROR(dev, "unknown chipset, disabling acceleration\n");
pgraph->accel_blocked = true;
return 0;
}
nv_wr32(dev, NV03_PMC_ENABLE,
nv_rd32(dev, NV03_PMC_ENABLE) & ~NV_PMC_ENABLE_PGRAPH);
nv_wr32(dev, NV03_PMC_ENABLE,
nv_rd32(dev, NV03_PMC_ENABLE) | NV_PMC_ENABLE_PGRAPH);
if (!pgraph->ctx_table) {
/* Create Context Pointer Table */
ret = nouveau_gpuobj_new(dev, NULL, 32 * 4, 16,
NVOBJ_FLAG_ZERO_ALLOC,
&pgraph->ctx_table);
if (ret)
return ret;
}
nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE,
pgraph->ctx_table->pinst >> 4);
nv_wr32(dev, NV03_PGRAPH_INTR , 0xFFFFFFFF);
nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
nv_wr32(dev, NV04_PGRAPH_DEBUG_0, 0xFFFFFFFF);
nv_wr32(dev, NV04_PGRAPH_DEBUG_0, 0x00000000);
nv_wr32(dev, NV04_PGRAPH_DEBUG_1, 0x401287c0);
nv_wr32(dev, 0x400890, 0x01b463ff);
nv_wr32(dev, NV04_PGRAPH_DEBUG_3, 0xf2de0475);
nv_wr32(dev, NV10_PGRAPH_DEBUG_4, 0x00008000);
nv_wr32(dev, NV04_PGRAPH_LIMIT_VIOL_PIX, 0xf04bdff6);
nv_wr32(dev, 0x400B80, 0x1003d888);
nv_wr32(dev, 0x400B84, 0x0c000000);
nv_wr32(dev, 0x400098, 0x00000000);
nv_wr32(dev, 0x40009C, 0x0005ad00);
nv_wr32(dev, 0x400B88, 0x62ff00ff); /* suspiciously like PGRAPH_DEBUG_2 */
nv_wr32(dev, 0x4000a0, 0x00000000);
nv_wr32(dev, 0x4000a4, 0x00000008);
nv_wr32(dev, 0x4008a8, 0xb784a400);
nv_wr32(dev, 0x400ba0, 0x002f8685);
nv_wr32(dev, 0x400ba4, 0x00231f3f);
nv_wr32(dev, 0x4008a4, 0x40000020);
if (dev_priv->chipset == 0x34) {
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0004);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA , 0x00200201);
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0008);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA , 0x00000008);
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0000);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA , 0x00000032);
nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00E00004);
nv_wr32(dev, NV10_PGRAPH_RDI_DATA , 0x00000002);
}
nv_wr32(dev, 0x4000c0, 0x00000016);
/* Turn all the tiling regions off. */
for (i = 0; i < NV10_PFB_TILE__SIZE; i++)
nv20_graph_set_region_tiling(dev, i, 0, 0, 0);
nv_wr32(dev, NV10_PGRAPH_CTX_CONTROL, 0x10000100);
nv_wr32(dev, NV10_PGRAPH_STATE , 0xFFFFFFFF);
nv_wr32(dev, 0x0040075c , 0x00000001);
/* begin RAM config */
/* vramsz = pci_resource_len(dev->pdev, 0) - 1; */
nv_wr32(dev, 0x4009A4, nv_rd32(dev, NV04_PFB_CFG0));
nv_wr32(dev, 0x4009A8, nv_rd32(dev, NV04_PFB_CFG1));
if (dev_priv->chipset != 0x34) {
nv_wr32(dev, 0x400750, 0x00EA0000);
nv_wr32(dev, 0x400754, nv_rd32(dev, NV04_PFB_CFG0));
nv_wr32(dev, 0x400750, 0x00EA0004);
nv_wr32(dev, 0x400754, nv_rd32(dev, NV04_PFB_CFG1));
}
return 0;
}
int
nvc0_instmem_init(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
struct pci_dev *pdev = dev->pdev;
struct nvc0_instmem_priv *priv;
struct nouveau_vm *vm = NULL;
int ret;
priv = kzalloc(sizeof(*priv), GFP_KERNEL);
if (!priv)
return -ENOMEM;
pinstmem->priv = priv;
/* BAR3 VM */
ret = nouveau_vm_new(dev, 0, pci_resource_len(pdev, 3), 0,
&dev_priv->bar3_vm);
if (ret)
goto error;
ret = nouveau_gpuobj_new(dev, NULL,
(pci_resource_len(pdev, 3) >> 12) * 8, 0,
NVOBJ_FLAG_DONT_MAP |
NVOBJ_FLAG_ZERO_ALLOC,
&dev_priv->bar3_vm->pgt[0].obj[0]);
if (ret)
goto error;
dev_priv->bar3_vm->pgt[0].refcount[0] = 1;
nv50_instmem_map(dev_priv->bar3_vm->pgt[0].obj[0]);
ret = nouveau_gpuobj_new(dev, NULL, 0x8000, 4096,
NVOBJ_FLAG_ZERO_ALLOC, &priv->bar3_pgd);
if (ret)
goto error;
ret = nouveau_vm_ref(dev_priv->bar3_vm, &vm, priv->bar3_pgd);
if (ret)
goto error;
nouveau_vm_ref(NULL, &vm, NULL);
ret = nvc0_channel_new(dev, 8192, dev_priv->bar3_vm, &priv->bar3,
priv->bar3_pgd, pci_resource_len(dev->pdev, 3));
if (ret)
goto error;
/* BAR1 VM */
ret = nouveau_vm_new(dev, 0, pci_resource_len(pdev, 1), 0, &vm);
if (ret)
goto error;
ret = nouveau_gpuobj_new(dev, NULL, 0x8000, 4096,
NVOBJ_FLAG_ZERO_ALLOC, &priv->bar1_pgd);
if (ret)
goto error;
ret = nouveau_vm_ref(vm, &dev_priv->bar1_vm, priv->bar1_pgd);
if (ret)
goto error;
nouveau_vm_ref(NULL, &vm, NULL);
ret = nvc0_channel_new(dev, 8192, dev_priv->bar1_vm, &priv->bar1,
priv->bar1_pgd, pci_resource_len(dev->pdev, 1));
if (ret)
goto error;
/* channel vm */
ret = nouveau_vm_new(dev, 0, (1ULL << 40), 0x0008000000ULL, &vm);
if (ret)
goto error;
ret = nouveau_gpuobj_new(dev, NULL, 0x8000, 4096, 0, &priv->chan_pgd);
if (ret)
goto error;
nouveau_vm_ref(vm, &dev_priv->chan_vm, priv->chan_pgd);
nouveau_vm_ref(NULL, &vm, NULL);
nvc0_instmem_resume(dev);
return 0;
error:
nvc0_instmem_takedown(dev);
return ret;
}
static int
nv40_instmem_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
struct nouveau_oclass *oclass, void *data, u32 size,
struct nouveau_object **pobject)
{
struct nouveau_device *device = nv_device(parent);
struct nv04_instmem_priv *priv;
int ret, bar, vs;
ret = nouveau_instmem_create(parent, engine, oclass, &priv);
*pobject = nv_object(priv);
if (ret)
return ret;
/* map bar */
if (nv_device_resource_len(device, 2))
bar = 2;
else
bar = 3;
#ifdef __NetBSD__
priv->iomemt = nv_device_resource_tag(device, bar);
priv->iomemsz = nv_device_resource_len(device, bar);
ret = bus_space_map(priv->iomemt,
nv_device_resource_start(device, bar),
priv->iomemsz, 0, &priv->iomemh);
if (ret) {
priv->iomemsz = 0;
nv_error(priv, "unable to map PRAMIN BAR: %d\n", ret);
return -EFAULT;
}
#else
priv->iomem = ioremap(nv_device_resource_start(device, bar),
nv_device_resource_len(device, bar));
if (!priv->iomem) {
nv_error(priv, "unable to map PRAMIN BAR\n");
return -EFAULT;
}
#endif
/* PRAMIN aperture maps over the end of vram, reserve enough space
* to fit graphics contexts for every channel, the magics come
* from engine/graph/nv40.c
*/
vs = hweight8((nv_rd32(priv, 0x001540) & 0x0000ff00) >> 8);
if (device->chipset == 0x40) priv->base.reserved = 0x6aa0 * vs;
else if (device->chipset < 0x43) priv->base.reserved = 0x4f00 * vs;
else if (nv44_graph_class(priv)) priv->base.reserved = 0x4980 * vs;
else priv->base.reserved = 0x4a40 * vs;
priv->base.reserved += 16 * 1024;
priv->base.reserved *= 32; /* per-channel */
priv->base.reserved += 512 * 1024; /* pci(e)gart table */
priv->base.reserved += 512 * 1024; /* object storage */
priv->base.reserved = round_up(priv->base.reserved, 4096);
ret = nouveau_mm_init(&priv->heap, 0, priv->base.reserved, 1);
if (ret)
return ret;
/* 0x00000-0x10000: reserve for probable vbios image */
ret = nouveau_gpuobj_new(nv_object(priv), NULL, 0x10000, 0, 0,
&priv->vbios);
if (ret)
return ret;
/* 0x10000-0x18000: reserve for RAMHT */
ret = nouveau_ramht_new(nv_object(priv), NULL, 0x08000, 0,
&priv->ramht);
if (ret)
return ret;
/* 0x18000-0x18200: reserve for RAMRO
* 0x18200-0x20000: padding
*/
ret = nouveau_gpuobj_new(nv_object(priv), NULL, 0x08000, 0, 0,
&priv->ramro);
if (ret)
return ret;
/* 0x20000-0x21000: reserve for RAMFC
* 0x21000-0x40000: padding and some unknown crap
*/
ret = nouveau_gpuobj_new(nv_object(priv), NULL, 0x20000, 0,
NVOBJ_FLAG_ZERO_ALLOC, &priv->ramfc);
if (ret)
return ret;
return 0;
}
static void
nouveau_accel_init(struct nouveau_drm *drm)
{
struct nvif_device *device = &drm->device;
u32 arg0, arg1;
u32 sclass[16];
int ret, i;
if (nouveau_noaccel)
return;
/* initialise synchronisation routines */
/*XXX: this is crap, but the fence/channel stuff is a little
* backwards in some places. this will be fixed.
*/
ret = nvif_object_sclass(&device->base, sclass, ARRAY_SIZE(sclass));
if (ret < 0)
return;
for (ret = -ENOSYS, i = 0; ret && i < ARRAY_SIZE(sclass); i++) {
switch (sclass[i]) {
case NV03_CHANNEL_DMA:
ret = nv04_fence_create(drm);
break;
case NV10_CHANNEL_DMA:
ret = nv10_fence_create(drm);
break;
case NV17_CHANNEL_DMA:
case NV40_CHANNEL_DMA:
ret = nv17_fence_create(drm);
break;
case NV50_CHANNEL_GPFIFO:
ret = nv50_fence_create(drm);
break;
case G82_CHANNEL_GPFIFO:
ret = nv84_fence_create(drm);
break;
case FERMI_CHANNEL_GPFIFO:
case KEPLER_CHANNEL_GPFIFO_A:
ret = nvc0_fence_create(drm);
break;
default:
break;
}
}
if (ret) {
NV_ERROR(drm, "failed to initialise sync subsystem, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->info.family >= NV_DEVICE_INFO_V0_KEPLER) {
ret = nouveau_channel_new(drm, &drm->device, NVDRM_CHAN + 1,
KEPLER_CHANNEL_GPFIFO_A_V0_ENGINE_CE0|
KEPLER_CHANNEL_GPFIFO_A_V0_ENGINE_CE1,
0, &drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = KEPLER_CHANNEL_GPFIFO_A_V0_ENGINE_GR;
arg1 = 1;
} else
if (device->info.chipset >= 0xa3 &&
device->info.chipset != 0xaa &&
device->info.chipset != 0xac) {
ret = nouveau_channel_new(drm, &drm->device, NVDRM_CHAN + 1,
NvDmaFB, NvDmaTT, &drm->cechan);
if (ret)
NV_ERROR(drm, "failed to create ce channel, %d\n", ret);
arg0 = NvDmaFB;
arg1 = NvDmaTT;
} else {
arg0 = NvDmaFB;
arg1 = NvDmaTT;
}
ret = nouveau_channel_new(drm, &drm->device, NVDRM_CHAN, arg0, arg1,
&drm->channel);
if (ret) {
NV_ERROR(drm, "failed to create kernel channel, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nvif_object_init(drm->channel->object, NULL, NVDRM_NVSW,
nouveau_abi16_swclass(drm), NULL, 0, &drm->nvsw);
if (ret == 0) {
struct nouveau_software_chan *swch;
ret = RING_SPACE(drm->channel, 2);
if (ret == 0) {
if (device->info.family < NV_DEVICE_INFO_V0_FERMI) {
BEGIN_NV04(drm->channel, NvSubSw, 0, 1);
OUT_RING (drm->channel, NVDRM_NVSW);
} else
if (device->info.family < NV_DEVICE_INFO_V0_KEPLER) {
BEGIN_NVC0(drm->channel, FermiSw, 0, 1);
OUT_RING (drm->channel, 0x001f0000);
}
}
swch = (void *)nvkm_object(&drm->nvsw)->parent;
swch->flip = nouveau_flip_complete;
swch->flip_data = drm->channel;
}
if (ret) {
NV_ERROR(drm, "failed to allocate software object, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
if (device->info.family < NV_DEVICE_INFO_V0_FERMI) {
ret = nouveau_gpuobj_new(nvkm_object(&drm->device), NULL, 32,
0, 0, &drm->notify);
if (ret) {
NV_ERROR(drm, "failed to allocate notifier, %d\n", ret);
nouveau_accel_fini(drm);
return;
}
ret = nvif_object_init(drm->channel->object, NULL, NvNotify0,
NV_DMA_IN_MEMORY,
&(struct nv_dma_v0) {
.target = NV_DMA_V0_TARGET_VRAM,
.access = NV_DMA_V0_ACCESS_RDWR,
.start = drm->notify->addr,
.limit = drm->notify->addr + 31
}, sizeof(struct nv_dma_v0),
