dect
/
linux-2.6
Archived
13
0
Fork 0

drm/nvc0: initial vm implementation, use for bar1/bar3 management

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This commit is contained in:
Ben Skeggs 2010-11-15 11:48:33 +10:00
parent 4c74eb7ff2
commit 8984e04615
8 changed files with 277 additions and 215 deletions

View File

@ -28,7 +28,8 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
nv10_gpio.o nv50_gpio.o \
nv50_calc.o \
nv04_pm.o nv50_pm.o nva3_pm.o \
nv50_vram.o nv50_vm.o nvc0_vm.o
nv50_vram.o nvc0_vram.o \
nv50_vm.o nvc0_vm.o
nouveau-$(CONFIG_DRM_NOUVEAU_DEBUG) += nouveau_debugfs.o
nouveau-$(CONFIG_COMPAT) += nouveau_ioc32.o

View File

@ -120,6 +120,9 @@ nouveau_bo_new(struct drm_device *dev, struct nouveau_channel *chan,
align >>= PAGE_SHIFT;
if (!nvbo->no_vm && dev_priv->chan_vm) {
if (dev_priv->card_type == NV_C0)
page_shift = 12;
ret = nouveau_vm_get(dev_priv->chan_vm, size, page_shift,
NV_MEM_ACCESS_RW, &nvbo->vma);
if (ret) {
@ -413,7 +416,7 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
man->default_caching = TTM_PL_FLAG_CACHED;
break;
case TTM_PL_VRAM:
if (dev_priv->card_type == NV_50) {
if (dev_priv->card_type >= NV_50) {
man->func = &nouveau_vram_manager;
man->io_reserve_fastpath = false;
man->use_io_reserve_lru = true;
@ -901,6 +904,7 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
case TTM_PL_VRAM:
{
struct nouveau_vram *vram = mem->mm_node;
u8 page_shift;
if (!dev_priv->bar1_vm) {
mem->bus.offset = mem->start << PAGE_SHIFT;
@ -909,8 +913,13 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
break;
}
if (dev_priv->card_type == NV_C0)
page_shift = vram->page_shift;
else
page_shift = 12;
ret = nouveau_vm_get(dev_priv->bar1_vm, mem->bus.size,
vram->page_shift, NV_MEM_ACCESS_RW,
page_shift, NV_MEM_ACCESS_RW,
&vram->bar_vma);
if (ret)
return ret;
@ -921,8 +930,9 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
return ret;
}
mem->bus.offset = vram->bar_vma.offset;
mem->bus.offset -= 0x0020000000ULL;
mem->bus.offset = vram->bar_vma.offset;
if (dev_priv->card_type == NV_50) /*XXX*/
mem->bus.offset -= 0x0020000000ULL;
mem->bus.base = pci_resource_start(dev->pdev, 1);
mem->bus.is_iomem = true;
}

View File

@ -842,6 +842,9 @@ extern void nv10_mem_put_tile_region(struct drm_device *dev,
struct nouveau_fence *fence);
extern const struct ttm_mem_type_manager_func nouveau_vram_manager;
/* nvc0_vram.c */
extern const struct ttm_mem_type_manager_func nvc0_vram_manager;
/* nouveau_notifier.c */
extern int nouveau_notifier_init_channel(struct nouveau_channel *);
extern void nouveau_notifier_takedown_channel(struct nouveau_channel *);
@ -1229,11 +1232,6 @@ extern int nvc0_instmem_init(struct drm_device *);
extern void nvc0_instmem_takedown(struct drm_device *);
extern int nvc0_instmem_suspend(struct drm_device *);
extern void nvc0_instmem_resume(struct drm_device *);
extern int nvc0_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
extern void nvc0_instmem_put(struct nouveau_gpuobj *);
extern int nvc0_instmem_map(struct nouveau_gpuobj *);
extern void nvc0_instmem_unmap(struct nouveau_gpuobj *);
extern void nvc0_instmem_flush(struct drm_device *);
/* nv04_mc.c */
extern int nv04_mc_init(struct drm_device *);

View File

@ -255,9 +255,6 @@ nouveau_mem_detect(struct drm_device *dev)
if (dev_priv->card_type < NV_50) {
dev_priv->vram_size = nv_rd32(dev, NV04_PFB_FIFO_DATA);
dev_priv->vram_size &= NV10_PFB_FIFO_DATA_RAM_AMOUNT_MB_MASK;
} else {
dev_priv->vram_size = nv_rd32(dev, 0x10f20c) << 20;
dev_priv->vram_size *= nv_rd32(dev, 0x121c74);
}
if (dev_priv->vram_size)

View File

@ -59,4 +59,9 @@ int nv50_vram_new(struct drm_device *, u64 size, u32 align, u32 size_nc,
void nv50_vram_del(struct drm_device *, struct nouveau_vram **);
bool nv50_vram_flags_valid(struct drm_device *, u32 tile_flags);
int nvc0_vram_init(struct drm_device *);
int nvc0_vram_new(struct drm_device *, u64 size, u32 align, u32 ncmin,
u32 memtype, struct nouveau_vram **);
bool nvc0_vram_flags_valid(struct drm_device *, u32 tile_flags);
#endif

View File

@ -464,11 +464,11 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->instmem.takedown = nvc0_instmem_takedown;
engine->instmem.suspend = nvc0_instmem_suspend;
engine->instmem.resume = nvc0_instmem_resume;
engine->instmem.get = nvc0_instmem_get;
engine->instmem.put = nvc0_instmem_put;
engine->instmem.map = nvc0_instmem_map;
engine->instmem.unmap = nvc0_instmem_unmap;
engine->instmem.flush = nvc0_instmem_flush;
engine->instmem.get = nv50_instmem_get;
engine->instmem.put = nv50_instmem_put;
engine->instmem.map = nv50_instmem_map;
engine->instmem.unmap = nv50_instmem_unmap;
engine->instmem.flush = nv84_instmem_flush;
engine->mc.init = nv50_mc_init;
engine->mc.takedown = nv50_mc_takedown;
engine->timer.init = nv04_timer_init;
@ -509,8 +509,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->gpio.irq_enable = nv50_gpio_irq_enable;
engine->crypt.init = nouveau_stub_init;
engine->crypt.takedown = nouveau_stub_takedown;
engine->vram.init = nouveau_mem_detect;
engine->vram.flags_valid = nouveau_mem_flags_valid;
engine->vram.init = nvc0_vram_init;
engine->vram.get = nvc0_vram_new;
engine->vram.put = nv50_vram_del;
engine->vram.flags_valid = nvc0_vram_flags_valid;
break;
default:
NV_ERROR(dev, "NV%02x unsupported\n", dev_priv->chipset);

View File

@ -25,159 +25,21 @@
#include "drmP.h"
#include "nouveau_drv.h"
#include "nouveau_vm.h"
struct nvc0_gpuobj_node {
struct nouveau_bo *vram;
struct drm_mm_node *ramin;
u32 align;
struct nvc0_instmem_priv {
struct nouveau_gpuobj *bar1_pgd;
struct nouveau_channel *bar1;
struct nouveau_gpuobj *bar3_pgd;
struct nouveau_channel *bar3;
};
int
nvc0_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
{
struct drm_device *dev = gpuobj->dev;
struct nvc0_gpuobj_node *node = NULL;
int ret;
node = kzalloc(sizeof(*node), GFP_KERNEL);
if (!node)
return -ENOMEM;
node->align = align;
ret = nouveau_bo_new(dev, NULL, size, align, TTM_PL_FLAG_VRAM,
0, 0x0000, true, false, &node->vram);
if (ret) {
NV_ERROR(dev, "error getting PRAMIN backing pages: %d\n", ret);
return ret;
}
ret = nouveau_bo_pin(node->vram, TTM_PL_FLAG_VRAM);
if (ret) {
NV_ERROR(dev, "error pinning PRAMIN backing VRAM: %d\n", ret);
nouveau_bo_ref(NULL, &node->vram);
return ret;
}
gpuobj->vinst = node->vram->bo.mem.start << PAGE_SHIFT;
gpuobj->size = node->vram->bo.mem.num_pages << PAGE_SHIFT;
gpuobj->node = node;
return 0;
}
void
nvc0_instmem_put(struct nouveau_gpuobj *gpuobj)
{
struct nvc0_gpuobj_node *node;
node = gpuobj->node;
gpuobj->node = NULL;
nouveau_bo_unpin(node->vram);
nouveau_bo_ref(NULL, &node->vram);
kfree(node);
}
int
nvc0_instmem_map(struct nouveau_gpuobj *gpuobj)
{
struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
struct nvc0_gpuobj_node *node = gpuobj->node;
struct drm_device *dev = gpuobj->dev;
struct drm_mm_node *ramin = NULL;
u32 pte, pte_end;
u64 vram;
do {
if (drm_mm_pre_get(&dev_priv->ramin_heap))
return -ENOMEM;
spin_lock(&dev_priv->ramin_lock);
ramin = drm_mm_search_free(&dev_priv->ramin_heap, gpuobj->size,
node->align, 0);
if (ramin == NULL) {
spin_unlock(&dev_priv->ramin_lock);
return -ENOMEM;
}
ramin = drm_mm_get_block_atomic(ramin, gpuobj->size, node->align);
spin_unlock(&dev_priv->ramin_lock);
} while (ramin == NULL);
pte = (ramin->start >> 12) << 1;
pte_end = ((ramin->size >> 12) << 1) + pte;
vram = gpuobj->vinst;
NV_DEBUG(dev, "pramin=0x%lx, pte=%d, pte_end=%d\n",
ramin->start, pte, pte_end);
NV_DEBUG(dev, "first vram page: 0x%010llx\n", gpuobj->vinst);
while (pte < pte_end) {
nv_wr32(dev, 0x702000 + (pte * 8), (vram >> 8) | 1);
nv_wr32(dev, 0x702004 + (pte * 8), 0);
vram += 4096;
pte++;
}
dev_priv->engine.instmem.flush(dev);
if (1) {
u32 chan = nv_rd32(dev, 0x1700) << 16;
nv_wr32(dev, 0x100cb8, (chan + 0x1000) >> 8);
nv_wr32(dev, 0x100cbc, 0x80000005);
}
node->ramin = ramin;
gpuobj->pinst = ramin->start;
return 0;
}
void
nvc0_instmem_unmap(struct nouveau_gpuobj *gpuobj)
{
struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
struct nvc0_gpuobj_node *node = gpuobj->node;
u32 pte, pte_end;
if (!node->ramin || !dev_priv->ramin_available)
return;
pte = (node->ramin->start >> 12) << 1;
pte_end = ((node->ramin->size >> 12) << 1) + pte;
while (pte < pte_end) {
nv_wr32(gpuobj->dev, 0x702000 + (pte * 8), 0);
nv_wr32(gpuobj->dev, 0x702004 + (pte * 8), 0);
pte++;
}
dev_priv->engine.instmem.flush(gpuobj->dev);
spin_lock(&dev_priv->ramin_lock);
drm_mm_put_block(node->ramin);
node->ramin = NULL;
spin_unlock(&dev_priv->ramin_lock);
}
void
nvc0_instmem_flush(struct drm_device *dev)
{
nv_wr32(dev, 0x070000, 1);
if (!nv_wait(dev, 0x070000, 0x00000002, 0x00000000))
NV_ERROR(dev, "PRAMIN flush timeout\n");
}
int
nvc0_instmem_suspend(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
u32 *buf;
int i;
dev_priv->susres.ramin_copy = vmalloc(65536);
if (!dev_priv->susres.ramin_copy)
return -ENOMEM;
buf = dev_priv->susres.ramin_copy;
for (i = 0; i < 65536; i += 4)
buf[i/4] = nv_rd32(dev, NV04_PRAMIN + i);
dev_priv->ramin_available = false;
return 0;
}
@ -185,73 +47,169 @@ void
nvc0_instmem_resume(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
u32 *buf = dev_priv->susres.ramin_copy;
u64 chan;
int i;
struct nvc0_instmem_priv *priv = dev_priv->engine.instmem.priv;
chan = dev_priv->vram_size - dev_priv->ramin_rsvd_vram;
nv_wr32(dev, 0x001700, chan >> 16);
nv_mask(dev, 0x100c80, 0x00000001, 0x00000000);
nv_wr32(dev, 0x001704, 0x80000000 | priv->bar1->ramin->vinst >> 12);
nv_wr32(dev, 0x001714, 0xc0000000 | priv->bar3->ramin->vinst >> 12);
dev_priv->ramin_available = true;
}
for (i = 0; i < 65536; i += 4)
nv_wr32(dev, NV04_PRAMIN + i, buf[i/4]);
vfree(dev_priv->susres.ramin_copy);
dev_priv->susres.ramin_copy = NULL;
static void
nvc0_channel_del(struct nouveau_channel **pchan)
{
struct nouveau_channel *chan;
nv_wr32(dev, 0x001714, 0xc0000000 | (chan >> 12));
chan = *pchan;
*pchan = NULL;
if (!chan)
return;
nouveau_vm_ref(NULL, &chan->vm, NULL);
if (chan->ramin_heap.free_stack.next)
drm_mm_takedown(&chan->ramin_heap);
nouveau_gpuobj_ref(NULL, &chan->ramin);
kfree(chan);
}
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
nvc0_instmem_init(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
u64 chan, pgt3, imem, lim3 = dev_priv->ramin_size - 1;
int ret, i;
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;
dev_priv->ramin_rsvd_vram = 1 * 1024 * 1024;
chan = dev_priv->vram_size - dev_priv->ramin_rsvd_vram;
imem = 4096 + 4096 + 32768;
nv_wr32(dev, 0x001700, chan >> 16);
/* channel setup */
nv_wr32(dev, 0x700200, lower_32_bits(chan + 0x1000));
nv_wr32(dev, 0x700204, upper_32_bits(chan + 0x1000));
nv_wr32(dev, 0x700208, lower_32_bits(lim3));
nv_wr32(dev, 0x70020c, upper_32_bits(lim3));
/* point pgd -> pgt */
nv_wr32(dev, 0x701000, 0);
nv_wr32(dev, 0x701004, ((chan + 0x2000) >> 8) | 1);
/* point pgt -> physical vram for channel */
pgt3 = 0x2000;
for (i = 0; i < dev_priv->ramin_rsvd_vram; i += 4096, pgt3 += 8) {
nv_wr32(dev, 0x700000 + pgt3, ((chan + i) >> 8) | 1);
nv_wr32(dev, 0x700004 + pgt3, 0);
}
/* clear rest of pgt */
for (; i < dev_priv->ramin_size; i += 4096, pgt3 += 8) {
nv_wr32(dev, 0x700000 + pgt3, 0);
nv_wr32(dev, 0x700004 + pgt3, 0);
}
/* point bar3 at the channel */
nv_wr32(dev, 0x001714, 0xc0000000 | (chan >> 12));
/* Global PRAMIN heap */
ret = drm_mm_init(&dev_priv->ramin_heap, imem,
dev_priv->ramin_size - imem);
if (ret) {
NV_ERROR(dev, "Failed to init RAMIN heap\n");
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;
nvc0_instmem_resume(dev);
return 0;
error:
nvc0_instmem_takedown(dev);
return ret;
}
void
nvc0_instmem_takedown(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nvc0_instmem_priv *priv = dev_priv->engine.instmem.priv;
struct nouveau_vm *vm = NULL;
nvc0_instmem_suspend(dev);
nv_wr32(dev, 0x1704, 0x00000000);
nv_wr32(dev, 0x1714, 0x00000000);
nvc0_channel_del(&priv->bar1);
nouveau_vm_ref(NULL, &dev_priv->bar1_vm, priv->bar1_pgd);
nouveau_gpuobj_ref(NULL, &priv->bar1_pgd);
nvc0_channel_del(&priv->bar3);
nouveau_vm_ref(dev_priv->bar3_vm, &vm, NULL);
nouveau_vm_ref(NULL, &vm, priv->bar3_pgd);
nouveau_gpuobj_ref(NULL, &priv->bar3_pgd);
nouveau_gpuobj_ref(NULL, &dev_priv->bar3_vm->pgt[0].obj[0]);
nouveau_vm_ref(NULL, &dev_priv->bar3_vm, NULL);
dev_priv->engine.instmem.priv = NULL;
kfree(priv);
}

View File

@ -0,0 +1,91 @@
/*
* Copyright 2010 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 "drmP.h"
#include "nouveau_drv.h"
#include "nouveau_mm.h"
bool
nvc0_vram_flags_valid(struct drm_device *dev, u32 tile_flags)
{
if (likely(!(tile_flags & NOUVEAU_GEM_TILE_LAYOUT_MASK)))
return true;
return false;
}
int
nvc0_vram_new(struct drm_device *dev, u64 size, u32 align, u32 ncmin,
u32 type, struct nouveau_vram **pvram)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
struct ttm_mem_type_manager *man = &bdev->man[TTM_PL_VRAM];
struct nouveau_mm *mm = man->priv;
struct nouveau_mm_node *r;
struct nouveau_vram *vram;
int ret;
size >>= 12;
align >>= 12;
ncmin >>= 12;
vram = kzalloc(sizeof(*vram), GFP_KERNEL);
if (!vram)
return -ENOMEM;
INIT_LIST_HEAD(&vram->regions);
vram->dev = dev_priv->dev;
vram->memtype = type;
vram->size = size;
mutex_lock(&mm->mutex);
do {
ret = nouveau_mm_get(mm, 1, size, ncmin, align, &r);
if (ret) {
mutex_unlock(&mm->mutex);
nv50_vram_del(dev, &vram);
return ret;
}
list_add_tail(&r->rl_entry, &vram->regions);
size -= r->length;
} while (size);
mutex_unlock(&mm->mutex);
r = list_first_entry(&vram->regions, struct nouveau_mm_node, rl_entry);
vram->offset = (u64)r->offset << 12;
*pvram = vram;
return 0;
}
int
nvc0_vram_init(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
dev_priv->vram_size = nv_rd32(dev, 0x10f20c) << 20;
dev_priv->vram_size *= nv_rd32(dev, 0x121c74);
dev_priv->vram_rblock_size = 4096;
return 0;
}