Commit f869ef88 authored by Ben Skeggs's avatar Ben Skeggs

drm/nv50: implement BAR1/BAR3 management on top of new VM code

Signed-off-by: default avatarBen Skeggs <bskeggs@redhat.com>
parent a11c3198
...@@ -32,6 +32,8 @@ ...@@ -32,6 +32,8 @@
#include "nouveau_drm.h" #include "nouveau_drm.h"
#include "nouveau_drv.h" #include "nouveau_drv.h"
#include "nouveau_dma.h" #include "nouveau_dma.h"
#include "nouveau_mm.h"
#include "nouveau_vm.h"
#include <linux/log2.h> #include <linux/log2.h>
#include <linux/slab.h> #include <linux/slab.h>
...@@ -386,10 +388,13 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, ...@@ -386,10 +388,13 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
man->default_caching = TTM_PL_FLAG_CACHED; man->default_caching = TTM_PL_FLAG_CACHED;
break; break;
case TTM_PL_VRAM: case TTM_PL_VRAM:
if (dev_priv->card_type == NV_50) if (dev_priv->card_type == NV_50) {
man->func = &nouveau_vram_manager; man->func = &nouveau_vram_manager;
else man->io_reserve_fastpath = false;
man->use_io_reserve_lru = true;
} else {
man->func = &ttm_bo_manager_func; man->func = &ttm_bo_manager_func;
}
man->flags = TTM_MEMTYPE_FLAG_FIXED | man->flags = TTM_MEMTYPE_FLAG_FIXED |
TTM_MEMTYPE_FLAG_MAPPABLE; TTM_MEMTYPE_FLAG_MAPPABLE;
man->available_caching = TTM_PL_FLAG_UNCACHED | man->available_caching = TTM_PL_FLAG_UNCACHED |
...@@ -858,6 +863,7 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) ...@@ -858,6 +863,7 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev); struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
struct drm_device *dev = dev_priv->dev; struct drm_device *dev = dev_priv->dev;
int ret;
mem->bus.addr = NULL; mem->bus.addr = NULL;
mem->bus.offset = 0; mem->bus.offset = 0;
...@@ -880,9 +886,32 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) ...@@ -880,9 +886,32 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
#endif #endif
break; break;
case TTM_PL_VRAM: case TTM_PL_VRAM:
mem->bus.offset = mem->start << PAGE_SHIFT; {
struct nouveau_vram *vram = mem->mm_node;
if (!dev_priv->bar1_vm) {
mem->bus.offset = mem->start << PAGE_SHIFT;
mem->bus.base = pci_resource_start(dev->pdev, 1);
mem->bus.is_iomem = true;
break;
}
ret = nouveau_vm_get(dev_priv->bar1_vm, mem->bus.size, 12,
NV_MEM_ACCESS_RW, &vram->bar_vma);
if (ret)
return ret;
nouveau_vm_map(&vram->bar_vma, vram);
if (ret) {
nouveau_vm_put(&vram->bar_vma);
return ret;
}
mem->bus.offset = vram->bar_vma.offset;
mem->bus.offset -= 0x0020000000ULL;
mem->bus.base = pci_resource_start(dev->pdev, 1); mem->bus.base = pci_resource_start(dev->pdev, 1);
mem->bus.is_iomem = true; mem->bus.is_iomem = true;
}
break; break;
default: default:
return -EINVAL; return -EINVAL;
...@@ -893,6 +922,17 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) ...@@ -893,6 +922,17 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
static void static void
nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
{ {
struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
struct nouveau_vram *vram = mem->mm_node;
if (!dev_priv->bar1_vm || mem->mem_type != TTM_PL_VRAM)
return;
if (!vram->bar_vma.node)
return;
nouveau_vm_unmap(&vram->bar_vma);
nouveau_vm_put(&vram->bar_vma);
} }
static int static int
......
...@@ -55,7 +55,10 @@ struct nouveau_fpriv { ...@@ -55,7 +55,10 @@ struct nouveau_fpriv {
#include "nouveau_reg.h" #include "nouveau_reg.h"
#include "nouveau_bios.h" #include "nouveau_bios.h"
#include "nouveau_util.h" #include "nouveau_util.h"
struct nouveau_grctx; struct nouveau_grctx;
struct nouveau_vram;
#include "nouveau_vm.h"
#define MAX_NUM_DCB_ENTRIES 16 #define MAX_NUM_DCB_ENTRIES 16
...@@ -69,6 +72,8 @@ struct nouveau_grctx; ...@@ -69,6 +72,8 @@ struct nouveau_grctx;
struct nouveau_vram { struct nouveau_vram {
struct drm_device *dev; struct drm_device *dev;
struct nouveau_vma bar_vma;
struct list_head regions; struct list_head regions;
u32 memtype; u32 memtype;
u64 offset; u64 offset;
...@@ -244,6 +249,7 @@ struct nouveau_channel { ...@@ -244,6 +249,7 @@ struct nouveau_channel {
void *pgraph_ctx; void *pgraph_ctx;
/* NV50 VM */ /* NV50 VM */
struct nouveau_vm *vm;
struct nouveau_gpuobj *vm_pd; struct nouveau_gpuobj *vm_pd;
struct nouveau_gpuobj *vm_gart_pt; struct nouveau_gpuobj *vm_gart_pt;
struct nouveau_gpuobj *vm_vram_pt[NV50_VM_VRAM_NR]; struct nouveau_gpuobj *vm_vram_pt[NV50_VM_VRAM_NR];
...@@ -701,6 +707,10 @@ struct drm_nouveau_private { ...@@ -701,6 +707,10 @@ struct drm_nouveau_private {
uint64_t fb_aper_free; uint64_t fb_aper_free;
int fb_mtrr; int fb_mtrr;
/* BAR control (NV50-) */
struct nouveau_vm *bar1_vm;
struct nouveau_vm *bar3_vm;
/* G8x/G9x virtual address space */ /* G8x/G9x virtual address space */
uint64_t vm_gart_base; uint64_t vm_gart_base;
uint64_t vm_gart_size; uint64_t vm_gart_size;
......
...@@ -31,12 +31,16 @@ ...@@ -31,12 +31,16 @@
#include "nouveau_drv.h" #include "nouveau_drv.h"
#include "nouveau_vm.h" #include "nouveau_vm.h"
#define BAR1_VM_BASE 0x0020000000ULL
#define BAR1_VM_SIZE pci_resource_len(dev->pdev, 1)
#define BAR3_VM_BASE 0x0000000000ULL
#define BAR3_VM_SIZE pci_resource_len(dev->pdev, 3)
struct nv50_instmem_priv { struct nv50_instmem_priv {
uint32_t save1700[5]; /* 0x1700->0x1710 */ uint32_t save1700[5]; /* 0x1700->0x1710 */
struct nouveau_gpuobj *pramin_pt; struct nouveau_gpuobj *bar1_dmaobj;
struct nouveau_gpuobj *pramin_bar; struct nouveau_gpuobj *bar3_dmaobj;
struct nouveau_gpuobj *fb_bar;
}; };
static void static void
...@@ -50,6 +54,7 @@ nv50_channel_del(struct nouveau_channel **pchan) ...@@ -50,6 +54,7 @@ nv50_channel_del(struct nouveau_channel **pchan)
return; return;
nouveau_gpuobj_ref(NULL, &chan->ramfc); nouveau_gpuobj_ref(NULL, &chan->ramfc);
nouveau_vm_ref(NULL, &chan->vm, chan->vm_pd);
nouveau_gpuobj_ref(NULL, &chan->vm_pd); nouveau_gpuobj_ref(NULL, &chan->vm_pd);
if (chan->ramin_heap.free_stack.next) if (chan->ramin_heap.free_stack.next)
drm_mm_takedown(&chan->ramin_heap); drm_mm_takedown(&chan->ramin_heap);
...@@ -58,14 +63,14 @@ nv50_channel_del(struct nouveau_channel **pchan) ...@@ -58,14 +63,14 @@ nv50_channel_del(struct nouveau_channel **pchan)
} }
static int static int
nv50_channel_new(struct drm_device *dev, u32 size, nv50_channel_new(struct drm_device *dev, u32 size, struct nouveau_vm *vm,
struct nouveau_channel **pchan) struct nouveau_channel **pchan)
{ {
struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_nouveau_private *dev_priv = dev->dev_private;
u32 pgd = (dev_priv->chipset == 0x50) ? 0x1400 : 0x0200; u32 pgd = (dev_priv->chipset == 0x50) ? 0x1400 : 0x0200;
u32 fc = (dev_priv->chipset == 0x50) ? 0x0000 : 0x4200; u32 fc = (dev_priv->chipset == 0x50) ? 0x0000 : 0x4200;
struct nouveau_channel *chan; struct nouveau_channel *chan;
int ret; int ret, i;
chan = kzalloc(sizeof(*chan), GFP_KERNEL); chan = kzalloc(sizeof(*chan), GFP_KERNEL);
if (!chan) if (!chan)
...@@ -94,6 +99,17 @@ nv50_channel_new(struct drm_device *dev, u32 size, ...@@ -94,6 +99,17 @@ nv50_channel_new(struct drm_device *dev, u32 size,
return ret; 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 : ret = nouveau_gpuobj_new_fake(dev, chan->ramin->pinst == ~0 ? ~0 :
chan->ramin->pinst + fc, chan->ramin->pinst + fc,
chan->ramin->vinst + fc, 0x100, chan->ramin->vinst + fc, 0x100,
...@@ -113,6 +129,7 @@ nv50_instmem_init(struct drm_device *dev) ...@@ -113,6 +129,7 @@ nv50_instmem_init(struct drm_device *dev)
struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nv50_instmem_priv *priv; struct nv50_instmem_priv *priv;
struct nouveau_channel *chan; struct nouveau_channel *chan;
struct nouveau_vm *vm;
int ret, i; int ret, i;
u32 tmp; u32 tmp;
...@@ -129,53 +146,75 @@ nv50_instmem_init(struct drm_device *dev) ...@@ -129,53 +146,75 @@ nv50_instmem_init(struct drm_device *dev)
ret = drm_mm_init(&dev_priv->ramin_heap, 0, dev_priv->ramin_size); ret = drm_mm_init(&dev_priv->ramin_heap, 0, dev_priv->ramin_size);
if (ret) { if (ret) {
NV_ERROR(dev, "Failed to init RAMIN heap\n"); NV_ERROR(dev, "Failed to init RAMIN heap\n");
return -ENOMEM; goto error;
} }
/* we need a channel to plug into the hw to control the BARs */ /* BAR3 */
ret = nv50_channel_new(dev, 128*1024, &dev_priv->channels.ptr[0]); ret = nouveau_vm_new(dev, BAR3_VM_BASE, BAR3_VM_SIZE, BAR3_VM_BASE,
29, 12, 16, &dev_priv->bar3_vm);
if (ret) if (ret)
return ret; goto error;
chan = dev_priv->channels.ptr[127] = dev_priv->channels.ptr[0];
/* allocate page table for PRAMIN BAR */ ret = nouveau_gpuobj_new(dev, NULL, (BAR3_VM_SIZE >> 12) * 8,
ret = nouveau_gpuobj_new(dev, chan, (dev_priv->ramin_size >> 12) * 8, 0x1000, NVOBJ_FLAG_DONT_MAP |
0x1000, NVOBJ_FLAG_ZERO_ALLOC, NVOBJ_FLAG_ZERO_ALLOC,
&priv->pramin_pt); &dev_priv->bar3_vm->pgt[0].obj);
if (ret) if (ret)
return ret; goto error;
dev_priv->bar3_vm->pgt[0].page_shift = 12;
dev_priv->bar3_vm->pgt[0].refcount = 1;
nv_wo32(chan->vm_pd, 0x0000, priv->pramin_pt->vinst | 0x63); nv50_instmem_map(dev_priv->bar3_vm->pgt[0].obj);
nv_wo32(chan->vm_pd, 0x0004, 0);
/* DMA object for PRAMIN BAR */ ret = nv50_channel_new(dev, 128 * 1024, dev_priv->bar3_vm, &chan);
ret = nouveau_gpuobj_new(dev, chan, 6*4, 16, 0, &priv->pramin_bar);
if (ret) if (ret)
return ret; goto error;
nv_wo32(priv->pramin_bar, 0x00, 0x7fc00000); dev_priv->channels.ptr[0] = dev_priv->channels.ptr[127] = chan;
nv_wo32(priv->pramin_bar, 0x04, dev_priv->ramin_size - 1);
nv_wo32(priv->pramin_bar, 0x08, 0x00000000);
nv_wo32(priv->pramin_bar, 0x0c, 0x00000000);
nv_wo32(priv->pramin_bar, 0x10, 0x00000000);
nv_wo32(priv->pramin_bar, 0x14, 0x00000000);
nv50_instmem_map(chan->ramin); ret = nv50_gpuobj_dma_new(chan, 0x0000, BAR3_VM_BASE, BAR3_VM_SIZE,
NV_MEM_TARGET_VM, NV_MEM_ACCESS_VM,
NV_MEM_TYPE_VM, NV_MEM_COMP_VM,
&priv->bar3_dmaobj);
if (ret)
goto error;
/* poke regs... */
nv_wr32(dev, 0x001704, 0x00000000 | (chan->ramin->vinst >> 12)); nv_wr32(dev, 0x001704, 0x00000000 | (chan->ramin->vinst >> 12));
nv_wr32(dev, 0x001704, 0x40000000 | (chan->ramin->vinst >> 12)); nv_wr32(dev, 0x001704, 0x40000000 | (chan->ramin->vinst >> 12));
nv_wr32(dev, 0x00170c, 0x80000000 | (priv->pramin_bar->cinst >> 4)); nv_wr32(dev, 0x00170c, 0x80000000 | (priv->bar3_dmaobj->cinst >> 4));
tmp = nv_ri32(dev, 0); tmp = nv_ri32(dev, 0);
nv_wi32(dev, 0, ~tmp); nv_wi32(dev, 0, ~tmp);
if (nv_ri32(dev, 0) != ~tmp) { if (nv_ri32(dev, 0) != ~tmp) {
NV_ERROR(dev, "PRAMIN readback failed\n"); NV_ERROR(dev, "PRAMIN readback failed\n");
return -EIO; ret = -EIO;
goto error;
} }
nv_wi32(dev, 0, tmp); nv_wi32(dev, 0, tmp);
dev_priv->ramin_available = true; dev_priv->ramin_available = true;
/* BAR1 */
ret = nouveau_vm_new(dev, BAR1_VM_BASE, BAR1_VM_SIZE, BAR1_VM_BASE,
29, 12, 16, &vm);
if (ret)
goto error;
ret = nouveau_vm_ref(vm, &dev_priv->bar1_vm, chan->vm_pd);
if (ret)
goto error;
nouveau_vm_ref(NULL, &vm, NULL);
ret = nv50_gpuobj_dma_new(chan, 0x0000, BAR1_VM_BASE, BAR1_VM_SIZE,
NV_MEM_TARGET_VM, NV_MEM_ACCESS_VM,
NV_MEM_TYPE_VM, NV_MEM_COMP_VM,
&priv->bar1_dmaobj);
if (ret)
goto error;
nv_wr32(dev, 0x001708, 0x80000000 | (priv->bar1_dmaobj->cinst >> 4));
for (i = 0; i < 8; i++)
nv_wr32(dev, 0x1900 + (i*4), 0);
/* Determine VM layout */ /* Determine VM layout */
dev_priv->vm_gart_base = roundup(NV50_VM_BLOCK, NV50_VM_BLOCK); dev_priv->vm_gart_base = roundup(NV50_VM_BLOCK, NV50_VM_BLOCK);
dev_priv->vm_gart_size = NV50_VM_BLOCK; dev_priv->vm_gart_size = NV50_VM_BLOCK;
...@@ -200,38 +239,19 @@ nv50_instmem_init(struct drm_device *dev) ...@@ -200,38 +239,19 @@ nv50_instmem_init(struct drm_device *dev)
for (i = 0; i < dev_priv->vm_vram_pt_nr; i++) { for (i = 0; i < dev_priv->vm_vram_pt_nr; i++) {
ret = nouveau_gpuobj_new(dev, NULL, NV50_VM_BLOCK / 0x10000 * 8, ret = nouveau_gpuobj_new(dev, NULL, NV50_VM_BLOCK / 0x10000 * 8,
0, NVOBJ_FLAG_ZERO_ALLOC, 0, NVOBJ_FLAG_ZERO_ALLOC,
&chan->vm_vram_pt[i]); &dev_priv->vm_vram_pt[i]);
if (ret) { if (ret) {
NV_ERROR(dev, "Error creating VRAM PGT: %d\n", ret); NV_ERROR(dev, "Error creating VRAM PGT: %d\n", ret);
dev_priv->vm_vram_pt_nr = i; dev_priv->vm_vram_pt_nr = i;
return ret; return ret;
} }
dev_priv->vm_vram_pt[i] = chan->vm_vram_pt[i];
nv_wo32(chan->vm_pd, 0x10 + (i*8),
chan->vm_vram_pt[i]->vinst | 0x61);
nv_wo32(chan->vm_pd, 0x14 + (i*8), 0);
} }
/* DMA object for FB BAR */
ret = nouveau_gpuobj_new(dev, chan, 6*4, 16, 0, &priv->fb_bar);
if (ret)
return ret;
nv_wo32(priv->fb_bar, 0x00, 0x7fc00000);
nv_wo32(priv->fb_bar, 0x04, 0x40000000 +
pci_resource_len(dev->pdev, 1) - 1);
nv_wo32(priv->fb_bar, 0x08, 0x40000000);
nv_wo32(priv->fb_bar, 0x0c, 0x00000000);
nv_wo32(priv->fb_bar, 0x10, 0x00000000);
nv_wo32(priv->fb_bar, 0x14, 0x00000000);
dev_priv->engine.instmem.flush(dev);
nv_wr32(dev, 0x001708, 0x80000000 | (priv->fb_bar->cinst >> 4));
for (i = 0; i < 8; i++)
nv_wr32(dev, 0x1900 + (i*4), 0);
return 0; return 0;
error:
nv50_instmem_takedown(dev);
return ret;
} }
void void
...@@ -249,23 +269,25 @@ nv50_instmem_takedown(struct drm_device *dev) ...@@ -249,23 +269,25 @@ nv50_instmem_takedown(struct drm_device *dev)
dev_priv->ramin_available = false; dev_priv->ramin_available = false;
/* Restore state from before init */ for (i = 0; i < dev_priv->vm_vram_pt_nr; i++)
nouveau_gpuobj_ref(NULL, &dev_priv->vm_vram_pt[i]);
dev_priv->vm_vram_pt_nr = 0;
for (i = 0x1700; i <= 0x1710; i += 4) for (i = 0x1700; i <= 0x1710; i += 4)
nv_wr32(dev, i, priv->save1700[(i - 0x1700) / 4]); nv_wr32(dev, i, priv->save1700[(i - 0x1700) / 4]);
nouveau_gpuobj_ref(NULL, &priv->fb_bar); nouveau_gpuobj_ref(NULL, &priv->bar3_dmaobj);
nouveau_gpuobj_ref(NULL, &priv->pramin_bar); nouveau_gpuobj_ref(NULL, &priv->bar1_dmaobj);
nouveau_gpuobj_ref(NULL, &priv->pramin_pt);
/* Destroy dummy channel */ nouveau_vm_ref(NULL, &dev_priv->bar1_vm, chan->vm_pd);
if (chan) { dev_priv->channels.ptr[127] = 0;
for (i = 0; i < dev_priv->vm_vram_pt_nr; i++) nv50_channel_del(&dev_priv->channels.ptr[0]);
nouveau_gpuobj_ref(NULL, &chan->vm_vram_pt[i]);
dev_priv->vm_vram_pt_nr = 0;
nv50_channel_del(&dev_priv->channels.ptr[0]); nouveau_gpuobj_ref(NULL, &dev_priv->bar3_vm->pgt[0].obj);
dev_priv->channels.ptr[127] = NULL; nouveau_vm_ref(NULL, &dev_priv->bar3_vm, NULL);
}
if (dev_priv->ramin_heap.free_stack.next)
drm_mm_takedown(&dev_priv->ramin_heap);
dev_priv->engine.instmem.priv = NULL; dev_priv->engine.instmem.priv = NULL;
kfree(priv); kfree(priv);
...@@ -293,9 +315,9 @@ nv50_instmem_resume(struct drm_device *dev) ...@@ -293,9 +315,9 @@ nv50_instmem_resume(struct drm_device *dev)
nv_wr32(dev, NV50_PUNK_UNK1710, 0); nv_wr32(dev, NV50_PUNK_UNK1710, 0);
nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->vinst >> 12) | nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->vinst >> 12) |
NV50_PUNK_BAR_CFG_BASE_VALID); NV50_PUNK_BAR_CFG_BASE_VALID);
nv_wr32(dev, NV50_PUNK_BAR1_CTXDMA, (priv->fb_bar->cinst >> 4) | nv_wr32(dev, NV50_PUNK_BAR1_CTXDMA, (priv->bar1_dmaobj->cinst >> 4) |
NV50_PUNK_BAR1_CTXDMA_VALID); NV50_PUNK_BAR1_CTXDMA_VALID);
nv_wr32(dev, NV50_PUNK_BAR3_CTXDMA, (priv->pramin_bar->cinst >> 4) | nv_wr32(dev, NV50_PUNK_BAR3_CTXDMA, (priv->bar3_dmaobj->cinst >> 4) |
NV50_PUNK_BAR3_CTXDMA_VALID); NV50_PUNK_BAR3_CTXDMA_VALID);
for (i = 0; i < 8; i++) for (i = 0; i < 8; i++)
...@@ -305,8 +327,7 @@ nv50_instmem_resume(struct drm_device *dev) ...@@ -305,8 +327,7 @@ nv50_instmem_resume(struct drm_device *dev)
} }
struct nv50_gpuobj_node { struct nv50_gpuobj_node {
struct nouveau_bo *vram; struct nouveau_vram *vram;
struct drm_mm_node *ramin;
u32 align; u32 align;
}; };
...@@ -323,23 +344,17 @@ nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align) ...@@ -323,23 +344,17 @@ nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
return -ENOMEM; return -ENOMEM;
node->align = align; node->align = align;
ret = nouveau_bo_new(dev, NULL, size, align, TTM_PL_FLAG_VRAM, size = (size + 4095) & ~4095;
0, 0x0000, true, false, &node->vram); align = max(align, (u32)4096);
if (ret) {
NV_ERROR(dev, "error getting PRAMIN backing pages: %d\n", ret);
WARN_ON(1);
return ret;
}
ret = nouveau_bo_pin(node->vram, TTM_PL_FLAG_VRAM); ret = nv50_vram_new(dev, size, align, 0, 0, &node->vram);
if (ret) { if (ret) {
NV_ERROR(dev, "error pinning PRAMIN backing VRAM: %d\n", ret); kfree(node);
nouveau_bo_ref(NULL, &node->vram);
return ret; return ret;
} }
gpuobj->vinst = node->vram->bo.mem.start << PAGE_SHIFT; gpuobj->vinst = node->vram->offset;
gpuobj->size = node->vram->bo.mem.num_pages << PAGE_SHIFT; gpuobj->size = size;
gpuobj->node = node; gpuobj->node = node;
return 0; return 0;
} }
...@@ -347,13 +362,13 @@ nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align) ...@@ -347,13 +362,13 @@ nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
void void
nv50_instmem_put(struct nouveau_gpuobj *gpuobj) nv50_instmem_put(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_device *dev = gpuobj->dev;
struct nv50_gpuobj_node *node; struct nv50_gpuobj_node *node;
node = gpuobj->node; node = gpuobj->node;
gpuobj->node = NULL; gpuobj->node = NULL;
nouveau_bo_unpin(node->vram); nv50_vram_del(dev, &node->vram);
nouveau_bo_ref(NULL, &node->vram);
kfree(node); kfree(node);
} }
...@@ -361,83 +376,28 @@ int ...@@ -361,83 +376,28 @@ int
nv50_instmem_map(struct nouveau_gpuobj *gpuobj) nv50_instmem_map(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private; struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
struct nv50_gpuobj_node *node = gpuobj->node; struct nv50_gpuobj_node *node = gpuobj->node;
struct drm_device *dev = gpuobj->dev; int ret;
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);
vram |= 1;
if (dev_priv->vram_sys_base) {
vram += dev_priv->vram_sys_base;
vram |= 0x30;
}
while (pte < pte_end) {
nv_wo32(priv->pramin_pt, (pte * 4) + 0, lower_32_bits(vram));
nv_wo32(priv->pramin_pt, (pte * 4) + 4, upper_32_bits(vram));
vram += 0x1000;
pte += 2;
}
dev_priv->engine.instmem.flush(dev);
nv50_vm_flush_engine(dev, 6); ret = nouveau_vm_get(dev_priv->bar3_vm, gpuobj->size, 12,
NV_MEM_ACCESS_RW, &node->vram->bar_vma);
if (ret)
return ret;
node->ramin = ramin; nouveau_vm_map(&node->vram->bar_vma, node->vram);
gpuobj->pinst = ramin->start; gpuobj->pinst = node->vram->bar_vma.offset;
return 0; return 0;
} }
void void
nv50_instmem_unmap(struct nouveau_gpuobj *gpuobj) nv50_instmem_unmap(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
struct nv50_gpuobj_node *node = gpuobj->node; struct nv50_gpuobj_node *node = gpuobj->node;
u32 pte, pte_end;
if (!node->ramin || !dev_priv->ramin_available) if (node->vram->bar_vma.node) {
return; nouveau_vm_unmap(&node->vram->bar_vma);
nouveau_vm_put(&node->vram->bar_vma);
pte = (node->ramin->start >> 12) << 1;
pte_end = ((node->ramin->size >> 12) << 1) + pte;
while (pte < pte_end) {
nv_wo32(priv->pramin_pt, (pte * 4) + 0, 0x00000000);
nv_wo32(priv->pramin_pt, (pte * 4) + 4, 0x00000000);
pte += 2;
} }
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 void
......
...@@ -151,7 +151,6 @@ nv50_vm_flush(struct nouveau_vm *vm) ...@@ -151,7 +151,6 @@ nv50_vm_flush(struct nouveau_vm *vm)
struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem; struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
pinstmem->flush(vm->dev); pinstmem->flush(vm->dev);
nv50_vm_flush_engine(vm->dev, 6); nv50_vm_flush_engine(vm->dev, 6);
} }
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment