1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
|
#include "drmP.h"
#include "nouveau_drv.h"
#include <linux/pagemap.h>
#include <linux/slab.h>
#define NV_CTXDMA_PAGE_SHIFT 12
#define NV_CTXDMA_PAGE_SIZE (1 << NV_CTXDMA_PAGE_SHIFT)
#define NV_CTXDMA_PAGE_MASK (NV_CTXDMA_PAGE_SIZE - 1)
struct nouveau_sgdma_be {
struct ttm_backend backend;
struct drm_device *dev;
dma_addr_t *pages;
unsigned nr_pages;
unsigned pte_start;
bool bound;
};
static int
nouveau_sgdma_populate(struct ttm_backend *be, unsigned long num_pages,
struct page **pages, struct page *dummy_read_page)
{
struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
struct drm_device *dev = nvbe->dev;
NV_DEBUG(nvbe->dev, "num_pages = %ld\n", num_pages);
if (nvbe->pages)
return -EINVAL;
nvbe->pages = kmalloc(sizeof(dma_addr_t) * num_pages, GFP_KERNEL);
if (!nvbe->pages)
return -ENOMEM;
nvbe->nr_pages = 0;
while (num_pages--) {
nvbe->pages[nvbe->nr_pages] =
pci_map_page(dev->pdev, pages[nvbe->nr_pages], 0,
PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
if (pci_dma_mapping_error(dev->pdev,
nvbe->pages[nvbe->nr_pages])) {
be->func->clear(be);
return -EFAULT;
}
nvbe->nr_pages++;
}
return 0;
}
static void
nouveau_sgdma_clear(struct ttm_backend *be)
{
struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
struct drm_device *dev;
if (nvbe && nvbe->pages) {
dev = nvbe->dev;
NV_DEBUG(dev, "\n");
if (nvbe->bound)
be->func->unbind(be);
while (nvbe->nr_pages--) {
pci_unmap_page(dev->pdev, nvbe->pages[nvbe->nr_pages],
PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
}
kfree(nvbe->pages);
nvbe->pages = NULL;
nvbe->nr_pages = 0;
}
}
static inline unsigned
nouveau_sgdma_pte(struct drm_device *dev, uint64_t offset)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
unsigned pte = (offset >> NV_CTXDMA_PAGE_SHIFT);
if (dev_priv->card_type < NV_50)
return pte + 2;
return pte << 1;
}
static int
nouveau_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
{
struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
struct drm_device *dev = nvbe->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *gpuobj = dev_priv->gart_info.sg_ctxdma;
unsigned i, j, pte;
NV_DEBUG(dev, "pg=0x%lx\n", mem->start);
pte = nouveau_sgdma_pte(nvbe->dev, mem->start << PAGE_SHIFT);
nvbe->pte_start = pte;
for (i = 0; i < nvbe->nr_pages; i++) {
dma_addr_t dma_offset = nvbe->pages[i];
uint32_t offset_l = lower_32_bits(dma_offset);
uint32_t offset_h = upper_32_bits(dma_offset);
for (j = 0; j < PAGE_SIZE / NV_CTXDMA_PAGE_SIZE; j++) {
if (dev_priv->card_type < NV_50) {
nv_wo32(gpuobj, (pte * 4) + 0, offset_l | 3);
pte += 1;
} else {
nv_wo32(gpuobj, (pte * 4) + 0, offset_l | 0x21);
nv_wo32(gpuobj, (pte * 4) + 4, offset_h & 0xff);
pte += 2;
}
dma_offset += NV_CTXDMA_PAGE_SIZE;
}
}
dev_priv->engine.instmem.flush(nvbe->dev);
if (dev_priv->card_type == NV_50) {
dev_priv->engine.fifo.tlb_flush(dev);
dev_priv->engine.graph.tlb_flush(dev);
}
nvbe->bound = true;
return 0;
}
static int
nouveau_sgdma_unbind(struct ttm_backend *be)
{
struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
struct drm_device *dev = nvbe->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *gpuobj = dev_priv->gart_info.sg_ctxdma;
unsigned i, j, pte;
NV_DEBUG(dev, "\n");
if (!nvbe->bound)
return 0;
pte = nvbe->pte_start;
for (i = 0; i < nvbe->nr_pages; i++) {
for (j = 0; j < PAGE_SIZE / NV_CTXDMA_PAGE_SIZE; j++) {
if (dev_priv->card_type < NV_50) {
nv_wo32(gpuobj, (pte * 4) + 0, 0x00000000);
pte += 1;
} else {
nv_wo32(gpuobj, (pte * 4) + 0, 0x00000000);
nv_wo32(gpuobj, (pte * 4) + 4, 0x00000000);
pte += 2;
}
}
}
dev_priv->engine.instmem.flush(nvbe->dev);
if (dev_priv->card_type == NV_50) {
dev_priv->engine.fifo.tlb_flush(dev);
dev_priv->engine.graph.tlb_flush(dev);
}
nvbe->bound = false;
return 0;
}
static void
nouveau_sgdma_destroy(struct ttm_backend *be)
{
struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
if (be) {
NV_DEBUG(nvbe->dev, "\n");
if (nvbe) {
if (nvbe->pages)
be->func->clear(be);
kfree(nvbe);
}
}
}
static struct ttm_backend_func nouveau_sgdma_backend = {
.populate = nouveau_sgdma_populate,
.clear = nouveau_sgdma_clear,
.bind = nouveau_sgdma_bind,
.unbind = nouveau_sgdma_unbind,
.destroy = nouveau_sgdma_destroy
};
struct ttm_backend *
nouveau_sgdma_init_ttm(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_sgdma_be *nvbe;
if (!dev_priv->gart_info.sg_ctxdma)
return NULL;
nvbe = kzalloc(sizeof(*nvbe), GFP_KERNEL);
if (!nvbe)
return NULL;
nvbe->dev = dev;
nvbe->backend.func = &nouveau_sgdma_backend;
return &nvbe->backend;
}
int
nouveau_sgdma_init(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *gpuobj = NULL;
uint32_t aper_size, obj_size;
int i, ret;
if (dev_priv->card_type < NV_50) {
if(dev_priv->ramin_rsvd_vram < 2 * 1024 * 1024)
aper_size = 64 * 1024 * 1024;
else
aper_size = 512 * 1024 * 1024;
obj_size = (aper_size >> NV_CTXDMA_PAGE_SHIFT) * 4;
obj_size += 8; /* ctxdma header */
} else {
/* 1 entire VM page table */
aper_size = (512 * 1024 * 1024);
obj_size = (aper_size >> NV_CTXDMA_PAGE_SHIFT) * 8;
}
ret = nouveau_gpuobj_new(dev, NULL, obj_size, 16,
NVOBJ_FLAG_ZERO_ALLOC |
NVOBJ_FLAG_ZERO_FREE, &gpuobj);
if (ret) {
NV_ERROR(dev, "Error creating sgdma object: %d\n", ret);
return ret;
}
if (dev_priv->card_type < NV_50) {
/* special case, allocated from global instmem heap so
* cinst is invalid, we use it on all channels though so
* cinst needs to be valid, set it the same as pinst
*/
gpuobj->cinst = gpuobj->pinst;
/* Maybe use NV_DMA_TARGET_AGP for PCIE? NVIDIA do this, and
* confirmed to work on c51. Perhaps means NV_DMA_TARGET_PCIE
* on those cards? */
nv_wo32(gpuobj, 0, NV_CLASS_DMA_IN_MEMORY |
(1 << 12) /* PT present */ |
(0 << 13) /* PT *not* linear */ |
(NV_DMA_ACCESS_RW << 14) |
(NV_DMA_TARGET_PCI << 16));
nv_wo32(gpuobj, 4, aper_size - 1);
for (i = 2; i < 2 + (aper_size >> 12); i++)
nv_wo32(gpuobj, i * 4, 0x00000000);
} else {
for (i = 0; i < obj_size; i += 8) {
nv_wo32(gpuobj, i + 0, 0x00000000);
nv_wo32(gpuobj, i + 4, 0x00000000);
}
}
dev_priv->engine.instmem.flush(dev);
dev_priv->gart_info.type = NOUVEAU_GART_SGDMA;
dev_priv->gart_info.aper_base = 0;
dev_priv->gart_info.aper_size = aper_size;
dev_priv->gart_info.sg_ctxdma = gpuobj;
return 0;
}
void
nouveau_sgdma_takedown(struct drm_device *dev)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
nouveau_gpuobj_ref(NULL, &dev_priv->gart_info.sg_ctxdma);
}
int
nouveau_sgdma_get_page(struct drm_device *dev, uint32_t offset, uint32_t *page)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_gpuobj *gpuobj = dev_priv->gart_info.sg_ctxdma;
int pte;
pte = (offset >> NV_CTXDMA_PAGE_SHIFT) << 2;
if (dev_priv->card_type < NV_50) {
*page = nv_ro32(gpuobj, (pte + 8)) & ~NV_CTXDMA_PAGE_MASK;
return 0;
}
NV_ERROR(dev, "Unimplemented on NV50\n");
return -EINVAL;
}
|