Rev 6321 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
1120 | serge | 1 | /* |
2 | * Copyright 2008 Advanced Micro Devices, Inc. |
||
3 | * Copyright 2008 Red Hat Inc. |
||
4 | * Copyright 2009 Jerome Glisse. |
||
5 | * |
||
6 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
7 | * copy of this software and associated documentation files (the "Software"), |
||
8 | * to deal in the Software without restriction, including without limitation |
||
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
||
10 | * and/or sell copies of the Software, and to permit persons to whom the |
||
11 | * Software is furnished to do so, subject to the following conditions: |
||
12 | * |
||
13 | * The above copyright notice and this permission notice shall be included in |
||
14 | * all copies or substantial portions of the Software. |
||
15 | * |
||
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
||
19 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
||
20 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
||
21 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
||
22 | * OTHER DEALINGS IN THE SOFTWARE. |
||
23 | * |
||
24 | * Authors: Dave Airlie |
||
25 | * Alex Deucher |
||
26 | * Jerome Glisse |
||
27 | */ |
||
2997 | Serge | 28 | #include |
29 | #include |
||
1120 | serge | 30 | #include "radeon.h" |
31 | |||
32 | /* |
||
2997 | Serge | 33 | * GART |
34 | * The GART (Graphics Aperture Remapping Table) is an aperture |
||
35 | * in the GPU's address space. System pages can be mapped into |
||
36 | * the aperture and look like contiguous pages from the GPU's |
||
37 | * perspective. A page table maps the pages in the aperture |
||
38 | * to the actual backing pages in system memory. |
||
39 | * |
||
40 | * Radeon GPUs support both an internal GART, as described above, |
||
41 | * and AGP. AGP works similarly, but the GART table is configured |
||
42 | * and maintained by the northbridge rather than the driver. |
||
43 | * Radeon hw has a separate AGP aperture that is programmed to |
||
44 | * point to the AGP aperture provided by the northbridge and the |
||
45 | * requests are passed through to the northbridge aperture. |
||
46 | * Both AGP and internal GART can be used at the same time, however |
||
47 | * that is not currently supported by the driver. |
||
48 | * |
||
49 | * This file handles the common internal GART management. |
||
50 | */ |
||
51 | |||
52 | /* |
||
1120 | serge | 53 | * Common GART table functions. |
54 | */ |
||
2997 | Serge | 55 | /** |
56 | * radeon_gart_table_ram_alloc - allocate system ram for gart page table |
||
57 | * |
||
58 | * @rdev: radeon_device pointer |
||
59 | * |
||
60 | * Allocate system memory for GART page table |
||
61 | * (r1xx-r3xx, non-pcie r4xx, rs400). These asics require the |
||
62 | * gart table to be in system memory. |
||
63 | * Returns 0 for success, -ENOMEM for failure. |
||
64 | */ |
||
1120 | serge | 65 | int radeon_gart_table_ram_alloc(struct radeon_device *rdev) |
66 | { |
||
67 | void *ptr; |
||
68 | |||
1246 | serge | 69 | ptr = pci_alloc_consistent(rdev->pdev, rdev->gart.table_size, |
70 | &rdev->gart.table_addr); |
||
1120 | serge | 71 | if (ptr == NULL) { |
72 | return -ENOMEM; |
||
73 | } |
||
74 | #ifdef CONFIG_X86 |
||
75 | if (rdev->family == CHIP_RS400 || rdev->family == CHIP_RS480 || |
||
76 | rdev->family == CHIP_RS690 || rdev->family == CHIP_RS740) { |
||
77 | set_memory_uc((unsigned long)ptr, |
||
78 | rdev->gart.table_size >> PAGE_SHIFT); |
||
79 | } |
||
80 | #endif |
||
2997 | Serge | 81 | rdev->gart.ptr = ptr; |
82 | memset((void *)rdev->gart.ptr, 0, rdev->gart.table_size); |
||
1120 | serge | 83 | return 0; |
84 | } |
||
85 | |||
2997 | Serge | 86 | /** |
87 | * radeon_gart_table_ram_free - free system ram for gart page table |
||
88 | * |
||
89 | * @rdev: radeon_device pointer |
||
90 | * |
||
91 | * Free system memory for GART page table |
||
92 | * (r1xx-r3xx, non-pcie r4xx, rs400). These asics require the |
||
93 | * gart table to be in system memory. |
||
94 | */ |
||
1120 | serge | 95 | void radeon_gart_table_ram_free(struct radeon_device *rdev) |
96 | { |
||
2997 | Serge | 97 | if (rdev->gart.ptr == NULL) { |
1120 | serge | 98 | return; |
99 | } |
||
100 | #ifdef CONFIG_X86 |
||
101 | if (rdev->family == CHIP_RS400 || rdev->family == CHIP_RS480 || |
||
102 | rdev->family == CHIP_RS690 || rdev->family == CHIP_RS740) { |
||
2997 | Serge | 103 | set_memory_wb((unsigned long)rdev->gart.ptr, |
1120 | serge | 104 | rdev->gart.table_size >> PAGE_SHIFT); |
105 | } |
||
106 | #endif |
||
2997 | Serge | 107 | rdev->gart.ptr = NULL; |
1120 | serge | 108 | rdev->gart.table_addr = 0; |
109 | } |
||
110 | |||
2997 | Serge | 111 | /** |
112 | * radeon_gart_table_vram_alloc - allocate vram for gart page table |
||
113 | * |
||
114 | * @rdev: radeon_device pointer |
||
115 | * |
||
116 | * Allocate video memory for GART page table |
||
117 | * (pcie r4xx, r5xx+). These asics require the |
||
118 | * gart table to be in video memory. |
||
119 | * Returns 0 for success, error for failure. |
||
120 | */ |
||
1120 | serge | 121 | int radeon_gart_table_vram_alloc(struct radeon_device *rdev) |
122 | { |
||
6104 | serge | 123 | int r; |
1120 | serge | 124 | |
2997 | Serge | 125 | if (rdev->gart.robj == NULL) { |
1963 | serge | 126 | r = radeon_bo_create(rdev, rdev->gart.table_size, |
127 | PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM, |
||
5271 | serge | 128 | 0, NULL, NULL, &rdev->gart.robj); |
6104 | serge | 129 | if (r) { |
130 | return r; |
||
131 | } |
||
132 | } |
||
1179 | serge | 133 | return 0; |
134 | } |
||
135 | |||
2997 | Serge | 136 | /** |
137 | * radeon_gart_table_vram_pin - pin gart page table in vram |
||
138 | * |
||
139 | * @rdev: radeon_device pointer |
||
140 | * |
||
141 | * Pin the GART page table in vram so it will not be moved |
||
142 | * by the memory manager (pcie r4xx, r5xx+). These asics require the |
||
143 | * gart table to be in video memory. |
||
144 | * Returns 0 for success, error for failure. |
||
145 | */ |
||
1179 | serge | 146 | int radeon_gart_table_vram_pin(struct radeon_device *rdev) |
147 | { |
||
148 | uint64_t gpu_addr; |
||
149 | int r; |
||
150 | |||
2997 | Serge | 151 | r = radeon_bo_reserve(rdev->gart.robj, false); |
1404 | serge | 152 | if (unlikely(r != 0)) |
153 | return r; |
||
2997 | Serge | 154 | r = radeon_bo_pin(rdev->gart.robj, |
6104 | serge | 155 | RADEON_GEM_DOMAIN_VRAM, &gpu_addr); |
156 | if (r) { |
||
2997 | Serge | 157 | radeon_bo_unreserve(rdev->gart.robj); |
6104 | serge | 158 | return r; |
159 | } |
||
2997 | Serge | 160 | r = radeon_bo_kmap(rdev->gart.robj, &rdev->gart.ptr); |
1404 | serge | 161 | if (r) |
2997 | Serge | 162 | radeon_bo_unpin(rdev->gart.robj); |
163 | radeon_bo_unreserve(rdev->gart.robj); |
||
1404 | serge | 164 | rdev->gart.table_addr = gpu_addr; |
6104 | serge | 165 | |
166 | if (!r) { |
||
167 | int i; |
||
168 | |||
169 | /* We might have dropped some GART table updates while it wasn't |
||
170 | * mapped, restore all entries |
||
171 | */ |
||
172 | for (i = 0; i < rdev->gart.num_gpu_pages; i++) |
||
173 | radeon_gart_set_page(rdev, i, rdev->gart.pages_entry[i]); |
||
174 | mb(); |
||
175 | radeon_gart_tlb_flush(rdev); |
||
176 | } |
||
177 | |||
178 | return r; |
||
1120 | serge | 179 | } |
180 | |||
2997 | Serge | 181 | /** |
182 | * radeon_gart_table_vram_unpin - unpin gart page table in vram |
||
183 | * |
||
184 | * @rdev: radeon_device pointer |
||
185 | * |
||
186 | * Unpin the GART page table in vram (pcie r4xx, r5xx+). |
||
187 | * These asics require the gart table to be in video memory. |
||
188 | */ |
||
189 | void radeon_gart_table_vram_unpin(struct radeon_device *rdev) |
||
1120 | serge | 190 | { |
1404 | serge | 191 | int r; |
192 | |||
2997 | Serge | 193 | if (rdev->gart.robj == NULL) { |
1120 | serge | 194 | return; |
195 | } |
||
2997 | Serge | 196 | r = radeon_bo_reserve(rdev->gart.robj, false); |
1404 | serge | 197 | if (likely(r == 0)) { |
2997 | Serge | 198 | radeon_bo_kunmap(rdev->gart.robj); |
199 | radeon_bo_unpin(rdev->gart.robj); |
||
200 | radeon_bo_unreserve(rdev->gart.robj); |
||
201 | rdev->gart.ptr = NULL; |
||
1404 | serge | 202 | } |
1120 | serge | 203 | } |
204 | |||
2997 | Serge | 205 | /** |
206 | * radeon_gart_table_vram_free - free gart page table vram |
||
207 | * |
||
208 | * @rdev: radeon_device pointer |
||
209 | * |
||
210 | * Free the video memory used for the GART page table |
||
211 | * (pcie r4xx, r5xx+). These asics require the gart table to |
||
212 | * be in video memory. |
||
213 | */ |
||
214 | void radeon_gart_table_vram_free(struct radeon_device *rdev) |
||
215 | { |
||
216 | if (rdev->gart.robj == NULL) { |
||
217 | return; |
||
218 | } |
||
219 | radeon_bo_unref(&rdev->gart.robj); |
||
220 | } |
||
1120 | serge | 221 | |
222 | /* |
||
223 | * Common gart functions. |
||
224 | */ |
||
2997 | Serge | 225 | /** |
226 | * radeon_gart_unbind - unbind pages from the gart page table |
||
227 | * |
||
228 | * @rdev: radeon_device pointer |
||
229 | * @offset: offset into the GPU's gart aperture |
||
230 | * @pages: number of pages to unbind |
||
231 | * |
||
232 | * Unbinds the requested pages from the gart page table and |
||
233 | * replaces them with the dummy page (all asics). |
||
234 | */ |
||
1120 | serge | 235 | void radeon_gart_unbind(struct radeon_device *rdev, unsigned offset, |
236 | int pages) |
||
237 | { |
||
238 | unsigned t; |
||
239 | unsigned p; |
||
240 | int i, j; |
||
241 | |||
242 | if (!rdev->gart.ready) { |
||
2997 | Serge | 243 | WARN(1, "trying to unbind memory from uninitialized GART !\n"); |
1120 | serge | 244 | return; |
245 | } |
||
1268 | serge | 246 | t = offset / RADEON_GPU_PAGE_SIZE; |
247 | p = t / (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); |
||
1120 | serge | 248 | for (i = 0; i < pages; i++, p++) { |
249 | if (rdev->gart.pages[p]) { |
||
250 | rdev->gart.pages[p] = NULL; |
||
1268 | serge | 251 | for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) { |
6104 | serge | 252 | rdev->gart.pages_entry[t] = rdev->dummy_page.entry; |
2997 | Serge | 253 | if (rdev->gart.ptr) { |
6104 | serge | 254 | radeon_gart_set_page(rdev, t, |
255 | rdev->dummy_page.entry); |
||
2997 | Serge | 256 | } |
1120 | serge | 257 | } |
258 | } |
||
259 | } |
||
6104 | serge | 260 | if (rdev->gart.ptr) { |
261 | mb(); |
||
262 | radeon_gart_tlb_flush(rdev); |
||
263 | } |
||
1120 | serge | 264 | } |
265 | |||
2997 | Serge | 266 | /** |
267 | * radeon_gart_bind - bind pages into the gart page table |
||
268 | * |
||
269 | * @rdev: radeon_device pointer |
||
270 | * @offset: offset into the GPU's gart aperture |
||
271 | * @pages: number of pages to bind |
||
272 | * @pagelist: pages to bind |
||
273 | * @dma_addr: DMA addresses of pages |
||
5078 | serge | 274 | * @flags: RADEON_GART_PAGE_* flags |
2997 | Serge | 275 | * |
276 | * Binds the requested pages to the gart page table |
||
277 | * (all asics). |
||
278 | * Returns 0 for success, -EINVAL for failure. |
||
279 | */ |
||
1120 | serge | 280 | int radeon_gart_bind(struct radeon_device *rdev, unsigned offset, |
5078 | serge | 281 | int pages, struct page **pagelist, dma_addr_t *dma_addr, |
282 | uint32_t flags) |
||
1120 | serge | 283 | { |
6104 | serge | 284 | unsigned t; |
285 | unsigned p; |
||
286 | uint64_t page_base, page_entry; |
||
287 | int i, j; |
||
1120 | serge | 288 | |
6104 | serge | 289 | if (!rdev->gart.ready) { |
2997 | Serge | 290 | WARN(1, "trying to bind memory to uninitialized GART !\n"); |
6104 | serge | 291 | return -EINVAL; |
292 | } |
||
1268 | serge | 293 | t = offset / RADEON_GPU_PAGE_SIZE; |
294 | p = t / (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); |
||
1120 | serge | 295 | |
6104 | serge | 296 | for (i = 0; i < pages; i++, p++) { |
297 | rdev->gart.pages[p] = pagelist[i]; |
||
298 | page_base = dma_addr[i]; |
||
1268 | serge | 299 | for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) { |
6104 | serge | 300 | page_entry = radeon_gart_get_page_entry(page_base, flags); |
301 | rdev->gart.pages_entry[t] = page_entry; |
||
302 | if (rdev->gart.ptr) { |
||
303 | radeon_gart_set_page(rdev, t, page_entry); |
||
304 | } |
||
1268 | serge | 305 | page_base += RADEON_GPU_PAGE_SIZE; |
6104 | serge | 306 | } |
2997 | Serge | 307 | } |
6104 | serge | 308 | if (rdev->gart.ptr) { |
309 | mb(); |
||
310 | radeon_gart_tlb_flush(rdev); |
||
311 | } |
||
312 | return 0; |
||
1120 | serge | 313 | } |
314 | |||
2997 | Serge | 315 | /** |
316 | * radeon_gart_init - init the driver info for managing the gart |
||
317 | * |
||
318 | * @rdev: radeon_device pointer |
||
319 | * |
||
320 | * Allocate the dummy page and init the gart driver info (all asics). |
||
321 | * Returns 0 for success, error for failure. |
||
322 | */ |
||
1120 | serge | 323 | int radeon_gart_init(struct radeon_device *rdev) |
324 | { |
||
1430 | serge | 325 | int r, i; |
326 | |||
6104 | serge | 327 | if (rdev->gart.pages) { |
328 | return 0; |
||
329 | } |
||
1268 | serge | 330 | /* We need PAGE_SIZE >= RADEON_GPU_PAGE_SIZE */ |
331 | if (PAGE_SIZE < RADEON_GPU_PAGE_SIZE) { |
||
6104 | serge | 332 | DRM_ERROR("Page size is smaller than GPU page size!\n"); |
333 | return -EINVAL; |
||
334 | } |
||
1430 | serge | 335 | r = radeon_dummy_page_init(rdev); |
336 | if (r) |
||
337 | return r; |
||
6104 | serge | 338 | /* Compute table size */ |
339 | rdev->gart.num_cpu_pages = rdev->mc.gtt_size / PAGE_SIZE; |
||
1268 | serge | 340 | rdev->gart.num_gpu_pages = rdev->mc.gtt_size / RADEON_GPU_PAGE_SIZE; |
6104 | serge | 341 | DRM_INFO("GART: num cpu pages %u, num gpu pages %u\n", |
342 | rdev->gart.num_cpu_pages, rdev->gart.num_gpu_pages); |
||
343 | /* Allocate pages table */ |
||
2997 | Serge | 344 | rdev->gart.pages = vzalloc(sizeof(void *) * rdev->gart.num_cpu_pages); |
6104 | serge | 345 | if (rdev->gart.pages == NULL) { |
1404 | serge | 346 | radeon_gart_fini(rdev); |
6104 | serge | 347 | return -ENOMEM; |
348 | } |
||
349 | rdev->gart.pages_entry = KernelAlloc(sizeof(uint64_t) * |
||
350 | rdev->gart.num_gpu_pages); |
||
351 | if (rdev->gart.pages_entry == NULL) { |
||
1404 | serge | 352 | radeon_gart_fini(rdev); |
6104 | serge | 353 | return -ENOMEM; |
354 | } |
||
1430 | serge | 355 | /* set GART entry to point to the dummy page by default */ |
6104 | serge | 356 | for (i = 0; i < rdev->gart.num_gpu_pages; i++) |
357 | rdev->gart.pages_entry[i] = rdev->dummy_page.entry; |
||
358 | return 0; |
||
1120 | serge | 359 | } |
360 | |||
2997 | Serge | 361 | /** |
362 | * radeon_gart_fini - tear down the driver info for managing the gart |
||
363 | * |
||
364 | * @rdev: radeon_device pointer |
||
365 | * |
||
366 | * Tear down the gart driver info and free the dummy page (all asics). |
||
367 | */ |
||
1120 | serge | 368 | void radeon_gart_fini(struct radeon_device *rdev) |
369 | { |
||
6104 | serge | 370 | if (rdev->gart.ready) { |
1120 | serge | 371 | /* unbind pages */ |
372 | radeon_gart_unbind(rdev, 0, rdev->gart.num_cpu_pages); |
||
373 | } |
||
374 | rdev->gart.ready = false; |
||
2997 | Serge | 375 | vfree(rdev->gart.pages); |
6104 | serge | 376 | vfree(rdev->gart.pages_entry); |
1120 | serge | 377 | rdev->gart.pages = NULL; |
6104 | serge | 378 | rdev->gart.pages_entry = NULL; |
2997 | Serge | 379 | |
5078 | serge | 380 | radeon_dummy_page_fini(rdev); |
2997 | Serge | 381 | }>>>>>>> |