Rev 3764 | Rev 5271 | 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 | |||
1631 | serge | 32 | |
33 | static inline void * |
||
34 | pci_alloc_consistent(struct pci_dev *hwdev, size_t size, |
||
35 | addr_t *dma_handle) |
||
36 | { |
||
37 | |||
38 | size = (size + 0x7FFF) & ~0x7FFF; |
||
39 | |||
40 | *dma_handle = AllocPages(size >> 12); |
||
41 | return (void*)MapIoMem(*dma_handle, size, PG_SW+PG_NOCACHE); |
||
42 | } |
||
43 | |||
1120 | serge | 44 | /* |
2997 | Serge | 45 | * GART |
46 | * The GART (Graphics Aperture Remapping Table) is an aperture |
||
47 | * in the GPU's address space. System pages can be mapped into |
||
48 | * the aperture and look like contiguous pages from the GPU's |
||
49 | * perspective. A page table maps the pages in the aperture |
||
50 | * to the actual backing pages in system memory. |
||
51 | * |
||
52 | * Radeon GPUs support both an internal GART, as described above, |
||
53 | * and AGP. AGP works similarly, but the GART table is configured |
||
54 | * and maintained by the northbridge rather than the driver. |
||
55 | * Radeon hw has a separate AGP aperture that is programmed to |
||
56 | * point to the AGP aperture provided by the northbridge and the |
||
57 | * requests are passed through to the northbridge aperture. |
||
58 | * Both AGP and internal GART can be used at the same time, however |
||
59 | * that is not currently supported by the driver. |
||
60 | * |
||
61 | * This file handles the common internal GART management. |
||
62 | */ |
||
63 | |||
64 | /* |
||
1120 | serge | 65 | * Common GART table functions. |
66 | */ |
||
2997 | Serge | 67 | /** |
68 | * radeon_gart_table_ram_alloc - allocate system ram for gart page table |
||
69 | * |
||
70 | * @rdev: radeon_device pointer |
||
71 | * |
||
72 | * Allocate system memory for GART page table |
||
73 | * (r1xx-r3xx, non-pcie r4xx, rs400). These asics require the |
||
74 | * gart table to be in system memory. |
||
75 | * Returns 0 for success, -ENOMEM for failure. |
||
76 | */ |
||
1120 | serge | 77 | int radeon_gart_table_ram_alloc(struct radeon_device *rdev) |
78 | { |
||
79 | void *ptr; |
||
80 | |||
1246 | serge | 81 | ptr = pci_alloc_consistent(rdev->pdev, rdev->gart.table_size, |
82 | &rdev->gart.table_addr); |
||
1120 | serge | 83 | if (ptr == NULL) { |
84 | return -ENOMEM; |
||
85 | } |
||
86 | #ifdef CONFIG_X86 |
||
87 | if (rdev->family == CHIP_RS400 || rdev->family == CHIP_RS480 || |
||
88 | rdev->family == CHIP_RS690 || rdev->family == CHIP_RS740) { |
||
89 | set_memory_uc((unsigned long)ptr, |
||
90 | rdev->gart.table_size >> PAGE_SHIFT); |
||
91 | } |
||
92 | #endif |
||
2997 | Serge | 93 | rdev->gart.ptr = ptr; |
94 | memset((void *)rdev->gart.ptr, 0, rdev->gart.table_size); |
||
1120 | serge | 95 | return 0; |
96 | } |
||
97 | |||
2997 | Serge | 98 | /** |
99 | * radeon_gart_table_ram_free - free system ram for gart page table |
||
100 | * |
||
101 | * @rdev: radeon_device pointer |
||
102 | * |
||
103 | * Free system memory for GART page table |
||
104 | * (r1xx-r3xx, non-pcie r4xx, rs400). These asics require the |
||
105 | * gart table to be in system memory. |
||
106 | */ |
||
1120 | serge | 107 | void radeon_gart_table_ram_free(struct radeon_device *rdev) |
108 | { |
||
2997 | Serge | 109 | if (rdev->gart.ptr == NULL) { |
1120 | serge | 110 | return; |
111 | } |
||
112 | #ifdef CONFIG_X86 |
||
113 | if (rdev->family == CHIP_RS400 || rdev->family == CHIP_RS480 || |
||
114 | rdev->family == CHIP_RS690 || rdev->family == CHIP_RS740) { |
||
2997 | Serge | 115 | set_memory_wb((unsigned long)rdev->gart.ptr, |
1120 | serge | 116 | rdev->gart.table_size >> PAGE_SHIFT); |
117 | } |
||
118 | #endif |
||
2997 | Serge | 119 | rdev->gart.ptr = NULL; |
1120 | serge | 120 | rdev->gart.table_addr = 0; |
121 | } |
||
122 | |||
2997 | Serge | 123 | /** |
124 | * radeon_gart_table_vram_alloc - allocate vram for gart page table |
||
125 | * |
||
126 | * @rdev: radeon_device pointer |
||
127 | * |
||
128 | * Allocate video memory for GART page table |
||
129 | * (pcie r4xx, r5xx+). These asics require the |
||
130 | * gart table to be in video memory. |
||
131 | * Returns 0 for success, error for failure. |
||
132 | */ |
||
1120 | serge | 133 | int radeon_gart_table_vram_alloc(struct radeon_device *rdev) |
134 | { |
||
135 | int r; |
||
136 | |||
2997 | Serge | 137 | if (rdev->gart.robj == NULL) { |
1963 | serge | 138 | r = radeon_bo_create(rdev, rdev->gart.table_size, |
139 | PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM, |
||
5078 | serge | 140 | 0, NULL, &rdev->gart.robj); |
1120 | serge | 141 | if (r) { |
142 | return r; |
||
143 | } |
||
144 | } |
||
1179 | serge | 145 | return 0; |
146 | } |
||
147 | |||
2997 | Serge | 148 | /** |
149 | * radeon_gart_table_vram_pin - pin gart page table in vram |
||
150 | * |
||
151 | * @rdev: radeon_device pointer |
||
152 | * |
||
153 | * Pin the GART page table in vram so it will not be moved |
||
154 | * by the memory manager (pcie r4xx, r5xx+). These asics require the |
||
155 | * gart table to be in video memory. |
||
156 | * Returns 0 for success, error for failure. |
||
157 | */ |
||
1179 | serge | 158 | int radeon_gart_table_vram_pin(struct radeon_device *rdev) |
159 | { |
||
160 | uint64_t gpu_addr; |
||
161 | int r; |
||
162 | |||
2997 | Serge | 163 | r = radeon_bo_reserve(rdev->gart.robj, false); |
1404 | serge | 164 | if (unlikely(r != 0)) |
165 | return r; |
||
2997 | Serge | 166 | r = radeon_bo_pin(rdev->gart.robj, |
1120 | serge | 167 | RADEON_GEM_DOMAIN_VRAM, &gpu_addr); |
168 | if (r) { |
||
2997 | Serge | 169 | radeon_bo_unreserve(rdev->gart.robj); |
1120 | serge | 170 | return r; |
171 | } |
||
2997 | Serge | 172 | r = radeon_bo_kmap(rdev->gart.robj, &rdev->gart.ptr); |
1404 | serge | 173 | if (r) |
2997 | Serge | 174 | radeon_bo_unpin(rdev->gart.robj); |
175 | radeon_bo_unreserve(rdev->gart.robj); |
||
1404 | serge | 176 | rdev->gart.table_addr = gpu_addr; |
177 | return r; |
||
1120 | serge | 178 | } |
179 | |||
2997 | Serge | 180 | /** |
181 | * radeon_gart_table_vram_unpin - unpin gart page table in vram |
||
182 | * |
||
183 | * @rdev: radeon_device pointer |
||
184 | * |
||
185 | * Unpin the GART page table in vram (pcie r4xx, r5xx+). |
||
186 | * These asics require the gart table to be in video memory. |
||
187 | */ |
||
188 | void radeon_gart_table_vram_unpin(struct radeon_device *rdev) |
||
1120 | serge | 189 | { |
1404 | serge | 190 | int r; |
191 | |||
2997 | Serge | 192 | if (rdev->gart.robj == NULL) { |
1120 | serge | 193 | return; |
194 | } |
||
2997 | Serge | 195 | r = radeon_bo_reserve(rdev->gart.robj, false); |
1404 | serge | 196 | if (likely(r == 0)) { |
2997 | Serge | 197 | radeon_bo_kunmap(rdev->gart.robj); |
198 | radeon_bo_unpin(rdev->gart.robj); |
||
199 | radeon_bo_unreserve(rdev->gart.robj); |
||
200 | rdev->gart.ptr = NULL; |
||
1404 | serge | 201 | } |
1120 | serge | 202 | } |
203 | |||
2997 | Serge | 204 | /** |
205 | * radeon_gart_table_vram_free - free gart page table vram |
||
206 | * |
||
207 | * @rdev: radeon_device pointer |
||
208 | * |
||
209 | * Free the video memory used for the GART page table |
||
210 | * (pcie r4xx, r5xx+). These asics require the gart table to |
||
211 | * be in video memory. |
||
212 | */ |
||
213 | void radeon_gart_table_vram_free(struct radeon_device *rdev) |
||
214 | { |
||
215 | if (rdev->gart.robj == NULL) { |
||
216 | return; |
||
217 | } |
||
218 | radeon_bo_unref(&rdev->gart.robj); |
||
219 | } |
||
1120 | serge | 220 | |
221 | /* |
||
222 | * Common gart functions. |
||
223 | */ |
||
2997 | Serge | 224 | /** |
225 | * radeon_gart_unbind - unbind pages from the gart page table |
||
226 | * |
||
227 | * @rdev: radeon_device pointer |
||
228 | * @offset: offset into the GPU's gart aperture |
||
229 | * @pages: number of pages to unbind |
||
230 | * |
||
231 | * Unbinds the requested pages from the gart page table and |
||
232 | * replaces them with the dummy page (all asics). |
||
233 | */ |
||
1120 | serge | 234 | void radeon_gart_unbind(struct radeon_device *rdev, unsigned offset, |
235 | int pages) |
||
236 | { |
||
237 | unsigned t; |
||
238 | unsigned p; |
||
239 | int i, j; |
||
1430 | serge | 240 | u64 page_base; |
1120 | serge | 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; |
||
1430 | serge | 251 | rdev->gart.pages_addr[p] = rdev->dummy_page.addr; |
252 | page_base = rdev->gart.pages_addr[p]; |
||
1268 | serge | 253 | for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) { |
2997 | Serge | 254 | if (rdev->gart.ptr) { |
5078 | serge | 255 | radeon_gart_set_page(rdev, t, page_base, |
256 | RADEON_GART_PAGE_DUMMY); |
||
2997 | Serge | 257 | } |
1430 | serge | 258 | page_base += RADEON_GPU_PAGE_SIZE; |
1120 | serge | 259 | } |
260 | } |
||
261 | } |
||
262 | mb(); |
||
263 | radeon_gart_tlb_flush(rdev); |
||
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 | { |
284 | unsigned t; |
||
285 | unsigned p; |
||
286 | uint64_t page_base; |
||
287 | int i, j; |
||
288 | |||
289 | if (!rdev->gart.ready) { |
||
2997 | Serge | 290 | WARN(1, "trying to bind memory to uninitialized GART !\n"); |
1120 | 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 | |
296 | for (i = 0; i < pages; i++, p++) { |
||
5078 | serge | 297 | rdev->gart.pages_addr[p] = dma_addr[i]; |
1120 | serge | 298 | rdev->gart.pages[p] = pagelist[i]; |
2997 | Serge | 299 | if (rdev->gart.ptr) { |
1268 | serge | 300 | page_base = rdev->gart.pages_addr[p]; |
301 | for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) { |
||
5078 | serge | 302 | radeon_gart_set_page(rdev, t, page_base, flags); |
1268 | serge | 303 | page_base += RADEON_GPU_PAGE_SIZE; |
1120 | serge | 304 | } |
305 | } |
||
2997 | Serge | 306 | } |
1120 | serge | 307 | mb(); |
308 | radeon_gart_tlb_flush(rdev); |
||
309 | return 0; |
||
310 | } |
||
311 | |||
2997 | Serge | 312 | /** |
313 | * radeon_gart_init - init the driver info for managing the gart |
||
314 | * |
||
315 | * @rdev: radeon_device pointer |
||
316 | * |
||
317 | * Allocate the dummy page and init the gart driver info (all asics). |
||
318 | * Returns 0 for success, error for failure. |
||
319 | */ |
||
1120 | serge | 320 | int radeon_gart_init(struct radeon_device *rdev) |
321 | { |
||
1430 | serge | 322 | int r, i; |
323 | |||
1120 | serge | 324 | if (rdev->gart.pages) { |
325 | return 0; |
||
326 | } |
||
1268 | serge | 327 | /* We need PAGE_SIZE >= RADEON_GPU_PAGE_SIZE */ |
328 | if (PAGE_SIZE < RADEON_GPU_PAGE_SIZE) { |
||
1120 | serge | 329 | DRM_ERROR("Page size is smaller than GPU page size!\n"); |
330 | return -EINVAL; |
||
331 | } |
||
1430 | serge | 332 | r = radeon_dummy_page_init(rdev); |
333 | if (r) |
||
334 | return r; |
||
1120 | serge | 335 | /* Compute table size */ |
336 | rdev->gart.num_cpu_pages = rdev->mc.gtt_size / PAGE_SIZE; |
||
1268 | serge | 337 | rdev->gart.num_gpu_pages = rdev->mc.gtt_size / RADEON_GPU_PAGE_SIZE; |
1120 | serge | 338 | DRM_INFO("GART: num cpu pages %u, num gpu pages %u\n", |
339 | rdev->gart.num_cpu_pages, rdev->gart.num_gpu_pages); |
||
340 | /* Allocate pages table */ |
||
2997 | Serge | 341 | rdev->gart.pages = vzalloc(sizeof(void *) * rdev->gart.num_cpu_pages); |
1120 | serge | 342 | if (rdev->gart.pages == NULL) { |
1404 | serge | 343 | radeon_gart_fini(rdev); |
1120 | serge | 344 | return -ENOMEM; |
345 | } |
||
2997 | Serge | 346 | rdev->gart.pages_addr = vzalloc(sizeof(dma_addr_t) * |
347 | rdev->gart.num_cpu_pages); |
||
1120 | serge | 348 | if (rdev->gart.pages_addr == NULL) { |
1404 | serge | 349 | radeon_gart_fini(rdev); |
1120 | serge | 350 | return -ENOMEM; |
351 | } |
||
1430 | serge | 352 | /* set GART entry to point to the dummy page by default */ |
353 | for (i = 0; i < rdev->gart.num_cpu_pages; i++) { |
||
354 | rdev->gart.pages_addr[i] = rdev->dummy_page.addr; |
||
355 | } |
||
1120 | serge | 356 | return 0; |
357 | } |
||
358 | |||
2997 | Serge | 359 | /** |
360 | * radeon_gart_fini - tear down the driver info for managing the gart |
||
361 | * |
||
362 | * @rdev: radeon_device pointer |
||
363 | * |
||
364 | * Tear down the gart driver info and free the dummy page (all asics). |
||
365 | */ |
||
1120 | serge | 366 | void radeon_gart_fini(struct radeon_device *rdev) |
367 | { |
||
368 | if (rdev->gart.pages && rdev->gart.pages_addr && rdev->gart.ready) { |
||
369 | /* unbind pages */ |
||
370 | radeon_gart_unbind(rdev, 0, rdev->gart.num_cpu_pages); |
||
371 | } |
||
372 | rdev->gart.ready = false; |
||
2997 | Serge | 373 | vfree(rdev->gart.pages); |
374 | vfree(rdev->gart.pages_addr); |
||
1120 | serge | 375 | rdev->gart.pages = NULL; |
376 | rdev->gart.pages_addr = NULL; |
||
2997 | Serge | 377 | |
5078 | serge | 378 | radeon_dummy_page_fini(rdev); |
2997 | Serge | 379 | }>>>>>> |