Commit | Line | Data |
---|---|---|
771fe6b9 JG |
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 | */ | |
3ce0a23d JG |
28 | #include <linux/seq_file.h> |
29 | #include <linux/firmware.h> | |
30 | #include <linux/platform_device.h> | |
771fe6b9 | 31 | #include "drmP.h" |
3ce0a23d | 32 | #include "radeon_drm.h" |
771fe6b9 | 33 | #include "radeon.h" |
3ce0a23d | 34 | #include "radeon_mode.h" |
3ce0a23d JG |
35 | #include "r600d.h" |
36 | #include "avivod.h" | |
37 | #include "atom.h" | |
771fe6b9 | 38 | |
3ce0a23d JG |
39 | #define PFP_UCODE_SIZE 576 |
40 | #define PM4_UCODE_SIZE 1792 | |
41 | #define R700_PFP_UCODE_SIZE 848 | |
42 | #define R700_PM4_UCODE_SIZE 1360 | |
43 | ||
44 | /* Firmware Names */ | |
45 | MODULE_FIRMWARE("radeon/R600_pfp.bin"); | |
46 | MODULE_FIRMWARE("radeon/R600_me.bin"); | |
47 | MODULE_FIRMWARE("radeon/RV610_pfp.bin"); | |
48 | MODULE_FIRMWARE("radeon/RV610_me.bin"); | |
49 | MODULE_FIRMWARE("radeon/RV630_pfp.bin"); | |
50 | MODULE_FIRMWARE("radeon/RV630_me.bin"); | |
51 | MODULE_FIRMWARE("radeon/RV620_pfp.bin"); | |
52 | MODULE_FIRMWARE("radeon/RV620_me.bin"); | |
53 | MODULE_FIRMWARE("radeon/RV635_pfp.bin"); | |
54 | MODULE_FIRMWARE("radeon/RV635_me.bin"); | |
55 | MODULE_FIRMWARE("radeon/RV670_pfp.bin"); | |
56 | MODULE_FIRMWARE("radeon/RV670_me.bin"); | |
57 | MODULE_FIRMWARE("radeon/RS780_pfp.bin"); | |
58 | MODULE_FIRMWARE("radeon/RS780_me.bin"); | |
59 | MODULE_FIRMWARE("radeon/RV770_pfp.bin"); | |
60 | MODULE_FIRMWARE("radeon/RV770_me.bin"); | |
61 | MODULE_FIRMWARE("radeon/RV730_pfp.bin"); | |
62 | MODULE_FIRMWARE("radeon/RV730_me.bin"); | |
63 | MODULE_FIRMWARE("radeon/RV710_pfp.bin"); | |
64 | MODULE_FIRMWARE("radeon/RV710_me.bin"); | |
65 | ||
66 | int r600_debugfs_mc_info_init(struct radeon_device *rdev); | |
771fe6b9 JG |
67 | |
68 | /* This files gather functions specifics to: | |
69 | * r600,rv610,rv630,rv620,rv635,rv670 | |
70 | * | |
71 | * Some of these functions might be used by newer ASICs. | |
72 | */ | |
73 | int r600_mc_wait_for_idle(struct radeon_device *rdev); | |
74 | void r600_gpu_init(struct radeon_device *rdev); | |
3ce0a23d | 75 | void r600_fini(struct radeon_device *rdev); |
771fe6b9 JG |
76 | |
77 | ||
78 | /* | |
3ce0a23d | 79 | * R600 PCIE GART |
771fe6b9 | 80 | */ |
3ce0a23d | 81 | int r600_gart_clear_page(struct radeon_device *rdev, int i) |
771fe6b9 | 82 | { |
3ce0a23d JG |
83 | void __iomem *ptr = (void *)rdev->gart.table.vram.ptr; |
84 | u64 pte; | |
771fe6b9 | 85 | |
3ce0a23d JG |
86 | if (i < 0 || i > rdev->gart.num_gpu_pages) |
87 | return -EINVAL; | |
88 | pte = 0; | |
89 | writeq(pte, ((void __iomem *)ptr) + (i * 8)); | |
90 | return 0; | |
91 | } | |
771fe6b9 | 92 | |
3ce0a23d JG |
93 | void r600_pcie_gart_tlb_flush(struct radeon_device *rdev) |
94 | { | |
95 | unsigned i; | |
96 | u32 tmp; | |
97 | ||
98 | WREG32(VM_CONTEXT0_INVALIDATION_LOW_ADDR, rdev->mc.gtt_start >> 12); | |
99 | WREG32(VM_CONTEXT0_INVALIDATION_HIGH_ADDR, (rdev->mc.gtt_end - 1) >> 12); | |
100 | WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1)); | |
101 | for (i = 0; i < rdev->usec_timeout; i++) { | |
102 | /* read MC_STATUS */ | |
103 | tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE); | |
104 | tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT; | |
105 | if (tmp == 2) { | |
106 | printk(KERN_WARNING "[drm] r600 flush TLB failed\n"); | |
107 | return; | |
108 | } | |
109 | if (tmp) { | |
110 | return; | |
111 | } | |
112 | udelay(1); | |
113 | } | |
114 | } | |
115 | ||
4aac0473 | 116 | int r600_pcie_gart_init(struct radeon_device *rdev) |
3ce0a23d | 117 | { |
4aac0473 | 118 | int r; |
3ce0a23d | 119 | |
4aac0473 JG |
120 | if (rdev->gart.table.vram.robj) { |
121 | WARN(1, "R600 PCIE GART already initialized.\n"); | |
122 | return 0; | |
123 | } | |
3ce0a23d JG |
124 | /* Initialize common gart structure */ |
125 | r = radeon_gart_init(rdev); | |
4aac0473 | 126 | if (r) |
3ce0a23d | 127 | return r; |
3ce0a23d | 128 | rdev->gart.table_size = rdev->gart.num_gpu_pages * 8; |
4aac0473 JG |
129 | return radeon_gart_table_vram_alloc(rdev); |
130 | } | |
131 | ||
132 | int r600_pcie_gart_enable(struct radeon_device *rdev) | |
133 | { | |
134 | u32 tmp; | |
135 | int r, i; | |
136 | ||
137 | if (rdev->gart.table.vram.robj == NULL) { | |
138 | dev_err(rdev->dev, "No VRAM object for PCIE GART.\n"); | |
139 | return -EINVAL; | |
771fe6b9 | 140 | } |
4aac0473 JG |
141 | r = radeon_gart_table_vram_pin(rdev); |
142 | if (r) | |
143 | return r; | |
3ce0a23d JG |
144 | for (i = 0; i < rdev->gart.num_gpu_pages; i++) |
145 | r600_gart_clear_page(rdev, i); | |
146 | /* Setup L2 cache */ | |
147 | WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING | | |
148 | ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | | |
149 | EFFECTIVE_L2_QUEUE_SIZE(7)); | |
150 | WREG32(VM_L2_CNTL2, 0); | |
151 | WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1)); | |
152 | /* Setup TLB control */ | |
153 | tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING | | |
154 | SYSTEM_ACCESS_MODE_NOT_IN_SYS | | |
155 | EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) | | |
156 | ENABLE_WAIT_L2_QUERY; | |
157 | WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp); | |
158 | WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp); | |
159 | WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING); | |
160 | WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp); | |
161 | WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp); | |
162 | WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp); | |
163 | WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp); | |
164 | WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp); | |
165 | WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp); | |
166 | WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp); | |
167 | WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp); | |
168 | WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp); | |
169 | WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE); | |
170 | WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE); | |
171 | WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12); | |
172 | WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, (rdev->mc.gtt_end - 1) >> 12); | |
173 | WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12); | |
174 | WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) | | |
175 | RANGE_PROTECTION_FAULT_ENABLE_DEFAULT); | |
176 | WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR, | |
177 | (u32)(rdev->dummy_page.addr >> 12)); | |
178 | for (i = 1; i < 7; i++) | |
179 | WREG32(VM_CONTEXT0_CNTL + (i * 4), 0); | |
771fe6b9 | 180 | |
3ce0a23d JG |
181 | r600_pcie_gart_tlb_flush(rdev); |
182 | rdev->gart.ready = true; | |
771fe6b9 JG |
183 | return 0; |
184 | } | |
185 | ||
3ce0a23d | 186 | void r600_pcie_gart_disable(struct radeon_device *rdev) |
771fe6b9 | 187 | { |
3ce0a23d JG |
188 | u32 tmp; |
189 | int i; | |
771fe6b9 | 190 | |
3ce0a23d JG |
191 | /* Disable all tables */ |
192 | for (i = 0; i < 7; i++) | |
193 | WREG32(VM_CONTEXT0_CNTL + (i * 4), 0); | |
771fe6b9 | 194 | |
3ce0a23d JG |
195 | /* Disable L2 cache */ |
196 | WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING | | |
197 | EFFECTIVE_L2_QUEUE_SIZE(7)); | |
198 | WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1)); | |
199 | /* Setup L1 TLB control */ | |
200 | tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) | | |
201 | ENABLE_WAIT_L2_QUERY; | |
202 | WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp); | |
203 | WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp); | |
204 | WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp); | |
205 | WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp); | |
206 | WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp); | |
207 | WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp); | |
208 | WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp); | |
209 | WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp); | |
210 | WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp); | |
211 | WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp); | |
212 | WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp); | |
213 | WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp); | |
214 | WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp); | |
215 | WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp); | |
4aac0473 JG |
216 | if (rdev->gart.table.vram.robj) { |
217 | radeon_object_kunmap(rdev->gart.table.vram.robj); | |
218 | radeon_object_unpin(rdev->gart.table.vram.robj); | |
219 | } | |
220 | } | |
221 | ||
222 | void r600_pcie_gart_fini(struct radeon_device *rdev) | |
223 | { | |
224 | r600_pcie_gart_disable(rdev); | |
225 | radeon_gart_table_vram_free(rdev); | |
226 | radeon_gart_fini(rdev); | |
771fe6b9 JG |
227 | } |
228 | ||
229 | int r600_mc_wait_for_idle(struct radeon_device *rdev) | |
230 | { | |
3ce0a23d JG |
231 | unsigned i; |
232 | u32 tmp; | |
233 | ||
234 | for (i = 0; i < rdev->usec_timeout; i++) { | |
235 | /* read MC_STATUS */ | |
236 | tmp = RREG32(R_000E50_SRBM_STATUS) & 0x3F00; | |
237 | if (!tmp) | |
238 | return 0; | |
239 | udelay(1); | |
240 | } | |
241 | return -1; | |
771fe6b9 JG |
242 | } |
243 | ||
3ce0a23d | 244 | static void r600_mc_resume(struct radeon_device *rdev) |
771fe6b9 | 245 | { |
3ce0a23d JG |
246 | u32 d1vga_control, d2vga_control; |
247 | u32 vga_render_control, vga_hdp_control; | |
248 | u32 d1crtc_control, d2crtc_control; | |
249 | u32 new_d1grph_primary, new_d1grph_secondary; | |
250 | u32 new_d2grph_primary, new_d2grph_secondary; | |
251 | u64 old_vram_start; | |
252 | u32 tmp; | |
253 | int i, j; | |
771fe6b9 | 254 | |
3ce0a23d JG |
255 | /* Initialize HDP */ |
256 | for (i = 0, j = 0; i < 32; i++, j += 0x18) { | |
257 | WREG32((0x2c14 + j), 0x00000000); | |
258 | WREG32((0x2c18 + j), 0x00000000); | |
259 | WREG32((0x2c1c + j), 0x00000000); | |
260 | WREG32((0x2c20 + j), 0x00000000); | |
261 | WREG32((0x2c24 + j), 0x00000000); | |
262 | } | |
263 | WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0); | |
771fe6b9 | 264 | |
3ce0a23d JG |
265 | d1vga_control = RREG32(D1VGA_CONTROL); |
266 | d2vga_control = RREG32(D2VGA_CONTROL); | |
267 | vga_render_control = RREG32(VGA_RENDER_CONTROL); | |
268 | vga_hdp_control = RREG32(VGA_HDP_CONTROL); | |
269 | d1crtc_control = RREG32(D1CRTC_CONTROL); | |
270 | d2crtc_control = RREG32(D2CRTC_CONTROL); | |
271 | old_vram_start = (u64)(RREG32(MC_VM_FB_LOCATION) & 0xFFFF) << 24; | |
272 | new_d1grph_primary = RREG32(D1GRPH_PRIMARY_SURFACE_ADDRESS); | |
273 | new_d1grph_secondary = RREG32(D1GRPH_SECONDARY_SURFACE_ADDRESS); | |
274 | new_d1grph_primary += rdev->mc.vram_start - old_vram_start; | |
275 | new_d1grph_secondary += rdev->mc.vram_start - old_vram_start; | |
276 | new_d2grph_primary = RREG32(D2GRPH_PRIMARY_SURFACE_ADDRESS); | |
277 | new_d2grph_secondary = RREG32(D2GRPH_SECONDARY_SURFACE_ADDRESS); | |
278 | new_d2grph_primary += rdev->mc.vram_start - old_vram_start; | |
279 | new_d2grph_secondary += rdev->mc.vram_start - old_vram_start; | |
280 | ||
281 | /* Stop all video */ | |
282 | WREG32(D1VGA_CONTROL, 0); | |
283 | WREG32(D2VGA_CONTROL, 0); | |
284 | WREG32(VGA_RENDER_CONTROL, 0); | |
285 | WREG32(D1CRTC_UPDATE_LOCK, 1); | |
286 | WREG32(D2CRTC_UPDATE_LOCK, 1); | |
287 | WREG32(D1CRTC_CONTROL, 0); | |
288 | WREG32(D2CRTC_CONTROL, 0); | |
289 | WREG32(D1CRTC_UPDATE_LOCK, 0); | |
290 | WREG32(D2CRTC_UPDATE_LOCK, 0); | |
291 | ||
292 | mdelay(1); | |
293 | if (r600_mc_wait_for_idle(rdev)) { | |
294 | printk(KERN_WARNING "[drm] MC not idle !\n"); | |
295 | } | |
296 | ||
297 | /* Lockout access through VGA aperture*/ | |
298 | WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE); | |
299 | ||
300 | /* Update configuration */ | |
301 | WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, rdev->mc.vram_start >> 12); | |
302 | WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, (rdev->mc.vram_end - 1) >> 12); | |
303 | WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0); | |
304 | tmp = (((rdev->mc.vram_end - 1) >> 24) & 0xFFFF) << 16; | |
305 | tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF); | |
306 | WREG32(MC_VM_FB_LOCATION, tmp); | |
307 | WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8)); | |
308 | WREG32(HDP_NONSURFACE_INFO, (2 << 7)); | |
309 | WREG32(HDP_NONSURFACE_SIZE, (rdev->mc.mc_vram_size - 1) | 0x3FF); | |
310 | if (rdev->flags & RADEON_IS_AGP) { | |
311 | WREG32(MC_VM_AGP_TOP, (rdev->mc.gtt_end - 1) >> 16); | |
312 | WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 16); | |
313 | WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22); | |
314 | } else { | |
315 | WREG32(MC_VM_AGP_BASE, 0); | |
316 | WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF); | |
317 | WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF); | |
318 | } | |
319 | WREG32(D1GRPH_PRIMARY_SURFACE_ADDRESS, new_d1grph_primary); | |
320 | WREG32(D1GRPH_SECONDARY_SURFACE_ADDRESS, new_d1grph_secondary); | |
321 | WREG32(D2GRPH_PRIMARY_SURFACE_ADDRESS, new_d2grph_primary); | |
322 | WREG32(D2GRPH_SECONDARY_SURFACE_ADDRESS, new_d2grph_secondary); | |
323 | WREG32(VGA_MEMORY_BASE_ADDRESS, rdev->mc.vram_start); | |
324 | ||
325 | /* Unlock host access */ | |
326 | WREG32(VGA_HDP_CONTROL, vga_hdp_control); | |
327 | ||
328 | mdelay(1); | |
329 | if (r600_mc_wait_for_idle(rdev)) { | |
330 | printk(KERN_WARNING "[drm] MC not idle !\n"); | |
331 | } | |
332 | ||
333 | /* Restore video state */ | |
334 | WREG32(D1CRTC_UPDATE_LOCK, 1); | |
335 | WREG32(D2CRTC_UPDATE_LOCK, 1); | |
336 | WREG32(D1CRTC_CONTROL, d1crtc_control); | |
337 | WREG32(D2CRTC_CONTROL, d2crtc_control); | |
338 | WREG32(D1CRTC_UPDATE_LOCK, 0); | |
339 | WREG32(D2CRTC_UPDATE_LOCK, 0); | |
340 | WREG32(D1VGA_CONTROL, d1vga_control); | |
341 | WREG32(D2VGA_CONTROL, d2vga_control); | |
342 | WREG32(VGA_RENDER_CONTROL, vga_render_control); | |
343 | } | |
344 | ||
345 | int r600_mc_init(struct radeon_device *rdev) | |
771fe6b9 | 346 | { |
3ce0a23d JG |
347 | fixed20_12 a; |
348 | u32 tmp; | |
771fe6b9 | 349 | int chansize; |
3ce0a23d | 350 | int r; |
771fe6b9 | 351 | |
3ce0a23d | 352 | /* Get VRAM informations */ |
771fe6b9 JG |
353 | rdev->mc.vram_width = 128; |
354 | rdev->mc.vram_is_ddr = true; | |
3ce0a23d JG |
355 | tmp = RREG32(RAMCFG); |
356 | if (tmp & CHANSIZE_OVERRIDE) { | |
771fe6b9 | 357 | chansize = 16; |
3ce0a23d | 358 | } else if (tmp & CHANSIZE_MASK) { |
771fe6b9 JG |
359 | chansize = 64; |
360 | } else { | |
361 | chansize = 32; | |
362 | } | |
363 | if (rdev->family == CHIP_R600) { | |
364 | rdev->mc.vram_width = 8 * chansize; | |
365 | } else if (rdev->family == CHIP_RV670) { | |
366 | rdev->mc.vram_width = 4 * chansize; | |
367 | } else if ((rdev->family == CHIP_RV610) || | |
368 | (rdev->family == CHIP_RV620)) { | |
369 | rdev->mc.vram_width = chansize; | |
370 | } else if ((rdev->family == CHIP_RV630) || | |
371 | (rdev->family == CHIP_RV635)) { | |
372 | rdev->mc.vram_width = 2 * chansize; | |
373 | } | |
3ce0a23d JG |
374 | /* Could aper size report 0 ? */ |
375 | rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0); | |
376 | rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0); | |
377 | /* Setup GPU memory space */ | |
378 | rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE); | |
379 | rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE); | |
380 | if (rdev->flags & RADEON_IS_AGP) { | |
381 | r = radeon_agp_init(rdev); | |
382 | if (r) | |
383 | return r; | |
384 | /* gtt_size is setup by radeon_agp_init */ | |
385 | rdev->mc.gtt_location = rdev->mc.agp_base; | |
386 | tmp = 0xFFFFFFFFUL - rdev->mc.agp_base - rdev->mc.gtt_size; | |
387 | /* Try to put vram before or after AGP because we | |
388 | * we want SYSTEM_APERTURE to cover both VRAM and | |
389 | * AGP so that GPU can catch out of VRAM/AGP access | |
390 | */ | |
391 | if (rdev->mc.gtt_location > rdev->mc.mc_vram_size) { | |
392 | /* Enought place before */ | |
393 | rdev->mc.vram_location = rdev->mc.gtt_location - | |
394 | rdev->mc.mc_vram_size; | |
395 | } else if (tmp > rdev->mc.mc_vram_size) { | |
396 | /* Enought place after */ | |
397 | rdev->mc.vram_location = rdev->mc.gtt_location + | |
398 | rdev->mc.gtt_size; | |
399 | } else { | |
400 | /* Try to setup VRAM then AGP might not | |
401 | * not work on some card | |
402 | */ | |
403 | rdev->mc.vram_location = 0x00000000UL; | |
404 | rdev->mc.gtt_location = rdev->mc.mc_vram_size; | |
405 | } | |
406 | } else { | |
407 | if (rdev->family == CHIP_RS780 || rdev->family == CHIP_RS880) { | |
408 | rdev->mc.vram_location = (RREG32(MC_VM_FB_LOCATION) & | |
409 | 0xFFFF) << 24; | |
410 | rdev->mc.gtt_size = radeon_gart_size * 1024 * 1024; | |
411 | tmp = rdev->mc.vram_location + rdev->mc.mc_vram_size; | |
412 | if ((0xFFFFFFFFUL - tmp) >= rdev->mc.gtt_size) { | |
413 | /* Enough place after vram */ | |
414 | rdev->mc.gtt_location = tmp; | |
415 | } else if (rdev->mc.vram_location >= rdev->mc.gtt_size) { | |
416 | /* Enough place before vram */ | |
417 | rdev->mc.gtt_location = 0; | |
418 | } else { | |
419 | /* Not enough place after or before shrink | |
420 | * gart size | |
421 | */ | |
422 | if (rdev->mc.vram_location > (0xFFFFFFFFUL - tmp)) { | |
423 | rdev->mc.gtt_location = 0; | |
424 | rdev->mc.gtt_size = rdev->mc.vram_location; | |
425 | } else { | |
426 | rdev->mc.gtt_location = tmp; | |
427 | rdev->mc.gtt_size = 0xFFFFFFFFUL - tmp; | |
428 | } | |
429 | } | |
430 | rdev->mc.gtt_location = rdev->mc.mc_vram_size; | |
431 | } else { | |
432 | rdev->mc.vram_location = 0x00000000UL; | |
433 | rdev->mc.gtt_location = rdev->mc.mc_vram_size; | |
434 | rdev->mc.gtt_size = radeon_gart_size * 1024 * 1024; | |
435 | } | |
436 | } | |
437 | rdev->mc.vram_start = rdev->mc.vram_location; | |
438 | rdev->mc.vram_end = rdev->mc.vram_location + rdev->mc.mc_vram_size; | |
439 | rdev->mc.gtt_start = rdev->mc.gtt_location; | |
440 | rdev->mc.gtt_end = rdev->mc.gtt_location + rdev->mc.gtt_size; | |
441 | /* FIXME: we should enforce default clock in case GPU is not in | |
442 | * default setup | |
443 | */ | |
444 | a.full = rfixed_const(100); | |
445 | rdev->pm.sclk.full = rfixed_const(rdev->clock.default_sclk); | |
446 | rdev->pm.sclk.full = rfixed_div(rdev->pm.sclk, a); | |
447 | return 0; | |
771fe6b9 JG |
448 | } |
449 | ||
3ce0a23d JG |
450 | /* We doesn't check that the GPU really needs a reset we simply do the |
451 | * reset, it's up to the caller to determine if the GPU needs one. We | |
452 | * might add an helper function to check that. | |
453 | */ | |
454 | int r600_gpu_soft_reset(struct radeon_device *rdev) | |
771fe6b9 | 455 | { |
3ce0a23d JG |
456 | u32 grbm_busy_mask = S_008010_VC_BUSY(1) | S_008010_VGT_BUSY_NO_DMA(1) | |
457 | S_008010_VGT_BUSY(1) | S_008010_TA03_BUSY(1) | | |
458 | S_008010_TC_BUSY(1) | S_008010_SX_BUSY(1) | | |
459 | S_008010_SH_BUSY(1) | S_008010_SPI03_BUSY(1) | | |
460 | S_008010_SMX_BUSY(1) | S_008010_SC_BUSY(1) | | |
461 | S_008010_PA_BUSY(1) | S_008010_DB03_BUSY(1) | | |
462 | S_008010_CR_BUSY(1) | S_008010_CB03_BUSY(1) | | |
463 | S_008010_GUI_ACTIVE(1); | |
464 | u32 grbm2_busy_mask = S_008014_SPI0_BUSY(1) | S_008014_SPI1_BUSY(1) | | |
465 | S_008014_SPI2_BUSY(1) | S_008014_SPI3_BUSY(1) | | |
466 | S_008014_TA0_BUSY(1) | S_008014_TA1_BUSY(1) | | |
467 | S_008014_TA2_BUSY(1) | S_008014_TA3_BUSY(1) | | |
468 | S_008014_DB0_BUSY(1) | S_008014_DB1_BUSY(1) | | |
469 | S_008014_DB2_BUSY(1) | S_008014_DB3_BUSY(1) | | |
470 | S_008014_CB0_BUSY(1) | S_008014_CB1_BUSY(1) | | |
471 | S_008014_CB2_BUSY(1) | S_008014_CB3_BUSY(1); | |
472 | u32 srbm_reset = 0; | |
771fe6b9 | 473 | |
3ce0a23d JG |
474 | /* Disable CP parsing/prefetching */ |
475 | WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(0xff)); | |
476 | /* Check if any of the rendering block is busy and reset it */ | |
477 | if ((RREG32(R_008010_GRBM_STATUS) & grbm_busy_mask) || | |
478 | (RREG32(R_008014_GRBM_STATUS2) & grbm2_busy_mask)) { | |
479 | WREG32(R_008020_GRBM_SOFT_RESET, S_008020_SOFT_RESET_CR(1) | | |
480 | S_008020_SOFT_RESET_DB(1) | | |
481 | S_008020_SOFT_RESET_CB(1) | | |
482 | S_008020_SOFT_RESET_PA(1) | | |
483 | S_008020_SOFT_RESET_SC(1) | | |
484 | S_008020_SOFT_RESET_SMX(1) | | |
485 | S_008020_SOFT_RESET_SPI(1) | | |
486 | S_008020_SOFT_RESET_SX(1) | | |
487 | S_008020_SOFT_RESET_SH(1) | | |
488 | S_008020_SOFT_RESET_TC(1) | | |
489 | S_008020_SOFT_RESET_TA(1) | | |
490 | S_008020_SOFT_RESET_VC(1) | | |
491 | S_008020_SOFT_RESET_VGT(1)); | |
492 | (void)RREG32(R_008020_GRBM_SOFT_RESET); | |
493 | udelay(50); | |
494 | WREG32(R_008020_GRBM_SOFT_RESET, 0); | |
495 | (void)RREG32(R_008020_GRBM_SOFT_RESET); | |
496 | } | |
497 | /* Reset CP (we always reset CP) */ | |
498 | WREG32(R_008020_GRBM_SOFT_RESET, S_008020_SOFT_RESET_CP(1)); | |
499 | (void)RREG32(R_008020_GRBM_SOFT_RESET); | |
500 | udelay(50); | |
501 | WREG32(R_008020_GRBM_SOFT_RESET, 0); | |
502 | (void)RREG32(R_008020_GRBM_SOFT_RESET); | |
503 | /* Reset others GPU block if necessary */ | |
504 | if (G_000E50_RLC_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
505 | srbm_reset |= S_000E60_SOFT_RESET_RLC(1); | |
506 | if (G_000E50_GRBM_RQ_PENDING(RREG32(R_000E50_SRBM_STATUS))) | |
507 | srbm_reset |= S_000E60_SOFT_RESET_GRBM(1); | |
508 | if (G_000E50_HI_RQ_PENDING(RREG32(R_000E50_SRBM_STATUS))) | |
509 | srbm_reset |= S_000E60_SOFT_RESET_IH(1); | |
510 | if (G_000E50_VMC_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
511 | srbm_reset |= S_000E60_SOFT_RESET_VMC(1); | |
512 | if (G_000E50_MCB_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
513 | srbm_reset |= S_000E60_SOFT_RESET_MC(1); | |
514 | if (G_000E50_MCDZ_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
515 | srbm_reset |= S_000E60_SOFT_RESET_MC(1); | |
516 | if (G_000E50_MCDY_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
517 | srbm_reset |= S_000E60_SOFT_RESET_MC(1); | |
518 | if (G_000E50_MCDX_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
519 | srbm_reset |= S_000E60_SOFT_RESET_MC(1); | |
520 | if (G_000E50_MCDW_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
521 | srbm_reset |= S_000E60_SOFT_RESET_MC(1); | |
522 | if (G_000E50_RLC_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
523 | srbm_reset |= S_000E60_SOFT_RESET_RLC(1); | |
524 | if (G_000E50_SEM_BUSY(RREG32(R_000E50_SRBM_STATUS))) | |
525 | srbm_reset |= S_000E60_SOFT_RESET_SEM(1); | |
526 | WREG32(R_000E60_SRBM_SOFT_RESET, srbm_reset); | |
527 | (void)RREG32(R_000E60_SRBM_SOFT_RESET); | |
528 | udelay(50); | |
529 | WREG32(R_000E60_SRBM_SOFT_RESET, 0); | |
530 | (void)RREG32(R_000E60_SRBM_SOFT_RESET); | |
531 | /* Wait a little for things to settle down */ | |
532 | udelay(50); | |
533 | return 0; | |
534 | } | |
535 | ||
536 | int r600_gpu_reset(struct radeon_device *rdev) | |
537 | { | |
538 | return r600_gpu_soft_reset(rdev); | |
539 | } | |
540 | ||
541 | static u32 r600_get_tile_pipe_to_backend_map(u32 num_tile_pipes, | |
542 | u32 num_backends, | |
543 | u32 backend_disable_mask) | |
544 | { | |
545 | u32 backend_map = 0; | |
546 | u32 enabled_backends_mask; | |
547 | u32 enabled_backends_count; | |
548 | u32 cur_pipe; | |
549 | u32 swizzle_pipe[R6XX_MAX_PIPES]; | |
550 | u32 cur_backend; | |
551 | u32 i; | |
552 | ||
553 | if (num_tile_pipes > R6XX_MAX_PIPES) | |
554 | num_tile_pipes = R6XX_MAX_PIPES; | |
555 | if (num_tile_pipes < 1) | |
556 | num_tile_pipes = 1; | |
557 | if (num_backends > R6XX_MAX_BACKENDS) | |
558 | num_backends = R6XX_MAX_BACKENDS; | |
559 | if (num_backends < 1) | |
560 | num_backends = 1; | |
561 | ||
562 | enabled_backends_mask = 0; | |
563 | enabled_backends_count = 0; | |
564 | for (i = 0; i < R6XX_MAX_BACKENDS; ++i) { | |
565 | if (((backend_disable_mask >> i) & 1) == 0) { | |
566 | enabled_backends_mask |= (1 << i); | |
567 | ++enabled_backends_count; | |
568 | } | |
569 | if (enabled_backends_count == num_backends) | |
570 | break; | |
571 | } | |
572 | ||
573 | if (enabled_backends_count == 0) { | |
574 | enabled_backends_mask = 1; | |
575 | enabled_backends_count = 1; | |
576 | } | |
577 | ||
578 | if (enabled_backends_count != num_backends) | |
579 | num_backends = enabled_backends_count; | |
580 | ||
581 | memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R6XX_MAX_PIPES); | |
582 | switch (num_tile_pipes) { | |
583 | case 1: | |
584 | swizzle_pipe[0] = 0; | |
585 | break; | |
586 | case 2: | |
587 | swizzle_pipe[0] = 0; | |
588 | swizzle_pipe[1] = 1; | |
589 | break; | |
590 | case 3: | |
591 | swizzle_pipe[0] = 0; | |
592 | swizzle_pipe[1] = 1; | |
593 | swizzle_pipe[2] = 2; | |
594 | break; | |
595 | case 4: | |
596 | swizzle_pipe[0] = 0; | |
597 | swizzle_pipe[1] = 1; | |
598 | swizzle_pipe[2] = 2; | |
599 | swizzle_pipe[3] = 3; | |
600 | break; | |
601 | case 5: | |
602 | swizzle_pipe[0] = 0; | |
603 | swizzle_pipe[1] = 1; | |
604 | swizzle_pipe[2] = 2; | |
605 | swizzle_pipe[3] = 3; | |
606 | swizzle_pipe[4] = 4; | |
607 | break; | |
608 | case 6: | |
609 | swizzle_pipe[0] = 0; | |
610 | swizzle_pipe[1] = 2; | |
611 | swizzle_pipe[2] = 4; | |
612 | swizzle_pipe[3] = 5; | |
613 | swizzle_pipe[4] = 1; | |
614 | swizzle_pipe[5] = 3; | |
615 | break; | |
616 | case 7: | |
617 | swizzle_pipe[0] = 0; | |
618 | swizzle_pipe[1] = 2; | |
619 | swizzle_pipe[2] = 4; | |
620 | swizzle_pipe[3] = 6; | |
621 | swizzle_pipe[4] = 1; | |
622 | swizzle_pipe[5] = 3; | |
623 | swizzle_pipe[6] = 5; | |
624 | break; | |
625 | case 8: | |
626 | swizzle_pipe[0] = 0; | |
627 | swizzle_pipe[1] = 2; | |
628 | swizzle_pipe[2] = 4; | |
629 | swizzle_pipe[3] = 6; | |
630 | swizzle_pipe[4] = 1; | |
631 | swizzle_pipe[5] = 3; | |
632 | swizzle_pipe[6] = 5; | |
633 | swizzle_pipe[7] = 7; | |
634 | break; | |
635 | } | |
636 | ||
637 | cur_backend = 0; | |
638 | for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) { | |
639 | while (((1 << cur_backend) & enabled_backends_mask) == 0) | |
640 | cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS; | |
641 | ||
642 | backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2))); | |
643 | ||
644 | cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS; | |
645 | } | |
646 | ||
647 | return backend_map; | |
648 | } | |
649 | ||
650 | int r600_count_pipe_bits(uint32_t val) | |
651 | { | |
652 | int i, ret = 0; | |
653 | ||
654 | for (i = 0; i < 32; i++) { | |
655 | ret += val & 1; | |
656 | val >>= 1; | |
657 | } | |
658 | return ret; | |
771fe6b9 JG |
659 | } |
660 | ||
3ce0a23d JG |
661 | void r600_gpu_init(struct radeon_device *rdev) |
662 | { | |
663 | u32 tiling_config; | |
664 | u32 ramcfg; | |
665 | u32 tmp; | |
666 | int i, j; | |
667 | u32 sq_config; | |
668 | u32 sq_gpr_resource_mgmt_1 = 0; | |
669 | u32 sq_gpr_resource_mgmt_2 = 0; | |
670 | u32 sq_thread_resource_mgmt = 0; | |
671 | u32 sq_stack_resource_mgmt_1 = 0; | |
672 | u32 sq_stack_resource_mgmt_2 = 0; | |
673 | ||
674 | /* FIXME: implement */ | |
675 | switch (rdev->family) { | |
676 | case CHIP_R600: | |
677 | rdev->config.r600.max_pipes = 4; | |
678 | rdev->config.r600.max_tile_pipes = 8; | |
679 | rdev->config.r600.max_simds = 4; | |
680 | rdev->config.r600.max_backends = 4; | |
681 | rdev->config.r600.max_gprs = 256; | |
682 | rdev->config.r600.max_threads = 192; | |
683 | rdev->config.r600.max_stack_entries = 256; | |
684 | rdev->config.r600.max_hw_contexts = 8; | |
685 | rdev->config.r600.max_gs_threads = 16; | |
686 | rdev->config.r600.sx_max_export_size = 128; | |
687 | rdev->config.r600.sx_max_export_pos_size = 16; | |
688 | rdev->config.r600.sx_max_export_smx_size = 128; | |
689 | rdev->config.r600.sq_num_cf_insts = 2; | |
690 | break; | |
691 | case CHIP_RV630: | |
692 | case CHIP_RV635: | |
693 | rdev->config.r600.max_pipes = 2; | |
694 | rdev->config.r600.max_tile_pipes = 2; | |
695 | rdev->config.r600.max_simds = 3; | |
696 | rdev->config.r600.max_backends = 1; | |
697 | rdev->config.r600.max_gprs = 128; | |
698 | rdev->config.r600.max_threads = 192; | |
699 | rdev->config.r600.max_stack_entries = 128; | |
700 | rdev->config.r600.max_hw_contexts = 8; | |
701 | rdev->config.r600.max_gs_threads = 4; | |
702 | rdev->config.r600.sx_max_export_size = 128; | |
703 | rdev->config.r600.sx_max_export_pos_size = 16; | |
704 | rdev->config.r600.sx_max_export_smx_size = 128; | |
705 | rdev->config.r600.sq_num_cf_insts = 2; | |
706 | break; | |
707 | case CHIP_RV610: | |
708 | case CHIP_RV620: | |
709 | case CHIP_RS780: | |
710 | case CHIP_RS880: | |
711 | rdev->config.r600.max_pipes = 1; | |
712 | rdev->config.r600.max_tile_pipes = 1; | |
713 | rdev->config.r600.max_simds = 2; | |
714 | rdev->config.r600.max_backends = 1; | |
715 | rdev->config.r600.max_gprs = 128; | |
716 | rdev->config.r600.max_threads = 192; | |
717 | rdev->config.r600.max_stack_entries = 128; | |
718 | rdev->config.r600.max_hw_contexts = 4; | |
719 | rdev->config.r600.max_gs_threads = 4; | |
720 | rdev->config.r600.sx_max_export_size = 128; | |
721 | rdev->config.r600.sx_max_export_pos_size = 16; | |
722 | rdev->config.r600.sx_max_export_smx_size = 128; | |
723 | rdev->config.r600.sq_num_cf_insts = 1; | |
724 | break; | |
725 | case CHIP_RV670: | |
726 | rdev->config.r600.max_pipes = 4; | |
727 | rdev->config.r600.max_tile_pipes = 4; | |
728 | rdev->config.r600.max_simds = 4; | |
729 | rdev->config.r600.max_backends = 4; | |
730 | rdev->config.r600.max_gprs = 192; | |
731 | rdev->config.r600.max_threads = 192; | |
732 | rdev->config.r600.max_stack_entries = 256; | |
733 | rdev->config.r600.max_hw_contexts = 8; | |
734 | rdev->config.r600.max_gs_threads = 16; | |
735 | rdev->config.r600.sx_max_export_size = 128; | |
736 | rdev->config.r600.sx_max_export_pos_size = 16; | |
737 | rdev->config.r600.sx_max_export_smx_size = 128; | |
738 | rdev->config.r600.sq_num_cf_insts = 2; | |
739 | break; | |
740 | default: | |
741 | break; | |
742 | } | |
743 | ||
744 | /* Initialize HDP */ | |
745 | for (i = 0, j = 0; i < 32; i++, j += 0x18) { | |
746 | WREG32((0x2c14 + j), 0x00000000); | |
747 | WREG32((0x2c18 + j), 0x00000000); | |
748 | WREG32((0x2c1c + j), 0x00000000); | |
749 | WREG32((0x2c20 + j), 0x00000000); | |
750 | WREG32((0x2c24 + j), 0x00000000); | |
751 | } | |
752 | ||
753 | WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff)); | |
754 | ||
755 | /* Setup tiling */ | |
756 | tiling_config = 0; | |
757 | ramcfg = RREG32(RAMCFG); | |
758 | switch (rdev->config.r600.max_tile_pipes) { | |
759 | case 1: | |
760 | tiling_config |= PIPE_TILING(0); | |
761 | break; | |
762 | case 2: | |
763 | tiling_config |= PIPE_TILING(1); | |
764 | break; | |
765 | case 4: | |
766 | tiling_config |= PIPE_TILING(2); | |
767 | break; | |
768 | case 8: | |
769 | tiling_config |= PIPE_TILING(3); | |
770 | break; | |
771 | default: | |
772 | break; | |
773 | } | |
774 | tiling_config |= BANK_TILING((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT); | |
775 | tiling_config |= GROUP_SIZE(0); | |
776 | tmp = (ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT; | |
777 | if (tmp > 3) { | |
778 | tiling_config |= ROW_TILING(3); | |
779 | tiling_config |= SAMPLE_SPLIT(3); | |
780 | } else { | |
781 | tiling_config |= ROW_TILING(tmp); | |
782 | tiling_config |= SAMPLE_SPLIT(tmp); | |
783 | } | |
784 | tiling_config |= BANK_SWAPS(1); | |
785 | tmp = r600_get_tile_pipe_to_backend_map(rdev->config.r600.max_tile_pipes, | |
786 | rdev->config.r600.max_backends, | |
787 | (0xff << rdev->config.r600.max_backends) & 0xff); | |
788 | tiling_config |= BACKEND_MAP(tmp); | |
789 | WREG32(GB_TILING_CONFIG, tiling_config); | |
790 | WREG32(DCP_TILING_CONFIG, tiling_config & 0xffff); | |
791 | WREG32(HDP_TILING_CONFIG, tiling_config & 0xffff); | |
792 | ||
793 | tmp = BACKEND_DISABLE((R6XX_MAX_BACKENDS_MASK << rdev->config.r600.max_backends) & R6XX_MAX_BACKENDS_MASK); | |
794 | WREG32(CC_RB_BACKEND_DISABLE, tmp); | |
795 | ||
796 | /* Setup pipes */ | |
797 | tmp = INACTIVE_QD_PIPES((R6XX_MAX_PIPES_MASK << rdev->config.r600.max_pipes) & R6XX_MAX_PIPES_MASK); | |
798 | tmp |= INACTIVE_SIMDS((R6XX_MAX_SIMDS_MASK << rdev->config.r600.max_simds) & R6XX_MAX_SIMDS_MASK); | |
799 | WREG32(CC_GC_SHADER_PIPE_CONFIG, tmp); | |
800 | WREG32(GC_USER_SHADER_PIPE_CONFIG, tmp); | |
801 | ||
802 | tmp = R6XX_MAX_BACKENDS - r600_count_pipe_bits(tmp & INACTIVE_QD_PIPES_MASK); | |
803 | WREG32(VGT_OUT_DEALLOC_CNTL, (tmp * 4) & DEALLOC_DIST_MASK); | |
804 | WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, ((tmp * 4) - 2) & VTX_REUSE_DEPTH_MASK); | |
805 | ||
806 | /* Setup some CP states */ | |
807 | WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) | ROQ_IB2_START(0x2b))); | |
808 | WREG32(CP_MEQ_THRESHOLDS, (MEQ_END(0x40) | ROQ_END(0x40))); | |
809 | ||
810 | WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO | SYNC_GRADIENT | | |
811 | SYNC_WALKER | SYNC_ALIGNER)); | |
812 | /* Setup various GPU states */ | |
813 | if (rdev->family == CHIP_RV670) | |
814 | WREG32(ARB_GDEC_RD_CNTL, 0x00000021); | |
815 | ||
816 | tmp = RREG32(SX_DEBUG_1); | |
817 | tmp |= SMX_EVENT_RELEASE; | |
818 | if ((rdev->family > CHIP_R600)) | |
819 | tmp |= ENABLE_NEW_SMX_ADDRESS; | |
820 | WREG32(SX_DEBUG_1, tmp); | |
821 | ||
822 | if (((rdev->family) == CHIP_R600) || | |
823 | ((rdev->family) == CHIP_RV630) || | |
824 | ((rdev->family) == CHIP_RV610) || | |
825 | ((rdev->family) == CHIP_RV620) || | |
826 | ((rdev->family) == CHIP_RS780)) { | |
827 | WREG32(DB_DEBUG, PREZ_MUST_WAIT_FOR_POSTZ_DONE); | |
828 | } else { | |
829 | WREG32(DB_DEBUG, 0); | |
830 | } | |
831 | WREG32(DB_WATERMARKS, (DEPTH_FREE(4) | DEPTH_CACHELINE_FREE(16) | | |
832 | DEPTH_FLUSH(16) | DEPTH_PENDING_FREE(4))); | |
833 | ||
834 | WREG32(PA_SC_MULTI_CHIP_CNTL, 0); | |
835 | WREG32(VGT_NUM_INSTANCES, 0); | |
836 | ||
837 | WREG32(SPI_CONFIG_CNTL, GPR_WRITE_PRIORITY(0)); | |
838 | WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(0)); | |
839 | ||
840 | tmp = RREG32(SQ_MS_FIFO_SIZES); | |
841 | if (((rdev->family) == CHIP_RV610) || | |
842 | ((rdev->family) == CHIP_RV620) || | |
843 | ((rdev->family) == CHIP_RS780)) { | |
844 | tmp = (CACHE_FIFO_SIZE(0xa) | | |
845 | FETCH_FIFO_HIWATER(0xa) | | |
846 | DONE_FIFO_HIWATER(0xe0) | | |
847 | ALU_UPDATE_FIFO_HIWATER(0x8)); | |
848 | } else if (((rdev->family) == CHIP_R600) || | |
849 | ((rdev->family) == CHIP_RV630)) { | |
850 | tmp &= ~DONE_FIFO_HIWATER(0xff); | |
851 | tmp |= DONE_FIFO_HIWATER(0x4); | |
852 | } | |
853 | WREG32(SQ_MS_FIFO_SIZES, tmp); | |
854 | ||
855 | /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT | |
856 | * should be adjusted as needed by the 2D/3D drivers. This just sets default values | |
857 | */ | |
858 | sq_config = RREG32(SQ_CONFIG); | |
859 | sq_config &= ~(PS_PRIO(3) | | |
860 | VS_PRIO(3) | | |
861 | GS_PRIO(3) | | |
862 | ES_PRIO(3)); | |
863 | sq_config |= (DX9_CONSTS | | |
864 | VC_ENABLE | | |
865 | PS_PRIO(0) | | |
866 | VS_PRIO(1) | | |
867 | GS_PRIO(2) | | |
868 | ES_PRIO(3)); | |
869 | ||
870 | if ((rdev->family) == CHIP_R600) { | |
871 | sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(124) | | |
872 | NUM_VS_GPRS(124) | | |
873 | NUM_CLAUSE_TEMP_GPRS(4)); | |
874 | sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(0) | | |
875 | NUM_ES_GPRS(0)); | |
876 | sq_thread_resource_mgmt = (NUM_PS_THREADS(136) | | |
877 | NUM_VS_THREADS(48) | | |
878 | NUM_GS_THREADS(4) | | |
879 | NUM_ES_THREADS(4)); | |
880 | sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(128) | | |
881 | NUM_VS_STACK_ENTRIES(128)); | |
882 | sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(0) | | |
883 | NUM_ES_STACK_ENTRIES(0)); | |
884 | } else if (((rdev->family) == CHIP_RV610) || | |
885 | ((rdev->family) == CHIP_RV620) || | |
886 | ((rdev->family) == CHIP_RS780)) { | |
887 | /* no vertex cache */ | |
888 | sq_config &= ~VC_ENABLE; | |
889 | ||
890 | sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) | | |
891 | NUM_VS_GPRS(44) | | |
892 | NUM_CLAUSE_TEMP_GPRS(2)); | |
893 | sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) | | |
894 | NUM_ES_GPRS(17)); | |
895 | sq_thread_resource_mgmt = (NUM_PS_THREADS(79) | | |
896 | NUM_VS_THREADS(78) | | |
897 | NUM_GS_THREADS(4) | | |
898 | NUM_ES_THREADS(31)); | |
899 | sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) | | |
900 | NUM_VS_STACK_ENTRIES(40)); | |
901 | sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) | | |
902 | NUM_ES_STACK_ENTRIES(16)); | |
903 | } else if (((rdev->family) == CHIP_RV630) || | |
904 | ((rdev->family) == CHIP_RV635)) { | |
905 | sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) | | |
906 | NUM_VS_GPRS(44) | | |
907 | NUM_CLAUSE_TEMP_GPRS(2)); | |
908 | sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(18) | | |
909 | NUM_ES_GPRS(18)); | |
910 | sq_thread_resource_mgmt = (NUM_PS_THREADS(79) | | |
911 | NUM_VS_THREADS(78) | | |
912 | NUM_GS_THREADS(4) | | |
913 | NUM_ES_THREADS(31)); | |
914 | sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) | | |
915 | NUM_VS_STACK_ENTRIES(40)); | |
916 | sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) | | |
917 | NUM_ES_STACK_ENTRIES(16)); | |
918 | } else if ((rdev->family) == CHIP_RV670) { | |
919 | sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) | | |
920 | NUM_VS_GPRS(44) | | |
921 | NUM_CLAUSE_TEMP_GPRS(2)); | |
922 | sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) | | |
923 | NUM_ES_GPRS(17)); | |
924 | sq_thread_resource_mgmt = (NUM_PS_THREADS(79) | | |
925 | NUM_VS_THREADS(78) | | |
926 | NUM_GS_THREADS(4) | | |
927 | NUM_ES_THREADS(31)); | |
928 | sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(64) | | |
929 | NUM_VS_STACK_ENTRIES(64)); | |
930 | sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(64) | | |
931 | NUM_ES_STACK_ENTRIES(64)); | |
932 | } | |
933 | ||
934 | WREG32(SQ_CONFIG, sq_config); | |
935 | WREG32(SQ_GPR_RESOURCE_MGMT_1, sq_gpr_resource_mgmt_1); | |
936 | WREG32(SQ_GPR_RESOURCE_MGMT_2, sq_gpr_resource_mgmt_2); | |
937 | WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt); | |
938 | WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1); | |
939 | WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2); | |
940 | ||
941 | if (((rdev->family) == CHIP_RV610) || | |
942 | ((rdev->family) == CHIP_RV620) || | |
943 | ((rdev->family) == CHIP_RS780)) { | |
944 | WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(TC_ONLY)); | |
945 | } else { | |
946 | WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC)); | |
947 | } | |
948 | ||
949 | /* More default values. 2D/3D driver should adjust as needed */ | |
950 | WREG32(PA_SC_AA_SAMPLE_LOCS_2S, (S0_X(0xc) | S0_Y(0x4) | | |
951 | S1_X(0x4) | S1_Y(0xc))); | |
952 | WREG32(PA_SC_AA_SAMPLE_LOCS_4S, (S0_X(0xe) | S0_Y(0xe) | | |
953 | S1_X(0x2) | S1_Y(0x2) | | |
954 | S2_X(0xa) | S2_Y(0x6) | | |
955 | S3_X(0x6) | S3_Y(0xa))); | |
956 | WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD0, (S0_X(0xe) | S0_Y(0xb) | | |
957 | S1_X(0x4) | S1_Y(0xc) | | |
958 | S2_X(0x1) | S2_Y(0x6) | | |
959 | S3_X(0xa) | S3_Y(0xe))); | |
960 | WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD1, (S4_X(0x6) | S4_Y(0x1) | | |
961 | S5_X(0x0) | S5_Y(0x0) | | |
962 | S6_X(0xb) | S6_Y(0x4) | | |
963 | S7_X(0x7) | S7_Y(0x8))); | |
964 | ||
965 | WREG32(VGT_STRMOUT_EN, 0); | |
966 | tmp = rdev->config.r600.max_pipes * 16; | |
967 | switch (rdev->family) { | |
968 | case CHIP_RV610: | |
969 | case CHIP_RS780: | |
970 | case CHIP_RV620: | |
971 | tmp += 32; | |
972 | break; | |
973 | case CHIP_RV670: | |
974 | tmp += 128; | |
975 | break; | |
976 | default: | |
977 | break; | |
978 | } | |
979 | if (tmp > 256) { | |
980 | tmp = 256; | |
981 | } | |
982 | WREG32(VGT_ES_PER_GS, 128); | |
983 | WREG32(VGT_GS_PER_ES, tmp); | |
984 | WREG32(VGT_GS_PER_VS, 2); | |
985 | WREG32(VGT_GS_VERTEX_REUSE, 16); | |
986 | ||
987 | /* more default values. 2D/3D driver should adjust as needed */ | |
988 | WREG32(PA_SC_LINE_STIPPLE_STATE, 0); | |
989 | WREG32(VGT_STRMOUT_EN, 0); | |
990 | WREG32(SX_MISC, 0); | |
991 | WREG32(PA_SC_MODE_CNTL, 0); | |
992 | WREG32(PA_SC_AA_CONFIG, 0); | |
993 | WREG32(PA_SC_LINE_STIPPLE, 0); | |
994 | WREG32(SPI_INPUT_Z, 0); | |
995 | WREG32(SPI_PS_IN_CONTROL_0, NUM_INTERP(2)); | |
996 | WREG32(CB_COLOR7_FRAG, 0); | |
997 | ||
998 | /* Clear render buffer base addresses */ | |
999 | WREG32(CB_COLOR0_BASE, 0); | |
1000 | WREG32(CB_COLOR1_BASE, 0); | |
1001 | WREG32(CB_COLOR2_BASE, 0); | |
1002 | WREG32(CB_COLOR3_BASE, 0); | |
1003 | WREG32(CB_COLOR4_BASE, 0); | |
1004 | WREG32(CB_COLOR5_BASE, 0); | |
1005 | WREG32(CB_COLOR6_BASE, 0); | |
1006 | WREG32(CB_COLOR7_BASE, 0); | |
1007 | WREG32(CB_COLOR7_FRAG, 0); | |
1008 | ||
1009 | switch (rdev->family) { | |
1010 | case CHIP_RV610: | |
1011 | case CHIP_RS780: | |
1012 | case CHIP_RV620: | |
1013 | tmp = TC_L2_SIZE(8); | |
1014 | break; | |
1015 | case CHIP_RV630: | |
1016 | case CHIP_RV635: | |
1017 | tmp = TC_L2_SIZE(4); | |
1018 | break; | |
1019 | case CHIP_R600: | |
1020 | tmp = TC_L2_SIZE(0) | L2_DISABLE_LATE_HIT; | |
1021 | break; | |
1022 | default: | |
1023 | tmp = TC_L2_SIZE(0); | |
1024 | break; | |
1025 | } | |
1026 | WREG32(TC_CNTL, tmp); | |
1027 | ||
1028 | tmp = RREG32(HDP_HOST_PATH_CNTL); | |
1029 | WREG32(HDP_HOST_PATH_CNTL, tmp); | |
1030 | ||
1031 | tmp = RREG32(ARB_POP); | |
1032 | tmp |= ENABLE_TC128; | |
1033 | WREG32(ARB_POP, tmp); | |
1034 | ||
1035 | WREG32(PA_SC_MULTI_CHIP_CNTL, 0); | |
1036 | WREG32(PA_CL_ENHANCE, (CLIP_VTX_REORDER_ENA | | |
1037 | NUM_CLIP_SEQ(3))); | |
1038 | WREG32(PA_SC_ENHANCE, FORCE_EOV_MAX_CLK_CNT(4095)); | |
1039 | } | |
1040 | ||
1041 | ||
771fe6b9 JG |
1042 | /* |
1043 | * Indirect registers accessor | |
1044 | */ | |
3ce0a23d JG |
1045 | u32 r600_pciep_rreg(struct radeon_device *rdev, u32 reg) |
1046 | { | |
1047 | u32 r; | |
1048 | ||
1049 | WREG32(PCIE_PORT_INDEX, ((reg) & 0xff)); | |
1050 | (void)RREG32(PCIE_PORT_INDEX); | |
1051 | r = RREG32(PCIE_PORT_DATA); | |
1052 | return r; | |
1053 | } | |
1054 | ||
1055 | void r600_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v) | |
1056 | { | |
1057 | WREG32(PCIE_PORT_INDEX, ((reg) & 0xff)); | |
1058 | (void)RREG32(PCIE_PORT_INDEX); | |
1059 | WREG32(PCIE_PORT_DATA, (v)); | |
1060 | (void)RREG32(PCIE_PORT_DATA); | |
1061 | } | |
1062 | ||
1063 | ||
1064 | /* | |
1065 | * CP & Ring | |
1066 | */ | |
1067 | void r600_cp_stop(struct radeon_device *rdev) | |
1068 | { | |
1069 | WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1)); | |
1070 | } | |
1071 | ||
1072 | int r600_cp_init_microcode(struct radeon_device *rdev) | |
1073 | { | |
1074 | struct platform_device *pdev; | |
1075 | const char *chip_name; | |
1076 | size_t pfp_req_size, me_req_size; | |
1077 | char fw_name[30]; | |
1078 | int err; | |
1079 | ||
1080 | DRM_DEBUG("\n"); | |
1081 | ||
1082 | pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0); | |
1083 | err = IS_ERR(pdev); | |
1084 | if (err) { | |
1085 | printk(KERN_ERR "radeon_cp: Failed to register firmware\n"); | |
1086 | return -EINVAL; | |
1087 | } | |
1088 | ||
1089 | switch (rdev->family) { | |
1090 | case CHIP_R600: chip_name = "R600"; break; | |
1091 | case CHIP_RV610: chip_name = "RV610"; break; | |
1092 | case CHIP_RV630: chip_name = "RV630"; break; | |
1093 | case CHIP_RV620: chip_name = "RV620"; break; | |
1094 | case CHIP_RV635: chip_name = "RV635"; break; | |
1095 | case CHIP_RV670: chip_name = "RV670"; break; | |
1096 | case CHIP_RS780: | |
1097 | case CHIP_RS880: chip_name = "RS780"; break; | |
1098 | case CHIP_RV770: chip_name = "RV770"; break; | |
1099 | case CHIP_RV730: | |
1100 | case CHIP_RV740: chip_name = "RV730"; break; | |
1101 | case CHIP_RV710: chip_name = "RV710"; break; | |
1102 | default: BUG(); | |
1103 | } | |
1104 | ||
1105 | if (rdev->family >= CHIP_RV770) { | |
1106 | pfp_req_size = R700_PFP_UCODE_SIZE * 4; | |
1107 | me_req_size = R700_PM4_UCODE_SIZE * 4; | |
1108 | } else { | |
1109 | pfp_req_size = PFP_UCODE_SIZE * 4; | |
1110 | me_req_size = PM4_UCODE_SIZE * 12; | |
1111 | } | |
1112 | ||
1113 | DRM_INFO("Loading %s CP Microcode\n", chip_name); | |
1114 | ||
1115 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name); | |
1116 | err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev); | |
1117 | if (err) | |
1118 | goto out; | |
1119 | if (rdev->pfp_fw->size != pfp_req_size) { | |
1120 | printk(KERN_ERR | |
1121 | "r600_cp: Bogus length %zu in firmware \"%s\"\n", | |
1122 | rdev->pfp_fw->size, fw_name); | |
1123 | err = -EINVAL; | |
1124 | goto out; | |
1125 | } | |
1126 | ||
1127 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name); | |
1128 | err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev); | |
1129 | if (err) | |
1130 | goto out; | |
1131 | if (rdev->me_fw->size != me_req_size) { | |
1132 | printk(KERN_ERR | |
1133 | "r600_cp: Bogus length %zu in firmware \"%s\"\n", | |
1134 | rdev->me_fw->size, fw_name); | |
1135 | err = -EINVAL; | |
1136 | } | |
1137 | out: | |
1138 | platform_device_unregister(pdev); | |
1139 | ||
1140 | if (err) { | |
1141 | if (err != -EINVAL) | |
1142 | printk(KERN_ERR | |
1143 | "r600_cp: Failed to load firmware \"%s\"\n", | |
1144 | fw_name); | |
1145 | release_firmware(rdev->pfp_fw); | |
1146 | rdev->pfp_fw = NULL; | |
1147 | release_firmware(rdev->me_fw); | |
1148 | rdev->me_fw = NULL; | |
1149 | } | |
1150 | return err; | |
1151 | } | |
1152 | ||
1153 | static int r600_cp_load_microcode(struct radeon_device *rdev) | |
1154 | { | |
1155 | const __be32 *fw_data; | |
1156 | int i; | |
1157 | ||
1158 | if (!rdev->me_fw || !rdev->pfp_fw) | |
1159 | return -EINVAL; | |
1160 | ||
1161 | r600_cp_stop(rdev); | |
1162 | ||
1163 | WREG32(CP_RB_CNTL, RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3)); | |
1164 | ||
1165 | /* Reset cp */ | |
1166 | WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP); | |
1167 | RREG32(GRBM_SOFT_RESET); | |
1168 | mdelay(15); | |
1169 | WREG32(GRBM_SOFT_RESET, 0); | |
1170 | ||
1171 | WREG32(CP_ME_RAM_WADDR, 0); | |
1172 | ||
1173 | fw_data = (const __be32 *)rdev->me_fw->data; | |
1174 | WREG32(CP_ME_RAM_WADDR, 0); | |
1175 | for (i = 0; i < PM4_UCODE_SIZE * 3; i++) | |
1176 | WREG32(CP_ME_RAM_DATA, | |
1177 | be32_to_cpup(fw_data++)); | |
1178 | ||
1179 | fw_data = (const __be32 *)rdev->pfp_fw->data; | |
1180 | WREG32(CP_PFP_UCODE_ADDR, 0); | |
1181 | for (i = 0; i < PFP_UCODE_SIZE; i++) | |
1182 | WREG32(CP_PFP_UCODE_DATA, | |
1183 | be32_to_cpup(fw_data++)); | |
1184 | ||
1185 | WREG32(CP_PFP_UCODE_ADDR, 0); | |
1186 | WREG32(CP_ME_RAM_WADDR, 0); | |
1187 | WREG32(CP_ME_RAM_RADDR, 0); | |
1188 | return 0; | |
1189 | } | |
1190 | ||
1191 | int r600_cp_start(struct radeon_device *rdev) | |
1192 | { | |
1193 | int r; | |
1194 | uint32_t cp_me; | |
1195 | ||
1196 | r = radeon_ring_lock(rdev, 7); | |
1197 | if (r) { | |
1198 | DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); | |
1199 | return r; | |
1200 | } | |
1201 | radeon_ring_write(rdev, PACKET3(PACKET3_ME_INITIALIZE, 5)); | |
1202 | radeon_ring_write(rdev, 0x1); | |
1203 | if (rdev->family < CHIP_RV770) { | |
1204 | radeon_ring_write(rdev, 0x3); | |
1205 | radeon_ring_write(rdev, rdev->config.r600.max_hw_contexts - 1); | |
1206 | } else { | |
1207 | radeon_ring_write(rdev, 0x0); | |
1208 | radeon_ring_write(rdev, rdev->config.rv770.max_hw_contexts - 1); | |
1209 | } | |
1210 | radeon_ring_write(rdev, PACKET3_ME_INITIALIZE_DEVICE_ID(1)); | |
1211 | radeon_ring_write(rdev, 0); | |
1212 | radeon_ring_write(rdev, 0); | |
1213 | radeon_ring_unlock_commit(rdev); | |
1214 | ||
1215 | cp_me = 0xff; | |
1216 | WREG32(R_0086D8_CP_ME_CNTL, cp_me); | |
1217 | return 0; | |
1218 | } | |
1219 | ||
1220 | int r600_cp_resume(struct radeon_device *rdev) | |
1221 | { | |
1222 | u32 tmp; | |
1223 | u32 rb_bufsz; | |
1224 | int r; | |
1225 | ||
1226 | /* Reset cp */ | |
1227 | WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP); | |
1228 | RREG32(GRBM_SOFT_RESET); | |
1229 | mdelay(15); | |
1230 | WREG32(GRBM_SOFT_RESET, 0); | |
1231 | ||
1232 | /* Set ring buffer size */ | |
1233 | rb_bufsz = drm_order(rdev->cp.ring_size / 8); | |
1234 | #ifdef __BIG_ENDIAN | |
1235 | WREG32(CP_RB_CNTL, BUF_SWAP_32BIT | RB_NO_UPDATE | | |
1236 | (drm_order(4096/8) << 8) | rb_bufsz); | |
1237 | #else | |
1238 | WREG32(CP_RB_CNTL, RB_NO_UPDATE | (drm_order(4096/8) << 8) | rb_bufsz); | |
1239 | #endif | |
1240 | WREG32(CP_SEM_WAIT_TIMER, 0x4); | |
1241 | ||
1242 | /* Set the write pointer delay */ | |
1243 | WREG32(CP_RB_WPTR_DELAY, 0); | |
1244 | ||
1245 | /* Initialize the ring buffer's read and write pointers */ | |
1246 | tmp = RREG32(CP_RB_CNTL); | |
1247 | WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA); | |
1248 | WREG32(CP_RB_RPTR_WR, 0); | |
1249 | WREG32(CP_RB_WPTR, 0); | |
1250 | WREG32(CP_RB_RPTR_ADDR, rdev->cp.gpu_addr & 0xFFFFFFFF); | |
1251 | WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->cp.gpu_addr)); | |
1252 | mdelay(1); | |
1253 | WREG32(CP_RB_CNTL, tmp); | |
1254 | ||
1255 | WREG32(CP_RB_BASE, rdev->cp.gpu_addr >> 8); | |
1256 | WREG32(CP_DEBUG, (1 << 27) | (1 << 28)); | |
1257 | ||
1258 | rdev->cp.rptr = RREG32(CP_RB_RPTR); | |
1259 | rdev->cp.wptr = RREG32(CP_RB_WPTR); | |
1260 | ||
1261 | r600_cp_start(rdev); | |
1262 | rdev->cp.ready = true; | |
1263 | r = radeon_ring_test(rdev); | |
1264 | if (r) { | |
1265 | rdev->cp.ready = false; | |
1266 | return r; | |
1267 | } | |
1268 | return 0; | |
1269 | } | |
1270 | ||
1271 | void r600_cp_commit(struct radeon_device *rdev) | |
1272 | { | |
1273 | WREG32(CP_RB_WPTR, rdev->cp.wptr); | |
1274 | (void)RREG32(CP_RB_WPTR); | |
1275 | } | |
1276 | ||
1277 | void r600_ring_init(struct radeon_device *rdev, unsigned ring_size) | |
1278 | { | |
1279 | u32 rb_bufsz; | |
1280 | ||
1281 | /* Align ring size */ | |
1282 | rb_bufsz = drm_order(ring_size / 8); | |
1283 | ring_size = (1 << (rb_bufsz + 1)) * 4; | |
1284 | rdev->cp.ring_size = ring_size; | |
1285 | rdev->cp.align_mask = 16 - 1; | |
1286 | } | |
1287 | ||
1288 | ||
1289 | /* | |
1290 | * GPU scratch registers helpers function. | |
1291 | */ | |
1292 | void r600_scratch_init(struct radeon_device *rdev) | |
1293 | { | |
1294 | int i; | |
1295 | ||
1296 | rdev->scratch.num_reg = 7; | |
1297 | for (i = 0; i < rdev->scratch.num_reg; i++) { | |
1298 | rdev->scratch.free[i] = true; | |
1299 | rdev->scratch.reg[i] = SCRATCH_REG0 + (i * 4); | |
1300 | } | |
1301 | } | |
1302 | ||
1303 | int r600_ring_test(struct radeon_device *rdev) | |
1304 | { | |
1305 | uint32_t scratch; | |
1306 | uint32_t tmp = 0; | |
1307 | unsigned i; | |
1308 | int r; | |
1309 | ||
1310 | r = radeon_scratch_get(rdev, &scratch); | |
1311 | if (r) { | |
1312 | DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r); | |
1313 | return r; | |
1314 | } | |
1315 | WREG32(scratch, 0xCAFEDEAD); | |
1316 | r = radeon_ring_lock(rdev, 3); | |
1317 | if (r) { | |
1318 | DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); | |
1319 | radeon_scratch_free(rdev, scratch); | |
1320 | return r; | |
1321 | } | |
1322 | radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1)); | |
1323 | radeon_ring_write(rdev, ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2)); | |
1324 | radeon_ring_write(rdev, 0xDEADBEEF); | |
1325 | radeon_ring_unlock_commit(rdev); | |
1326 | for (i = 0; i < rdev->usec_timeout; i++) { | |
1327 | tmp = RREG32(scratch); | |
1328 | if (tmp == 0xDEADBEEF) | |
1329 | break; | |
1330 | DRM_UDELAY(1); | |
1331 | } | |
1332 | if (i < rdev->usec_timeout) { | |
1333 | DRM_INFO("ring test succeeded in %d usecs\n", i); | |
1334 | } else { | |
1335 | DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n", | |
1336 | scratch, tmp); | |
1337 | r = -EINVAL; | |
1338 | } | |
1339 | radeon_scratch_free(rdev, scratch); | |
1340 | return r; | |
1341 | } | |
1342 | ||
1343 | /* | |
1344 | * Writeback | |
1345 | */ | |
1346 | int r600_wb_init(struct radeon_device *rdev) | |
1347 | { | |
1348 | int r; | |
1349 | ||
1350 | if (rdev->wb.wb_obj == NULL) { | |
1351 | r = radeon_object_create(rdev, NULL, 4096, | |
1352 | true, | |
1353 | RADEON_GEM_DOMAIN_GTT, | |
1354 | false, &rdev->wb.wb_obj); | |
1355 | if (r) { | |
1356 | DRM_ERROR("radeon: failed to create WB buffer (%d).\n", r); | |
1357 | return r; | |
1358 | } | |
1359 | r = radeon_object_pin(rdev->wb.wb_obj, | |
1360 | RADEON_GEM_DOMAIN_GTT, | |
1361 | &rdev->wb.gpu_addr); | |
1362 | if (r) { | |
1363 | DRM_ERROR("radeon: failed to pin WB buffer (%d).\n", r); | |
1364 | return r; | |
1365 | } | |
1366 | r = radeon_object_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb); | |
1367 | if (r) { | |
1368 | DRM_ERROR("radeon: failed to map WB buffer (%d).\n", r); | |
1369 | return r; | |
1370 | } | |
1371 | } | |
1372 | WREG32(SCRATCH_ADDR, (rdev->wb.gpu_addr >> 8) & 0xFFFFFFFF); | |
1373 | WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + 1024) & 0xFFFFFFFC); | |
1374 | WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + 1024) & 0xFF); | |
1375 | WREG32(SCRATCH_UMSK, 0xff); | |
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | void r600_wb_fini(struct radeon_device *rdev) | |
1380 | { | |
1381 | if (rdev->wb.wb_obj) { | |
1382 | radeon_object_kunmap(rdev->wb.wb_obj); | |
1383 | radeon_object_unpin(rdev->wb.wb_obj); | |
1384 | radeon_object_unref(&rdev->wb.wb_obj); | |
1385 | rdev->wb.wb = NULL; | |
1386 | rdev->wb.wb_obj = NULL; | |
1387 | } | |
1388 | } | |
1389 | ||
1390 | ||
1391 | /* | |
1392 | * CS | |
1393 | */ | |
1394 | void r600_fence_ring_emit(struct radeon_device *rdev, | |
1395 | struct radeon_fence *fence) | |
1396 | { | |
1397 | /* Emit fence sequence & fire IRQ */ | |
1398 | radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1)); | |
1399 | radeon_ring_write(rdev, ((rdev->fence_drv.scratch_reg - PACKET3_SET_CONFIG_REG_OFFSET) >> 2)); | |
1400 | radeon_ring_write(rdev, fence->seq); | |
1401 | } | |
1402 | ||
1403 | int r600_copy_dma(struct radeon_device *rdev, | |
1404 | uint64_t src_offset, | |
1405 | uint64_t dst_offset, | |
1406 | unsigned num_pages, | |
1407 | struct radeon_fence *fence) | |
1408 | { | |
1409 | /* FIXME: implement */ | |
1410 | return 0; | |
1411 | } | |
1412 | ||
1413 | int r600_copy_blit(struct radeon_device *rdev, | |
1414 | uint64_t src_offset, uint64_t dst_offset, | |
1415 | unsigned num_pages, struct radeon_fence *fence) | |
1416 | { | |
1417 | r600_blit_prepare_copy(rdev, num_pages * 4096); | |
1418 | r600_kms_blit_copy(rdev, src_offset, dst_offset, num_pages * 4096); | |
1419 | r600_blit_done_copy(rdev, fence); | |
1420 | return 0; | |
1421 | } | |
1422 | ||
1423 | int r600_irq_process(struct radeon_device *rdev) | |
1424 | { | |
1425 | /* FIXME: implement */ | |
1426 | return 0; | |
1427 | } | |
1428 | ||
1429 | int r600_irq_set(struct radeon_device *rdev) | |
1430 | { | |
1431 | /* FIXME: implement */ | |
1432 | return 0; | |
1433 | } | |
1434 | ||
1435 | int r600_set_surface_reg(struct radeon_device *rdev, int reg, | |
1436 | uint32_t tiling_flags, uint32_t pitch, | |
1437 | uint32_t offset, uint32_t obj_size) | |
1438 | { | |
1439 | /* FIXME: implement */ | |
1440 | return 0; | |
1441 | } | |
1442 | ||
1443 | void r600_clear_surface_reg(struct radeon_device *rdev, int reg) | |
1444 | { | |
1445 | /* FIXME: implement */ | |
1446 | } | |
1447 | ||
1448 | ||
1449 | bool r600_card_posted(struct radeon_device *rdev) | |
1450 | { | |
1451 | uint32_t reg; | |
1452 | ||
1453 | /* first check CRTCs */ | |
1454 | reg = RREG32(D1CRTC_CONTROL) | | |
1455 | RREG32(D2CRTC_CONTROL); | |
1456 | if (reg & CRTC_EN) | |
1457 | return true; | |
1458 | ||
1459 | /* then check MEM_SIZE, in case the crtcs are off */ | |
1460 | if (RREG32(CONFIG_MEMSIZE)) | |
1461 | return true; | |
1462 | ||
1463 | return false; | |
1464 | } | |
1465 | ||
1466 | int r600_resume(struct radeon_device *rdev) | |
1467 | { | |
1468 | int r; | |
1469 | ||
1470 | r600_gpu_reset(rdev); | |
1471 | r600_mc_resume(rdev); | |
1472 | r = r600_pcie_gart_enable(rdev); | |
1473 | if (r) | |
1474 | return r; | |
1475 | r600_gpu_init(rdev); | |
1476 | r = radeon_ring_init(rdev, rdev->cp.ring_size); | |
1477 | if (r) | |
1478 | return r; | |
1479 | r = r600_cp_load_microcode(rdev); | |
1480 | if (r) | |
1481 | return r; | |
1482 | r = r600_cp_resume(rdev); | |
1483 | if (r) | |
1484 | return r; | |
1485 | r = r600_wb_init(rdev); | |
1486 | if (r) | |
1487 | return r; | |
1488 | return 0; | |
1489 | } | |
1490 | ||
1491 | int r600_suspend(struct radeon_device *rdev) | |
1492 | { | |
1493 | /* FIXME: we should wait for ring to be empty */ | |
1494 | r600_cp_stop(rdev); | |
4aac0473 | 1495 | r600_pcie_gart_disable(rdev); |
3ce0a23d JG |
1496 | return 0; |
1497 | } | |
1498 | ||
1499 | /* Plan is to move initialization in that function and use | |
1500 | * helper function so that radeon_device_init pretty much | |
1501 | * do nothing more than calling asic specific function. This | |
1502 | * should also allow to remove a bunch of callback function | |
1503 | * like vram_info. | |
1504 | */ | |
1505 | int r600_init(struct radeon_device *rdev) | |
771fe6b9 | 1506 | { |
3ce0a23d | 1507 | int r; |
771fe6b9 | 1508 | |
3ce0a23d JG |
1509 | rdev->new_init_path = true; |
1510 | r = radeon_dummy_page_init(rdev); | |
1511 | if (r) | |
1512 | return r; | |
1513 | if (r600_debugfs_mc_info_init(rdev)) { | |
1514 | DRM_ERROR("Failed to register debugfs file for mc !\n"); | |
1515 | } | |
1516 | /* This don't do much */ | |
1517 | r = radeon_gem_init(rdev); | |
1518 | if (r) | |
1519 | return r; | |
1520 | /* Read BIOS */ | |
1521 | if (!radeon_get_bios(rdev)) { | |
1522 | if (ASIC_IS_AVIVO(rdev)) | |
1523 | return -EINVAL; | |
1524 | } | |
1525 | /* Must be an ATOMBIOS */ | |
1526 | if (!rdev->is_atom_bios) | |
1527 | return -EINVAL; | |
1528 | r = radeon_atombios_init(rdev); | |
1529 | if (r) | |
1530 | return r; | |
1531 | /* Post card if necessary */ | |
1532 | if (!r600_card_posted(rdev) && rdev->bios) { | |
1533 | DRM_INFO("GPU not posted. posting now...\n"); | |
1534 | atom_asic_init(rdev->mode_info.atom_context); | |
1535 | } | |
1536 | /* Initialize scratch registers */ | |
1537 | r600_scratch_init(rdev); | |
1538 | /* Initialize surface registers */ | |
1539 | radeon_surface_init(rdev); | |
1540 | r = radeon_clocks_init(rdev); | |
1541 | if (r) | |
1542 | return r; | |
1543 | /* Fence driver */ | |
1544 | r = radeon_fence_driver_init(rdev); | |
1545 | if (r) | |
1546 | return r; | |
1547 | r = r600_mc_init(rdev); | |
1548 | if (r) { | |
1549 | if (rdev->flags & RADEON_IS_AGP) { | |
1550 | /* Retry with disabling AGP */ | |
1551 | r600_fini(rdev); | |
1552 | rdev->flags &= ~RADEON_IS_AGP; | |
1553 | return r600_init(rdev); | |
1554 | } | |
1555 | return r; | |
1556 | } | |
1557 | /* Memory manager */ | |
1558 | r = radeon_object_init(rdev); | |
1559 | if (r) | |
1560 | return r; | |
1561 | rdev->cp.ring_obj = NULL; | |
1562 | r600_ring_init(rdev, 1024 * 1024); | |
1563 | ||
1564 | if (!rdev->me_fw || !rdev->pfp_fw) { | |
1565 | r = r600_cp_init_microcode(rdev); | |
1566 | if (r) { | |
1567 | DRM_ERROR("Failed to load firmware!\n"); | |
1568 | return r; | |
1569 | } | |
1570 | } | |
1571 | ||
4aac0473 JG |
1572 | r = r600_pcie_gart_init(rdev); |
1573 | if (r) | |
1574 | return r; | |
1575 | ||
733289c2 | 1576 | rdev->accel_working = true; |
3ce0a23d JG |
1577 | r = r600_resume(rdev); |
1578 | if (r) { | |
1579 | if (rdev->flags & RADEON_IS_AGP) { | |
1580 | /* Retry with disabling AGP */ | |
1581 | r600_fini(rdev); | |
1582 | rdev->flags &= ~RADEON_IS_AGP; | |
1583 | return r600_init(rdev); | |
1584 | } | |
733289c2 | 1585 | rdev->accel_working = false; |
3ce0a23d | 1586 | } |
733289c2 JG |
1587 | if (rdev->accel_working) { |
1588 | r = radeon_ib_pool_init(rdev); | |
1589 | if (r) { | |
1590 | DRM_ERROR("radeon: failled initializing IB pool (%d).\n", r); | |
1591 | rdev->accel_working = false; | |
1592 | } | |
1593 | r = r600_blit_init(rdev); | |
1594 | if (r) { | |
1595 | DRM_ERROR("radeon: failled blitter (%d).\n", r); | |
1596 | rdev->accel_working = false; | |
1597 | } | |
1598 | r = radeon_ib_test(rdev); | |
1599 | if (r) { | |
1600 | DRM_ERROR("radeon: failled testing IB (%d).\n", r); | |
1601 | rdev->accel_working = false; | |
1602 | } | |
3ce0a23d JG |
1603 | } |
1604 | return 0; | |
1605 | } | |
1606 | ||
1607 | void r600_fini(struct radeon_device *rdev) | |
1608 | { | |
1609 | /* Suspend operations */ | |
1610 | r600_suspend(rdev); | |
1611 | ||
1612 | r600_blit_fini(rdev); | |
1613 | radeon_ring_fini(rdev); | |
4aac0473 | 1614 | r600_pcie_gart_fini(rdev); |
3ce0a23d JG |
1615 | radeon_gem_fini(rdev); |
1616 | radeon_fence_driver_fini(rdev); | |
1617 | radeon_clocks_fini(rdev); | |
1618 | #if __OS_HAS_AGP | |
1619 | if (rdev->flags & RADEON_IS_AGP) | |
1620 | radeon_agp_fini(rdev); | |
1621 | #endif | |
1622 | radeon_object_fini(rdev); | |
1623 | if (rdev->is_atom_bios) | |
1624 | radeon_atombios_fini(rdev); | |
1625 | else | |
1626 | radeon_combios_fini(rdev); | |
1627 | kfree(rdev->bios); | |
1628 | rdev->bios = NULL; | |
1629 | radeon_dummy_page_fini(rdev); | |
1630 | } | |
1631 | ||
1632 | ||
1633 | /* | |
1634 | * CS stuff | |
1635 | */ | |
1636 | void r600_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib) | |
1637 | { | |
1638 | /* FIXME: implement */ | |
1639 | radeon_ring_write(rdev, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); | |
1640 | radeon_ring_write(rdev, ib->gpu_addr & 0xFFFFFFFC); | |
1641 | radeon_ring_write(rdev, upper_32_bits(ib->gpu_addr) & 0xFF); | |
1642 | radeon_ring_write(rdev, ib->length_dw); | |
1643 | } | |
1644 | ||
1645 | int r600_ib_test(struct radeon_device *rdev) | |
1646 | { | |
1647 | struct radeon_ib *ib; | |
1648 | uint32_t scratch; | |
1649 | uint32_t tmp = 0; | |
1650 | unsigned i; | |
1651 | int r; | |
1652 | ||
1653 | r = radeon_scratch_get(rdev, &scratch); | |
1654 | if (r) { | |
1655 | DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r); | |
1656 | return r; | |
1657 | } | |
1658 | WREG32(scratch, 0xCAFEDEAD); | |
1659 | r = radeon_ib_get(rdev, &ib); | |
1660 | if (r) { | |
1661 | DRM_ERROR("radeon: failed to get ib (%d).\n", r); | |
1662 | return r; | |
1663 | } | |
1664 | ib->ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1); | |
1665 | ib->ptr[1] = ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2); | |
1666 | ib->ptr[2] = 0xDEADBEEF; | |
1667 | ib->ptr[3] = PACKET2(0); | |
1668 | ib->ptr[4] = PACKET2(0); | |
1669 | ib->ptr[5] = PACKET2(0); | |
1670 | ib->ptr[6] = PACKET2(0); | |
1671 | ib->ptr[7] = PACKET2(0); | |
1672 | ib->ptr[8] = PACKET2(0); | |
1673 | ib->ptr[9] = PACKET2(0); | |
1674 | ib->ptr[10] = PACKET2(0); | |
1675 | ib->ptr[11] = PACKET2(0); | |
1676 | ib->ptr[12] = PACKET2(0); | |
1677 | ib->ptr[13] = PACKET2(0); | |
1678 | ib->ptr[14] = PACKET2(0); | |
1679 | ib->ptr[15] = PACKET2(0); | |
1680 | ib->length_dw = 16; | |
1681 | r = radeon_ib_schedule(rdev, ib); | |
1682 | if (r) { | |
1683 | radeon_scratch_free(rdev, scratch); | |
1684 | radeon_ib_free(rdev, &ib); | |
1685 | DRM_ERROR("radeon: failed to schedule ib (%d).\n", r); | |
1686 | return r; | |
1687 | } | |
1688 | r = radeon_fence_wait(ib->fence, false); | |
1689 | if (r) { | |
1690 | DRM_ERROR("radeon: fence wait failed (%d).\n", r); | |
1691 | return r; | |
1692 | } | |
1693 | for (i = 0; i < rdev->usec_timeout; i++) { | |
1694 | tmp = RREG32(scratch); | |
1695 | if (tmp == 0xDEADBEEF) | |
1696 | break; | |
1697 | DRM_UDELAY(1); | |
1698 | } | |
1699 | if (i < rdev->usec_timeout) { | |
1700 | DRM_INFO("ib test succeeded in %u usecs\n", i); | |
1701 | } else { | |
1702 | DRM_ERROR("radeon: ib test failed (sracth(0x%04X)=0x%08X)\n", | |
1703 | scratch, tmp); | |
1704 | r = -EINVAL; | |
1705 | } | |
1706 | radeon_scratch_free(rdev, scratch); | |
1707 | radeon_ib_free(rdev, &ib); | |
771fe6b9 JG |
1708 | return r; |
1709 | } | |
1710 | ||
3ce0a23d JG |
1711 | |
1712 | ||
1713 | ||
1714 | /* | |
1715 | * Debugfs info | |
1716 | */ | |
1717 | #if defined(CONFIG_DEBUG_FS) | |
1718 | ||
1719 | static int r600_debugfs_cp_ring_info(struct seq_file *m, void *data) | |
771fe6b9 | 1720 | { |
3ce0a23d JG |
1721 | struct drm_info_node *node = (struct drm_info_node *) m->private; |
1722 | struct drm_device *dev = node->minor->dev; | |
1723 | struct radeon_device *rdev = dev->dev_private; | |
1724 | uint32_t rdp, wdp; | |
1725 | unsigned count, i, j; | |
1726 | ||
1727 | radeon_ring_free_size(rdev); | |
1728 | rdp = RREG32(CP_RB_RPTR); | |
1729 | wdp = RREG32(CP_RB_WPTR); | |
1730 | count = (rdp + rdev->cp.ring_size - wdp) & rdev->cp.ptr_mask; | |
1731 | seq_printf(m, "CP_STAT 0x%08x\n", RREG32(CP_STAT)); | |
1732 | seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp); | |
1733 | seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp); | |
1734 | seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw); | |
1735 | seq_printf(m, "%u dwords in ring\n", count); | |
1736 | for (j = 0; j <= count; j++) { | |
1737 | i = (rdp + j) & rdev->cp.ptr_mask; | |
1738 | seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]); | |
1739 | } | |
1740 | return 0; | |
1741 | } | |
1742 | ||
1743 | static int r600_debugfs_mc_info(struct seq_file *m, void *data) | |
1744 | { | |
1745 | struct drm_info_node *node = (struct drm_info_node *) m->private; | |
1746 | struct drm_device *dev = node->minor->dev; | |
1747 | struct radeon_device *rdev = dev->dev_private; | |
1748 | ||
1749 | DREG32_SYS(m, rdev, R_000E50_SRBM_STATUS); | |
1750 | DREG32_SYS(m, rdev, VM_L2_STATUS); | |
1751 | return 0; | |
1752 | } | |
1753 | ||
1754 | static struct drm_info_list r600_mc_info_list[] = { | |
1755 | {"r600_mc_info", r600_debugfs_mc_info, 0, NULL}, | |
1756 | {"r600_ring_info", r600_debugfs_cp_ring_info, 0, NULL}, | |
1757 | }; | |
1758 | #endif | |
1759 | ||
1760 | int r600_debugfs_mc_info_init(struct radeon_device *rdev) | |
1761 | { | |
1762 | #if defined(CONFIG_DEBUG_FS) | |
1763 | return radeon_debugfs_add_files(rdev, r600_mc_info_list, ARRAY_SIZE(r600_mc_info_list)); | |
1764 | #else | |
1765 | return 0; | |
1766 | #endif | |
771fe6b9 | 1767 | } |