]>
Commit | Line | Data |
---|---|---|
2002c9c3 | 1 | /* Copyright (c) 2016-2017 The Linux Foundation. All rights reserved. |
b5f103ab JC |
2 | * |
3 | * This program is free software; you can redistribute it and/or modify | |
4 | * it under the terms of the GNU General Public License version 2 and | |
5 | * only version 2 as published by the Free Software Foundation. | |
6 | * | |
7 | * This program is distributed in the hope that it will be useful, | |
8 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
10 | * GNU General Public License for more details. | |
11 | * | |
12 | */ | |
13 | ||
14 | #include "msm_gem.h" | |
7f8036b7 | 15 | #include "msm_mmu.h" |
b5f103ab JC |
16 | #include "a5xx_gpu.h" |
17 | ||
18 | extern bool hang_debug; | |
19 | static void a5xx_dump(struct msm_gpu *gpu); | |
20 | ||
21 | static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, | |
22 | struct msm_file_private *ctx) | |
23 | { | |
24 | struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); | |
25 | struct msm_drm_private *priv = gpu->dev->dev_private; | |
26 | struct msm_ringbuffer *ring = gpu->rb; | |
27 | unsigned int i, ibs = 0; | |
28 | ||
29 | for (i = 0; i < submit->nr_cmds; i++) { | |
30 | switch (submit->cmd[i].type) { | |
31 | case MSM_SUBMIT_CMD_IB_TARGET_BUF: | |
32 | break; | |
33 | case MSM_SUBMIT_CMD_CTX_RESTORE_BUF: | |
34 | if (priv->lastctx == ctx) | |
35 | break; | |
36 | case MSM_SUBMIT_CMD_BUF: | |
37 | OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3); | |
38 | OUT_RING(ring, lower_32_bits(submit->cmd[i].iova)); | |
39 | OUT_RING(ring, upper_32_bits(submit->cmd[i].iova)); | |
40 | OUT_RING(ring, submit->cmd[i].size); | |
41 | ibs++; | |
42 | break; | |
43 | } | |
44 | } | |
45 | ||
46 | OUT_PKT4(ring, REG_A5XX_CP_SCRATCH_REG(2), 1); | |
47 | OUT_RING(ring, submit->fence->seqno); | |
48 | ||
49 | OUT_PKT7(ring, CP_EVENT_WRITE, 4); | |
50 | OUT_RING(ring, CACHE_FLUSH_TS | (1 << 31)); | |
51 | OUT_RING(ring, lower_32_bits(rbmemptr(adreno_gpu, fence))); | |
52 | OUT_RING(ring, upper_32_bits(rbmemptr(adreno_gpu, fence))); | |
53 | OUT_RING(ring, submit->fence->seqno); | |
54 | ||
55 | gpu->funcs->flush(gpu); | |
56 | } | |
57 | ||
58 | struct a5xx_hwcg { | |
59 | u32 offset; | |
60 | u32 value; | |
61 | }; | |
62 | ||
63 | static const struct a5xx_hwcg a530_hwcg[] = { | |
64 | {REG_A5XX_RBBM_CLOCK_CNTL_SP0, 0x02222222}, | |
65 | {REG_A5XX_RBBM_CLOCK_CNTL_SP1, 0x02222222}, | |
66 | {REG_A5XX_RBBM_CLOCK_CNTL_SP2, 0x02222222}, | |
67 | {REG_A5XX_RBBM_CLOCK_CNTL_SP3, 0x02222222}, | |
68 | {REG_A5XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220}, | |
69 | {REG_A5XX_RBBM_CLOCK_CNTL2_SP1, 0x02222220}, | |
70 | {REG_A5XX_RBBM_CLOCK_CNTL2_SP2, 0x02222220}, | |
71 | {REG_A5XX_RBBM_CLOCK_CNTL2_SP3, 0x02222220}, | |
72 | {REG_A5XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF}, | |
73 | {REG_A5XX_RBBM_CLOCK_HYST_SP1, 0x0000F3CF}, | |
74 | {REG_A5XX_RBBM_CLOCK_HYST_SP2, 0x0000F3CF}, | |
75 | {REG_A5XX_RBBM_CLOCK_HYST_SP3, 0x0000F3CF}, | |
76 | {REG_A5XX_RBBM_CLOCK_DELAY_SP0, 0x00000080}, | |
77 | {REG_A5XX_RBBM_CLOCK_DELAY_SP1, 0x00000080}, | |
78 | {REG_A5XX_RBBM_CLOCK_DELAY_SP2, 0x00000080}, | |
79 | {REG_A5XX_RBBM_CLOCK_DELAY_SP3, 0x00000080}, | |
80 | {REG_A5XX_RBBM_CLOCK_CNTL_TP0, 0x22222222}, | |
81 | {REG_A5XX_RBBM_CLOCK_CNTL_TP1, 0x22222222}, | |
82 | {REG_A5XX_RBBM_CLOCK_CNTL_TP2, 0x22222222}, | |
83 | {REG_A5XX_RBBM_CLOCK_CNTL_TP3, 0x22222222}, | |
84 | {REG_A5XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222}, | |
85 | {REG_A5XX_RBBM_CLOCK_CNTL2_TP1, 0x22222222}, | |
86 | {REG_A5XX_RBBM_CLOCK_CNTL2_TP2, 0x22222222}, | |
87 | {REG_A5XX_RBBM_CLOCK_CNTL2_TP3, 0x22222222}, | |
88 | {REG_A5XX_RBBM_CLOCK_CNTL3_TP0, 0x00002222}, | |
89 | {REG_A5XX_RBBM_CLOCK_CNTL3_TP1, 0x00002222}, | |
90 | {REG_A5XX_RBBM_CLOCK_CNTL3_TP2, 0x00002222}, | |
91 | {REG_A5XX_RBBM_CLOCK_CNTL3_TP3, 0x00002222}, | |
92 | {REG_A5XX_RBBM_CLOCK_HYST_TP0, 0x77777777}, | |
93 | {REG_A5XX_RBBM_CLOCK_HYST_TP1, 0x77777777}, | |
94 | {REG_A5XX_RBBM_CLOCK_HYST_TP2, 0x77777777}, | |
95 | {REG_A5XX_RBBM_CLOCK_HYST_TP3, 0x77777777}, | |
96 | {REG_A5XX_RBBM_CLOCK_HYST2_TP0, 0x77777777}, | |
97 | {REG_A5XX_RBBM_CLOCK_HYST2_TP1, 0x77777777}, | |
98 | {REG_A5XX_RBBM_CLOCK_HYST2_TP2, 0x77777777}, | |
99 | {REG_A5XX_RBBM_CLOCK_HYST2_TP3, 0x77777777}, | |
100 | {REG_A5XX_RBBM_CLOCK_HYST3_TP0, 0x00007777}, | |
101 | {REG_A5XX_RBBM_CLOCK_HYST3_TP1, 0x00007777}, | |
102 | {REG_A5XX_RBBM_CLOCK_HYST3_TP2, 0x00007777}, | |
103 | {REG_A5XX_RBBM_CLOCK_HYST3_TP3, 0x00007777}, | |
104 | {REG_A5XX_RBBM_CLOCK_DELAY_TP0, 0x11111111}, | |
105 | {REG_A5XX_RBBM_CLOCK_DELAY_TP1, 0x11111111}, | |
106 | {REG_A5XX_RBBM_CLOCK_DELAY_TP2, 0x11111111}, | |
107 | {REG_A5XX_RBBM_CLOCK_DELAY_TP3, 0x11111111}, | |
108 | {REG_A5XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111}, | |
109 | {REG_A5XX_RBBM_CLOCK_DELAY2_TP1, 0x11111111}, | |
110 | {REG_A5XX_RBBM_CLOCK_DELAY2_TP2, 0x11111111}, | |
111 | {REG_A5XX_RBBM_CLOCK_DELAY2_TP3, 0x11111111}, | |
112 | {REG_A5XX_RBBM_CLOCK_DELAY3_TP0, 0x00001111}, | |
113 | {REG_A5XX_RBBM_CLOCK_DELAY3_TP1, 0x00001111}, | |
114 | {REG_A5XX_RBBM_CLOCK_DELAY3_TP2, 0x00001111}, | |
115 | {REG_A5XX_RBBM_CLOCK_DELAY3_TP3, 0x00001111}, | |
116 | {REG_A5XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222}, | |
117 | {REG_A5XX_RBBM_CLOCK_CNTL2_UCHE, 0x22222222}, | |
118 | {REG_A5XX_RBBM_CLOCK_CNTL3_UCHE, 0x22222222}, | |
119 | {REG_A5XX_RBBM_CLOCK_CNTL4_UCHE, 0x00222222}, | |
120 | {REG_A5XX_RBBM_CLOCK_HYST_UCHE, 0x00444444}, | |
121 | {REG_A5XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002}, | |
122 | {REG_A5XX_RBBM_CLOCK_CNTL_RB0, 0x22222222}, | |
123 | {REG_A5XX_RBBM_CLOCK_CNTL_RB1, 0x22222222}, | |
124 | {REG_A5XX_RBBM_CLOCK_CNTL_RB2, 0x22222222}, | |
125 | {REG_A5XX_RBBM_CLOCK_CNTL_RB3, 0x22222222}, | |
126 | {REG_A5XX_RBBM_CLOCK_CNTL2_RB0, 0x00222222}, | |
127 | {REG_A5XX_RBBM_CLOCK_CNTL2_RB1, 0x00222222}, | |
128 | {REG_A5XX_RBBM_CLOCK_CNTL2_RB2, 0x00222222}, | |
129 | {REG_A5XX_RBBM_CLOCK_CNTL2_RB3, 0x00222222}, | |
130 | {REG_A5XX_RBBM_CLOCK_CNTL_CCU0, 0x00022220}, | |
131 | {REG_A5XX_RBBM_CLOCK_CNTL_CCU1, 0x00022220}, | |
132 | {REG_A5XX_RBBM_CLOCK_CNTL_CCU2, 0x00022220}, | |
133 | {REG_A5XX_RBBM_CLOCK_CNTL_CCU3, 0x00022220}, | |
134 | {REG_A5XX_RBBM_CLOCK_CNTL_RAC, 0x05522222}, | |
135 | {REG_A5XX_RBBM_CLOCK_CNTL2_RAC, 0x00505555}, | |
136 | {REG_A5XX_RBBM_CLOCK_HYST_RB_CCU0, 0x04040404}, | |
137 | {REG_A5XX_RBBM_CLOCK_HYST_RB_CCU1, 0x04040404}, | |
138 | {REG_A5XX_RBBM_CLOCK_HYST_RB_CCU2, 0x04040404}, | |
139 | {REG_A5XX_RBBM_CLOCK_HYST_RB_CCU3, 0x04040404}, | |
140 | {REG_A5XX_RBBM_CLOCK_HYST_RAC, 0x07444044}, | |
141 | {REG_A5XX_RBBM_CLOCK_DELAY_RB_CCU_L1_0, 0x00000002}, | |
142 | {REG_A5XX_RBBM_CLOCK_DELAY_RB_CCU_L1_1, 0x00000002}, | |
143 | {REG_A5XX_RBBM_CLOCK_DELAY_RB_CCU_L1_2, 0x00000002}, | |
144 | {REG_A5XX_RBBM_CLOCK_DELAY_RB_CCU_L1_3, 0x00000002}, | |
145 | {REG_A5XX_RBBM_CLOCK_DELAY_RAC, 0x00010011}, | |
146 | {REG_A5XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222}, | |
147 | {REG_A5XX_RBBM_CLOCK_MODE_GPC, 0x02222222}, | |
148 | {REG_A5XX_RBBM_CLOCK_MODE_VFD, 0x00002222}, | |
149 | {REG_A5XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000}, | |
150 | {REG_A5XX_RBBM_CLOCK_HYST_GPC, 0x04104004}, | |
151 | {REG_A5XX_RBBM_CLOCK_HYST_VFD, 0x00000000}, | |
152 | {REG_A5XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000}, | |
153 | {REG_A5XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000}, | |
154 | {REG_A5XX_RBBM_CLOCK_DELAY_GPC, 0x00000200}, | |
155 | {REG_A5XX_RBBM_CLOCK_DELAY_VFD, 0x00002222} | |
156 | }; | |
157 | ||
158 | static const struct { | |
159 | int (*test)(struct adreno_gpu *gpu); | |
160 | const struct a5xx_hwcg *regs; | |
161 | unsigned int count; | |
162 | } a5xx_hwcg_regs[] = { | |
163 | { adreno_is_a530, a530_hwcg, ARRAY_SIZE(a530_hwcg), }, | |
164 | }; | |
165 | ||
166 | static void _a5xx_enable_hwcg(struct msm_gpu *gpu, | |
167 | const struct a5xx_hwcg *regs, unsigned int count) | |
168 | { | |
169 | unsigned int i; | |
170 | ||
171 | for (i = 0; i < count; i++) | |
172 | gpu_write(gpu, regs[i].offset, regs[i].value); | |
173 | ||
174 | gpu_write(gpu, REG_A5XX_RBBM_CLOCK_CNTL, 0xAAA8AA00); | |
175 | gpu_write(gpu, REG_A5XX_RBBM_ISDB_CNT, 0x182); | |
176 | } | |
177 | ||
178 | static void a5xx_enable_hwcg(struct msm_gpu *gpu) | |
179 | { | |
180 | struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); | |
181 | unsigned int i; | |
182 | ||
183 | for (i = 0; i < ARRAY_SIZE(a5xx_hwcg_regs); i++) { | |
184 | if (a5xx_hwcg_regs[i].test(adreno_gpu)) { | |
185 | _a5xx_enable_hwcg(gpu, a5xx_hwcg_regs[i].regs, | |
186 | a5xx_hwcg_regs[i].count); | |
187 | return; | |
188 | } | |
189 | } | |
190 | } | |
191 | ||
192 | static int a5xx_me_init(struct msm_gpu *gpu) | |
193 | { | |
194 | struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); | |
195 | struct msm_ringbuffer *ring = gpu->rb; | |
196 | ||
197 | OUT_PKT7(ring, CP_ME_INIT, 8); | |
198 | ||
199 | OUT_RING(ring, 0x0000002F); | |
200 | ||
201 | /* Enable multiple hardware contexts */ | |
202 | OUT_RING(ring, 0x00000003); | |
203 | ||
204 | /* Enable error detection */ | |
205 | OUT_RING(ring, 0x20000000); | |
206 | ||
207 | /* Don't enable header dump */ | |
208 | OUT_RING(ring, 0x00000000); | |
209 | OUT_RING(ring, 0x00000000); | |
210 | ||
211 | /* Specify workarounds for various microcode issues */ | |
212 | if (adreno_is_a530(adreno_gpu)) { | |
213 | /* Workaround for token end syncs | |
214 | * Force a WFI after every direct-render 3D mode draw and every | |
215 | * 2D mode 3 draw | |
216 | */ | |
217 | OUT_RING(ring, 0x0000000B); | |
218 | } else { | |
219 | /* No workarounds enabled */ | |
220 | OUT_RING(ring, 0x00000000); | |
221 | } | |
222 | ||
223 | OUT_RING(ring, 0x00000000); | |
224 | OUT_RING(ring, 0x00000000); | |
225 | ||
226 | gpu->funcs->flush(gpu); | |
227 | ||
228 | return gpu->funcs->idle(gpu) ? 0 : -EINVAL; | |
229 | } | |
230 | ||
231 | static struct drm_gem_object *a5xx_ucode_load_bo(struct msm_gpu *gpu, | |
232 | const struct firmware *fw, u64 *iova) | |
233 | { | |
234 | struct drm_device *drm = gpu->dev; | |
235 | struct drm_gem_object *bo; | |
236 | void *ptr; | |
237 | ||
238 | mutex_lock(&drm->struct_mutex); | |
239 | bo = msm_gem_new(drm, fw->size - 4, MSM_BO_UNCACHED); | |
240 | mutex_unlock(&drm->struct_mutex); | |
241 | ||
242 | if (IS_ERR(bo)) | |
243 | return bo; | |
244 | ||
245 | ptr = msm_gem_get_vaddr(bo); | |
246 | if (!ptr) { | |
247 | drm_gem_object_unreference_unlocked(bo); | |
248 | return ERR_PTR(-ENOMEM); | |
249 | } | |
250 | ||
251 | if (iova) { | |
252 | int ret = msm_gem_get_iova(bo, gpu->id, iova); | |
253 | ||
254 | if (ret) { | |
255 | drm_gem_object_unreference_unlocked(bo); | |
256 | return ERR_PTR(ret); | |
257 | } | |
258 | } | |
259 | ||
260 | memcpy(ptr, &fw->data[4], fw->size - 4); | |
261 | ||
262 | msm_gem_put_vaddr(bo); | |
263 | return bo; | |
264 | } | |
265 | ||
266 | static int a5xx_ucode_init(struct msm_gpu *gpu) | |
267 | { | |
268 | struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); | |
269 | struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); | |
270 | int ret; | |
271 | ||
272 | if (!a5xx_gpu->pm4_bo) { | |
273 | a5xx_gpu->pm4_bo = a5xx_ucode_load_bo(gpu, adreno_gpu->pm4, | |
274 | &a5xx_gpu->pm4_iova); | |
275 | ||
276 | if (IS_ERR(a5xx_gpu->pm4_bo)) { | |
277 | ret = PTR_ERR(a5xx_gpu->pm4_bo); | |
278 | a5xx_gpu->pm4_bo = NULL; | |
279 | dev_err(gpu->dev->dev, "could not allocate PM4: %d\n", | |
280 | ret); | |
281 | return ret; | |
282 | } | |
283 | } | |
284 | ||
285 | if (!a5xx_gpu->pfp_bo) { | |
286 | a5xx_gpu->pfp_bo = a5xx_ucode_load_bo(gpu, adreno_gpu->pfp, | |
287 | &a5xx_gpu->pfp_iova); | |
288 | ||
289 | if (IS_ERR(a5xx_gpu->pfp_bo)) { | |
290 | ret = PTR_ERR(a5xx_gpu->pfp_bo); | |
291 | a5xx_gpu->pfp_bo = NULL; | |
292 | dev_err(gpu->dev->dev, "could not allocate PFP: %d\n", | |
293 | ret); | |
294 | return ret; | |
295 | } | |
296 | } | |
297 | ||
298 | gpu_write64(gpu, REG_A5XX_CP_ME_INSTR_BASE_LO, | |
299 | REG_A5XX_CP_ME_INSTR_BASE_HI, a5xx_gpu->pm4_iova); | |
300 | ||
301 | gpu_write64(gpu, REG_A5XX_CP_PFP_INSTR_BASE_LO, | |
302 | REG_A5XX_CP_PFP_INSTR_BASE_HI, a5xx_gpu->pfp_iova); | |
303 | ||
304 | return 0; | |
305 | } | |
306 | ||
307 | #define A5XX_INT_MASK (A5XX_RBBM_INT_0_MASK_RBBM_AHB_ERROR | \ | |
308 | A5XX_RBBM_INT_0_MASK_RBBM_TRANSFER_TIMEOUT | \ | |
309 | A5XX_RBBM_INT_0_MASK_RBBM_ME_MS_TIMEOUT | \ | |
310 | A5XX_RBBM_INT_0_MASK_RBBM_PFP_MS_TIMEOUT | \ | |
311 | A5XX_RBBM_INT_0_MASK_RBBM_ETS_MS_TIMEOUT | \ | |
312 | A5XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNC_OVERFLOW | \ | |
313 | A5XX_RBBM_INT_0_MASK_CP_HW_ERROR | \ | |
314 | A5XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \ | |
315 | A5XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \ | |
316 | A5XX_RBBM_INT_0_MASK_GPMU_VOLTAGE_DROOP) | |
317 | ||
318 | static int a5xx_hw_init(struct msm_gpu *gpu) | |
319 | { | |
320 | struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); | |
321 | int ret; | |
322 | ||
323 | gpu_write(gpu, REG_A5XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x00000003); | |
324 | ||
325 | /* Make all blocks contribute to the GPU BUSY perf counter */ | |
326 | gpu_write(gpu, REG_A5XX_RBBM_PERFCTR_GPU_BUSY_MASKED, 0xFFFFFFFF); | |
327 | ||
328 | /* Enable RBBM error reporting bits */ | |
329 | gpu_write(gpu, REG_A5XX_RBBM_AHB_CNTL0, 0x00000001); | |
330 | ||
4e09b95d | 331 | if (adreno_gpu->info->quirks & ADRENO_QUIRK_FAULT_DETECT_MASK) { |
b5f103ab JC |
332 | /* |
333 | * Mask out the activity signals from RB1-3 to avoid false | |
334 | * positives | |
335 | */ | |
336 | ||
337 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_MASK_CNTL11, | |
338 | 0xF0000000); | |
339 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_MASK_CNTL12, | |
340 | 0xFFFFFFFF); | |
341 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_MASK_CNTL13, | |
342 | 0xFFFFFFFF); | |
343 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_MASK_CNTL14, | |
344 | 0xFFFFFFFF); | |
345 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_MASK_CNTL15, | |
346 | 0xFFFFFFFF); | |
347 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_MASK_CNTL16, | |
348 | 0xFFFFFFFF); | |
349 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_MASK_CNTL17, | |
350 | 0xFFFFFFFF); | |
351 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_MASK_CNTL18, | |
352 | 0xFFFFFFFF); | |
353 | } | |
354 | ||
355 | /* Enable fault detection */ | |
356 | gpu_write(gpu, REG_A5XX_RBBM_INTERFACE_HANG_INT_CNTL, | |
357 | (1 << 30) | 0xFFFF); | |
358 | ||
359 | /* Turn on performance counters */ | |
360 | gpu_write(gpu, REG_A5XX_RBBM_PERFCTR_CNTL, 0x01); | |
361 | ||
362 | /* Increase VFD cache access so LRZ and other data gets evicted less */ | |
363 | gpu_write(gpu, REG_A5XX_UCHE_CACHE_WAYS, 0x02); | |
364 | ||
365 | /* Disable L2 bypass in the UCHE */ | |
366 | gpu_write(gpu, REG_A5XX_UCHE_TRAP_BASE_LO, 0xFFFF0000); | |
367 | gpu_write(gpu, REG_A5XX_UCHE_TRAP_BASE_HI, 0x0001FFFF); | |
368 | gpu_write(gpu, REG_A5XX_UCHE_WRITE_THRU_BASE_LO, 0xFFFF0000); | |
369 | gpu_write(gpu, REG_A5XX_UCHE_WRITE_THRU_BASE_HI, 0x0001FFFF); | |
370 | ||
371 | /* Set the GMEM VA range (0 to gpu->gmem) */ | |
372 | gpu_write(gpu, REG_A5XX_UCHE_GMEM_RANGE_MIN_LO, 0x00100000); | |
373 | gpu_write(gpu, REG_A5XX_UCHE_GMEM_RANGE_MIN_HI, 0x00000000); | |
374 | gpu_write(gpu, REG_A5XX_UCHE_GMEM_RANGE_MAX_LO, | |
375 | 0x00100000 + adreno_gpu->gmem - 1); | |
376 | gpu_write(gpu, REG_A5XX_UCHE_GMEM_RANGE_MAX_HI, 0x00000000); | |
377 | ||
378 | gpu_write(gpu, REG_A5XX_CP_MEQ_THRESHOLDS, 0x40); | |
379 | gpu_write(gpu, REG_A5XX_CP_MERCIU_SIZE, 0x40); | |
380 | gpu_write(gpu, REG_A5XX_CP_ROQ_THRESHOLDS_2, 0x80000060); | |
381 | gpu_write(gpu, REG_A5XX_CP_ROQ_THRESHOLDS_1, 0x40201B16); | |
382 | ||
383 | gpu_write(gpu, REG_A5XX_PC_DBG_ECO_CNTL, (0x400 << 11 | 0x300 << 22)); | |
384 | ||
4e09b95d | 385 | if (adreno_gpu->info->quirks & ADRENO_QUIRK_TWO_PASS_USE_WFI) |
b5f103ab JC |
386 | gpu_rmw(gpu, REG_A5XX_PC_DBG_ECO_CNTL, 0, (1 << 8)); |
387 | ||
388 | gpu_write(gpu, REG_A5XX_PC_DBG_ECO_CNTL, 0xc0200100); | |
389 | ||
390 | /* Enable USE_RETENTION_FLOPS */ | |
391 | gpu_write(gpu, REG_A5XX_CP_CHICKEN_DBG, 0x02000000); | |
392 | ||
393 | /* Enable ME/PFP split notification */ | |
394 | gpu_write(gpu, REG_A5XX_RBBM_AHB_CNTL1, 0xA6FFFFFF); | |
395 | ||
396 | /* Enable HWCG */ | |
397 | a5xx_enable_hwcg(gpu); | |
398 | ||
399 | gpu_write(gpu, REG_A5XX_RBBM_AHB_CNTL2, 0x0000003F); | |
400 | ||
401 | /* Set the highest bank bit */ | |
402 | gpu_write(gpu, REG_A5XX_TPL1_MODE_CNTL, 2 << 7); | |
403 | gpu_write(gpu, REG_A5XX_RB_MODE_CNTL, 2 << 1); | |
404 | ||
405 | /* Protect registers from the CP */ | |
406 | gpu_write(gpu, REG_A5XX_CP_PROTECT_CNTL, 0x00000007); | |
407 | ||
408 | /* RBBM */ | |
409 | gpu_write(gpu, REG_A5XX_CP_PROTECT(0), ADRENO_PROTECT_RW(0x04, 4)); | |
410 | gpu_write(gpu, REG_A5XX_CP_PROTECT(1), ADRENO_PROTECT_RW(0x08, 8)); | |
411 | gpu_write(gpu, REG_A5XX_CP_PROTECT(2), ADRENO_PROTECT_RW(0x10, 16)); | |
412 | gpu_write(gpu, REG_A5XX_CP_PROTECT(3), ADRENO_PROTECT_RW(0x20, 32)); | |
413 | gpu_write(gpu, REG_A5XX_CP_PROTECT(4), ADRENO_PROTECT_RW(0x40, 64)); | |
414 | gpu_write(gpu, REG_A5XX_CP_PROTECT(5), ADRENO_PROTECT_RW(0x80, 64)); | |
415 | ||
416 | /* Content protect */ | |
417 | gpu_write(gpu, REG_A5XX_CP_PROTECT(6), | |
418 | ADRENO_PROTECT_RW(REG_A5XX_RBBM_SECVID_TSB_TRUSTED_BASE_LO, | |
419 | 16)); | |
420 | gpu_write(gpu, REG_A5XX_CP_PROTECT(7), | |
421 | ADRENO_PROTECT_RW(REG_A5XX_RBBM_SECVID_TRUST_CNTL, 2)); | |
422 | ||
423 | /* CP */ | |
424 | gpu_write(gpu, REG_A5XX_CP_PROTECT(8), ADRENO_PROTECT_RW(0x800, 64)); | |
425 | gpu_write(gpu, REG_A5XX_CP_PROTECT(9), ADRENO_PROTECT_RW(0x840, 8)); | |
426 | gpu_write(gpu, REG_A5XX_CP_PROTECT(10), ADRENO_PROTECT_RW(0x880, 32)); | |
427 | gpu_write(gpu, REG_A5XX_CP_PROTECT(11), ADRENO_PROTECT_RW(0xAA0, 1)); | |
428 | ||
429 | /* RB */ | |
430 | gpu_write(gpu, REG_A5XX_CP_PROTECT(12), ADRENO_PROTECT_RW(0xCC0, 1)); | |
431 | gpu_write(gpu, REG_A5XX_CP_PROTECT(13), ADRENO_PROTECT_RW(0xCF0, 2)); | |
432 | ||
433 | /* VPC */ | |
434 | gpu_write(gpu, REG_A5XX_CP_PROTECT(14), ADRENO_PROTECT_RW(0xE68, 8)); | |
435 | gpu_write(gpu, REG_A5XX_CP_PROTECT(15), ADRENO_PROTECT_RW(0xE70, 4)); | |
436 | ||
437 | /* UCHE */ | |
438 | gpu_write(gpu, REG_A5XX_CP_PROTECT(16), ADRENO_PROTECT_RW(0xE80, 16)); | |
439 | ||
440 | if (adreno_is_a530(adreno_gpu)) | |
441 | gpu_write(gpu, REG_A5XX_CP_PROTECT(17), | |
442 | ADRENO_PROTECT_RW(0x10000, 0x8000)); | |
443 | ||
444 | gpu_write(gpu, REG_A5XX_RBBM_SECVID_TSB_CNTL, 0); | |
445 | /* | |
446 | * Disable the trusted memory range - we don't actually supported secure | |
447 | * memory rendering at this point in time and we don't want to block off | |
448 | * part of the virtual memory space. | |
449 | */ | |
450 | gpu_write64(gpu, REG_A5XX_RBBM_SECVID_TSB_TRUSTED_BASE_LO, | |
451 | REG_A5XX_RBBM_SECVID_TSB_TRUSTED_BASE_HI, 0x00000000); | |
452 | gpu_write(gpu, REG_A5XX_RBBM_SECVID_TSB_TRUSTED_SIZE, 0x00000000); | |
453 | ||
2401a008 JC |
454 | /* Load the GPMU firmware before starting the HW init */ |
455 | a5xx_gpmu_ucode_init(gpu); | |
456 | ||
b5f103ab JC |
457 | ret = adreno_hw_init(gpu); |
458 | if (ret) | |
459 | return ret; | |
460 | ||
461 | ret = a5xx_ucode_init(gpu); | |
462 | if (ret) | |
463 | return ret; | |
464 | ||
465 | /* Disable the interrupts through the initial bringup stage */ | |
466 | gpu_write(gpu, REG_A5XX_RBBM_INT_0_MASK, A5XX_INT_MASK); | |
467 | ||
468 | /* Clear ME_HALT to start the micro engine */ | |
469 | gpu_write(gpu, REG_A5XX_CP_PFP_ME_CNTL, 0); | |
470 | ret = a5xx_me_init(gpu); | |
471 | if (ret) | |
472 | return ret; | |
473 | ||
2401a008 JC |
474 | ret = a5xx_power_init(gpu); |
475 | if (ret) | |
476 | return ret; | |
b5f103ab JC |
477 | |
478 | /* | |
479 | * Send a pipeline event stat to get misbehaving counters to start | |
480 | * ticking correctly | |
481 | */ | |
482 | if (adreno_is_a530(adreno_gpu)) { | |
483 | OUT_PKT7(gpu->rb, CP_EVENT_WRITE, 1); | |
484 | OUT_RING(gpu->rb, 0x0F); | |
485 | ||
486 | gpu->funcs->flush(gpu); | |
487 | if (!gpu->funcs->idle(gpu)) | |
488 | return -EINVAL; | |
489 | } | |
490 | ||
2401a008 JC |
491 | /* Put the GPU into unsecure mode */ |
492 | gpu_write(gpu, REG_A5XX_RBBM_SECVID_TRUST_CNTL, 0x0); | |
493 | ||
b5f103ab JC |
494 | return 0; |
495 | } | |
496 | ||
497 | static void a5xx_recover(struct msm_gpu *gpu) | |
498 | { | |
499 | int i; | |
500 | ||
501 | adreno_dump_info(gpu); | |
502 | ||
503 | for (i = 0; i < 8; i++) { | |
504 | printk("CP_SCRATCH_REG%d: %u\n", i, | |
505 | gpu_read(gpu, REG_A5XX_CP_SCRATCH_REG(i))); | |
506 | } | |
507 | ||
508 | if (hang_debug) | |
509 | a5xx_dump(gpu); | |
510 | ||
511 | gpu_write(gpu, REG_A5XX_RBBM_SW_RESET_CMD, 1); | |
512 | gpu_read(gpu, REG_A5XX_RBBM_SW_RESET_CMD); | |
513 | gpu_write(gpu, REG_A5XX_RBBM_SW_RESET_CMD, 0); | |
514 | adreno_recover(gpu); | |
515 | } | |
516 | ||
517 | static void a5xx_destroy(struct msm_gpu *gpu) | |
518 | { | |
519 | struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); | |
520 | struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu); | |
521 | ||
522 | DBG("%s", gpu->name); | |
523 | ||
524 | if (a5xx_gpu->pm4_bo) { | |
525 | if (a5xx_gpu->pm4_iova) | |
526 | msm_gem_put_iova(a5xx_gpu->pm4_bo, gpu->id); | |
527 | drm_gem_object_unreference_unlocked(a5xx_gpu->pm4_bo); | |
528 | } | |
529 | ||
530 | if (a5xx_gpu->pfp_bo) { | |
531 | if (a5xx_gpu->pfp_iova) | |
532 | msm_gem_put_iova(a5xx_gpu->pfp_bo, gpu->id); | |
533 | drm_gem_object_unreference_unlocked(a5xx_gpu->pfp_bo); | |
534 | } | |
535 | ||
2401a008 | 536 | if (a5xx_gpu->gpmu_bo) { |
2002c9c3 | 537 | if (a5xx_gpu->gpmu_iova) |
2401a008 JC |
538 | msm_gem_put_iova(a5xx_gpu->gpmu_bo, gpu->id); |
539 | drm_gem_object_unreference_unlocked(a5xx_gpu->gpmu_bo); | |
540 | } | |
541 | ||
b5f103ab JC |
542 | adreno_gpu_cleanup(adreno_gpu); |
543 | kfree(a5xx_gpu); | |
544 | } | |
545 | ||
546 | static inline bool _a5xx_check_idle(struct msm_gpu *gpu) | |
547 | { | |
548 | if (gpu_read(gpu, REG_A5XX_RBBM_STATUS) & ~A5XX_RBBM_STATUS_HI_BUSY) | |
549 | return false; | |
550 | ||
551 | /* | |
552 | * Nearly every abnormality ends up pausing the GPU and triggering a | |
553 | * fault so we can safely just watch for this one interrupt to fire | |
554 | */ | |
555 | return !(gpu_read(gpu, REG_A5XX_RBBM_INT_0_STATUS) & | |
556 | A5XX_RBBM_INT_0_MASK_MISC_HANG_DETECT); | |
557 | } | |
558 | ||
559 | static bool a5xx_idle(struct msm_gpu *gpu) | |
560 | { | |
561 | /* wait for CP to drain ringbuffer: */ | |
562 | if (!adreno_idle(gpu)) | |
563 | return false; | |
564 | ||
565 | if (spin_until(_a5xx_check_idle(gpu))) { | |
566 | DRM_ERROR("%s: %ps: timeout waiting for GPU to idle: status %8.8X irq %8.8X\n", | |
567 | gpu->name, __builtin_return_address(0), | |
568 | gpu_read(gpu, REG_A5XX_RBBM_STATUS), | |
569 | gpu_read(gpu, REG_A5XX_RBBM_INT_0_STATUS)); | |
570 | ||
571 | return false; | |
572 | } | |
573 | ||
574 | return true; | |
575 | } | |
576 | ||
7f8036b7 RC |
577 | static int a5xx_fault_handler(void *arg, unsigned long iova, int flags) |
578 | { | |
579 | struct msm_gpu *gpu = arg; | |
580 | pr_warn_ratelimited("*** gpu fault: iova=%08lx, flags=%d (%u,%u,%u,%u)\n", | |
581 | iova, flags, | |
582 | gpu_read(gpu, REG_A5XX_CP_SCRATCH_REG(4)), | |
583 | gpu_read(gpu, REG_A5XX_CP_SCRATCH_REG(5)), | |
584 | gpu_read(gpu, REG_A5XX_CP_SCRATCH_REG(6)), | |
585 | gpu_read(gpu, REG_A5XX_CP_SCRATCH_REG(7))); | |
586 | ||
587 | return -EFAULT; | |
588 | } | |
589 | ||
b5f103ab JC |
590 | static void a5xx_cp_err_irq(struct msm_gpu *gpu) |
591 | { | |
592 | u32 status = gpu_read(gpu, REG_A5XX_CP_INTERRUPT_STATUS); | |
593 | ||
594 | if (status & A5XX_CP_INT_CP_OPCODE_ERROR) { | |
595 | u32 val; | |
596 | ||
597 | gpu_write(gpu, REG_A5XX_CP_PFP_STAT_ADDR, 0); | |
598 | ||
599 | /* | |
600 | * REG_A5XX_CP_PFP_STAT_DATA is indexed, and we want index 1 so | |
601 | * read it twice | |
602 | */ | |
603 | ||
604 | gpu_read(gpu, REG_A5XX_CP_PFP_STAT_DATA); | |
605 | val = gpu_read(gpu, REG_A5XX_CP_PFP_STAT_DATA); | |
606 | ||
607 | dev_err_ratelimited(gpu->dev->dev, "CP | opcode error | possible opcode=0x%8.8X\n", | |
608 | val); | |
609 | } | |
610 | ||
611 | if (status & A5XX_CP_INT_CP_HW_FAULT_ERROR) | |
612 | dev_err_ratelimited(gpu->dev->dev, "CP | HW fault | status=0x%8.8X\n", | |
613 | gpu_read(gpu, REG_A5XX_CP_HW_FAULT)); | |
614 | ||
615 | if (status & A5XX_CP_INT_CP_DMA_ERROR) | |
616 | dev_err_ratelimited(gpu->dev->dev, "CP | DMA error\n"); | |
617 | ||
618 | if (status & A5XX_CP_INT_CP_REGISTER_PROTECTION_ERROR) { | |
619 | u32 val = gpu_read(gpu, REG_A5XX_CP_PROTECT_STATUS); | |
620 | ||
621 | dev_err_ratelimited(gpu->dev->dev, | |
622 | "CP | protected mode error | %s | addr=0x%8.8X | status=0x%8.8X\n", | |
623 | val & (1 << 24) ? "WRITE" : "READ", | |
624 | (val & 0xFFFFF) >> 2, val); | |
625 | } | |
626 | ||
627 | if (status & A5XX_CP_INT_CP_AHB_ERROR) { | |
628 | u32 status = gpu_read(gpu, REG_A5XX_CP_AHB_FAULT); | |
629 | const char *access[16] = { "reserved", "reserved", | |
630 | "timestamp lo", "timestamp hi", "pfp read", "pfp write", | |
631 | "", "", "me read", "me write", "", "", "crashdump read", | |
632 | "crashdump write" }; | |
633 | ||
634 | dev_err_ratelimited(gpu->dev->dev, | |
635 | "CP | AHB error | addr=%X access=%s error=%d | status=0x%8.8X\n", | |
636 | status & 0xFFFFF, access[(status >> 24) & 0xF], | |
637 | (status & (1 << 31)), status); | |
638 | } | |
639 | } | |
640 | ||
7352fb5a | 641 | static void a5xx_rbbm_err_irq(struct msm_gpu *gpu, u32 status) |
b5f103ab | 642 | { |
b5f103ab JC |
643 | if (status & A5XX_RBBM_INT_0_MASK_RBBM_AHB_ERROR) { |
644 | u32 val = gpu_read(gpu, REG_A5XX_RBBM_AHB_ERROR_STATUS); | |
645 | ||
646 | dev_err_ratelimited(gpu->dev->dev, | |
647 | "RBBM | AHB bus error | %s | addr=0x%X | ports=0x%X:0x%X\n", | |
648 | val & (1 << 28) ? "WRITE" : "READ", | |
649 | (val & 0xFFFFF) >> 2, (val >> 20) & 0x3, | |
650 | (val >> 24) & 0xF); | |
651 | ||
652 | /* Clear the error */ | |
653 | gpu_write(gpu, REG_A5XX_RBBM_AHB_CMD, (1 << 4)); | |
7352fb5a JC |
654 | |
655 | /* Clear the interrupt */ | |
656 | gpu_write(gpu, REG_A5XX_RBBM_INT_CLEAR_CMD, | |
657 | A5XX_RBBM_INT_0_MASK_RBBM_AHB_ERROR); | |
b5f103ab JC |
658 | } |
659 | ||
660 | if (status & A5XX_RBBM_INT_0_MASK_RBBM_TRANSFER_TIMEOUT) | |
661 | dev_err_ratelimited(gpu->dev->dev, "RBBM | AHB transfer timeout\n"); | |
662 | ||
663 | if (status & A5XX_RBBM_INT_0_MASK_RBBM_ME_MS_TIMEOUT) | |
664 | dev_err_ratelimited(gpu->dev->dev, "RBBM | ME master split | status=0x%X\n", | |
665 | gpu_read(gpu, REG_A5XX_RBBM_AHB_ME_SPLIT_STATUS)); | |
666 | ||
667 | if (status & A5XX_RBBM_INT_0_MASK_RBBM_PFP_MS_TIMEOUT) | |
668 | dev_err_ratelimited(gpu->dev->dev, "RBBM | PFP master split | status=0x%X\n", | |
669 | gpu_read(gpu, REG_A5XX_RBBM_AHB_PFP_SPLIT_STATUS)); | |
670 | ||
671 | if (status & A5XX_RBBM_INT_0_MASK_RBBM_ETS_MS_TIMEOUT) | |
672 | dev_err_ratelimited(gpu->dev->dev, "RBBM | ETS master split | status=0x%X\n", | |
673 | gpu_read(gpu, REG_A5XX_RBBM_AHB_ETS_SPLIT_STATUS)); | |
674 | ||
675 | if (status & A5XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNC_OVERFLOW) | |
676 | dev_err_ratelimited(gpu->dev->dev, "RBBM | ATB ASYNC overflow\n"); | |
677 | ||
678 | if (status & A5XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW) | |
679 | dev_err_ratelimited(gpu->dev->dev, "RBBM | ATB bus overflow\n"); | |
680 | } | |
681 | ||
682 | static void a5xx_uche_err_irq(struct msm_gpu *gpu) | |
683 | { | |
684 | uint64_t addr = (uint64_t) gpu_read(gpu, REG_A5XX_UCHE_TRAP_LOG_HI); | |
685 | ||
686 | addr |= gpu_read(gpu, REG_A5XX_UCHE_TRAP_LOG_LO); | |
687 | ||
688 | dev_err_ratelimited(gpu->dev->dev, "UCHE | Out of bounds access | addr=0x%llX\n", | |
689 | addr); | |
690 | } | |
691 | ||
692 | static void a5xx_gpmu_err_irq(struct msm_gpu *gpu) | |
693 | { | |
694 | dev_err_ratelimited(gpu->dev->dev, "GPMU | voltage droop\n"); | |
695 | } | |
696 | ||
697 | #define RBBM_ERROR_MASK \ | |
698 | (A5XX_RBBM_INT_0_MASK_RBBM_AHB_ERROR | \ | |
699 | A5XX_RBBM_INT_0_MASK_RBBM_TRANSFER_TIMEOUT | \ | |
700 | A5XX_RBBM_INT_0_MASK_RBBM_ME_MS_TIMEOUT | \ | |
701 | A5XX_RBBM_INT_0_MASK_RBBM_PFP_MS_TIMEOUT | \ | |
702 | A5XX_RBBM_INT_0_MASK_RBBM_ETS_MS_TIMEOUT | \ | |
703 | A5XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNC_OVERFLOW) | |
704 | ||
705 | static irqreturn_t a5xx_irq(struct msm_gpu *gpu) | |
706 | { | |
707 | u32 status = gpu_read(gpu, REG_A5XX_RBBM_INT_0_STATUS); | |
708 | ||
7352fb5a JC |
709 | /* |
710 | * Clear all the interrupts except RBBM_AHB_ERROR - if we clear it | |
711 | * before the source is cleared the interrupt will storm. | |
712 | */ | |
713 | gpu_write(gpu, REG_A5XX_RBBM_INT_CLEAR_CMD, | |
714 | status & ~A5XX_RBBM_INT_0_MASK_RBBM_AHB_ERROR); | |
b5f103ab | 715 | |
7352fb5a | 716 | /* Pass status to a5xx_rbbm_err_irq because we've already cleared it */ |
b5f103ab | 717 | if (status & RBBM_ERROR_MASK) |
7352fb5a | 718 | a5xx_rbbm_err_irq(gpu, status); |
b5f103ab JC |
719 | |
720 | if (status & A5XX_RBBM_INT_0_MASK_CP_HW_ERROR) | |
721 | a5xx_cp_err_irq(gpu); | |
722 | ||
723 | if (status & A5XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS) | |
724 | a5xx_uche_err_irq(gpu); | |
725 | ||
726 | if (status & A5XX_RBBM_INT_0_MASK_GPMU_VOLTAGE_DROOP) | |
727 | a5xx_gpmu_err_irq(gpu); | |
728 | ||
729 | if (status & A5XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS) | |
730 | msm_gpu_retire(gpu); | |
731 | ||
732 | return IRQ_HANDLED; | |
733 | } | |
734 | ||
735 | static const u32 a5xx_register_offsets[REG_ADRENO_REGISTER_MAX] = { | |
736 | REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_BASE, REG_A5XX_CP_RB_BASE), | |
737 | REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_BASE_HI, REG_A5XX_CP_RB_BASE_HI), | |
738 | REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_RPTR_ADDR, REG_A5XX_CP_RB_RPTR_ADDR), | |
739 | REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_RPTR_ADDR_HI, | |
740 | REG_A5XX_CP_RB_RPTR_ADDR_HI), | |
741 | REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_RPTR, REG_A5XX_CP_RB_RPTR), | |
742 | REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_WPTR, REG_A5XX_CP_RB_WPTR), | |
743 | REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_CNTL, REG_A5XX_CP_RB_CNTL), | |
744 | }; | |
745 | ||
746 | static const u32 a5xx_registers[] = { | |
747 | 0x0000, 0x0002, 0x0004, 0x0020, 0x0022, 0x0026, 0x0029, 0x002B, | |
748 | 0x002E, 0x0035, 0x0038, 0x0042, 0x0044, 0x0044, 0x0047, 0x0095, | |
749 | 0x0097, 0x00BB, 0x03A0, 0x0464, 0x0469, 0x046F, 0x04D2, 0x04D3, | |
750 | 0x04E0, 0x0533, 0x0540, 0x0555, 0xF400, 0xF400, 0xF800, 0xF807, | |
751 | 0x0800, 0x081A, 0x081F, 0x0841, 0x0860, 0x0860, 0x0880, 0x08A0, | |
752 | 0x0B00, 0x0B12, 0x0B15, 0x0B28, 0x0B78, 0x0B7F, 0x0BB0, 0x0BBD, | |
753 | 0x0BC0, 0x0BC6, 0x0BD0, 0x0C53, 0x0C60, 0x0C61, 0x0C80, 0x0C82, | |
754 | 0x0C84, 0x0C85, 0x0C90, 0x0C98, 0x0CA0, 0x0CA0, 0x0CB0, 0x0CB2, | |
755 | 0x2180, 0x2185, 0x2580, 0x2585, 0x0CC1, 0x0CC1, 0x0CC4, 0x0CC7, | |
756 | 0x0CCC, 0x0CCC, 0x0CD0, 0x0CD8, 0x0CE0, 0x0CE5, 0x0CE8, 0x0CE8, | |
757 | 0x0CEC, 0x0CF1, 0x0CFB, 0x0D0E, 0x2100, 0x211E, 0x2140, 0x2145, | |
758 | 0x2500, 0x251E, 0x2540, 0x2545, 0x0D10, 0x0D17, 0x0D20, 0x0D23, | |
759 | 0x0D30, 0x0D30, 0x20C0, 0x20C0, 0x24C0, 0x24C0, 0x0E40, 0x0E43, | |
760 | 0x0E4A, 0x0E4A, 0x0E50, 0x0E57, 0x0E60, 0x0E7C, 0x0E80, 0x0E8E, | |
761 | 0x0E90, 0x0E96, 0x0EA0, 0x0EA8, 0x0EB0, 0x0EB2, 0xE140, 0xE147, | |
762 | 0xE150, 0xE187, 0xE1A0, 0xE1A9, 0xE1B0, 0xE1B6, 0xE1C0, 0xE1C7, | |
763 | 0xE1D0, 0xE1D1, 0xE200, 0xE201, 0xE210, 0xE21C, 0xE240, 0xE268, | |
764 | 0xE000, 0xE006, 0xE010, 0xE09A, 0xE0A0, 0xE0A4, 0xE0AA, 0xE0EB, | |
765 | 0xE100, 0xE105, 0xE380, 0xE38F, 0xE3B0, 0xE3B0, 0xE400, 0xE405, | |
766 | 0xE408, 0xE4E9, 0xE4F0, 0xE4F0, 0xE280, 0xE280, 0xE282, 0xE2A3, | |
767 | 0xE2A5, 0xE2C2, 0xE940, 0xE947, 0xE950, 0xE987, 0xE9A0, 0xE9A9, | |
768 | 0xE9B0, 0xE9B6, 0xE9C0, 0xE9C7, 0xE9D0, 0xE9D1, 0xEA00, 0xEA01, | |
769 | 0xEA10, 0xEA1C, 0xEA40, 0xEA68, 0xE800, 0xE806, 0xE810, 0xE89A, | |
770 | 0xE8A0, 0xE8A4, 0xE8AA, 0xE8EB, 0xE900, 0xE905, 0xEB80, 0xEB8F, | |
771 | 0xEBB0, 0xEBB0, 0xEC00, 0xEC05, 0xEC08, 0xECE9, 0xECF0, 0xECF0, | |
772 | 0xEA80, 0xEA80, 0xEA82, 0xEAA3, 0xEAA5, 0xEAC2, 0xA800, 0xA8FF, | |
773 | 0xAC60, 0xAC60, 0xB000, 0xB97F, 0xB9A0, 0xB9BF, | |
774 | ~0 | |
775 | }; | |
776 | ||
777 | static void a5xx_dump(struct msm_gpu *gpu) | |
778 | { | |
779 | dev_info(gpu->dev->dev, "status: %08x\n", | |
780 | gpu_read(gpu, REG_A5XX_RBBM_STATUS)); | |
781 | adreno_dump(gpu); | |
782 | } | |
783 | ||
784 | static int a5xx_pm_resume(struct msm_gpu *gpu) | |
785 | { | |
2401a008 JC |
786 | int ret; |
787 | ||
788 | /* Turn on the core power */ | |
789 | ret = msm_gpu_pm_resume(gpu); | |
790 | if (ret) | |
791 | return ret; | |
792 | ||
793 | /* Turn the RBCCU domain first to limit the chances of voltage droop */ | |
794 | gpu_write(gpu, REG_A5XX_GPMU_RBCCU_POWER_CNTL, 0x778000); | |
795 | ||
796 | /* Wait 3 usecs before polling */ | |
797 | udelay(3); | |
798 | ||
799 | ret = spin_usecs(gpu, 20, REG_A5XX_GPMU_RBCCU_PWR_CLK_STATUS, | |
800 | (1 << 20), (1 << 20)); | |
801 | if (ret) { | |
802 | DRM_ERROR("%s: timeout waiting for RBCCU GDSC enable: %X\n", | |
803 | gpu->name, | |
804 | gpu_read(gpu, REG_A5XX_GPMU_RBCCU_PWR_CLK_STATUS)); | |
805 | return ret; | |
806 | } | |
807 | ||
808 | /* Turn on the SP domain */ | |
809 | gpu_write(gpu, REG_A5XX_GPMU_SP_POWER_CNTL, 0x778000); | |
810 | ret = spin_usecs(gpu, 20, REG_A5XX_GPMU_SP_PWR_CLK_STATUS, | |
811 | (1 << 20), (1 << 20)); | |
812 | if (ret) | |
813 | DRM_ERROR("%s: timeout waiting for SP GDSC enable\n", | |
814 | gpu->name); | |
815 | ||
816 | return ret; | |
b5f103ab JC |
817 | } |
818 | ||
819 | static int a5xx_pm_suspend(struct msm_gpu *gpu) | |
820 | { | |
2401a008 JC |
821 | /* Clear the VBIF pipe before shutting down */ |
822 | gpu_write(gpu, REG_A5XX_VBIF_XIN_HALT_CTRL0, 0xF); | |
823 | spin_until((gpu_read(gpu, REG_A5XX_VBIF_XIN_HALT_CTRL1) & 0xF) == 0xF); | |
824 | ||
825 | gpu_write(gpu, REG_A5XX_VBIF_XIN_HALT_CTRL0, 0); | |
826 | ||
827 | /* | |
828 | * Reset the VBIF before power collapse to avoid issue with FIFO | |
829 | * entries | |
830 | */ | |
831 | gpu_write(gpu, REG_A5XX_RBBM_BLOCK_SW_RESET_CMD, 0x003C0000); | |
832 | gpu_write(gpu, REG_A5XX_RBBM_BLOCK_SW_RESET_CMD, 0x00000000); | |
833 | ||
b5f103ab JC |
834 | return msm_gpu_pm_suspend(gpu); |
835 | } | |
836 | ||
837 | static int a5xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value) | |
838 | { | |
839 | *value = gpu_read64(gpu, REG_A5XX_RBBM_PERFCTR_CP_0_LO, | |
840 | REG_A5XX_RBBM_PERFCTR_CP_0_HI); | |
841 | ||
842 | return 0; | |
843 | } | |
844 | ||
845 | #ifdef CONFIG_DEBUG_FS | |
846 | static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m) | |
847 | { | |
b5f103ab JC |
848 | seq_printf(m, "status: %08x\n", |
849 | gpu_read(gpu, REG_A5XX_RBBM_STATUS)); | |
b5f103ab JC |
850 | adreno_show(gpu, m); |
851 | } | |
852 | #endif | |
853 | ||
854 | static const struct adreno_gpu_funcs funcs = { | |
855 | .base = { | |
856 | .get_param = adreno_get_param, | |
857 | .hw_init = a5xx_hw_init, | |
858 | .pm_suspend = a5xx_pm_suspend, | |
859 | .pm_resume = a5xx_pm_resume, | |
860 | .recover = a5xx_recover, | |
861 | .last_fence = adreno_last_fence, | |
862 | .submit = a5xx_submit, | |
863 | .flush = adreno_flush, | |
864 | .idle = a5xx_idle, | |
865 | .irq = a5xx_irq, | |
866 | .destroy = a5xx_destroy, | |
0c3eaf1f | 867 | #ifdef CONFIG_DEBUG_FS |
b5f103ab | 868 | .show = a5xx_show, |
0c3eaf1f | 869 | #endif |
b5f103ab JC |
870 | }, |
871 | .get_timestamp = a5xx_get_timestamp, | |
872 | }; | |
873 | ||
874 | struct msm_gpu *a5xx_gpu_init(struct drm_device *dev) | |
875 | { | |
876 | struct msm_drm_private *priv = dev->dev_private; | |
877 | struct platform_device *pdev = priv->gpu_pdev; | |
878 | struct a5xx_gpu *a5xx_gpu = NULL; | |
879 | struct adreno_gpu *adreno_gpu; | |
880 | struct msm_gpu *gpu; | |
881 | int ret; | |
882 | ||
883 | if (!pdev) { | |
884 | dev_err(dev->dev, "No A5XX device is defined\n"); | |
885 | return ERR_PTR(-ENXIO); | |
886 | } | |
887 | ||
888 | a5xx_gpu = kzalloc(sizeof(*a5xx_gpu), GFP_KERNEL); | |
889 | if (!a5xx_gpu) | |
890 | return ERR_PTR(-ENOMEM); | |
891 | ||
892 | adreno_gpu = &a5xx_gpu->base; | |
893 | gpu = &adreno_gpu->base; | |
894 | ||
895 | a5xx_gpu->pdev = pdev; | |
896 | adreno_gpu->registers = a5xx_registers; | |
897 | adreno_gpu->reg_offsets = a5xx_register_offsets; | |
898 | ||
2401a008 JC |
899 | a5xx_gpu->lm_leakage = 0x4E001A; |
900 | ||
b5f103ab JC |
901 | ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs); |
902 | if (ret) { | |
903 | a5xx_destroy(&(a5xx_gpu->base.base)); | |
904 | return ERR_PTR(ret); | |
905 | } | |
906 | ||
7f8036b7 RC |
907 | if (gpu->aspace) |
908 | msm_mmu_set_fault_handler(gpu->aspace->mmu, gpu, a5xx_fault_handler); | |
909 | ||
b5f103ab JC |
910 | return gpu; |
911 | } |