Commit | Line | Data |
---|---|---|
7198e6b0 RC |
1 | /* |
2 | * Copyright (C) 2013 Red Hat | |
3 | * Author: Rob Clark <robdclark@gmail.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms of the GNU General Public License version 2 as published by | |
7 | * the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License along with | |
15 | * this program. If not, see <http://www.gnu.org/licenses/>. | |
16 | */ | |
17 | ||
55459968 RC |
18 | #ifdef CONFIG_MSM_OCMEM |
19 | # include <mach/ocmem.h> | |
20 | #endif | |
21 | ||
7198e6b0 RC |
22 | #include "a3xx_gpu.h" |
23 | ||
24 | #define A3XX_INT0_MASK \ | |
25 | (A3XX_INT0_RBBM_AHB_ERROR | \ | |
26 | A3XX_INT0_RBBM_ATB_BUS_OVERFLOW | \ | |
27 | A3XX_INT0_CP_T0_PACKET_IN_IB | \ | |
28 | A3XX_INT0_CP_OPCODE_ERROR | \ | |
29 | A3XX_INT0_CP_RESERVED_BIT_ERROR | \ | |
30 | A3XX_INT0_CP_HW_FAULT | \ | |
31 | A3XX_INT0_CP_IB1_INT | \ | |
32 | A3XX_INT0_CP_IB2_INT | \ | |
33 | A3XX_INT0_CP_RB_INT | \ | |
34 | A3XX_INT0_CP_REG_PROTECT_FAULT | \ | |
35 | A3XX_INT0_CP_AHB_ERROR_HALT | \ | |
36 | A3XX_INT0_UCHE_OOB_ACCESS) | |
37 | ||
5b6ef08e RC |
38 | |
39 | static bool hang_debug = false; | |
40 | MODULE_PARM_DESC(hang_debug, "Dump registers when hang is detected (can be slow!)"); | |
41 | module_param_named(hang_debug, hang_debug, bool, 0600); | |
42 | static void a3xx_dump(struct msm_gpu *gpu); | |
43 | ||
7198e6b0 RC |
44 | static struct platform_device *a3xx_pdev; |
45 | ||
46 | static void a3xx_me_init(struct msm_gpu *gpu) | |
47 | { | |
48 | struct msm_ringbuffer *ring = gpu->rb; | |
49 | ||
50 | OUT_PKT3(ring, CP_ME_INIT, 17); | |
51 | OUT_RING(ring, 0x000003f7); | |
52 | OUT_RING(ring, 0x00000000); | |
53 | OUT_RING(ring, 0x00000000); | |
54 | OUT_RING(ring, 0x00000000); | |
55 | OUT_RING(ring, 0x00000080); | |
56 | OUT_RING(ring, 0x00000100); | |
57 | OUT_RING(ring, 0x00000180); | |
58 | OUT_RING(ring, 0x00006600); | |
59 | OUT_RING(ring, 0x00000150); | |
60 | OUT_RING(ring, 0x0000014e); | |
61 | OUT_RING(ring, 0x00000154); | |
62 | OUT_RING(ring, 0x00000001); | |
63 | OUT_RING(ring, 0x00000000); | |
64 | OUT_RING(ring, 0x00000000); | |
65 | OUT_RING(ring, 0x00000000); | |
66 | OUT_RING(ring, 0x00000000); | |
67 | OUT_RING(ring, 0x00000000); | |
68 | ||
69 | gpu->funcs->flush(gpu); | |
70 | gpu->funcs->idle(gpu); | |
71 | } | |
72 | ||
73 | static int a3xx_hw_init(struct msm_gpu *gpu) | |
74 | { | |
75 | struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); | |
55459968 | 76 | struct a3xx_gpu *a3xx_gpu = to_a3xx_gpu(adreno_gpu); |
7198e6b0 RC |
77 | uint32_t *ptr, len; |
78 | int i, ret; | |
79 | ||
80 | DBG("%s", gpu->name); | |
81 | ||
82 | if (adreno_is_a305(adreno_gpu)) { | |
83 | /* Set up 16 deep read/write request queues: */ | |
84 | gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x10101010); | |
85 | gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x10101010); | |
86 | gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x10101010); | |
87 | gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x10101010); | |
88 | gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303); | |
89 | gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x10101010); | |
90 | gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x10101010); | |
91 | /* Enable WR-REQ: */ | |
92 | gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x0000ff); | |
93 | /* Set up round robin arbitration between both AXI ports: */ | |
94 | gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030); | |
95 | /* Set up AOOO: */ | |
96 | gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003c); | |
97 | gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003c003c); | |
98 | ||
99 | } else if (adreno_is_a320(adreno_gpu)) { | |
100 | /* Set up 16 deep read/write request queues: */ | |
101 | gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x10101010); | |
102 | gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x10101010); | |
103 | gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x10101010); | |
104 | gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x10101010); | |
105 | gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303); | |
106 | gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x10101010); | |
107 | gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x10101010); | |
108 | /* Enable WR-REQ: */ | |
109 | gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x0000ff); | |
110 | /* Set up round robin arbitration between both AXI ports: */ | |
111 | gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030); | |
112 | /* Set up AOOO: */ | |
113 | gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003c); | |
114 | gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003c003c); | |
115 | /* Enable 1K sort: */ | |
116 | gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x000000ff); | |
117 | gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4); | |
118 | ||
55459968 RC |
119 | } else if (adreno_is_a330v2(adreno_gpu)) { |
120 | /* | |
121 | * Most of the VBIF registers on 8974v2 have the correct | |
122 | * values at power on, so we won't modify those if we don't | |
123 | * need to | |
124 | */ | |
125 | /* Enable 1k sort: */ | |
126 | gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001003f); | |
127 | gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4); | |
128 | /* Enable WR-REQ: */ | |
129 | gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x00003f); | |
130 | gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303); | |
131 | /* Set up VBIF_ROUND_ROBIN_QOS_ARB: */ | |
132 | gpu_write(gpu, REG_A3XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x0003); | |
133 | ||
7198e6b0 RC |
134 | } else if (adreno_is_a330(adreno_gpu)) { |
135 | /* Set up 16 deep read/write request queues: */ | |
136 | gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF0, 0x18181818); | |
137 | gpu_write(gpu, REG_A3XX_VBIF_IN_RD_LIM_CONF1, 0x18181818); | |
138 | gpu_write(gpu, REG_A3XX_VBIF_OUT_RD_LIM_CONF0, 0x18181818); | |
139 | gpu_write(gpu, REG_A3XX_VBIF_OUT_WR_LIM_CONF0, 0x18181818); | |
140 | gpu_write(gpu, REG_A3XX_VBIF_DDR_OUT_MAX_BURST, 0x0000303); | |
141 | gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF0, 0x18181818); | |
142 | gpu_write(gpu, REG_A3XX_VBIF_IN_WR_LIM_CONF1, 0x18181818); | |
143 | /* Enable WR-REQ: */ | |
144 | gpu_write(gpu, REG_A3XX_VBIF_GATE_OFF_WRREQ_EN, 0x00003f); | |
145 | /* Set up round robin arbitration between both AXI ports: */ | |
146 | gpu_write(gpu, REG_A3XX_VBIF_ARB_CTL, 0x00000030); | |
147 | /* Set up VBIF_ROUND_ROBIN_QOS_ARB: */ | |
148 | gpu_write(gpu, REG_A3XX_VBIF_ROUND_ROBIN_QOS_ARB, 0x0001); | |
149 | /* Set up AOOO: */ | |
55459968 RC |
150 | gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO_EN, 0x0000003f); |
151 | gpu_write(gpu, REG_A3XX_VBIF_OUT_AXI_AOOO, 0x003f003f); | |
7198e6b0 | 152 | /* Enable 1K sort: */ |
55459968 | 153 | gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT, 0x0001003f); |
7198e6b0 RC |
154 | gpu_write(gpu, REG_A3XX_VBIF_ABIT_SORT_CONF, 0x000000a4); |
155 | /* Disable VBIF clock gating. This is to enable AXI running | |
156 | * higher frequency than GPU: | |
157 | */ | |
158 | gpu_write(gpu, REG_A3XX_VBIF_CLKON, 0x00000001); | |
159 | ||
160 | } else { | |
161 | BUG(); | |
162 | } | |
163 | ||
164 | /* Make all blocks contribute to the GPU BUSY perf counter: */ | |
165 | gpu_write(gpu, REG_A3XX_RBBM_GPU_BUSY_MASKED, 0xffffffff); | |
166 | ||
167 | /* Tune the hystersis counters for SP and CP idle detection: */ | |
168 | gpu_write(gpu, REG_A3XX_RBBM_SP_HYST_CNT, 0x10); | |
169 | gpu_write(gpu, REG_A3XX_RBBM_WAIT_IDLE_CLOCKS_CTL, 0x10); | |
170 | ||
171 | /* Enable the RBBM error reporting bits. This lets us get | |
172 | * useful information on failure: | |
173 | */ | |
174 | gpu_write(gpu, REG_A3XX_RBBM_AHB_CTL0, 0x00000001); | |
175 | ||
176 | /* Enable AHB error reporting: */ | |
177 | gpu_write(gpu, REG_A3XX_RBBM_AHB_CTL1, 0xa6ffffff); | |
178 | ||
179 | /* Turn on the power counters: */ | |
180 | gpu_write(gpu, REG_A3XX_RBBM_RBBM_CTL, 0x00030000); | |
181 | ||
182 | /* Turn on hang detection - this spews a lot of useful information | |
183 | * into the RBBM registers on a hang: | |
184 | */ | |
185 | gpu_write(gpu, REG_A3XX_RBBM_INTERFACE_HANG_INT_CTL, 0x00010fff); | |
186 | ||
187 | /* Enable 64-byte cacheline size. HW Default is 32-byte (0x000000E0): */ | |
188 | gpu_write(gpu, REG_A3XX_UCHE_CACHE_MODE_CONTROL_REG, 0x00000001); | |
189 | ||
190 | /* Enable Clock gating: */ | |
55459968 RC |
191 | if (adreno_is_a320(adreno_gpu)) |
192 | gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbfffffff); | |
193 | else if (adreno_is_a330v2(adreno_gpu)) | |
194 | gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xaaaaaaaa); | |
195 | else if (adreno_is_a330(adreno_gpu)) | |
196 | gpu_write(gpu, REG_A3XX_RBBM_CLOCK_CTL, 0xbffcffff); | |
197 | ||
198 | if (adreno_is_a330v2(adreno_gpu)) | |
199 | gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x05515455); | |
200 | else if (adreno_is_a330(adreno_gpu)) | |
201 | gpu_write(gpu, REG_A3XX_RBBM_GPR0_CTL, 0x00000000); | |
202 | ||
203 | /* Set the OCMEM base address for A330, etc */ | |
204 | if (a3xx_gpu->ocmem_hdl) { | |
205 | gpu_write(gpu, REG_A3XX_RB_GMEM_BASE_ADDR, | |
206 | (unsigned int)(a3xx_gpu->ocmem_base >> 14)); | |
207 | } | |
7198e6b0 RC |
208 | |
209 | /* Turn on performance counters: */ | |
210 | gpu_write(gpu, REG_A3XX_RBBM_PERFCTR_CTL, 0x01); | |
211 | ||
212 | /* Set SP perfcounter 7 to count SP_FS_FULL_ALU_INSTRUCTIONS | |
213 | * we will use this to augment our hang detection: | |
214 | */ | |
215 | gpu_write(gpu, REG_A3XX_SP_PERFCOUNTER7_SELECT, | |
216 | SP_FS_FULL_ALU_INSTRUCTIONS); | |
217 | ||
218 | gpu_write(gpu, REG_A3XX_RBBM_INT_0_MASK, A3XX_INT0_MASK); | |
219 | ||
220 | ret = adreno_hw_init(gpu); | |
221 | if (ret) | |
222 | return ret; | |
223 | ||
224 | /* setup access protection: */ | |
225 | gpu_write(gpu, REG_A3XX_CP_PROTECT_CTRL, 0x00000007); | |
226 | ||
227 | /* RBBM registers */ | |
228 | gpu_write(gpu, REG_A3XX_CP_PROTECT(0), 0x63000040); | |
229 | gpu_write(gpu, REG_A3XX_CP_PROTECT(1), 0x62000080); | |
230 | gpu_write(gpu, REG_A3XX_CP_PROTECT(2), 0x600000cc); | |
231 | gpu_write(gpu, REG_A3XX_CP_PROTECT(3), 0x60000108); | |
232 | gpu_write(gpu, REG_A3XX_CP_PROTECT(4), 0x64000140); | |
233 | gpu_write(gpu, REG_A3XX_CP_PROTECT(5), 0x66000400); | |
234 | ||
235 | /* CP registers */ | |
236 | gpu_write(gpu, REG_A3XX_CP_PROTECT(6), 0x65000700); | |
237 | gpu_write(gpu, REG_A3XX_CP_PROTECT(7), 0x610007d8); | |
238 | gpu_write(gpu, REG_A3XX_CP_PROTECT(8), 0x620007e0); | |
239 | gpu_write(gpu, REG_A3XX_CP_PROTECT(9), 0x61001178); | |
240 | gpu_write(gpu, REG_A3XX_CP_PROTECT(10), 0x64001180); | |
241 | ||
242 | /* RB registers */ | |
243 | gpu_write(gpu, REG_A3XX_CP_PROTECT(11), 0x60003300); | |
244 | ||
245 | /* VBIF registers */ | |
246 | gpu_write(gpu, REG_A3XX_CP_PROTECT(12), 0x6b00c000); | |
247 | ||
248 | /* NOTE: PM4/micro-engine firmware registers look to be the same | |
249 | * for a2xx and a3xx.. we could possibly push that part down to | |
250 | * adreno_gpu base class. Or push both PM4 and PFP but | |
251 | * parameterize the pfp ucode addr/data registers.. | |
252 | */ | |
253 | ||
254 | /* Load PM4: */ | |
255 | ptr = (uint32_t *)(adreno_gpu->pm4->data); | |
256 | len = adreno_gpu->pm4->size / 4; | |
e529c7e6 | 257 | DBG("loading PM4 ucode version: %x", ptr[1]); |
7198e6b0 RC |
258 | |
259 | gpu_write(gpu, REG_AXXX_CP_DEBUG, | |
260 | AXXX_CP_DEBUG_DYNAMIC_CLK_DISABLE | | |
261 | AXXX_CP_DEBUG_MIU_128BIT_WRITE_ENABLE); | |
262 | gpu_write(gpu, REG_AXXX_CP_ME_RAM_WADDR, 0); | |
263 | for (i = 1; i < len; i++) | |
264 | gpu_write(gpu, REG_AXXX_CP_ME_RAM_DATA, ptr[i]); | |
265 | ||
266 | /* Load PFP: */ | |
267 | ptr = (uint32_t *)(adreno_gpu->pfp->data); | |
268 | len = adreno_gpu->pfp->size / 4; | |
e529c7e6 | 269 | DBG("loading PFP ucode version: %x", ptr[5]); |
7198e6b0 RC |
270 | |
271 | gpu_write(gpu, REG_A3XX_CP_PFP_UCODE_ADDR, 0); | |
272 | for (i = 1; i < len; i++) | |
273 | gpu_write(gpu, REG_A3XX_CP_PFP_UCODE_DATA, ptr[i]); | |
274 | ||
275 | /* CP ROQ queue sizes (bytes) - RB:16, ST:16, IB1:32, IB2:64 */ | |
55459968 | 276 | if (adreno_is_a305(adreno_gpu) || adreno_is_a320(adreno_gpu)) { |
7198e6b0 RC |
277 | gpu_write(gpu, REG_AXXX_CP_QUEUE_THRESHOLDS, |
278 | AXXX_CP_QUEUE_THRESHOLDS_CSQ_IB1_START(2) | | |
279 | AXXX_CP_QUEUE_THRESHOLDS_CSQ_IB2_START(6) | | |
280 | AXXX_CP_QUEUE_THRESHOLDS_CSQ_ST_START(14)); | |
55459968 RC |
281 | } else if (adreno_is_a330(adreno_gpu)) { |
282 | /* NOTE: this (value take from downstream android driver) | |
283 | * includes some bits outside of the known bitfields. But | |
284 | * A330 has this "MERCIU queue" thing too, which might | |
285 | * explain a new bitfield or reshuffling: | |
286 | */ | |
287 | gpu_write(gpu, REG_AXXX_CP_QUEUE_THRESHOLDS, 0x003e2008); | |
288 | } | |
7198e6b0 RC |
289 | |
290 | /* clear ME_HALT to start micro engine */ | |
291 | gpu_write(gpu, REG_AXXX_CP_ME_CNTL, 0); | |
292 | ||
293 | a3xx_me_init(gpu); | |
294 | ||
295 | return 0; | |
296 | } | |
297 | ||
55459968 RC |
298 | static void a3xx_recover(struct msm_gpu *gpu) |
299 | { | |
5b6ef08e RC |
300 | /* dump registers before resetting gpu, if enabled: */ |
301 | if (hang_debug) | |
302 | a3xx_dump(gpu); | |
55459968 RC |
303 | gpu_write(gpu, REG_A3XX_RBBM_SW_RESET_CMD, 1); |
304 | gpu_read(gpu, REG_A3XX_RBBM_SW_RESET_CMD); | |
305 | gpu_write(gpu, REG_A3XX_RBBM_SW_RESET_CMD, 0); | |
306 | adreno_recover(gpu); | |
307 | } | |
308 | ||
7198e6b0 RC |
309 | static void a3xx_destroy(struct msm_gpu *gpu) |
310 | { | |
311 | struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); | |
312 | struct a3xx_gpu *a3xx_gpu = to_a3xx_gpu(adreno_gpu); | |
313 | ||
314 | DBG("%s", gpu->name); | |
315 | ||
316 | adreno_gpu_cleanup(adreno_gpu); | |
55459968 RC |
317 | |
318 | #ifdef CONFIG_MSM_OCMEM | |
319 | if (a3xx_gpu->ocmem_base) | |
320 | ocmem_free(OCMEM_GRAPHICS, a3xx_gpu->ocmem_hdl); | |
321 | #endif | |
322 | ||
7198e6b0 RC |
323 | put_device(&a3xx_gpu->pdev->dev); |
324 | kfree(a3xx_gpu); | |
325 | } | |
326 | ||
327 | static void a3xx_idle(struct msm_gpu *gpu) | |
328 | { | |
7198e6b0 RC |
329 | /* wait for ringbuffer to drain: */ |
330 | adreno_idle(gpu); | |
331 | ||
7198e6b0 | 332 | /* then wait for GPU to finish: */ |
0963756f RC |
333 | if (spin_until(!(gpu_read(gpu, REG_A3XX_RBBM_STATUS) & |
334 | A3XX_RBBM_STATUS_GPU_BUSY))) | |
335 | DRM_ERROR("%s: timeout waiting for GPU to idle!\n", gpu->name); | |
7198e6b0 RC |
336 | |
337 | /* TODO maybe we need to reset GPU here to recover from hang? */ | |
338 | } | |
339 | ||
340 | static irqreturn_t a3xx_irq(struct msm_gpu *gpu) | |
341 | { | |
342 | uint32_t status; | |
343 | ||
344 | status = gpu_read(gpu, REG_A3XX_RBBM_INT_0_STATUS); | |
345 | DBG("%s: %08x", gpu->name, status); | |
346 | ||
347 | // TODO | |
348 | ||
349 | gpu_write(gpu, REG_A3XX_RBBM_INT_CLEAR_CMD, status); | |
350 | ||
351 | msm_gpu_retire(gpu); | |
352 | ||
353 | return IRQ_HANDLED; | |
354 | } | |
355 | ||
7198e6b0 RC |
356 | static const unsigned int a3xx_registers[] = { |
357 | 0x0000, 0x0002, 0x0010, 0x0012, 0x0018, 0x0018, 0x0020, 0x0027, | |
358 | 0x0029, 0x002b, 0x002e, 0x0033, 0x0040, 0x0042, 0x0050, 0x005c, | |
359 | 0x0060, 0x006c, 0x0080, 0x0082, 0x0084, 0x0088, 0x0090, 0x00e5, | |
360 | 0x00ea, 0x00ed, 0x0100, 0x0100, 0x0110, 0x0123, 0x01c0, 0x01c1, | |
361 | 0x01c3, 0x01c5, 0x01c7, 0x01c7, 0x01d5, 0x01d9, 0x01dc, 0x01dd, | |
362 | 0x01ea, 0x01ea, 0x01ee, 0x01f1, 0x01f5, 0x01f5, 0x01fc, 0x01ff, | |
363 | 0x0440, 0x0440, 0x0443, 0x0443, 0x0445, 0x0445, 0x044d, 0x044f, | |
364 | 0x0452, 0x0452, 0x0454, 0x046f, 0x047c, 0x047c, 0x047f, 0x047f, | |
365 | 0x0578, 0x057f, 0x0600, 0x0602, 0x0605, 0x0607, 0x060a, 0x060e, | |
366 | 0x0612, 0x0614, 0x0c01, 0x0c02, 0x0c06, 0x0c1d, 0x0c3d, 0x0c3f, | |
367 | 0x0c48, 0x0c4b, 0x0c80, 0x0c80, 0x0c88, 0x0c8b, 0x0ca0, 0x0cb7, | |
368 | 0x0cc0, 0x0cc1, 0x0cc6, 0x0cc7, 0x0ce4, 0x0ce5, 0x0e00, 0x0e05, | |
369 | 0x0e0c, 0x0e0c, 0x0e22, 0x0e23, 0x0e41, 0x0e45, 0x0e64, 0x0e65, | |
370 | 0x0e80, 0x0e82, 0x0e84, 0x0e89, 0x0ea0, 0x0ea1, 0x0ea4, 0x0ea7, | |
371 | 0x0ec4, 0x0ecb, 0x0ee0, 0x0ee0, 0x0f00, 0x0f01, 0x0f03, 0x0f09, | |
372 | 0x2040, 0x2040, 0x2044, 0x2044, 0x2048, 0x204d, 0x2068, 0x2069, | |
373 | 0x206c, 0x206d, 0x2070, 0x2070, 0x2072, 0x2072, 0x2074, 0x2075, | |
374 | 0x2079, 0x207a, 0x20c0, 0x20d3, 0x20e4, 0x20ef, 0x2100, 0x2109, | |
375 | 0x210c, 0x210c, 0x210e, 0x210e, 0x2110, 0x2111, 0x2114, 0x2115, | |
376 | 0x21e4, 0x21e4, 0x21ea, 0x21ea, 0x21ec, 0x21ed, 0x21f0, 0x21f0, | |
377 | 0x2200, 0x2212, 0x2214, 0x2217, 0x221a, 0x221a, 0x2240, 0x227e, | |
378 | 0x2280, 0x228b, 0x22c0, 0x22c0, 0x22c4, 0x22ce, 0x22d0, 0x22d8, | |
379 | 0x22df, 0x22e6, 0x22e8, 0x22e9, 0x22ec, 0x22ec, 0x22f0, 0x22f7, | |
380 | 0x22ff, 0x22ff, 0x2340, 0x2343, 0x2348, 0x2349, 0x2350, 0x2356, | |
381 | 0x2360, 0x2360, 0x2440, 0x2440, 0x2444, 0x2444, 0x2448, 0x244d, | |
382 | 0x2468, 0x2469, 0x246c, 0x246d, 0x2470, 0x2470, 0x2472, 0x2472, | |
383 | 0x2474, 0x2475, 0x2479, 0x247a, 0x24c0, 0x24d3, 0x24e4, 0x24ef, | |
384 | 0x2500, 0x2509, 0x250c, 0x250c, 0x250e, 0x250e, 0x2510, 0x2511, | |
385 | 0x2514, 0x2515, 0x25e4, 0x25e4, 0x25ea, 0x25ea, 0x25ec, 0x25ed, | |
386 | 0x25f0, 0x25f0, 0x2600, 0x2612, 0x2614, 0x2617, 0x261a, 0x261a, | |
387 | 0x2640, 0x267e, 0x2680, 0x268b, 0x26c0, 0x26c0, 0x26c4, 0x26ce, | |
388 | 0x26d0, 0x26d8, 0x26df, 0x26e6, 0x26e8, 0x26e9, 0x26ec, 0x26ec, | |
389 | 0x26f0, 0x26f7, 0x26ff, 0x26ff, 0x2740, 0x2743, 0x2748, 0x2749, | |
390 | 0x2750, 0x2756, 0x2760, 0x2760, 0x300c, 0x300e, 0x301c, 0x301d, | |
391 | 0x302a, 0x302a, 0x302c, 0x302d, 0x3030, 0x3031, 0x3034, 0x3036, | |
392 | 0x303c, 0x303c, 0x305e, 0x305f, | |
393 | }; | |
394 | ||
5b6ef08e | 395 | #ifdef CONFIG_DEBUG_FS |
7198e6b0 RC |
396 | static void a3xx_show(struct msm_gpu *gpu, struct seq_file *m) |
397 | { | |
398 | int i; | |
399 | ||
400 | adreno_show(gpu, m); | |
401 | seq_printf(m, "status: %08x\n", | |
402 | gpu_read(gpu, REG_A3XX_RBBM_STATUS)); | |
403 | ||
404 | /* dump these out in a form that can be parsed by demsm: */ | |
405 | seq_printf(m, "IO:region %s 00000000 00020000\n", gpu->name); | |
406 | for (i = 0; i < ARRAY_SIZE(a3xx_registers); i += 2) { | |
407 | uint32_t start = a3xx_registers[i]; | |
408 | uint32_t end = a3xx_registers[i+1]; | |
409 | uint32_t addr; | |
410 | ||
411 | for (addr = start; addr <= end; addr++) { | |
412 | uint32_t val = gpu_read(gpu, addr); | |
413 | seq_printf(m, "IO:R %08x %08x\n", addr<<2, val); | |
414 | } | |
415 | } | |
416 | } | |
417 | #endif | |
418 | ||
5b6ef08e RC |
419 | /* would be nice to not have to duplicate the _show() stuff with printk(): */ |
420 | static void a3xx_dump(struct msm_gpu *gpu) | |
421 | { | |
422 | int i; | |
423 | ||
424 | adreno_dump(gpu); | |
425 | printk("status: %08x\n", | |
426 | gpu_read(gpu, REG_A3XX_RBBM_STATUS)); | |
427 | ||
428 | /* dump these out in a form that can be parsed by demsm: */ | |
429 | printk("IO:region %s 00000000 00020000\n", gpu->name); | |
430 | for (i = 0; i < ARRAY_SIZE(a3xx_registers); i += 2) { | |
431 | uint32_t start = a3xx_registers[i]; | |
432 | uint32_t end = a3xx_registers[i+1]; | |
433 | uint32_t addr; | |
434 | ||
435 | for (addr = start; addr <= end; addr++) { | |
436 | uint32_t val = gpu_read(gpu, addr); | |
437 | printk("IO:R %08x %08x\n", addr<<2, val); | |
438 | } | |
439 | } | |
440 | } | |
441 | ||
7198e6b0 RC |
442 | static const struct adreno_gpu_funcs funcs = { |
443 | .base = { | |
444 | .get_param = adreno_get_param, | |
445 | .hw_init = a3xx_hw_init, | |
446 | .pm_suspend = msm_gpu_pm_suspend, | |
447 | .pm_resume = msm_gpu_pm_resume, | |
55459968 | 448 | .recover = a3xx_recover, |
7198e6b0 RC |
449 | .last_fence = adreno_last_fence, |
450 | .submit = adreno_submit, | |
451 | .flush = adreno_flush, | |
452 | .idle = a3xx_idle, | |
453 | .irq = a3xx_irq, | |
454 | .destroy = a3xx_destroy, | |
455 | #ifdef CONFIG_DEBUG_FS | |
456 | .show = a3xx_show, | |
457 | #endif | |
458 | }, | |
459 | }; | |
460 | ||
461 | struct msm_gpu *a3xx_gpu_init(struct drm_device *dev) | |
462 | { | |
463 | struct a3xx_gpu *a3xx_gpu = NULL; | |
55459968 | 464 | struct adreno_gpu *adreno_gpu; |
7198e6b0 RC |
465 | struct msm_gpu *gpu; |
466 | struct platform_device *pdev = a3xx_pdev; | |
467 | struct adreno_platform_config *config; | |
468 | int ret; | |
469 | ||
470 | if (!pdev) { | |
471 | dev_err(dev->dev, "no a3xx device\n"); | |
472 | ret = -ENXIO; | |
473 | goto fail; | |
474 | } | |
475 | ||
476 | config = pdev->dev.platform_data; | |
477 | ||
478 | a3xx_gpu = kzalloc(sizeof(*a3xx_gpu), GFP_KERNEL); | |
479 | if (!a3xx_gpu) { | |
480 | ret = -ENOMEM; | |
481 | goto fail; | |
482 | } | |
483 | ||
55459968 RC |
484 | adreno_gpu = &a3xx_gpu->base; |
485 | gpu = &adreno_gpu->base; | |
7198e6b0 RC |
486 | |
487 | get_device(&pdev->dev); | |
488 | a3xx_gpu->pdev = pdev; | |
489 | ||
490 | gpu->fast_rate = config->fast_rate; | |
491 | gpu->slow_rate = config->slow_rate; | |
492 | gpu->bus_freq = config->bus_freq; | |
bf2b33af RC |
493 | #ifdef CONFIG_MSM_BUS_SCALING |
494 | gpu->bus_scale_table = config->bus_scale_table; | |
495 | #endif | |
7198e6b0 RC |
496 | |
497 | DBG("fast_rate=%u, slow_rate=%u, bus_freq=%u", | |
498 | gpu->fast_rate, gpu->slow_rate, gpu->bus_freq); | |
499 | ||
55459968 | 500 | ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, config->rev); |
7198e6b0 RC |
501 | if (ret) |
502 | goto fail; | |
503 | ||
55459968 RC |
504 | /* if needed, allocate gmem: */ |
505 | if (adreno_is_a330(adreno_gpu)) { | |
506 | #ifdef CONFIG_MSM_OCMEM | |
507 | /* TODO this is different/missing upstream: */ | |
508 | struct ocmem_buf *ocmem_hdl = | |
509 | ocmem_allocate(OCMEM_GRAPHICS, adreno_gpu->gmem); | |
510 | ||
511 | a3xx_gpu->ocmem_hdl = ocmem_hdl; | |
512 | a3xx_gpu->ocmem_base = ocmem_hdl->addr; | |
513 | adreno_gpu->gmem = ocmem_hdl->len; | |
514 | DBG("using %dK of OCMEM at 0x%08x", adreno_gpu->gmem / 1024, | |
515 | a3xx_gpu->ocmem_base); | |
516 | #endif | |
517 | } | |
518 | ||
871d812a RC |
519 | if (!gpu->mmu) { |
520 | /* TODO we think it is possible to configure the GPU to | |
521 | * restrict access to VRAM carveout. But the required | |
522 | * registers are unknown. For now just bail out and | |
523 | * limp along with just modesetting. If it turns out | |
524 | * to not be possible to restrict access, then we must | |
525 | * implement a cmdstream validator. | |
526 | */ | |
527 | dev_err(dev->dev, "No memory protection without IOMMU\n"); | |
528 | ret = -ENXIO; | |
529 | goto fail; | |
530 | } | |
531 | ||
532 | return gpu; | |
7198e6b0 RC |
533 | |
534 | fail: | |
535 | if (a3xx_gpu) | |
536 | a3xx_destroy(&a3xx_gpu->base.base); | |
537 | ||
538 | return ERR_PTR(ret); | |
539 | } | |
540 | ||
541 | /* | |
542 | * The a3xx device: | |
543 | */ | |
544 | ||
bf2b33af RC |
545 | #if defined(CONFIG_MSM_BUS_SCALING) && !defined(CONFIG_OF) |
546 | # include <mach/kgsl.h> | |
547 | #endif | |
548 | ||
7198e6b0 RC |
549 | static int a3xx_probe(struct platform_device *pdev) |
550 | { | |
551 | static struct adreno_platform_config config = {}; | |
552 | #ifdef CONFIG_OF | |
55459968 RC |
553 | struct device_node *child, *node = pdev->dev.of_node; |
554 | u32 val; | |
555 | int ret; | |
556 | ||
557 | ret = of_property_read_u32(node, "qcom,chipid", &val); | |
558 | if (ret) { | |
559 | dev_err(&pdev->dev, "could not find chipid: %d\n", ret); | |
560 | return ret; | |
561 | } | |
562 | ||
563 | config.rev = ADRENO_REV((val >> 24) & 0xff, | |
564 | (val >> 16) & 0xff, (val >> 8) & 0xff, val & 0xff); | |
565 | ||
566 | /* find clock rates: */ | |
567 | config.fast_rate = 0; | |
568 | config.slow_rate = ~0; | |
569 | for_each_child_of_node(node, child) { | |
570 | if (of_device_is_compatible(child, "qcom,gpu-pwrlevels")) { | |
571 | struct device_node *pwrlvl; | |
572 | for_each_child_of_node(child, pwrlvl) { | |
573 | ret = of_property_read_u32(pwrlvl, "qcom,gpu-freq", &val); | |
574 | if (ret) { | |
575 | dev_err(&pdev->dev, "could not find gpu-freq: %d\n", ret); | |
576 | return ret; | |
577 | } | |
578 | config.fast_rate = max(config.fast_rate, val); | |
579 | config.slow_rate = min(config.slow_rate, val); | |
580 | } | |
581 | } | |
582 | } | |
583 | ||
584 | if (!config.fast_rate) { | |
585 | dev_err(&pdev->dev, "could not find clk rates\n"); | |
586 | return -ENXIO; | |
587 | } | |
588 | ||
7198e6b0 | 589 | #else |
bf2b33af | 590 | struct kgsl_device_platform_data *pdata = pdev->dev.platform_data; |
7198e6b0 RC |
591 | uint32_t version = socinfo_get_version(); |
592 | if (cpu_is_apq8064ab()) { | |
593 | config.fast_rate = 450000000; | |
594 | config.slow_rate = 27000000; | |
595 | config.bus_freq = 4; | |
596 | config.rev = ADRENO_REV(3, 2, 1, 0); | |
e529c7e6 | 597 | } else if (cpu_is_apq8064()) { |
7198e6b0 RC |
598 | config.fast_rate = 400000000; |
599 | config.slow_rate = 27000000; | |
600 | config.bus_freq = 4; | |
601 | ||
602 | if (SOCINFO_VERSION_MAJOR(version) == 2) | |
603 | config.rev = ADRENO_REV(3, 2, 0, 2); | |
604 | else if ((SOCINFO_VERSION_MAJOR(version) == 1) && | |
605 | (SOCINFO_VERSION_MINOR(version) == 1)) | |
606 | config.rev = ADRENO_REV(3, 2, 0, 1); | |
607 | else | |
608 | config.rev = ADRENO_REV(3, 2, 0, 0); | |
609 | ||
e529c7e6 RC |
610 | } else if (cpu_is_msm8960ab()) { |
611 | config.fast_rate = 400000000; | |
612 | config.slow_rate = 320000000; | |
613 | config.bus_freq = 4; | |
614 | ||
615 | if (SOCINFO_VERSION_MINOR(version) == 0) | |
616 | config.rev = ADRENO_REV(3, 2, 1, 0); | |
617 | else | |
618 | config.rev = ADRENO_REV(3, 2, 1, 1); | |
619 | ||
7198e6b0 RC |
620 | } else if (cpu_is_msm8930()) { |
621 | config.fast_rate = 400000000; | |
622 | config.slow_rate = 27000000; | |
623 | config.bus_freq = 3; | |
624 | ||
625 | if ((SOCINFO_VERSION_MAJOR(version) == 1) && | |
626 | (SOCINFO_VERSION_MINOR(version) == 2)) | |
627 | config.rev = ADRENO_REV(3, 0, 5, 2); | |
628 | else | |
629 | config.rev = ADRENO_REV(3, 0, 5, 0); | |
630 | ||
631 | } | |
bf2b33af RC |
632 | # ifdef CONFIG_MSM_BUS_SCALING |
633 | config.bus_scale_table = pdata->bus_scale_table; | |
634 | # endif | |
7198e6b0 RC |
635 | #endif |
636 | pdev->dev.platform_data = &config; | |
637 | a3xx_pdev = pdev; | |
638 | return 0; | |
639 | } | |
640 | ||
641 | static int a3xx_remove(struct platform_device *pdev) | |
642 | { | |
643 | a3xx_pdev = NULL; | |
644 | return 0; | |
645 | } | |
646 | ||
55459968 RC |
647 | static const struct of_device_id dt_match[] = { |
648 | { .compatible = "qcom,kgsl-3d0" }, | |
649 | {} | |
650 | }; | |
651 | MODULE_DEVICE_TABLE(of, dt_match); | |
652 | ||
7198e6b0 RC |
653 | static struct platform_driver a3xx_driver = { |
654 | .probe = a3xx_probe, | |
655 | .remove = a3xx_remove, | |
55459968 RC |
656 | .driver = { |
657 | .name = "kgsl-3d0", | |
658 | .of_match_table = dt_match, | |
659 | }, | |
7198e6b0 RC |
660 | }; |
661 | ||
662 | void __init a3xx_register(void) | |
663 | { | |
664 | platform_driver_register(&a3xx_driver); | |
665 | } | |
666 | ||
667 | void __exit a3xx_unregister(void) | |
668 | { | |
669 | platform_driver_unregister(&a3xx_driver); | |
670 | } |