1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2017-2019 The Linux Foundation. All rights reserved. */
3 
4 
5 #include "msm_gem.h"
6 #include "msm_mmu.h"
7 #include "msm_gpu_trace.h"
8 #include "a6xx_gpu.h"
9 #include "a6xx_gmu.xml.h"
10 
11 #include <linux/bitfield.h>
12 #include <linux/devfreq.h>
13 #include <linux/pm_domain.h>
14 #include <linux/soc/qcom/llcc-qcom.h>
15 
16 #define GPU_PAS_ID 13
17 
_a6xx_check_idle(struct msm_gpu * gpu)18 static inline bool _a6xx_check_idle(struct msm_gpu *gpu)
19 {
20 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
21 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
22 
23 	/* Check that the GMU is idle */
24 	if (!adreno_has_gmu_wrapper(adreno_gpu) && !a6xx_gmu_isidle(&a6xx_gpu->gmu))
25 		return false;
26 
27 	/* Check tha the CX master is idle */
28 	if (gpu_read(gpu, REG_A6XX_RBBM_STATUS) &
29 			~A6XX_RBBM_STATUS_CP_AHB_BUSY_CX_MASTER)
30 		return false;
31 
32 	return !(gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS) &
33 		A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT);
34 }
35 
a6xx_idle(struct msm_gpu * gpu,struct msm_ringbuffer * ring)36 static bool a6xx_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
37 {
38 	/* wait for CP to drain ringbuffer: */
39 	if (!adreno_idle(gpu, ring))
40 		return false;
41 
42 	if (spin_until(_a6xx_check_idle(gpu))) {
43 		DRM_ERROR("%s: %ps: timeout waiting for GPU to idle: status %8.8X irq %8.8X rptr/wptr %d/%d\n",
44 			gpu->name, __builtin_return_address(0),
45 			gpu_read(gpu, REG_A6XX_RBBM_STATUS),
46 			gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS),
47 			gpu_read(gpu, REG_A6XX_CP_RB_RPTR),
48 			gpu_read(gpu, REG_A6XX_CP_RB_WPTR));
49 		return false;
50 	}
51 
52 	return true;
53 }
54 
update_shadow_rptr(struct msm_gpu * gpu,struct msm_ringbuffer * ring)55 static void update_shadow_rptr(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
56 {
57 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
58 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
59 
60 	/* Expanded APRIV doesn't need to issue the WHERE_AM_I opcode */
61 	if (a6xx_gpu->has_whereami && !adreno_gpu->base.hw_apriv) {
62 		OUT_PKT7(ring, CP_WHERE_AM_I, 2);
63 		OUT_RING(ring, lower_32_bits(shadowptr(a6xx_gpu, ring)));
64 		OUT_RING(ring, upper_32_bits(shadowptr(a6xx_gpu, ring)));
65 	}
66 }
67 
a6xx_flush(struct msm_gpu * gpu,struct msm_ringbuffer * ring)68 static void a6xx_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
69 {
70 	uint32_t wptr;
71 	unsigned long flags;
72 
73 	update_shadow_rptr(gpu, ring);
74 
75 	spin_lock_irqsave(&ring->preempt_lock, flags);
76 
77 	/* Copy the shadow to the actual register */
78 	ring->cur = ring->next;
79 
80 	/* Make sure to wrap wptr if we need to */
81 	wptr = get_wptr(ring);
82 
83 	spin_unlock_irqrestore(&ring->preempt_lock, flags);
84 
85 	/* Make sure everything is posted before making a decision */
86 	mb();
87 
88 	gpu_write(gpu, REG_A6XX_CP_RB_WPTR, wptr);
89 }
90 
get_stats_counter(struct msm_ringbuffer * ring,u32 counter,u64 iova)91 static void get_stats_counter(struct msm_ringbuffer *ring, u32 counter,
92 		u64 iova)
93 {
94 	OUT_PKT7(ring, CP_REG_TO_MEM, 3);
95 	OUT_RING(ring, CP_REG_TO_MEM_0_REG(counter) |
96 		CP_REG_TO_MEM_0_CNT(2) |
97 		CP_REG_TO_MEM_0_64B);
98 	OUT_RING(ring, lower_32_bits(iova));
99 	OUT_RING(ring, upper_32_bits(iova));
100 }
101 
a6xx_set_pagetable(struct a6xx_gpu * a6xx_gpu,struct msm_ringbuffer * ring,struct msm_file_private * ctx)102 static void a6xx_set_pagetable(struct a6xx_gpu *a6xx_gpu,
103 		struct msm_ringbuffer *ring, struct msm_file_private *ctx)
104 {
105 	bool sysprof = refcount_read(&a6xx_gpu->base.base.sysprof_active) > 1;
106 	phys_addr_t ttbr;
107 	u32 asid;
108 	u64 memptr = rbmemptr(ring, ttbr0);
109 
110 	if (ctx->seqno == a6xx_gpu->base.base.cur_ctx_seqno)
111 		return;
112 
113 	if (msm_iommu_pagetable_params(ctx->aspace->mmu, &ttbr, &asid))
114 		return;
115 
116 	if (!sysprof) {
117 		/* Turn off protected mode to write to special registers */
118 		OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1);
119 		OUT_RING(ring, 0);
120 
121 		OUT_PKT4(ring, REG_A6XX_RBBM_PERFCTR_SRAM_INIT_CMD, 1);
122 		OUT_RING(ring, 1);
123 	}
124 
125 	/* Execute the table update */
126 	OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
127 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_0_TTBR0_LO(lower_32_bits(ttbr)));
128 
129 	OUT_RING(ring,
130 		CP_SMMU_TABLE_UPDATE_1_TTBR0_HI(upper_32_bits(ttbr)) |
131 		CP_SMMU_TABLE_UPDATE_1_ASID(asid));
132 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_2_CONTEXTIDR(0));
133 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_3_CONTEXTBANK(0));
134 
135 	/*
136 	 * Write the new TTBR0 to the memstore. This is good for debugging.
137 	 */
138 	OUT_PKT7(ring, CP_MEM_WRITE, 4);
139 	OUT_RING(ring, CP_MEM_WRITE_0_ADDR_LO(lower_32_bits(memptr)));
140 	OUT_RING(ring, CP_MEM_WRITE_1_ADDR_HI(upper_32_bits(memptr)));
141 	OUT_RING(ring, lower_32_bits(ttbr));
142 	OUT_RING(ring, (asid << 16) | upper_32_bits(ttbr));
143 
144 	/*
145 	 * And finally, trigger a uche flush to be sure there isn't anything
146 	 * lingering in that part of the GPU
147 	 */
148 
149 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
150 	OUT_RING(ring, CACHE_INVALIDATE);
151 
152 	if (!sysprof) {
153 		/*
154 		 * Wait for SRAM clear after the pgtable update, so the
155 		 * two can happen in parallel:
156 		 */
157 		OUT_PKT7(ring, CP_WAIT_REG_MEM, 6);
158 		OUT_RING(ring, CP_WAIT_REG_MEM_0_FUNCTION(WRITE_EQ));
159 		OUT_RING(ring, CP_WAIT_REG_MEM_1_POLL_ADDR_LO(
160 				REG_A6XX_RBBM_PERFCTR_SRAM_INIT_STATUS));
161 		OUT_RING(ring, CP_WAIT_REG_MEM_2_POLL_ADDR_HI(0));
162 		OUT_RING(ring, CP_WAIT_REG_MEM_3_REF(0x1));
163 		OUT_RING(ring, CP_WAIT_REG_MEM_4_MASK(0x1));
164 		OUT_RING(ring, CP_WAIT_REG_MEM_5_DELAY_LOOP_CYCLES(0));
165 
166 		/* Re-enable protected mode: */
167 		OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1);
168 		OUT_RING(ring, 1);
169 	}
170 }
171 
a6xx_submit(struct msm_gpu * gpu,struct msm_gem_submit * submit)172 static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit)
173 {
174 	unsigned int index = submit->seqno % MSM_GPU_SUBMIT_STATS_COUNT;
175 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
176 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
177 	struct msm_ringbuffer *ring = submit->ring;
178 	unsigned int i, ibs = 0;
179 
180 	a6xx_set_pagetable(a6xx_gpu, ring, submit->queue->ctx);
181 
182 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP(0),
183 		rbmemptr_stats(ring, index, cpcycles_start));
184 
185 	/*
186 	 * For PM4 the GMU register offsets are calculated from the base of the
187 	 * GPU registers so we need to add 0x1a800 to the register value on A630
188 	 * to get the right value from PM4.
189 	 */
190 	get_stats_counter(ring, REG_A6XX_CP_ALWAYS_ON_COUNTER,
191 		rbmemptr_stats(ring, index, alwayson_start));
192 
193 	/* Invalidate CCU depth and color */
194 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
195 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(PC_CCU_INVALIDATE_DEPTH));
196 
197 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
198 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(PC_CCU_INVALIDATE_COLOR));
199 
200 	/* Submit the commands */
201 	for (i = 0; i < submit->nr_cmds; i++) {
202 		switch (submit->cmd[i].type) {
203 		case MSM_SUBMIT_CMD_IB_TARGET_BUF:
204 			break;
205 		case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
206 			if (gpu->cur_ctx_seqno == submit->queue->ctx->seqno)
207 				break;
208 			fallthrough;
209 		case MSM_SUBMIT_CMD_BUF:
210 			OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3);
211 			OUT_RING(ring, lower_32_bits(submit->cmd[i].iova));
212 			OUT_RING(ring, upper_32_bits(submit->cmd[i].iova));
213 			OUT_RING(ring, submit->cmd[i].size);
214 			ibs++;
215 			break;
216 		}
217 
218 		/*
219 		 * Periodically update shadow-wptr if needed, so that we
220 		 * can see partial progress of submits with large # of
221 		 * cmds.. otherwise we could needlessly stall waiting for
222 		 * ringbuffer state, simply due to looking at a shadow
223 		 * rptr value that has not been updated
224 		 */
225 		if ((ibs % 32) == 0)
226 			update_shadow_rptr(gpu, ring);
227 	}
228 
229 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP(0),
230 		rbmemptr_stats(ring, index, cpcycles_end));
231 	get_stats_counter(ring, REG_A6XX_CP_ALWAYS_ON_COUNTER,
232 		rbmemptr_stats(ring, index, alwayson_end));
233 
234 	/* Write the fence to the scratch register */
235 	OUT_PKT4(ring, REG_A6XX_CP_SCRATCH_REG(2), 1);
236 	OUT_RING(ring, submit->seqno);
237 
238 	/*
239 	 * Execute a CACHE_FLUSH_TS event. This will ensure that the
240 	 * timestamp is written to the memory and then triggers the interrupt
241 	 */
242 	OUT_PKT7(ring, CP_EVENT_WRITE, 4);
243 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(CACHE_FLUSH_TS) |
244 		CP_EVENT_WRITE_0_IRQ);
245 	OUT_RING(ring, lower_32_bits(rbmemptr(ring, fence)));
246 	OUT_RING(ring, upper_32_bits(rbmemptr(ring, fence)));
247 	OUT_RING(ring, submit->seqno);
248 
249 	trace_msm_gpu_submit_flush(submit,
250 		gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER));
251 
252 	a6xx_flush(gpu, ring);
253 }
254 
255 const struct adreno_reglist a612_hwcg[] = {
256 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222},
257 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
258 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000081},
259 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000f3cf},
260 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222222},
261 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
262 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
263 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
264 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
265 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
266 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
267 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
268 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
269 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
270 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
271 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
272 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
273 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01202222},
274 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
275 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040f00},
276 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05522022},
277 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
278 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
279 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
280 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
281 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
282 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x02222222},
283 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
284 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
285 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
286 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
287 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
288 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
289 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
290 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
291 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
292 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
293 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
294 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
295 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
296 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
297 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
298 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
299 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
300 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
301 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
302 	{},
303 };
304 
305 /* For a615 family (a615, a616, a618 and a619) */
306 const struct adreno_reglist a615_hwcg[] = {
307 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0,  0x02222222},
308 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
309 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
310 	{REG_A6XX_RBBM_CLOCK_HYST_SP0,  0x0000F3CF},
311 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0,  0x02222222},
312 	{REG_A6XX_RBBM_CLOCK_CNTL_TP1,  0x02222222},
313 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
314 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP1, 0x22222222},
315 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
316 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP1, 0x22222222},
317 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
318 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP1, 0x00022222},
319 	{REG_A6XX_RBBM_CLOCK_HYST_TP0,  0x77777777},
320 	{REG_A6XX_RBBM_CLOCK_HYST_TP1,  0x77777777},
321 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
322 	{REG_A6XX_RBBM_CLOCK_HYST2_TP1, 0x77777777},
323 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
324 	{REG_A6XX_RBBM_CLOCK_HYST3_TP1, 0x77777777},
325 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
326 	{REG_A6XX_RBBM_CLOCK_HYST4_TP1, 0x00077777},
327 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
328 	{REG_A6XX_RBBM_CLOCK_DELAY_TP1, 0x11111111},
329 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
330 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP1, 0x11111111},
331 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
332 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP1, 0x11111111},
333 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
334 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP1, 0x00011111},
335 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE,  0x22222222},
336 	{REG_A6XX_RBBM_CLOCK_CNTL2_UCHE, 0x22222222},
337 	{REG_A6XX_RBBM_CLOCK_CNTL3_UCHE, 0x22222222},
338 	{REG_A6XX_RBBM_CLOCK_CNTL4_UCHE, 0x00222222},
339 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE,  0x00000004},
340 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
341 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
342 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x00002222},
343 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002020},
344 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU1, 0x00002220},
345 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU2, 0x00002220},
346 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU3, 0x00002220},
347 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
348 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU1, 0x00040F00},
349 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU2, 0x00040F00},
350 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU3, 0x00040F00},
351 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05022022},
352 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
353 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
354 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
355 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
356 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
357 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
358 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
359 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
360 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
361 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
362 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
363 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
364 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
365 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
366 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
367 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
368 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
369 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
370 	{},
371 };
372 
373 const struct adreno_reglist a630_hwcg[] = {
374 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222},
375 	{REG_A6XX_RBBM_CLOCK_CNTL_SP1, 0x22222222},
376 	{REG_A6XX_RBBM_CLOCK_CNTL_SP2, 0x22222222},
377 	{REG_A6XX_RBBM_CLOCK_CNTL_SP3, 0x22222222},
378 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02022220},
379 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP1, 0x02022220},
380 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP2, 0x02022220},
381 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP3, 0x02022220},
382 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
383 	{REG_A6XX_RBBM_CLOCK_DELAY_SP1, 0x00000080},
384 	{REG_A6XX_RBBM_CLOCK_DELAY_SP2, 0x00000080},
385 	{REG_A6XX_RBBM_CLOCK_DELAY_SP3, 0x00000080},
386 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000f3cf},
387 	{REG_A6XX_RBBM_CLOCK_HYST_SP1, 0x0000f3cf},
388 	{REG_A6XX_RBBM_CLOCK_HYST_SP2, 0x0000f3cf},
389 	{REG_A6XX_RBBM_CLOCK_HYST_SP3, 0x0000f3cf},
390 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
391 	{REG_A6XX_RBBM_CLOCK_CNTL_TP1, 0x02222222},
392 	{REG_A6XX_RBBM_CLOCK_CNTL_TP2, 0x02222222},
393 	{REG_A6XX_RBBM_CLOCK_CNTL_TP3, 0x02222222},
394 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
395 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP1, 0x22222222},
396 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP2, 0x22222222},
397 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP3, 0x22222222},
398 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
399 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP1, 0x22222222},
400 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP2, 0x22222222},
401 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP3, 0x22222222},
402 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
403 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP1, 0x00022222},
404 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP2, 0x00022222},
405 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP3, 0x00022222},
406 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
407 	{REG_A6XX_RBBM_CLOCK_HYST_TP1, 0x77777777},
408 	{REG_A6XX_RBBM_CLOCK_HYST_TP2, 0x77777777},
409 	{REG_A6XX_RBBM_CLOCK_HYST_TP3, 0x77777777},
410 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
411 	{REG_A6XX_RBBM_CLOCK_HYST2_TP1, 0x77777777},
412 	{REG_A6XX_RBBM_CLOCK_HYST2_TP2, 0x77777777},
413 	{REG_A6XX_RBBM_CLOCK_HYST2_TP3, 0x77777777},
414 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
415 	{REG_A6XX_RBBM_CLOCK_HYST3_TP1, 0x77777777},
416 	{REG_A6XX_RBBM_CLOCK_HYST3_TP2, 0x77777777},
417 	{REG_A6XX_RBBM_CLOCK_HYST3_TP3, 0x77777777},
418 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
419 	{REG_A6XX_RBBM_CLOCK_HYST4_TP1, 0x00077777},
420 	{REG_A6XX_RBBM_CLOCK_HYST4_TP2, 0x00077777},
421 	{REG_A6XX_RBBM_CLOCK_HYST4_TP3, 0x00077777},
422 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
423 	{REG_A6XX_RBBM_CLOCK_DELAY_TP1, 0x11111111},
424 	{REG_A6XX_RBBM_CLOCK_DELAY_TP2, 0x11111111},
425 	{REG_A6XX_RBBM_CLOCK_DELAY_TP3, 0x11111111},
426 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
427 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP1, 0x11111111},
428 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP2, 0x11111111},
429 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP3, 0x11111111},
430 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
431 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP1, 0x11111111},
432 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP2, 0x11111111},
433 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP3, 0x11111111},
434 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
435 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP1, 0x00011111},
436 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP2, 0x00011111},
437 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP3, 0x00011111},
438 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
439 	{REG_A6XX_RBBM_CLOCK_CNTL2_UCHE, 0x22222222},
440 	{REG_A6XX_RBBM_CLOCK_CNTL3_UCHE, 0x22222222},
441 	{REG_A6XX_RBBM_CLOCK_CNTL4_UCHE, 0x00222222},
442 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
443 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
444 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
445 	{REG_A6XX_RBBM_CLOCK_CNTL_RB1, 0x22222222},
446 	{REG_A6XX_RBBM_CLOCK_CNTL_RB2, 0x22222222},
447 	{REG_A6XX_RBBM_CLOCK_CNTL_RB3, 0x22222222},
448 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x00002222},
449 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB1, 0x00002222},
450 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB2, 0x00002222},
451 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB3, 0x00002222},
452 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
453 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU1, 0x00002220},
454 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU2, 0x00002220},
455 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU3, 0x00002220},
456 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040f00},
457 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU1, 0x00040f00},
458 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU2, 0x00040f00},
459 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU3, 0x00040f00},
460 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05022022},
461 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
462 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
463 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
464 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
465 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
466 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
467 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
468 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
469 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
470 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
471 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
472 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
473 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
474 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
475 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
476 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
477 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
478 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
479 	{},
480 };
481 
482 const struct adreno_reglist a640_hwcg[] = {
483 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
484 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
485 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
486 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
487 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
488 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
489 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
490 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
491 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
492 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
493 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
494 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
495 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
496 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
497 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
498 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
499 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
500 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
501 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
502 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
503 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05222022},
504 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
505 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
506 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
507 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
508 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
509 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
510 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
511 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
512 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
513 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
514 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
515 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
516 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
517 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
518 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
519 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
520 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
521 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
522 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000000},
523 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
524 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
525 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
526 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
527 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
528 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
529 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
530 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
531 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
532 	{},
533 };
534 
535 const struct adreno_reglist a650_hwcg[] = {
536 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
537 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
538 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
539 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
540 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
541 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
542 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
543 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
544 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
545 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
546 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
547 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
548 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
549 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
550 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
551 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
552 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
553 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
554 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
555 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
556 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022},
557 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
558 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
559 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
560 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
561 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
562 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
563 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
564 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
565 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
566 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
567 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
568 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
569 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
570 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
571 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
572 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
573 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
574 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
575 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000777},
576 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
577 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
578 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
579 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
580 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
581 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
582 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
583 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
584 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
585 	{},
586 };
587 
588 const struct adreno_reglist a660_hwcg[] = {
589 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
590 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
591 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
592 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
593 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222222},
594 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
595 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
596 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
597 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
598 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
599 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
600 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
601 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
602 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
603 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
604 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
605 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
606 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
607 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
608 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
609 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022},
610 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
611 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
612 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
613 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
614 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
615 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
616 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
617 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
618 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
619 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
620 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
621 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
622 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
623 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
624 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
625 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
626 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
627 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
628 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000000},
629 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
630 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
631 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
632 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
633 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
634 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
635 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
636 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
637 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
638 	{},
639 };
640 
641 const struct adreno_reglist a690_hwcg[] = {
642 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
643 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
644 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
645 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
646 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222222},
647 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
648 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
649 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
650 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
651 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
652 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
653 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
654 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
655 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
656 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
657 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
658 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
659 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
660 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
661 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
662 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022},
663 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
664 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
665 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
666 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
667 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
668 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
669 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
670 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
671 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
672 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
673 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
674 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
675 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
676 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
677 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
678 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
679 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
680 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
681 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000000},
682 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
683 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
684 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
685 	{REG_A6XX_RBBM_CLOCK_CNTL, 0x8AA8AA82},
686 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
687 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
688 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
689 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
690 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
691 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
692 	{REG_A6XX_GPU_GMU_AO_GMU_CGC_MODE_CNTL, 0x20200},
693 	{REG_A6XX_GPU_GMU_AO_GMU_CGC_DELAY_CNTL, 0x10111},
694 	{REG_A6XX_GPU_GMU_AO_GMU_CGC_HYST_CNTL, 0x5555},
695 	{}
696 };
697 
a6xx_set_hwcg(struct msm_gpu * gpu,bool state)698 static void a6xx_set_hwcg(struct msm_gpu *gpu, bool state)
699 {
700 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
701 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
702 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
703 	const struct adreno_reglist *reg;
704 	unsigned int i;
705 	u32 val, clock_cntl_on;
706 
707 	if (!adreno_gpu->info->hwcg)
708 		return;
709 
710 	if (adreno_is_a630(adreno_gpu))
711 		clock_cntl_on = 0x8aa8aa02;
712 	else if (adreno_is_a610(adreno_gpu))
713 		clock_cntl_on = 0xaaa8aa82;
714 	else
715 		clock_cntl_on = 0x8aa8aa82;
716 
717 	val = gpu_read(gpu, REG_A6XX_RBBM_CLOCK_CNTL);
718 
719 	/* Don't re-program the registers if they are already correct */
720 	if ((!state && !val) || (state && (val == clock_cntl_on)))
721 		return;
722 
723 	/* Disable SP clock before programming HWCG registers */
724 	if (!adreno_is_a610(adreno_gpu))
725 		gmu_rmw(gmu, REG_A6XX_GPU_GMU_GX_SPTPRAC_CLOCK_CONTROL, 1, 0);
726 
727 	for (i = 0; (reg = &adreno_gpu->info->hwcg[i], reg->offset); i++)
728 		gpu_write(gpu, reg->offset, state ? reg->value : 0);
729 
730 	/* Enable SP clock */
731 	if (!adreno_is_a610(adreno_gpu))
732 		gmu_rmw(gmu, REG_A6XX_GPU_GMU_GX_SPTPRAC_CLOCK_CONTROL, 0, 1);
733 
734 	gpu_write(gpu, REG_A6XX_RBBM_CLOCK_CNTL, state ? clock_cntl_on : 0);
735 }
736 
737 /* For a615, a616, a618, a619, a630, a640 and a680 */
738 static const u32 a6xx_protect[] = {
739 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
740 	A6XX_PROTECT_RDONLY(0x00501, 0x0005),
741 	A6XX_PROTECT_RDONLY(0x0050b, 0x02f4),
742 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
743 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
744 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
745 	A6XX_PROTECT_NORDWR(0x00800, 0x0082),
746 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
747 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
748 	A6XX_PROTECT_RDONLY(0x008de, 0x00ae),
749 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
750 	A6XX_PROTECT_NORDWR(0x0098d, 0x0272),
751 	A6XX_PROTECT_NORDWR(0x00e00, 0x0001),
752 	A6XX_PROTECT_NORDWR(0x00e03, 0x000c),
753 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
754 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
755 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
756 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
757 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
758 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
759 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
760 	A6XX_PROTECT_NORDWR(0x09e70, 0x0001),
761 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
762 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
763 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
764 	A6XX_PROTECT_NORDWR(0x0ae50, 0x032f),
765 	A6XX_PROTECT_NORDWR(0x0b604, 0x0000),
766 	A6XX_PROTECT_NORDWR(0x0be02, 0x0001),
767 	A6XX_PROTECT_NORDWR(0x0be20, 0x17df),
768 	A6XX_PROTECT_NORDWR(0x0f000, 0x0bff),
769 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
770 	A6XX_PROTECT_NORDWR(0x11c00, 0x0000), /* note: infinite range */
771 };
772 
773 /* These are for a620 and a650 */
774 static const u32 a650_protect[] = {
775 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
776 	A6XX_PROTECT_RDONLY(0x00501, 0x0005),
777 	A6XX_PROTECT_RDONLY(0x0050b, 0x02f4),
778 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
779 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
780 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
781 	A6XX_PROTECT_NORDWR(0x00800, 0x0082),
782 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
783 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
784 	A6XX_PROTECT_RDONLY(0x008de, 0x00ae),
785 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
786 	A6XX_PROTECT_NORDWR(0x0098d, 0x0272),
787 	A6XX_PROTECT_NORDWR(0x00e00, 0x0001),
788 	A6XX_PROTECT_NORDWR(0x00e03, 0x000c),
789 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
790 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
791 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
792 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
793 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
794 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
795 	A6XX_PROTECT_NORDWR(0x08e80, 0x027f),
796 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
797 	A6XX_PROTECT_NORDWR(0x09e60, 0x0011),
798 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
799 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
800 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
801 	A6XX_PROTECT_NORDWR(0x0ae50, 0x032f),
802 	A6XX_PROTECT_NORDWR(0x0b604, 0x0000),
803 	A6XX_PROTECT_NORDWR(0x0b608, 0x0007),
804 	A6XX_PROTECT_NORDWR(0x0be02, 0x0001),
805 	A6XX_PROTECT_NORDWR(0x0be20, 0x17df),
806 	A6XX_PROTECT_NORDWR(0x0f000, 0x0bff),
807 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
808 	A6XX_PROTECT_NORDWR(0x18400, 0x1fff),
809 	A6XX_PROTECT_NORDWR(0x1a800, 0x1fff),
810 	A6XX_PROTECT_NORDWR(0x1f400, 0x0443),
811 	A6XX_PROTECT_RDONLY(0x1f844, 0x007b),
812 	A6XX_PROTECT_NORDWR(0x1f887, 0x001b),
813 	A6XX_PROTECT_NORDWR(0x1f8c0, 0x0000), /* note: infinite range */
814 };
815 
816 /* These are for a635 and a660 */
817 static const u32 a660_protect[] = {
818 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
819 	A6XX_PROTECT_RDONLY(0x00501, 0x0005),
820 	A6XX_PROTECT_RDONLY(0x0050b, 0x02f4),
821 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
822 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
823 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
824 	A6XX_PROTECT_NORDWR(0x00800, 0x0082),
825 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
826 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
827 	A6XX_PROTECT_RDONLY(0x008de, 0x00ae),
828 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
829 	A6XX_PROTECT_NORDWR(0x0098d, 0x0272),
830 	A6XX_PROTECT_NORDWR(0x00e00, 0x0001),
831 	A6XX_PROTECT_NORDWR(0x00e03, 0x000c),
832 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
833 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
834 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
835 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
836 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
837 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
838 	A6XX_PROTECT_NORDWR(0x08e80, 0x027f),
839 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
840 	A6XX_PROTECT_NORDWR(0x09e60, 0x0011),
841 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
842 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
843 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
844 	A6XX_PROTECT_NORDWR(0x0ae50, 0x012f),
845 	A6XX_PROTECT_NORDWR(0x0b604, 0x0000),
846 	A6XX_PROTECT_NORDWR(0x0b608, 0x0006),
847 	A6XX_PROTECT_NORDWR(0x0be02, 0x0001),
848 	A6XX_PROTECT_NORDWR(0x0be20, 0x015f),
849 	A6XX_PROTECT_NORDWR(0x0d000, 0x05ff),
850 	A6XX_PROTECT_NORDWR(0x0f000, 0x0bff),
851 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
852 	A6XX_PROTECT_NORDWR(0x18400, 0x1fff),
853 	A6XX_PROTECT_NORDWR(0x1a400, 0x1fff),
854 	A6XX_PROTECT_NORDWR(0x1f400, 0x0443),
855 	A6XX_PROTECT_RDONLY(0x1f844, 0x007b),
856 	A6XX_PROTECT_NORDWR(0x1f860, 0x0000),
857 	A6XX_PROTECT_NORDWR(0x1f887, 0x001b),
858 	A6XX_PROTECT_NORDWR(0x1f8c0, 0x0000), /* note: infinite range */
859 };
860 
861 /* These are for a690 */
862 static const u32 a690_protect[] = {
863 	A6XX_PROTECT_RDONLY(0x00000, 0x004ff),
864 	A6XX_PROTECT_RDONLY(0x00501, 0x00001),
865 	A6XX_PROTECT_RDONLY(0x0050b, 0x002f4),
866 	A6XX_PROTECT_NORDWR(0x0050e, 0x00000),
867 	A6XX_PROTECT_NORDWR(0x00510, 0x00000),
868 	A6XX_PROTECT_NORDWR(0x00534, 0x00000),
869 	A6XX_PROTECT_NORDWR(0x00800, 0x00082),
870 	A6XX_PROTECT_NORDWR(0x008a0, 0x00008),
871 	A6XX_PROTECT_NORDWR(0x008ab, 0x00024),
872 	A6XX_PROTECT_RDONLY(0x008de, 0x000ae),
873 	A6XX_PROTECT_NORDWR(0x00900, 0x0004d),
874 	A6XX_PROTECT_NORDWR(0x0098d, 0x00272),
875 	A6XX_PROTECT_NORDWR(0x00e00, 0x00001),
876 	A6XX_PROTECT_NORDWR(0x00e03, 0x0000c),
877 	A6XX_PROTECT_NORDWR(0x03c00, 0x000c3),
878 	A6XX_PROTECT_RDONLY(0x03cc4, 0x01fff),
879 	A6XX_PROTECT_NORDWR(0x08630, 0x001cf),
880 	A6XX_PROTECT_NORDWR(0x08e00, 0x00000),
881 	A6XX_PROTECT_NORDWR(0x08e08, 0x00007),
882 	A6XX_PROTECT_NORDWR(0x08e50, 0x0001f),
883 	A6XX_PROTECT_NORDWR(0x08e80, 0x0027f),
884 	A6XX_PROTECT_NORDWR(0x09624, 0x001db),
885 	A6XX_PROTECT_NORDWR(0x09e60, 0x00011),
886 	A6XX_PROTECT_NORDWR(0x09e78, 0x00187),
887 	A6XX_PROTECT_NORDWR(0x0a630, 0x001cf),
888 	A6XX_PROTECT_NORDWR(0x0ae02, 0x00000),
889 	A6XX_PROTECT_NORDWR(0x0ae50, 0x0012f),
890 	A6XX_PROTECT_NORDWR(0x0b604, 0x00000),
891 	A6XX_PROTECT_NORDWR(0x0b608, 0x00006),
892 	A6XX_PROTECT_NORDWR(0x0be02, 0x00001),
893 	A6XX_PROTECT_NORDWR(0x0be20, 0x0015f),
894 	A6XX_PROTECT_NORDWR(0x0d000, 0x005ff),
895 	A6XX_PROTECT_NORDWR(0x0f000, 0x00bff),
896 	A6XX_PROTECT_RDONLY(0x0fc00, 0x01fff),
897 	A6XX_PROTECT_NORDWR(0x11c00, 0x00000), /*note: infiite range */
898 };
899 
a6xx_set_cp_protect(struct msm_gpu * gpu)900 static void a6xx_set_cp_protect(struct msm_gpu *gpu)
901 {
902 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
903 	const u32 *regs = a6xx_protect;
904 	unsigned i, count, count_max;
905 
906 	if (adreno_is_a650(adreno_gpu)) {
907 		regs = a650_protect;
908 		count = ARRAY_SIZE(a650_protect);
909 		count_max = 48;
910 		BUILD_BUG_ON(ARRAY_SIZE(a650_protect) > 48);
911 	} else if (adreno_is_a690(adreno_gpu)) {
912 		regs = a690_protect;
913 		count = ARRAY_SIZE(a690_protect);
914 		count_max = 48;
915 		BUILD_BUG_ON(ARRAY_SIZE(a690_protect) > 48);
916 	} else if (adreno_is_a660_family(adreno_gpu)) {
917 		regs = a660_protect;
918 		count = ARRAY_SIZE(a660_protect);
919 		count_max = 48;
920 		BUILD_BUG_ON(ARRAY_SIZE(a660_protect) > 48);
921 	} else {
922 		regs = a6xx_protect;
923 		count = ARRAY_SIZE(a6xx_protect);
924 		count_max = 32;
925 		BUILD_BUG_ON(ARRAY_SIZE(a6xx_protect) > 32);
926 	}
927 
928 	/*
929 	 * Enable access protection to privileged registers, fault on an access
930 	 * protect violation and select the last span to protect from the start
931 	 * address all the way to the end of the register address space
932 	 */
933 	gpu_write(gpu, REG_A6XX_CP_PROTECT_CNTL,
934 		  A6XX_CP_PROTECT_CNTL_ACCESS_PROT_EN |
935 		  A6XX_CP_PROTECT_CNTL_ACCESS_FAULT_ON_VIOL_EN |
936 		  A6XX_CP_PROTECT_CNTL_LAST_SPAN_INF_RANGE);
937 
938 	for (i = 0; i < count - 1; i++) {
939 		/* Intentionally skip writing to some registers */
940 		if (regs[i])
941 			gpu_write(gpu, REG_A6XX_CP_PROTECT(i), regs[i]);
942 	}
943 	/* last CP_PROTECT to have "infinite" length on the last entry */
944 	gpu_write(gpu, REG_A6XX_CP_PROTECT(count_max - 1), regs[i]);
945 }
946 
a6xx_set_ubwc_config(struct msm_gpu * gpu)947 static void a6xx_set_ubwc_config(struct msm_gpu *gpu)
948 {
949 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
950 	/* Unknown, introduced with A650 family, related to UBWC mode/ver 4 */
951 	u32 rgb565_predicator = 0;
952 	/* Unknown, introduced with A650 family */
953 	u32 uavflagprd_inv = 0;
954 	/* Whether the minimum access length is 64 bits */
955 	u32 min_acc_len = 0;
956 	/* Entirely magic, per-GPU-gen value */
957 	u32 ubwc_mode = 0;
958 	/*
959 	 * The Highest Bank Bit value represents the bit of the highest DDR bank.
960 	 * We then subtract 13 from it (13 is the minimum value allowed by hw) and
961 	 * write the lowest two bits of the remaining value as hbb_lo and the
962 	 * one above it as hbb_hi to the hardware. This should ideally use DRAM
963 	 * type detection.
964 	 */
965 	u32 hbb_hi = 0;
966 	u32 hbb_lo = 2;
967 	/* Unknown, introduced with A640/680 */
968 	u32 amsbc = 0;
969 
970 	if (adreno_is_a610(adreno_gpu)) {
971 		/* HBB = 14 */
972 		hbb_lo = 1;
973 		min_acc_len = 1;
974 		ubwc_mode = 1;
975 	}
976 
977 	/* a618 is using the hw default values */
978 	if (adreno_is_a618(adreno_gpu))
979 		return;
980 
981 	if (adreno_is_a619_holi(adreno_gpu))
982 		hbb_lo = 0;
983 
984 	if (adreno_is_a640_family(adreno_gpu))
985 		amsbc = 1;
986 
987 	if (adreno_is_a650(adreno_gpu) || adreno_is_a660(adreno_gpu)) {
988 		/* TODO: get ddr type from bootloader and use 2 for LPDDR4 */
989 		hbb_lo = 3;
990 		amsbc = 1;
991 		rgb565_predicator = 1;
992 		uavflagprd_inv = 2;
993 	}
994 
995 	if (adreno_is_a690(adreno_gpu)) {
996 		hbb_lo = 2;
997 		amsbc = 1;
998 		rgb565_predicator = 1;
999 		uavflagprd_inv = 2;
1000 	}
1001 
1002 	if (adreno_is_7c3(adreno_gpu)) {
1003 		hbb_lo = 1;
1004 		amsbc = 1;
1005 		rgb565_predicator = 1;
1006 		uavflagprd_inv = 2;
1007 	}
1008 
1009 	gpu_write(gpu, REG_A6XX_RB_NC_MODE_CNTL,
1010 		  rgb565_predicator << 11 | hbb_hi << 10 | amsbc << 4 |
1011 		  min_acc_len << 3 | hbb_lo << 1 | ubwc_mode);
1012 
1013 	gpu_write(gpu, REG_A6XX_TPL1_NC_MODE_CNTL, hbb_hi << 4 |
1014 		  min_acc_len << 3 | hbb_lo << 1 | ubwc_mode);
1015 
1016 	gpu_write(gpu, REG_A6XX_SP_NC_MODE_CNTL, hbb_hi << 10 |
1017 		  uavflagprd_inv << 4 | min_acc_len << 3 |
1018 		  hbb_lo << 1 | ubwc_mode);
1019 
1020 	gpu_write(gpu, REG_A6XX_UCHE_MODE_CNTL, min_acc_len << 23 | hbb_lo << 21);
1021 }
1022 
a6xx_cp_init(struct msm_gpu * gpu)1023 static int a6xx_cp_init(struct msm_gpu *gpu)
1024 {
1025 	struct msm_ringbuffer *ring = gpu->rb[0];
1026 
1027 	OUT_PKT7(ring, CP_ME_INIT, 8);
1028 
1029 	OUT_RING(ring, 0x0000002f);
1030 
1031 	/* Enable multiple hardware contexts */
1032 	OUT_RING(ring, 0x00000003);
1033 
1034 	/* Enable error detection */
1035 	OUT_RING(ring, 0x20000000);
1036 
1037 	/* Don't enable header dump */
1038 	OUT_RING(ring, 0x00000000);
1039 	OUT_RING(ring, 0x00000000);
1040 
1041 	/* No workarounds enabled */
1042 	OUT_RING(ring, 0x00000000);
1043 
1044 	/* Pad rest of the cmds with 0's */
1045 	OUT_RING(ring, 0x00000000);
1046 	OUT_RING(ring, 0x00000000);
1047 
1048 	a6xx_flush(gpu, ring);
1049 	return a6xx_idle(gpu, ring) ? 0 : -EINVAL;
1050 }
1051 
1052 /*
1053  * Check that the microcode version is new enough to include several key
1054  * security fixes. Return true if the ucode is safe.
1055  */
a6xx_ucode_check_version(struct a6xx_gpu * a6xx_gpu,struct drm_gem_object * obj)1056 static bool a6xx_ucode_check_version(struct a6xx_gpu *a6xx_gpu,
1057 		struct drm_gem_object *obj)
1058 {
1059 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
1060 	struct msm_gpu *gpu = &adreno_gpu->base;
1061 	const char *sqe_name = adreno_gpu->info->fw[ADRENO_FW_SQE];
1062 	u32 *buf = msm_gem_get_vaddr(obj);
1063 	bool ret = false;
1064 
1065 	if (IS_ERR(buf))
1066 		return false;
1067 
1068 	/*
1069 	 * Targets up to a640 (a618, a630 and a640) need to check for a
1070 	 * microcode version that is patched to support the whereami opcode or
1071 	 * one that is new enough to include it by default.
1072 	 *
1073 	 * a650 tier targets don't need whereami but still need to be
1074 	 * equal to or newer than 0.95 for other security fixes
1075 	 *
1076 	 * a660 targets have all the critical security fixes from the start
1077 	 */
1078 	if (!strcmp(sqe_name, "a630_sqe.fw")) {
1079 		/*
1080 		 * If the lowest nibble is 0xa that is an indication that this
1081 		 * microcode has been patched. The actual version is in dword
1082 		 * [3] but we only care about the patchlevel which is the lowest
1083 		 * nibble of dword [3]
1084 		 *
1085 		 * Otherwise check that the firmware is greater than or equal
1086 		 * to 1.90 which was the first version that had this fix built
1087 		 * in
1088 		 */
1089 		if ((((buf[0] & 0xf) == 0xa) && (buf[2] & 0xf) >= 1) ||
1090 			(buf[0] & 0xfff) >= 0x190) {
1091 			a6xx_gpu->has_whereami = true;
1092 			ret = true;
1093 			goto out;
1094 		}
1095 
1096 		DRM_DEV_ERROR(&gpu->pdev->dev,
1097 			"a630 SQE ucode is too old. Have version %x need at least %x\n",
1098 			buf[0] & 0xfff, 0x190);
1099 	} else if (!strcmp(sqe_name, "a650_sqe.fw")) {
1100 		if ((buf[0] & 0xfff) >= 0x095) {
1101 			ret = true;
1102 			goto out;
1103 		}
1104 
1105 		DRM_DEV_ERROR(&gpu->pdev->dev,
1106 			"a650 SQE ucode is too old. Have version %x need at least %x\n",
1107 			buf[0] & 0xfff, 0x095);
1108 	} else if (!strcmp(sqe_name, "a660_sqe.fw")) {
1109 		ret = true;
1110 	} else {
1111 		DRM_DEV_ERROR(&gpu->pdev->dev,
1112 			"unknown GPU, add it to a6xx_ucode_check_version()!!\n");
1113 	}
1114 out:
1115 	msm_gem_put_vaddr(obj);
1116 	return ret;
1117 }
1118 
a6xx_ucode_load(struct msm_gpu * gpu)1119 static int a6xx_ucode_load(struct msm_gpu *gpu)
1120 {
1121 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1122 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1123 
1124 	if (!a6xx_gpu->sqe_bo) {
1125 		a6xx_gpu->sqe_bo = adreno_fw_create_bo(gpu,
1126 			adreno_gpu->fw[ADRENO_FW_SQE], &a6xx_gpu->sqe_iova);
1127 
1128 		if (IS_ERR(a6xx_gpu->sqe_bo)) {
1129 			int ret = PTR_ERR(a6xx_gpu->sqe_bo);
1130 
1131 			a6xx_gpu->sqe_bo = NULL;
1132 			DRM_DEV_ERROR(&gpu->pdev->dev,
1133 				"Could not allocate SQE ucode: %d\n", ret);
1134 
1135 			return ret;
1136 		}
1137 
1138 		msm_gem_object_set_name(a6xx_gpu->sqe_bo, "sqefw");
1139 		if (!a6xx_ucode_check_version(a6xx_gpu, a6xx_gpu->sqe_bo)) {
1140 			msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
1141 			drm_gem_object_put(a6xx_gpu->sqe_bo);
1142 
1143 			a6xx_gpu->sqe_bo = NULL;
1144 			return -EPERM;
1145 		}
1146 	}
1147 
1148 	/*
1149 	 * Expanded APRIV and targets that support WHERE_AM_I both need a
1150 	 * privileged buffer to store the RPTR shadow
1151 	 */
1152 	if ((adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami) &&
1153 	    !a6xx_gpu->shadow_bo) {
1154 		a6xx_gpu->shadow = msm_gem_kernel_new(gpu->dev,
1155 						      sizeof(u32) * gpu->nr_rings,
1156 						      MSM_BO_WC | MSM_BO_MAP_PRIV,
1157 						      gpu->aspace, &a6xx_gpu->shadow_bo,
1158 						      &a6xx_gpu->shadow_iova);
1159 
1160 		if (IS_ERR(a6xx_gpu->shadow))
1161 			return PTR_ERR(a6xx_gpu->shadow);
1162 
1163 		msm_gem_object_set_name(a6xx_gpu->shadow_bo, "shadow");
1164 	}
1165 
1166 	return 0;
1167 }
1168 
a6xx_zap_shader_init(struct msm_gpu * gpu)1169 static int a6xx_zap_shader_init(struct msm_gpu *gpu)
1170 {
1171 	static bool loaded;
1172 	int ret;
1173 
1174 	if (loaded)
1175 		return 0;
1176 
1177 	ret = adreno_zap_shader_load(gpu, GPU_PAS_ID);
1178 
1179 	loaded = !ret;
1180 	return ret;
1181 }
1182 
1183 #define A6XX_INT_MASK (A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR | \
1184 	  A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW | \
1185 	  A6XX_RBBM_INT_0_MASK_CP_HW_ERROR | \
1186 	  A6XX_RBBM_INT_0_MASK_CP_IB2 | \
1187 	  A6XX_RBBM_INT_0_MASK_CP_IB1 | \
1188 	  A6XX_RBBM_INT_0_MASK_CP_RB | \
1189 	  A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \
1190 	  A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW | \
1191 	  A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT | \
1192 	  A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \
1193 	  A6XX_RBBM_INT_0_MASK_UCHE_TRAP_INTR)
1194 
hw_init(struct msm_gpu * gpu)1195 static int hw_init(struct msm_gpu *gpu)
1196 {
1197 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1198 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1199 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
1200 	int ret;
1201 
1202 	if (!adreno_has_gmu_wrapper(adreno_gpu)) {
1203 		/* Make sure the GMU keeps the GPU on while we set it up */
1204 		ret = a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
1205 		if (ret)
1206 			return ret;
1207 	}
1208 
1209 	/* Clear GBIF halt in case GX domain was not collapsed */
1210 	if (adreno_is_a619_holi(adreno_gpu)) {
1211 		gpu_write(gpu, REG_A6XX_GBIF_HALT, 0);
1212 		gpu_write(gpu, REG_A6XX_RBBM_GPR0_CNTL, 0);
1213 		/* Let's make extra sure that the GPU can access the memory.. */
1214 		mb();
1215 	} else if (a6xx_has_gbif(adreno_gpu)) {
1216 		gpu_write(gpu, REG_A6XX_GBIF_HALT, 0);
1217 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_HALT, 0);
1218 		/* Let's make extra sure that the GPU can access the memory.. */
1219 		mb();
1220 	}
1221 
1222 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_CNTL, 0);
1223 
1224 	if (adreno_is_a619_holi(adreno_gpu))
1225 		a6xx_sptprac_enable(gmu);
1226 
1227 	/*
1228 	 * Disable the trusted memory range - we don't actually supported secure
1229 	 * memory rendering at this point in time and we don't want to block off
1230 	 * part of the virtual memory space.
1231 	 */
1232 	gpu_write64(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_BASE, 0x00000000);
1233 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_SIZE, 0x00000000);
1234 
1235 	/* Turn on 64 bit addressing for all blocks */
1236 	gpu_write(gpu, REG_A6XX_CP_ADDR_MODE_CNTL, 0x1);
1237 	gpu_write(gpu, REG_A6XX_VSC_ADDR_MODE_CNTL, 0x1);
1238 	gpu_write(gpu, REG_A6XX_GRAS_ADDR_MODE_CNTL, 0x1);
1239 	gpu_write(gpu, REG_A6XX_RB_ADDR_MODE_CNTL, 0x1);
1240 	gpu_write(gpu, REG_A6XX_PC_ADDR_MODE_CNTL, 0x1);
1241 	gpu_write(gpu, REG_A6XX_HLSQ_ADDR_MODE_CNTL, 0x1);
1242 	gpu_write(gpu, REG_A6XX_VFD_ADDR_MODE_CNTL, 0x1);
1243 	gpu_write(gpu, REG_A6XX_VPC_ADDR_MODE_CNTL, 0x1);
1244 	gpu_write(gpu, REG_A6XX_UCHE_ADDR_MODE_CNTL, 0x1);
1245 	gpu_write(gpu, REG_A6XX_SP_ADDR_MODE_CNTL, 0x1);
1246 	gpu_write(gpu, REG_A6XX_TPL1_ADDR_MODE_CNTL, 0x1);
1247 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_ADDR_MODE_CNTL, 0x1);
1248 
1249 	/* enable hardware clockgating */
1250 	a6xx_set_hwcg(gpu, true);
1251 
1252 	/* VBIF/GBIF start*/
1253 	if (adreno_is_a610(adreno_gpu) ||
1254 	    adreno_is_a640_family(adreno_gpu) ||
1255 	    adreno_is_a650_family(adreno_gpu)) {
1256 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE0, 0x00071620);
1257 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE1, 0x00071620);
1258 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE2, 0x00071620);
1259 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
1260 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL, 0x3);
1261 	} else {
1262 		gpu_write(gpu, REG_A6XX_RBBM_VBIF_CLIENT_QOS_CNTL, 0x3);
1263 	}
1264 
1265 	if (adreno_is_a630(adreno_gpu))
1266 		gpu_write(gpu, REG_A6XX_VBIF_GATE_OFF_WRREQ_EN, 0x00000009);
1267 
1268 	/* Make all blocks contribute to the GPU BUSY perf counter */
1269 	gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_GPU_BUSY_MASKED, 0xffffffff);
1270 
1271 	/* Disable L2 bypass in the UCHE */
1272 	gpu_write64(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX, 0x0001ffffffffffc0llu);
1273 	gpu_write64(gpu, REG_A6XX_UCHE_TRAP_BASE, 0x0001fffffffff000llu);
1274 	gpu_write64(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE, 0x0001fffffffff000llu);
1275 
1276 	if (!adreno_is_a650_family(adreno_gpu)) {
1277 		/* Set the GMEM VA range [0x100000:0x100000 + gpu->gmem - 1] */
1278 		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MIN, 0x00100000);
1279 
1280 		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MAX,
1281 			0x00100000 + adreno_gpu->info->gmem - 1);
1282 	}
1283 
1284 	gpu_write(gpu, REG_A6XX_UCHE_FILTER_CNTL, 0x804);
1285 	gpu_write(gpu, REG_A6XX_UCHE_CACHE_WAYS, 0x4);
1286 
1287 	if (adreno_is_a640_family(adreno_gpu) || adreno_is_a650_family(adreno_gpu)) {
1288 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x02000140);
1289 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x8040362c);
1290 	} else if (adreno_is_a610(adreno_gpu)) {
1291 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x00800060);
1292 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x40201b16);
1293 	} else {
1294 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x010000c0);
1295 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x8040362c);
1296 	}
1297 
1298 	if (adreno_is_a660_family(adreno_gpu))
1299 		gpu_write(gpu, REG_A6XX_CP_LPAC_PROG_FIFO_SIZE, 0x00000020);
1300 
1301 	/* Setting the mem pool size */
1302 	if (adreno_is_a610(adreno_gpu)) {
1303 		gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 48);
1304 		gpu_write(gpu, REG_A6XX_CP_MEM_POOL_DBG_ADDR, 47);
1305 	} else
1306 		gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 128);
1307 
1308 	/* Setting the primFifo thresholds default values,
1309 	 * and vccCacheSkipDis=1 bit (0x200) for A640 and newer
1310 	*/
1311 	if (adreno_is_a650(adreno_gpu) || adreno_is_a660(adreno_gpu) || adreno_is_a690(adreno_gpu))
1312 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00300200);
1313 	else if (adreno_is_a640_family(adreno_gpu) || adreno_is_7c3(adreno_gpu))
1314 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00200200);
1315 	else if (adreno_is_a650(adreno_gpu) || adreno_is_a660(adreno_gpu))
1316 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00300200);
1317 	else if (adreno_is_a619(adreno_gpu))
1318 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00018000);
1319 	else if (adreno_is_a610(adreno_gpu))
1320 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00080000);
1321 	else
1322 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00180000);
1323 
1324 	/* Set the AHB default slave response to "ERROR" */
1325 	gpu_write(gpu, REG_A6XX_CP_AHB_CNTL, 0x1);
1326 
1327 	/* Turn on performance counters */
1328 	gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_CNTL, 0x1);
1329 
1330 	/* Select CP0 to always count cycles */
1331 	gpu_write(gpu, REG_A6XX_CP_PERFCTR_CP_SEL(0), PERF_CP_ALWAYS_COUNT);
1332 
1333 	a6xx_set_ubwc_config(gpu);
1334 
1335 	/* Enable fault detection */
1336 	if (adreno_is_a619(adreno_gpu))
1337 		gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL, (1 << 30) | 0x3fffff);
1338 	else if (adreno_is_a610(adreno_gpu))
1339 		gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL, (1 << 30) | 0x3ffff);
1340 	else
1341 		gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL, (1 << 30) | 0x1fffff);
1342 
1343 	gpu_write(gpu, REG_A6XX_UCHE_CLIENT_PF, 1);
1344 
1345 	/* Set weights for bicubic filtering */
1346 	if (adreno_is_a650_family(adreno_gpu)) {
1347 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_0, 0);
1348 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_1,
1349 			0x3fe05ff4);
1350 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_2,
1351 			0x3fa0ebee);
1352 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_3,
1353 			0x3f5193ed);
1354 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_4,
1355 			0x3f0243f0);
1356 	}
1357 
1358 	/* Set up the CX GMU counter 0 to count busy ticks */
1359 	gmu_write(gmu, REG_A6XX_GPU_GMU_AO_GPU_CX_BUSY_MASK, 0xff000000);
1360 
1361 	/* Enable the power counter */
1362 	gmu_rmw(gmu, REG_A6XX_GMU_CX_GMU_POWER_COUNTER_SELECT_0, 0xff, BIT(5));
1363 	gmu_write(gmu, REG_A6XX_GMU_CX_GMU_POWER_COUNTER_ENABLE, 1);
1364 
1365 	/* Protect registers from the CP */
1366 	a6xx_set_cp_protect(gpu);
1367 
1368 	if (adreno_is_a660_family(adreno_gpu)) {
1369 		gpu_write(gpu, REG_A6XX_CP_CHICKEN_DBG, 0x1);
1370 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL, 0x0);
1371 	}
1372 
1373 	/* Set dualQ + disable afull for A660 GPU */
1374 	if (adreno_is_a660(adreno_gpu))
1375 		gpu_write(gpu, REG_A6XX_UCHE_CMDQ_CONFIG, 0x66906);
1376 
1377 	/* Enable expanded apriv for targets that support it */
1378 	if (gpu->hw_apriv) {
1379 		gpu_write(gpu, REG_A6XX_CP_APRIV_CNTL,
1380 			(1 << 6) | (1 << 5) | (1 << 3) | (1 << 2) | (1 << 1));
1381 	}
1382 
1383 	/* Enable interrupts */
1384 	gpu_write(gpu, REG_A6XX_RBBM_INT_0_MASK, A6XX_INT_MASK);
1385 
1386 	ret = adreno_hw_init(gpu);
1387 	if (ret)
1388 		goto out;
1389 
1390 	gpu_write64(gpu, REG_A6XX_CP_SQE_INSTR_BASE, a6xx_gpu->sqe_iova);
1391 
1392 	/* Set the ringbuffer address */
1393 	gpu_write64(gpu, REG_A6XX_CP_RB_BASE, gpu->rb[0]->iova);
1394 
1395 	/* Targets that support extended APRIV can use the RPTR shadow from
1396 	 * hardware but all the other ones need to disable the feature. Targets
1397 	 * that support the WHERE_AM_I opcode can use that instead
1398 	 */
1399 	if (adreno_gpu->base.hw_apriv)
1400 		gpu_write(gpu, REG_A6XX_CP_RB_CNTL, MSM_GPU_RB_CNTL_DEFAULT);
1401 	else
1402 		gpu_write(gpu, REG_A6XX_CP_RB_CNTL,
1403 			MSM_GPU_RB_CNTL_DEFAULT | AXXX_CP_RB_CNTL_NO_UPDATE);
1404 
1405 	/* Configure the RPTR shadow if needed: */
1406 	if (a6xx_gpu->shadow_bo) {
1407 		gpu_write64(gpu, REG_A6XX_CP_RB_RPTR_ADDR,
1408 			shadowptr(a6xx_gpu, gpu->rb[0]));
1409 	}
1410 
1411 	/* Always come up on rb 0 */
1412 	a6xx_gpu->cur_ring = gpu->rb[0];
1413 
1414 	gpu->cur_ctx_seqno = 0;
1415 
1416 	/* Enable the SQE_to start the CP engine */
1417 	gpu_write(gpu, REG_A6XX_CP_SQE_CNTL, 1);
1418 
1419 	ret = a6xx_cp_init(gpu);
1420 	if (ret)
1421 		goto out;
1422 
1423 	/*
1424 	 * Try to load a zap shader into the secure world. If successful
1425 	 * we can use the CP to switch out of secure mode. If not then we
1426 	 * have no resource but to try to switch ourselves out manually. If we
1427 	 * guessed wrong then access to the RBBM_SECVID_TRUST_CNTL register will
1428 	 * be blocked and a permissions violation will soon follow.
1429 	 */
1430 	ret = a6xx_zap_shader_init(gpu);
1431 	if (!ret) {
1432 		OUT_PKT7(gpu->rb[0], CP_SET_SECURE_MODE, 1);
1433 		OUT_RING(gpu->rb[0], 0x00000000);
1434 
1435 		a6xx_flush(gpu, gpu->rb[0]);
1436 		if (!a6xx_idle(gpu, gpu->rb[0]))
1437 			return -EINVAL;
1438 	} else if (ret == -ENODEV) {
1439 		/*
1440 		 * This device does not use zap shader (but print a warning
1441 		 * just in case someone got their dt wrong.. hopefully they
1442 		 * have a debug UART to realize the error of their ways...
1443 		 * if you mess this up you are about to crash horribly)
1444 		 */
1445 		dev_warn_once(gpu->dev->dev,
1446 			"Zap shader not enabled - using SECVID_TRUST_CNTL instead\n");
1447 		gpu_write(gpu, REG_A6XX_RBBM_SECVID_TRUST_CNTL, 0x0);
1448 		ret = 0;
1449 	} else {
1450 		return ret;
1451 	}
1452 
1453 out:
1454 	if (adreno_has_gmu_wrapper(adreno_gpu))
1455 		return ret;
1456 	/*
1457 	 * Tell the GMU that we are done touching the GPU and it can start power
1458 	 * management
1459 	 */
1460 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
1461 
1462 	if (a6xx_gpu->gmu.legacy) {
1463 		/* Take the GMU out of its special boot mode */
1464 		a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
1465 	}
1466 
1467 	return ret;
1468 }
1469 
a6xx_hw_init(struct msm_gpu * gpu)1470 static int a6xx_hw_init(struct msm_gpu *gpu)
1471 {
1472 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1473 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1474 	int ret;
1475 
1476 	mutex_lock(&a6xx_gpu->gmu.lock);
1477 	ret = hw_init(gpu);
1478 	mutex_unlock(&a6xx_gpu->gmu.lock);
1479 
1480 	return ret;
1481 }
1482 
a6xx_dump(struct msm_gpu * gpu)1483 static void a6xx_dump(struct msm_gpu *gpu)
1484 {
1485 	DRM_DEV_INFO(&gpu->pdev->dev, "status:   %08x\n",
1486 			gpu_read(gpu, REG_A6XX_RBBM_STATUS));
1487 	adreno_dump(gpu);
1488 }
1489 
a6xx_recover(struct msm_gpu * gpu)1490 static void a6xx_recover(struct msm_gpu *gpu)
1491 {
1492 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1493 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1494 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
1495 	int i, active_submits;
1496 
1497 	adreno_dump_info(gpu);
1498 
1499 	for (i = 0; i < 8; i++)
1500 		DRM_DEV_INFO(&gpu->pdev->dev, "CP_SCRATCH_REG%d: %u\n", i,
1501 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(i)));
1502 
1503 	if (hang_debug)
1504 		a6xx_dump(gpu);
1505 
1506 	/*
1507 	 * To handle recovery specific sequences during the rpm suspend we are
1508 	 * about to trigger
1509 	 */
1510 	a6xx_gpu->hung = true;
1511 
1512 	/* Halt SQE first */
1513 	gpu_write(gpu, REG_A6XX_CP_SQE_CNTL, 3);
1514 
1515 	pm_runtime_dont_use_autosuspend(&gpu->pdev->dev);
1516 
1517 	/* active_submit won't change until we make a submission */
1518 	mutex_lock(&gpu->active_lock);
1519 	active_submits = gpu->active_submits;
1520 
1521 	/*
1522 	 * Temporarily clear active_submits count to silence a WARN() in the
1523 	 * runtime suspend cb
1524 	 */
1525 	gpu->active_submits = 0;
1526 
1527 	if (adreno_has_gmu_wrapper(adreno_gpu)) {
1528 		/* Drain the outstanding traffic on memory buses */
1529 		a6xx_bus_clear_pending_transactions(adreno_gpu, true);
1530 
1531 		/* Reset the GPU to a clean state */
1532 		a6xx_gpu_sw_reset(gpu, true);
1533 		a6xx_gpu_sw_reset(gpu, false);
1534 	}
1535 
1536 	reinit_completion(&gmu->pd_gate);
1537 	dev_pm_genpd_add_notifier(gmu->cxpd, &gmu->pd_nb);
1538 	dev_pm_genpd_synced_poweroff(gmu->cxpd);
1539 
1540 	/* Drop the rpm refcount from active submits */
1541 	if (active_submits)
1542 		pm_runtime_put(&gpu->pdev->dev);
1543 
1544 	/* And the final one from recover worker */
1545 	pm_runtime_put_sync(&gpu->pdev->dev);
1546 
1547 	if (!wait_for_completion_timeout(&gmu->pd_gate, msecs_to_jiffies(1000)))
1548 		DRM_DEV_ERROR(&gpu->pdev->dev, "cx gdsc didn't collapse\n");
1549 
1550 	dev_pm_genpd_remove_notifier(gmu->cxpd);
1551 
1552 	pm_runtime_use_autosuspend(&gpu->pdev->dev);
1553 
1554 	if (active_submits)
1555 		pm_runtime_get(&gpu->pdev->dev);
1556 
1557 	pm_runtime_get_sync(&gpu->pdev->dev);
1558 
1559 	gpu->active_submits = active_submits;
1560 	mutex_unlock(&gpu->active_lock);
1561 
1562 	msm_gpu_hw_init(gpu);
1563 	a6xx_gpu->hung = false;
1564 }
1565 
a6xx_uche_fault_block(struct msm_gpu * gpu,u32 mid)1566 static const char *a6xx_uche_fault_block(struct msm_gpu *gpu, u32 mid)
1567 {
1568 	static const char *uche_clients[7] = {
1569 		"VFD", "SP", "VSC", "VPC", "HLSQ", "PC", "LRZ",
1570 	};
1571 	u32 val;
1572 
1573 	if (mid < 1 || mid > 3)
1574 		return "UNKNOWN";
1575 
1576 	/*
1577 	 * The source of the data depends on the mid ID read from FSYNR1.
1578 	 * and the client ID read from the UCHE block
1579 	 */
1580 	val = gpu_read(gpu, REG_A6XX_UCHE_CLIENT_PF);
1581 
1582 	/* mid = 3 is most precise and refers to only one block per client */
1583 	if (mid == 3)
1584 		return uche_clients[val & 7];
1585 
1586 	/* For mid=2 the source is TP or VFD except when the client id is 0 */
1587 	if (mid == 2)
1588 		return ((val & 7) == 0) ? "TP" : "TP|VFD";
1589 
1590 	/* For mid=1 just return "UCHE" as a catchall for everything else */
1591 	return "UCHE";
1592 }
1593 
a6xx_fault_block(struct msm_gpu * gpu,u32 id)1594 static const char *a6xx_fault_block(struct msm_gpu *gpu, u32 id)
1595 {
1596 	if (id == 0)
1597 		return "CP";
1598 	else if (id == 4)
1599 		return "CCU";
1600 	else if (id == 6)
1601 		return "CDP Prefetch";
1602 
1603 	return a6xx_uche_fault_block(gpu, id);
1604 }
1605 
a6xx_fault_handler(void * arg,unsigned long iova,int flags,void * data)1606 static int a6xx_fault_handler(void *arg, unsigned long iova, int flags, void *data)
1607 {
1608 	struct msm_gpu *gpu = arg;
1609 	struct adreno_smmu_fault_info *info = data;
1610 	const char *block = "unknown";
1611 
1612 	u32 scratch[] = {
1613 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(4)),
1614 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(5)),
1615 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(6)),
1616 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(7)),
1617 	};
1618 
1619 	if (info)
1620 		block = a6xx_fault_block(gpu, info->fsynr1 & 0xff);
1621 
1622 	return adreno_fault_handler(gpu, iova, flags, info, block, scratch);
1623 }
1624 
a6xx_cp_hw_err_irq(struct msm_gpu * gpu)1625 static void a6xx_cp_hw_err_irq(struct msm_gpu *gpu)
1626 {
1627 	u32 status = gpu_read(gpu, REG_A6XX_CP_INTERRUPT_STATUS);
1628 
1629 	if (status & A6XX_CP_INT_CP_OPCODE_ERROR) {
1630 		u32 val;
1631 
1632 		gpu_write(gpu, REG_A6XX_CP_SQE_STAT_ADDR, 1);
1633 		val = gpu_read(gpu, REG_A6XX_CP_SQE_STAT_DATA);
1634 		dev_err_ratelimited(&gpu->pdev->dev,
1635 			"CP | opcode error | possible opcode=0x%8.8X\n",
1636 			val);
1637 	}
1638 
1639 	if (status & A6XX_CP_INT_CP_UCODE_ERROR)
1640 		dev_err_ratelimited(&gpu->pdev->dev,
1641 			"CP ucode error interrupt\n");
1642 
1643 	if (status & A6XX_CP_INT_CP_HW_FAULT_ERROR)
1644 		dev_err_ratelimited(&gpu->pdev->dev, "CP | HW fault | status=0x%8.8X\n",
1645 			gpu_read(gpu, REG_A6XX_CP_HW_FAULT));
1646 
1647 	if (status & A6XX_CP_INT_CP_REGISTER_PROTECTION_ERROR) {
1648 		u32 val = gpu_read(gpu, REG_A6XX_CP_PROTECT_STATUS);
1649 
1650 		dev_err_ratelimited(&gpu->pdev->dev,
1651 			"CP | protected mode error | %s | addr=0x%8.8X | status=0x%8.8X\n",
1652 			val & (1 << 20) ? "READ" : "WRITE",
1653 			(val & 0x3ffff), val);
1654 	}
1655 
1656 	if (status & A6XX_CP_INT_CP_AHB_ERROR)
1657 		dev_err_ratelimited(&gpu->pdev->dev, "CP AHB error interrupt\n");
1658 
1659 	if (status & A6XX_CP_INT_CP_VSD_PARITY_ERROR)
1660 		dev_err_ratelimited(&gpu->pdev->dev, "CP VSD decoder parity error\n");
1661 
1662 	if (status & A6XX_CP_INT_CP_ILLEGAL_INSTR_ERROR)
1663 		dev_err_ratelimited(&gpu->pdev->dev, "CP illegal instruction error\n");
1664 
1665 }
1666 
a6xx_fault_detect_irq(struct msm_gpu * gpu)1667 static void a6xx_fault_detect_irq(struct msm_gpu *gpu)
1668 {
1669 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1670 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1671 	struct msm_ringbuffer *ring = gpu->funcs->active_ring(gpu);
1672 
1673 	/*
1674 	 * If stalled on SMMU fault, we could trip the GPU's hang detection,
1675 	 * but the fault handler will trigger the devcore dump, and we want
1676 	 * to otherwise resume normally rather than killing the submit, so
1677 	 * just bail.
1678 	 */
1679 	if (gpu_read(gpu, REG_A6XX_RBBM_STATUS3) & A6XX_RBBM_STATUS3_SMMU_STALLED_ON_FAULT)
1680 		return;
1681 
1682 	/*
1683 	 * Force the GPU to stay on until after we finish
1684 	 * collecting information
1685 	 */
1686 	if (!adreno_has_gmu_wrapper(adreno_gpu))
1687 		gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 1);
1688 
1689 	DRM_DEV_ERROR(&gpu->pdev->dev,
1690 		"gpu fault ring %d fence %x status %8.8X rb %4.4x/%4.4x ib1 %16.16llX/%4.4x ib2 %16.16llX/%4.4x\n",
1691 		ring ? ring->id : -1, ring ? ring->fctx->last_fence : 0,
1692 		gpu_read(gpu, REG_A6XX_RBBM_STATUS),
1693 		gpu_read(gpu, REG_A6XX_CP_RB_RPTR),
1694 		gpu_read(gpu, REG_A6XX_CP_RB_WPTR),
1695 		gpu_read64(gpu, REG_A6XX_CP_IB1_BASE),
1696 		gpu_read(gpu, REG_A6XX_CP_IB1_REM_SIZE),
1697 		gpu_read64(gpu, REG_A6XX_CP_IB2_BASE),
1698 		gpu_read(gpu, REG_A6XX_CP_IB2_REM_SIZE));
1699 
1700 	/* Turn off the hangcheck timer to keep it from bothering us */
1701 	del_timer(&gpu->hangcheck_timer);
1702 
1703 	kthread_queue_work(gpu->worker, &gpu->recover_work);
1704 }
1705 
a6xx_irq(struct msm_gpu * gpu)1706 static irqreturn_t a6xx_irq(struct msm_gpu *gpu)
1707 {
1708 	struct msm_drm_private *priv = gpu->dev->dev_private;
1709 	u32 status = gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS);
1710 
1711 	gpu_write(gpu, REG_A6XX_RBBM_INT_CLEAR_CMD, status);
1712 
1713 	if (priv->disable_err_irq)
1714 		status &= A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS;
1715 
1716 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT)
1717 		a6xx_fault_detect_irq(gpu);
1718 
1719 	if (status & A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR)
1720 		dev_err_ratelimited(&gpu->pdev->dev, "CP | AHB bus error\n");
1721 
1722 	if (status & A6XX_RBBM_INT_0_MASK_CP_HW_ERROR)
1723 		a6xx_cp_hw_err_irq(gpu);
1724 
1725 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW)
1726 		dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB ASYNC overflow\n");
1727 
1728 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW)
1729 		dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB bus overflow\n");
1730 
1731 	if (status & A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS)
1732 		dev_err_ratelimited(&gpu->pdev->dev, "UCHE | Out of bounds access\n");
1733 
1734 	if (status & A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS)
1735 		msm_gpu_retire(gpu);
1736 
1737 	return IRQ_HANDLED;
1738 }
1739 
a6xx_llc_deactivate(struct a6xx_gpu * a6xx_gpu)1740 static void a6xx_llc_deactivate(struct a6xx_gpu *a6xx_gpu)
1741 {
1742 	llcc_slice_deactivate(a6xx_gpu->llc_slice);
1743 	llcc_slice_deactivate(a6xx_gpu->htw_llc_slice);
1744 }
1745 
a6xx_llc_activate(struct a6xx_gpu * a6xx_gpu)1746 static void a6xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
1747 {
1748 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
1749 	struct msm_gpu *gpu = &adreno_gpu->base;
1750 	u32 cntl1_regval = 0;
1751 
1752 	if (IS_ERR(a6xx_gpu->llc_mmio))
1753 		return;
1754 
1755 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
1756 		u32 gpu_scid = llcc_get_slice_id(a6xx_gpu->llc_slice);
1757 
1758 		gpu_scid &= 0x1f;
1759 		cntl1_regval = (gpu_scid << 0) | (gpu_scid << 5) | (gpu_scid << 10) |
1760 			       (gpu_scid << 15) | (gpu_scid << 20);
1761 
1762 		/* On A660, the SCID programming for UCHE traffic is done in
1763 		 * A6XX_GBIF_SCACHE_CNTL0[14:10]
1764 		 */
1765 		if (adreno_is_a660_family(adreno_gpu))
1766 			gpu_rmw(gpu, REG_A6XX_GBIF_SCACHE_CNTL0, (0x1f << 10) |
1767 				(1 << 8), (gpu_scid << 10) | (1 << 8));
1768 	}
1769 
1770 	/*
1771 	 * For targets with a MMU500, activate the slice but don't program the
1772 	 * register.  The XBL will take care of that.
1773 	 */
1774 	if (!llcc_slice_activate(a6xx_gpu->htw_llc_slice)) {
1775 		if (!a6xx_gpu->have_mmu500) {
1776 			u32 gpuhtw_scid = llcc_get_slice_id(a6xx_gpu->htw_llc_slice);
1777 
1778 			gpuhtw_scid &= 0x1f;
1779 			cntl1_regval |= FIELD_PREP(GENMASK(29, 25), gpuhtw_scid);
1780 		}
1781 	}
1782 
1783 	if (!cntl1_regval)
1784 		return;
1785 
1786 	/*
1787 	 * Program the slice IDs for the various GPU blocks and GPU MMU
1788 	 * pagetables
1789 	 */
1790 	if (!a6xx_gpu->have_mmu500) {
1791 		a6xx_llc_write(a6xx_gpu,
1792 			REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_1, cntl1_regval);
1793 
1794 		/*
1795 		 * Program cacheability overrides to not allocate cache
1796 		 * lines on a write miss
1797 		 */
1798 		a6xx_llc_rmw(a6xx_gpu,
1799 			REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_0, 0xF, 0x03);
1800 		return;
1801 	}
1802 
1803 	gpu_rmw(gpu, REG_A6XX_GBIF_SCACHE_CNTL1, GENMASK(24, 0), cntl1_regval);
1804 }
1805 
a6xx_llc_slices_destroy(struct a6xx_gpu * a6xx_gpu)1806 static void a6xx_llc_slices_destroy(struct a6xx_gpu *a6xx_gpu)
1807 {
1808 	/* No LLCC on non-RPMh (and by extension, non-GMU) SoCs */
1809 	if (adreno_has_gmu_wrapper(&a6xx_gpu->base))
1810 		return;
1811 
1812 	llcc_slice_putd(a6xx_gpu->llc_slice);
1813 	llcc_slice_putd(a6xx_gpu->htw_llc_slice);
1814 }
1815 
a6xx_llc_slices_init(struct platform_device * pdev,struct a6xx_gpu * a6xx_gpu)1816 static void a6xx_llc_slices_init(struct platform_device *pdev,
1817 		struct a6xx_gpu *a6xx_gpu)
1818 {
1819 	struct device_node *phandle;
1820 
1821 	/* No LLCC on non-RPMh (and by extension, non-GMU) SoCs */
1822 	if (adreno_has_gmu_wrapper(&a6xx_gpu->base))
1823 		return;
1824 
1825 	/*
1826 	 * There is a different programming path for targets with an mmu500
1827 	 * attached, so detect if that is the case
1828 	 */
1829 	phandle = of_parse_phandle(pdev->dev.of_node, "iommus", 0);
1830 	a6xx_gpu->have_mmu500 = (phandle &&
1831 		of_device_is_compatible(phandle, "arm,mmu-500"));
1832 	of_node_put(phandle);
1833 
1834 	if (a6xx_gpu->have_mmu500)
1835 		a6xx_gpu->llc_mmio = NULL;
1836 	else
1837 		a6xx_gpu->llc_mmio = msm_ioremap(pdev, "cx_mem");
1838 
1839 	a6xx_gpu->llc_slice = llcc_slice_getd(LLCC_GPU);
1840 	a6xx_gpu->htw_llc_slice = llcc_slice_getd(LLCC_GPUHTW);
1841 
1842 	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
1843 		a6xx_gpu->llc_mmio = ERR_PTR(-EINVAL);
1844 }
1845 
1846 #define GBIF_CLIENT_HALT_MASK		BIT(0)
1847 #define GBIF_ARB_HALT_MASK		BIT(1)
1848 #define VBIF_XIN_HALT_CTRL0_MASK	GENMASK(3, 0)
1849 #define VBIF_RESET_ACK_MASK		0xF0
1850 #define GPR0_GBIF_HALT_REQUEST		0x1E0
1851 
a6xx_bus_clear_pending_transactions(struct adreno_gpu * adreno_gpu,bool gx_off)1852 void a6xx_bus_clear_pending_transactions(struct adreno_gpu *adreno_gpu, bool gx_off)
1853 {
1854 	struct msm_gpu *gpu = &adreno_gpu->base;
1855 
1856 	if (adreno_is_a619_holi(adreno_gpu)) {
1857 		gpu_write(gpu, REG_A6XX_RBBM_GPR0_CNTL, GPR0_GBIF_HALT_REQUEST);
1858 		spin_until((gpu_read(gpu, REG_A6XX_RBBM_VBIF_GX_RESET_STATUS) &
1859 				(VBIF_RESET_ACK_MASK)) == VBIF_RESET_ACK_MASK);
1860 	} else if (!a6xx_has_gbif(adreno_gpu)) {
1861 		gpu_write(gpu, REG_A6XX_VBIF_XIN_HALT_CTRL0, VBIF_XIN_HALT_CTRL0_MASK);
1862 		spin_until((gpu_read(gpu, REG_A6XX_VBIF_XIN_HALT_CTRL1) &
1863 				(VBIF_XIN_HALT_CTRL0_MASK)) == VBIF_XIN_HALT_CTRL0_MASK);
1864 		gpu_write(gpu, REG_A6XX_VBIF_XIN_HALT_CTRL0, 0);
1865 
1866 		return;
1867 	}
1868 
1869 	if (gx_off) {
1870 		/* Halt the gx side of GBIF */
1871 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_HALT, 1);
1872 		spin_until(gpu_read(gpu, REG_A6XX_RBBM_GBIF_HALT_ACK) & 1);
1873 	}
1874 
1875 	/* Halt new client requests on GBIF */
1876 	gpu_write(gpu, REG_A6XX_GBIF_HALT, GBIF_CLIENT_HALT_MASK);
1877 	spin_until((gpu_read(gpu, REG_A6XX_GBIF_HALT_ACK) &
1878 			(GBIF_CLIENT_HALT_MASK)) == GBIF_CLIENT_HALT_MASK);
1879 
1880 	/* Halt all AXI requests on GBIF */
1881 	gpu_write(gpu, REG_A6XX_GBIF_HALT, GBIF_ARB_HALT_MASK);
1882 	spin_until((gpu_read(gpu,  REG_A6XX_GBIF_HALT_ACK) &
1883 			(GBIF_ARB_HALT_MASK)) == GBIF_ARB_HALT_MASK);
1884 
1885 	/* The GBIF halt needs to be explicitly cleared */
1886 	gpu_write(gpu, REG_A6XX_GBIF_HALT, 0x0);
1887 }
1888 
a6xx_gpu_sw_reset(struct msm_gpu * gpu,bool assert)1889 void a6xx_gpu_sw_reset(struct msm_gpu *gpu, bool assert)
1890 {
1891 	/* 11nm chips (e.g. ones with A610) have hw issues with the reset line! */
1892 	if (adreno_is_a610(to_adreno_gpu(gpu)))
1893 		return;
1894 
1895 	gpu_write(gpu, REG_A6XX_RBBM_SW_RESET_CMD, assert);
1896 	/* Perform a bogus read and add a brief delay to ensure ordering. */
1897 	gpu_read(gpu, REG_A6XX_RBBM_SW_RESET_CMD);
1898 	udelay(1);
1899 
1900 	/* The reset line needs to be asserted for at least 100 us */
1901 	if (assert)
1902 		udelay(100);
1903 }
1904 
a6xx_gmu_pm_resume(struct msm_gpu * gpu)1905 static int a6xx_gmu_pm_resume(struct msm_gpu *gpu)
1906 {
1907 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1908 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1909 	int ret;
1910 
1911 	gpu->needs_hw_init = true;
1912 
1913 	trace_msm_gpu_resume(0);
1914 
1915 	mutex_lock(&a6xx_gpu->gmu.lock);
1916 	ret = a6xx_gmu_resume(a6xx_gpu);
1917 	mutex_unlock(&a6xx_gpu->gmu.lock);
1918 	if (ret)
1919 		return ret;
1920 
1921 	msm_devfreq_resume(gpu);
1922 
1923 	a6xx_llc_activate(a6xx_gpu);
1924 
1925 	return ret;
1926 }
1927 
a6xx_pm_resume(struct msm_gpu * gpu)1928 static int a6xx_pm_resume(struct msm_gpu *gpu)
1929 {
1930 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1931 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1932 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
1933 	unsigned long freq = gpu->fast_rate;
1934 	struct dev_pm_opp *opp;
1935 	int ret;
1936 
1937 	gpu->needs_hw_init = true;
1938 
1939 	trace_msm_gpu_resume(0);
1940 
1941 	mutex_lock(&a6xx_gpu->gmu.lock);
1942 
1943 	opp = dev_pm_opp_find_freq_ceil(&gpu->pdev->dev, &freq);
1944 	if (IS_ERR(opp)) {
1945 		ret = PTR_ERR(opp);
1946 		goto err_set_opp;
1947 	}
1948 	dev_pm_opp_put(opp);
1949 
1950 	/* Set the core clock and bus bw, having VDD scaling in mind */
1951 	dev_pm_opp_set_opp(&gpu->pdev->dev, opp);
1952 
1953 	pm_runtime_resume_and_get(gmu->dev);
1954 	pm_runtime_resume_and_get(gmu->gxpd);
1955 
1956 	ret = clk_bulk_prepare_enable(gpu->nr_clocks, gpu->grp_clks);
1957 	if (ret)
1958 		goto err_bulk_clk;
1959 
1960 	if (adreno_is_a619_holi(adreno_gpu))
1961 		a6xx_sptprac_enable(gmu);
1962 
1963 	/* If anything goes south, tear the GPU down piece by piece.. */
1964 	if (ret) {
1965 err_bulk_clk:
1966 		pm_runtime_put(gmu->gxpd);
1967 		pm_runtime_put(gmu->dev);
1968 		dev_pm_opp_set_opp(&gpu->pdev->dev, NULL);
1969 	}
1970 err_set_opp:
1971 	mutex_unlock(&a6xx_gpu->gmu.lock);
1972 
1973 	if (!ret)
1974 		msm_devfreq_resume(gpu);
1975 
1976 	return ret;
1977 }
1978 
a6xx_gmu_pm_suspend(struct msm_gpu * gpu)1979 static int a6xx_gmu_pm_suspend(struct msm_gpu *gpu)
1980 {
1981 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1982 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1983 	int i, ret;
1984 
1985 	trace_msm_gpu_suspend(0);
1986 
1987 	a6xx_llc_deactivate(a6xx_gpu);
1988 
1989 	msm_devfreq_suspend(gpu);
1990 
1991 	mutex_lock(&a6xx_gpu->gmu.lock);
1992 	ret = a6xx_gmu_stop(a6xx_gpu);
1993 	mutex_unlock(&a6xx_gpu->gmu.lock);
1994 	if (ret)
1995 		return ret;
1996 
1997 	if (a6xx_gpu->shadow_bo)
1998 		for (i = 0; i < gpu->nr_rings; i++)
1999 			a6xx_gpu->shadow[i] = 0;
2000 
2001 	gpu->suspend_count++;
2002 
2003 	return 0;
2004 }
2005 
a6xx_pm_suspend(struct msm_gpu * gpu)2006 static int a6xx_pm_suspend(struct msm_gpu *gpu)
2007 {
2008 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2009 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2010 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
2011 	int i;
2012 
2013 	trace_msm_gpu_suspend(0);
2014 
2015 	msm_devfreq_suspend(gpu);
2016 
2017 	mutex_lock(&a6xx_gpu->gmu.lock);
2018 
2019 	/* Drain the outstanding traffic on memory buses */
2020 	a6xx_bus_clear_pending_transactions(adreno_gpu, true);
2021 
2022 	if (adreno_is_a619_holi(adreno_gpu))
2023 		a6xx_sptprac_disable(gmu);
2024 
2025 	clk_bulk_disable_unprepare(gpu->nr_clocks, gpu->grp_clks);
2026 
2027 	pm_runtime_put_sync(gmu->gxpd);
2028 	dev_pm_opp_set_opp(&gpu->pdev->dev, NULL);
2029 	pm_runtime_put_sync(gmu->dev);
2030 
2031 	mutex_unlock(&a6xx_gpu->gmu.lock);
2032 
2033 	if (a6xx_gpu->shadow_bo)
2034 		for (i = 0; i < gpu->nr_rings; i++)
2035 			a6xx_gpu->shadow[i] = 0;
2036 
2037 	gpu->suspend_count++;
2038 
2039 	return 0;
2040 }
2041 
a6xx_gmu_get_timestamp(struct msm_gpu * gpu,uint64_t * value)2042 static int a6xx_gmu_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
2043 {
2044 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2045 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2046 
2047 	mutex_lock(&a6xx_gpu->gmu.lock);
2048 
2049 	/* Force the GPU power on so we can read this register */
2050 	a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
2051 
2052 	*value = gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER);
2053 
2054 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
2055 
2056 	mutex_unlock(&a6xx_gpu->gmu.lock);
2057 
2058 	return 0;
2059 }
2060 
a6xx_get_timestamp(struct msm_gpu * gpu,uint64_t * value)2061 static int a6xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
2062 {
2063 	*value = gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER);
2064 	return 0;
2065 }
2066 
a6xx_active_ring(struct msm_gpu * gpu)2067 static struct msm_ringbuffer *a6xx_active_ring(struct msm_gpu *gpu)
2068 {
2069 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2070 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2071 
2072 	return a6xx_gpu->cur_ring;
2073 }
2074 
a6xx_destroy(struct msm_gpu * gpu)2075 static void a6xx_destroy(struct msm_gpu *gpu)
2076 {
2077 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2078 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2079 
2080 	if (a6xx_gpu->sqe_bo) {
2081 		msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
2082 		drm_gem_object_put(a6xx_gpu->sqe_bo);
2083 	}
2084 
2085 	if (a6xx_gpu->shadow_bo) {
2086 		msm_gem_unpin_iova(a6xx_gpu->shadow_bo, gpu->aspace);
2087 		drm_gem_object_put(a6xx_gpu->shadow_bo);
2088 	}
2089 
2090 	a6xx_llc_slices_destroy(a6xx_gpu);
2091 
2092 	a6xx_gmu_remove(a6xx_gpu);
2093 
2094 	adreno_gpu_cleanup(adreno_gpu);
2095 
2096 	kfree(a6xx_gpu);
2097 }
2098 
a6xx_gpu_busy(struct msm_gpu * gpu,unsigned long * out_sample_rate)2099 static u64 a6xx_gpu_busy(struct msm_gpu *gpu, unsigned long *out_sample_rate)
2100 {
2101 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2102 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2103 	u64 busy_cycles;
2104 
2105 	/* 19.2MHz */
2106 	*out_sample_rate = 19200000;
2107 
2108 	busy_cycles = gmu_read64(&a6xx_gpu->gmu,
2109 			REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_L,
2110 			REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_H);
2111 
2112 	return busy_cycles;
2113 }
2114 
a6xx_gpu_set_freq(struct msm_gpu * gpu,struct dev_pm_opp * opp,bool suspended)2115 static void a6xx_gpu_set_freq(struct msm_gpu *gpu, struct dev_pm_opp *opp,
2116 			      bool suspended)
2117 {
2118 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2119 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2120 
2121 	mutex_lock(&a6xx_gpu->gmu.lock);
2122 	a6xx_gmu_set_freq(gpu, opp, suspended);
2123 	mutex_unlock(&a6xx_gpu->gmu.lock);
2124 }
2125 
2126 static struct msm_gem_address_space *
a6xx_create_address_space(struct msm_gpu * gpu,struct platform_device * pdev)2127 a6xx_create_address_space(struct msm_gpu *gpu, struct platform_device *pdev)
2128 {
2129 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2130 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2131 	unsigned long quirks = 0;
2132 
2133 	/*
2134 	 * This allows GPU to set the bus attributes required to use system
2135 	 * cache on behalf of the iommu page table walker.
2136 	 */
2137 	if (!IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice) &&
2138 	    !device_iommu_capable(&pdev->dev, IOMMU_CAP_CACHE_COHERENCY))
2139 		quirks |= IO_PGTABLE_QUIRK_ARM_OUTER_WBWA;
2140 
2141 	return adreno_iommu_create_address_space(gpu, pdev, quirks);
2142 }
2143 
2144 static struct msm_gem_address_space *
a6xx_create_private_address_space(struct msm_gpu * gpu)2145 a6xx_create_private_address_space(struct msm_gpu *gpu)
2146 {
2147 	struct msm_mmu *mmu;
2148 
2149 	mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
2150 
2151 	if (IS_ERR(mmu))
2152 		return ERR_CAST(mmu);
2153 
2154 	return msm_gem_address_space_create(mmu,
2155 		"gpu", 0x100000000ULL,
2156 		adreno_private_address_space_size(gpu));
2157 }
2158 
a6xx_get_rptr(struct msm_gpu * gpu,struct msm_ringbuffer * ring)2159 static uint32_t a6xx_get_rptr(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
2160 {
2161 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2162 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2163 
2164 	if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami)
2165 		return a6xx_gpu->shadow[ring->id];
2166 
2167 	return ring->memptrs->rptr = gpu_read(gpu, REG_A6XX_CP_RB_RPTR);
2168 }
2169 
a6xx_progress(struct msm_gpu * gpu,struct msm_ringbuffer * ring)2170 static bool a6xx_progress(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
2171 {
2172 	struct msm_cp_state cp_state = {
2173 		.ib1_base = gpu_read64(gpu, REG_A6XX_CP_IB1_BASE),
2174 		.ib2_base = gpu_read64(gpu, REG_A6XX_CP_IB2_BASE),
2175 		.ib1_rem  = gpu_read(gpu, REG_A6XX_CP_IB1_REM_SIZE),
2176 		.ib2_rem  = gpu_read(gpu, REG_A6XX_CP_IB2_REM_SIZE),
2177 	};
2178 	bool progress;
2179 
2180 	/*
2181 	 * Adjust the remaining data to account for what has already been
2182 	 * fetched from memory, but not yet consumed by the SQE.
2183 	 *
2184 	 * This is not *technically* correct, the amount buffered could
2185 	 * exceed the IB size due to hw prefetching ahead, but:
2186 	 *
2187 	 * (1) We aren't trying to find the exact position, just whether
2188 	 *     progress has been made
2189 	 * (2) The CP_REG_TO_MEM at the end of a submit should be enough
2190 	 *     to prevent prefetching into an unrelated submit.  (And
2191 	 *     either way, at some point the ROQ will be full.)
2192 	 */
2193 	cp_state.ib1_rem += gpu_read(gpu, REG_A6XX_CP_ROQ_AVAIL_IB1) >> 16;
2194 	cp_state.ib2_rem += gpu_read(gpu, REG_A6XX_CP_ROQ_AVAIL_IB2) >> 16;
2195 
2196 	progress = !!memcmp(&cp_state, &ring->last_cp_state, sizeof(cp_state));
2197 
2198 	ring->last_cp_state = cp_state;
2199 
2200 	return progress;
2201 }
2202 
fuse_to_supp_hw(const struct adreno_info * info,u32 fuse)2203 static u32 fuse_to_supp_hw(const struct adreno_info *info, u32 fuse)
2204 {
2205 	if (!info->speedbins)
2206 		return UINT_MAX;
2207 
2208 	for (int i = 0; info->speedbins[i].fuse != SHRT_MAX; i++)
2209 		if (info->speedbins[i].fuse == fuse)
2210 			return BIT(info->speedbins[i].speedbin);
2211 
2212 	return UINT_MAX;
2213 }
2214 
a6xx_set_supported_hw(struct device * dev,const struct adreno_info * info)2215 static int a6xx_set_supported_hw(struct device *dev, const struct adreno_info *info)
2216 {
2217 	u32 supp_hw;
2218 	u32 speedbin;
2219 	int ret;
2220 
2221 	ret = adreno_read_speedbin(dev, &speedbin);
2222 	/*
2223 	 * -ENOENT means that the platform doesn't support speedbin which is
2224 	 * fine
2225 	 */
2226 	if (ret == -ENOENT) {
2227 		return 0;
2228 	} else if (ret) {
2229 		dev_err_probe(dev, ret,
2230 			      "failed to read speed-bin. Some OPPs may not be supported by hardware\n");
2231 		return ret;
2232 	}
2233 
2234 	supp_hw = fuse_to_supp_hw(info, speedbin);
2235 
2236 	if (supp_hw == UINT_MAX) {
2237 		DRM_DEV_ERROR(dev,
2238 			"missing support for speed-bin: %u. Some OPPs may not be supported by hardware\n",
2239 			speedbin);
2240 		supp_hw = BIT(0); /* Default */
2241 	}
2242 
2243 	ret = devm_pm_opp_set_supported_hw(dev, &supp_hw, 1);
2244 	if (ret)
2245 		return ret;
2246 
2247 	return 0;
2248 }
2249 
2250 static const struct adreno_gpu_funcs funcs = {
2251 	.base = {
2252 		.get_param = adreno_get_param,
2253 		.set_param = adreno_set_param,
2254 		.hw_init = a6xx_hw_init,
2255 		.ucode_load = a6xx_ucode_load,
2256 		.pm_suspend = a6xx_gmu_pm_suspend,
2257 		.pm_resume = a6xx_gmu_pm_resume,
2258 		.recover = a6xx_recover,
2259 		.submit = a6xx_submit,
2260 		.active_ring = a6xx_active_ring,
2261 		.irq = a6xx_irq,
2262 		.destroy = a6xx_destroy,
2263 #if defined(CONFIG_DRM_MSM_GPU_STATE)
2264 		.show = a6xx_show,
2265 #endif
2266 		.gpu_busy = a6xx_gpu_busy,
2267 		.gpu_get_freq = a6xx_gmu_get_freq,
2268 		.gpu_set_freq = a6xx_gpu_set_freq,
2269 #if defined(CONFIG_DRM_MSM_GPU_STATE)
2270 		.gpu_state_get = a6xx_gpu_state_get,
2271 		.gpu_state_put = a6xx_gpu_state_put,
2272 #endif
2273 		.create_address_space = a6xx_create_address_space,
2274 		.create_private_address_space = a6xx_create_private_address_space,
2275 		.get_rptr = a6xx_get_rptr,
2276 		.progress = a6xx_progress,
2277 	},
2278 	.get_timestamp = a6xx_gmu_get_timestamp,
2279 };
2280 
2281 static const struct adreno_gpu_funcs funcs_gmuwrapper = {
2282 	.base = {
2283 		.get_param = adreno_get_param,
2284 		.set_param = adreno_set_param,
2285 		.hw_init = a6xx_hw_init,
2286 		.ucode_load = a6xx_ucode_load,
2287 		.pm_suspend = a6xx_pm_suspend,
2288 		.pm_resume = a6xx_pm_resume,
2289 		.recover = a6xx_recover,
2290 		.submit = a6xx_submit,
2291 		.active_ring = a6xx_active_ring,
2292 		.irq = a6xx_irq,
2293 		.destroy = a6xx_destroy,
2294 #if defined(CONFIG_DRM_MSM_GPU_STATE)
2295 		.show = a6xx_show,
2296 #endif
2297 		.gpu_busy = a6xx_gpu_busy,
2298 #if defined(CONFIG_DRM_MSM_GPU_STATE)
2299 		.gpu_state_get = a6xx_gpu_state_get,
2300 		.gpu_state_put = a6xx_gpu_state_put,
2301 #endif
2302 		.create_address_space = a6xx_create_address_space,
2303 		.create_private_address_space = a6xx_create_private_address_space,
2304 		.get_rptr = a6xx_get_rptr,
2305 		.progress = a6xx_progress,
2306 	},
2307 	.get_timestamp = a6xx_get_timestamp,
2308 };
2309 
a6xx_gpu_init(struct drm_device * dev)2310 struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
2311 {
2312 	struct msm_drm_private *priv = dev->dev_private;
2313 	struct platform_device *pdev = priv->gpu_pdev;
2314 	struct adreno_platform_config *config = pdev->dev.platform_data;
2315 	struct device_node *node;
2316 	struct a6xx_gpu *a6xx_gpu;
2317 	struct adreno_gpu *adreno_gpu;
2318 	struct msm_gpu *gpu;
2319 	int ret;
2320 
2321 	a6xx_gpu = kzalloc(sizeof(*a6xx_gpu), GFP_KERNEL);
2322 	if (!a6xx_gpu)
2323 		return ERR_PTR(-ENOMEM);
2324 
2325 	adreno_gpu = &a6xx_gpu->base;
2326 	gpu = &adreno_gpu->base;
2327 
2328 	mutex_init(&a6xx_gpu->gmu.lock);
2329 
2330 	adreno_gpu->registers = NULL;
2331 
2332 	/* Check if there is a GMU phandle and set it up */
2333 	node = of_parse_phandle(pdev->dev.of_node, "qcom,gmu", 0);
2334 	/* FIXME: How do we gracefully handle this? */
2335 	BUG_ON(!node);
2336 
2337 	adreno_gpu->gmu_is_wrapper = of_device_is_compatible(node, "qcom,adreno-gmu-wrapper");
2338 
2339 	adreno_gpu->base.hw_apriv =
2340 		!!(config->info->quirks & ADRENO_QUIRK_HAS_HW_APRIV);
2341 
2342 	a6xx_llc_slices_init(pdev, a6xx_gpu);
2343 
2344 	ret = a6xx_set_supported_hw(&pdev->dev, config->info);
2345 	if (ret) {
2346 		a6xx_destroy(&(a6xx_gpu->base.base));
2347 		return ERR_PTR(ret);
2348 	}
2349 
2350 	if (adreno_has_gmu_wrapper(adreno_gpu))
2351 		ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs_gmuwrapper, 1);
2352 	else
2353 		ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1);
2354 	if (ret) {
2355 		a6xx_destroy(&(a6xx_gpu->base.base));
2356 		return ERR_PTR(ret);
2357 	}
2358 
2359 	/*
2360 	 * For now only clamp to idle freq for devices where this is known not
2361 	 * to cause power supply issues:
2362 	 */
2363 	if (adreno_is_a618(adreno_gpu) || adreno_is_7c3(adreno_gpu))
2364 		priv->gpu_clamp_to_idle = true;
2365 
2366 	if (adreno_has_gmu_wrapper(adreno_gpu))
2367 		ret = a6xx_gmu_wrapper_init(a6xx_gpu, node);
2368 	else
2369 		ret = a6xx_gmu_init(a6xx_gpu, node);
2370 	of_node_put(node);
2371 	if (ret) {
2372 		a6xx_destroy(&(a6xx_gpu->base.base));
2373 		return ERR_PTR(ret);
2374 	}
2375 
2376 	if (gpu->aspace)
2377 		msm_mmu_set_fault_handler(gpu->aspace->mmu, gpu,
2378 				a6xx_fault_handler);
2379 
2380 	return gpu;
2381 }
2382