xref: /linux/drivers/gpu/drm/msm/adreno/a6xx_gpu.c (revision 5c1672705a1a2389f5ad78e0fea6f08ed32d6f18)
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 
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 
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 
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 
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 
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 
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 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
107 	phys_addr_t ttbr;
108 	u32 asid;
109 	u64 memptr = rbmemptr(ring, ttbr0);
110 
111 	if (ctx->seqno == a6xx_gpu->base.base.cur_ctx_seqno)
112 		return;
113 
114 	if (msm_iommu_pagetable_params(ctx->aspace->mmu, &ttbr, &asid))
115 		return;
116 
117 	if (!sysprof) {
118 		if (!adreno_is_a7xx(adreno_gpu)) {
119 			/* Turn off protected mode to write to special registers */
120 			OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1);
121 			OUT_RING(ring, 0);
122 		}
123 
124 		OUT_PKT4(ring, REG_A6XX_RBBM_PERFCTR_SRAM_INIT_CMD, 1);
125 		OUT_RING(ring, 1);
126 	}
127 
128 	/* Execute the table update */
129 	OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
130 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_0_TTBR0_LO(lower_32_bits(ttbr)));
131 
132 	OUT_RING(ring,
133 		CP_SMMU_TABLE_UPDATE_1_TTBR0_HI(upper_32_bits(ttbr)) |
134 		CP_SMMU_TABLE_UPDATE_1_ASID(asid));
135 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_2_CONTEXTIDR(0));
136 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_3_CONTEXTBANK(0));
137 
138 	/*
139 	 * Write the new TTBR0 to the memstore. This is good for debugging.
140 	 */
141 	OUT_PKT7(ring, CP_MEM_WRITE, 4);
142 	OUT_RING(ring, CP_MEM_WRITE_0_ADDR_LO(lower_32_bits(memptr)));
143 	OUT_RING(ring, CP_MEM_WRITE_1_ADDR_HI(upper_32_bits(memptr)));
144 	OUT_RING(ring, lower_32_bits(ttbr));
145 	OUT_RING(ring, (asid << 16) | upper_32_bits(ttbr));
146 
147 	/*
148 	 * Sync both threads after switching pagetables and enable BR only
149 	 * to make sure BV doesn't race ahead while BR is still switching
150 	 * pagetables.
151 	 */
152 	if (adreno_is_a7xx(&a6xx_gpu->base)) {
153 		OUT_PKT7(ring, CP_THREAD_CONTROL, 1);
154 		OUT_RING(ring, CP_THREAD_CONTROL_0_SYNC_THREADS | CP_SET_THREAD_BR);
155 	}
156 
157 	/*
158 	 * And finally, trigger a uche flush to be sure there isn't anything
159 	 * lingering in that part of the GPU
160 	 */
161 
162 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
163 	OUT_RING(ring, CACHE_INVALIDATE);
164 
165 	if (!sysprof) {
166 		/*
167 		 * Wait for SRAM clear after the pgtable update, so the
168 		 * two can happen in parallel:
169 		 */
170 		OUT_PKT7(ring, CP_WAIT_REG_MEM, 6);
171 		OUT_RING(ring, CP_WAIT_REG_MEM_0_FUNCTION(WRITE_EQ));
172 		OUT_RING(ring, CP_WAIT_REG_MEM_1_POLL_ADDR_LO(
173 				REG_A6XX_RBBM_PERFCTR_SRAM_INIT_STATUS));
174 		OUT_RING(ring, CP_WAIT_REG_MEM_2_POLL_ADDR_HI(0));
175 		OUT_RING(ring, CP_WAIT_REG_MEM_3_REF(0x1));
176 		OUT_RING(ring, CP_WAIT_REG_MEM_4_MASK(0x1));
177 		OUT_RING(ring, CP_WAIT_REG_MEM_5_DELAY_LOOP_CYCLES(0));
178 
179 		if (!adreno_is_a7xx(adreno_gpu)) {
180 			/* Re-enable protected mode: */
181 			OUT_PKT7(ring, CP_SET_PROTECTED_MODE, 1);
182 			OUT_RING(ring, 1);
183 		}
184 	}
185 }
186 
187 static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit)
188 {
189 	unsigned int index = submit->seqno % MSM_GPU_SUBMIT_STATS_COUNT;
190 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
191 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
192 	struct msm_ringbuffer *ring = submit->ring;
193 	unsigned int i, ibs = 0;
194 
195 	a6xx_set_pagetable(a6xx_gpu, ring, submit->queue->ctx);
196 
197 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP(0),
198 		rbmemptr_stats(ring, index, cpcycles_start));
199 
200 	/*
201 	 * For PM4 the GMU register offsets are calculated from the base of the
202 	 * GPU registers so we need to add 0x1a800 to the register value on A630
203 	 * to get the right value from PM4.
204 	 */
205 	get_stats_counter(ring, REG_A6XX_CP_ALWAYS_ON_COUNTER,
206 		rbmemptr_stats(ring, index, alwayson_start));
207 
208 	/* Invalidate CCU depth and color */
209 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
210 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(PC_CCU_INVALIDATE_DEPTH));
211 
212 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
213 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(PC_CCU_INVALIDATE_COLOR));
214 
215 	/* Submit the commands */
216 	for (i = 0; i < submit->nr_cmds; i++) {
217 		switch (submit->cmd[i].type) {
218 		case MSM_SUBMIT_CMD_IB_TARGET_BUF:
219 			break;
220 		case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
221 			if (gpu->cur_ctx_seqno == submit->queue->ctx->seqno)
222 				break;
223 			fallthrough;
224 		case MSM_SUBMIT_CMD_BUF:
225 			OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3);
226 			OUT_RING(ring, lower_32_bits(submit->cmd[i].iova));
227 			OUT_RING(ring, upper_32_bits(submit->cmd[i].iova));
228 			OUT_RING(ring, submit->cmd[i].size);
229 			ibs++;
230 			break;
231 		}
232 
233 		/*
234 		 * Periodically update shadow-wptr if needed, so that we
235 		 * can see partial progress of submits with large # of
236 		 * cmds.. otherwise we could needlessly stall waiting for
237 		 * ringbuffer state, simply due to looking at a shadow
238 		 * rptr value that has not been updated
239 		 */
240 		if ((ibs % 32) == 0)
241 			update_shadow_rptr(gpu, ring);
242 	}
243 
244 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP(0),
245 		rbmemptr_stats(ring, index, cpcycles_end));
246 	get_stats_counter(ring, REG_A6XX_CP_ALWAYS_ON_COUNTER,
247 		rbmemptr_stats(ring, index, alwayson_end));
248 
249 	/* Write the fence to the scratch register */
250 	OUT_PKT4(ring, REG_A6XX_CP_SCRATCH_REG(2), 1);
251 	OUT_RING(ring, submit->seqno);
252 
253 	/*
254 	 * Execute a CACHE_FLUSH_TS event. This will ensure that the
255 	 * timestamp is written to the memory and then triggers the interrupt
256 	 */
257 	OUT_PKT7(ring, CP_EVENT_WRITE, 4);
258 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(CACHE_FLUSH_TS) |
259 		CP_EVENT_WRITE_0_IRQ);
260 	OUT_RING(ring, lower_32_bits(rbmemptr(ring, fence)));
261 	OUT_RING(ring, upper_32_bits(rbmemptr(ring, fence)));
262 	OUT_RING(ring, submit->seqno);
263 
264 	trace_msm_gpu_submit_flush(submit,
265 		gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER));
266 
267 	a6xx_flush(gpu, ring);
268 }
269 
270 static void a7xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit)
271 {
272 	unsigned int index = submit->seqno % MSM_GPU_SUBMIT_STATS_COUNT;
273 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
274 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
275 	struct msm_ringbuffer *ring = submit->ring;
276 	unsigned int i, ibs = 0;
277 
278 	/*
279 	 * Toggle concurrent binning for pagetable switch and set the thread to
280 	 * BR since only it can execute the pagetable switch packets.
281 	 */
282 	OUT_PKT7(ring, CP_THREAD_CONTROL, 1);
283 	OUT_RING(ring, CP_THREAD_CONTROL_0_SYNC_THREADS | CP_SET_THREAD_BR);
284 
285 	a6xx_set_pagetable(a6xx_gpu, ring, submit->queue->ctx);
286 
287 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP(0),
288 		rbmemptr_stats(ring, index, cpcycles_start));
289 	get_stats_counter(ring, REG_A6XX_CP_ALWAYS_ON_COUNTER,
290 		rbmemptr_stats(ring, index, alwayson_start));
291 
292 	OUT_PKT7(ring, CP_THREAD_CONTROL, 1);
293 	OUT_RING(ring, CP_SET_THREAD_BOTH);
294 
295 	OUT_PKT7(ring, CP_SET_MARKER, 1);
296 	OUT_RING(ring, 0x101); /* IFPC disable */
297 
298 	OUT_PKT7(ring, CP_SET_MARKER, 1);
299 	OUT_RING(ring, 0x00d); /* IB1LIST start */
300 
301 	/* Submit the commands */
302 	for (i = 0; i < submit->nr_cmds; i++) {
303 		switch (submit->cmd[i].type) {
304 		case MSM_SUBMIT_CMD_IB_TARGET_BUF:
305 			break;
306 		case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
307 			if (gpu->cur_ctx_seqno == submit->queue->ctx->seqno)
308 				break;
309 			fallthrough;
310 		case MSM_SUBMIT_CMD_BUF:
311 			OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3);
312 			OUT_RING(ring, lower_32_bits(submit->cmd[i].iova));
313 			OUT_RING(ring, upper_32_bits(submit->cmd[i].iova));
314 			OUT_RING(ring, submit->cmd[i].size);
315 			ibs++;
316 			break;
317 		}
318 
319 		/*
320 		 * Periodically update shadow-wptr if needed, so that we
321 		 * can see partial progress of submits with large # of
322 		 * cmds.. otherwise we could needlessly stall waiting for
323 		 * ringbuffer state, simply due to looking at a shadow
324 		 * rptr value that has not been updated
325 		 */
326 		if ((ibs % 32) == 0)
327 			update_shadow_rptr(gpu, ring);
328 	}
329 
330 	OUT_PKT7(ring, CP_SET_MARKER, 1);
331 	OUT_RING(ring, 0x00e); /* IB1LIST end */
332 
333 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP(0),
334 		rbmemptr_stats(ring, index, cpcycles_end));
335 	get_stats_counter(ring, REG_A6XX_CP_ALWAYS_ON_COUNTER,
336 		rbmemptr_stats(ring, index, alwayson_end));
337 
338 	/* Write the fence to the scratch register */
339 	OUT_PKT4(ring, REG_A6XX_CP_SCRATCH_REG(2), 1);
340 	OUT_RING(ring, submit->seqno);
341 
342 	OUT_PKT7(ring, CP_THREAD_CONTROL, 1);
343 	OUT_RING(ring, CP_SET_THREAD_BR);
344 
345 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
346 	OUT_RING(ring, CCU_INVALIDATE_DEPTH);
347 
348 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
349 	OUT_RING(ring, CCU_INVALIDATE_COLOR);
350 
351 	OUT_PKT7(ring, CP_THREAD_CONTROL, 1);
352 	OUT_RING(ring, CP_SET_THREAD_BV);
353 
354 	/*
355 	 * Make sure the timestamp is committed once BV pipe is
356 	 * completely done with this submission.
357 	 */
358 	OUT_PKT7(ring, CP_EVENT_WRITE, 4);
359 	OUT_RING(ring, CACHE_CLEAN | BIT(27));
360 	OUT_RING(ring, lower_32_bits(rbmemptr(ring, bv_fence)));
361 	OUT_RING(ring, upper_32_bits(rbmemptr(ring, bv_fence)));
362 	OUT_RING(ring, submit->seqno);
363 
364 	OUT_PKT7(ring, CP_THREAD_CONTROL, 1);
365 	OUT_RING(ring, CP_SET_THREAD_BR);
366 
367 	/*
368 	 * This makes sure that BR doesn't race ahead and commit
369 	 * timestamp to memstore while BV is still processing
370 	 * this submission.
371 	 */
372 	OUT_PKT7(ring, CP_WAIT_TIMESTAMP, 4);
373 	OUT_RING(ring, 0);
374 	OUT_RING(ring, lower_32_bits(rbmemptr(ring, bv_fence)));
375 	OUT_RING(ring, upper_32_bits(rbmemptr(ring, bv_fence)));
376 	OUT_RING(ring, submit->seqno);
377 
378 	/* write the ringbuffer timestamp */
379 	OUT_PKT7(ring, CP_EVENT_WRITE, 4);
380 	OUT_RING(ring, CACHE_CLEAN | CP_EVENT_WRITE_0_IRQ | BIT(27));
381 	OUT_RING(ring, lower_32_bits(rbmemptr(ring, fence)));
382 	OUT_RING(ring, upper_32_bits(rbmemptr(ring, fence)));
383 	OUT_RING(ring, submit->seqno);
384 
385 	OUT_PKT7(ring, CP_THREAD_CONTROL, 1);
386 	OUT_RING(ring, CP_SET_THREAD_BOTH);
387 
388 	OUT_PKT7(ring, CP_SET_MARKER, 1);
389 	OUT_RING(ring, 0x100); /* IFPC enable */
390 
391 	trace_msm_gpu_submit_flush(submit,
392 		gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER));
393 
394 	a6xx_flush(gpu, ring);
395 }
396 
397 const struct adreno_reglist a612_hwcg[] = {
398 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222},
399 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
400 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000081},
401 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000f3cf},
402 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222222},
403 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
404 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
405 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
406 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
407 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
408 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
409 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
410 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
411 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
412 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
413 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
414 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
415 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01202222},
416 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
417 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040f00},
418 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05522022},
419 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
420 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
421 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
422 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
423 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
424 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x02222222},
425 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
426 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
427 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
428 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
429 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
430 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
431 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
432 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
433 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
434 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
435 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
436 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
437 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
438 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
439 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
440 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
441 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
442 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
443 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
444 	{},
445 };
446 
447 /* For a615 family (a615, a616, a618 and a619) */
448 const struct adreno_reglist a615_hwcg[] = {
449 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0,  0x02222222},
450 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
451 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
452 	{REG_A6XX_RBBM_CLOCK_HYST_SP0,  0x0000F3CF},
453 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0,  0x02222222},
454 	{REG_A6XX_RBBM_CLOCK_CNTL_TP1,  0x02222222},
455 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
456 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP1, 0x22222222},
457 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
458 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP1, 0x22222222},
459 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
460 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP1, 0x00022222},
461 	{REG_A6XX_RBBM_CLOCK_HYST_TP0,  0x77777777},
462 	{REG_A6XX_RBBM_CLOCK_HYST_TP1,  0x77777777},
463 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
464 	{REG_A6XX_RBBM_CLOCK_HYST2_TP1, 0x77777777},
465 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
466 	{REG_A6XX_RBBM_CLOCK_HYST3_TP1, 0x77777777},
467 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
468 	{REG_A6XX_RBBM_CLOCK_HYST4_TP1, 0x00077777},
469 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
470 	{REG_A6XX_RBBM_CLOCK_DELAY_TP1, 0x11111111},
471 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
472 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP1, 0x11111111},
473 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
474 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP1, 0x11111111},
475 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
476 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP1, 0x00011111},
477 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE,  0x22222222},
478 	{REG_A6XX_RBBM_CLOCK_CNTL2_UCHE, 0x22222222},
479 	{REG_A6XX_RBBM_CLOCK_CNTL3_UCHE, 0x22222222},
480 	{REG_A6XX_RBBM_CLOCK_CNTL4_UCHE, 0x00222222},
481 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE,  0x00000004},
482 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
483 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
484 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x00002222},
485 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002020},
486 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU1, 0x00002220},
487 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU2, 0x00002220},
488 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU3, 0x00002220},
489 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
490 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU1, 0x00040F00},
491 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU2, 0x00040F00},
492 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU3, 0x00040F00},
493 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05022022},
494 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
495 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
496 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
497 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
498 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
499 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
500 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
501 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
502 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
503 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
504 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
505 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
506 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
507 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
508 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
509 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
510 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
511 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
512 	{},
513 };
514 
515 const struct adreno_reglist a630_hwcg[] = {
516 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222},
517 	{REG_A6XX_RBBM_CLOCK_CNTL_SP1, 0x22222222},
518 	{REG_A6XX_RBBM_CLOCK_CNTL_SP2, 0x22222222},
519 	{REG_A6XX_RBBM_CLOCK_CNTL_SP3, 0x22222222},
520 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02022220},
521 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP1, 0x02022220},
522 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP2, 0x02022220},
523 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP3, 0x02022220},
524 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
525 	{REG_A6XX_RBBM_CLOCK_DELAY_SP1, 0x00000080},
526 	{REG_A6XX_RBBM_CLOCK_DELAY_SP2, 0x00000080},
527 	{REG_A6XX_RBBM_CLOCK_DELAY_SP3, 0x00000080},
528 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000f3cf},
529 	{REG_A6XX_RBBM_CLOCK_HYST_SP1, 0x0000f3cf},
530 	{REG_A6XX_RBBM_CLOCK_HYST_SP2, 0x0000f3cf},
531 	{REG_A6XX_RBBM_CLOCK_HYST_SP3, 0x0000f3cf},
532 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
533 	{REG_A6XX_RBBM_CLOCK_CNTL_TP1, 0x02222222},
534 	{REG_A6XX_RBBM_CLOCK_CNTL_TP2, 0x02222222},
535 	{REG_A6XX_RBBM_CLOCK_CNTL_TP3, 0x02222222},
536 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
537 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP1, 0x22222222},
538 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP2, 0x22222222},
539 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP3, 0x22222222},
540 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
541 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP1, 0x22222222},
542 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP2, 0x22222222},
543 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP3, 0x22222222},
544 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
545 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP1, 0x00022222},
546 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP2, 0x00022222},
547 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP3, 0x00022222},
548 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
549 	{REG_A6XX_RBBM_CLOCK_HYST_TP1, 0x77777777},
550 	{REG_A6XX_RBBM_CLOCK_HYST_TP2, 0x77777777},
551 	{REG_A6XX_RBBM_CLOCK_HYST_TP3, 0x77777777},
552 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
553 	{REG_A6XX_RBBM_CLOCK_HYST2_TP1, 0x77777777},
554 	{REG_A6XX_RBBM_CLOCK_HYST2_TP2, 0x77777777},
555 	{REG_A6XX_RBBM_CLOCK_HYST2_TP3, 0x77777777},
556 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
557 	{REG_A6XX_RBBM_CLOCK_HYST3_TP1, 0x77777777},
558 	{REG_A6XX_RBBM_CLOCK_HYST3_TP2, 0x77777777},
559 	{REG_A6XX_RBBM_CLOCK_HYST3_TP3, 0x77777777},
560 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
561 	{REG_A6XX_RBBM_CLOCK_HYST4_TP1, 0x00077777},
562 	{REG_A6XX_RBBM_CLOCK_HYST4_TP2, 0x00077777},
563 	{REG_A6XX_RBBM_CLOCK_HYST4_TP3, 0x00077777},
564 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
565 	{REG_A6XX_RBBM_CLOCK_DELAY_TP1, 0x11111111},
566 	{REG_A6XX_RBBM_CLOCK_DELAY_TP2, 0x11111111},
567 	{REG_A6XX_RBBM_CLOCK_DELAY_TP3, 0x11111111},
568 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
569 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP1, 0x11111111},
570 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP2, 0x11111111},
571 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP3, 0x11111111},
572 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
573 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP1, 0x11111111},
574 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP2, 0x11111111},
575 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP3, 0x11111111},
576 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
577 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP1, 0x00011111},
578 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP2, 0x00011111},
579 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP3, 0x00011111},
580 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
581 	{REG_A6XX_RBBM_CLOCK_CNTL2_UCHE, 0x22222222},
582 	{REG_A6XX_RBBM_CLOCK_CNTL3_UCHE, 0x22222222},
583 	{REG_A6XX_RBBM_CLOCK_CNTL4_UCHE, 0x00222222},
584 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
585 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
586 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
587 	{REG_A6XX_RBBM_CLOCK_CNTL_RB1, 0x22222222},
588 	{REG_A6XX_RBBM_CLOCK_CNTL_RB2, 0x22222222},
589 	{REG_A6XX_RBBM_CLOCK_CNTL_RB3, 0x22222222},
590 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x00002222},
591 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB1, 0x00002222},
592 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB2, 0x00002222},
593 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB3, 0x00002222},
594 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
595 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU1, 0x00002220},
596 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU2, 0x00002220},
597 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU3, 0x00002220},
598 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040f00},
599 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU1, 0x00040f00},
600 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU2, 0x00040f00},
601 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU3, 0x00040f00},
602 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05022022},
603 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
604 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
605 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
606 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
607 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
608 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
609 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
610 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
611 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
612 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
613 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
614 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
615 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
616 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
617 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
618 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
619 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
620 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
621 	{},
622 };
623 
624 const struct adreno_reglist a640_hwcg[] = {
625 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
626 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
627 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
628 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
629 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
630 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
631 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
632 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
633 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
634 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
635 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
636 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
637 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
638 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
639 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
640 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
641 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
642 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
643 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
644 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
645 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05222022},
646 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
647 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
648 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
649 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
650 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
651 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
652 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
653 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
654 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
655 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
656 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
657 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
658 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
659 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
660 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
661 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
662 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
663 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
664 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000000},
665 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
666 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
667 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
668 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
669 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
670 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
671 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
672 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
673 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
674 	{},
675 };
676 
677 const struct adreno_reglist a650_hwcg[] = {
678 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
679 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
680 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
681 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
682 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
683 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
684 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
685 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
686 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
687 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
688 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
689 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
690 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
691 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
692 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
693 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
694 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
695 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
696 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
697 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
698 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022},
699 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
700 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
701 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
702 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
703 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
704 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
705 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
706 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
707 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
708 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
709 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
710 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
711 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
712 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
713 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
714 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
715 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
716 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
717 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000777},
718 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
719 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
720 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
721 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
722 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
723 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
724 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
725 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
726 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
727 	{},
728 };
729 
730 const struct adreno_reglist a660_hwcg[] = {
731 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
732 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
733 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
734 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
735 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222222},
736 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
737 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
738 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
739 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
740 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
741 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
742 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
743 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
744 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
745 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
746 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
747 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
748 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
749 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
750 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
751 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022},
752 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
753 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
754 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
755 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
756 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
757 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
758 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
759 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
760 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
761 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
762 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
763 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
764 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
765 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
766 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
767 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
768 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
769 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
770 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000000},
771 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
772 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
773 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
774 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
775 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
776 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
777 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
778 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
779 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
780 	{},
781 };
782 
783 const struct adreno_reglist a690_hwcg[] = {
784 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
785 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
786 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
787 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
788 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222222},
789 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
790 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
791 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
792 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
793 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
794 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
795 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
796 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
797 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
798 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
799 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
800 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
801 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
802 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
803 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
804 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022},
805 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
806 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
807 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
808 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
809 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
810 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
811 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
812 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
813 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
814 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
815 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
816 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
817 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
818 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
819 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
820 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
821 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
822 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
823 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000000},
824 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
825 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
826 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
827 	{REG_A6XX_RBBM_CLOCK_CNTL, 0x8AA8AA82},
828 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
829 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
830 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
831 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
832 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
833 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
834 	{REG_A6XX_GPU_GMU_AO_GMU_CGC_MODE_CNTL, 0x20200},
835 	{REG_A6XX_GPU_GMU_AO_GMU_CGC_DELAY_CNTL, 0x10111},
836 	{REG_A6XX_GPU_GMU_AO_GMU_CGC_HYST_CNTL, 0x5555},
837 	{}
838 };
839 
840 const struct adreno_reglist a702_hwcg[] = {
841 	{ REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222 },
842 	{ REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220 },
843 	{ REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000081 },
844 	{ REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000f3cf },
845 	{ REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222222 },
846 	{ REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222 },
847 	{ REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222 },
848 	{ REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222 },
849 	{ REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111 },
850 	{ REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111 },
851 	{ REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111 },
852 	{ REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111 },
853 	{ REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777 },
854 	{ REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777 },
855 	{ REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777 },
856 	{ REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777 },
857 	{ REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222 },
858 	{ REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01202222 },
859 	{ REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220 },
860 	{ REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040f00 },
861 	{ REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05522022 },
862 	{ REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555 },
863 	{ REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011 },
864 	{ REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044 },
865 	{ REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222 },
866 	{ REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222 },
867 	{ REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x02222222 },
868 	{ REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002 },
869 	{ REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222 },
870 	{ REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000 },
871 	{ REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222 },
872 	{ REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200 },
873 	{ REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000 },
874 	{ REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000 },
875 	{ REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000 },
876 	{ REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004 },
877 	{ REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000 },
878 	{ REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222 },
879 	{ REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004 },
880 	{ REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002 },
881 	{ REG_A6XX_RBBM_ISDB_CNT, 0x00000182 },
882 	{ REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000 },
883 	{ REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000 },
884 	{ REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222 },
885 	{ REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111 },
886 	{ REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555 },
887 	{ REG_A6XX_RBBM_CLOCK_CNTL_FCHE, 0x00000222 },
888 	{ REG_A6XX_RBBM_CLOCK_DELAY_FCHE, 0x00000000 },
889 	{ REG_A6XX_RBBM_CLOCK_HYST_FCHE, 0x00000000 },
890 	{ REG_A6XX_RBBM_CLOCK_CNTL_GLC, 0x00222222 },
891 	{ REG_A6XX_RBBM_CLOCK_DELAY_GLC, 0x00000000 },
892 	{ REG_A6XX_RBBM_CLOCK_HYST_GLC, 0x00000000 },
893 	{ REG_A6XX_RBBM_CLOCK_CNTL_MHUB, 0x00000002 },
894 	{ REG_A6XX_RBBM_CLOCK_DELAY_MHUB, 0x00000000 },
895 	{ REG_A6XX_RBBM_CLOCK_HYST_MHUB, 0x00000000 },
896 	{}
897 };
898 
899 const struct adreno_reglist a730_hwcg[] = {
900 	{ REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222 },
901 	{ REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02022222 },
902 	{ REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000f3cf },
903 	{ REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080 },
904 	{ REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222220 },
905 	{ REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222 },
906 	{ REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222 },
907 	{ REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00222222 },
908 	{ REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777 },
909 	{ REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777 },
910 	{ REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777 },
911 	{ REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777 },
912 	{ REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111 },
913 	{ REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111 },
914 	{ REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111 },
915 	{ REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111 },
916 	{ REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222 },
917 	{ REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004 },
918 	{ REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002 },
919 	{ REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222 },
920 	{ REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222 },
921 	{ REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220 },
922 	{ REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x44000f00 },
923 	{ REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022 },
924 	{ REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00555555 },
925 	{ REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011 },
926 	{ REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00440044 },
927 	{ REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222 },
928 	{ REG_A7XX_RBBM_CLOCK_MODE2_GRAS, 0x00000222 },
929 	{ REG_A7XX_RBBM_CLOCK_MODE_BV_GRAS, 0x00222222 },
930 	{ REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x02222223 },
931 	{ REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222 },
932 	{ REG_A7XX_RBBM_CLOCK_MODE_BV_GPC, 0x00222222 },
933 	{ REG_A7XX_RBBM_CLOCK_MODE_BV_VFD, 0x00002222 },
934 	{ REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000 },
935 	{ REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004 },
936 	{ REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000 },
937 	{ REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000 },
938 	{ REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200 },
939 	{ REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222 },
940 	{ REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222 },
941 	{ REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000 },
942 	{ REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000 },
943 	{ REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002 },
944 	{ REG_A7XX_RBBM_CLOCK_MODE_BV_LRZ, 0x55555552 },
945 	{ REG_A7XX_RBBM_CLOCK_MODE_CP, 0x00000223 },
946 	{ REG_A6XX_RBBM_CLOCK_CNTL, 0x8aa8aa82 },
947 	{ REG_A6XX_RBBM_ISDB_CNT, 0x00000182 },
948 	{ REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000 },
949 	{ REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000 },
950 	{ REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222 },
951 	{ REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111 },
952 	{ REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555 },
953 	{},
954 };
955 
956 const struct adreno_reglist a740_hwcg[] = {
957 	{ REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222 },
958 	{ REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x22022222 },
959 	{ REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x003cf3cf },
960 	{ REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080 },
961 	{ REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x22222220 },
962 	{ REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222 },
963 	{ REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222 },
964 	{ REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00222222 },
965 	{ REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777 },
966 	{ REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777 },
967 	{ REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777 },
968 	{ REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777 },
969 	{ REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111 },
970 	{ REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111 },
971 	{ REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111 },
972 	{ REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111 },
973 	{ REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222 },
974 	{ REG_A6XX_RBBM_CLOCK_CNTL2_UCHE, 0x00222222 },
975 	{ REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000444 },
976 	{ REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000222 },
977 	{ REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222 },
978 	{ REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222 },
979 	{ REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220 },
980 	{ REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x44000f00 },
981 	{ REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022 },
982 	{ REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00555555 },
983 	{ REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011 },
984 	{ REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00440044 },
985 	{ REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222 },
986 	{ REG_A7XX_RBBM_CLOCK_MODE2_GRAS, 0x00000222 },
987 	{ REG_A7XX_RBBM_CLOCK_MODE_BV_GRAS, 0x00222222 },
988 	{ REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x02222223 },
989 	{ REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00222222 },
990 	{ REG_A7XX_RBBM_CLOCK_MODE_BV_GPC, 0x00222222 },
991 	{ REG_A7XX_RBBM_CLOCK_MODE_BV_VFD, 0x00002222 },
992 	{ REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000 },
993 	{ REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004 },
994 	{ REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000 },
995 	{ REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00000000 },
996 	{ REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200 },
997 	{ REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00000000 },
998 	{ REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222 },
999 	{ REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000 },
1000 	{ REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000 },
1001 	{ REG_A7XX_RBBM_CLOCK_MODE_BV_LRZ, 0x55555552 },
1002 	{ REG_A7XX_RBBM_CLOCK_HYST2_VFD, 0x00000000 },
1003 	{ REG_A7XX_RBBM_CLOCK_MODE_CP, 0x00000222 },
1004 	{ REG_A6XX_RBBM_CLOCK_CNTL, 0x8aa8aa82 },
1005 	{ REG_A6XX_RBBM_ISDB_CNT, 0x00000182 },
1006 	{ REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000 },
1007 	{ REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000 },
1008 	{ REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222 },
1009 	{ REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111 },
1010 	{ REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555 },
1011 	{},
1012 };
1013 
1014 static void a6xx_set_hwcg(struct msm_gpu *gpu, bool state)
1015 {
1016 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1017 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1018 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
1019 	const struct adreno_reglist *reg;
1020 	unsigned int i;
1021 	u32 val, clock_cntl_on, cgc_mode;
1022 
1023 	if (!(adreno_gpu->info->hwcg || adreno_is_a7xx(adreno_gpu)))
1024 		return;
1025 
1026 	if (adreno_is_a630(adreno_gpu))
1027 		clock_cntl_on = 0x8aa8aa02;
1028 	else if (adreno_is_a610(adreno_gpu))
1029 		clock_cntl_on = 0xaaa8aa82;
1030 	else if (adreno_is_a702(adreno_gpu))
1031 		clock_cntl_on = 0xaaaaaa82;
1032 	else
1033 		clock_cntl_on = 0x8aa8aa82;
1034 
1035 	if (adreno_is_a7xx(adreno_gpu)) {
1036 		cgc_mode = adreno_is_a740_family(adreno_gpu) ? 0x20222 : 0x20000;
1037 
1038 		gmu_write(&a6xx_gpu->gmu, REG_A6XX_GPU_GMU_AO_GMU_CGC_MODE_CNTL,
1039 			  state ? cgc_mode : 0);
1040 		gmu_write(&a6xx_gpu->gmu, REG_A6XX_GPU_GMU_AO_GMU_CGC_DELAY_CNTL,
1041 			  state ? 0x10111 : 0);
1042 		gmu_write(&a6xx_gpu->gmu, REG_A6XX_GPU_GMU_AO_GMU_CGC_HYST_CNTL,
1043 			  state ? 0x5555 : 0);
1044 	}
1045 
1046 	if (!adreno_gpu->info->hwcg) {
1047 		gpu_write(gpu, REG_A7XX_RBBM_CLOCK_CNTL_GLOBAL, 1);
1048 		gpu_write(gpu, REG_A7XX_RBBM_CGC_GLOBAL_LOAD_CMD, state ? 1 : 0);
1049 
1050 		if (state) {
1051 			gpu_write(gpu, REG_A7XX_RBBM_CGC_P2S_TRIG_CMD, 1);
1052 
1053 			if (gpu_poll_timeout(gpu, REG_A7XX_RBBM_CGC_P2S_STATUS, val,
1054 					     val & A7XX_RBBM_CGC_P2S_STATUS_TXDONE, 1, 10)) {
1055 				dev_err(&gpu->pdev->dev, "RBBM_CGC_P2S_STATUS TXDONE Poll failed\n");
1056 				return;
1057 			}
1058 
1059 			gpu_write(gpu, REG_A7XX_RBBM_CLOCK_CNTL_GLOBAL, 0);
1060 		}
1061 
1062 		return;
1063 	}
1064 
1065 	val = gpu_read(gpu, REG_A6XX_RBBM_CLOCK_CNTL);
1066 
1067 	/* Don't re-program the registers if they are already correct */
1068 	if ((!state && !val) || (state && (val == clock_cntl_on)))
1069 		return;
1070 
1071 	/* Disable SP clock before programming HWCG registers */
1072 	if (!adreno_is_a610_family(adreno_gpu) && !adreno_is_a7xx(adreno_gpu))
1073 		gmu_rmw(gmu, REG_A6XX_GPU_GMU_GX_SPTPRAC_CLOCK_CONTROL, 1, 0);
1074 
1075 	for (i = 0; (reg = &adreno_gpu->info->hwcg[i], reg->offset); i++)
1076 		gpu_write(gpu, reg->offset, state ? reg->value : 0);
1077 
1078 	/* Enable SP clock */
1079 	if (!adreno_is_a610_family(adreno_gpu) && !adreno_is_a7xx(adreno_gpu))
1080 		gmu_rmw(gmu, REG_A6XX_GPU_GMU_GX_SPTPRAC_CLOCK_CONTROL, 0, 1);
1081 
1082 	gpu_write(gpu, REG_A6XX_RBBM_CLOCK_CNTL, state ? clock_cntl_on : 0);
1083 }
1084 
1085 /* For a615, a616, a618, a619, a630, a640 and a680 */
1086 static const u32 a6xx_protect[] = {
1087 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
1088 	A6XX_PROTECT_RDONLY(0x00501, 0x0005),
1089 	A6XX_PROTECT_RDONLY(0x0050b, 0x02f4),
1090 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
1091 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
1092 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
1093 	A6XX_PROTECT_NORDWR(0x00800, 0x0082),
1094 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
1095 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
1096 	A6XX_PROTECT_RDONLY(0x008de, 0x00ae),
1097 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
1098 	A6XX_PROTECT_NORDWR(0x0098d, 0x0272),
1099 	A6XX_PROTECT_NORDWR(0x00e00, 0x0001),
1100 	A6XX_PROTECT_NORDWR(0x00e03, 0x000c),
1101 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
1102 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
1103 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
1104 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
1105 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
1106 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
1107 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
1108 	A6XX_PROTECT_NORDWR(0x09e70, 0x0001),
1109 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
1110 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
1111 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
1112 	A6XX_PROTECT_NORDWR(0x0ae50, 0x032f),
1113 	A6XX_PROTECT_NORDWR(0x0b604, 0x0000),
1114 	A6XX_PROTECT_NORDWR(0x0be02, 0x0001),
1115 	A6XX_PROTECT_NORDWR(0x0be20, 0x17df),
1116 	A6XX_PROTECT_NORDWR(0x0f000, 0x0bff),
1117 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
1118 	A6XX_PROTECT_NORDWR(0x11c00, 0x0000), /* note: infinite range */
1119 };
1120 
1121 /* These are for a620 and a650 */
1122 static const u32 a650_protect[] = {
1123 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
1124 	A6XX_PROTECT_RDONLY(0x00501, 0x0005),
1125 	A6XX_PROTECT_RDONLY(0x0050b, 0x02f4),
1126 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
1127 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
1128 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
1129 	A6XX_PROTECT_NORDWR(0x00800, 0x0082),
1130 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
1131 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
1132 	A6XX_PROTECT_RDONLY(0x008de, 0x00ae),
1133 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
1134 	A6XX_PROTECT_NORDWR(0x0098d, 0x0272),
1135 	A6XX_PROTECT_NORDWR(0x00e00, 0x0001),
1136 	A6XX_PROTECT_NORDWR(0x00e03, 0x000c),
1137 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
1138 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
1139 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
1140 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
1141 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
1142 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
1143 	A6XX_PROTECT_NORDWR(0x08e80, 0x027f),
1144 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
1145 	A6XX_PROTECT_NORDWR(0x09e60, 0x0011),
1146 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
1147 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
1148 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
1149 	A6XX_PROTECT_NORDWR(0x0ae50, 0x032f),
1150 	A6XX_PROTECT_NORDWR(0x0b604, 0x0000),
1151 	A6XX_PROTECT_NORDWR(0x0b608, 0x0007),
1152 	A6XX_PROTECT_NORDWR(0x0be02, 0x0001),
1153 	A6XX_PROTECT_NORDWR(0x0be20, 0x17df),
1154 	A6XX_PROTECT_NORDWR(0x0f000, 0x0bff),
1155 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
1156 	A6XX_PROTECT_NORDWR(0x18400, 0x1fff),
1157 	A6XX_PROTECT_NORDWR(0x1a800, 0x1fff),
1158 	A6XX_PROTECT_NORDWR(0x1f400, 0x0443),
1159 	A6XX_PROTECT_RDONLY(0x1f844, 0x007b),
1160 	A6XX_PROTECT_NORDWR(0x1f887, 0x001b),
1161 	A6XX_PROTECT_NORDWR(0x1f8c0, 0x0000), /* note: infinite range */
1162 };
1163 
1164 /* These are for a635 and a660 */
1165 static const u32 a660_protect[] = {
1166 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
1167 	A6XX_PROTECT_RDONLY(0x00501, 0x0005),
1168 	A6XX_PROTECT_RDONLY(0x0050b, 0x02f4),
1169 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
1170 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
1171 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
1172 	A6XX_PROTECT_NORDWR(0x00800, 0x0082),
1173 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
1174 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
1175 	A6XX_PROTECT_RDONLY(0x008de, 0x00ae),
1176 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
1177 	A6XX_PROTECT_NORDWR(0x0098d, 0x0272),
1178 	A6XX_PROTECT_NORDWR(0x00e00, 0x0001),
1179 	A6XX_PROTECT_NORDWR(0x00e03, 0x000c),
1180 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
1181 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
1182 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
1183 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
1184 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
1185 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
1186 	A6XX_PROTECT_NORDWR(0x08e80, 0x027f),
1187 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
1188 	A6XX_PROTECT_NORDWR(0x09e60, 0x0011),
1189 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
1190 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
1191 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
1192 	A6XX_PROTECT_NORDWR(0x0ae50, 0x012f),
1193 	A6XX_PROTECT_NORDWR(0x0b604, 0x0000),
1194 	A6XX_PROTECT_NORDWR(0x0b608, 0x0006),
1195 	A6XX_PROTECT_NORDWR(0x0be02, 0x0001),
1196 	A6XX_PROTECT_NORDWR(0x0be20, 0x015f),
1197 	A6XX_PROTECT_NORDWR(0x0d000, 0x05ff),
1198 	A6XX_PROTECT_NORDWR(0x0f000, 0x0bff),
1199 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
1200 	A6XX_PROTECT_NORDWR(0x18400, 0x1fff),
1201 	A6XX_PROTECT_NORDWR(0x1a400, 0x1fff),
1202 	A6XX_PROTECT_NORDWR(0x1f400, 0x0443),
1203 	A6XX_PROTECT_RDONLY(0x1f844, 0x007b),
1204 	A6XX_PROTECT_NORDWR(0x1f860, 0x0000),
1205 	A6XX_PROTECT_NORDWR(0x1f887, 0x001b),
1206 	A6XX_PROTECT_NORDWR(0x1f8c0, 0x0000), /* note: infinite range */
1207 };
1208 
1209 /* These are for a690 */
1210 static const u32 a690_protect[] = {
1211 	A6XX_PROTECT_RDONLY(0x00000, 0x004ff),
1212 	A6XX_PROTECT_RDONLY(0x00501, 0x00001),
1213 	A6XX_PROTECT_RDONLY(0x0050b, 0x002f4),
1214 	A6XX_PROTECT_NORDWR(0x0050e, 0x00000),
1215 	A6XX_PROTECT_NORDWR(0x00510, 0x00000),
1216 	A6XX_PROTECT_NORDWR(0x00534, 0x00000),
1217 	A6XX_PROTECT_NORDWR(0x00800, 0x00082),
1218 	A6XX_PROTECT_NORDWR(0x008a0, 0x00008),
1219 	A6XX_PROTECT_NORDWR(0x008ab, 0x00024),
1220 	A6XX_PROTECT_RDONLY(0x008de, 0x000ae),
1221 	A6XX_PROTECT_NORDWR(0x00900, 0x0004d),
1222 	A6XX_PROTECT_NORDWR(0x0098d, 0x00272),
1223 	A6XX_PROTECT_NORDWR(0x00e00, 0x00001),
1224 	A6XX_PROTECT_NORDWR(0x00e03, 0x0000c),
1225 	A6XX_PROTECT_NORDWR(0x03c00, 0x000c3),
1226 	A6XX_PROTECT_RDONLY(0x03cc4, 0x01fff),
1227 	A6XX_PROTECT_NORDWR(0x08630, 0x001cf),
1228 	A6XX_PROTECT_NORDWR(0x08e00, 0x00000),
1229 	A6XX_PROTECT_NORDWR(0x08e08, 0x00007),
1230 	A6XX_PROTECT_NORDWR(0x08e50, 0x0001f),
1231 	A6XX_PROTECT_NORDWR(0x08e80, 0x0027f),
1232 	A6XX_PROTECT_NORDWR(0x09624, 0x001db),
1233 	A6XX_PROTECT_NORDWR(0x09e60, 0x00011),
1234 	A6XX_PROTECT_NORDWR(0x09e78, 0x00187),
1235 	A6XX_PROTECT_NORDWR(0x0a630, 0x001cf),
1236 	A6XX_PROTECT_NORDWR(0x0ae02, 0x00000),
1237 	A6XX_PROTECT_NORDWR(0x0ae50, 0x0012f),
1238 	A6XX_PROTECT_NORDWR(0x0b604, 0x00000),
1239 	A6XX_PROTECT_NORDWR(0x0b608, 0x00006),
1240 	A6XX_PROTECT_NORDWR(0x0be02, 0x00001),
1241 	A6XX_PROTECT_NORDWR(0x0be20, 0x0015f),
1242 	A6XX_PROTECT_NORDWR(0x0d000, 0x005ff),
1243 	A6XX_PROTECT_NORDWR(0x0f000, 0x00bff),
1244 	A6XX_PROTECT_RDONLY(0x0fc00, 0x01fff),
1245 	A6XX_PROTECT_NORDWR(0x11c00, 0x00000), /*note: infiite range */
1246 };
1247 
1248 static const u32 a730_protect[] = {
1249 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
1250 	A6XX_PROTECT_RDONLY(0x0050b, 0x0058),
1251 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
1252 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
1253 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
1254 	A6XX_PROTECT_RDONLY(0x005fb, 0x009d),
1255 	A6XX_PROTECT_NORDWR(0x00699, 0x01e9),
1256 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
1257 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
1258 	/* 0x008d0-0x008dd are unprotected on purpose for tools like perfetto */
1259 	A6XX_PROTECT_RDONLY(0x008de, 0x0154),
1260 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
1261 	A6XX_PROTECT_NORDWR(0x0098d, 0x00b2),
1262 	A6XX_PROTECT_NORDWR(0x00a41, 0x01be),
1263 	A6XX_PROTECT_NORDWR(0x00df0, 0x0001),
1264 	A6XX_PROTECT_NORDWR(0x00e01, 0x0000),
1265 	A6XX_PROTECT_NORDWR(0x00e07, 0x0008),
1266 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
1267 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
1268 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
1269 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
1270 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
1271 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
1272 	A6XX_PROTECT_NORDWR(0x08e80, 0x0280),
1273 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
1274 	A6XX_PROTECT_NORDWR(0x09e40, 0x0000),
1275 	A6XX_PROTECT_NORDWR(0x09e64, 0x000d),
1276 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
1277 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
1278 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
1279 	A6XX_PROTECT_NORDWR(0x0ae50, 0x000f),
1280 	A6XX_PROTECT_NORDWR(0x0ae66, 0x0003),
1281 	A6XX_PROTECT_NORDWR(0x0ae6f, 0x0003),
1282 	A6XX_PROTECT_NORDWR(0x0b604, 0x0003),
1283 	A6XX_PROTECT_NORDWR(0x0ec00, 0x0fff),
1284 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
1285 	A6XX_PROTECT_NORDWR(0x18400, 0x0053),
1286 	A6XX_PROTECT_RDONLY(0x18454, 0x0004),
1287 	A6XX_PROTECT_NORDWR(0x18459, 0x1fff),
1288 	A6XX_PROTECT_NORDWR(0x1a459, 0x1fff),
1289 	A6XX_PROTECT_NORDWR(0x1c459, 0x1fff),
1290 	A6XX_PROTECT_NORDWR(0x1f400, 0x0443),
1291 	A6XX_PROTECT_RDONLY(0x1f844, 0x007b),
1292 	A6XX_PROTECT_NORDWR(0x1f860, 0x0000),
1293 	A6XX_PROTECT_NORDWR(0x1f878, 0x002a),
1294 	/* CP_PROTECT_REG[44, 46] are left untouched! */
1295 	0,
1296 	0,
1297 	0,
1298 	A6XX_PROTECT_NORDWR(0x1f8c0, 0x00000),
1299 };
1300 
1301 static void a6xx_set_cp_protect(struct msm_gpu *gpu)
1302 {
1303 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1304 	const u32 *regs = a6xx_protect;
1305 	unsigned i, count, count_max;
1306 
1307 	if (adreno_is_a650(adreno_gpu) || adreno_is_a702(adreno_gpu)) {
1308 		regs = a650_protect;
1309 		count = ARRAY_SIZE(a650_protect);
1310 		count_max = 48;
1311 		BUILD_BUG_ON(ARRAY_SIZE(a650_protect) > 48);
1312 	} else if (adreno_is_a690(adreno_gpu)) {
1313 		regs = a690_protect;
1314 		count = ARRAY_SIZE(a690_protect);
1315 		count_max = 48;
1316 		BUILD_BUG_ON(ARRAY_SIZE(a690_protect) > 48);
1317 	} else if (adreno_is_a660_family(adreno_gpu)) {
1318 		regs = a660_protect;
1319 		count = ARRAY_SIZE(a660_protect);
1320 		count_max = 48;
1321 		BUILD_BUG_ON(ARRAY_SIZE(a660_protect) > 48);
1322 	} else if (adreno_is_a730(adreno_gpu) ||
1323 		   adreno_is_a740(adreno_gpu) ||
1324 		   adreno_is_a750(adreno_gpu)) {
1325 		regs = a730_protect;
1326 		count = ARRAY_SIZE(a730_protect);
1327 		count_max = 48;
1328 		BUILD_BUG_ON(ARRAY_SIZE(a730_protect) > 48);
1329 	} else {
1330 		regs = a6xx_protect;
1331 		count = ARRAY_SIZE(a6xx_protect);
1332 		count_max = 32;
1333 		BUILD_BUG_ON(ARRAY_SIZE(a6xx_protect) > 32);
1334 	}
1335 
1336 	/*
1337 	 * Enable access protection to privileged registers, fault on an access
1338 	 * protect violation and select the last span to protect from the start
1339 	 * address all the way to the end of the register address space
1340 	 */
1341 	gpu_write(gpu, REG_A6XX_CP_PROTECT_CNTL,
1342 		  A6XX_CP_PROTECT_CNTL_ACCESS_PROT_EN |
1343 		  A6XX_CP_PROTECT_CNTL_ACCESS_FAULT_ON_VIOL_EN |
1344 		  A6XX_CP_PROTECT_CNTL_LAST_SPAN_INF_RANGE);
1345 
1346 	for (i = 0; i < count - 1; i++) {
1347 		/* Intentionally skip writing to some registers */
1348 		if (regs[i])
1349 			gpu_write(gpu, REG_A6XX_CP_PROTECT(i), regs[i]);
1350 	}
1351 	/* last CP_PROTECT to have "infinite" length on the last entry */
1352 	gpu_write(gpu, REG_A6XX_CP_PROTECT(count_max - 1), regs[i]);
1353 }
1354 
1355 static void a6xx_calc_ubwc_config(struct adreno_gpu *gpu)
1356 {
1357 	/* Unknown, introduced with A650 family, related to UBWC mode/ver 4 */
1358 	gpu->ubwc_config.rgb565_predicator = 0;
1359 	/* Unknown, introduced with A650 family */
1360 	gpu->ubwc_config.uavflagprd_inv = 0;
1361 	/* Whether the minimum access length is 64 bits */
1362 	gpu->ubwc_config.min_acc_len = 0;
1363 	/* Entirely magic, per-GPU-gen value */
1364 	gpu->ubwc_config.ubwc_mode = 0;
1365 	/*
1366 	 * The Highest Bank Bit value represents the bit of the highest DDR bank.
1367 	 * This should ideally use DRAM type detection.
1368 	 */
1369 	gpu->ubwc_config.highest_bank_bit = 15;
1370 
1371 	if (adreno_is_a610(gpu)) {
1372 		gpu->ubwc_config.highest_bank_bit = 13;
1373 		gpu->ubwc_config.min_acc_len = 1;
1374 		gpu->ubwc_config.ubwc_mode = 1;
1375 	}
1376 
1377 	if (adreno_is_a618(gpu))
1378 		gpu->ubwc_config.highest_bank_bit = 14;
1379 
1380 	if (adreno_is_a619(gpu))
1381 		/* TODO: Should be 14 but causes corruption at e.g. 1920x1200 on DP */
1382 		gpu->ubwc_config.highest_bank_bit = 13;
1383 
1384 	if (adreno_is_a619_holi(gpu))
1385 		gpu->ubwc_config.highest_bank_bit = 13;
1386 
1387 	if (adreno_is_a640_family(gpu))
1388 		gpu->ubwc_config.amsbc = 1;
1389 
1390 	if (adreno_is_a650(gpu) ||
1391 	    adreno_is_a660(gpu) ||
1392 	    adreno_is_a690(gpu) ||
1393 	    adreno_is_a730(gpu) ||
1394 	    adreno_is_a740_family(gpu)) {
1395 		/* TODO: get ddr type from bootloader and use 2 for LPDDR4 */
1396 		gpu->ubwc_config.highest_bank_bit = 16;
1397 		gpu->ubwc_config.amsbc = 1;
1398 		gpu->ubwc_config.rgb565_predicator = 1;
1399 		gpu->ubwc_config.uavflagprd_inv = 2;
1400 	}
1401 
1402 	if (adreno_is_7c3(gpu)) {
1403 		gpu->ubwc_config.highest_bank_bit = 14;
1404 		gpu->ubwc_config.amsbc = 1;
1405 		gpu->ubwc_config.rgb565_predicator = 1;
1406 		gpu->ubwc_config.uavflagprd_inv = 2;
1407 	}
1408 
1409 	if (adreno_is_a702(gpu)) {
1410 		gpu->ubwc_config.highest_bank_bit = 14;
1411 		gpu->ubwc_config.min_acc_len = 1;
1412 		gpu->ubwc_config.ubwc_mode = 2;
1413 	}
1414 }
1415 
1416 static void a6xx_set_ubwc_config(struct msm_gpu *gpu)
1417 {
1418 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1419 	/*
1420 	 * We subtract 13 from the highest bank bit (13 is the minimum value
1421 	 * allowed by hw) and write the lowest two bits of the remaining value
1422 	 * as hbb_lo and the one above it as hbb_hi to the hardware.
1423 	 */
1424 	BUG_ON(adreno_gpu->ubwc_config.highest_bank_bit < 13);
1425 	u32 hbb = adreno_gpu->ubwc_config.highest_bank_bit - 13;
1426 	u32 hbb_hi = hbb >> 2;
1427 	u32 hbb_lo = hbb & 3;
1428 
1429 	gpu_write(gpu, REG_A6XX_RB_NC_MODE_CNTL,
1430 		  adreno_gpu->ubwc_config.rgb565_predicator << 11 |
1431 		  hbb_hi << 10 | adreno_gpu->ubwc_config.amsbc << 4 |
1432 		  adreno_gpu->ubwc_config.min_acc_len << 3 |
1433 		  hbb_lo << 1 | adreno_gpu->ubwc_config.ubwc_mode);
1434 
1435 	gpu_write(gpu, REG_A6XX_TPL1_NC_MODE_CNTL, hbb_hi << 4 |
1436 		  adreno_gpu->ubwc_config.min_acc_len << 3 |
1437 		  hbb_lo << 1 | adreno_gpu->ubwc_config.ubwc_mode);
1438 
1439 	gpu_write(gpu, REG_A6XX_SP_NC_MODE_CNTL, hbb_hi << 10 |
1440 		  adreno_gpu->ubwc_config.uavflagprd_inv << 4 |
1441 		  adreno_gpu->ubwc_config.min_acc_len << 3 |
1442 		  hbb_lo << 1 | adreno_gpu->ubwc_config.ubwc_mode);
1443 
1444 	if (adreno_is_a7xx(adreno_gpu))
1445 		gpu_write(gpu, REG_A7XX_GRAS_NC_MODE_CNTL,
1446 			  FIELD_PREP(GENMASK(8, 5), hbb_lo));
1447 
1448 	gpu_write(gpu, REG_A6XX_UCHE_MODE_CNTL,
1449 		  adreno_gpu->ubwc_config.min_acc_len << 23 | hbb_lo << 21);
1450 }
1451 
1452 static int a6xx_cp_init(struct msm_gpu *gpu)
1453 {
1454 	struct msm_ringbuffer *ring = gpu->rb[0];
1455 
1456 	OUT_PKT7(ring, CP_ME_INIT, 8);
1457 
1458 	OUT_RING(ring, 0x0000002f);
1459 
1460 	/* Enable multiple hardware contexts */
1461 	OUT_RING(ring, 0x00000003);
1462 
1463 	/* Enable error detection */
1464 	OUT_RING(ring, 0x20000000);
1465 
1466 	/* Don't enable header dump */
1467 	OUT_RING(ring, 0x00000000);
1468 	OUT_RING(ring, 0x00000000);
1469 
1470 	/* No workarounds enabled */
1471 	OUT_RING(ring, 0x00000000);
1472 
1473 	/* Pad rest of the cmds with 0's */
1474 	OUT_RING(ring, 0x00000000);
1475 	OUT_RING(ring, 0x00000000);
1476 
1477 	a6xx_flush(gpu, ring);
1478 	return a6xx_idle(gpu, ring) ? 0 : -EINVAL;
1479 }
1480 
1481 static int a7xx_cp_init(struct msm_gpu *gpu)
1482 {
1483 	struct msm_ringbuffer *ring = gpu->rb[0];
1484 	u32 mask;
1485 
1486 	/* Disable concurrent binning before sending CP init */
1487 	OUT_PKT7(ring, CP_THREAD_CONTROL, 1);
1488 	OUT_RING(ring, BIT(27));
1489 
1490 	OUT_PKT7(ring, CP_ME_INIT, 7);
1491 
1492 	/* Use multiple HW contexts */
1493 	mask = BIT(0);
1494 
1495 	/* Enable error detection */
1496 	mask |= BIT(1);
1497 
1498 	/* Set default reset state */
1499 	mask |= BIT(3);
1500 
1501 	/* Disable save/restore of performance counters across preemption */
1502 	mask |= BIT(6);
1503 
1504 	/* Enable the register init list with the spinlock */
1505 	mask |= BIT(8);
1506 
1507 	OUT_RING(ring, mask);
1508 
1509 	/* Enable multiple hardware contexts */
1510 	OUT_RING(ring, 0x00000003);
1511 
1512 	/* Enable error detection */
1513 	OUT_RING(ring, 0x20000000);
1514 
1515 	/* Operation mode mask */
1516 	OUT_RING(ring, 0x00000002);
1517 
1518 	/* *Don't* send a power up reg list for concurrent binning (TODO) */
1519 	/* Lo address */
1520 	OUT_RING(ring, 0x00000000);
1521 	/* Hi address */
1522 	OUT_RING(ring, 0x00000000);
1523 	/* BIT(31) set => read the regs from the list */
1524 	OUT_RING(ring, 0x00000000);
1525 
1526 	a6xx_flush(gpu, ring);
1527 	return a6xx_idle(gpu, ring) ? 0 : -EINVAL;
1528 }
1529 
1530 /*
1531  * Check that the microcode version is new enough to include several key
1532  * security fixes. Return true if the ucode is safe.
1533  */
1534 static bool a6xx_ucode_check_version(struct a6xx_gpu *a6xx_gpu,
1535 		struct drm_gem_object *obj)
1536 {
1537 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
1538 	struct msm_gpu *gpu = &adreno_gpu->base;
1539 	const char *sqe_name = adreno_gpu->info->fw[ADRENO_FW_SQE];
1540 	u32 *buf = msm_gem_get_vaddr(obj);
1541 	bool ret = false;
1542 
1543 	if (IS_ERR(buf))
1544 		return false;
1545 
1546 	/* A7xx is safe! */
1547 	if (adreno_is_a7xx(adreno_gpu) || adreno_is_a702(adreno_gpu))
1548 		return true;
1549 
1550 	/*
1551 	 * Targets up to a640 (a618, a630 and a640) need to check for a
1552 	 * microcode version that is patched to support the whereami opcode or
1553 	 * one that is new enough to include it by default.
1554 	 *
1555 	 * a650 tier targets don't need whereami but still need to be
1556 	 * equal to or newer than 0.95 for other security fixes
1557 	 *
1558 	 * a660 targets have all the critical security fixes from the start
1559 	 */
1560 	if (!strcmp(sqe_name, "a630_sqe.fw")) {
1561 		/*
1562 		 * If the lowest nibble is 0xa that is an indication that this
1563 		 * microcode has been patched. The actual version is in dword
1564 		 * [3] but we only care about the patchlevel which is the lowest
1565 		 * nibble of dword [3]
1566 		 *
1567 		 * Otherwise check that the firmware is greater than or equal
1568 		 * to 1.90 which was the first version that had this fix built
1569 		 * in
1570 		 */
1571 		if ((((buf[0] & 0xf) == 0xa) && (buf[2] & 0xf) >= 1) ||
1572 			(buf[0] & 0xfff) >= 0x190) {
1573 			a6xx_gpu->has_whereami = true;
1574 			ret = true;
1575 			goto out;
1576 		}
1577 
1578 		DRM_DEV_ERROR(&gpu->pdev->dev,
1579 			"a630 SQE ucode is too old. Have version %x need at least %x\n",
1580 			buf[0] & 0xfff, 0x190);
1581 	} else if (!strcmp(sqe_name, "a650_sqe.fw")) {
1582 		if ((buf[0] & 0xfff) >= 0x095) {
1583 			ret = true;
1584 			goto out;
1585 		}
1586 
1587 		DRM_DEV_ERROR(&gpu->pdev->dev,
1588 			"a650 SQE ucode is too old. Have version %x need at least %x\n",
1589 			buf[0] & 0xfff, 0x095);
1590 	} else if (!strcmp(sqe_name, "a660_sqe.fw")) {
1591 		ret = true;
1592 	} else {
1593 		DRM_DEV_ERROR(&gpu->pdev->dev,
1594 			"unknown GPU, add it to a6xx_ucode_check_version()!!\n");
1595 	}
1596 out:
1597 	msm_gem_put_vaddr(obj);
1598 	return ret;
1599 }
1600 
1601 static int a6xx_ucode_load(struct msm_gpu *gpu)
1602 {
1603 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1604 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1605 
1606 	if (!a6xx_gpu->sqe_bo) {
1607 		a6xx_gpu->sqe_bo = adreno_fw_create_bo(gpu,
1608 			adreno_gpu->fw[ADRENO_FW_SQE], &a6xx_gpu->sqe_iova);
1609 
1610 		if (IS_ERR(a6xx_gpu->sqe_bo)) {
1611 			int ret = PTR_ERR(a6xx_gpu->sqe_bo);
1612 
1613 			a6xx_gpu->sqe_bo = NULL;
1614 			DRM_DEV_ERROR(&gpu->pdev->dev,
1615 				"Could not allocate SQE ucode: %d\n", ret);
1616 
1617 			return ret;
1618 		}
1619 
1620 		msm_gem_object_set_name(a6xx_gpu->sqe_bo, "sqefw");
1621 		if (!a6xx_ucode_check_version(a6xx_gpu, a6xx_gpu->sqe_bo)) {
1622 			msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
1623 			drm_gem_object_put(a6xx_gpu->sqe_bo);
1624 
1625 			a6xx_gpu->sqe_bo = NULL;
1626 			return -EPERM;
1627 		}
1628 	}
1629 
1630 	/*
1631 	 * Expanded APRIV and targets that support WHERE_AM_I both need a
1632 	 * privileged buffer to store the RPTR shadow
1633 	 */
1634 	if ((adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami) &&
1635 	    !a6xx_gpu->shadow_bo) {
1636 		a6xx_gpu->shadow = msm_gem_kernel_new(gpu->dev,
1637 						      sizeof(u32) * gpu->nr_rings,
1638 						      MSM_BO_WC | MSM_BO_MAP_PRIV,
1639 						      gpu->aspace, &a6xx_gpu->shadow_bo,
1640 						      &a6xx_gpu->shadow_iova);
1641 
1642 		if (IS_ERR(a6xx_gpu->shadow))
1643 			return PTR_ERR(a6xx_gpu->shadow);
1644 
1645 		msm_gem_object_set_name(a6xx_gpu->shadow_bo, "shadow");
1646 	}
1647 
1648 	return 0;
1649 }
1650 
1651 static int a6xx_zap_shader_init(struct msm_gpu *gpu)
1652 {
1653 	static bool loaded;
1654 	int ret;
1655 
1656 	if (loaded)
1657 		return 0;
1658 
1659 	ret = adreno_zap_shader_load(gpu, GPU_PAS_ID);
1660 
1661 	loaded = !ret;
1662 	return ret;
1663 }
1664 
1665 #define A6XX_INT_MASK (A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR | \
1666 		       A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW | \
1667 		       A6XX_RBBM_INT_0_MASK_CP_HW_ERROR | \
1668 		       A6XX_RBBM_INT_0_MASK_CP_IB2 | \
1669 		       A6XX_RBBM_INT_0_MASK_CP_IB1 | \
1670 		       A6XX_RBBM_INT_0_MASK_CP_RB | \
1671 		       A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \
1672 		       A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW | \
1673 		       A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT | \
1674 		       A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \
1675 		       A6XX_RBBM_INT_0_MASK_UCHE_TRAP_INTR)
1676 
1677 #define A7XX_INT_MASK (A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR | \
1678 		       A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW | \
1679 		       A6XX_RBBM_INT_0_MASK_RBBM_GPC_ERROR | \
1680 		       A6XX_RBBM_INT_0_MASK_CP_SW | \
1681 		       A6XX_RBBM_INT_0_MASK_CP_HW_ERROR | \
1682 		       A6XX_RBBM_INT_0_MASK_PM4CPINTERRUPT | \
1683 		       A6XX_RBBM_INT_0_MASK_CP_RB_DONE_TS | \
1684 		       A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \
1685 		       A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW | \
1686 		       A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT | \
1687 		       A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \
1688 		       A6XX_RBBM_INT_0_MASK_UCHE_TRAP_INTR | \
1689 		       A6XX_RBBM_INT_0_MASK_TSBWRITEERROR)
1690 
1691 #define A7XX_APRIV_MASK (A6XX_CP_APRIV_CNTL_ICACHE | \
1692 			 A6XX_CP_APRIV_CNTL_RBFETCH | \
1693 			 A6XX_CP_APRIV_CNTL_RBPRIVLEVEL | \
1694 			 A6XX_CP_APRIV_CNTL_RBRPWB)
1695 
1696 #define A7XX_BR_APRIVMASK (A7XX_APRIV_MASK | \
1697 			   A6XX_CP_APRIV_CNTL_CDREAD | \
1698 			   A6XX_CP_APRIV_CNTL_CDWRITE)
1699 
1700 static int hw_init(struct msm_gpu *gpu)
1701 {
1702 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1703 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1704 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
1705 	u64 gmem_range_min;
1706 	int ret;
1707 
1708 	if (!adreno_has_gmu_wrapper(adreno_gpu)) {
1709 		/* Make sure the GMU keeps the GPU on while we set it up */
1710 		ret = a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
1711 		if (ret)
1712 			return ret;
1713 	}
1714 
1715 	/* Clear GBIF halt in case GX domain was not collapsed */
1716 	if (adreno_is_a619_holi(adreno_gpu)) {
1717 		gpu_write(gpu, REG_A6XX_GBIF_HALT, 0);
1718 		gpu_write(gpu, REG_A6XX_RBBM_GPR0_CNTL, 0);
1719 		/* Let's make extra sure that the GPU can access the memory.. */
1720 		mb();
1721 	} else if (a6xx_has_gbif(adreno_gpu)) {
1722 		gpu_write(gpu, REG_A6XX_GBIF_HALT, 0);
1723 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_HALT, 0);
1724 		/* Let's make extra sure that the GPU can access the memory.. */
1725 		mb();
1726 	}
1727 
1728 	/* Some GPUs are stubborn and take their sweet time to unhalt GBIF! */
1729 	if (adreno_is_a7xx(adreno_gpu) && a6xx_has_gbif(adreno_gpu))
1730 		spin_until(!gpu_read(gpu, REG_A6XX_GBIF_HALT_ACK));
1731 
1732 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_CNTL, 0);
1733 
1734 	if (adreno_is_a619_holi(adreno_gpu))
1735 		a6xx_sptprac_enable(gmu);
1736 
1737 	/*
1738 	 * Disable the trusted memory range - we don't actually supported secure
1739 	 * memory rendering at this point in time and we don't want to block off
1740 	 * part of the virtual memory space.
1741 	 */
1742 	gpu_write64(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_BASE, 0x00000000);
1743 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_SIZE, 0x00000000);
1744 
1745 	if (!adreno_is_a7xx(adreno_gpu)) {
1746 		/* Turn on 64 bit addressing for all blocks */
1747 		gpu_write(gpu, REG_A6XX_CP_ADDR_MODE_CNTL, 0x1);
1748 		gpu_write(gpu, REG_A6XX_VSC_ADDR_MODE_CNTL, 0x1);
1749 		gpu_write(gpu, REG_A6XX_GRAS_ADDR_MODE_CNTL, 0x1);
1750 		gpu_write(gpu, REG_A6XX_RB_ADDR_MODE_CNTL, 0x1);
1751 		gpu_write(gpu, REG_A6XX_PC_ADDR_MODE_CNTL, 0x1);
1752 		gpu_write(gpu, REG_A6XX_HLSQ_ADDR_MODE_CNTL, 0x1);
1753 		gpu_write(gpu, REG_A6XX_VFD_ADDR_MODE_CNTL, 0x1);
1754 		gpu_write(gpu, REG_A6XX_VPC_ADDR_MODE_CNTL, 0x1);
1755 		gpu_write(gpu, REG_A6XX_UCHE_ADDR_MODE_CNTL, 0x1);
1756 		gpu_write(gpu, REG_A6XX_SP_ADDR_MODE_CNTL, 0x1);
1757 		gpu_write(gpu, REG_A6XX_TPL1_ADDR_MODE_CNTL, 0x1);
1758 		gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_ADDR_MODE_CNTL, 0x1);
1759 	}
1760 
1761 	/* enable hardware clockgating */
1762 	a6xx_set_hwcg(gpu, true);
1763 
1764 	/* VBIF/GBIF start*/
1765 	if (adreno_is_a610_family(adreno_gpu) ||
1766 	    adreno_is_a640_family(adreno_gpu) ||
1767 	    adreno_is_a650_family(adreno_gpu) ||
1768 	    adreno_is_a7xx(adreno_gpu)) {
1769 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE0, 0x00071620);
1770 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE1, 0x00071620);
1771 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE2, 0x00071620);
1772 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
1773 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL,
1774 			  adreno_is_a7xx(adreno_gpu) ? 0x2120212 : 0x3);
1775 	} else {
1776 		gpu_write(gpu, REG_A6XX_RBBM_VBIF_CLIENT_QOS_CNTL, 0x3);
1777 	}
1778 
1779 	if (adreno_is_a630(adreno_gpu))
1780 		gpu_write(gpu, REG_A6XX_VBIF_GATE_OFF_WRREQ_EN, 0x00000009);
1781 
1782 	if (adreno_is_a7xx(adreno_gpu))
1783 		gpu_write(gpu, REG_A6XX_UCHE_GBIF_GX_CONFIG, 0x10240e0);
1784 
1785 	/* Make all blocks contribute to the GPU BUSY perf counter */
1786 	gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_GPU_BUSY_MASKED, 0xffffffff);
1787 
1788 	/* Disable L2 bypass in the UCHE */
1789 	if (adreno_is_a7xx(adreno_gpu)) {
1790 		gpu_write64(gpu, REG_A6XX_UCHE_TRAP_BASE, 0x0001fffffffff000llu);
1791 		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE, 0x0001fffffffff000llu);
1792 	} else {
1793 		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX, 0x0001ffffffffffc0llu);
1794 		gpu_write64(gpu, REG_A6XX_UCHE_TRAP_BASE, 0x0001fffffffff000llu);
1795 		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE, 0x0001fffffffff000llu);
1796 	}
1797 
1798 	if (!(adreno_is_a650_family(adreno_gpu) ||
1799 	      adreno_is_a702(adreno_gpu) ||
1800 	      adreno_is_a730(adreno_gpu))) {
1801 		gmem_range_min = adreno_is_a740_family(adreno_gpu) ? SZ_16M : SZ_1M;
1802 
1803 		/* Set the GMEM VA range [0x100000:0x100000 + gpu->gmem - 1] */
1804 		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MIN, gmem_range_min);
1805 
1806 		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MAX,
1807 			gmem_range_min + adreno_gpu->info->gmem - 1);
1808 	}
1809 
1810 	if (adreno_is_a7xx(adreno_gpu))
1811 		gpu_write(gpu, REG_A6XX_UCHE_CACHE_WAYS, BIT(23));
1812 	else {
1813 		gpu_write(gpu, REG_A6XX_UCHE_FILTER_CNTL, 0x804);
1814 		gpu_write(gpu, REG_A6XX_UCHE_CACHE_WAYS, 0x4);
1815 	}
1816 
1817 	if (adreno_is_a640_family(adreno_gpu) || adreno_is_a650_family(adreno_gpu)) {
1818 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x02000140);
1819 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x8040362c);
1820 	} else if (adreno_is_a610_family(adreno_gpu)) {
1821 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x00800060);
1822 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x40201b16);
1823 	} else if (!adreno_is_a7xx(adreno_gpu)) {
1824 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x010000c0);
1825 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x8040362c);
1826 	}
1827 
1828 	if (adreno_is_a660_family(adreno_gpu))
1829 		gpu_write(gpu, REG_A6XX_CP_LPAC_PROG_FIFO_SIZE, 0x00000020);
1830 
1831 	/* Setting the mem pool size */
1832 	if (adreno_is_a610(adreno_gpu)) {
1833 		gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 48);
1834 		gpu_write(gpu, REG_A6XX_CP_MEM_POOL_DBG_ADDR, 47);
1835 	} else if (adreno_is_a702(adreno_gpu)) {
1836 		gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 64);
1837 		gpu_write(gpu, REG_A6XX_CP_MEM_POOL_DBG_ADDR, 63);
1838 	} else if (!adreno_is_a7xx(adreno_gpu))
1839 		gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 128);
1840 
1841 	/* Setting the primFifo thresholds default values,
1842 	 * and vccCacheSkipDis=1 bit (0x200) for A640 and newer
1843 	*/
1844 	if (adreno_is_a702(adreno_gpu))
1845 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x0000c000);
1846 	else if (adreno_is_a690(adreno_gpu))
1847 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00800200);
1848 	else if (adreno_is_a650(adreno_gpu) || adreno_is_a660(adreno_gpu))
1849 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00300200);
1850 	else if (adreno_is_a640_family(adreno_gpu) || adreno_is_7c3(adreno_gpu))
1851 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00200200);
1852 	else if (adreno_is_a650(adreno_gpu) || adreno_is_a660(adreno_gpu))
1853 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00300200);
1854 	else if (adreno_is_a619(adreno_gpu))
1855 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00018000);
1856 	else if (adreno_is_a610(adreno_gpu))
1857 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00080000);
1858 	else if (!adreno_is_a7xx(adreno_gpu))
1859 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00180000);
1860 
1861 	/* Set the AHB default slave response to "ERROR" */
1862 	gpu_write(gpu, REG_A6XX_CP_AHB_CNTL, 0x1);
1863 
1864 	/* Turn on performance counters */
1865 	gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_CNTL, 0x1);
1866 
1867 	if (adreno_is_a7xx(adreno_gpu)) {
1868 		/* Turn on the IFPC counter (countable 4 on XOCLK4) */
1869 		gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_CX_GMU_POWER_COUNTER_SELECT_1,
1870 			  FIELD_PREP(GENMASK(7, 0), 0x4));
1871 	}
1872 
1873 	/* Select CP0 to always count cycles */
1874 	gpu_write(gpu, REG_A6XX_CP_PERFCTR_CP_SEL(0), PERF_CP_ALWAYS_COUNT);
1875 
1876 	a6xx_set_ubwc_config(gpu);
1877 
1878 	/* Enable fault detection */
1879 	if (adreno_is_a730(adreno_gpu) ||
1880 	    adreno_is_a740_family(adreno_gpu))
1881 		gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL, (1 << 30) | 0xcfffff);
1882 	else if (adreno_is_a690(adreno_gpu))
1883 		gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL, (1 << 30) | 0x4fffff);
1884 	else if (adreno_is_a619(adreno_gpu))
1885 		gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL, (1 << 30) | 0x3fffff);
1886 	else if (adreno_is_a610(adreno_gpu) || adreno_is_a702(adreno_gpu))
1887 		gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL, (1 << 30) | 0x3ffff);
1888 	else
1889 		gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL, (1 << 30) | 0x1fffff);
1890 
1891 	gpu_write(gpu, REG_A6XX_UCHE_CLIENT_PF, BIT(7) | 0x1);
1892 
1893 	/* Set weights for bicubic filtering */
1894 	if (adreno_is_a650_family(adreno_gpu)) {
1895 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_0, 0);
1896 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_1,
1897 			0x3fe05ff4);
1898 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_2,
1899 			0x3fa0ebee);
1900 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_3,
1901 			0x3f5193ed);
1902 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_4,
1903 			0x3f0243f0);
1904 	}
1905 
1906 	/* Set up the CX GMU counter 0 to count busy ticks */
1907 	gmu_write(gmu, REG_A6XX_GPU_GMU_AO_GPU_CX_BUSY_MASK, 0xff000000);
1908 
1909 	/* Enable the power counter */
1910 	gmu_rmw(gmu, REG_A6XX_GMU_CX_GMU_POWER_COUNTER_SELECT_0, 0xff, BIT(5));
1911 	gmu_write(gmu, REG_A6XX_GMU_CX_GMU_POWER_COUNTER_ENABLE, 1);
1912 
1913 	/* Protect registers from the CP */
1914 	a6xx_set_cp_protect(gpu);
1915 
1916 	if (adreno_is_a660_family(adreno_gpu)) {
1917 		if (adreno_is_a690(adreno_gpu))
1918 			gpu_write(gpu, REG_A6XX_CP_CHICKEN_DBG, 0x00028801);
1919 		else
1920 			gpu_write(gpu, REG_A6XX_CP_CHICKEN_DBG, 0x1);
1921 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL, 0x0);
1922 	} else if (adreno_is_a702(adreno_gpu)) {
1923 		/* Something to do with the HLSQ cluster */
1924 		gpu_write(gpu, REG_A6XX_CP_CHICKEN_DBG, BIT(24));
1925 	}
1926 
1927 	if (adreno_is_a690(adreno_gpu))
1928 		gpu_write(gpu, REG_A6XX_UCHE_CMDQ_CONFIG, 0x90);
1929 	/* Set dualQ + disable afull for A660 GPU */
1930 	else if (adreno_is_a660(adreno_gpu))
1931 		gpu_write(gpu, REG_A6XX_UCHE_CMDQ_CONFIG, 0x66906);
1932 	else if (adreno_is_a7xx(adreno_gpu))
1933 		gpu_write(gpu, REG_A6XX_UCHE_CMDQ_CONFIG,
1934 			  FIELD_PREP(GENMASK(19, 16), 6) |
1935 			  FIELD_PREP(GENMASK(15, 12), 6) |
1936 			  FIELD_PREP(GENMASK(11, 8), 9) |
1937 			  BIT(3) | BIT(2) |
1938 			  FIELD_PREP(GENMASK(1, 0), 2));
1939 
1940 	/* Enable expanded apriv for targets that support it */
1941 	if (gpu->hw_apriv) {
1942 		if (adreno_is_a7xx(adreno_gpu)) {
1943 			gpu_write(gpu, REG_A6XX_CP_APRIV_CNTL,
1944 				  A7XX_BR_APRIVMASK);
1945 			gpu_write(gpu, REG_A7XX_CP_BV_APRIV_CNTL,
1946 				  A7XX_APRIV_MASK);
1947 			gpu_write(gpu, REG_A7XX_CP_LPAC_APRIV_CNTL,
1948 				  A7XX_APRIV_MASK);
1949 		} else
1950 			gpu_write(gpu, REG_A6XX_CP_APRIV_CNTL,
1951 				  BIT(6) | BIT(5) | BIT(3) | BIT(2) | BIT(1));
1952 	}
1953 
1954 	/* Enable interrupts */
1955 	gpu_write(gpu, REG_A6XX_RBBM_INT_0_MASK,
1956 		  adreno_is_a7xx(adreno_gpu) ? A7XX_INT_MASK : A6XX_INT_MASK);
1957 
1958 	ret = adreno_hw_init(gpu);
1959 	if (ret)
1960 		goto out;
1961 
1962 	gpu_write64(gpu, REG_A6XX_CP_SQE_INSTR_BASE, a6xx_gpu->sqe_iova);
1963 
1964 	/* Set the ringbuffer address */
1965 	gpu_write64(gpu, REG_A6XX_CP_RB_BASE, gpu->rb[0]->iova);
1966 
1967 	/* Targets that support extended APRIV can use the RPTR shadow from
1968 	 * hardware but all the other ones need to disable the feature. Targets
1969 	 * that support the WHERE_AM_I opcode can use that instead
1970 	 */
1971 	if (adreno_gpu->base.hw_apriv)
1972 		gpu_write(gpu, REG_A6XX_CP_RB_CNTL, MSM_GPU_RB_CNTL_DEFAULT);
1973 	else
1974 		gpu_write(gpu, REG_A6XX_CP_RB_CNTL,
1975 			MSM_GPU_RB_CNTL_DEFAULT | AXXX_CP_RB_CNTL_NO_UPDATE);
1976 
1977 	/* Configure the RPTR shadow if needed: */
1978 	if (a6xx_gpu->shadow_bo) {
1979 		gpu_write64(gpu, REG_A6XX_CP_RB_RPTR_ADDR,
1980 			shadowptr(a6xx_gpu, gpu->rb[0]));
1981 	}
1982 
1983 	/* ..which means "always" on A7xx, also for BV shadow */
1984 	if (adreno_is_a7xx(adreno_gpu)) {
1985 		gpu_write64(gpu, REG_A7XX_CP_BV_RB_RPTR_ADDR,
1986 			    rbmemptr(gpu->rb[0], bv_fence));
1987 	}
1988 
1989 	/* Always come up on rb 0 */
1990 	a6xx_gpu->cur_ring = gpu->rb[0];
1991 
1992 	gpu->cur_ctx_seqno = 0;
1993 
1994 	/* Enable the SQE_to start the CP engine */
1995 	gpu_write(gpu, REG_A6XX_CP_SQE_CNTL, 1);
1996 
1997 	ret = adreno_is_a7xx(adreno_gpu) ? a7xx_cp_init(gpu) : a6xx_cp_init(gpu);
1998 	if (ret)
1999 		goto out;
2000 
2001 	/*
2002 	 * Try to load a zap shader into the secure world. If successful
2003 	 * we can use the CP to switch out of secure mode. If not then we
2004 	 * have no resource but to try to switch ourselves out manually. If we
2005 	 * guessed wrong then access to the RBBM_SECVID_TRUST_CNTL register will
2006 	 * be blocked and a permissions violation will soon follow.
2007 	 */
2008 	ret = a6xx_zap_shader_init(gpu);
2009 	if (!ret) {
2010 		OUT_PKT7(gpu->rb[0], CP_SET_SECURE_MODE, 1);
2011 		OUT_RING(gpu->rb[0], 0x00000000);
2012 
2013 		a6xx_flush(gpu, gpu->rb[0]);
2014 		if (!a6xx_idle(gpu, gpu->rb[0]))
2015 			return -EINVAL;
2016 	} else if (ret == -ENODEV) {
2017 		/*
2018 		 * This device does not use zap shader (but print a warning
2019 		 * just in case someone got their dt wrong.. hopefully they
2020 		 * have a debug UART to realize the error of their ways...
2021 		 * if you mess this up you are about to crash horribly)
2022 		 */
2023 		dev_warn_once(gpu->dev->dev,
2024 			"Zap shader not enabled - using SECVID_TRUST_CNTL instead\n");
2025 		gpu_write(gpu, REG_A6XX_RBBM_SECVID_TRUST_CNTL, 0x0);
2026 		ret = 0;
2027 	} else {
2028 		return ret;
2029 	}
2030 
2031 out:
2032 	if (adreno_has_gmu_wrapper(adreno_gpu))
2033 		return ret;
2034 	/*
2035 	 * Tell the GMU that we are done touching the GPU and it can start power
2036 	 * management
2037 	 */
2038 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
2039 
2040 	if (a6xx_gpu->gmu.legacy) {
2041 		/* Take the GMU out of its special boot mode */
2042 		a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
2043 	}
2044 
2045 	return ret;
2046 }
2047 
2048 static int a6xx_hw_init(struct msm_gpu *gpu)
2049 {
2050 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2051 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2052 	int ret;
2053 
2054 	mutex_lock(&a6xx_gpu->gmu.lock);
2055 	ret = hw_init(gpu);
2056 	mutex_unlock(&a6xx_gpu->gmu.lock);
2057 
2058 	return ret;
2059 }
2060 
2061 static void a6xx_dump(struct msm_gpu *gpu)
2062 {
2063 	DRM_DEV_INFO(&gpu->pdev->dev, "status:   %08x\n",
2064 			gpu_read(gpu, REG_A6XX_RBBM_STATUS));
2065 	adreno_dump(gpu);
2066 }
2067 
2068 static void a6xx_recover(struct msm_gpu *gpu)
2069 {
2070 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2071 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2072 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
2073 	int i, active_submits;
2074 
2075 	adreno_dump_info(gpu);
2076 
2077 	for (i = 0; i < 8; i++)
2078 		DRM_DEV_INFO(&gpu->pdev->dev, "CP_SCRATCH_REG%d: %u\n", i,
2079 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(i)));
2080 
2081 	if (hang_debug)
2082 		a6xx_dump(gpu);
2083 
2084 	/*
2085 	 * To handle recovery specific sequences during the rpm suspend we are
2086 	 * about to trigger
2087 	 */
2088 	a6xx_gpu->hung = true;
2089 
2090 	/* Halt SQE first */
2091 	gpu_write(gpu, REG_A6XX_CP_SQE_CNTL, 3);
2092 
2093 	pm_runtime_dont_use_autosuspend(&gpu->pdev->dev);
2094 
2095 	/* active_submit won't change until we make a submission */
2096 	mutex_lock(&gpu->active_lock);
2097 	active_submits = gpu->active_submits;
2098 
2099 	/*
2100 	 * Temporarily clear active_submits count to silence a WARN() in the
2101 	 * runtime suspend cb
2102 	 */
2103 	gpu->active_submits = 0;
2104 
2105 	if (adreno_has_gmu_wrapper(adreno_gpu)) {
2106 		/* Drain the outstanding traffic on memory buses */
2107 		a6xx_bus_clear_pending_transactions(adreno_gpu, true);
2108 
2109 		/* Reset the GPU to a clean state */
2110 		a6xx_gpu_sw_reset(gpu, true);
2111 		a6xx_gpu_sw_reset(gpu, false);
2112 	}
2113 
2114 	reinit_completion(&gmu->pd_gate);
2115 	dev_pm_genpd_add_notifier(gmu->cxpd, &gmu->pd_nb);
2116 	dev_pm_genpd_synced_poweroff(gmu->cxpd);
2117 
2118 	/* Drop the rpm refcount from active submits */
2119 	if (active_submits)
2120 		pm_runtime_put(&gpu->pdev->dev);
2121 
2122 	/* And the final one from recover worker */
2123 	pm_runtime_put_sync(&gpu->pdev->dev);
2124 
2125 	if (!wait_for_completion_timeout(&gmu->pd_gate, msecs_to_jiffies(1000)))
2126 		DRM_DEV_ERROR(&gpu->pdev->dev, "cx gdsc didn't collapse\n");
2127 
2128 	dev_pm_genpd_remove_notifier(gmu->cxpd);
2129 
2130 	pm_runtime_use_autosuspend(&gpu->pdev->dev);
2131 
2132 	if (active_submits)
2133 		pm_runtime_get(&gpu->pdev->dev);
2134 
2135 	pm_runtime_get_sync(&gpu->pdev->dev);
2136 
2137 	gpu->active_submits = active_submits;
2138 	mutex_unlock(&gpu->active_lock);
2139 
2140 	msm_gpu_hw_init(gpu);
2141 	a6xx_gpu->hung = false;
2142 }
2143 
2144 static const char *a6xx_uche_fault_block(struct msm_gpu *gpu, u32 mid)
2145 {
2146 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2147 	static const char *uche_clients[7] = {
2148 		"VFD", "SP", "VSC", "VPC", "HLSQ", "PC", "LRZ",
2149 	};
2150 	u32 val;
2151 
2152 	if (adreno_is_a7xx(adreno_gpu)) {
2153 		if (mid != 1 && mid != 2 && mid != 3 && mid != 8)
2154 			return "UNKNOWN";
2155 	} else {
2156 		if (mid < 1 || mid > 3)
2157 			return "UNKNOWN";
2158 	}
2159 
2160 	/*
2161 	 * The source of the data depends on the mid ID read from FSYNR1.
2162 	 * and the client ID read from the UCHE block
2163 	 */
2164 	val = gpu_read(gpu, REG_A6XX_UCHE_CLIENT_PF);
2165 
2166 	if (adreno_is_a7xx(adreno_gpu)) {
2167 		/* Bit 3 for mid=3 indicates BR or BV */
2168 		static const char *uche_clients_a7xx[16] = {
2169 			"BR_VFD", "BR_SP", "BR_VSC", "BR_VPC",
2170 			"BR_HLSQ", "BR_PC", "BR_LRZ", "BR_TP",
2171 			"BV_VFD", "BV_SP", "BV_VSC", "BV_VPC",
2172 			"BV_HLSQ", "BV_PC", "BV_LRZ", "BV_TP",
2173 		};
2174 
2175 		/* LPAC has the same clients as BR and BV, but because it is
2176 		 * compute-only some of them do not exist and there are holes
2177 		 * in the array.
2178 		 */
2179 		static const char *uche_clients_lpac_a7xx[8] = {
2180 			"-", "LPAC_SP", "-", "-",
2181 			"LPAC_HLSQ", "-", "-", "LPAC_TP",
2182 		};
2183 
2184 		val &= GENMASK(6, 0);
2185 
2186 		/* mid=3 refers to BR or BV */
2187 		if (mid == 3) {
2188 			if (val < ARRAY_SIZE(uche_clients_a7xx))
2189 				return uche_clients_a7xx[val];
2190 			else
2191 				return "UCHE";
2192 		}
2193 
2194 		/* mid=8 refers to LPAC */
2195 		if (mid == 8) {
2196 			if (val < ARRAY_SIZE(uche_clients_lpac_a7xx))
2197 				return uche_clients_lpac_a7xx[val];
2198 			else
2199 				return "UCHE_LPAC";
2200 		}
2201 
2202 		/* mid=2 is a catchall for everything else in LPAC */
2203 		if (mid == 2)
2204 			return "UCHE_LPAC";
2205 
2206 		/* mid=1 is a catchall for everything else in BR/BV */
2207 		return "UCHE";
2208 	} else if (adreno_is_a660_family(adreno_gpu)) {
2209 		static const char *uche_clients_a660[8] = {
2210 			"VFD", "SP", "VSC", "VPC", "HLSQ", "PC", "LRZ", "TP",
2211 		};
2212 
2213 		static const char *uche_clients_a660_not[8] = {
2214 			"not VFD", "not SP", "not VSC", "not VPC",
2215 			"not HLSQ", "not PC", "not LRZ", "not TP",
2216 		};
2217 
2218 		val &= GENMASK(6, 0);
2219 
2220 		if (mid == 3 && val < ARRAY_SIZE(uche_clients_a660))
2221 			return uche_clients_a660[val];
2222 
2223 		if (mid == 1 && val < ARRAY_SIZE(uche_clients_a660_not))
2224 			return uche_clients_a660_not[val];
2225 
2226 		return "UCHE";
2227 	} else {
2228 		/* mid = 3 is most precise and refers to only one block per client */
2229 		if (mid == 3)
2230 			return uche_clients[val & 7];
2231 
2232 		/* For mid=2 the source is TP or VFD except when the client id is 0 */
2233 		if (mid == 2)
2234 			return ((val & 7) == 0) ? "TP" : "TP|VFD";
2235 
2236 		/* For mid=1 just return "UCHE" as a catchall for everything else */
2237 		return "UCHE";
2238 	}
2239 }
2240 
2241 static const char *a6xx_fault_block(struct msm_gpu *gpu, u32 id)
2242 {
2243 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2244 
2245 	if (id == 0)
2246 		return "CP";
2247 	else if (id == 4)
2248 		return "CCU";
2249 	else if (id == 6)
2250 		return "CDP Prefetch";
2251 	else if (id == 7)
2252 		return "GMU";
2253 	else if (id == 5 && adreno_is_a7xx(adreno_gpu))
2254 		return "Flag cache";
2255 
2256 	return a6xx_uche_fault_block(gpu, id);
2257 }
2258 
2259 static int a6xx_fault_handler(void *arg, unsigned long iova, int flags, void *data)
2260 {
2261 	struct msm_gpu *gpu = arg;
2262 	struct adreno_smmu_fault_info *info = data;
2263 	const char *block = "unknown";
2264 
2265 	u32 scratch[] = {
2266 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(4)),
2267 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(5)),
2268 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(6)),
2269 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(7)),
2270 	};
2271 
2272 	if (info)
2273 		block = a6xx_fault_block(gpu, info->fsynr1 & 0xff);
2274 
2275 	return adreno_fault_handler(gpu, iova, flags, info, block, scratch);
2276 }
2277 
2278 static void a6xx_cp_hw_err_irq(struct msm_gpu *gpu)
2279 {
2280 	u32 status = gpu_read(gpu, REG_A6XX_CP_INTERRUPT_STATUS);
2281 
2282 	if (status & A6XX_CP_INT_CP_OPCODE_ERROR) {
2283 		u32 val;
2284 
2285 		gpu_write(gpu, REG_A6XX_CP_SQE_STAT_ADDR, 1);
2286 		val = gpu_read(gpu, REG_A6XX_CP_SQE_STAT_DATA);
2287 		dev_err_ratelimited(&gpu->pdev->dev,
2288 			"CP | opcode error | possible opcode=0x%8.8X\n",
2289 			val);
2290 	}
2291 
2292 	if (status & A6XX_CP_INT_CP_UCODE_ERROR)
2293 		dev_err_ratelimited(&gpu->pdev->dev,
2294 			"CP ucode error interrupt\n");
2295 
2296 	if (status & A6XX_CP_INT_CP_HW_FAULT_ERROR)
2297 		dev_err_ratelimited(&gpu->pdev->dev, "CP | HW fault | status=0x%8.8X\n",
2298 			gpu_read(gpu, REG_A6XX_CP_HW_FAULT));
2299 
2300 	if (status & A6XX_CP_INT_CP_REGISTER_PROTECTION_ERROR) {
2301 		u32 val = gpu_read(gpu, REG_A6XX_CP_PROTECT_STATUS);
2302 
2303 		dev_err_ratelimited(&gpu->pdev->dev,
2304 			"CP | protected mode error | %s | addr=0x%8.8X | status=0x%8.8X\n",
2305 			val & (1 << 20) ? "READ" : "WRITE",
2306 			(val & 0x3ffff), val);
2307 	}
2308 
2309 	if (status & A6XX_CP_INT_CP_AHB_ERROR && !adreno_is_a7xx(to_adreno_gpu(gpu)))
2310 		dev_err_ratelimited(&gpu->pdev->dev, "CP AHB error interrupt\n");
2311 
2312 	if (status & A6XX_CP_INT_CP_VSD_PARITY_ERROR)
2313 		dev_err_ratelimited(&gpu->pdev->dev, "CP VSD decoder parity error\n");
2314 
2315 	if (status & A6XX_CP_INT_CP_ILLEGAL_INSTR_ERROR)
2316 		dev_err_ratelimited(&gpu->pdev->dev, "CP illegal instruction error\n");
2317 
2318 }
2319 
2320 static void a6xx_fault_detect_irq(struct msm_gpu *gpu)
2321 {
2322 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2323 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2324 	struct msm_ringbuffer *ring = gpu->funcs->active_ring(gpu);
2325 
2326 	/*
2327 	 * If stalled on SMMU fault, we could trip the GPU's hang detection,
2328 	 * but the fault handler will trigger the devcore dump, and we want
2329 	 * to otherwise resume normally rather than killing the submit, so
2330 	 * just bail.
2331 	 */
2332 	if (gpu_read(gpu, REG_A6XX_RBBM_STATUS3) & A6XX_RBBM_STATUS3_SMMU_STALLED_ON_FAULT)
2333 		return;
2334 
2335 	/*
2336 	 * Force the GPU to stay on until after we finish
2337 	 * collecting information
2338 	 */
2339 	if (!adreno_has_gmu_wrapper(adreno_gpu))
2340 		gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 1);
2341 
2342 	DRM_DEV_ERROR(&gpu->pdev->dev,
2343 		"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",
2344 		ring ? ring->id : -1, ring ? ring->fctx->last_fence : 0,
2345 		gpu_read(gpu, REG_A6XX_RBBM_STATUS),
2346 		gpu_read(gpu, REG_A6XX_CP_RB_RPTR),
2347 		gpu_read(gpu, REG_A6XX_CP_RB_WPTR),
2348 		gpu_read64(gpu, REG_A6XX_CP_IB1_BASE),
2349 		gpu_read(gpu, REG_A6XX_CP_IB1_REM_SIZE),
2350 		gpu_read64(gpu, REG_A6XX_CP_IB2_BASE),
2351 		gpu_read(gpu, REG_A6XX_CP_IB2_REM_SIZE));
2352 
2353 	/* Turn off the hangcheck timer to keep it from bothering us */
2354 	del_timer(&gpu->hangcheck_timer);
2355 
2356 	kthread_queue_work(gpu->worker, &gpu->recover_work);
2357 }
2358 
2359 static irqreturn_t a6xx_irq(struct msm_gpu *gpu)
2360 {
2361 	struct msm_drm_private *priv = gpu->dev->dev_private;
2362 	u32 status = gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS);
2363 
2364 	gpu_write(gpu, REG_A6XX_RBBM_INT_CLEAR_CMD, status);
2365 
2366 	if (priv->disable_err_irq)
2367 		status &= A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS;
2368 
2369 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT)
2370 		a6xx_fault_detect_irq(gpu);
2371 
2372 	if (status & A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR)
2373 		dev_err_ratelimited(&gpu->pdev->dev, "CP | AHB bus error\n");
2374 
2375 	if (status & A6XX_RBBM_INT_0_MASK_CP_HW_ERROR)
2376 		a6xx_cp_hw_err_irq(gpu);
2377 
2378 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW)
2379 		dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB ASYNC overflow\n");
2380 
2381 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW)
2382 		dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB bus overflow\n");
2383 
2384 	if (status & A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS)
2385 		dev_err_ratelimited(&gpu->pdev->dev, "UCHE | Out of bounds access\n");
2386 
2387 	if (status & A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS)
2388 		msm_gpu_retire(gpu);
2389 
2390 	return IRQ_HANDLED;
2391 }
2392 
2393 static void a6xx_llc_deactivate(struct a6xx_gpu *a6xx_gpu)
2394 {
2395 	llcc_slice_deactivate(a6xx_gpu->llc_slice);
2396 	llcc_slice_deactivate(a6xx_gpu->htw_llc_slice);
2397 }
2398 
2399 static void a6xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
2400 {
2401 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
2402 	struct msm_gpu *gpu = &adreno_gpu->base;
2403 	u32 cntl1_regval = 0;
2404 
2405 	if (IS_ERR(a6xx_gpu->llc_mmio))
2406 		return;
2407 
2408 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
2409 		u32 gpu_scid = llcc_get_slice_id(a6xx_gpu->llc_slice);
2410 
2411 		gpu_scid &= 0x1f;
2412 		cntl1_regval = (gpu_scid << 0) | (gpu_scid << 5) | (gpu_scid << 10) |
2413 			       (gpu_scid << 15) | (gpu_scid << 20);
2414 
2415 		/* On A660, the SCID programming for UCHE traffic is done in
2416 		 * A6XX_GBIF_SCACHE_CNTL0[14:10]
2417 		 */
2418 		if (adreno_is_a660_family(adreno_gpu))
2419 			gpu_rmw(gpu, REG_A6XX_GBIF_SCACHE_CNTL0, (0x1f << 10) |
2420 				(1 << 8), (gpu_scid << 10) | (1 << 8));
2421 	}
2422 
2423 	/*
2424 	 * For targets with a MMU500, activate the slice but don't program the
2425 	 * register.  The XBL will take care of that.
2426 	 */
2427 	if (!llcc_slice_activate(a6xx_gpu->htw_llc_slice)) {
2428 		if (!a6xx_gpu->have_mmu500) {
2429 			u32 gpuhtw_scid = llcc_get_slice_id(a6xx_gpu->htw_llc_slice);
2430 
2431 			gpuhtw_scid &= 0x1f;
2432 			cntl1_regval |= FIELD_PREP(GENMASK(29, 25), gpuhtw_scid);
2433 		}
2434 	}
2435 
2436 	if (!cntl1_regval)
2437 		return;
2438 
2439 	/*
2440 	 * Program the slice IDs for the various GPU blocks and GPU MMU
2441 	 * pagetables
2442 	 */
2443 	if (!a6xx_gpu->have_mmu500) {
2444 		a6xx_llc_write(a6xx_gpu,
2445 			REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_1, cntl1_regval);
2446 
2447 		/*
2448 		 * Program cacheability overrides to not allocate cache
2449 		 * lines on a write miss
2450 		 */
2451 		a6xx_llc_rmw(a6xx_gpu,
2452 			REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_0, 0xF, 0x03);
2453 		return;
2454 	}
2455 
2456 	gpu_rmw(gpu, REG_A6XX_GBIF_SCACHE_CNTL1, GENMASK(24, 0), cntl1_regval);
2457 }
2458 
2459 static void a7xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
2460 {
2461 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
2462 	struct msm_gpu *gpu = &adreno_gpu->base;
2463 
2464 	if (IS_ERR(a6xx_gpu->llc_mmio))
2465 		return;
2466 
2467 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
2468 		u32 gpu_scid = llcc_get_slice_id(a6xx_gpu->llc_slice);
2469 
2470 		gpu_scid &= GENMASK(4, 0);
2471 
2472 		gpu_write(gpu, REG_A6XX_GBIF_SCACHE_CNTL1,
2473 			  FIELD_PREP(GENMASK(29, 25), gpu_scid) |
2474 			  FIELD_PREP(GENMASK(24, 20), gpu_scid) |
2475 			  FIELD_PREP(GENMASK(19, 15), gpu_scid) |
2476 			  FIELD_PREP(GENMASK(14, 10), gpu_scid) |
2477 			  FIELD_PREP(GENMASK(9, 5), gpu_scid) |
2478 			  FIELD_PREP(GENMASK(4, 0), gpu_scid));
2479 
2480 		gpu_write(gpu, REG_A6XX_GBIF_SCACHE_CNTL0,
2481 			  FIELD_PREP(GENMASK(14, 10), gpu_scid) |
2482 			  BIT(8));
2483 	}
2484 
2485 	llcc_slice_activate(a6xx_gpu->htw_llc_slice);
2486 }
2487 
2488 static void a6xx_llc_slices_destroy(struct a6xx_gpu *a6xx_gpu)
2489 {
2490 	/* No LLCC on non-RPMh (and by extension, non-GMU) SoCs */
2491 	if (adreno_has_gmu_wrapper(&a6xx_gpu->base))
2492 		return;
2493 
2494 	llcc_slice_putd(a6xx_gpu->llc_slice);
2495 	llcc_slice_putd(a6xx_gpu->htw_llc_slice);
2496 }
2497 
2498 static void a6xx_llc_slices_init(struct platform_device *pdev,
2499 		struct a6xx_gpu *a6xx_gpu, bool is_a7xx)
2500 {
2501 	struct device_node *phandle;
2502 
2503 	/* No LLCC on non-RPMh (and by extension, non-GMU) SoCs */
2504 	if (adreno_has_gmu_wrapper(&a6xx_gpu->base))
2505 		return;
2506 
2507 	/*
2508 	 * There is a different programming path for A6xx targets with an
2509 	 * mmu500 attached, so detect if that is the case
2510 	 */
2511 	phandle = of_parse_phandle(pdev->dev.of_node, "iommus", 0);
2512 	a6xx_gpu->have_mmu500 = (phandle &&
2513 		of_device_is_compatible(phandle, "arm,mmu-500"));
2514 	of_node_put(phandle);
2515 
2516 	if (is_a7xx || !a6xx_gpu->have_mmu500)
2517 		a6xx_gpu->llc_mmio = msm_ioremap(pdev, "cx_mem");
2518 	else
2519 		a6xx_gpu->llc_mmio = NULL;
2520 
2521 	a6xx_gpu->llc_slice = llcc_slice_getd(LLCC_GPU);
2522 	a6xx_gpu->htw_llc_slice = llcc_slice_getd(LLCC_GPUHTW);
2523 
2524 	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
2525 		a6xx_gpu->llc_mmio = ERR_PTR(-EINVAL);
2526 }
2527 
2528 #define GBIF_CLIENT_HALT_MASK		BIT(0)
2529 #define GBIF_ARB_HALT_MASK		BIT(1)
2530 #define VBIF_XIN_HALT_CTRL0_MASK	GENMASK(3, 0)
2531 #define VBIF_RESET_ACK_MASK		0xF0
2532 #define GPR0_GBIF_HALT_REQUEST		0x1E0
2533 
2534 void a6xx_bus_clear_pending_transactions(struct adreno_gpu *adreno_gpu, bool gx_off)
2535 {
2536 	struct msm_gpu *gpu = &adreno_gpu->base;
2537 
2538 	if (adreno_is_a619_holi(adreno_gpu)) {
2539 		gpu_write(gpu, REG_A6XX_RBBM_GPR0_CNTL, GPR0_GBIF_HALT_REQUEST);
2540 		spin_until((gpu_read(gpu, REG_A6XX_RBBM_VBIF_GX_RESET_STATUS) &
2541 				(VBIF_RESET_ACK_MASK)) == VBIF_RESET_ACK_MASK);
2542 	} else if (!a6xx_has_gbif(adreno_gpu)) {
2543 		gpu_write(gpu, REG_A6XX_VBIF_XIN_HALT_CTRL0, VBIF_XIN_HALT_CTRL0_MASK);
2544 		spin_until((gpu_read(gpu, REG_A6XX_VBIF_XIN_HALT_CTRL1) &
2545 				(VBIF_XIN_HALT_CTRL0_MASK)) == VBIF_XIN_HALT_CTRL0_MASK);
2546 		gpu_write(gpu, REG_A6XX_VBIF_XIN_HALT_CTRL0, 0);
2547 
2548 		return;
2549 	}
2550 
2551 	if (gx_off) {
2552 		/* Halt the gx side of GBIF */
2553 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_HALT, 1);
2554 		spin_until(gpu_read(gpu, REG_A6XX_RBBM_GBIF_HALT_ACK) & 1);
2555 	}
2556 
2557 	/* Halt new client requests on GBIF */
2558 	gpu_write(gpu, REG_A6XX_GBIF_HALT, GBIF_CLIENT_HALT_MASK);
2559 	spin_until((gpu_read(gpu, REG_A6XX_GBIF_HALT_ACK) &
2560 			(GBIF_CLIENT_HALT_MASK)) == GBIF_CLIENT_HALT_MASK);
2561 
2562 	/* Halt all AXI requests on GBIF */
2563 	gpu_write(gpu, REG_A6XX_GBIF_HALT, GBIF_ARB_HALT_MASK);
2564 	spin_until((gpu_read(gpu,  REG_A6XX_GBIF_HALT_ACK) &
2565 			(GBIF_ARB_HALT_MASK)) == GBIF_ARB_HALT_MASK);
2566 
2567 	/* The GBIF halt needs to be explicitly cleared */
2568 	gpu_write(gpu, REG_A6XX_GBIF_HALT, 0x0);
2569 }
2570 
2571 void a6xx_gpu_sw_reset(struct msm_gpu *gpu, bool assert)
2572 {
2573 	/* 11nm chips (e.g. ones with A610) have hw issues with the reset line! */
2574 	if (adreno_is_a610(to_adreno_gpu(gpu)))
2575 		return;
2576 
2577 	gpu_write(gpu, REG_A6XX_RBBM_SW_RESET_CMD, assert);
2578 	/* Perform a bogus read and add a brief delay to ensure ordering. */
2579 	gpu_read(gpu, REG_A6XX_RBBM_SW_RESET_CMD);
2580 	udelay(1);
2581 
2582 	/* The reset line needs to be asserted for at least 100 us */
2583 	if (assert)
2584 		udelay(100);
2585 }
2586 
2587 static int a6xx_gmu_pm_resume(struct msm_gpu *gpu)
2588 {
2589 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2590 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2591 	int ret;
2592 
2593 	gpu->needs_hw_init = true;
2594 
2595 	trace_msm_gpu_resume(0);
2596 
2597 	mutex_lock(&a6xx_gpu->gmu.lock);
2598 	ret = a6xx_gmu_resume(a6xx_gpu);
2599 	mutex_unlock(&a6xx_gpu->gmu.lock);
2600 	if (ret)
2601 		return ret;
2602 
2603 	msm_devfreq_resume(gpu);
2604 
2605 	adreno_is_a7xx(adreno_gpu) ? a7xx_llc_activate(a6xx_gpu) : a6xx_llc_activate(a6xx_gpu);
2606 
2607 	return ret;
2608 }
2609 
2610 static int a6xx_pm_resume(struct msm_gpu *gpu)
2611 {
2612 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2613 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2614 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
2615 	unsigned long freq = gpu->fast_rate;
2616 	struct dev_pm_opp *opp;
2617 	int ret;
2618 
2619 	gpu->needs_hw_init = true;
2620 
2621 	trace_msm_gpu_resume(0);
2622 
2623 	mutex_lock(&a6xx_gpu->gmu.lock);
2624 
2625 	opp = dev_pm_opp_find_freq_ceil(&gpu->pdev->dev, &freq);
2626 	if (IS_ERR(opp)) {
2627 		ret = PTR_ERR(opp);
2628 		goto err_set_opp;
2629 	}
2630 	dev_pm_opp_put(opp);
2631 
2632 	/* Set the core clock and bus bw, having VDD scaling in mind */
2633 	dev_pm_opp_set_opp(&gpu->pdev->dev, opp);
2634 
2635 	pm_runtime_resume_and_get(gmu->dev);
2636 	pm_runtime_resume_and_get(gmu->gxpd);
2637 
2638 	ret = clk_bulk_prepare_enable(gpu->nr_clocks, gpu->grp_clks);
2639 	if (ret)
2640 		goto err_bulk_clk;
2641 
2642 	if (adreno_is_a619_holi(adreno_gpu))
2643 		a6xx_sptprac_enable(gmu);
2644 
2645 	/* If anything goes south, tear the GPU down piece by piece.. */
2646 	if (ret) {
2647 err_bulk_clk:
2648 		pm_runtime_put(gmu->gxpd);
2649 		pm_runtime_put(gmu->dev);
2650 		dev_pm_opp_set_opp(&gpu->pdev->dev, NULL);
2651 	}
2652 err_set_opp:
2653 	mutex_unlock(&a6xx_gpu->gmu.lock);
2654 
2655 	if (!ret)
2656 		msm_devfreq_resume(gpu);
2657 
2658 	return ret;
2659 }
2660 
2661 static int a6xx_gmu_pm_suspend(struct msm_gpu *gpu)
2662 {
2663 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2664 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2665 	int i, ret;
2666 
2667 	trace_msm_gpu_suspend(0);
2668 
2669 	a6xx_llc_deactivate(a6xx_gpu);
2670 
2671 	msm_devfreq_suspend(gpu);
2672 
2673 	mutex_lock(&a6xx_gpu->gmu.lock);
2674 	ret = a6xx_gmu_stop(a6xx_gpu);
2675 	mutex_unlock(&a6xx_gpu->gmu.lock);
2676 	if (ret)
2677 		return ret;
2678 
2679 	if (a6xx_gpu->shadow_bo)
2680 		for (i = 0; i < gpu->nr_rings; i++)
2681 			a6xx_gpu->shadow[i] = 0;
2682 
2683 	gpu->suspend_count++;
2684 
2685 	return 0;
2686 }
2687 
2688 static int a6xx_pm_suspend(struct msm_gpu *gpu)
2689 {
2690 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2691 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2692 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
2693 	int i;
2694 
2695 	trace_msm_gpu_suspend(0);
2696 
2697 	msm_devfreq_suspend(gpu);
2698 
2699 	mutex_lock(&a6xx_gpu->gmu.lock);
2700 
2701 	/* Drain the outstanding traffic on memory buses */
2702 	a6xx_bus_clear_pending_transactions(adreno_gpu, true);
2703 
2704 	if (adreno_is_a619_holi(adreno_gpu))
2705 		a6xx_sptprac_disable(gmu);
2706 
2707 	clk_bulk_disable_unprepare(gpu->nr_clocks, gpu->grp_clks);
2708 
2709 	pm_runtime_put_sync(gmu->gxpd);
2710 	dev_pm_opp_set_opp(&gpu->pdev->dev, NULL);
2711 	pm_runtime_put_sync(gmu->dev);
2712 
2713 	mutex_unlock(&a6xx_gpu->gmu.lock);
2714 
2715 	if (a6xx_gpu->shadow_bo)
2716 		for (i = 0; i < gpu->nr_rings; i++)
2717 			a6xx_gpu->shadow[i] = 0;
2718 
2719 	gpu->suspend_count++;
2720 
2721 	return 0;
2722 }
2723 
2724 static int a6xx_gmu_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
2725 {
2726 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2727 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2728 
2729 	mutex_lock(&a6xx_gpu->gmu.lock);
2730 
2731 	/* Force the GPU power on so we can read this register */
2732 	a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
2733 
2734 	*value = gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER);
2735 
2736 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
2737 
2738 	mutex_unlock(&a6xx_gpu->gmu.lock);
2739 
2740 	return 0;
2741 }
2742 
2743 static int a6xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
2744 {
2745 	*value = gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER);
2746 	return 0;
2747 }
2748 
2749 static struct msm_ringbuffer *a6xx_active_ring(struct msm_gpu *gpu)
2750 {
2751 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2752 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2753 
2754 	return a6xx_gpu->cur_ring;
2755 }
2756 
2757 static void a6xx_destroy(struct msm_gpu *gpu)
2758 {
2759 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2760 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2761 
2762 	if (a6xx_gpu->sqe_bo) {
2763 		msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
2764 		drm_gem_object_put(a6xx_gpu->sqe_bo);
2765 	}
2766 
2767 	if (a6xx_gpu->shadow_bo) {
2768 		msm_gem_unpin_iova(a6xx_gpu->shadow_bo, gpu->aspace);
2769 		drm_gem_object_put(a6xx_gpu->shadow_bo);
2770 	}
2771 
2772 	a6xx_llc_slices_destroy(a6xx_gpu);
2773 
2774 	a6xx_gmu_remove(a6xx_gpu);
2775 
2776 	adreno_gpu_cleanup(adreno_gpu);
2777 
2778 	kfree(a6xx_gpu);
2779 }
2780 
2781 static u64 a6xx_gpu_busy(struct msm_gpu *gpu, unsigned long *out_sample_rate)
2782 {
2783 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2784 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2785 	u64 busy_cycles;
2786 
2787 	/* 19.2MHz */
2788 	*out_sample_rate = 19200000;
2789 
2790 	busy_cycles = gmu_read64(&a6xx_gpu->gmu,
2791 			REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_L,
2792 			REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_H);
2793 
2794 	return busy_cycles;
2795 }
2796 
2797 static void a6xx_gpu_set_freq(struct msm_gpu *gpu, struct dev_pm_opp *opp,
2798 			      bool suspended)
2799 {
2800 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2801 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2802 
2803 	mutex_lock(&a6xx_gpu->gmu.lock);
2804 	a6xx_gmu_set_freq(gpu, opp, suspended);
2805 	mutex_unlock(&a6xx_gpu->gmu.lock);
2806 }
2807 
2808 static struct msm_gem_address_space *
2809 a6xx_create_address_space(struct msm_gpu *gpu, struct platform_device *pdev)
2810 {
2811 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2812 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2813 	unsigned long quirks = 0;
2814 
2815 	/*
2816 	 * This allows GPU to set the bus attributes required to use system
2817 	 * cache on behalf of the iommu page table walker.
2818 	 */
2819 	if (!IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice) &&
2820 	    !device_iommu_capable(&pdev->dev, IOMMU_CAP_CACHE_COHERENCY))
2821 		quirks |= IO_PGTABLE_QUIRK_ARM_OUTER_WBWA;
2822 
2823 	return adreno_iommu_create_address_space(gpu, pdev, quirks);
2824 }
2825 
2826 static struct msm_gem_address_space *
2827 a6xx_create_private_address_space(struct msm_gpu *gpu)
2828 {
2829 	struct msm_mmu *mmu;
2830 
2831 	mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
2832 
2833 	if (IS_ERR(mmu))
2834 		return ERR_CAST(mmu);
2835 
2836 	return msm_gem_address_space_create(mmu,
2837 		"gpu", 0x100000000ULL,
2838 		adreno_private_address_space_size(gpu));
2839 }
2840 
2841 static uint32_t a6xx_get_rptr(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
2842 {
2843 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
2844 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
2845 
2846 	if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami)
2847 		return a6xx_gpu->shadow[ring->id];
2848 
2849 	return ring->memptrs->rptr = gpu_read(gpu, REG_A6XX_CP_RB_RPTR);
2850 }
2851 
2852 static bool a6xx_progress(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
2853 {
2854 	struct msm_cp_state cp_state = {
2855 		.ib1_base = gpu_read64(gpu, REG_A6XX_CP_IB1_BASE),
2856 		.ib2_base = gpu_read64(gpu, REG_A6XX_CP_IB2_BASE),
2857 		.ib1_rem  = gpu_read(gpu, REG_A6XX_CP_IB1_REM_SIZE),
2858 		.ib2_rem  = gpu_read(gpu, REG_A6XX_CP_IB2_REM_SIZE),
2859 	};
2860 	bool progress;
2861 
2862 	/*
2863 	 * Adjust the remaining data to account for what has already been
2864 	 * fetched from memory, but not yet consumed by the SQE.
2865 	 *
2866 	 * This is not *technically* correct, the amount buffered could
2867 	 * exceed the IB size due to hw prefetching ahead, but:
2868 	 *
2869 	 * (1) We aren't trying to find the exact position, just whether
2870 	 *     progress has been made
2871 	 * (2) The CP_REG_TO_MEM at the end of a submit should be enough
2872 	 *     to prevent prefetching into an unrelated submit.  (And
2873 	 *     either way, at some point the ROQ will be full.)
2874 	 */
2875 	cp_state.ib1_rem += gpu_read(gpu, REG_A6XX_CP_ROQ_AVAIL_IB1) >> 16;
2876 	cp_state.ib2_rem += gpu_read(gpu, REG_A6XX_CP_ROQ_AVAIL_IB2) >> 16;
2877 
2878 	progress = !!memcmp(&cp_state, &ring->last_cp_state, sizeof(cp_state));
2879 
2880 	ring->last_cp_state = cp_state;
2881 
2882 	return progress;
2883 }
2884 
2885 static u32 fuse_to_supp_hw(const struct adreno_info *info, u32 fuse)
2886 {
2887 	if (!info->speedbins)
2888 		return UINT_MAX;
2889 
2890 	for (int i = 0; info->speedbins[i].fuse != SHRT_MAX; i++)
2891 		if (info->speedbins[i].fuse == fuse)
2892 			return BIT(info->speedbins[i].speedbin);
2893 
2894 	return UINT_MAX;
2895 }
2896 
2897 static int a6xx_set_supported_hw(struct device *dev, const struct adreno_info *info)
2898 {
2899 	u32 supp_hw;
2900 	u32 speedbin;
2901 	int ret;
2902 
2903 	ret = adreno_read_speedbin(dev, &speedbin);
2904 	/*
2905 	 * -ENOENT means that the platform doesn't support speedbin which is
2906 	 * fine
2907 	 */
2908 	if (ret == -ENOENT) {
2909 		return 0;
2910 	} else if (ret) {
2911 		dev_err_probe(dev, ret,
2912 			      "failed to read speed-bin. Some OPPs may not be supported by hardware\n");
2913 		return ret;
2914 	}
2915 
2916 	supp_hw = fuse_to_supp_hw(info, speedbin);
2917 
2918 	if (supp_hw == UINT_MAX) {
2919 		DRM_DEV_ERROR(dev,
2920 			"missing support for speed-bin: %u. Some OPPs may not be supported by hardware\n",
2921 			speedbin);
2922 		supp_hw = BIT(0); /* Default */
2923 	}
2924 
2925 	ret = devm_pm_opp_set_supported_hw(dev, &supp_hw, 1);
2926 	if (ret)
2927 		return ret;
2928 
2929 	return 0;
2930 }
2931 
2932 static const struct adreno_gpu_funcs funcs = {
2933 	.base = {
2934 		.get_param = adreno_get_param,
2935 		.set_param = adreno_set_param,
2936 		.hw_init = a6xx_hw_init,
2937 		.ucode_load = a6xx_ucode_load,
2938 		.pm_suspend = a6xx_gmu_pm_suspend,
2939 		.pm_resume = a6xx_gmu_pm_resume,
2940 		.recover = a6xx_recover,
2941 		.submit = a6xx_submit,
2942 		.active_ring = a6xx_active_ring,
2943 		.irq = a6xx_irq,
2944 		.destroy = a6xx_destroy,
2945 #if defined(CONFIG_DRM_MSM_GPU_STATE)
2946 		.show = a6xx_show,
2947 #endif
2948 		.gpu_busy = a6xx_gpu_busy,
2949 		.gpu_get_freq = a6xx_gmu_get_freq,
2950 		.gpu_set_freq = a6xx_gpu_set_freq,
2951 #if defined(CONFIG_DRM_MSM_GPU_STATE)
2952 		.gpu_state_get = a6xx_gpu_state_get,
2953 		.gpu_state_put = a6xx_gpu_state_put,
2954 #endif
2955 		.create_address_space = a6xx_create_address_space,
2956 		.create_private_address_space = a6xx_create_private_address_space,
2957 		.get_rptr = a6xx_get_rptr,
2958 		.progress = a6xx_progress,
2959 	},
2960 	.get_timestamp = a6xx_gmu_get_timestamp,
2961 };
2962 
2963 static const struct adreno_gpu_funcs funcs_gmuwrapper = {
2964 	.base = {
2965 		.get_param = adreno_get_param,
2966 		.set_param = adreno_set_param,
2967 		.hw_init = a6xx_hw_init,
2968 		.ucode_load = a6xx_ucode_load,
2969 		.pm_suspend = a6xx_pm_suspend,
2970 		.pm_resume = a6xx_pm_resume,
2971 		.recover = a6xx_recover,
2972 		.submit = a6xx_submit,
2973 		.active_ring = a6xx_active_ring,
2974 		.irq = a6xx_irq,
2975 		.destroy = a6xx_destroy,
2976 #if defined(CONFIG_DRM_MSM_GPU_STATE)
2977 		.show = a6xx_show,
2978 #endif
2979 		.gpu_busy = a6xx_gpu_busy,
2980 #if defined(CONFIG_DRM_MSM_GPU_STATE)
2981 		.gpu_state_get = a6xx_gpu_state_get,
2982 		.gpu_state_put = a6xx_gpu_state_put,
2983 #endif
2984 		.create_address_space = a6xx_create_address_space,
2985 		.create_private_address_space = a6xx_create_private_address_space,
2986 		.get_rptr = a6xx_get_rptr,
2987 		.progress = a6xx_progress,
2988 	},
2989 	.get_timestamp = a6xx_get_timestamp,
2990 };
2991 
2992 static const struct adreno_gpu_funcs funcs_a7xx = {
2993 	.base = {
2994 		.get_param = adreno_get_param,
2995 		.set_param = adreno_set_param,
2996 		.hw_init = a6xx_hw_init,
2997 		.ucode_load = a6xx_ucode_load,
2998 		.pm_suspend = a6xx_gmu_pm_suspend,
2999 		.pm_resume = a6xx_gmu_pm_resume,
3000 		.recover = a6xx_recover,
3001 		.submit = a7xx_submit,
3002 		.active_ring = a6xx_active_ring,
3003 		.irq = a6xx_irq,
3004 		.destroy = a6xx_destroy,
3005 #if defined(CONFIG_DRM_MSM_GPU_STATE)
3006 		.show = a6xx_show,
3007 #endif
3008 		.gpu_busy = a6xx_gpu_busy,
3009 		.gpu_get_freq = a6xx_gmu_get_freq,
3010 		.gpu_set_freq = a6xx_gpu_set_freq,
3011 #if defined(CONFIG_DRM_MSM_GPU_STATE)
3012 		.gpu_state_get = a6xx_gpu_state_get,
3013 		.gpu_state_put = a6xx_gpu_state_put,
3014 #endif
3015 		.create_address_space = a6xx_create_address_space,
3016 		.create_private_address_space = a6xx_create_private_address_space,
3017 		.get_rptr = a6xx_get_rptr,
3018 		.progress = a6xx_progress,
3019 	},
3020 	.get_timestamp = a6xx_gmu_get_timestamp,
3021 };
3022 
3023 struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
3024 {
3025 	struct msm_drm_private *priv = dev->dev_private;
3026 	struct platform_device *pdev = priv->gpu_pdev;
3027 	struct adreno_platform_config *config = pdev->dev.platform_data;
3028 	struct device_node *node;
3029 	struct a6xx_gpu *a6xx_gpu;
3030 	struct adreno_gpu *adreno_gpu;
3031 	struct msm_gpu *gpu;
3032 	bool is_a7xx;
3033 	int ret;
3034 
3035 	a6xx_gpu = kzalloc(sizeof(*a6xx_gpu), GFP_KERNEL);
3036 	if (!a6xx_gpu)
3037 		return ERR_PTR(-ENOMEM);
3038 
3039 	adreno_gpu = &a6xx_gpu->base;
3040 	gpu = &adreno_gpu->base;
3041 
3042 	mutex_init(&a6xx_gpu->gmu.lock);
3043 
3044 	adreno_gpu->registers = NULL;
3045 
3046 	/* Check if there is a GMU phandle and set it up */
3047 	node = of_parse_phandle(pdev->dev.of_node, "qcom,gmu", 0);
3048 	/* FIXME: How do we gracefully handle this? */
3049 	BUG_ON(!node);
3050 
3051 	adreno_gpu->gmu_is_wrapper = of_device_is_compatible(node, "qcom,adreno-gmu-wrapper");
3052 
3053 	adreno_gpu->base.hw_apriv =
3054 		!!(config->info->quirks & ADRENO_QUIRK_HAS_HW_APRIV);
3055 
3056 	/* gpu->info only gets assigned in adreno_gpu_init() */
3057 	is_a7xx = config->info->family == ADRENO_7XX_GEN1 ||
3058 		  config->info->family == ADRENO_7XX_GEN2 ||
3059 		  config->info->family == ADRENO_7XX_GEN3;
3060 
3061 	a6xx_llc_slices_init(pdev, a6xx_gpu, is_a7xx);
3062 
3063 	ret = a6xx_set_supported_hw(&pdev->dev, config->info);
3064 	if (ret) {
3065 		a6xx_destroy(&(a6xx_gpu->base.base));
3066 		return ERR_PTR(ret);
3067 	}
3068 
3069 	if (is_a7xx)
3070 		ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs_a7xx, 1);
3071 	else if (adreno_has_gmu_wrapper(adreno_gpu))
3072 		ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs_gmuwrapper, 1);
3073 	else
3074 		ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1);
3075 	if (ret) {
3076 		a6xx_destroy(&(a6xx_gpu->base.base));
3077 		return ERR_PTR(ret);
3078 	}
3079 
3080 	/*
3081 	 * For now only clamp to idle freq for devices where this is known not
3082 	 * to cause power supply issues:
3083 	 */
3084 	if (adreno_is_a618(adreno_gpu) || adreno_is_7c3(adreno_gpu))
3085 		priv->gpu_clamp_to_idle = true;
3086 
3087 	if (adreno_has_gmu_wrapper(adreno_gpu))
3088 		ret = a6xx_gmu_wrapper_init(a6xx_gpu, node);
3089 	else
3090 		ret = a6xx_gmu_init(a6xx_gpu, node);
3091 	of_node_put(node);
3092 	if (ret) {
3093 		a6xx_destroy(&(a6xx_gpu->base.base));
3094 		return ERR_PTR(ret);
3095 	}
3096 
3097 	if (gpu->aspace)
3098 		msm_mmu_set_fault_handler(gpu->aspace->mmu, gpu,
3099 				a6xx_fault_handler);
3100 
3101 	a6xx_calc_ubwc_config(adreno_gpu);
3102 
3103 	return gpu;
3104 }
3105