xref: /openbmc/linux/drivers/gpu/drm/msm/adreno/a6xx_gpu.c (revision bf459478)
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/nvmem-consumer.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 (!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 a6xx_flush(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 	uint32_t wptr;
60 	unsigned long flags;
61 
62 	/* Expanded APRIV doesn't need to issue the WHERE_AM_I opcode */
63 	if (a6xx_gpu->has_whereami && !adreno_gpu->base.hw_apriv) {
64 		struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
65 
66 		OUT_PKT7(ring, CP_WHERE_AM_I, 2);
67 		OUT_RING(ring, lower_32_bits(shadowptr(a6xx_gpu, ring)));
68 		OUT_RING(ring, upper_32_bits(shadowptr(a6xx_gpu, ring)));
69 	}
70 
71 	spin_lock_irqsave(&ring->preempt_lock, flags);
72 
73 	/* Copy the shadow to the actual register */
74 	ring->cur = ring->next;
75 
76 	/* Make sure to wrap wptr if we need to */
77 	wptr = get_wptr(ring);
78 
79 	spin_unlock_irqrestore(&ring->preempt_lock, flags);
80 
81 	/* Make sure everything is posted before making a decision */
82 	mb();
83 
84 	gpu_write(gpu, REG_A6XX_CP_RB_WPTR, wptr);
85 }
86 
87 static void get_stats_counter(struct msm_ringbuffer *ring, u32 counter,
88 		u64 iova)
89 {
90 	OUT_PKT7(ring, CP_REG_TO_MEM, 3);
91 	OUT_RING(ring, CP_REG_TO_MEM_0_REG(counter) |
92 		CP_REG_TO_MEM_0_CNT(2) |
93 		CP_REG_TO_MEM_0_64B);
94 	OUT_RING(ring, lower_32_bits(iova));
95 	OUT_RING(ring, upper_32_bits(iova));
96 }
97 
98 static void a6xx_set_pagetable(struct a6xx_gpu *a6xx_gpu,
99 		struct msm_ringbuffer *ring, struct msm_file_private *ctx)
100 {
101 	phys_addr_t ttbr;
102 	u32 asid;
103 	u64 memptr = rbmemptr(ring, ttbr0);
104 
105 	if (ctx == a6xx_gpu->cur_ctx)
106 		return;
107 
108 	if (msm_iommu_pagetable_params(ctx->aspace->mmu, &ttbr, &asid))
109 		return;
110 
111 	/* Execute the table update */
112 	OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
113 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_0_TTBR0_LO(lower_32_bits(ttbr)));
114 
115 	OUT_RING(ring,
116 		CP_SMMU_TABLE_UPDATE_1_TTBR0_HI(upper_32_bits(ttbr)) |
117 		CP_SMMU_TABLE_UPDATE_1_ASID(asid));
118 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_2_CONTEXTIDR(0));
119 	OUT_RING(ring, CP_SMMU_TABLE_UPDATE_3_CONTEXTBANK(0));
120 
121 	/*
122 	 * Write the new TTBR0 to the memstore. This is good for debugging.
123 	 */
124 	OUT_PKT7(ring, CP_MEM_WRITE, 4);
125 	OUT_RING(ring, CP_MEM_WRITE_0_ADDR_LO(lower_32_bits(memptr)));
126 	OUT_RING(ring, CP_MEM_WRITE_1_ADDR_HI(upper_32_bits(memptr)));
127 	OUT_RING(ring, lower_32_bits(ttbr));
128 	OUT_RING(ring, (asid << 16) | upper_32_bits(ttbr));
129 
130 	/*
131 	 * And finally, trigger a uche flush to be sure there isn't anything
132 	 * lingering in that part of the GPU
133 	 */
134 
135 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
136 	OUT_RING(ring, 0x31);
137 
138 	a6xx_gpu->cur_ctx = ctx;
139 }
140 
141 static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit)
142 {
143 	unsigned int index = submit->seqno % MSM_GPU_SUBMIT_STATS_COUNT;
144 	struct msm_drm_private *priv = gpu->dev->dev_private;
145 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
146 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
147 	struct msm_ringbuffer *ring = submit->ring;
148 	unsigned int i;
149 
150 	a6xx_set_pagetable(a6xx_gpu, ring, submit->queue->ctx);
151 
152 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
153 		rbmemptr_stats(ring, index, cpcycles_start));
154 
155 	/*
156 	 * For PM4 the GMU register offsets are calculated from the base of the
157 	 * GPU registers so we need to add 0x1a800 to the register value on A630
158 	 * to get the right value from PM4.
159 	 */
160 	get_stats_counter(ring, REG_A6XX_GMU_ALWAYS_ON_COUNTER_L + 0x1a800,
161 		rbmemptr_stats(ring, index, alwayson_start));
162 
163 	/* Invalidate CCU depth and color */
164 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
165 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(PC_CCU_INVALIDATE_DEPTH));
166 
167 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
168 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(PC_CCU_INVALIDATE_COLOR));
169 
170 	/* Submit the commands */
171 	for (i = 0; i < submit->nr_cmds; i++) {
172 		switch (submit->cmd[i].type) {
173 		case MSM_SUBMIT_CMD_IB_TARGET_BUF:
174 			break;
175 		case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
176 			if (priv->lastctx == submit->queue->ctx)
177 				break;
178 			fallthrough;
179 		case MSM_SUBMIT_CMD_BUF:
180 			OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3);
181 			OUT_RING(ring, lower_32_bits(submit->cmd[i].iova));
182 			OUT_RING(ring, upper_32_bits(submit->cmd[i].iova));
183 			OUT_RING(ring, submit->cmd[i].size);
184 			break;
185 		}
186 	}
187 
188 	get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
189 		rbmemptr_stats(ring, index, cpcycles_end));
190 	get_stats_counter(ring, REG_A6XX_GMU_ALWAYS_ON_COUNTER_L + 0x1a800,
191 		rbmemptr_stats(ring, index, alwayson_end));
192 
193 	/* Write the fence to the scratch register */
194 	OUT_PKT4(ring, REG_A6XX_CP_SCRATCH_REG(2), 1);
195 	OUT_RING(ring, submit->seqno);
196 
197 	/*
198 	 * Execute a CACHE_FLUSH_TS event. This will ensure that the
199 	 * timestamp is written to the memory and then triggers the interrupt
200 	 */
201 	OUT_PKT7(ring, CP_EVENT_WRITE, 4);
202 	OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(CACHE_FLUSH_TS) |
203 		CP_EVENT_WRITE_0_IRQ);
204 	OUT_RING(ring, lower_32_bits(rbmemptr(ring, fence)));
205 	OUT_RING(ring, upper_32_bits(rbmemptr(ring, fence)));
206 	OUT_RING(ring, submit->seqno);
207 
208 	trace_msm_gpu_submit_flush(submit,
209 		gmu_read64(&a6xx_gpu->gmu, REG_A6XX_GMU_ALWAYS_ON_COUNTER_L,
210 			REG_A6XX_GMU_ALWAYS_ON_COUNTER_H));
211 
212 	a6xx_flush(gpu, ring);
213 }
214 
215 const struct adreno_reglist a630_hwcg[] = {
216 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222},
217 	{REG_A6XX_RBBM_CLOCK_CNTL_SP1, 0x22222222},
218 	{REG_A6XX_RBBM_CLOCK_CNTL_SP2, 0x22222222},
219 	{REG_A6XX_RBBM_CLOCK_CNTL_SP3, 0x22222222},
220 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02022220},
221 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP1, 0x02022220},
222 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP2, 0x02022220},
223 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP3, 0x02022220},
224 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
225 	{REG_A6XX_RBBM_CLOCK_DELAY_SP1, 0x00000080},
226 	{REG_A6XX_RBBM_CLOCK_DELAY_SP2, 0x00000080},
227 	{REG_A6XX_RBBM_CLOCK_DELAY_SP3, 0x00000080},
228 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000f3cf},
229 	{REG_A6XX_RBBM_CLOCK_HYST_SP1, 0x0000f3cf},
230 	{REG_A6XX_RBBM_CLOCK_HYST_SP2, 0x0000f3cf},
231 	{REG_A6XX_RBBM_CLOCK_HYST_SP3, 0x0000f3cf},
232 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
233 	{REG_A6XX_RBBM_CLOCK_CNTL_TP1, 0x02222222},
234 	{REG_A6XX_RBBM_CLOCK_CNTL_TP2, 0x02222222},
235 	{REG_A6XX_RBBM_CLOCK_CNTL_TP3, 0x02222222},
236 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
237 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP1, 0x22222222},
238 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP2, 0x22222222},
239 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP3, 0x22222222},
240 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
241 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP1, 0x22222222},
242 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP2, 0x22222222},
243 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP3, 0x22222222},
244 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
245 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP1, 0x00022222},
246 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP2, 0x00022222},
247 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP3, 0x00022222},
248 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
249 	{REG_A6XX_RBBM_CLOCK_HYST_TP1, 0x77777777},
250 	{REG_A6XX_RBBM_CLOCK_HYST_TP2, 0x77777777},
251 	{REG_A6XX_RBBM_CLOCK_HYST_TP3, 0x77777777},
252 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
253 	{REG_A6XX_RBBM_CLOCK_HYST2_TP1, 0x77777777},
254 	{REG_A6XX_RBBM_CLOCK_HYST2_TP2, 0x77777777},
255 	{REG_A6XX_RBBM_CLOCK_HYST2_TP3, 0x77777777},
256 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
257 	{REG_A6XX_RBBM_CLOCK_HYST3_TP1, 0x77777777},
258 	{REG_A6XX_RBBM_CLOCK_HYST3_TP2, 0x77777777},
259 	{REG_A6XX_RBBM_CLOCK_HYST3_TP3, 0x77777777},
260 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
261 	{REG_A6XX_RBBM_CLOCK_HYST4_TP1, 0x00077777},
262 	{REG_A6XX_RBBM_CLOCK_HYST4_TP2, 0x00077777},
263 	{REG_A6XX_RBBM_CLOCK_HYST4_TP3, 0x00077777},
264 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
265 	{REG_A6XX_RBBM_CLOCK_DELAY_TP1, 0x11111111},
266 	{REG_A6XX_RBBM_CLOCK_DELAY_TP2, 0x11111111},
267 	{REG_A6XX_RBBM_CLOCK_DELAY_TP3, 0x11111111},
268 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
269 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP1, 0x11111111},
270 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP2, 0x11111111},
271 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP3, 0x11111111},
272 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
273 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP1, 0x11111111},
274 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP2, 0x11111111},
275 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP3, 0x11111111},
276 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
277 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP1, 0x00011111},
278 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP2, 0x00011111},
279 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP3, 0x00011111},
280 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
281 	{REG_A6XX_RBBM_CLOCK_CNTL2_UCHE, 0x22222222},
282 	{REG_A6XX_RBBM_CLOCK_CNTL3_UCHE, 0x22222222},
283 	{REG_A6XX_RBBM_CLOCK_CNTL4_UCHE, 0x00222222},
284 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
285 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
286 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
287 	{REG_A6XX_RBBM_CLOCK_CNTL_RB1, 0x22222222},
288 	{REG_A6XX_RBBM_CLOCK_CNTL_RB2, 0x22222222},
289 	{REG_A6XX_RBBM_CLOCK_CNTL_RB3, 0x22222222},
290 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x00002222},
291 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB1, 0x00002222},
292 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB2, 0x00002222},
293 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB3, 0x00002222},
294 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
295 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU1, 0x00002220},
296 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU2, 0x00002220},
297 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU3, 0x00002220},
298 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040f00},
299 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU1, 0x00040f00},
300 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU2, 0x00040f00},
301 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU3, 0x00040f00},
302 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05022022},
303 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
304 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
305 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
306 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
307 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
308 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
309 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
310 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
311 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
312 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
313 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
314 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
315 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
316 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
317 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
318 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
319 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
320 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
321 	{},
322 };
323 
324 const struct adreno_reglist a640_hwcg[] = {
325 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
326 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
327 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
328 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
329 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
330 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
331 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
332 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
333 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
334 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
335 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
336 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
337 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
338 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
339 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
340 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
341 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
342 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
343 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
344 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
345 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05222022},
346 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
347 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
348 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
349 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
350 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
351 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
352 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
353 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
354 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
355 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
356 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
357 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
358 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
359 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
360 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
361 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
362 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
363 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
364 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000000},
365 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
366 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
367 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
368 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
369 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
370 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
371 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
372 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
373 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
374 	{},
375 };
376 
377 const struct adreno_reglist a650_hwcg[] = {
378 	{REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
379 	{REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
380 	{REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
381 	{REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
382 	{REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
383 	{REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
384 	{REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
385 	{REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
386 	{REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
387 	{REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
388 	{REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
389 	{REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
390 	{REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
391 	{REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
392 	{REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
393 	{REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
394 	{REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
395 	{REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
396 	{REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
397 	{REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
398 	{REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022},
399 	{REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
400 	{REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
401 	{REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
402 	{REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
403 	{REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
404 	{REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
405 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
406 	{REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
407 	{REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
408 	{REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
409 	{REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
410 	{REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
411 	{REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
412 	{REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
413 	{REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
414 	{REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
415 	{REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
416 	{REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
417 	{REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000777},
418 	{REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
419 	{REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
420 	{REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
421 	{REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
422 	{REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
423 	{REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
424 	{REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
425 	{REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
426 	{REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
427 	{},
428 };
429 
430 static void a6xx_set_hwcg(struct msm_gpu *gpu, bool state)
431 {
432 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
433 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
434 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
435 	const struct adreno_reglist *reg;
436 	unsigned int i;
437 	u32 val, clock_cntl_on;
438 
439 	if (!adreno_gpu->info->hwcg)
440 		return;
441 
442 	if (adreno_is_a630(adreno_gpu))
443 		clock_cntl_on = 0x8aa8aa02;
444 	else
445 		clock_cntl_on = 0x8aa8aa82;
446 
447 	val = gpu_read(gpu, REG_A6XX_RBBM_CLOCK_CNTL);
448 
449 	/* Don't re-program the registers if they are already correct */
450 	if ((!state && !val) || (state && (val == clock_cntl_on)))
451 		return;
452 
453 	/* Disable SP clock before programming HWCG registers */
454 	gmu_rmw(gmu, REG_A6XX_GPU_GMU_GX_SPTPRAC_CLOCK_CONTROL, 1, 0);
455 
456 	for (i = 0; (reg = &adreno_gpu->info->hwcg[i], reg->offset); i++)
457 		gpu_write(gpu, reg->offset, state ? reg->value : 0);
458 
459 	/* Enable SP clock */
460 	gmu_rmw(gmu, REG_A6XX_GPU_GMU_GX_SPTPRAC_CLOCK_CONTROL, 0, 1);
461 
462 	gpu_write(gpu, REG_A6XX_RBBM_CLOCK_CNTL, state ? clock_cntl_on : 0);
463 }
464 
465 static void a6xx_set_ubwc_config(struct msm_gpu *gpu)
466 {
467 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
468 	u32 lower_bit = 2;
469 	u32 amsbc = 0;
470 	u32 rgb565_predicator = 0;
471 	u32 uavflagprd_inv = 0;
472 
473 	/* a618 is using the hw default values */
474 	if (adreno_is_a618(adreno_gpu))
475 		return;
476 
477 	if (adreno_is_a640(adreno_gpu))
478 		amsbc = 1;
479 
480 	if (adreno_is_a650(adreno_gpu)) {
481 		/* TODO: get ddr type from bootloader and use 2 for LPDDR4 */
482 		lower_bit = 3;
483 		amsbc = 1;
484 		rgb565_predicator = 1;
485 		uavflagprd_inv = 2;
486 	}
487 
488 	gpu_write(gpu, REG_A6XX_RB_NC_MODE_CNTL,
489 		rgb565_predicator << 11 | amsbc << 4 | lower_bit << 1);
490 	gpu_write(gpu, REG_A6XX_TPL1_NC_MODE_CNTL, lower_bit << 1);
491 	gpu_write(gpu, REG_A6XX_SP_NC_MODE_CNTL,
492 		uavflagprd_inv >> 4 | lower_bit << 1);
493 	gpu_write(gpu, REG_A6XX_UCHE_MODE_CNTL, lower_bit << 21);
494 }
495 
496 static int a6xx_cp_init(struct msm_gpu *gpu)
497 {
498 	struct msm_ringbuffer *ring = gpu->rb[0];
499 
500 	OUT_PKT7(ring, CP_ME_INIT, 8);
501 
502 	OUT_RING(ring, 0x0000002f);
503 
504 	/* Enable multiple hardware contexts */
505 	OUT_RING(ring, 0x00000003);
506 
507 	/* Enable error detection */
508 	OUT_RING(ring, 0x20000000);
509 
510 	/* Don't enable header dump */
511 	OUT_RING(ring, 0x00000000);
512 	OUT_RING(ring, 0x00000000);
513 
514 	/* No workarounds enabled */
515 	OUT_RING(ring, 0x00000000);
516 
517 	/* Pad rest of the cmds with 0's */
518 	OUT_RING(ring, 0x00000000);
519 	OUT_RING(ring, 0x00000000);
520 
521 	a6xx_flush(gpu, ring);
522 	return a6xx_idle(gpu, ring) ? 0 : -EINVAL;
523 }
524 
525 static void a6xx_ucode_check_version(struct a6xx_gpu *a6xx_gpu,
526 		struct drm_gem_object *obj)
527 {
528 	u32 *buf = msm_gem_get_vaddr(obj);
529 
530 	if (IS_ERR(buf))
531 		return;
532 
533 	/*
534 	 * If the lowest nibble is 0xa that is an indication that this microcode
535 	 * has been patched. The actual version is in dword [3] but we only care
536 	 * about the patchlevel which is the lowest nibble of dword [3]
537 	 *
538 	 * Otherwise check that the firmware is greater than or equal to 1.90
539 	 * which was the first version that had this fix built in
540 	 */
541 	if (((buf[0] & 0xf) == 0xa) && (buf[2] & 0xf) >= 1)
542 		a6xx_gpu->has_whereami = true;
543 	else if ((buf[0] & 0xfff) > 0x190)
544 		a6xx_gpu->has_whereami = true;
545 
546 	msm_gem_put_vaddr(obj);
547 }
548 
549 static int a6xx_ucode_init(struct msm_gpu *gpu)
550 {
551 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
552 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
553 
554 	if (!a6xx_gpu->sqe_bo) {
555 		a6xx_gpu->sqe_bo = adreno_fw_create_bo(gpu,
556 			adreno_gpu->fw[ADRENO_FW_SQE], &a6xx_gpu->sqe_iova);
557 
558 		if (IS_ERR(a6xx_gpu->sqe_bo)) {
559 			int ret = PTR_ERR(a6xx_gpu->sqe_bo);
560 
561 			a6xx_gpu->sqe_bo = NULL;
562 			DRM_DEV_ERROR(&gpu->pdev->dev,
563 				"Could not allocate SQE ucode: %d\n", ret);
564 
565 			return ret;
566 		}
567 
568 		msm_gem_object_set_name(a6xx_gpu->sqe_bo, "sqefw");
569 		a6xx_ucode_check_version(a6xx_gpu, a6xx_gpu->sqe_bo);
570 	}
571 
572 	gpu_write64(gpu, REG_A6XX_CP_SQE_INSTR_BASE_LO,
573 		REG_A6XX_CP_SQE_INSTR_BASE_HI, a6xx_gpu->sqe_iova);
574 
575 	return 0;
576 }
577 
578 static int a6xx_zap_shader_init(struct msm_gpu *gpu)
579 {
580 	static bool loaded;
581 	int ret;
582 
583 	if (loaded)
584 		return 0;
585 
586 	ret = adreno_zap_shader_load(gpu, GPU_PAS_ID);
587 
588 	loaded = !ret;
589 	return ret;
590 }
591 
592 #define A6XX_INT_MASK (A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR | \
593 	  A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW | \
594 	  A6XX_RBBM_INT_0_MASK_CP_HW_ERROR | \
595 	  A6XX_RBBM_INT_0_MASK_CP_IB2 | \
596 	  A6XX_RBBM_INT_0_MASK_CP_IB1 | \
597 	  A6XX_RBBM_INT_0_MASK_CP_RB | \
598 	  A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \
599 	  A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW | \
600 	  A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT | \
601 	  A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \
602 	  A6XX_RBBM_INT_0_MASK_UCHE_TRAP_INTR)
603 
604 static int a6xx_hw_init(struct msm_gpu *gpu)
605 {
606 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
607 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
608 	int ret;
609 
610 	/* Make sure the GMU keeps the GPU on while we set it up */
611 	a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
612 
613 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_CNTL, 0);
614 
615 	/*
616 	 * Disable the trusted memory range - we don't actually supported secure
617 	 * memory rendering at this point in time and we don't want to block off
618 	 * part of the virtual memory space.
619 	 */
620 	gpu_write64(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_BASE_LO,
621 		REG_A6XX_RBBM_SECVID_TSB_TRUSTED_BASE_HI, 0x00000000);
622 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_SIZE, 0x00000000);
623 
624 	/* Turn on 64 bit addressing for all blocks */
625 	gpu_write(gpu, REG_A6XX_CP_ADDR_MODE_CNTL, 0x1);
626 	gpu_write(gpu, REG_A6XX_VSC_ADDR_MODE_CNTL, 0x1);
627 	gpu_write(gpu, REG_A6XX_GRAS_ADDR_MODE_CNTL, 0x1);
628 	gpu_write(gpu, REG_A6XX_RB_ADDR_MODE_CNTL, 0x1);
629 	gpu_write(gpu, REG_A6XX_PC_ADDR_MODE_CNTL, 0x1);
630 	gpu_write(gpu, REG_A6XX_HLSQ_ADDR_MODE_CNTL, 0x1);
631 	gpu_write(gpu, REG_A6XX_VFD_ADDR_MODE_CNTL, 0x1);
632 	gpu_write(gpu, REG_A6XX_VPC_ADDR_MODE_CNTL, 0x1);
633 	gpu_write(gpu, REG_A6XX_UCHE_ADDR_MODE_CNTL, 0x1);
634 	gpu_write(gpu, REG_A6XX_SP_ADDR_MODE_CNTL, 0x1);
635 	gpu_write(gpu, REG_A6XX_TPL1_ADDR_MODE_CNTL, 0x1);
636 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_ADDR_MODE_CNTL, 0x1);
637 
638 	/* enable hardware clockgating */
639 	a6xx_set_hwcg(gpu, true);
640 
641 	/* VBIF/GBIF start*/
642 	if (adreno_is_a640(adreno_gpu) || adreno_is_a650(adreno_gpu)) {
643 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE0, 0x00071620);
644 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE1, 0x00071620);
645 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE2, 0x00071620);
646 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
647 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
648 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL, 0x3);
649 	} else {
650 		gpu_write(gpu, REG_A6XX_RBBM_VBIF_CLIENT_QOS_CNTL, 0x3);
651 	}
652 
653 	if (adreno_is_a630(adreno_gpu))
654 		gpu_write(gpu, REG_A6XX_VBIF_GATE_OFF_WRREQ_EN, 0x00000009);
655 
656 	/* Make all blocks contribute to the GPU BUSY perf counter */
657 	gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_GPU_BUSY_MASKED, 0xffffffff);
658 
659 	/* Disable L2 bypass in the UCHE */
660 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX_LO, 0xffffffc0);
661 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX_HI, 0x0001ffff);
662 	gpu_write(gpu, REG_A6XX_UCHE_TRAP_BASE_LO, 0xfffff000);
663 	gpu_write(gpu, REG_A6XX_UCHE_TRAP_BASE_HI, 0x0001ffff);
664 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE_LO, 0xfffff000);
665 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE_HI, 0x0001ffff);
666 
667 	if (!adreno_is_a650(adreno_gpu)) {
668 		/* Set the GMEM VA range [0x100000:0x100000 + gpu->gmem - 1] */
669 		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MIN_LO,
670 			REG_A6XX_UCHE_GMEM_RANGE_MIN_HI, 0x00100000);
671 
672 		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MAX_LO,
673 			REG_A6XX_UCHE_GMEM_RANGE_MAX_HI,
674 			0x00100000 + adreno_gpu->gmem - 1);
675 	}
676 
677 	gpu_write(gpu, REG_A6XX_UCHE_FILTER_CNTL, 0x804);
678 	gpu_write(gpu, REG_A6XX_UCHE_CACHE_WAYS, 0x4);
679 
680 	if (adreno_is_a640(adreno_gpu) || adreno_is_a650(adreno_gpu))
681 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x02000140);
682 	else
683 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x010000c0);
684 	gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x8040362c);
685 
686 	/* Setting the mem pool size */
687 	gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 128);
688 
689 	/* Setting the primFifo thresholds default values */
690 	if (adreno_is_a650(adreno_gpu))
691 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00300000);
692 	else if (adreno_is_a640(adreno_gpu))
693 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00200000);
694 	else
695 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, (0x300 << 11));
696 
697 	/* Set the AHB default slave response to "ERROR" */
698 	gpu_write(gpu, REG_A6XX_CP_AHB_CNTL, 0x1);
699 
700 	/* Turn on performance counters */
701 	gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_CNTL, 0x1);
702 
703 	/* Select CP0 to always count cycles */
704 	gpu_write(gpu, REG_A6XX_CP_PERFCTR_CP_SEL_0, PERF_CP_ALWAYS_COUNT);
705 
706 	a6xx_set_ubwc_config(gpu);
707 
708 	/* Enable fault detection */
709 	gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL,
710 		(1 << 30) | 0x1fffff);
711 
712 	gpu_write(gpu, REG_A6XX_UCHE_CLIENT_PF, 1);
713 
714 	/* Set weights for bicubic filtering */
715 	if (adreno_is_a650(adreno_gpu)) {
716 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_0, 0);
717 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_1,
718 			0x3fe05ff4);
719 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_2,
720 			0x3fa0ebee);
721 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_3,
722 			0x3f5193ed);
723 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_4,
724 			0x3f0243f0);
725 	}
726 
727 	/* Protect registers from the CP */
728 	gpu_write(gpu, REG_A6XX_CP_PROTECT_CNTL, 0x00000003);
729 
730 	gpu_write(gpu, REG_A6XX_CP_PROTECT(0),
731 		A6XX_PROTECT_RDONLY(0x600, 0x51));
732 	gpu_write(gpu, REG_A6XX_CP_PROTECT(1), A6XX_PROTECT_RW(0xae50, 0x2));
733 	gpu_write(gpu, REG_A6XX_CP_PROTECT(2), A6XX_PROTECT_RW(0x9624, 0x13));
734 	gpu_write(gpu, REG_A6XX_CP_PROTECT(3), A6XX_PROTECT_RW(0x8630, 0x8));
735 	gpu_write(gpu, REG_A6XX_CP_PROTECT(4), A6XX_PROTECT_RW(0x9e70, 0x1));
736 	gpu_write(gpu, REG_A6XX_CP_PROTECT(5), A6XX_PROTECT_RW(0x9e78, 0x187));
737 	gpu_write(gpu, REG_A6XX_CP_PROTECT(6), A6XX_PROTECT_RW(0xf000, 0x810));
738 	gpu_write(gpu, REG_A6XX_CP_PROTECT(7),
739 		A6XX_PROTECT_RDONLY(0xfc00, 0x3));
740 	gpu_write(gpu, REG_A6XX_CP_PROTECT(8), A6XX_PROTECT_RW(0x50e, 0x0));
741 	gpu_write(gpu, REG_A6XX_CP_PROTECT(9), A6XX_PROTECT_RDONLY(0x50f, 0x0));
742 	gpu_write(gpu, REG_A6XX_CP_PROTECT(10), A6XX_PROTECT_RW(0x510, 0x0));
743 	gpu_write(gpu, REG_A6XX_CP_PROTECT(11),
744 		A6XX_PROTECT_RDONLY(0x0, 0x4f9));
745 	gpu_write(gpu, REG_A6XX_CP_PROTECT(12),
746 		A6XX_PROTECT_RDONLY(0x501, 0xa));
747 	gpu_write(gpu, REG_A6XX_CP_PROTECT(13),
748 		A6XX_PROTECT_RDONLY(0x511, 0x44));
749 	gpu_write(gpu, REG_A6XX_CP_PROTECT(14), A6XX_PROTECT_RW(0xe00, 0xe));
750 	gpu_write(gpu, REG_A6XX_CP_PROTECT(15), A6XX_PROTECT_RW(0x8e00, 0x0));
751 	gpu_write(gpu, REG_A6XX_CP_PROTECT(16), A6XX_PROTECT_RW(0x8e50, 0xf));
752 	gpu_write(gpu, REG_A6XX_CP_PROTECT(17), A6XX_PROTECT_RW(0xbe02, 0x0));
753 	gpu_write(gpu, REG_A6XX_CP_PROTECT(18),
754 		A6XX_PROTECT_RW(0xbe20, 0x11f3));
755 	gpu_write(gpu, REG_A6XX_CP_PROTECT(19), A6XX_PROTECT_RW(0x800, 0x82));
756 	gpu_write(gpu, REG_A6XX_CP_PROTECT(20), A6XX_PROTECT_RW(0x8a0, 0x8));
757 	gpu_write(gpu, REG_A6XX_CP_PROTECT(21), A6XX_PROTECT_RW(0x8ab, 0x19));
758 	gpu_write(gpu, REG_A6XX_CP_PROTECT(22), A6XX_PROTECT_RW(0x900, 0x4d));
759 	gpu_write(gpu, REG_A6XX_CP_PROTECT(23), A6XX_PROTECT_RW(0x98d, 0x76));
760 	gpu_write(gpu, REG_A6XX_CP_PROTECT(24),
761 			A6XX_PROTECT_RDONLY(0x980, 0x4));
762 	gpu_write(gpu, REG_A6XX_CP_PROTECT(25), A6XX_PROTECT_RW(0xa630, 0x0));
763 
764 	/* Enable expanded apriv for targets that support it */
765 	if (gpu->hw_apriv) {
766 		gpu_write(gpu, REG_A6XX_CP_APRIV_CNTL,
767 			(1 << 6) | (1 << 5) | (1 << 3) | (1 << 2) | (1 << 1));
768 	}
769 
770 	/* Enable interrupts */
771 	gpu_write(gpu, REG_A6XX_RBBM_INT_0_MASK, A6XX_INT_MASK);
772 
773 	ret = adreno_hw_init(gpu);
774 	if (ret)
775 		goto out;
776 
777 	ret = a6xx_ucode_init(gpu);
778 	if (ret)
779 		goto out;
780 
781 	/* Set the ringbuffer address */
782 	gpu_write64(gpu, REG_A6XX_CP_RB_BASE, REG_A6XX_CP_RB_BASE_HI,
783 		gpu->rb[0]->iova);
784 
785 	/* Targets that support extended APRIV can use the RPTR shadow from
786 	 * hardware but all the other ones need to disable the feature. Targets
787 	 * that support the WHERE_AM_I opcode can use that instead
788 	 */
789 	if (adreno_gpu->base.hw_apriv)
790 		gpu_write(gpu, REG_A6XX_CP_RB_CNTL, MSM_GPU_RB_CNTL_DEFAULT);
791 	else
792 		gpu_write(gpu, REG_A6XX_CP_RB_CNTL,
793 			MSM_GPU_RB_CNTL_DEFAULT | AXXX_CP_RB_CNTL_NO_UPDATE);
794 
795 	/*
796 	 * Expanded APRIV and targets that support WHERE_AM_I both need a
797 	 * privileged buffer to store the RPTR shadow
798 	 */
799 
800 	if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami) {
801 		if (!a6xx_gpu->shadow_bo) {
802 			a6xx_gpu->shadow = msm_gem_kernel_new_locked(gpu->dev,
803 				sizeof(u32) * gpu->nr_rings,
804 				MSM_BO_UNCACHED | MSM_BO_MAP_PRIV,
805 				gpu->aspace, &a6xx_gpu->shadow_bo,
806 				&a6xx_gpu->shadow_iova);
807 
808 			if (IS_ERR(a6xx_gpu->shadow))
809 				return PTR_ERR(a6xx_gpu->shadow);
810 		}
811 
812 		gpu_write64(gpu, REG_A6XX_CP_RB_RPTR_ADDR_LO,
813 			REG_A6XX_CP_RB_RPTR_ADDR_HI,
814 			shadowptr(a6xx_gpu, gpu->rb[0]));
815 	}
816 
817 	/* Always come up on rb 0 */
818 	a6xx_gpu->cur_ring = gpu->rb[0];
819 
820 	a6xx_gpu->cur_ctx = NULL;
821 
822 	/* Enable the SQE_to start the CP engine */
823 	gpu_write(gpu, REG_A6XX_CP_SQE_CNTL, 1);
824 
825 	ret = a6xx_cp_init(gpu);
826 	if (ret)
827 		goto out;
828 
829 	/*
830 	 * Try to load a zap shader into the secure world. If successful
831 	 * we can use the CP to switch out of secure mode. If not then we
832 	 * have no resource but to try to switch ourselves out manually. If we
833 	 * guessed wrong then access to the RBBM_SECVID_TRUST_CNTL register will
834 	 * be blocked and a permissions violation will soon follow.
835 	 */
836 	ret = a6xx_zap_shader_init(gpu);
837 	if (!ret) {
838 		OUT_PKT7(gpu->rb[0], CP_SET_SECURE_MODE, 1);
839 		OUT_RING(gpu->rb[0], 0x00000000);
840 
841 		a6xx_flush(gpu, gpu->rb[0]);
842 		if (!a6xx_idle(gpu, gpu->rb[0]))
843 			return -EINVAL;
844 	} else if (ret == -ENODEV) {
845 		/*
846 		 * This device does not use zap shader (but print a warning
847 		 * just in case someone got their dt wrong.. hopefully they
848 		 * have a debug UART to realize the error of their ways...
849 		 * if you mess this up you are about to crash horribly)
850 		 */
851 		dev_warn_once(gpu->dev->dev,
852 			"Zap shader not enabled - using SECVID_TRUST_CNTL instead\n");
853 		gpu_write(gpu, REG_A6XX_RBBM_SECVID_TRUST_CNTL, 0x0);
854 		ret = 0;
855 	} else {
856 		return ret;
857 	}
858 
859 out:
860 	/*
861 	 * Tell the GMU that we are done touching the GPU and it can start power
862 	 * management
863 	 */
864 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
865 
866 	if (a6xx_gpu->gmu.legacy) {
867 		/* Take the GMU out of its special boot mode */
868 		a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
869 	}
870 
871 	return ret;
872 }
873 
874 static void a6xx_dump(struct msm_gpu *gpu)
875 {
876 	DRM_DEV_INFO(&gpu->pdev->dev, "status:   %08x\n",
877 			gpu_read(gpu, REG_A6XX_RBBM_STATUS));
878 	adreno_dump(gpu);
879 }
880 
881 #define VBIF_RESET_ACK_TIMEOUT	100
882 #define VBIF_RESET_ACK_MASK	0x00f0
883 
884 static void a6xx_recover(struct msm_gpu *gpu)
885 {
886 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
887 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
888 	int i;
889 
890 	adreno_dump_info(gpu);
891 
892 	for (i = 0; i < 8; i++)
893 		DRM_DEV_INFO(&gpu->pdev->dev, "CP_SCRATCH_REG%d: %u\n", i,
894 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(i)));
895 
896 	if (hang_debug)
897 		a6xx_dump(gpu);
898 
899 	/*
900 	 * Turn off keep alive that might have been enabled by the hang
901 	 * interrupt
902 	 */
903 	gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 0);
904 
905 	gpu->funcs->pm_suspend(gpu);
906 	gpu->funcs->pm_resume(gpu);
907 
908 	msm_gpu_hw_init(gpu);
909 }
910 
911 static int a6xx_fault_handler(void *arg, unsigned long iova, int flags)
912 {
913 	struct msm_gpu *gpu = arg;
914 
915 	pr_warn_ratelimited("*** gpu fault: iova=%08lx, flags=%d (%u,%u,%u,%u)\n",
916 			iova, flags,
917 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(4)),
918 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(5)),
919 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(6)),
920 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(7)));
921 
922 	return -EFAULT;
923 }
924 
925 static void a6xx_cp_hw_err_irq(struct msm_gpu *gpu)
926 {
927 	u32 status = gpu_read(gpu, REG_A6XX_CP_INTERRUPT_STATUS);
928 
929 	if (status & A6XX_CP_INT_CP_OPCODE_ERROR) {
930 		u32 val;
931 
932 		gpu_write(gpu, REG_A6XX_CP_SQE_STAT_ADDR, 1);
933 		val = gpu_read(gpu, REG_A6XX_CP_SQE_STAT_DATA);
934 		dev_err_ratelimited(&gpu->pdev->dev,
935 			"CP | opcode error | possible opcode=0x%8.8X\n",
936 			val);
937 	}
938 
939 	if (status & A6XX_CP_INT_CP_UCODE_ERROR)
940 		dev_err_ratelimited(&gpu->pdev->dev,
941 			"CP ucode error interrupt\n");
942 
943 	if (status & A6XX_CP_INT_CP_HW_FAULT_ERROR)
944 		dev_err_ratelimited(&gpu->pdev->dev, "CP | HW fault | status=0x%8.8X\n",
945 			gpu_read(gpu, REG_A6XX_CP_HW_FAULT));
946 
947 	if (status & A6XX_CP_INT_CP_REGISTER_PROTECTION_ERROR) {
948 		u32 val = gpu_read(gpu, REG_A6XX_CP_PROTECT_STATUS);
949 
950 		dev_err_ratelimited(&gpu->pdev->dev,
951 			"CP | protected mode error | %s | addr=0x%8.8X | status=0x%8.8X\n",
952 			val & (1 << 20) ? "READ" : "WRITE",
953 			(val & 0x3ffff), val);
954 	}
955 
956 	if (status & A6XX_CP_INT_CP_AHB_ERROR)
957 		dev_err_ratelimited(&gpu->pdev->dev, "CP AHB error interrupt\n");
958 
959 	if (status & A6XX_CP_INT_CP_VSD_PARITY_ERROR)
960 		dev_err_ratelimited(&gpu->pdev->dev, "CP VSD decoder parity error\n");
961 
962 	if (status & A6XX_CP_INT_CP_ILLEGAL_INSTR_ERROR)
963 		dev_err_ratelimited(&gpu->pdev->dev, "CP illegal instruction error\n");
964 
965 }
966 
967 static void a6xx_fault_detect_irq(struct msm_gpu *gpu)
968 {
969 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
970 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
971 	struct msm_ringbuffer *ring = gpu->funcs->active_ring(gpu);
972 
973 	/*
974 	 * Force the GPU to stay on until after we finish
975 	 * collecting information
976 	 */
977 	gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 1);
978 
979 	DRM_DEV_ERROR(&gpu->pdev->dev,
980 		"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",
981 		ring ? ring->id : -1, ring ? ring->seqno : 0,
982 		gpu_read(gpu, REG_A6XX_RBBM_STATUS),
983 		gpu_read(gpu, REG_A6XX_CP_RB_RPTR),
984 		gpu_read(gpu, REG_A6XX_CP_RB_WPTR),
985 		gpu_read64(gpu, REG_A6XX_CP_IB1_BASE, REG_A6XX_CP_IB1_BASE_HI),
986 		gpu_read(gpu, REG_A6XX_CP_IB1_REM_SIZE),
987 		gpu_read64(gpu, REG_A6XX_CP_IB2_BASE, REG_A6XX_CP_IB2_BASE_HI),
988 		gpu_read(gpu, REG_A6XX_CP_IB2_REM_SIZE));
989 
990 	/* Turn off the hangcheck timer to keep it from bothering us */
991 	del_timer(&gpu->hangcheck_timer);
992 
993 	kthread_queue_work(gpu->worker, &gpu->recover_work);
994 }
995 
996 static irqreturn_t a6xx_irq(struct msm_gpu *gpu)
997 {
998 	u32 status = gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS);
999 
1000 	gpu_write(gpu, REG_A6XX_RBBM_INT_CLEAR_CMD, status);
1001 
1002 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT)
1003 		a6xx_fault_detect_irq(gpu);
1004 
1005 	if (status & A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR)
1006 		dev_err_ratelimited(&gpu->pdev->dev, "CP | AHB bus error\n");
1007 
1008 	if (status & A6XX_RBBM_INT_0_MASK_CP_HW_ERROR)
1009 		a6xx_cp_hw_err_irq(gpu);
1010 
1011 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW)
1012 		dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB ASYNC overflow\n");
1013 
1014 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW)
1015 		dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB bus overflow\n");
1016 
1017 	if (status & A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS)
1018 		dev_err_ratelimited(&gpu->pdev->dev, "UCHE | Out of bounds access\n");
1019 
1020 	if (status & A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS)
1021 		msm_gpu_retire(gpu);
1022 
1023 	return IRQ_HANDLED;
1024 }
1025 
1026 static void a6xx_llc_rmw(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 mask, u32 or)
1027 {
1028 	return msm_rmw(a6xx_gpu->llc_mmio + (reg << 2), mask, or);
1029 }
1030 
1031 static void a6xx_llc_write(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 value)
1032 {
1033 	return msm_writel(value, a6xx_gpu->llc_mmio + (reg << 2));
1034 }
1035 
1036 static void a6xx_llc_deactivate(struct a6xx_gpu *a6xx_gpu)
1037 {
1038 	llcc_slice_deactivate(a6xx_gpu->llc_slice);
1039 	llcc_slice_deactivate(a6xx_gpu->htw_llc_slice);
1040 }
1041 
1042 static void a6xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
1043 {
1044 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
1045 	struct msm_gpu *gpu = &adreno_gpu->base;
1046 	u32 cntl1_regval = 0;
1047 
1048 	if (IS_ERR(a6xx_gpu->llc_mmio))
1049 		return;
1050 
1051 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
1052 		u32 gpu_scid = llcc_get_slice_id(a6xx_gpu->llc_slice);
1053 
1054 		gpu_scid &= 0x1f;
1055 		cntl1_regval = (gpu_scid << 0) | (gpu_scid << 5) | (gpu_scid << 10) |
1056 			       (gpu_scid << 15) | (gpu_scid << 20);
1057 	}
1058 
1059 	/*
1060 	 * For targets with a MMU500, activate the slice but don't program the
1061 	 * register.  The XBL will take care of that.
1062 	 */
1063 	if (!llcc_slice_activate(a6xx_gpu->htw_llc_slice)) {
1064 		if (!a6xx_gpu->have_mmu500) {
1065 			u32 gpuhtw_scid = llcc_get_slice_id(a6xx_gpu->htw_llc_slice);
1066 
1067 			gpuhtw_scid &= 0x1f;
1068 			cntl1_regval |= FIELD_PREP(GENMASK(29, 25), gpuhtw_scid);
1069 		}
1070 	}
1071 
1072 	if (cntl1_regval) {
1073 		/*
1074 		 * Program the slice IDs for the various GPU blocks and GPU MMU
1075 		 * pagetables
1076 		 */
1077 		if (a6xx_gpu->have_mmu500)
1078 			gpu_rmw(gpu, REG_A6XX_GBIF_SCACHE_CNTL1, GENMASK(24, 0),
1079 				cntl1_regval);
1080 		else {
1081 			a6xx_llc_write(a6xx_gpu,
1082 				REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_1, cntl1_regval);
1083 
1084 			/*
1085 			 * Program cacheability overrides to not allocate cache
1086 			 * lines on a write miss
1087 			 */
1088 			a6xx_llc_rmw(a6xx_gpu,
1089 				REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_0, 0xF, 0x03);
1090 		}
1091 	}
1092 }
1093 
1094 static void a6xx_llc_slices_destroy(struct a6xx_gpu *a6xx_gpu)
1095 {
1096 	llcc_slice_putd(a6xx_gpu->llc_slice);
1097 	llcc_slice_putd(a6xx_gpu->htw_llc_slice);
1098 }
1099 
1100 static void a6xx_llc_slices_init(struct platform_device *pdev,
1101 		struct a6xx_gpu *a6xx_gpu)
1102 {
1103 	struct device_node *phandle;
1104 
1105 	a6xx_gpu->llc_mmio = msm_ioremap(pdev, "cx_mem", "gpu_cx");
1106 	if (IS_ERR(a6xx_gpu->llc_mmio))
1107 		return;
1108 
1109 	/*
1110 	 * There is a different programming path for targets with an mmu500
1111 	 * attached, so detect if that is the case
1112 	 */
1113 	phandle = of_parse_phandle(pdev->dev.of_node, "iommus", 0);
1114 	a6xx_gpu->have_mmu500 = (phandle &&
1115 		of_device_is_compatible(phandle, "arm,mmu-500"));
1116 	of_node_put(phandle);
1117 
1118 	a6xx_gpu->llc_slice = llcc_slice_getd(LLCC_GPU);
1119 	a6xx_gpu->htw_llc_slice = llcc_slice_getd(LLCC_GPUHTW);
1120 
1121 	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
1122 		a6xx_gpu->llc_mmio = ERR_PTR(-EINVAL);
1123 }
1124 
1125 static int a6xx_pm_resume(struct msm_gpu *gpu)
1126 {
1127 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1128 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1129 	int ret;
1130 
1131 	gpu->needs_hw_init = true;
1132 
1133 	trace_msm_gpu_resume(0);
1134 
1135 	ret = a6xx_gmu_resume(a6xx_gpu);
1136 	if (ret)
1137 		return ret;
1138 
1139 	msm_gpu_resume_devfreq(gpu);
1140 
1141 	a6xx_llc_activate(a6xx_gpu);
1142 
1143 	return 0;
1144 }
1145 
1146 static int a6xx_pm_suspend(struct msm_gpu *gpu)
1147 {
1148 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1149 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1150 	int i, ret;
1151 
1152 	trace_msm_gpu_suspend(0);
1153 
1154 	a6xx_llc_deactivate(a6xx_gpu);
1155 
1156 	devfreq_suspend_device(gpu->devfreq.devfreq);
1157 
1158 	ret = a6xx_gmu_stop(a6xx_gpu);
1159 	if (ret)
1160 		return ret;
1161 
1162 	if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami)
1163 		for (i = 0; i < gpu->nr_rings; i++)
1164 			a6xx_gpu->shadow[i] = 0;
1165 
1166 	return 0;
1167 }
1168 
1169 static int a6xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
1170 {
1171 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1172 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1173 	static DEFINE_MUTEX(perfcounter_oob);
1174 
1175 	mutex_lock(&perfcounter_oob);
1176 
1177 	/* Force the GPU power on so we can read this register */
1178 	a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
1179 
1180 	*value = gpu_read64(gpu, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
1181 		REG_A6XX_RBBM_PERFCTR_CP_0_HI);
1182 
1183 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
1184 	mutex_unlock(&perfcounter_oob);
1185 	return 0;
1186 }
1187 
1188 static struct msm_ringbuffer *a6xx_active_ring(struct msm_gpu *gpu)
1189 {
1190 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1191 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1192 
1193 	return a6xx_gpu->cur_ring;
1194 }
1195 
1196 static void a6xx_destroy(struct msm_gpu *gpu)
1197 {
1198 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1199 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1200 
1201 	if (a6xx_gpu->sqe_bo) {
1202 		msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
1203 		drm_gem_object_put(a6xx_gpu->sqe_bo);
1204 	}
1205 
1206 	if (a6xx_gpu->shadow_bo) {
1207 		msm_gem_unpin_iova(a6xx_gpu->shadow_bo, gpu->aspace);
1208 		drm_gem_object_put(a6xx_gpu->shadow_bo);
1209 	}
1210 
1211 	a6xx_llc_slices_destroy(a6xx_gpu);
1212 
1213 	a6xx_gmu_remove(a6xx_gpu);
1214 
1215 	adreno_gpu_cleanup(adreno_gpu);
1216 
1217 	if (a6xx_gpu->opp_table)
1218 		dev_pm_opp_put_supported_hw(a6xx_gpu->opp_table);
1219 
1220 	kfree(a6xx_gpu);
1221 }
1222 
1223 static unsigned long a6xx_gpu_busy(struct msm_gpu *gpu)
1224 {
1225 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1226 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1227 	u64 busy_cycles, busy_time;
1228 
1229 
1230 	/* Only read the gpu busy if the hardware is already active */
1231 	if (pm_runtime_get_if_in_use(a6xx_gpu->gmu.dev) == 0)
1232 		return 0;
1233 
1234 	busy_cycles = gmu_read64(&a6xx_gpu->gmu,
1235 			REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_L,
1236 			REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_H);
1237 
1238 	busy_time = (busy_cycles - gpu->devfreq.busy_cycles) * 10;
1239 	do_div(busy_time, 192);
1240 
1241 	gpu->devfreq.busy_cycles = busy_cycles;
1242 
1243 	pm_runtime_put(a6xx_gpu->gmu.dev);
1244 
1245 	if (WARN_ON(busy_time > ~0LU))
1246 		return ~0LU;
1247 
1248 	return (unsigned long)busy_time;
1249 }
1250 
1251 static struct msm_gem_address_space *
1252 a6xx_create_address_space(struct msm_gpu *gpu, struct platform_device *pdev)
1253 {
1254 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1255 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1256 	struct iommu_domain *iommu;
1257 	struct msm_mmu *mmu;
1258 	struct msm_gem_address_space *aspace;
1259 	u64 start, size;
1260 
1261 	iommu = iommu_domain_alloc(&platform_bus_type);
1262 	if (!iommu)
1263 		return NULL;
1264 
1265 	/*
1266 	 * This allows GPU to set the bus attributes required to use system
1267 	 * cache on behalf of the iommu page table walker.
1268 	 */
1269 	if (!IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
1270 		adreno_set_llc_attributes(iommu);
1271 
1272 	mmu = msm_iommu_new(&pdev->dev, iommu);
1273 	if (IS_ERR(mmu)) {
1274 		iommu_domain_free(iommu);
1275 		return ERR_CAST(mmu);
1276 	}
1277 
1278 	/*
1279 	 * Use the aperture start or SZ_16M, whichever is greater. This will
1280 	 * ensure that we align with the allocated pagetable range while still
1281 	 * allowing room in the lower 32 bits for GMEM and whatnot
1282 	 */
1283 	start = max_t(u64, SZ_16M, iommu->geometry.aperture_start);
1284 	size = iommu->geometry.aperture_end - start + 1;
1285 
1286 	aspace = msm_gem_address_space_create(mmu, "gpu",
1287 		start & GENMASK_ULL(48, 0), size);
1288 
1289 	if (IS_ERR(aspace) && !IS_ERR(mmu))
1290 		mmu->funcs->destroy(mmu);
1291 
1292 	return aspace;
1293 }
1294 
1295 static struct msm_gem_address_space *
1296 a6xx_create_private_address_space(struct msm_gpu *gpu)
1297 {
1298 	struct msm_mmu *mmu;
1299 
1300 	mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
1301 
1302 	if (IS_ERR(mmu))
1303 		return ERR_CAST(mmu);
1304 
1305 	return msm_gem_address_space_create(mmu,
1306 		"gpu", 0x100000000ULL, 0x1ffffffffULL);
1307 }
1308 
1309 static uint32_t a6xx_get_rptr(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
1310 {
1311 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1312 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1313 
1314 	if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami)
1315 		return a6xx_gpu->shadow[ring->id];
1316 
1317 	return ring->memptrs->rptr = gpu_read(gpu, REG_A6XX_CP_RB_RPTR);
1318 }
1319 
1320 static u32 a618_get_speed_bin(u32 fuse)
1321 {
1322 	if (fuse == 0)
1323 		return 0;
1324 	else if (fuse == 169)
1325 		return 1;
1326 	else if (fuse == 174)
1327 		return 2;
1328 
1329 	return UINT_MAX;
1330 }
1331 
1332 static u32 fuse_to_supp_hw(struct device *dev, u32 revn, u32 fuse)
1333 {
1334 	u32 val = UINT_MAX;
1335 
1336 	if (revn == 618)
1337 		val = a618_get_speed_bin(fuse);
1338 
1339 	if (val == UINT_MAX) {
1340 		DRM_DEV_ERROR(dev,
1341 			"missing support for speed-bin: %u. Some OPPs may not be supported by hardware",
1342 			fuse);
1343 		return UINT_MAX;
1344 	}
1345 
1346 	return (1 << val);
1347 }
1348 
1349 static int a6xx_set_supported_hw(struct device *dev, struct a6xx_gpu *a6xx_gpu,
1350 		u32 revn)
1351 {
1352 	struct opp_table *opp_table;
1353 	struct nvmem_cell *cell;
1354 	u32 supp_hw = UINT_MAX;
1355 	void *buf;
1356 
1357 	cell = nvmem_cell_get(dev, "speed_bin");
1358 	/*
1359 	 * -ENOENT means that the platform doesn't support speedbin which is
1360 	 * fine
1361 	 */
1362 	if (PTR_ERR(cell) == -ENOENT)
1363 		return 0;
1364 	else if (IS_ERR(cell)) {
1365 		DRM_DEV_ERROR(dev,
1366 				"failed to read speed-bin. Some OPPs may not be supported by hardware");
1367 		goto done;
1368 	}
1369 
1370 	buf = nvmem_cell_read(cell, NULL);
1371 	if (IS_ERR(buf)) {
1372 		nvmem_cell_put(cell);
1373 		DRM_DEV_ERROR(dev,
1374 				"failed to read speed-bin. Some OPPs may not be supported by hardware");
1375 		goto done;
1376 	}
1377 
1378 	supp_hw = fuse_to_supp_hw(dev, revn, *((u32 *) buf));
1379 
1380 	kfree(buf);
1381 	nvmem_cell_put(cell);
1382 
1383 done:
1384 	opp_table = dev_pm_opp_set_supported_hw(dev, &supp_hw, 1);
1385 	if (IS_ERR(opp_table))
1386 		return PTR_ERR(opp_table);
1387 
1388 	a6xx_gpu->opp_table = opp_table;
1389 	return 0;
1390 }
1391 
1392 static const struct adreno_gpu_funcs funcs = {
1393 	.base = {
1394 		.get_param = adreno_get_param,
1395 		.hw_init = a6xx_hw_init,
1396 		.pm_suspend = a6xx_pm_suspend,
1397 		.pm_resume = a6xx_pm_resume,
1398 		.recover = a6xx_recover,
1399 		.submit = a6xx_submit,
1400 		.active_ring = a6xx_active_ring,
1401 		.irq = a6xx_irq,
1402 		.destroy = a6xx_destroy,
1403 #if defined(CONFIG_DRM_MSM_GPU_STATE)
1404 		.show = a6xx_show,
1405 #endif
1406 		.gpu_busy = a6xx_gpu_busy,
1407 		.gpu_get_freq = a6xx_gmu_get_freq,
1408 		.gpu_set_freq = a6xx_gmu_set_freq,
1409 #if defined(CONFIG_DRM_MSM_GPU_STATE)
1410 		.gpu_state_get = a6xx_gpu_state_get,
1411 		.gpu_state_put = a6xx_gpu_state_put,
1412 #endif
1413 		.create_address_space = a6xx_create_address_space,
1414 		.create_private_address_space = a6xx_create_private_address_space,
1415 		.get_rptr = a6xx_get_rptr,
1416 	},
1417 	.get_timestamp = a6xx_get_timestamp,
1418 };
1419 
1420 struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
1421 {
1422 	struct msm_drm_private *priv = dev->dev_private;
1423 	struct platform_device *pdev = priv->gpu_pdev;
1424 	struct adreno_platform_config *config = pdev->dev.platform_data;
1425 	const struct adreno_info *info;
1426 	struct device_node *node;
1427 	struct a6xx_gpu *a6xx_gpu;
1428 	struct adreno_gpu *adreno_gpu;
1429 	struct msm_gpu *gpu;
1430 	int ret;
1431 
1432 	a6xx_gpu = kzalloc(sizeof(*a6xx_gpu), GFP_KERNEL);
1433 	if (!a6xx_gpu)
1434 		return ERR_PTR(-ENOMEM);
1435 
1436 	adreno_gpu = &a6xx_gpu->base;
1437 	gpu = &adreno_gpu->base;
1438 
1439 	adreno_gpu->registers = NULL;
1440 
1441 	/*
1442 	 * We need to know the platform type before calling into adreno_gpu_init
1443 	 * so that the hw_apriv flag can be correctly set. Snoop into the info
1444 	 * and grab the revision number
1445 	 */
1446 	info = adreno_info(config->rev);
1447 
1448 	if (info && info->revn == 650)
1449 		adreno_gpu->base.hw_apriv = true;
1450 
1451 	a6xx_llc_slices_init(pdev, a6xx_gpu);
1452 
1453 	ret = a6xx_set_supported_hw(&pdev->dev, a6xx_gpu, info->revn);
1454 	if (ret) {
1455 		a6xx_destroy(&(a6xx_gpu->base.base));
1456 		return ERR_PTR(ret);
1457 	}
1458 
1459 	ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1);
1460 	if (ret) {
1461 		a6xx_destroy(&(a6xx_gpu->base.base));
1462 		return ERR_PTR(ret);
1463 	}
1464 
1465 	/* Check if there is a GMU phandle and set it up */
1466 	node = of_parse_phandle(pdev->dev.of_node, "qcom,gmu", 0);
1467 
1468 	/* FIXME: How do we gracefully handle this? */
1469 	BUG_ON(!node);
1470 
1471 	ret = a6xx_gmu_init(a6xx_gpu, node);
1472 	if (ret) {
1473 		a6xx_destroy(&(a6xx_gpu->base.base));
1474 		return ERR_PTR(ret);
1475 	}
1476 
1477 	if (gpu->aspace)
1478 		msm_mmu_set_fault_handler(gpu->aspace->mmu, gpu,
1479 				a6xx_fault_handler);
1480 
1481 	return gpu;
1482 }
1483