xref: /openbmc/linux/drivers/gpu/drm/msm/adreno/a6xx_gpu.c (revision 65a0d3c1)
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_CP_ALWAYS_ON_COUNTER_LO,
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_CP_ALWAYS_ON_COUNTER_LO,
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 		gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER_LO,
210 			REG_A6XX_CP_ALWAYS_ON_COUNTER_HI));
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 /* For a615, a616, a618, A619, a630, a640 and a680 */
466 static const u32 a6xx_protect[] = {
467 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
468 	A6XX_PROTECT_RDONLY(0x00501, 0x0005),
469 	A6XX_PROTECT_RDONLY(0x0050b, 0x02f4),
470 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
471 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
472 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
473 	A6XX_PROTECT_NORDWR(0x00800, 0x0082),
474 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
475 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
476 	A6XX_PROTECT_RDONLY(0x008de, 0x00ae),
477 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
478 	A6XX_PROTECT_NORDWR(0x0098d, 0x0272),
479 	A6XX_PROTECT_NORDWR(0x00e00, 0x0001),
480 	A6XX_PROTECT_NORDWR(0x00e03, 0x000c),
481 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
482 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
483 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
484 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
485 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
486 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
487 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
488 	A6XX_PROTECT_NORDWR(0x09e70, 0x0001),
489 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
490 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
491 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
492 	A6XX_PROTECT_NORDWR(0x0ae50, 0x032f),
493 	A6XX_PROTECT_NORDWR(0x0b604, 0x0000),
494 	A6XX_PROTECT_NORDWR(0x0be02, 0x0001),
495 	A6XX_PROTECT_NORDWR(0x0be20, 0x17df),
496 	A6XX_PROTECT_NORDWR(0x0f000, 0x0bff),
497 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
498 	A6XX_PROTECT_NORDWR(0x11c00, 0x0000), /* note: infinite range */
499 };
500 
501 /* These are for a620 and a650 */
502 static const u32 a650_protect[] = {
503 	A6XX_PROTECT_RDONLY(0x00000, 0x04ff),
504 	A6XX_PROTECT_RDONLY(0x00501, 0x0005),
505 	A6XX_PROTECT_RDONLY(0x0050b, 0x02f4),
506 	A6XX_PROTECT_NORDWR(0x0050e, 0x0000),
507 	A6XX_PROTECT_NORDWR(0x00510, 0x0000),
508 	A6XX_PROTECT_NORDWR(0x00534, 0x0000),
509 	A6XX_PROTECT_NORDWR(0x00800, 0x0082),
510 	A6XX_PROTECT_NORDWR(0x008a0, 0x0008),
511 	A6XX_PROTECT_NORDWR(0x008ab, 0x0024),
512 	A6XX_PROTECT_RDONLY(0x008de, 0x00ae),
513 	A6XX_PROTECT_NORDWR(0x00900, 0x004d),
514 	A6XX_PROTECT_NORDWR(0x0098d, 0x0272),
515 	A6XX_PROTECT_NORDWR(0x00e00, 0x0001),
516 	A6XX_PROTECT_NORDWR(0x00e03, 0x000c),
517 	A6XX_PROTECT_NORDWR(0x03c00, 0x00c3),
518 	A6XX_PROTECT_RDONLY(0x03cc4, 0x1fff),
519 	A6XX_PROTECT_NORDWR(0x08630, 0x01cf),
520 	A6XX_PROTECT_NORDWR(0x08e00, 0x0000),
521 	A6XX_PROTECT_NORDWR(0x08e08, 0x0000),
522 	A6XX_PROTECT_NORDWR(0x08e50, 0x001f),
523 	A6XX_PROTECT_NORDWR(0x08e80, 0x027f),
524 	A6XX_PROTECT_NORDWR(0x09624, 0x01db),
525 	A6XX_PROTECT_NORDWR(0x09e60, 0x0011),
526 	A6XX_PROTECT_NORDWR(0x09e78, 0x0187),
527 	A6XX_PROTECT_NORDWR(0x0a630, 0x01cf),
528 	A6XX_PROTECT_NORDWR(0x0ae02, 0x0000),
529 	A6XX_PROTECT_NORDWR(0x0ae50, 0x032f),
530 	A6XX_PROTECT_NORDWR(0x0b604, 0x0000),
531 	A6XX_PROTECT_NORDWR(0x0b608, 0x0007),
532 	A6XX_PROTECT_NORDWR(0x0be02, 0x0001),
533 	A6XX_PROTECT_NORDWR(0x0be20, 0x17df),
534 	A6XX_PROTECT_NORDWR(0x0f000, 0x0bff),
535 	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
536 	A6XX_PROTECT_NORDWR(0x18400, 0x1fff),
537 	A6XX_PROTECT_NORDWR(0x1a800, 0x1fff),
538 	A6XX_PROTECT_NORDWR(0x1f400, 0x0443),
539 	A6XX_PROTECT_RDONLY(0x1f844, 0x007b),
540 	A6XX_PROTECT_NORDWR(0x1f887, 0x001b),
541 	A6XX_PROTECT_NORDWR(0x1f8c0, 0x0000), /* note: infinite range */
542 };
543 
544 static void a6xx_set_cp_protect(struct msm_gpu *gpu)
545 {
546 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
547 	const u32 *regs = a6xx_protect;
548 	unsigned i, count = ARRAY_SIZE(a6xx_protect), count_max = 32;
549 
550 	BUILD_BUG_ON(ARRAY_SIZE(a6xx_protect) > 32);
551 	BUILD_BUG_ON(ARRAY_SIZE(a650_protect) > 48);
552 
553 	if (adreno_is_a650(adreno_gpu)) {
554 		regs = a650_protect;
555 		count = ARRAY_SIZE(a650_protect);
556 		count_max = 48;
557 	}
558 
559 	/*
560 	 * Enable access protection to privileged registers, fault on an access
561 	 * protect violation and select the last span to protect from the start
562 	 * address all the way to the end of the register address space
563 	 */
564 	gpu_write(gpu, REG_A6XX_CP_PROTECT_CNTL, BIT(0) | BIT(1) | BIT(3));
565 
566 	for (i = 0; i < count - 1; i++)
567 		gpu_write(gpu, REG_A6XX_CP_PROTECT(i), regs[i]);
568 	/* last CP_PROTECT to have "infinite" length on the last entry */
569 	gpu_write(gpu, REG_A6XX_CP_PROTECT(count_max - 1), regs[i]);
570 }
571 
572 static void a6xx_set_ubwc_config(struct msm_gpu *gpu)
573 {
574 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
575 	u32 lower_bit = 2;
576 	u32 amsbc = 0;
577 	u32 rgb565_predicator = 0;
578 	u32 uavflagprd_inv = 0;
579 
580 	/* a618 is using the hw default values */
581 	if (adreno_is_a618(adreno_gpu))
582 		return;
583 
584 	if (adreno_is_a640(adreno_gpu))
585 		amsbc = 1;
586 
587 	if (adreno_is_a650(adreno_gpu)) {
588 		/* TODO: get ddr type from bootloader and use 2 for LPDDR4 */
589 		lower_bit = 3;
590 		amsbc = 1;
591 		rgb565_predicator = 1;
592 		uavflagprd_inv = 2;
593 	}
594 
595 	gpu_write(gpu, REG_A6XX_RB_NC_MODE_CNTL,
596 		rgb565_predicator << 11 | amsbc << 4 | lower_bit << 1);
597 	gpu_write(gpu, REG_A6XX_TPL1_NC_MODE_CNTL, lower_bit << 1);
598 	gpu_write(gpu, REG_A6XX_SP_NC_MODE_CNTL,
599 		uavflagprd_inv << 4 | lower_bit << 1);
600 	gpu_write(gpu, REG_A6XX_UCHE_MODE_CNTL, lower_bit << 21);
601 }
602 
603 static int a6xx_cp_init(struct msm_gpu *gpu)
604 {
605 	struct msm_ringbuffer *ring = gpu->rb[0];
606 
607 	OUT_PKT7(ring, CP_ME_INIT, 8);
608 
609 	OUT_RING(ring, 0x0000002f);
610 
611 	/* Enable multiple hardware contexts */
612 	OUT_RING(ring, 0x00000003);
613 
614 	/* Enable error detection */
615 	OUT_RING(ring, 0x20000000);
616 
617 	/* Don't enable header dump */
618 	OUT_RING(ring, 0x00000000);
619 	OUT_RING(ring, 0x00000000);
620 
621 	/* No workarounds enabled */
622 	OUT_RING(ring, 0x00000000);
623 
624 	/* Pad rest of the cmds with 0's */
625 	OUT_RING(ring, 0x00000000);
626 	OUT_RING(ring, 0x00000000);
627 
628 	a6xx_flush(gpu, ring);
629 	return a6xx_idle(gpu, ring) ? 0 : -EINVAL;
630 }
631 
632 /*
633  * Check that the microcode version is new enough to include several key
634  * security fixes. Return true if the ucode is safe.
635  */
636 static bool a6xx_ucode_check_version(struct a6xx_gpu *a6xx_gpu,
637 		struct drm_gem_object *obj)
638 {
639 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
640 	struct msm_gpu *gpu = &adreno_gpu->base;
641 	u32 *buf = msm_gem_get_vaddr(obj);
642 	bool ret = false;
643 
644 	if (IS_ERR(buf))
645 		return false;
646 
647 	/*
648 	 * Targets up to a640 (a618, a630 and a640) need to check for a
649 	 * microcode version that is patched to support the whereami opcode or
650 	 * one that is new enough to include it by default.
651 	 */
652 	if (adreno_is_a618(adreno_gpu) || adreno_is_a630(adreno_gpu) ||
653 		adreno_is_a640(adreno_gpu)) {
654 		/*
655 		 * If the lowest nibble is 0xa that is an indication that this
656 		 * microcode has been patched. The actual version is in dword
657 		 * [3] but we only care about the patchlevel which is the lowest
658 		 * nibble of dword [3]
659 		 *
660 		 * Otherwise check that the firmware is greater than or equal
661 		 * to 1.90 which was the first version that had this fix built
662 		 * in
663 		 */
664 		if ((((buf[0] & 0xf) == 0xa) && (buf[2] & 0xf) >= 1) ||
665 			(buf[0] & 0xfff) >= 0x190) {
666 			a6xx_gpu->has_whereami = true;
667 			ret = true;
668 			goto out;
669 		}
670 
671 		DRM_DEV_ERROR(&gpu->pdev->dev,
672 			"a630 SQE ucode is too old. Have version %x need at least %x\n",
673 			buf[0] & 0xfff, 0x190);
674 	}  else {
675 		/*
676 		 * a650 tier targets don't need whereami but still need to be
677 		 * equal to or newer than 0.95 for other security fixes
678 		 */
679 		if (adreno_is_a650(adreno_gpu)) {
680 			if ((buf[0] & 0xfff) >= 0x095) {
681 				ret = true;
682 				goto out;
683 			}
684 
685 			DRM_DEV_ERROR(&gpu->pdev->dev,
686 				"a650 SQE ucode is too old. Have version %x need at least %x\n",
687 				buf[0] & 0xfff, 0x095);
688 		}
689 
690 		/*
691 		 * When a660 is added those targets should return true here
692 		 * since those have all the critical security fixes built in
693 		 * from the start
694 		 */
695 	}
696 out:
697 	msm_gem_put_vaddr(obj);
698 	return ret;
699 }
700 
701 static int a6xx_ucode_init(struct msm_gpu *gpu)
702 {
703 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
704 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
705 
706 	if (!a6xx_gpu->sqe_bo) {
707 		a6xx_gpu->sqe_bo = adreno_fw_create_bo(gpu,
708 			adreno_gpu->fw[ADRENO_FW_SQE], &a6xx_gpu->sqe_iova);
709 
710 		if (IS_ERR(a6xx_gpu->sqe_bo)) {
711 			int ret = PTR_ERR(a6xx_gpu->sqe_bo);
712 
713 			a6xx_gpu->sqe_bo = NULL;
714 			DRM_DEV_ERROR(&gpu->pdev->dev,
715 				"Could not allocate SQE ucode: %d\n", ret);
716 
717 			return ret;
718 		}
719 
720 		msm_gem_object_set_name(a6xx_gpu->sqe_bo, "sqefw");
721 		if (!a6xx_ucode_check_version(a6xx_gpu, a6xx_gpu->sqe_bo)) {
722 			msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
723 			drm_gem_object_put(a6xx_gpu->sqe_bo);
724 
725 			a6xx_gpu->sqe_bo = NULL;
726 			return -EPERM;
727 		}
728 	}
729 
730 	gpu_write64(gpu, REG_A6XX_CP_SQE_INSTR_BASE_LO,
731 		REG_A6XX_CP_SQE_INSTR_BASE_HI, a6xx_gpu->sqe_iova);
732 
733 	return 0;
734 }
735 
736 static int a6xx_zap_shader_init(struct msm_gpu *gpu)
737 {
738 	static bool loaded;
739 	int ret;
740 
741 	if (loaded)
742 		return 0;
743 
744 	ret = adreno_zap_shader_load(gpu, GPU_PAS_ID);
745 
746 	loaded = !ret;
747 	return ret;
748 }
749 
750 #define A6XX_INT_MASK (A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR | \
751 	  A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW | \
752 	  A6XX_RBBM_INT_0_MASK_CP_HW_ERROR | \
753 	  A6XX_RBBM_INT_0_MASK_CP_IB2 | \
754 	  A6XX_RBBM_INT_0_MASK_CP_IB1 | \
755 	  A6XX_RBBM_INT_0_MASK_CP_RB | \
756 	  A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \
757 	  A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW | \
758 	  A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT | \
759 	  A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \
760 	  A6XX_RBBM_INT_0_MASK_UCHE_TRAP_INTR)
761 
762 static int a6xx_hw_init(struct msm_gpu *gpu)
763 {
764 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
765 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
766 	int ret;
767 
768 	/* Make sure the GMU keeps the GPU on while we set it up */
769 	a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
770 
771 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_CNTL, 0);
772 
773 	/*
774 	 * Disable the trusted memory range - we don't actually supported secure
775 	 * memory rendering at this point in time and we don't want to block off
776 	 * part of the virtual memory space.
777 	 */
778 	gpu_write64(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_BASE_LO,
779 		REG_A6XX_RBBM_SECVID_TSB_TRUSTED_BASE_HI, 0x00000000);
780 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_SIZE, 0x00000000);
781 
782 	/* Turn on 64 bit addressing for all blocks */
783 	gpu_write(gpu, REG_A6XX_CP_ADDR_MODE_CNTL, 0x1);
784 	gpu_write(gpu, REG_A6XX_VSC_ADDR_MODE_CNTL, 0x1);
785 	gpu_write(gpu, REG_A6XX_GRAS_ADDR_MODE_CNTL, 0x1);
786 	gpu_write(gpu, REG_A6XX_RB_ADDR_MODE_CNTL, 0x1);
787 	gpu_write(gpu, REG_A6XX_PC_ADDR_MODE_CNTL, 0x1);
788 	gpu_write(gpu, REG_A6XX_HLSQ_ADDR_MODE_CNTL, 0x1);
789 	gpu_write(gpu, REG_A6XX_VFD_ADDR_MODE_CNTL, 0x1);
790 	gpu_write(gpu, REG_A6XX_VPC_ADDR_MODE_CNTL, 0x1);
791 	gpu_write(gpu, REG_A6XX_UCHE_ADDR_MODE_CNTL, 0x1);
792 	gpu_write(gpu, REG_A6XX_SP_ADDR_MODE_CNTL, 0x1);
793 	gpu_write(gpu, REG_A6XX_TPL1_ADDR_MODE_CNTL, 0x1);
794 	gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_ADDR_MODE_CNTL, 0x1);
795 
796 	/* enable hardware clockgating */
797 	a6xx_set_hwcg(gpu, true);
798 
799 	/* VBIF/GBIF start*/
800 	if (adreno_is_a640(adreno_gpu) || adreno_is_a650(adreno_gpu)) {
801 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE0, 0x00071620);
802 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE1, 0x00071620);
803 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE2, 0x00071620);
804 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
805 		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
806 		gpu_write(gpu, REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL, 0x3);
807 	} else {
808 		gpu_write(gpu, REG_A6XX_RBBM_VBIF_CLIENT_QOS_CNTL, 0x3);
809 	}
810 
811 	if (adreno_is_a630(adreno_gpu))
812 		gpu_write(gpu, REG_A6XX_VBIF_GATE_OFF_WRREQ_EN, 0x00000009);
813 
814 	/* Make all blocks contribute to the GPU BUSY perf counter */
815 	gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_GPU_BUSY_MASKED, 0xffffffff);
816 
817 	/* Disable L2 bypass in the UCHE */
818 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX_LO, 0xffffffc0);
819 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX_HI, 0x0001ffff);
820 	gpu_write(gpu, REG_A6XX_UCHE_TRAP_BASE_LO, 0xfffff000);
821 	gpu_write(gpu, REG_A6XX_UCHE_TRAP_BASE_HI, 0x0001ffff);
822 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE_LO, 0xfffff000);
823 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE_HI, 0x0001ffff);
824 
825 	if (!adreno_is_a650(adreno_gpu)) {
826 		/* Set the GMEM VA range [0x100000:0x100000 + gpu->gmem - 1] */
827 		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MIN_LO,
828 			REG_A6XX_UCHE_GMEM_RANGE_MIN_HI, 0x00100000);
829 
830 		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MAX_LO,
831 			REG_A6XX_UCHE_GMEM_RANGE_MAX_HI,
832 			0x00100000 + adreno_gpu->gmem - 1);
833 	}
834 
835 	gpu_write(gpu, REG_A6XX_UCHE_FILTER_CNTL, 0x804);
836 	gpu_write(gpu, REG_A6XX_UCHE_CACHE_WAYS, 0x4);
837 
838 	if (adreno_is_a640(adreno_gpu) || adreno_is_a650(adreno_gpu))
839 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x02000140);
840 	else
841 		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x010000c0);
842 	gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x8040362c);
843 
844 	/* Setting the mem pool size */
845 	gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 128);
846 
847 	/* Setting the primFifo thresholds default values */
848 	if (adreno_is_a650(adreno_gpu))
849 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00300000);
850 	else if (adreno_is_a640(adreno_gpu))
851 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00200000);
852 	else
853 		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, (0x300 << 11));
854 
855 	/* Set the AHB default slave response to "ERROR" */
856 	gpu_write(gpu, REG_A6XX_CP_AHB_CNTL, 0x1);
857 
858 	/* Turn on performance counters */
859 	gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_CNTL, 0x1);
860 
861 	/* Select CP0 to always count cycles */
862 	gpu_write(gpu, REG_A6XX_CP_PERFCTR_CP_SEL_0, PERF_CP_ALWAYS_COUNT);
863 
864 	a6xx_set_ubwc_config(gpu);
865 
866 	/* Enable fault detection */
867 	gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL,
868 		(1 << 30) | 0x1fffff);
869 
870 	gpu_write(gpu, REG_A6XX_UCHE_CLIENT_PF, 1);
871 
872 	/* Set weights for bicubic filtering */
873 	if (adreno_is_a650(adreno_gpu)) {
874 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_0, 0);
875 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_1,
876 			0x3fe05ff4);
877 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_2,
878 			0x3fa0ebee);
879 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_3,
880 			0x3f5193ed);
881 		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_4,
882 			0x3f0243f0);
883 	}
884 
885 	/* Protect registers from the CP */
886 	a6xx_set_cp_protect(gpu);
887 
888 	/* Enable expanded apriv for targets that support it */
889 	if (gpu->hw_apriv) {
890 		gpu_write(gpu, REG_A6XX_CP_APRIV_CNTL,
891 			(1 << 6) | (1 << 5) | (1 << 3) | (1 << 2) | (1 << 1));
892 	}
893 
894 	/* Enable interrupts */
895 	gpu_write(gpu, REG_A6XX_RBBM_INT_0_MASK, A6XX_INT_MASK);
896 
897 	ret = adreno_hw_init(gpu);
898 	if (ret)
899 		goto out;
900 
901 	ret = a6xx_ucode_init(gpu);
902 	if (ret)
903 		goto out;
904 
905 	/* Set the ringbuffer address */
906 	gpu_write64(gpu, REG_A6XX_CP_RB_BASE, REG_A6XX_CP_RB_BASE_HI,
907 		gpu->rb[0]->iova);
908 
909 	/* Targets that support extended APRIV can use the RPTR shadow from
910 	 * hardware but all the other ones need to disable the feature. Targets
911 	 * that support the WHERE_AM_I opcode can use that instead
912 	 */
913 	if (adreno_gpu->base.hw_apriv)
914 		gpu_write(gpu, REG_A6XX_CP_RB_CNTL, MSM_GPU_RB_CNTL_DEFAULT);
915 	else
916 		gpu_write(gpu, REG_A6XX_CP_RB_CNTL,
917 			MSM_GPU_RB_CNTL_DEFAULT | AXXX_CP_RB_CNTL_NO_UPDATE);
918 
919 	/*
920 	 * Expanded APRIV and targets that support WHERE_AM_I both need a
921 	 * privileged buffer to store the RPTR shadow
922 	 */
923 
924 	if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami) {
925 		if (!a6xx_gpu->shadow_bo) {
926 			a6xx_gpu->shadow = msm_gem_kernel_new_locked(gpu->dev,
927 				sizeof(u32) * gpu->nr_rings,
928 				MSM_BO_UNCACHED | MSM_BO_MAP_PRIV,
929 				gpu->aspace, &a6xx_gpu->shadow_bo,
930 				&a6xx_gpu->shadow_iova);
931 
932 			if (IS_ERR(a6xx_gpu->shadow))
933 				return PTR_ERR(a6xx_gpu->shadow);
934 		}
935 
936 		gpu_write64(gpu, REG_A6XX_CP_RB_RPTR_ADDR_LO,
937 			REG_A6XX_CP_RB_RPTR_ADDR_HI,
938 			shadowptr(a6xx_gpu, gpu->rb[0]));
939 	}
940 
941 	/* Always come up on rb 0 */
942 	a6xx_gpu->cur_ring = gpu->rb[0];
943 
944 	a6xx_gpu->cur_ctx = NULL;
945 
946 	/* Enable the SQE_to start the CP engine */
947 	gpu_write(gpu, REG_A6XX_CP_SQE_CNTL, 1);
948 
949 	ret = a6xx_cp_init(gpu);
950 	if (ret)
951 		goto out;
952 
953 	/*
954 	 * Try to load a zap shader into the secure world. If successful
955 	 * we can use the CP to switch out of secure mode. If not then we
956 	 * have no resource but to try to switch ourselves out manually. If we
957 	 * guessed wrong then access to the RBBM_SECVID_TRUST_CNTL register will
958 	 * be blocked and a permissions violation will soon follow.
959 	 */
960 	ret = a6xx_zap_shader_init(gpu);
961 	if (!ret) {
962 		OUT_PKT7(gpu->rb[0], CP_SET_SECURE_MODE, 1);
963 		OUT_RING(gpu->rb[0], 0x00000000);
964 
965 		a6xx_flush(gpu, gpu->rb[0]);
966 		if (!a6xx_idle(gpu, gpu->rb[0]))
967 			return -EINVAL;
968 	} else if (ret == -ENODEV) {
969 		/*
970 		 * This device does not use zap shader (but print a warning
971 		 * just in case someone got their dt wrong.. hopefully they
972 		 * have a debug UART to realize the error of their ways...
973 		 * if you mess this up you are about to crash horribly)
974 		 */
975 		dev_warn_once(gpu->dev->dev,
976 			"Zap shader not enabled - using SECVID_TRUST_CNTL instead\n");
977 		gpu_write(gpu, REG_A6XX_RBBM_SECVID_TRUST_CNTL, 0x0);
978 		ret = 0;
979 	} else {
980 		return ret;
981 	}
982 
983 out:
984 	/*
985 	 * Tell the GMU that we are done touching the GPU and it can start power
986 	 * management
987 	 */
988 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
989 
990 	if (a6xx_gpu->gmu.legacy) {
991 		/* Take the GMU out of its special boot mode */
992 		a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
993 	}
994 
995 	return ret;
996 }
997 
998 static void a6xx_dump(struct msm_gpu *gpu)
999 {
1000 	DRM_DEV_INFO(&gpu->pdev->dev, "status:   %08x\n",
1001 			gpu_read(gpu, REG_A6XX_RBBM_STATUS));
1002 	adreno_dump(gpu);
1003 }
1004 
1005 #define VBIF_RESET_ACK_TIMEOUT	100
1006 #define VBIF_RESET_ACK_MASK	0x00f0
1007 
1008 static void a6xx_recover(struct msm_gpu *gpu)
1009 {
1010 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1011 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1012 	int i;
1013 
1014 	adreno_dump_info(gpu);
1015 
1016 	for (i = 0; i < 8; i++)
1017 		DRM_DEV_INFO(&gpu->pdev->dev, "CP_SCRATCH_REG%d: %u\n", i,
1018 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(i)));
1019 
1020 	if (hang_debug)
1021 		a6xx_dump(gpu);
1022 
1023 	/*
1024 	 * Turn off keep alive that might have been enabled by the hang
1025 	 * interrupt
1026 	 */
1027 	gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 0);
1028 
1029 	gpu->funcs->pm_suspend(gpu);
1030 	gpu->funcs->pm_resume(gpu);
1031 
1032 	msm_gpu_hw_init(gpu);
1033 }
1034 
1035 static int a6xx_fault_handler(void *arg, unsigned long iova, int flags)
1036 {
1037 	struct msm_gpu *gpu = arg;
1038 
1039 	pr_warn_ratelimited("*** gpu fault: iova=%08lx, flags=%d (%u,%u,%u,%u)\n",
1040 			iova, flags,
1041 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(4)),
1042 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(5)),
1043 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(6)),
1044 			gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(7)));
1045 
1046 	return -EFAULT;
1047 }
1048 
1049 static void a6xx_cp_hw_err_irq(struct msm_gpu *gpu)
1050 {
1051 	u32 status = gpu_read(gpu, REG_A6XX_CP_INTERRUPT_STATUS);
1052 
1053 	if (status & A6XX_CP_INT_CP_OPCODE_ERROR) {
1054 		u32 val;
1055 
1056 		gpu_write(gpu, REG_A6XX_CP_SQE_STAT_ADDR, 1);
1057 		val = gpu_read(gpu, REG_A6XX_CP_SQE_STAT_DATA);
1058 		dev_err_ratelimited(&gpu->pdev->dev,
1059 			"CP | opcode error | possible opcode=0x%8.8X\n",
1060 			val);
1061 	}
1062 
1063 	if (status & A6XX_CP_INT_CP_UCODE_ERROR)
1064 		dev_err_ratelimited(&gpu->pdev->dev,
1065 			"CP ucode error interrupt\n");
1066 
1067 	if (status & A6XX_CP_INT_CP_HW_FAULT_ERROR)
1068 		dev_err_ratelimited(&gpu->pdev->dev, "CP | HW fault | status=0x%8.8X\n",
1069 			gpu_read(gpu, REG_A6XX_CP_HW_FAULT));
1070 
1071 	if (status & A6XX_CP_INT_CP_REGISTER_PROTECTION_ERROR) {
1072 		u32 val = gpu_read(gpu, REG_A6XX_CP_PROTECT_STATUS);
1073 
1074 		dev_err_ratelimited(&gpu->pdev->dev,
1075 			"CP | protected mode error | %s | addr=0x%8.8X | status=0x%8.8X\n",
1076 			val & (1 << 20) ? "READ" : "WRITE",
1077 			(val & 0x3ffff), val);
1078 	}
1079 
1080 	if (status & A6XX_CP_INT_CP_AHB_ERROR)
1081 		dev_err_ratelimited(&gpu->pdev->dev, "CP AHB error interrupt\n");
1082 
1083 	if (status & A6XX_CP_INT_CP_VSD_PARITY_ERROR)
1084 		dev_err_ratelimited(&gpu->pdev->dev, "CP VSD decoder parity error\n");
1085 
1086 	if (status & A6XX_CP_INT_CP_ILLEGAL_INSTR_ERROR)
1087 		dev_err_ratelimited(&gpu->pdev->dev, "CP illegal instruction error\n");
1088 
1089 }
1090 
1091 static void a6xx_fault_detect_irq(struct msm_gpu *gpu)
1092 {
1093 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1094 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1095 	struct msm_ringbuffer *ring = gpu->funcs->active_ring(gpu);
1096 
1097 	/*
1098 	 * Force the GPU to stay on until after we finish
1099 	 * collecting information
1100 	 */
1101 	gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 1);
1102 
1103 	DRM_DEV_ERROR(&gpu->pdev->dev,
1104 		"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",
1105 		ring ? ring->id : -1, ring ? ring->seqno : 0,
1106 		gpu_read(gpu, REG_A6XX_RBBM_STATUS),
1107 		gpu_read(gpu, REG_A6XX_CP_RB_RPTR),
1108 		gpu_read(gpu, REG_A6XX_CP_RB_WPTR),
1109 		gpu_read64(gpu, REG_A6XX_CP_IB1_BASE, REG_A6XX_CP_IB1_BASE_HI),
1110 		gpu_read(gpu, REG_A6XX_CP_IB1_REM_SIZE),
1111 		gpu_read64(gpu, REG_A6XX_CP_IB2_BASE, REG_A6XX_CP_IB2_BASE_HI),
1112 		gpu_read(gpu, REG_A6XX_CP_IB2_REM_SIZE));
1113 
1114 	/* Turn off the hangcheck timer to keep it from bothering us */
1115 	del_timer(&gpu->hangcheck_timer);
1116 
1117 	kthread_queue_work(gpu->worker, &gpu->recover_work);
1118 }
1119 
1120 static irqreturn_t a6xx_irq(struct msm_gpu *gpu)
1121 {
1122 	u32 status = gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS);
1123 
1124 	gpu_write(gpu, REG_A6XX_RBBM_INT_CLEAR_CMD, status);
1125 
1126 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT)
1127 		a6xx_fault_detect_irq(gpu);
1128 
1129 	if (status & A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR)
1130 		dev_err_ratelimited(&gpu->pdev->dev, "CP | AHB bus error\n");
1131 
1132 	if (status & A6XX_RBBM_INT_0_MASK_CP_HW_ERROR)
1133 		a6xx_cp_hw_err_irq(gpu);
1134 
1135 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW)
1136 		dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB ASYNC overflow\n");
1137 
1138 	if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW)
1139 		dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB bus overflow\n");
1140 
1141 	if (status & A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS)
1142 		dev_err_ratelimited(&gpu->pdev->dev, "UCHE | Out of bounds access\n");
1143 
1144 	if (status & A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS)
1145 		msm_gpu_retire(gpu);
1146 
1147 	return IRQ_HANDLED;
1148 }
1149 
1150 static void a6xx_llc_rmw(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 mask, u32 or)
1151 {
1152 	return msm_rmw(a6xx_gpu->llc_mmio + (reg << 2), mask, or);
1153 }
1154 
1155 static void a6xx_llc_write(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 value)
1156 {
1157 	return msm_writel(value, a6xx_gpu->llc_mmio + (reg << 2));
1158 }
1159 
1160 static void a6xx_llc_deactivate(struct a6xx_gpu *a6xx_gpu)
1161 {
1162 	llcc_slice_deactivate(a6xx_gpu->llc_slice);
1163 	llcc_slice_deactivate(a6xx_gpu->htw_llc_slice);
1164 }
1165 
1166 static void a6xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
1167 {
1168 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
1169 	struct msm_gpu *gpu = &adreno_gpu->base;
1170 	u32 cntl1_regval = 0;
1171 
1172 	if (IS_ERR(a6xx_gpu->llc_mmio))
1173 		return;
1174 
1175 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
1176 		u32 gpu_scid = llcc_get_slice_id(a6xx_gpu->llc_slice);
1177 
1178 		gpu_scid &= 0x1f;
1179 		cntl1_regval = (gpu_scid << 0) | (gpu_scid << 5) | (gpu_scid << 10) |
1180 			       (gpu_scid << 15) | (gpu_scid << 20);
1181 	}
1182 
1183 	/*
1184 	 * For targets with a MMU500, activate the slice but don't program the
1185 	 * register.  The XBL will take care of that.
1186 	 */
1187 	if (!llcc_slice_activate(a6xx_gpu->htw_llc_slice)) {
1188 		if (!a6xx_gpu->have_mmu500) {
1189 			u32 gpuhtw_scid = llcc_get_slice_id(a6xx_gpu->htw_llc_slice);
1190 
1191 			gpuhtw_scid &= 0x1f;
1192 			cntl1_regval |= FIELD_PREP(GENMASK(29, 25), gpuhtw_scid);
1193 		}
1194 	}
1195 
1196 	if (cntl1_regval) {
1197 		/*
1198 		 * Program the slice IDs for the various GPU blocks and GPU MMU
1199 		 * pagetables
1200 		 */
1201 		if (a6xx_gpu->have_mmu500)
1202 			gpu_rmw(gpu, REG_A6XX_GBIF_SCACHE_CNTL1, GENMASK(24, 0),
1203 				cntl1_regval);
1204 		else {
1205 			a6xx_llc_write(a6xx_gpu,
1206 				REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_1, cntl1_regval);
1207 
1208 			/*
1209 			 * Program cacheability overrides to not allocate cache
1210 			 * lines on a write miss
1211 			 */
1212 			a6xx_llc_rmw(a6xx_gpu,
1213 				REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_0, 0xF, 0x03);
1214 		}
1215 	}
1216 }
1217 
1218 static void a6xx_llc_slices_destroy(struct a6xx_gpu *a6xx_gpu)
1219 {
1220 	llcc_slice_putd(a6xx_gpu->llc_slice);
1221 	llcc_slice_putd(a6xx_gpu->htw_llc_slice);
1222 }
1223 
1224 static void a6xx_llc_slices_init(struct platform_device *pdev,
1225 		struct a6xx_gpu *a6xx_gpu)
1226 {
1227 	struct device_node *phandle;
1228 
1229 	/*
1230 	 * There is a different programming path for targets with an mmu500
1231 	 * attached, so detect if that is the case
1232 	 */
1233 	phandle = of_parse_phandle(pdev->dev.of_node, "iommus", 0);
1234 	a6xx_gpu->have_mmu500 = (phandle &&
1235 		of_device_is_compatible(phandle, "arm,mmu-500"));
1236 	of_node_put(phandle);
1237 
1238 	if (a6xx_gpu->have_mmu500)
1239 		a6xx_gpu->llc_mmio = NULL;
1240 	else
1241 		a6xx_gpu->llc_mmio = msm_ioremap(pdev, "cx_mem", "gpu_cx");
1242 
1243 	a6xx_gpu->llc_slice = llcc_slice_getd(LLCC_GPU);
1244 	a6xx_gpu->htw_llc_slice = llcc_slice_getd(LLCC_GPUHTW);
1245 
1246 	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
1247 		a6xx_gpu->llc_mmio = ERR_PTR(-EINVAL);
1248 }
1249 
1250 static int a6xx_pm_resume(struct msm_gpu *gpu)
1251 {
1252 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1253 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1254 	int ret;
1255 
1256 	gpu->needs_hw_init = true;
1257 
1258 	trace_msm_gpu_resume(0);
1259 
1260 	ret = a6xx_gmu_resume(a6xx_gpu);
1261 	if (ret)
1262 		return ret;
1263 
1264 	msm_gpu_resume_devfreq(gpu);
1265 
1266 	a6xx_llc_activate(a6xx_gpu);
1267 
1268 	return 0;
1269 }
1270 
1271 static int a6xx_pm_suspend(struct msm_gpu *gpu)
1272 {
1273 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1274 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1275 	int i, ret;
1276 
1277 	trace_msm_gpu_suspend(0);
1278 
1279 	a6xx_llc_deactivate(a6xx_gpu);
1280 
1281 	devfreq_suspend_device(gpu->devfreq.devfreq);
1282 
1283 	ret = a6xx_gmu_stop(a6xx_gpu);
1284 	if (ret)
1285 		return ret;
1286 
1287 	if (a6xx_gpu->shadow_bo)
1288 		for (i = 0; i < gpu->nr_rings; i++)
1289 			a6xx_gpu->shadow[i] = 0;
1290 
1291 	return 0;
1292 }
1293 
1294 static int a6xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
1295 {
1296 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1297 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1298 	static DEFINE_MUTEX(perfcounter_oob);
1299 
1300 	mutex_lock(&perfcounter_oob);
1301 
1302 	/* Force the GPU power on so we can read this register */
1303 	a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
1304 
1305 	*value = gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER_LO,
1306 		REG_A6XX_CP_ALWAYS_ON_COUNTER_HI);
1307 
1308 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
1309 	mutex_unlock(&perfcounter_oob);
1310 	return 0;
1311 }
1312 
1313 static struct msm_ringbuffer *a6xx_active_ring(struct msm_gpu *gpu)
1314 {
1315 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1316 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1317 
1318 	return a6xx_gpu->cur_ring;
1319 }
1320 
1321 static void a6xx_destroy(struct msm_gpu *gpu)
1322 {
1323 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1324 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1325 
1326 	if (a6xx_gpu->sqe_bo) {
1327 		msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
1328 		drm_gem_object_put(a6xx_gpu->sqe_bo);
1329 	}
1330 
1331 	if (a6xx_gpu->shadow_bo) {
1332 		msm_gem_unpin_iova(a6xx_gpu->shadow_bo, gpu->aspace);
1333 		drm_gem_object_put(a6xx_gpu->shadow_bo);
1334 	}
1335 
1336 	a6xx_llc_slices_destroy(a6xx_gpu);
1337 
1338 	a6xx_gmu_remove(a6xx_gpu);
1339 
1340 	adreno_gpu_cleanup(adreno_gpu);
1341 
1342 	if (a6xx_gpu->opp_table)
1343 		dev_pm_opp_put_supported_hw(a6xx_gpu->opp_table);
1344 
1345 	kfree(a6xx_gpu);
1346 }
1347 
1348 static unsigned long a6xx_gpu_busy(struct msm_gpu *gpu)
1349 {
1350 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1351 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1352 	u64 busy_cycles, busy_time;
1353 
1354 
1355 	/* Only read the gpu busy if the hardware is already active */
1356 	if (pm_runtime_get_if_in_use(a6xx_gpu->gmu.dev) == 0)
1357 		return 0;
1358 
1359 	busy_cycles = gmu_read64(&a6xx_gpu->gmu,
1360 			REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_L,
1361 			REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_H);
1362 
1363 	busy_time = (busy_cycles - gpu->devfreq.busy_cycles) * 10;
1364 	do_div(busy_time, 192);
1365 
1366 	gpu->devfreq.busy_cycles = busy_cycles;
1367 
1368 	pm_runtime_put(a6xx_gpu->gmu.dev);
1369 
1370 	if (WARN_ON(busy_time > ~0LU))
1371 		return ~0LU;
1372 
1373 	return (unsigned long)busy_time;
1374 }
1375 
1376 static struct msm_gem_address_space *
1377 a6xx_create_address_space(struct msm_gpu *gpu, struct platform_device *pdev)
1378 {
1379 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1380 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1381 	struct iommu_domain *iommu;
1382 	struct msm_mmu *mmu;
1383 	struct msm_gem_address_space *aspace;
1384 	u64 start, size;
1385 
1386 	iommu = iommu_domain_alloc(&platform_bus_type);
1387 	if (!iommu)
1388 		return NULL;
1389 
1390 	/*
1391 	 * This allows GPU to set the bus attributes required to use system
1392 	 * cache on behalf of the iommu page table walker.
1393 	 */
1394 	if (!IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
1395 		adreno_set_llc_attributes(iommu);
1396 
1397 	mmu = msm_iommu_new(&pdev->dev, iommu);
1398 	if (IS_ERR(mmu)) {
1399 		iommu_domain_free(iommu);
1400 		return ERR_CAST(mmu);
1401 	}
1402 
1403 	/*
1404 	 * Use the aperture start or SZ_16M, whichever is greater. This will
1405 	 * ensure that we align with the allocated pagetable range while still
1406 	 * allowing room in the lower 32 bits for GMEM and whatnot
1407 	 */
1408 	start = max_t(u64, SZ_16M, iommu->geometry.aperture_start);
1409 	size = iommu->geometry.aperture_end - start + 1;
1410 
1411 	aspace = msm_gem_address_space_create(mmu, "gpu",
1412 		start & GENMASK_ULL(48, 0), size);
1413 
1414 	if (IS_ERR(aspace) && !IS_ERR(mmu))
1415 		mmu->funcs->destroy(mmu);
1416 
1417 	return aspace;
1418 }
1419 
1420 static struct msm_gem_address_space *
1421 a6xx_create_private_address_space(struct msm_gpu *gpu)
1422 {
1423 	struct msm_mmu *mmu;
1424 
1425 	mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
1426 
1427 	if (IS_ERR(mmu))
1428 		return ERR_CAST(mmu);
1429 
1430 	return msm_gem_address_space_create(mmu,
1431 		"gpu", 0x100000000ULL, 0x1ffffffffULL);
1432 }
1433 
1434 static uint32_t a6xx_get_rptr(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
1435 {
1436 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1437 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1438 
1439 	if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami)
1440 		return a6xx_gpu->shadow[ring->id];
1441 
1442 	return ring->memptrs->rptr = gpu_read(gpu, REG_A6XX_CP_RB_RPTR);
1443 }
1444 
1445 static u32 a618_get_speed_bin(u32 fuse)
1446 {
1447 	if (fuse == 0)
1448 		return 0;
1449 	else if (fuse == 169)
1450 		return 1;
1451 	else if (fuse == 174)
1452 		return 2;
1453 
1454 	return UINT_MAX;
1455 }
1456 
1457 static u32 fuse_to_supp_hw(struct device *dev, u32 revn, u32 fuse)
1458 {
1459 	u32 val = UINT_MAX;
1460 
1461 	if (revn == 618)
1462 		val = a618_get_speed_bin(fuse);
1463 
1464 	if (val == UINT_MAX) {
1465 		DRM_DEV_ERROR(dev,
1466 			"missing support for speed-bin: %u. Some OPPs may not be supported by hardware",
1467 			fuse);
1468 		return UINT_MAX;
1469 	}
1470 
1471 	return (1 << val);
1472 }
1473 
1474 static int a6xx_set_supported_hw(struct device *dev, struct a6xx_gpu *a6xx_gpu,
1475 		u32 revn)
1476 {
1477 	struct opp_table *opp_table;
1478 	u32 supp_hw = UINT_MAX;
1479 	u16 speedbin;
1480 	int ret;
1481 
1482 	ret = nvmem_cell_read_u16(dev, "speed_bin", &speedbin);
1483 	/*
1484 	 * -ENOENT means that the platform doesn't support speedbin which is
1485 	 * fine
1486 	 */
1487 	if (ret == -ENOENT) {
1488 		return 0;
1489 	} else if (ret) {
1490 		DRM_DEV_ERROR(dev,
1491 			      "failed to read speed-bin (%d). Some OPPs may not be supported by hardware",
1492 			      ret);
1493 		goto done;
1494 	}
1495 	speedbin = le16_to_cpu(speedbin);
1496 
1497 	supp_hw = fuse_to_supp_hw(dev, revn, speedbin);
1498 
1499 done:
1500 	opp_table = dev_pm_opp_set_supported_hw(dev, &supp_hw, 1);
1501 	if (IS_ERR(opp_table))
1502 		return PTR_ERR(opp_table);
1503 
1504 	a6xx_gpu->opp_table = opp_table;
1505 	return 0;
1506 }
1507 
1508 static const struct adreno_gpu_funcs funcs = {
1509 	.base = {
1510 		.get_param = adreno_get_param,
1511 		.hw_init = a6xx_hw_init,
1512 		.pm_suspend = a6xx_pm_suspend,
1513 		.pm_resume = a6xx_pm_resume,
1514 		.recover = a6xx_recover,
1515 		.submit = a6xx_submit,
1516 		.active_ring = a6xx_active_ring,
1517 		.irq = a6xx_irq,
1518 		.destroy = a6xx_destroy,
1519 #if defined(CONFIG_DRM_MSM_GPU_STATE)
1520 		.show = a6xx_show,
1521 #endif
1522 		.gpu_busy = a6xx_gpu_busy,
1523 		.gpu_get_freq = a6xx_gmu_get_freq,
1524 		.gpu_set_freq = a6xx_gmu_set_freq,
1525 #if defined(CONFIG_DRM_MSM_GPU_STATE)
1526 		.gpu_state_get = a6xx_gpu_state_get,
1527 		.gpu_state_put = a6xx_gpu_state_put,
1528 #endif
1529 		.create_address_space = a6xx_create_address_space,
1530 		.create_private_address_space = a6xx_create_private_address_space,
1531 		.get_rptr = a6xx_get_rptr,
1532 	},
1533 	.get_timestamp = a6xx_get_timestamp,
1534 };
1535 
1536 struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
1537 {
1538 	struct msm_drm_private *priv = dev->dev_private;
1539 	struct platform_device *pdev = priv->gpu_pdev;
1540 	struct adreno_platform_config *config = pdev->dev.platform_data;
1541 	const struct adreno_info *info;
1542 	struct device_node *node;
1543 	struct a6xx_gpu *a6xx_gpu;
1544 	struct adreno_gpu *adreno_gpu;
1545 	struct msm_gpu *gpu;
1546 	int ret;
1547 
1548 	a6xx_gpu = kzalloc(sizeof(*a6xx_gpu), GFP_KERNEL);
1549 	if (!a6xx_gpu)
1550 		return ERR_PTR(-ENOMEM);
1551 
1552 	adreno_gpu = &a6xx_gpu->base;
1553 	gpu = &adreno_gpu->base;
1554 
1555 	adreno_gpu->registers = NULL;
1556 
1557 	/*
1558 	 * We need to know the platform type before calling into adreno_gpu_init
1559 	 * so that the hw_apriv flag can be correctly set. Snoop into the info
1560 	 * and grab the revision number
1561 	 */
1562 	info = adreno_info(config->rev);
1563 
1564 	if (info && info->revn == 650)
1565 		adreno_gpu->base.hw_apriv = true;
1566 
1567 	a6xx_llc_slices_init(pdev, a6xx_gpu);
1568 
1569 	ret = a6xx_set_supported_hw(&pdev->dev, a6xx_gpu, info->revn);
1570 	if (ret) {
1571 		a6xx_destroy(&(a6xx_gpu->base.base));
1572 		return ERR_PTR(ret);
1573 	}
1574 
1575 	ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1);
1576 	if (ret) {
1577 		a6xx_destroy(&(a6xx_gpu->base.base));
1578 		return ERR_PTR(ret);
1579 	}
1580 
1581 	/* Check if there is a GMU phandle and set it up */
1582 	node = of_parse_phandle(pdev->dev.of_node, "qcom,gmu", 0);
1583 
1584 	/* FIXME: How do we gracefully handle this? */
1585 	BUG_ON(!node);
1586 
1587 	ret = a6xx_gmu_init(a6xx_gpu, node);
1588 	if (ret) {
1589 		a6xx_destroy(&(a6xx_gpu->base.base));
1590 		return ERR_PTR(ret);
1591 	}
1592 
1593 	if (gpu->aspace)
1594 		msm_mmu_set_fault_handler(gpu->aspace->mmu, gpu,
1595 				a6xx_fault_handler);
1596 
1597 	return gpu;
1598 }
1599