drivers/gpu/drm/msm/adreno/a6xx_gpu.c

   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/soc/qcom/llcc-qcom.h>
  14
  15 #define GPU_PAS_ID 13
  16
  17 static inline bool _a6xx_check_idle(struct msm_gpu *gpu)
  18 {
  19         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
  20         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
  21
  22         /* Check that the GMU is idle */
  23         if (!a6xx_gmu_isidle(&a6xx_gpu->gmu))
  24                 return false;
  25
  26         /* Check tha the CX master is idle */
  27         if (gpu_read(gpu, REG_A6XX_RBBM_STATUS) &
  28                         ~A6XX_RBBM_STATUS_CP_AHB_BUSY_CX_MASTER)
  29                 return false;
  30
  31         return !(gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS) &
  32                 A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT);
  33 }
  34
  35 static bool a6xx_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
  36 {
  37         /* wait for CP to drain ringbuffer: */
  38         if (!adreno_idle(gpu, ring))
  39                 return false;
  40
  41         if (spin_until(_a6xx_check_idle(gpu))) {
  42                 DRM_ERROR("%s: %ps: timeout waiting for GPU to idle: status %8.8X irq %8.8X rptr/wptr %d/%d\n",
  43                         gpu->name, __builtin_return_address(0),
  44                         gpu_read(gpu, REG_A6XX_RBBM_STATUS),
  45                         gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS),
  46                         gpu_read(gpu, REG_A6XX_CP_RB_RPTR),
  47                         gpu_read(gpu, REG_A6XX_CP_RB_WPTR));
  48                 return false;
  49         }
  50
  51         return true;
  52 }
  53
  54 static void a6xx_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
  55 {
  56         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
  57         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
  58         uint32_t wptr;
  59         unsigned long flags;
  60
  61         /* Expanded APRIV doesn't need to issue the WHERE_AM_I opcode */
  62         if (a6xx_gpu->has_whereami && !adreno_gpu->base.hw_apriv) {
  63                 struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
  64
  65                 OUT_PKT7(ring, CP_WHERE_AM_I, 2);
  66                 OUT_RING(ring, lower_32_bits(shadowptr(a6xx_gpu, ring)));
  67                 OUT_RING(ring, upper_32_bits(shadowptr(a6xx_gpu, ring)));
  68         }
  69
  70         spin_lock_irqsave(&ring->preempt_lock, flags);
  71
  72         /* Copy the shadow to the actual register */
  73         ring->cur = ring->next;
  74
  75         /* Make sure to wrap wptr if we need to */
  76         wptr = get_wptr(ring);
  77
  78         spin_unlock_irqrestore(&ring->preempt_lock, flags);
  79
  80         /* Make sure everything is posted before making a decision */
  81         mb();
  82
  83         gpu_write(gpu, REG_A6XX_CP_RB_WPTR, wptr);
  84 }
  85
  86 static void get_stats_counter(struct msm_ringbuffer *ring, u32 counter,
  87                 u64 iova)
  88 {
  89         OUT_PKT7(ring, CP_REG_TO_MEM, 3);
  90         OUT_RING(ring, CP_REG_TO_MEM_0_REG(counter) |
  91                 CP_REG_TO_MEM_0_CNT(2) |
  92                 CP_REG_TO_MEM_0_64B);
  93         OUT_RING(ring, lower_32_bits(iova));
  94         OUT_RING(ring, upper_32_bits(iova));
  95 }
  96
  97 static void a6xx_set_pagetable(struct a6xx_gpu *a6xx_gpu,
  98                 struct msm_ringbuffer *ring, struct msm_file_private *ctx)
  99 {
 100         phys_addr_t ttbr;
 101         u32 asid;
 102         u64 memptr = rbmemptr(ring, ttbr0);
 103
 104         if (ctx == a6xx_gpu->cur_ctx)
 105                 return;
 106
 107         if (msm_iommu_pagetable_params(ctx->aspace->mmu, &ttbr, &asid))
 108                 return;
 109
 110         /* Execute the table update */
 111         OUT_PKT7(ring, CP_SMMU_TABLE_UPDATE, 4);
 112         OUT_RING(ring, CP_SMMU_TABLE_UPDATE_0_TTBR0_LO(lower_32_bits(ttbr)));
 113
 114         OUT_RING(ring,
 115                 CP_SMMU_TABLE_UPDATE_1_TTBR0_HI(upper_32_bits(ttbr)) |
 116                 CP_SMMU_TABLE_UPDATE_1_ASID(asid));
 117         OUT_RING(ring, CP_SMMU_TABLE_UPDATE_2_CONTEXTIDR(0));
 118         OUT_RING(ring, CP_SMMU_TABLE_UPDATE_3_CONTEXTBANK(0));
 119
 120         /*
 121          * Write the new TTBR0 to the memstore. This is good for debugging.
 122          */
 123         OUT_PKT7(ring, CP_MEM_WRITE, 4);
 124         OUT_RING(ring, CP_MEM_WRITE_0_ADDR_LO(lower_32_bits(memptr)));
 125         OUT_RING(ring, CP_MEM_WRITE_1_ADDR_HI(upper_32_bits(memptr)));
 126         OUT_RING(ring, lower_32_bits(ttbr));
 127         OUT_RING(ring, (asid << 16) | upper_32_bits(ttbr));
 128
 129         /*
 130          * And finally, trigger a uche flush to be sure there isn't anything
 131          * lingering in that part of the GPU
 132          */
 133
 134         OUT_PKT7(ring, CP_EVENT_WRITE, 1);
 135         OUT_RING(ring, 0x31);
 136
 137         a6xx_gpu->cur_ctx = ctx;
 138 }
 139
 140 static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit)
 141 {
 142         unsigned int index = submit->seqno % MSM_GPU_SUBMIT_STATS_COUNT;
 143         struct msm_drm_private *priv = gpu->dev->dev_private;
 144         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 145         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
 146         struct msm_ringbuffer *ring = submit->ring;
 147         unsigned int i;
 148
 149         a6xx_set_pagetable(a6xx_gpu, ring, submit->queue->ctx);
 150
 151         get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
 152                 rbmemptr_stats(ring, index, cpcycles_start));
 153
 154         /*
 155          * For PM4 the GMU register offsets are calculated from the base of the
 156          * GPU registers so we need to add 0x1a800 to the register value on A630
 157          * to get the right value from PM4.
 158          */
 159         get_stats_counter(ring, REG_A6XX_GMU_ALWAYS_ON_COUNTER_L + 0x1a800,
 160                 rbmemptr_stats(ring, index, alwayson_start));
 161
 162         /* Invalidate CCU depth and color */
 163         OUT_PKT7(ring, CP_EVENT_WRITE, 1);
 164         OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(PC_CCU_INVALIDATE_DEPTH));
 165
 166         OUT_PKT7(ring, CP_EVENT_WRITE, 1);
 167         OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(PC_CCU_INVALIDATE_COLOR));
 168
 169         /* Submit the commands */
 170         for (i = 0; i < submit->nr_cmds; i++) {
 171                 switch (submit->cmd[i].type) {
 172                 case MSM_SUBMIT_CMD_IB_TARGET_BUF:
 173                         break;
 174                 case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
 175                         if (priv->lastctx == submit->queue->ctx)
 176                                 break;
 177                         fallthrough;
 178                 case MSM_SUBMIT_CMD_BUF:
 179                         OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3);
 180                         OUT_RING(ring, lower_32_bits(submit->cmd[i].iova));
 181                         OUT_RING(ring, upper_32_bits(submit->cmd[i].iova));
 182                         OUT_RING(ring, submit->cmd[i].size);
 183                         break;
 184                 }
 185         }
 186
 187         get_stats_counter(ring, REG_A6XX_RBBM_PERFCTR_CP_0_LO,
 188                 rbmemptr_stats(ring, index, cpcycles_end));
 189         get_stats_counter(ring, REG_A6XX_GMU_ALWAYS_ON_COUNTER_L + 0x1a800,
 190                 rbmemptr_stats(ring, index, alwayson_end));
 191
 192         /* Write the fence to the scratch register */
 193         OUT_PKT4(ring, REG_A6XX_CP_SCRATCH_REG(2), 1);
 194         OUT_RING(ring, submit->seqno);
 195
 196         /*
 197          * Execute a CACHE_FLUSH_TS event. This will ensure that the
 198          * timestamp is written to the memory and then triggers the interrupt
 199          */
 200         OUT_PKT7(ring, CP_EVENT_WRITE, 4);
 201         OUT_RING(ring, CP_EVENT_WRITE_0_EVENT(CACHE_FLUSH_TS) |
 202                 CP_EVENT_WRITE_0_IRQ);
 203         OUT_RING(ring, lower_32_bits(rbmemptr(ring, fence)));
 204         OUT_RING(ring, upper_32_bits(rbmemptr(ring, fence)));
 205         OUT_RING(ring, submit->seqno);
 206
 207         trace_msm_gpu_submit_flush(submit,
 208                 gmu_read64(&a6xx_gpu->gmu, REG_A6XX_GMU_ALWAYS_ON_COUNTER_L,
 209                         REG_A6XX_GMU_ALWAYS_ON_COUNTER_H));
 210
 211         a6xx_flush(gpu, ring);
 212 }
 213
 214 const struct adreno_reglist a630_hwcg[] = {
 215         {REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222},
 216         {REG_A6XX_RBBM_CLOCK_CNTL_SP1, 0x22222222},
 217         {REG_A6XX_RBBM_CLOCK_CNTL_SP2, 0x22222222},
 218         {REG_A6XX_RBBM_CLOCK_CNTL_SP3, 0x22222222},
 219         {REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02022220},
 220         {REG_A6XX_RBBM_CLOCK_CNTL2_SP1, 0x02022220},
 221         {REG_A6XX_RBBM_CLOCK_CNTL2_SP2, 0x02022220},
 222         {REG_A6XX_RBBM_CLOCK_CNTL2_SP3, 0x02022220},
 223         {REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
 224         {REG_A6XX_RBBM_CLOCK_DELAY_SP1, 0x00000080},
 225         {REG_A6XX_RBBM_CLOCK_DELAY_SP2, 0x00000080},
 226         {REG_A6XX_RBBM_CLOCK_DELAY_SP3, 0x00000080},
 227         {REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000f3cf},
 228         {REG_A6XX_RBBM_CLOCK_HYST_SP1, 0x0000f3cf},
 229         {REG_A6XX_RBBM_CLOCK_HYST_SP2, 0x0000f3cf},
 230         {REG_A6XX_RBBM_CLOCK_HYST_SP3, 0x0000f3cf},
 231         {REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
 232         {REG_A6XX_RBBM_CLOCK_CNTL_TP1, 0x02222222},
 233         {REG_A6XX_RBBM_CLOCK_CNTL_TP2, 0x02222222},
 234         {REG_A6XX_RBBM_CLOCK_CNTL_TP3, 0x02222222},
 235         {REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
 236         {REG_A6XX_RBBM_CLOCK_CNTL2_TP1, 0x22222222},
 237         {REG_A6XX_RBBM_CLOCK_CNTL2_TP2, 0x22222222},
 238         {REG_A6XX_RBBM_CLOCK_CNTL2_TP3, 0x22222222},
 239         {REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
 240         {REG_A6XX_RBBM_CLOCK_CNTL3_TP1, 0x22222222},
 241         {REG_A6XX_RBBM_CLOCK_CNTL3_TP2, 0x22222222},
 242         {REG_A6XX_RBBM_CLOCK_CNTL3_TP3, 0x22222222},
 243         {REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
 244         {REG_A6XX_RBBM_CLOCK_CNTL4_TP1, 0x00022222},
 245         {REG_A6XX_RBBM_CLOCK_CNTL4_TP2, 0x00022222},
 246         {REG_A6XX_RBBM_CLOCK_CNTL4_TP3, 0x00022222},
 247         {REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
 248         {REG_A6XX_RBBM_CLOCK_HYST_TP1, 0x77777777},
 249         {REG_A6XX_RBBM_CLOCK_HYST_TP2, 0x77777777},
 250         {REG_A6XX_RBBM_CLOCK_HYST_TP3, 0x77777777},
 251         {REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
 252         {REG_A6XX_RBBM_CLOCK_HYST2_TP1, 0x77777777},
 253         {REG_A6XX_RBBM_CLOCK_HYST2_TP2, 0x77777777},
 254         {REG_A6XX_RBBM_CLOCK_HYST2_TP3, 0x77777777},
 255         {REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
 256         {REG_A6XX_RBBM_CLOCK_HYST3_TP1, 0x77777777},
 257         {REG_A6XX_RBBM_CLOCK_HYST3_TP2, 0x77777777},
 258         {REG_A6XX_RBBM_CLOCK_HYST3_TP3, 0x77777777},
 259         {REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
 260         {REG_A6XX_RBBM_CLOCK_HYST4_TP1, 0x00077777},
 261         {REG_A6XX_RBBM_CLOCK_HYST4_TP2, 0x00077777},
 262         {REG_A6XX_RBBM_CLOCK_HYST4_TP3, 0x00077777},
 263         {REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
 264         {REG_A6XX_RBBM_CLOCK_DELAY_TP1, 0x11111111},
 265         {REG_A6XX_RBBM_CLOCK_DELAY_TP2, 0x11111111},
 266         {REG_A6XX_RBBM_CLOCK_DELAY_TP3, 0x11111111},
 267         {REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
 268         {REG_A6XX_RBBM_CLOCK_DELAY2_TP1, 0x11111111},
 269         {REG_A6XX_RBBM_CLOCK_DELAY2_TP2, 0x11111111},
 270         {REG_A6XX_RBBM_CLOCK_DELAY2_TP3, 0x11111111},
 271         {REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
 272         {REG_A6XX_RBBM_CLOCK_DELAY3_TP1, 0x11111111},
 273         {REG_A6XX_RBBM_CLOCK_DELAY3_TP2, 0x11111111},
 274         {REG_A6XX_RBBM_CLOCK_DELAY3_TP3, 0x11111111},
 275         {REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
 276         {REG_A6XX_RBBM_CLOCK_DELAY4_TP1, 0x00011111},
 277         {REG_A6XX_RBBM_CLOCK_DELAY4_TP2, 0x00011111},
 278         {REG_A6XX_RBBM_CLOCK_DELAY4_TP3, 0x00011111},
 279         {REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
 280         {REG_A6XX_RBBM_CLOCK_CNTL2_UCHE, 0x22222222},
 281         {REG_A6XX_RBBM_CLOCK_CNTL3_UCHE, 0x22222222},
 282         {REG_A6XX_RBBM_CLOCK_CNTL4_UCHE, 0x00222222},
 283         {REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
 284         {REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
 285         {REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
 286         {REG_A6XX_RBBM_CLOCK_CNTL_RB1, 0x22222222},
 287         {REG_A6XX_RBBM_CLOCK_CNTL_RB2, 0x22222222},
 288         {REG_A6XX_RBBM_CLOCK_CNTL_RB3, 0x22222222},
 289         {REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x00002222},
 290         {REG_A6XX_RBBM_CLOCK_CNTL2_RB1, 0x00002222},
 291         {REG_A6XX_RBBM_CLOCK_CNTL2_RB2, 0x00002222},
 292         {REG_A6XX_RBBM_CLOCK_CNTL2_RB3, 0x00002222},
 293         {REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
 294         {REG_A6XX_RBBM_CLOCK_CNTL_CCU1, 0x00002220},
 295         {REG_A6XX_RBBM_CLOCK_CNTL_CCU2, 0x00002220},
 296         {REG_A6XX_RBBM_CLOCK_CNTL_CCU3, 0x00002220},
 297         {REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040f00},
 298         {REG_A6XX_RBBM_CLOCK_HYST_RB_CCU1, 0x00040f00},
 299         {REG_A6XX_RBBM_CLOCK_HYST_RB_CCU2, 0x00040f00},
 300         {REG_A6XX_RBBM_CLOCK_HYST_RB_CCU3, 0x00040f00},
 301         {REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05022022},
 302         {REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
 303         {REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
 304         {REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
 305         {REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
 306         {REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
 307         {REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
 308         {REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
 309         {REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
 310         {REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
 311         {REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
 312         {REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
 313         {REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
 314         {REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
 315         {REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
 316         {REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
 317         {REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
 318         {REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
 319         {REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
 320         {},
 321 };
 322
 323 const struct adreno_reglist a640_hwcg[] = {
 324         {REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
 325         {REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
 326         {REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
 327         {REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
 328         {REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
 329         {REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
 330         {REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
 331         {REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
 332         {REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
 333         {REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
 334         {REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
 335         {REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
 336         {REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
 337         {REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
 338         {REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
 339         {REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
 340         {REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
 341         {REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
 342         {REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
 343         {REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
 344         {REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x05222022},
 345         {REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
 346         {REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
 347         {REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
 348         {REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
 349         {REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
 350         {REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
 351         {REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
 352         {REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
 353         {REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
 354         {REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
 355         {REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
 356         {REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
 357         {REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
 358         {REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
 359         {REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
 360         {REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
 361         {REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
 362         {REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
 363         {REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000000},
 364         {REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
 365         {REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
 366         {REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
 367         {REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
 368         {REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
 369         {REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
 370         {REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
 371         {REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
 372         {REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
 373         {},
 374 };
 375
 376 const struct adreno_reglist a650_hwcg[] = {
 377         {REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x02222222},
 378         {REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
 379         {REG_A6XX_RBBM_CLOCK_DELAY_SP0, 0x00000080},
 380         {REG_A6XX_RBBM_CLOCK_HYST_SP0, 0x0000F3CF},
 381         {REG_A6XX_RBBM_CLOCK_CNTL_TP0, 0x02222222},
 382         {REG_A6XX_RBBM_CLOCK_CNTL2_TP0, 0x22222222},
 383         {REG_A6XX_RBBM_CLOCK_CNTL3_TP0, 0x22222222},
 384         {REG_A6XX_RBBM_CLOCK_CNTL4_TP0, 0x00022222},
 385         {REG_A6XX_RBBM_CLOCK_DELAY_TP0, 0x11111111},
 386         {REG_A6XX_RBBM_CLOCK_DELAY2_TP0, 0x11111111},
 387         {REG_A6XX_RBBM_CLOCK_DELAY3_TP0, 0x11111111},
 388         {REG_A6XX_RBBM_CLOCK_DELAY4_TP0, 0x00011111},
 389         {REG_A6XX_RBBM_CLOCK_HYST_TP0, 0x77777777},
 390         {REG_A6XX_RBBM_CLOCK_HYST2_TP0, 0x77777777},
 391         {REG_A6XX_RBBM_CLOCK_HYST3_TP0, 0x77777777},
 392         {REG_A6XX_RBBM_CLOCK_HYST4_TP0, 0x00077777},
 393         {REG_A6XX_RBBM_CLOCK_CNTL_RB0, 0x22222222},
 394         {REG_A6XX_RBBM_CLOCK_CNTL2_RB0, 0x01002222},
 395         {REG_A6XX_RBBM_CLOCK_CNTL_CCU0, 0x00002220},
 396         {REG_A6XX_RBBM_CLOCK_HYST_RB_CCU0, 0x00040F00},
 397         {REG_A6XX_RBBM_CLOCK_CNTL_RAC, 0x25222022},
 398         {REG_A6XX_RBBM_CLOCK_CNTL2_RAC, 0x00005555},
 399         {REG_A6XX_RBBM_CLOCK_DELAY_RAC, 0x00000011},
 400         {REG_A6XX_RBBM_CLOCK_HYST_RAC, 0x00445044},
 401         {REG_A6XX_RBBM_CLOCK_CNTL_TSE_RAS_RBBM, 0x04222222},
 402         {REG_A6XX_RBBM_CLOCK_MODE_VFD, 0x00002222},
 403         {REG_A6XX_RBBM_CLOCK_MODE_GPC, 0x00222222},
 404         {REG_A6XX_RBBM_CLOCK_DELAY_HLSQ_2, 0x00000002},
 405         {REG_A6XX_RBBM_CLOCK_MODE_HLSQ, 0x00002222},
 406         {REG_A6XX_RBBM_CLOCK_DELAY_TSE_RAS_RBBM, 0x00004000},
 407         {REG_A6XX_RBBM_CLOCK_DELAY_VFD, 0x00002222},
 408         {REG_A6XX_RBBM_CLOCK_DELAY_GPC, 0x00000200},
 409         {REG_A6XX_RBBM_CLOCK_DELAY_HLSQ, 0x00000000},
 410         {REG_A6XX_RBBM_CLOCK_HYST_TSE_RAS_RBBM, 0x00000000},
 411         {REG_A6XX_RBBM_CLOCK_HYST_VFD, 0x00000000},
 412         {REG_A6XX_RBBM_CLOCK_HYST_GPC, 0x04104004},
 413         {REG_A6XX_RBBM_CLOCK_HYST_HLSQ, 0x00000000},
 414         {REG_A6XX_RBBM_CLOCK_CNTL_TEX_FCHE, 0x00000222},
 415         {REG_A6XX_RBBM_CLOCK_DELAY_TEX_FCHE, 0x00000111},
 416         {REG_A6XX_RBBM_CLOCK_HYST_TEX_FCHE, 0x00000777},
 417         {REG_A6XX_RBBM_CLOCK_CNTL_UCHE, 0x22222222},
 418         {REG_A6XX_RBBM_CLOCK_HYST_UCHE, 0x00000004},
 419         {REG_A6XX_RBBM_CLOCK_DELAY_UCHE, 0x00000002},
 420         {REG_A6XX_RBBM_ISDB_CNT, 0x00000182},
 421         {REG_A6XX_RBBM_RAC_THRESHOLD_CNT, 0x00000000},
 422         {REG_A6XX_RBBM_SP_HYST_CNT, 0x00000000},
 423         {REG_A6XX_RBBM_CLOCK_CNTL_GMU_GX, 0x00000222},
 424         {REG_A6XX_RBBM_CLOCK_DELAY_GMU_GX, 0x00000111},
 425         {REG_A6XX_RBBM_CLOCK_HYST_GMU_GX, 0x00000555},
 426         {},
 427 };
 428
 429 static void a6xx_set_hwcg(struct msm_gpu *gpu, bool state)
 430 {
 431         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 432         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
 433         struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
 434         const struct adreno_reglist *reg;
 435         unsigned int i;
 436         u32 val, clock_cntl_on;
 437
 438         if (!adreno_gpu->info->hwcg)
 439                 return;
 440
 441         if (adreno_is_a630(adreno_gpu))
 442                 clock_cntl_on = 0x8aa8aa02;
 443         else
 444                 clock_cntl_on = 0x8aa8aa82;
 445
 446         val = gpu_read(gpu, REG_A6XX_RBBM_CLOCK_CNTL);
 447
 448         /* Don't re-program the registers if they are already correct */
 449         if ((!state && !val) || (state && (val == clock_cntl_on)))
 450                 return;
 451
 452         /* Disable SP clock before programming HWCG registers */
 453         gmu_rmw(gmu, REG_A6XX_GPU_GMU_GX_SPTPRAC_CLOCK_CONTROL, 1, 0);
 454
 455         for (i = 0; (reg = &adreno_gpu->info->hwcg[i], reg->offset); i++)
 456                 gpu_write(gpu, reg->offset, state ? reg->value : 0);
 457
 458         /* Enable SP clock */
 459         gmu_rmw(gmu, REG_A6XX_GPU_GMU_GX_SPTPRAC_CLOCK_CONTROL, 0, 1);
 460
 461         gpu_write(gpu, REG_A6XX_RBBM_CLOCK_CNTL, state ? clock_cntl_on : 0);
 462 }
 463
 464 static void a6xx_set_ubwc_config(struct msm_gpu *gpu)
 465 {
 466         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 467         u32 lower_bit = 2;
 468         u32 amsbc = 0;
 469         u32 rgb565_predicator = 0;
 470         u32 uavflagprd_inv = 0;
 471
 472         /* a618 is using the hw default values */
 473         if (adreno_is_a618(adreno_gpu))
 474                 return;
 475
 476         if (adreno_is_a640(adreno_gpu))
 477                 amsbc = 1;
 478
 479         if (adreno_is_a650(adreno_gpu)) {
 480                 /* TODO: get ddr type from bootloader and use 2 for LPDDR4 */
 481                 lower_bit = 3;
 482                 amsbc = 1;
 483                 rgb565_predicator = 1;
 484                 uavflagprd_inv = 2;
 485         }
 486
 487         gpu_write(gpu, REG_A6XX_RB_NC_MODE_CNTL,
 488                 rgb565_predicator << 11 | amsbc << 4 | lower_bit << 1);
 489         gpu_write(gpu, REG_A6XX_TPL1_NC_MODE_CNTL, lower_bit << 1);
 490         gpu_write(gpu, REG_A6XX_SP_NC_MODE_CNTL,
 491                 uavflagprd_inv << 4 | lower_bit << 1);
 492         gpu_write(gpu, REG_A6XX_UCHE_MODE_CNTL, lower_bit << 21);
 493 }
 494
 495 static int a6xx_cp_init(struct msm_gpu *gpu)
 496 {
 497         struct msm_ringbuffer *ring = gpu->rb[0];
 498
 499         OUT_PKT7(ring, CP_ME_INIT, 8);
 500
 501         OUT_RING(ring, 0x0000002f);
 502
 503         /* Enable multiple hardware contexts */
 504         OUT_RING(ring, 0x00000003);
 505
 506         /* Enable error detection */
 507         OUT_RING(ring, 0x20000000);
 508
 509         /* Don't enable header dump */
 510         OUT_RING(ring, 0x00000000);
 511         OUT_RING(ring, 0x00000000);
 512
 513         /* No workarounds enabled */
 514         OUT_RING(ring, 0x00000000);
 515
 516         /* Pad rest of the cmds with 0's */
 517         OUT_RING(ring, 0x00000000);
 518         OUT_RING(ring, 0x00000000);
 519
 520         a6xx_flush(gpu, ring);
 521         return a6xx_idle(gpu, ring) ? 0 : -EINVAL;
 522 }
 523
 524 /*
 525  * Check that the microcode version is new enough to include several key
 526  * security fixes. Return true if the ucode is safe.
 527  */
 528 static bool a6xx_ucode_check_version(struct a6xx_gpu *a6xx_gpu,
 529                 struct drm_gem_object *obj)
 530 {
 531         struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 532         struct msm_gpu *gpu = &adreno_gpu->base;
 533         u32 *buf = msm_gem_get_vaddr(obj);
 534         bool ret = false;
 535
 536         if (IS_ERR(buf))
 537                 return false;
 538
 539         /*
 540          * Targets up to a640 (a618, a630 and a640) need to check for a
 541          * microcode version that is patched to support the whereami opcode or
 542          * one that is new enough to include it by default.
 543          */
 544         if (adreno_is_a618(adreno_gpu) || adreno_is_a630(adreno_gpu) ||
 545                 adreno_is_a640(adreno_gpu)) {
 546                 /*
 547                  * If the lowest nibble is 0xa that is an indication that this
 548                  * microcode has been patched. The actual version is in dword
 549                  * [3] but we only care about the patchlevel which is the lowest
 550                  * nibble of dword [3]
 551                  *
 552                  * Otherwise check that the firmware is greater than or equal
 553                  * to 1.90 which was the first version that had this fix built
 554                  * in
 555                  */
 556                 if ((((buf[0] & 0xf) == 0xa) && (buf[2] & 0xf) >= 1) ||
 557                         (buf[0] & 0xfff) >= 0x190) {
 558                         a6xx_gpu->has_whereami = true;
 559                         ret = true;
 560                         goto out;
 561                 }
 562
 563                 DRM_DEV_ERROR(&gpu->pdev->dev,
 564                         "a630 SQE ucode is too old. Have version %x need at least %x\n",
 565                         buf[0] & 0xfff, 0x190);
 566         }  else {
 567                 /*
 568                  * a650 tier targets don't need whereami but still need to be
 569                  * equal to or newer than 0.95 for other security fixes
 570                  */
 571                 if (adreno_is_a650(adreno_gpu)) {
 572                         if ((buf[0] & 0xfff) >= 0x095) {
 573                                 ret = true;
 574                                 goto out;
 575                         }
 576
 577                         DRM_DEV_ERROR(&gpu->pdev->dev,
 578                                 "a650 SQE ucode is too old. Have version %x need at least %x\n",
 579                                 buf[0] & 0xfff, 0x095);
 580                 }
 581
 582                 /*
 583                  * When a660 is added those targets should return true here
 584                  * since those have all the critical security fixes built in
 585                  * from the start
 586                  */
 587         }
 588 out:
 589         msm_gem_put_vaddr(obj);
 590         return ret;
 591 }
 592
 593 static int a6xx_ucode_init(struct msm_gpu *gpu)
 594 {
 595         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 596         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
 597
 598         if (!a6xx_gpu->sqe_bo) {
 599                 a6xx_gpu->sqe_bo = adreno_fw_create_bo(gpu,
 600                         adreno_gpu->fw[ADRENO_FW_SQE], &a6xx_gpu->sqe_iova);
 601
 602                 if (IS_ERR(a6xx_gpu->sqe_bo)) {
 603                         int ret = PTR_ERR(a6xx_gpu->sqe_bo);
 604
 605                         a6xx_gpu->sqe_bo = NULL;
 606                         DRM_DEV_ERROR(&gpu->pdev->dev,
 607                                 "Could not allocate SQE ucode: %d\n", ret);
 608
 609                         return ret;
 610                 }
 611
 612                 msm_gem_object_set_name(a6xx_gpu->sqe_bo, "sqefw");
 613                 if (!a6xx_ucode_check_version(a6xx_gpu, a6xx_gpu->sqe_bo)) {
 614                         msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
 615                         drm_gem_object_put(a6xx_gpu->sqe_bo);
 616
 617                         a6xx_gpu->sqe_bo = NULL;
 618                         return -EPERM;
 619                 }
 620         }
 621
 622         gpu_write64(gpu, REG_A6XX_CP_SQE_INSTR_BASE_LO,
 623                 REG_A6XX_CP_SQE_INSTR_BASE_HI, a6xx_gpu->sqe_iova);
 624
 625         return 0;
 626 }
 627
 628 static int a6xx_zap_shader_init(struct msm_gpu *gpu)
 629 {
 630         static bool loaded;
 631         int ret;
 632
 633         if (loaded)
 634                 return 0;
 635
 636         ret = adreno_zap_shader_load(gpu, GPU_PAS_ID);
 637
 638         loaded = !ret;
 639         return ret;
 640 }
 641
 642 #define A6XX_INT_MASK (A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR | \
 643           A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW | \
 644           A6XX_RBBM_INT_0_MASK_CP_HW_ERROR | \
 645           A6XX_RBBM_INT_0_MASK_CP_IB2 | \
 646           A6XX_RBBM_INT_0_MASK_CP_IB1 | \
 647           A6XX_RBBM_INT_0_MASK_CP_RB | \
 648           A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \
 649           A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW | \
 650           A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT | \
 651           A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \
 652           A6XX_RBBM_INT_0_MASK_UCHE_TRAP_INTR)
 653
 654 static int a6xx_hw_init(struct msm_gpu *gpu)
 655 {
 656         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 657         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
 658         int ret;
 659
 660         /* Make sure the GMU keeps the GPU on while we set it up */
 661         a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
 662
 663         gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_CNTL, 0);
 664
 665         /*
 666          * Disable the trusted memory range - we don't actually supported secure
 667          * memory rendering at this point in time and we don't want to block off
 668          * part of the virtual memory space.
 669          */
 670         gpu_write64(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_BASE_LO,
 671                 REG_A6XX_RBBM_SECVID_TSB_TRUSTED_BASE_HI, 0x00000000);
 672         gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_TRUSTED_SIZE, 0x00000000);
 673
 674         /* Turn on 64 bit addressing for all blocks */
 675         gpu_write(gpu, REG_A6XX_CP_ADDR_MODE_CNTL, 0x1);
 676         gpu_write(gpu, REG_A6XX_VSC_ADDR_MODE_CNTL, 0x1);
 677         gpu_write(gpu, REG_A6XX_GRAS_ADDR_MODE_CNTL, 0x1);
 678         gpu_write(gpu, REG_A6XX_RB_ADDR_MODE_CNTL, 0x1);
 679         gpu_write(gpu, REG_A6XX_PC_ADDR_MODE_CNTL, 0x1);
 680         gpu_write(gpu, REG_A6XX_HLSQ_ADDR_MODE_CNTL, 0x1);
 681         gpu_write(gpu, REG_A6XX_VFD_ADDR_MODE_CNTL, 0x1);
 682         gpu_write(gpu, REG_A6XX_VPC_ADDR_MODE_CNTL, 0x1);
 683         gpu_write(gpu, REG_A6XX_UCHE_ADDR_MODE_CNTL, 0x1);
 684         gpu_write(gpu, REG_A6XX_SP_ADDR_MODE_CNTL, 0x1);
 685         gpu_write(gpu, REG_A6XX_TPL1_ADDR_MODE_CNTL, 0x1);
 686         gpu_write(gpu, REG_A6XX_RBBM_SECVID_TSB_ADDR_MODE_CNTL, 0x1);
 687
 688         /* enable hardware clockgating */
 689         a6xx_set_hwcg(gpu, true);
 690
 691         /* VBIF/GBIF start*/
 692         if (adreno_is_a640(adreno_gpu) || adreno_is_a650(adreno_gpu)) {
 693                 gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE0, 0x00071620);
 694                 gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE1, 0x00071620);
 695                 gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE2, 0x00071620);
 696                 gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
 697                 gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
 698                 gpu_write(gpu, REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL, 0x3);
 699         } else {
 700                 gpu_write(gpu, REG_A6XX_RBBM_VBIF_CLIENT_QOS_CNTL, 0x3);
 701         }
 702
 703         if (adreno_is_a630(adreno_gpu))
 704                 gpu_write(gpu, REG_A6XX_VBIF_GATE_OFF_WRREQ_EN, 0x00000009);
 705
 706         /* Make all blocks contribute to the GPU BUSY perf counter */
 707         gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_GPU_BUSY_MASKED, 0xffffffff);
 708
 709         /* Disable L2 bypass in the UCHE */
 710         gpu_write(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX_LO, 0xffffffc0);
 711         gpu_write(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX_HI, 0x0001ffff);
 712         gpu_write(gpu, REG_A6XX_UCHE_TRAP_BASE_LO, 0xfffff000);
 713         gpu_write(gpu, REG_A6XX_UCHE_TRAP_BASE_HI, 0x0001ffff);
 714         gpu_write(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE_LO, 0xfffff000);
 715         gpu_write(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE_HI, 0x0001ffff);
 716
 717         if (!adreno_is_a650(adreno_gpu)) {
 718                 /* Set the GMEM VA range [0x100000:0x100000 + gpu->gmem - 1] */
 719                 gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MIN_LO,
 720                         REG_A6XX_UCHE_GMEM_RANGE_MIN_HI, 0x00100000);
 721
 722                 gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MAX_LO,
 723                         REG_A6XX_UCHE_GMEM_RANGE_MAX_HI,
 724                         0x00100000 + adreno_gpu->gmem - 1);
 725         }
 726
 727         gpu_write(gpu, REG_A6XX_UCHE_FILTER_CNTL, 0x804);
 728         gpu_write(gpu, REG_A6XX_UCHE_CACHE_WAYS, 0x4);
 729
 730         if (adreno_is_a640(adreno_gpu) || adreno_is_a650(adreno_gpu))
 731                 gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x02000140);
 732         else
 733                 gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x010000c0);
 734         gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x8040362c);
 735
 736         /* Setting the mem pool size */
 737         gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 128);
 738
 739         /* Setting the primFifo thresholds default values */
 740         if (adreno_is_a650(adreno_gpu))
 741                 gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00300000);
 742         else if (adreno_is_a640(adreno_gpu))
 743                 gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00200000);
 744         else
 745                 gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, (0x300 << 11));
 746
 747         /* Set the AHB default slave response to "ERROR" */
 748         gpu_write(gpu, REG_A6XX_CP_AHB_CNTL, 0x1);
 749
 750         /* Turn on performance counters */
 751         gpu_write(gpu, REG_A6XX_RBBM_PERFCTR_CNTL, 0x1);
 752
 753         /* Select CP0 to always count cycles */
 754         gpu_write(gpu, REG_A6XX_CP_PERFCTR_CP_SEL_0, PERF_CP_ALWAYS_COUNT);
 755
 756         a6xx_set_ubwc_config(gpu);
 757
 758         /* Enable fault detection */
 759         gpu_write(gpu, REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL,
 760                 (1 << 30) | 0x1fffff);
 761
 762         gpu_write(gpu, REG_A6XX_UCHE_CLIENT_PF, 1);
 763
 764         /* Set weights for bicubic filtering */
 765         if (adreno_is_a650(adreno_gpu)) {
 766                 gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_0, 0);
 767                 gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_1,
 768                         0x3fe05ff4);
 769                 gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_2,
 770                         0x3fa0ebee);
 771                 gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_3,
 772                         0x3f5193ed);
 773                 gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_4,
 774                         0x3f0243f0);
 775         }
 776
 777         /* Protect registers from the CP */
 778         gpu_write(gpu, REG_A6XX_CP_PROTECT_CNTL, 0x00000003);
 779
 780         gpu_write(gpu, REG_A6XX_CP_PROTECT(0),
 781                 A6XX_PROTECT_RDONLY(0x600, 0x51));
 782         gpu_write(gpu, REG_A6XX_CP_PROTECT(1), A6XX_PROTECT_RW(0xae50, 0x2));
 783         gpu_write(gpu, REG_A6XX_CP_PROTECT(2), A6XX_PROTECT_RW(0x9624, 0x13));
 784         gpu_write(gpu, REG_A6XX_CP_PROTECT(3), A6XX_PROTECT_RW(0x8630, 0x8));
 785         gpu_write(gpu, REG_A6XX_CP_PROTECT(4), A6XX_PROTECT_RW(0x9e70, 0x1));
 786         gpu_write(gpu, REG_A6XX_CP_PROTECT(5), A6XX_PROTECT_RW(0x9e78, 0x187));
 787         gpu_write(gpu, REG_A6XX_CP_PROTECT(6), A6XX_PROTECT_RW(0xf000, 0x810));
 788         gpu_write(gpu, REG_A6XX_CP_PROTECT(7),
 789                 A6XX_PROTECT_RDONLY(0xfc00, 0x3));
 790         gpu_write(gpu, REG_A6XX_CP_PROTECT(8), A6XX_PROTECT_RW(0x50e, 0x0));
 791         gpu_write(gpu, REG_A6XX_CP_PROTECT(9), A6XX_PROTECT_RDONLY(0x50f, 0x0));
 792         gpu_write(gpu, REG_A6XX_CP_PROTECT(10), A6XX_PROTECT_RW(0x510, 0x0));
 793         gpu_write(gpu, REG_A6XX_CP_PROTECT(11),
 794                 A6XX_PROTECT_RDONLY(0x0, 0x4f9));
 795         gpu_write(gpu, REG_A6XX_CP_PROTECT(12),
 796                 A6XX_PROTECT_RDONLY(0x501, 0xa));
 797         gpu_write(gpu, REG_A6XX_CP_PROTECT(13),
 798                 A6XX_PROTECT_RDONLY(0x511, 0x44));
 799         gpu_write(gpu, REG_A6XX_CP_PROTECT(14), A6XX_PROTECT_RW(0xe00, 0xe));
 800         gpu_write(gpu, REG_A6XX_CP_PROTECT(15), A6XX_PROTECT_RW(0x8e00, 0x0));
 801         gpu_write(gpu, REG_A6XX_CP_PROTECT(16), A6XX_PROTECT_RW(0x8e50, 0xf));
 802         gpu_write(gpu, REG_A6XX_CP_PROTECT(17), A6XX_PROTECT_RW(0xbe02, 0x0));
 803         gpu_write(gpu, REG_A6XX_CP_PROTECT(18),
 804                 A6XX_PROTECT_RW(0xbe20, 0x11f3));
 805         gpu_write(gpu, REG_A6XX_CP_PROTECT(19), A6XX_PROTECT_RW(0x800, 0x82));
 806         gpu_write(gpu, REG_A6XX_CP_PROTECT(20), A6XX_PROTECT_RW(0x8a0, 0x8));
 807         gpu_write(gpu, REG_A6XX_CP_PROTECT(21), A6XX_PROTECT_RW(0x8ab, 0x19));
 808         gpu_write(gpu, REG_A6XX_CP_PROTECT(22), A6XX_PROTECT_RW(0x900, 0x4d));
 809         gpu_write(gpu, REG_A6XX_CP_PROTECT(23), A6XX_PROTECT_RW(0x98d, 0x76));
 810         gpu_write(gpu, REG_A6XX_CP_PROTECT(24),
 811                         A6XX_PROTECT_RDONLY(0x980, 0x4));
 812         gpu_write(gpu, REG_A6XX_CP_PROTECT(25), A6XX_PROTECT_RW(0xa630, 0x0));
 813
 814         /* Enable expanded apriv for targets that support it */
 815         if (gpu->hw_apriv) {
 816                 gpu_write(gpu, REG_A6XX_CP_APRIV_CNTL,
 817                         (1 << 6) | (1 << 5) | (1 << 3) | (1 << 2) | (1 << 1));
 818         }
 819
 820         /* Enable interrupts */
 821         gpu_write(gpu, REG_A6XX_RBBM_INT_0_MASK, A6XX_INT_MASK);
 822
 823         ret = adreno_hw_init(gpu);
 824         if (ret)
 825                 goto out;
 826
 827         ret = a6xx_ucode_init(gpu);
 828         if (ret)
 829                 goto out;
 830
 831         /* Set the ringbuffer address */
 832         gpu_write64(gpu, REG_A6XX_CP_RB_BASE, REG_A6XX_CP_RB_BASE_HI,
 833                 gpu->rb[0]->iova);
 834
 835         /* Targets that support extended APRIV can use the RPTR shadow from
 836          * hardware but all the other ones need to disable the feature. Targets
 837          * that support the WHERE_AM_I opcode can use that instead
 838          */
 839         if (adreno_gpu->base.hw_apriv)
 840                 gpu_write(gpu, REG_A6XX_CP_RB_CNTL, MSM_GPU_RB_CNTL_DEFAULT);
 841         else
 842                 gpu_write(gpu, REG_A6XX_CP_RB_CNTL,
 843                         MSM_GPU_RB_CNTL_DEFAULT | AXXX_CP_RB_CNTL_NO_UPDATE);
 844
 845         /*
 846          * Expanded APRIV and targets that support WHERE_AM_I both need a
 847          * privileged buffer to store the RPTR shadow
 848          */
 849
 850         if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami) {
 851                 if (!a6xx_gpu->shadow_bo) {
 852                         a6xx_gpu->shadow = msm_gem_kernel_new_locked(gpu->dev,
 853                                 sizeof(u32) * gpu->nr_rings,
 854                                 MSM_BO_UNCACHED | MSM_BO_MAP_PRIV,
 855                                 gpu->aspace, &a6xx_gpu->shadow_bo,
 856                                 &a6xx_gpu->shadow_iova);
 857
 858                         if (IS_ERR(a6xx_gpu->shadow))
 859                                 return PTR_ERR(a6xx_gpu->shadow);
 860                 }
 861
 862                 gpu_write64(gpu, REG_A6XX_CP_RB_RPTR_ADDR_LO,
 863                         REG_A6XX_CP_RB_RPTR_ADDR_HI,
 864                         shadowptr(a6xx_gpu, gpu->rb[0]));
 865         }
 866
 867         /* Always come up on rb 0 */
 868         a6xx_gpu->cur_ring = gpu->rb[0];
 869
 870         a6xx_gpu->cur_ctx = NULL;
 871
 872         /* Enable the SQE_to start the CP engine */
 873         gpu_write(gpu, REG_A6XX_CP_SQE_CNTL, 1);
 874
 875         ret = a6xx_cp_init(gpu);
 876         if (ret)
 877                 goto out;
 878
 879         /*
 880          * Try to load a zap shader into the secure world. If successful
 881          * we can use the CP to switch out of secure mode. If not then we
 882          * have no resource but to try to switch ourselves out manually. If we
 883          * guessed wrong then access to the RBBM_SECVID_TRUST_CNTL register will
 884          * be blocked and a permissions violation will soon follow.
 885          */
 886         ret = a6xx_zap_shader_init(gpu);
 887         if (!ret) {
 888                 OUT_PKT7(gpu->rb[0], CP_SET_SECURE_MODE, 1);
 889                 OUT_RING(gpu->rb[0], 0x00000000);
 890
 891                 a6xx_flush(gpu, gpu->rb[0]);
 892                 if (!a6xx_idle(gpu, gpu->rb[0]))
 893                         return -EINVAL;
 894         } else if (ret == -ENODEV) {
 895                 /*
 896                  * This device does not use zap shader (but print a warning
 897                  * just in case someone got their dt wrong.. hopefully they
 898                  * have a debug UART to realize the error of their ways...
 899                  * if you mess this up you are about to crash horribly)
 900                  */
 901                 dev_warn_once(gpu->dev->dev,
 902                         "Zap shader not enabled - using SECVID_TRUST_CNTL instead\n");
 903                 gpu_write(gpu, REG_A6XX_RBBM_SECVID_TRUST_CNTL, 0x0);
 904                 ret = 0;
 905         } else {
 906                 return ret;
 907         }
 908
 909 out:
 910         /*
 911          * Tell the GMU that we are done touching the GPU and it can start power
 912          * management
 913          */
 914         a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
 915
 916         if (a6xx_gpu->gmu.legacy) {
 917                 /* Take the GMU out of its special boot mode */
 918                 a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
 919         }
 920
 921         return ret;
 922 }
 923
 924 static void a6xx_dump(struct msm_gpu *gpu)
 925 {
 926         DRM_DEV_INFO(&gpu->pdev->dev, "status:   %08x\n",
 927                         gpu_read(gpu, REG_A6XX_RBBM_STATUS));
 928         adreno_dump(gpu);
 929 }
 930
 931 #define VBIF_RESET_ACK_TIMEOUT  100
 932 #define VBIF_RESET_ACK_MASK     0x00f0
 933
 934 static void a6xx_recover(struct msm_gpu *gpu)
 935 {
 936         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 937         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
 938         int i;
 939
 940         adreno_dump_info(gpu);
 941
 942         for (i = 0; i < 8; i++)
 943                 DRM_DEV_INFO(&gpu->pdev->dev, "CP_SCRATCH_REG%d: %u\n", i,
 944                         gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(i)));
 945
 946         if (hang_debug)
 947                 a6xx_dump(gpu);
 948
 949         /*
 950          * Turn off keep alive that might have been enabled by the hang
 951          * interrupt
 952          */
 953         gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 0);
 954
 955         gpu->funcs->pm_suspend(gpu);
 956         gpu->funcs->pm_resume(gpu);
 957
 958         msm_gpu_hw_init(gpu);
 959 }
 960
 961 static int a6xx_fault_handler(void *arg, unsigned long iova, int flags)
 962 {
 963         struct msm_gpu *gpu = arg;
 964
 965         pr_warn_ratelimited("*** gpu fault: iova=%08lx, flags=%d (%u,%u,%u,%u)\n",
 966                         iova, flags,
 967                         gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(4)),
 968                         gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(5)),
 969                         gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(6)),
 970                         gpu_read(gpu, REG_A6XX_CP_SCRATCH_REG(7)));
 971
 972         return -EFAULT;
 973 }
 974
 975 static void a6xx_cp_hw_err_irq(struct msm_gpu *gpu)
 976 {
 977         u32 status = gpu_read(gpu, REG_A6XX_CP_INTERRUPT_STATUS);
 978
 979         if (status & A6XX_CP_INT_CP_OPCODE_ERROR) {
 980                 u32 val;
 981
 982                 gpu_write(gpu, REG_A6XX_CP_SQE_STAT_ADDR, 1);
 983                 val = gpu_read(gpu, REG_A6XX_CP_SQE_STAT_DATA);
 984                 dev_err_ratelimited(&gpu->pdev->dev,
 985                         "CP | opcode error | possible opcode=0x%8.8X\n",
 986                         val);
 987         }
 988
 989         if (status & A6XX_CP_INT_CP_UCODE_ERROR)
 990                 dev_err_ratelimited(&gpu->pdev->dev,
 991                         "CP ucode error interrupt\n");
 992
 993         if (status & A6XX_CP_INT_CP_HW_FAULT_ERROR)
 994                 dev_err_ratelimited(&gpu->pdev->dev, "CP | HW fault | status=0x%8.8X\n",
 995                         gpu_read(gpu, REG_A6XX_CP_HW_FAULT));
 996
 997         if (status & A6XX_CP_INT_CP_REGISTER_PROTECTION_ERROR) {
 998                 u32 val = gpu_read(gpu, REG_A6XX_CP_PROTECT_STATUS);
 999
1000                 dev_err_ratelimited(&gpu->pdev->dev,
1001                         "CP | protected mode error | %s | addr=0x%8.8X | status=0x%8.8X\n",
1002                         val & (1 << 20) ? "READ" : "WRITE",
1003                         (val & 0x3ffff), val);
1004         }
1005
1006         if (status & A6XX_CP_INT_CP_AHB_ERROR)
1007                 dev_err_ratelimited(&gpu->pdev->dev, "CP AHB error interrupt\n");
1008
1009         if (status & A6XX_CP_INT_CP_VSD_PARITY_ERROR)
1010                 dev_err_ratelimited(&gpu->pdev->dev, "CP VSD decoder parity error\n");
1011
1012         if (status & A6XX_CP_INT_CP_ILLEGAL_INSTR_ERROR)
1013                 dev_err_ratelimited(&gpu->pdev->dev, "CP illegal instruction error\n");
1014
1015 }
1016
1017 static void a6xx_fault_detect_irq(struct msm_gpu *gpu)
1018 {
1019         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1020         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1021         struct msm_ringbuffer *ring = gpu->funcs->active_ring(gpu);
1022
1023         /*
1024          * Force the GPU to stay on until after we finish
1025          * collecting information
1026          */
1027         gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 1);
1028
1029         DRM_DEV_ERROR(&gpu->pdev->dev,
1030                 "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",
1031                 ring ? ring->id : -1, ring ? ring->seqno : 0,
1032                 gpu_read(gpu, REG_A6XX_RBBM_STATUS),
1033                 gpu_read(gpu, REG_A6XX_CP_RB_RPTR),
1034                 gpu_read(gpu, REG_A6XX_CP_RB_WPTR),
1035                 gpu_read64(gpu, REG_A6XX_CP_IB1_BASE, REG_A6XX_CP_IB1_BASE_HI),
1036                 gpu_read(gpu, REG_A6XX_CP_IB1_REM_SIZE),
1037                 gpu_read64(gpu, REG_A6XX_CP_IB2_BASE, REG_A6XX_CP_IB2_BASE_HI),
1038                 gpu_read(gpu, REG_A6XX_CP_IB2_REM_SIZE));
1039
1040         /* Turn off the hangcheck timer to keep it from bothering us */
1041         del_timer(&gpu->hangcheck_timer);
1042
1043         kthread_queue_work(gpu->worker, &gpu->recover_work);
1044 }
1045
1046 static irqreturn_t a6xx_irq(struct msm_gpu *gpu)
1047 {
1048         u32 status = gpu_read(gpu, REG_A6XX_RBBM_INT_0_STATUS);
1049
1050         gpu_write(gpu, REG_A6XX_RBBM_INT_CLEAR_CMD, status);
1051
1052         if (status & A6XX_RBBM_INT_0_MASK_RBBM_HANG_DETECT)
1053                 a6xx_fault_detect_irq(gpu);
1054
1055         if (status & A6XX_RBBM_INT_0_MASK_CP_AHB_ERROR)
1056                 dev_err_ratelimited(&gpu->pdev->dev, "CP | AHB bus error\n");
1057
1058         if (status & A6XX_RBBM_INT_0_MASK_CP_HW_ERROR)
1059                 a6xx_cp_hw_err_irq(gpu);
1060
1061         if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNCFIFO_OVERFLOW)
1062                 dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB ASYNC overflow\n");
1063
1064         if (status & A6XX_RBBM_INT_0_MASK_RBBM_ATB_BUS_OVERFLOW)
1065                 dev_err_ratelimited(&gpu->pdev->dev, "RBBM | ATB bus overflow\n");
1066
1067         if (status & A6XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS)
1068                 dev_err_ratelimited(&gpu->pdev->dev, "UCHE | Out of bounds access\n");
1069
1070         if (status & A6XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS)
1071                 msm_gpu_retire(gpu);
1072
1073         return IRQ_HANDLED;
1074 }
1075
1076 static void a6xx_llc_rmw(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 mask, u32 or)
1077 {
1078         return msm_rmw(a6xx_gpu->llc_mmio + (reg << 2), mask, or);
1079 }
1080
1081 static void a6xx_llc_write(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 value)
1082 {
1083         return msm_writel(value, a6xx_gpu->llc_mmio + (reg << 2));
1084 }
1085
1086 static void a6xx_llc_deactivate(struct a6xx_gpu *a6xx_gpu)
1087 {
1088         llcc_slice_deactivate(a6xx_gpu->llc_slice);
1089         llcc_slice_deactivate(a6xx_gpu->htw_llc_slice);
1090 }
1091
1092 static void a6xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
1093 {
1094         struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
1095         struct msm_gpu *gpu = &adreno_gpu->base;
1096         u32 cntl1_regval = 0;
1097
1098         if (IS_ERR(a6xx_gpu->llc_mmio))
1099                 return;
1100
1101         if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
1102                 u32 gpu_scid = llcc_get_slice_id(a6xx_gpu->llc_slice);
1103
1104                 gpu_scid &= 0x1f;
1105                 cntl1_regval = (gpu_scid << 0) | (gpu_scid << 5) | (gpu_scid << 10) |
1106                                (gpu_scid << 15) | (gpu_scid << 20);
1107         }
1108
1109         /*
1110          * For targets with a MMU500, activate the slice but don't program the
1111          * register.  The XBL will take care of that.
1112          */
1113         if (!llcc_slice_activate(a6xx_gpu->htw_llc_slice)) {
1114                 if (!a6xx_gpu->have_mmu500) {
1115                         u32 gpuhtw_scid = llcc_get_slice_id(a6xx_gpu->htw_llc_slice);
1116
1117                         gpuhtw_scid &= 0x1f;
1118                         cntl1_regval |= FIELD_PREP(GENMASK(29, 25), gpuhtw_scid);
1119                 }
1120         }
1121
1122         if (cntl1_regval) {
1123                 /*
1124                  * Program the slice IDs for the various GPU blocks and GPU MMU
1125                  * pagetables
1126                  */
1127                 if (a6xx_gpu->have_mmu500)
1128                         gpu_rmw(gpu, REG_A6XX_GBIF_SCACHE_CNTL1, GENMASK(24, 0),
1129                                 cntl1_regval);
1130                 else {
1131                         a6xx_llc_write(a6xx_gpu,
1132                                 REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_1, cntl1_regval);
1133
1134                         /*
1135                          * Program cacheability overrides to not allocate cache
1136                          * lines on a write miss
1137                          */
1138                         a6xx_llc_rmw(a6xx_gpu,
1139                                 REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_0, 0xF, 0x03);
1140                 }
1141         }
1142 }
1143
1144 static void a6xx_llc_slices_destroy(struct a6xx_gpu *a6xx_gpu)
1145 {
1146         llcc_slice_putd(a6xx_gpu->llc_slice);
1147         llcc_slice_putd(a6xx_gpu->htw_llc_slice);
1148 }
1149
1150 static void a6xx_llc_slices_init(struct platform_device *pdev,
1151                 struct a6xx_gpu *a6xx_gpu)
1152 {
1153         struct device_node *phandle;
1154
1155         /*
1156          * There is a different programming path for targets with an mmu500
1157          * attached, so detect if that is the case
1158          */
1159         phandle = of_parse_phandle(pdev->dev.of_node, "iommus", 0);
1160         a6xx_gpu->have_mmu500 = (phandle &&
1161                 of_device_is_compatible(phandle, "arm,mmu-500"));
1162         of_node_put(phandle);
1163
1164         if (a6xx_gpu->have_mmu500)
1165                 a6xx_gpu->llc_mmio = NULL;
1166         else
1167                 a6xx_gpu->llc_mmio = msm_ioremap(pdev, "cx_mem", "gpu_cx");
1168
1169         a6xx_gpu->llc_slice = llcc_slice_getd(LLCC_GPU);
1170         a6xx_gpu->htw_llc_slice = llcc_slice_getd(LLCC_GPUHTW);
1171
1172         if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
1173                 a6xx_gpu->llc_mmio = ERR_PTR(-EINVAL);
1174 }
1175
1176 static int a6xx_pm_resume(struct msm_gpu *gpu)
1177 {
1178         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1179         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1180         int ret;
1181
1182         gpu->needs_hw_init = true;
1183
1184         trace_msm_gpu_resume(0);
1185
1186         ret = a6xx_gmu_resume(a6xx_gpu);
1187         if (ret)
1188                 return ret;
1189
1190         msm_gpu_resume_devfreq(gpu);
1191
1192         a6xx_llc_activate(a6xx_gpu);
1193
1194         return 0;
1195 }
1196
1197 static int a6xx_pm_suspend(struct msm_gpu *gpu)
1198 {
1199         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1200         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1201         int i, ret;
1202
1203         trace_msm_gpu_suspend(0);
1204
1205         a6xx_llc_deactivate(a6xx_gpu);
1206
1207         devfreq_suspend_device(gpu->devfreq.devfreq);
1208
1209         ret = a6xx_gmu_stop(a6xx_gpu);
1210         if (ret)
1211                 return ret;
1212
1213         if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami)
1214                 for (i = 0; i < gpu->nr_rings; i++)
1215                         a6xx_gpu->shadow[i] = 0;
1216
1217         return 0;
1218 }
1219
1220 static int a6xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
1221 {
1222         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1223         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1224         static DEFINE_MUTEX(perfcounter_oob);
1225
1226         mutex_lock(&perfcounter_oob);
1227
1228         /* Force the GPU power on so we can read this register */
1229         a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
1230
1231         *value = gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER_LO,
1232                 REG_A6XX_CP_ALWAYS_ON_COUNTER_HI);
1233
1234         a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET);
1235         mutex_unlock(&perfcounter_oob);
1236         return 0;
1237 }
1238
1239 static struct msm_ringbuffer *a6xx_active_ring(struct msm_gpu *gpu)
1240 {
1241         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1242         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1243
1244         return a6xx_gpu->cur_ring;
1245 }
1246
1247 static void a6xx_destroy(struct msm_gpu *gpu)
1248 {
1249         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1250         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1251
1252         if (a6xx_gpu->sqe_bo) {
1253                 msm_gem_unpin_iova(a6xx_gpu->sqe_bo, gpu->aspace);
1254                 drm_gem_object_put(a6xx_gpu->sqe_bo);
1255         }
1256
1257         if (a6xx_gpu->shadow_bo) {
1258                 msm_gem_unpin_iova(a6xx_gpu->shadow_bo, gpu->aspace);
1259                 drm_gem_object_put(a6xx_gpu->shadow_bo);
1260         }
1261
1262         a6xx_llc_slices_destroy(a6xx_gpu);
1263
1264         a6xx_gmu_remove(a6xx_gpu);
1265
1266         adreno_gpu_cleanup(adreno_gpu);
1267         kfree(a6xx_gpu);
1268 }
1269
1270 static unsigned long a6xx_gpu_busy(struct msm_gpu *gpu)
1271 {
1272         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1273         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1274         u64 busy_cycles, busy_time;
1275
1276
1277         /* Only read the gpu busy if the hardware is already active */
1278         if (pm_runtime_get_if_in_use(a6xx_gpu->gmu.dev) == 0)
1279                 return 0;
1280
1281         busy_cycles = gmu_read64(&a6xx_gpu->gmu,
1282                         REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_L,
1283                         REG_A6XX_GMU_CX_GMU_POWER_COUNTER_XOCLK_0_H);
1284
1285         busy_time = (busy_cycles - gpu->devfreq.busy_cycles) * 10;
1286         do_div(busy_time, 192);
1287
1288         gpu->devfreq.busy_cycles = busy_cycles;
1289
1290         pm_runtime_put(a6xx_gpu->gmu.dev);
1291
1292         if (WARN_ON(busy_time > ~0LU))
1293                 return ~0LU;
1294
1295         return (unsigned long)busy_time;
1296 }
1297
1298 static struct msm_gem_address_space *
1299 a6xx_create_private_address_space(struct msm_gpu *gpu)
1300 {
1301         struct msm_mmu *mmu;
1302
1303         mmu = msm_iommu_pagetable_create(gpu->aspace->mmu);
1304
1305         if (IS_ERR(mmu))
1306                 return ERR_CAST(mmu);
1307
1308         return msm_gem_address_space_create(mmu,
1309                 "gpu", 0x100000000ULL, 0x1ffffffffULL);
1310 }
1311
1312 static uint32_t a6xx_get_rptr(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
1313 {
1314         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1315         struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1316
1317         if (adreno_gpu->base.hw_apriv || a6xx_gpu->has_whereami)
1318                 return a6xx_gpu->shadow[ring->id];
1319
1320         return ring->memptrs->rptr = gpu_read(gpu, REG_A6XX_CP_RB_RPTR);
1321 }
1322
1323 static const struct adreno_gpu_funcs funcs = {
1324         .base = {
1325                 .get_param = adreno_get_param,
1326                 .hw_init = a6xx_hw_init,
1327                 .pm_suspend = a6xx_pm_suspend,
1328                 .pm_resume = a6xx_pm_resume,
1329                 .recover = a6xx_recover,
1330                 .submit = a6xx_submit,
1331                 .active_ring = a6xx_active_ring,
1332                 .irq = a6xx_irq,
1333                 .destroy = a6xx_destroy,
1334 #if defined(CONFIG_DRM_MSM_GPU_STATE)
1335                 .show = a6xx_show,
1336 #endif
1337                 .gpu_busy = a6xx_gpu_busy,
1338                 .gpu_get_freq = a6xx_gmu_get_freq,
1339                 .gpu_set_freq = a6xx_gmu_set_freq,
1340 #if defined(CONFIG_DRM_MSM_GPU_STATE)
1341                 .gpu_state_get = a6xx_gpu_state_get,
1342                 .gpu_state_put = a6xx_gpu_state_put,
1343 #endif
1344                 .create_address_space = adreno_iommu_create_address_space,
1345                 .create_private_address_space = a6xx_create_private_address_space,
1346                 .get_rptr = a6xx_get_rptr,
1347         },
1348         .get_timestamp = a6xx_get_timestamp,
1349 };
1350
1351 struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
1352 {
1353         struct msm_drm_private *priv = dev->dev_private;
1354         struct platform_device *pdev = priv->gpu_pdev;
1355         struct adreno_platform_config *config = pdev->dev.platform_data;
1356         const struct adreno_info *info;
1357         struct device_node *node;
1358         struct a6xx_gpu *a6xx_gpu;
1359         struct adreno_gpu *adreno_gpu;
1360         struct msm_gpu *gpu;
1361         int ret;
1362
1363         a6xx_gpu = kzalloc(sizeof(*a6xx_gpu), GFP_KERNEL);
1364         if (!a6xx_gpu)
1365                 return ERR_PTR(-ENOMEM);
1366
1367         adreno_gpu = &a6xx_gpu->base;
1368         gpu = &adreno_gpu->base;
1369
1370         adreno_gpu->registers = NULL;
1371
1372         /*
1373          * We need to know the platform type before calling into adreno_gpu_init
1374          * so that the hw_apriv flag can be correctly set. Snoop into the info
1375          * and grab the revision number
1376          */
1377         info = adreno_info(config->rev);
1378
1379         if (info && info->revn == 650)
1380                 adreno_gpu->base.hw_apriv = true;
1381
1382         a6xx_llc_slices_init(pdev, a6xx_gpu);
1383
1384         ret = adreno_gpu_init(dev, pdev, adreno_gpu, &funcs, 1);
1385         if (ret) {
1386                 a6xx_destroy(&(a6xx_gpu->base.base));
1387                 return ERR_PTR(ret);
1388         }
1389
1390         /* Check if there is a GMU phandle and set it up */
1391         node = of_parse_phandle(pdev->dev.of_node, "qcom,gmu", 0);
1392
1393         /* FIXME: How do we gracefully handle this? */
1394         BUG_ON(!node);
1395
1396         ret = a6xx_gmu_init(a6xx_gpu, node);
1397         if (ret) {
1398                 a6xx_destroy(&(a6xx_gpu->base.base));
1399                 return ERR_PTR(ret);
1400         }
1401
1402         if (gpu->aspace)
1403                 msm_mmu_set_fault_handler(gpu->aspace->mmu, gpu,
1404                                 a6xx_fault_handler);
1405
1406         return gpu;
1407 }