1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* Copyright (c) 2016-2017 The Linux Foundation. All rights reserved. 3 */ 4 #ifndef __A5XX_GPU_H__ 5 #define __A5XX_GPU_H__ 6 7 #include "adreno_gpu.h" 8 9 /* Bringing over the hack from the previous targets */ 10 #undef ROP_COPY 11 #undef ROP_XOR 12 13 #include "a5xx.xml.h" 14 15 struct a5xx_gpu { 16 struct adreno_gpu base; 17 18 struct drm_gem_object *pm4_bo; 19 uint64_t pm4_iova; 20 21 struct drm_gem_object *pfp_bo; 22 uint64_t pfp_iova; 23 24 struct drm_gem_object *gpmu_bo; 25 uint64_t gpmu_iova; 26 uint32_t gpmu_dwords; 27 28 uint32_t lm_leakage; 29 30 struct msm_ringbuffer *cur_ring; 31 struct msm_ringbuffer *next_ring; 32 33 struct drm_gem_object *preempt_bo[MSM_GPU_MAX_RINGS]; 34 struct drm_gem_object *preempt_counters_bo[MSM_GPU_MAX_RINGS]; 35 struct a5xx_preempt_record *preempt[MSM_GPU_MAX_RINGS]; 36 uint64_t preempt_iova[MSM_GPU_MAX_RINGS]; 37 uint32_t last_seqno[MSM_GPU_MAX_RINGS]; 38 39 uint64_t last_alu_cycles; 40 41 atomic_t preempt_state; 42 spinlock_t preempt_start_lock; 43 struct timer_list preempt_timer; 44 45 struct drm_gem_object *shadow_bo; 46 uint64_t shadow_iova; 47 uint32_t *shadow; 48 49 /* True if the microcode supports the WHERE_AM_I opcode */ 50 bool has_whereami; 51 }; 52 53 #define to_a5xx_gpu(x) container_of(x, struct a5xx_gpu, base) 54 55 #ifdef CONFIG_DEBUG_FS 56 void a5xx_debugfs_init(struct msm_gpu *gpu, struct drm_minor *minor); 57 #endif 58 59 /* 60 * In order to do lockless preemption we use a simple state machine to progress 61 * through the process. 62 * 63 * PREEMPT_NONE - no preemption in progress. Next state START. 64 * PREEMPT_START - The trigger is evaulating if preemption is possible. Next 65 * states: TRIGGERED, NONE 66 * PREEMPT_ABORT - An intermediate state before moving back to NONE. Next 67 * state: NONE. 68 * PREEMPT_TRIGGERED: A preemption has been executed on the hardware. Next 69 * states: FAULTED, PENDING 70 * PREEMPT_FAULTED: A preemption timed out (never completed). This will trigger 71 * recovery. Next state: N/A 72 * PREEMPT_PENDING: Preemption complete interrupt fired - the callback is 73 * checking the success of the operation. Next state: FAULTED, NONE. 74 */ 75 76 enum preempt_state { 77 PREEMPT_NONE = 0, 78 PREEMPT_START, 79 PREEMPT_ABORT, 80 PREEMPT_TRIGGERED, 81 PREEMPT_FAULTED, 82 PREEMPT_PENDING, 83 }; 84 85 /* 86 * struct a5xx_preempt_record is a shared buffer between the microcode and the 87 * CPU to store the state for preemption. The record itself is much larger 88 * (64k) but most of that is used by the CP for storage. 89 * 90 * There is a preemption record assigned per ringbuffer. When the CPU triggers a 91 * preemption, it fills out the record with the useful information (wptr, ring 92 * base, etc) and the microcode uses that information to set up the CP following 93 * the preemption. When a ring is switched out, the CP will save the ringbuffer 94 * state back to the record. In this way, once the records are properly set up 95 * the CPU can quickly switch back and forth between ringbuffers by only 96 * updating a few registers (often only the wptr). 97 * 98 * These are the CPU aware registers in the record: 99 * @magic: Must always be 0x27C4BAFC 100 * @info: Type of the record - written 0 by the CPU, updated by the CP 101 * @data: Data field from SET_RENDER_MODE or a checkpoint. Written and used by 102 * the CP 103 * @cntl: Value of RB_CNTL written by CPU, save/restored by CP 104 * @rptr: Value of RB_RPTR written by CPU, save/restored by CP 105 * @wptr: Value of RB_WPTR written by CPU, save/restored by CP 106 * @rptr_addr: Value of RB_RPTR_ADDR written by CPU, save/restored by CP 107 * @rbase: Value of RB_BASE written by CPU, save/restored by CP 108 * @counter: GPU address of the storage area for the performance counters 109 */ 110 struct a5xx_preempt_record { 111 uint32_t magic; 112 uint32_t info; 113 uint32_t data; 114 uint32_t cntl; 115 uint32_t rptr; 116 uint32_t wptr; 117 uint64_t rptr_addr; 118 uint64_t rbase; 119 uint64_t counter; 120 }; 121 122 /* Magic identifier for the preemption record */ 123 #define A5XX_PREEMPT_RECORD_MAGIC 0x27C4BAFCUL 124 125 /* 126 * Even though the structure above is only a few bytes, we need a full 64k to 127 * store the entire preemption record from the CP 128 */ 129 #define A5XX_PREEMPT_RECORD_SIZE (64 * 1024) 130 131 /* 132 * The preemption counter block is a storage area for the value of the 133 * preemption counters that are saved immediately before context switch. We 134 * append it on to the end of the allocation for the preemption record. 135 */ 136 #define A5XX_PREEMPT_COUNTER_SIZE (16 * 4) 137 138 extern const struct adreno_gpu_funcs a5xx_gpu_funcs; 139 140 int a5xx_power_init(struct msm_gpu *gpu); 141 void a5xx_gpmu_ucode_init(struct msm_gpu *gpu); 142 143 static inline int spin_usecs(struct msm_gpu *gpu, uint32_t usecs, 144 uint32_t reg, uint32_t mask, uint32_t value) 145 { 146 while (usecs--) { 147 udelay(1); 148 if ((gpu_read(gpu, reg) & mask) == value) 149 return 0; 150 cpu_relax(); 151 } 152 153 return -ETIMEDOUT; 154 } 155 156 #define shadowptr(a5xx_gpu, ring) ((a5xx_gpu)->shadow_iova + \ 157 ((ring)->id * sizeof(uint32_t))) 158 159 bool a5xx_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring); 160 void a5xx_set_hwcg(struct msm_gpu *gpu, bool state); 161 162 void a5xx_preempt_init(struct msm_gpu *gpu); 163 void a5xx_preempt_hw_init(struct msm_gpu *gpu); 164 void a5xx_preempt_trigger(struct msm_gpu *gpu); 165 void a5xx_preempt_irq(struct msm_gpu *gpu); 166 void a5xx_preempt_fini(struct msm_gpu *gpu); 167 168 void a5xx_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring, bool sync); 169 170 /* Return true if we are in a preempt state */ 171 static inline bool a5xx_in_preempt(struct a5xx_gpu *a5xx_gpu) 172 { 173 int preempt_state = atomic_read(&a5xx_gpu->preempt_state); 174 175 return !(preempt_state == PREEMPT_NONE || 176 preempt_state == PREEMPT_ABORT); 177 } 178 179 #endif /* __A5XX_GPU_H__ */ 180