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