xref: /linux/drivers/gpu/drm/msm/adreno/a6xx_gpu.h (revision 68a052239fc4b351e961f698b824f7654a346091)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Copyright (c) 2017, 2019 The Linux Foundation. All rights reserved. */
3 
4 #ifndef __A6XX_GPU_H__
5 #define __A6XX_GPU_H__
6 
7 
8 #include "adreno_gpu.h"
9 #include "a6xx_enums.xml.h"
10 #include "a7xx_enums.xml.h"
11 #include "a6xx_perfcntrs.xml.h"
12 #include "a7xx_perfcntrs.xml.h"
13 #include "a6xx.xml.h"
14 
15 #include "a6xx_gmu.h"
16 
17 extern bool hang_debug;
18 
19 struct cpu_gpu_lock {
20 	uint32_t gpu_req;
21 	uint32_t cpu_req;
22 	uint32_t turn;
23 	union {
24 		struct {
25 			uint16_t list_length;
26 			uint16_t list_offset;
27 		};
28 		struct {
29 			uint8_t ifpc_list_len;
30 			uint8_t preemption_list_len;
31 			uint16_t dynamic_list_len;
32 		};
33 	};
34 	uint64_t regs[62];
35 };
36 
37 /**
38  * struct a6xx_info - a6xx specific information from device table
39  *
40  * @hwcg: hw clock gating register sequence
41  * @protect: CP_PROTECT settings
42  * @pwrup_reglist pwrup reglist for preemption
43  */
44 struct a6xx_info {
45 	const struct adreno_reglist *hwcg;
46 	const struct adreno_protect *protect;
47 	const struct adreno_reglist_list *pwrup_reglist;
48 	const struct adreno_reglist_list *ifpc_reglist;
49 	u32 gmu_chipid;
50 	u32 gmu_cgc_mode;
51 	u32 prim_fifo_threshold;
52 	const struct a6xx_bcm *bcms;
53 };
54 
55 struct a6xx_gpu {
56 	struct adreno_gpu base;
57 
58 	struct drm_gem_object *sqe_bo;
59 	uint64_t sqe_iova;
60 
61 	struct msm_ringbuffer *cur_ring;
62 	struct msm_ringbuffer *next_ring;
63 
64 	struct drm_gem_object *preempt_bo[MSM_GPU_MAX_RINGS];
65 	void *preempt[MSM_GPU_MAX_RINGS];
66 	uint64_t preempt_iova[MSM_GPU_MAX_RINGS];
67 	struct drm_gem_object *preempt_smmu_bo[MSM_GPU_MAX_RINGS];
68 	void *preempt_smmu[MSM_GPU_MAX_RINGS];
69 	uint64_t preempt_smmu_iova[MSM_GPU_MAX_RINGS];
70 	uint32_t last_seqno[MSM_GPU_MAX_RINGS];
71 
72 	atomic_t preempt_state;
73 	spinlock_t eval_lock;
74 	struct timer_list preempt_timer;
75 
76 	unsigned int preempt_level;
77 	bool uses_gmem;
78 	bool skip_save_restore;
79 
80 	struct drm_gem_object *preempt_postamble_bo;
81 	void *preempt_postamble_ptr;
82 	uint64_t preempt_postamble_iova;
83 	uint64_t preempt_postamble_len;
84 	bool postamble_enabled;
85 
86 	struct a6xx_gmu gmu;
87 
88 	struct drm_gem_object *shadow_bo;
89 	uint64_t shadow_iova;
90 	uint32_t *shadow;
91 
92 	struct drm_gem_object *pwrup_reglist_bo;
93 	void *pwrup_reglist_ptr;
94 	uint64_t pwrup_reglist_iova;
95 	bool pwrup_reglist_emitted;
96 
97 	bool has_whereami;
98 
99 	void __iomem *llc_mmio;
100 	void *llc_slice;
101 	void *htw_llc_slice;
102 	bool have_mmu500;
103 	bool hung;
104 };
105 
106 #define to_a6xx_gpu(x) container_of(x, struct a6xx_gpu, base)
107 
108 /*
109  * In order to do lockless preemption we use a simple state machine to progress
110  * through the process.
111  *
112  * PREEMPT_NONE - no preemption in progress.  Next state START.
113  * PREEMPT_START - The trigger is evaluating if preemption is possible. Next
114  * states: TRIGGERED, NONE
115  * PREEMPT_FINISH - An intermediate state before moving back to NONE. Next
116  * state: NONE.
117  * PREEMPT_TRIGGERED: A preemption has been executed on the hardware. Next
118  * states: FAULTED, PENDING
119  * PREEMPT_FAULTED: A preemption timed out (never completed). This will trigger
120  * recovery.  Next state: N/A
121  * PREEMPT_PENDING: Preemption complete interrupt fired - the callback is
122  * checking the success of the operation. Next state: FAULTED, NONE.
123  */
124 
125 enum a6xx_preempt_state {
126 	PREEMPT_NONE = 0,
127 	PREEMPT_START,
128 	PREEMPT_FINISH,
129 	PREEMPT_TRIGGERED,
130 	PREEMPT_FAULTED,
131 	PREEMPT_PENDING,
132 };
133 
134 /*
135  * struct a6xx_preempt_record is a shared buffer between the microcode and the
136  * CPU to store the state for preemption. The record itself is much larger
137  * (2112k) but most of that is used by the CP for storage.
138  *
139  * There is a preemption record assigned per ringbuffer. When the CPU triggers a
140  * preemption, it fills out the record with the useful information (wptr, ring
141  * base, etc) and the microcode uses that information to set up the CP following
142  * the preemption.  When a ring is switched out, the CP will save the ringbuffer
143  * state back to the record. In this way, once the records are properly set up
144  * the CPU can quickly switch back and forth between ringbuffers by only
145  * updating a few registers (often only the wptr).
146  *
147  * These are the CPU aware registers in the record:
148  * @magic: Must always be 0xAE399D6EUL
149  * @info: Type of the record - written 0 by the CPU, updated by the CP
150  * @errno: preemption error record
151  * @data: Data field in YIELD and SET_MARKER packets, Written and used by CP
152  * @cntl: Value of RB_CNTL written by CPU, save/restored by CP
153  * @rptr: Value of RB_RPTR written by CPU, save/restored by CP
154  * @wptr: Value of RB_WPTR written by CPU, save/restored by CP
155  * @_pad: Reserved/padding
156  * @rptr_addr: Value of RB_RPTR_ADDR_LO|HI written by CPU, save/restored by CP
157  * @rbase: Value of RB_BASE written by CPU, save/restored by CP
158  * @counter: GPU address of the storage area for the preemption counters
159  * @bv_rptr_addr: Value of BV_RB_RPTR_ADDR_LO|HI written by CPU, save/restored by CP
160  */
161 struct a6xx_preempt_record {
162 	u32 magic;
163 	u32 info;
164 	u32 errno;
165 	u32 data;
166 	u32 cntl;
167 	u32 rptr;
168 	u32 wptr;
169 	u32 _pad;
170 	u64 rptr_addr;
171 	u64 rbase;
172 	u64 counter;
173 	u64 bv_rptr_addr;
174 };
175 
176 #define A6XX_PREEMPT_RECORD_MAGIC 0xAE399D6EUL
177 
178 #define PREEMPT_SMMU_INFO_SIZE 4096
179 
180 #define PREEMPT_RECORD_SIZE(adreno_gpu) \
181 	((adreno_gpu->info->preempt_record_size) == 0 ? \
182 	 4192 * SZ_1K : (adreno_gpu->info->preempt_record_size))
183 
184 /*
185  * The preemption counter block is a storage area for the value of the
186  * preemption counters that are saved immediately before context switch. We
187  * append it on to the end of the allocation for the preemption record.
188  */
189 #define A6XX_PREEMPT_COUNTER_SIZE (16 * 4)
190 
191 struct a7xx_cp_smmu_info {
192 	u32 magic;
193 	u32 _pad4;
194 	u64 ttbr0;
195 	u32 asid;
196 	u32 context_idr;
197 	u32 context_bank;
198 };
199 
200 #define GEN7_CP_SMMU_INFO_MAGIC 0x241350d5UL
201 
202 /*
203  * Given a register and a count, return a value to program into
204  * REG_CP_PROTECT_REG(n) - this will block both reads and writes for
205  * _len + 1 registers starting at _reg.
206  */
207 #define A6XX_PROTECT_NORDWR(_reg, _len) \
208 	((1 << 31) | \
209 	(((_len) & 0x3FFF) << 18) | ((_reg) & 0x3FFFF))
210 
211 /*
212  * Same as above, but allow reads over the range. For areas of mixed use (such
213  * as performance counters) this allows us to protect a much larger range with a
214  * single register
215  */
216 #define A6XX_PROTECT_RDONLY(_reg, _len) \
217 	((((_len) & 0x3FFF) << 18) | ((_reg) & 0x3FFFF))
218 
219 static inline bool a6xx_has_gbif(struct adreno_gpu *gpu)
220 {
221 	if(adreno_is_a630(gpu))
222 		return false;
223 
224 	return true;
225 }
226 
227 static inline void a6xx_llc_rmw(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 mask, u32 or)
228 {
229 	return msm_rmw(a6xx_gpu->llc_mmio + (reg << 2), mask, or);
230 }
231 
232 static inline u32 a6xx_llc_read(struct a6xx_gpu *a6xx_gpu, u32 reg)
233 {
234 	return readl(a6xx_gpu->llc_mmio + (reg << 2));
235 }
236 
237 static inline void a6xx_llc_write(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 value)
238 {
239 	writel(value, a6xx_gpu->llc_mmio + (reg << 2));
240 }
241 
242 #define shadowptr(_a6xx_gpu, _ring) ((_a6xx_gpu)->shadow_iova + \
243 		((_ring)->id * sizeof(uint32_t)))
244 
245 int a6xx_gmu_resume(struct a6xx_gpu *gpu);
246 int a6xx_gmu_stop(struct a6xx_gpu *gpu);
247 
248 int a6xx_gmu_wait_for_idle(struct a6xx_gmu *gmu);
249 
250 bool a6xx_gmu_isidle(struct a6xx_gmu *gmu);
251 
252 int a6xx_gmu_set_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state);
253 void a6xx_gmu_clear_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state);
254 
255 int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node);
256 int a6xx_gmu_wrapper_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node);
257 void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu);
258 void a6xx_gmu_sysprof_setup(struct msm_gpu *gpu);
259 
260 void a6xx_preempt_init(struct msm_gpu *gpu);
261 void a6xx_preempt_hw_init(struct msm_gpu *gpu);
262 void a6xx_preempt_trigger(struct msm_gpu *gpu);
263 void a6xx_preempt_irq(struct msm_gpu *gpu);
264 void a6xx_preempt_fini(struct msm_gpu *gpu);
265 int a6xx_preempt_submitqueue_setup(struct msm_gpu *gpu,
266 		struct msm_gpu_submitqueue *queue);
267 void a6xx_preempt_submitqueue_close(struct msm_gpu *gpu,
268 		struct msm_gpu_submitqueue *queue);
269 
270 /* Return true if we are in a preempt state */
271 static inline bool a6xx_in_preempt(struct a6xx_gpu *a6xx_gpu)
272 {
273 	/*
274 	 * Make sure the read to preempt_state is ordered with respect to reads
275 	 * of other variables before ...
276 	 */
277 	smp_rmb();
278 
279 	int preempt_state = atomic_read(&a6xx_gpu->preempt_state);
280 
281 	/* ... and after. */
282 	smp_rmb();
283 
284 	return !(preempt_state == PREEMPT_NONE ||
285 			preempt_state == PREEMPT_FINISH);
286 }
287 
288 void a6xx_gmu_set_freq(struct msm_gpu *gpu, struct dev_pm_opp *opp,
289 		       bool suspended);
290 unsigned long a6xx_gmu_get_freq(struct msm_gpu *gpu);
291 
292 void a6xx_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
293 		struct drm_printer *p);
294 
295 struct msm_gpu_state *a6xx_gpu_state_get(struct msm_gpu *gpu);
296 int a6xx_gpu_state_put(struct msm_gpu_state *state);
297 
298 void a6xx_bus_clear_pending_transactions(struct adreno_gpu *adreno_gpu, bool gx_off);
299 void a6xx_gpu_sw_reset(struct msm_gpu *gpu, bool assert);
300 int a6xx_fenced_write(struct a6xx_gpu *gpu, u32 offset, u64 value, u32 mask, bool is_64b);
301 
302 #endif /* __A6XX_GPU_H__ */
303