1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #ifndef __AMDGPU_GFX_H__
25 #define __AMDGPU_GFX_H__
26 
27 /*
28  * GFX stuff
29  */
30 #include "clearstate_defs.h"
31 #include "amdgpu_ring.h"
32 #include "amdgpu_rlc.h"
33 
34 /* GFX current status */
35 #define AMDGPU_GFX_NORMAL_MODE			0x00000000L
36 #define AMDGPU_GFX_SAFE_MODE			0x00000001L
37 #define AMDGPU_GFX_PG_DISABLED_MODE		0x00000002L
38 #define AMDGPU_GFX_CG_DISABLED_MODE		0x00000004L
39 #define AMDGPU_GFX_LBPW_DISABLED_MODE		0x00000008L
40 
41 #define AMDGPU_MAX_GFX_QUEUES KGD_MAX_QUEUES
42 #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES
43 
44 enum gfx_pipe_priority {
45 	AMDGPU_GFX_PIPE_PRIO_NORMAL = 1,
46 	AMDGPU_GFX_PIPE_PRIO_HIGH,
47 	AMDGPU_GFX_PIPE_PRIO_MAX
48 };
49 
50 #define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM  0
51 #define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM  15
52 
53 struct amdgpu_mec {
54 	struct amdgpu_bo	*hpd_eop_obj;
55 	u64			hpd_eop_gpu_addr;
56 	struct amdgpu_bo	*mec_fw_obj;
57 	u64			mec_fw_gpu_addr;
58 	u32 num_mec;
59 	u32 num_pipe_per_mec;
60 	u32 num_queue_per_pipe;
61 	void			*mqd_backup[AMDGPU_MAX_COMPUTE_RINGS + 1];
62 
63 	/* These are the resources for which amdgpu takes ownership */
64 	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
65 };
66 
67 enum amdgpu_unmap_queues_action {
68 	PREEMPT_QUEUES = 0,
69 	RESET_QUEUES,
70 	DISABLE_PROCESS_QUEUES,
71 	PREEMPT_QUEUES_NO_UNMAP,
72 };
73 
74 struct kiq_pm4_funcs {
75 	/* Support ASIC-specific kiq pm4 packets*/
76 	void (*kiq_set_resources)(struct amdgpu_ring *kiq_ring,
77 					uint64_t queue_mask);
78 	void (*kiq_map_queues)(struct amdgpu_ring *kiq_ring,
79 					struct amdgpu_ring *ring);
80 	void (*kiq_unmap_queues)(struct amdgpu_ring *kiq_ring,
81 				 struct amdgpu_ring *ring,
82 				 enum amdgpu_unmap_queues_action action,
83 				 u64 gpu_addr, u64 seq);
84 	void (*kiq_query_status)(struct amdgpu_ring *kiq_ring,
85 					struct amdgpu_ring *ring,
86 					u64 addr,
87 					u64 seq);
88 	void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
89 				uint16_t pasid, uint32_t flush_type,
90 				bool all_hub);
91 	/* Packet sizes */
92 	int set_resources_size;
93 	int map_queues_size;
94 	int unmap_queues_size;
95 	int query_status_size;
96 	int invalidate_tlbs_size;
97 };
98 
99 struct amdgpu_kiq {
100 	u64			eop_gpu_addr;
101 	struct amdgpu_bo	*eop_obj;
102 	spinlock_t              ring_lock;
103 	struct amdgpu_ring	ring;
104 	struct amdgpu_irq_src	irq;
105 	const struct kiq_pm4_funcs *pmf;
106 	uint32_t			reg_val_offs;
107 };
108 
109 /*
110  * GPU scratch registers structures, functions & helpers
111  */
112 struct amdgpu_scratch {
113 	unsigned		num_reg;
114 	uint32_t                reg_base;
115 	uint32_t		free_mask;
116 };
117 
118 /*
119  * GFX configurations
120  */
121 #define AMDGPU_GFX_MAX_SE 4
122 #define AMDGPU_GFX_MAX_SH_PER_SE 2
123 
124 struct amdgpu_rb_config {
125 	uint32_t rb_backend_disable;
126 	uint32_t user_rb_backend_disable;
127 	uint32_t raster_config;
128 	uint32_t raster_config_1;
129 };
130 
131 struct gb_addr_config {
132 	uint16_t pipe_interleave_size;
133 	uint8_t num_pipes;
134 	uint8_t max_compress_frags;
135 	uint8_t num_banks;
136 	uint8_t num_se;
137 	uint8_t num_rb_per_se;
138 };
139 
140 struct amdgpu_gfx_config {
141 	unsigned max_shader_engines;
142 	unsigned max_tile_pipes;
143 	unsigned max_cu_per_sh;
144 	unsigned max_sh_per_se;
145 	unsigned max_backends_per_se;
146 	unsigned max_texture_channel_caches;
147 	unsigned max_gprs;
148 	unsigned max_gs_threads;
149 	unsigned max_hw_contexts;
150 	unsigned sc_prim_fifo_size_frontend;
151 	unsigned sc_prim_fifo_size_backend;
152 	unsigned sc_hiz_tile_fifo_size;
153 	unsigned sc_earlyz_tile_fifo_size;
154 
155 	unsigned num_tile_pipes;
156 	unsigned backend_enable_mask;
157 	unsigned mem_max_burst_length_bytes;
158 	unsigned mem_row_size_in_kb;
159 	unsigned shader_engine_tile_size;
160 	unsigned num_gpus;
161 	unsigned multi_gpu_tile_size;
162 	unsigned mc_arb_ramcfg;
163 	unsigned num_banks;
164 	unsigned num_ranks;
165 	unsigned gb_addr_config;
166 	unsigned num_rbs;
167 	unsigned gs_vgt_table_depth;
168 	unsigned gs_prim_buffer_depth;
169 
170 	uint32_t tile_mode_array[32];
171 	uint32_t macrotile_mode_array[16];
172 
173 	struct gb_addr_config gb_addr_config_fields;
174 	struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
175 
176 	/* gfx configure feature */
177 	uint32_t double_offchip_lds_buf;
178 	/* cached value of DB_DEBUG2 */
179 	uint32_t db_debug2;
180 	/* gfx10 specific config */
181 	uint32_t num_sc_per_sh;
182 	uint32_t num_packer_per_sc;
183 	uint32_t pa_sc_tile_steering_override;
184 	uint64_t tcc_disabled_mask;
185 };
186 
187 struct amdgpu_cu_info {
188 	uint32_t simd_per_cu;
189 	uint32_t max_waves_per_simd;
190 	uint32_t wave_front_size;
191 	uint32_t max_scratch_slots_per_cu;
192 	uint32_t lds_size;
193 
194 	/* total active CU number */
195 	uint32_t number;
196 	uint32_t ao_cu_mask;
197 	uint32_t ao_cu_bitmap[4][4];
198 	uint32_t bitmap[4][4];
199 };
200 
201 struct amdgpu_gfx_funcs {
202 	/* get the gpu clock counter */
203 	uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev);
204 	void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num,
205 			     u32 sh_num, u32 instance);
206 	void (*read_wave_data)(struct amdgpu_device *adev, uint32_t simd,
207 			       uint32_t wave, uint32_t *dst, int *no_fields);
208 	void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t simd,
209 				uint32_t wave, uint32_t thread, uint32_t start,
210 				uint32_t size, uint32_t *dst);
211 	void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t simd,
212 				uint32_t wave, uint32_t start, uint32_t size,
213 				uint32_t *dst);
214 	void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe,
215 				 u32 queue, u32 vmid);
216 	int (*ras_error_inject)(struct amdgpu_device *adev, void *inject_if);
217 	int (*query_ras_error_count) (struct amdgpu_device *adev, void *ras_error_status);
218 	void (*reset_ras_error_count) (struct amdgpu_device *adev);
219 };
220 
221 struct sq_work {
222 	struct work_struct	work;
223 	unsigned ih_data;
224 };
225 
226 struct amdgpu_pfp {
227 	struct amdgpu_bo		*pfp_fw_obj;
228 	uint64_t			pfp_fw_gpu_addr;
229 	uint32_t			*pfp_fw_ptr;
230 };
231 
232 struct amdgpu_ce {
233 	struct amdgpu_bo		*ce_fw_obj;
234 	uint64_t			ce_fw_gpu_addr;
235 	uint32_t			*ce_fw_ptr;
236 };
237 
238 struct amdgpu_me {
239 	struct amdgpu_bo		*me_fw_obj;
240 	uint64_t			me_fw_gpu_addr;
241 	uint32_t			*me_fw_ptr;
242 	uint32_t			num_me;
243 	uint32_t			num_pipe_per_me;
244 	uint32_t			num_queue_per_pipe;
245 	void				*mqd_backup[AMDGPU_MAX_GFX_RINGS];
246 
247 	/* These are the resources for which amdgpu takes ownership */
248 	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
249 };
250 
251 struct amdgpu_gfx {
252 	struct mutex			gpu_clock_mutex;
253 	struct amdgpu_gfx_config	config;
254 	struct amdgpu_rlc		rlc;
255 	struct amdgpu_pfp		pfp;
256 	struct amdgpu_ce		ce;
257 	struct amdgpu_me		me;
258 	struct amdgpu_mec		mec;
259 	struct amdgpu_kiq		kiq;
260 	struct amdgpu_scratch		scratch;
261 	const struct firmware		*me_fw;	/* ME firmware */
262 	uint32_t			me_fw_version;
263 	const struct firmware		*pfp_fw; /* PFP firmware */
264 	uint32_t			pfp_fw_version;
265 	const struct firmware		*ce_fw;	/* CE firmware */
266 	uint32_t			ce_fw_version;
267 	const struct firmware		*rlc_fw; /* RLC firmware */
268 	uint32_t			rlc_fw_version;
269 	const struct firmware		*mec_fw; /* MEC firmware */
270 	uint32_t			mec_fw_version;
271 	const struct firmware		*mec2_fw; /* MEC2 firmware */
272 	uint32_t			mec2_fw_version;
273 	uint32_t			me_feature_version;
274 	uint32_t			ce_feature_version;
275 	uint32_t			pfp_feature_version;
276 	uint32_t			rlc_feature_version;
277 	uint32_t			rlc_srlc_fw_version;
278 	uint32_t			rlc_srlc_feature_version;
279 	uint32_t			rlc_srlg_fw_version;
280 	uint32_t			rlc_srlg_feature_version;
281 	uint32_t			rlc_srls_fw_version;
282 	uint32_t			rlc_srls_feature_version;
283 	uint32_t			mec_feature_version;
284 	uint32_t			mec2_feature_version;
285 	bool				mec_fw_write_wait;
286 	bool				me_fw_write_wait;
287 	bool				cp_fw_write_wait;
288 	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
289 	struct drm_gpu_scheduler	*gfx_sched[AMDGPU_MAX_GFX_RINGS];
290 	uint32_t			num_gfx_sched;
291 	unsigned			num_gfx_rings;
292 	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
293 	struct drm_gpu_scheduler        **compute_prio_sched[AMDGPU_GFX_PIPE_PRIO_MAX];
294 	struct drm_gpu_scheduler	*compute_sched[AMDGPU_MAX_COMPUTE_RINGS];
295 	uint32_t                        num_compute_sched[AMDGPU_GFX_PIPE_PRIO_MAX];
296 	unsigned			num_compute_rings;
297 	struct amdgpu_irq_src		eop_irq;
298 	struct amdgpu_irq_src		priv_reg_irq;
299 	struct amdgpu_irq_src		priv_inst_irq;
300 	struct amdgpu_irq_src		cp_ecc_error_irq;
301 	struct amdgpu_irq_src		sq_irq;
302 	struct sq_work			sq_work;
303 
304 	/* gfx status */
305 	uint32_t			gfx_current_status;
306 	/* ce ram size*/
307 	unsigned			ce_ram_size;
308 	struct amdgpu_cu_info		cu_info;
309 	const struct amdgpu_gfx_funcs	*funcs;
310 
311 	/* reset mask */
312 	uint32_t                        grbm_soft_reset;
313 	uint32_t                        srbm_soft_reset;
314 
315 	/* gfx off */
316 	bool                            gfx_off_state; /* true: enabled, false: disabled */
317 	struct mutex                    gfx_off_mutex;
318 	uint32_t                        gfx_off_req_count; /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */
319 	struct delayed_work             gfx_off_delay_work;
320 
321 	/* pipe reservation */
322 	struct mutex			pipe_reserve_mutex;
323 	DECLARE_BITMAP			(pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
324 
325 	/*ras */
326 	struct ras_common_if		*ras_if;
327 };
328 
329 #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))
330 #define amdgpu_gfx_select_se_sh(adev, se, sh, instance) (adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance))
331 #define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid) (adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid))
332 
333 /**
334  * amdgpu_gfx_create_bitmask - create a bitmask
335  *
336  * @bit_width: length of the mask
337  *
338  * create a variable length bit mask.
339  * Returns the bitmask.
340  */
341 static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width)
342 {
343 	return (u32)((1ULL << bit_width) - 1);
344 }
345 
346 int amdgpu_gfx_scratch_get(struct amdgpu_device *adev, uint32_t *reg);
347 void amdgpu_gfx_scratch_free(struct amdgpu_device *adev, uint32_t reg);
348 
349 void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se,
350 				 unsigned max_sh);
351 
352 int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev,
353 			     struct amdgpu_ring *ring,
354 			     struct amdgpu_irq_src *irq);
355 
356 void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring);
357 
358 void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev);
359 int amdgpu_gfx_kiq_init(struct amdgpu_device *adev,
360 			unsigned hpd_size);
361 
362 int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
363 			   unsigned mqd_size);
364 void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev);
365 int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev);
366 int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev);
367 
368 void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev);
369 void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev);
370 
371 int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec,
372 				int pipe, int queue);
373 void amdgpu_gfx_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
374 				 int *mec, int *pipe, int *queue);
375 bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int mec,
376 				     int pipe, int queue);
377 bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev,
378 					       int queue);
379 int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev, int me,
380 			       int pipe, int queue);
381 void amdgpu_gfx_bit_to_me_queue(struct amdgpu_device *adev, int bit,
382 				int *me, int *pipe, int *queue);
383 bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me,
384 				    int pipe, int queue);
385 void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable);
386 int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev);
387 void amdgpu_gfx_ras_fini(struct amdgpu_device *adev);
388 int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev,
389 		void *err_data,
390 		struct amdgpu_iv_entry *entry);
391 int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
392 				  struct amdgpu_irq_src *source,
393 				  struct amdgpu_iv_entry *entry);
394 uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg);
395 void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v);
396 #endif
397