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 #include "amdgpu_imu.h"
34 #include "soc15.h"
35 #include "amdgpu_ras.h"
36 
37 /* GFX current status */
38 #define AMDGPU_GFX_NORMAL_MODE			0x00000000L
39 #define AMDGPU_GFX_SAFE_MODE			0x00000001L
40 #define AMDGPU_GFX_PG_DISABLED_MODE		0x00000002L
41 #define AMDGPU_GFX_CG_DISABLED_MODE		0x00000004L
42 #define AMDGPU_GFX_LBPW_DISABLED_MODE		0x00000008L
43 
44 #define AMDGPU_MAX_GFX_QUEUES KGD_MAX_QUEUES
45 #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES
46 
47 enum amdgpu_gfx_pipe_priority {
48 	AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1,
49 	AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2
50 };
51 
52 #define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM  0
53 #define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM  15
54 
55 struct amdgpu_mec {
56 	struct amdgpu_bo	*hpd_eop_obj;
57 	u64			hpd_eop_gpu_addr;
58 	struct amdgpu_bo	*mec_fw_obj;
59 	u64			mec_fw_gpu_addr;
60 	struct amdgpu_bo	*mec_fw_data_obj;
61 	u64			mec_fw_data_gpu_addr;
62 
63 	u32 num_mec;
64 	u32 num_pipe_per_mec;
65 	u32 num_queue_per_pipe;
66 	void			*mqd_backup[AMDGPU_MAX_COMPUTE_RINGS + 1];
67 
68 	/* These are the resources for which amdgpu takes ownership */
69 	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
70 };
71 
72 enum amdgpu_unmap_queues_action {
73 	PREEMPT_QUEUES = 0,
74 	RESET_QUEUES,
75 	DISABLE_PROCESS_QUEUES,
76 	PREEMPT_QUEUES_NO_UNMAP,
77 };
78 
79 struct kiq_pm4_funcs {
80 	/* Support ASIC-specific kiq pm4 packets*/
81 	void (*kiq_set_resources)(struct amdgpu_ring *kiq_ring,
82 					uint64_t queue_mask);
83 	void (*kiq_map_queues)(struct amdgpu_ring *kiq_ring,
84 					struct amdgpu_ring *ring);
85 	void (*kiq_unmap_queues)(struct amdgpu_ring *kiq_ring,
86 				 struct amdgpu_ring *ring,
87 				 enum amdgpu_unmap_queues_action action,
88 				 u64 gpu_addr, u64 seq);
89 	void (*kiq_query_status)(struct amdgpu_ring *kiq_ring,
90 					struct amdgpu_ring *ring,
91 					u64 addr,
92 					u64 seq);
93 	void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
94 				uint16_t pasid, uint32_t flush_type,
95 				bool all_hub);
96 	/* Packet sizes */
97 	int set_resources_size;
98 	int map_queues_size;
99 	int unmap_queues_size;
100 	int query_status_size;
101 	int invalidate_tlbs_size;
102 };
103 
104 struct amdgpu_kiq {
105 	u64			eop_gpu_addr;
106 	struct amdgpu_bo	*eop_obj;
107 	spinlock_t              ring_lock;
108 	struct amdgpu_ring	ring;
109 	struct amdgpu_irq_src	irq;
110 	const struct kiq_pm4_funcs *pmf;
111 };
112 
113 /*
114  * GFX configurations
115  */
116 #define AMDGPU_GFX_MAX_SE 4
117 #define AMDGPU_GFX_MAX_SH_PER_SE 2
118 
119 struct amdgpu_rb_config {
120 	uint32_t rb_backend_disable;
121 	uint32_t user_rb_backend_disable;
122 	uint32_t raster_config;
123 	uint32_t raster_config_1;
124 };
125 
126 struct gb_addr_config {
127 	uint16_t pipe_interleave_size;
128 	uint8_t num_pipes;
129 	uint8_t max_compress_frags;
130 	uint8_t num_banks;
131 	uint8_t num_se;
132 	uint8_t num_rb_per_se;
133 	uint8_t num_pkrs;
134 };
135 
136 struct amdgpu_gfx_config {
137 	unsigned max_shader_engines;
138 	unsigned max_tile_pipes;
139 	unsigned max_cu_per_sh;
140 	unsigned max_sh_per_se;
141 	unsigned max_backends_per_se;
142 	unsigned max_texture_channel_caches;
143 	unsigned max_gprs;
144 	unsigned max_gs_threads;
145 	unsigned max_hw_contexts;
146 	unsigned sc_prim_fifo_size_frontend;
147 	unsigned sc_prim_fifo_size_backend;
148 	unsigned sc_hiz_tile_fifo_size;
149 	unsigned sc_earlyz_tile_fifo_size;
150 
151 	unsigned num_tile_pipes;
152 	unsigned backend_enable_mask;
153 	unsigned mem_max_burst_length_bytes;
154 	unsigned mem_row_size_in_kb;
155 	unsigned shader_engine_tile_size;
156 	unsigned num_gpus;
157 	unsigned multi_gpu_tile_size;
158 	unsigned mc_arb_ramcfg;
159 	unsigned num_banks;
160 	unsigned num_ranks;
161 	unsigned gb_addr_config;
162 	unsigned num_rbs;
163 	unsigned gs_vgt_table_depth;
164 	unsigned gs_prim_buffer_depth;
165 
166 	uint32_t tile_mode_array[32];
167 	uint32_t macrotile_mode_array[16];
168 
169 	struct gb_addr_config gb_addr_config_fields;
170 	struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
171 
172 	/* gfx configure feature */
173 	uint32_t double_offchip_lds_buf;
174 	/* cached value of DB_DEBUG2 */
175 	uint32_t db_debug2;
176 	/* gfx10 specific config */
177 	uint32_t num_sc_per_sh;
178 	uint32_t num_packer_per_sc;
179 	uint32_t pa_sc_tile_steering_override;
180 	uint64_t tcc_disabled_mask;
181 	uint32_t gc_num_tcp_per_sa;
182 	uint32_t gc_num_sdp_interface;
183 	uint32_t gc_num_tcps;
184 	uint32_t gc_num_tcp_per_wpg;
185 	uint32_t gc_tcp_l1_size;
186 	uint32_t gc_num_sqc_per_wgp;
187 	uint32_t gc_l1_instruction_cache_size_per_sqc;
188 	uint32_t gc_l1_data_cache_size_per_sqc;
189 	uint32_t gc_gl1c_per_sa;
190 	uint32_t gc_gl1c_size_per_instance;
191 	uint32_t gc_gl2c_per_gpu;
192 };
193 
194 struct amdgpu_cu_info {
195 	uint32_t simd_per_cu;
196 	uint32_t max_waves_per_simd;
197 	uint32_t wave_front_size;
198 	uint32_t max_scratch_slots_per_cu;
199 	uint32_t lds_size;
200 
201 	/* total active CU number */
202 	uint32_t number;
203 	uint32_t ao_cu_mask;
204 	uint32_t ao_cu_bitmap[4][4];
205 	uint32_t bitmap[4][4];
206 };
207 
208 struct amdgpu_gfx_ras {
209 	struct amdgpu_ras_block_object  ras_block;
210 	void (*enable_watchdog_timer)(struct amdgpu_device *adev);
211 	bool (*query_utcl2_poison_status)(struct amdgpu_device *adev);
212 };
213 
214 struct amdgpu_gfx_funcs {
215 	/* get the gpu clock counter */
216 	uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev);
217 	void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num,
218 			     u32 sh_num, u32 instance);
219 	void (*read_wave_data)(struct amdgpu_device *adev, uint32_t simd,
220 			       uint32_t wave, uint32_t *dst, int *no_fields);
221 	void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t simd,
222 				uint32_t wave, uint32_t thread, uint32_t start,
223 				uint32_t size, uint32_t *dst);
224 	void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t simd,
225 				uint32_t wave, uint32_t start, uint32_t size,
226 				uint32_t *dst);
227 	void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe,
228 				 u32 queue, u32 vmid);
229 	void (*init_spm_golden)(struct amdgpu_device *adev);
230 	void (*update_perfmon_mgcg)(struct amdgpu_device *adev, bool enable);
231 };
232 
233 struct sq_work {
234 	struct work_struct	work;
235 	unsigned ih_data;
236 };
237 
238 struct amdgpu_pfp {
239 	struct amdgpu_bo		*pfp_fw_obj;
240 	uint64_t			pfp_fw_gpu_addr;
241 	uint32_t			*pfp_fw_ptr;
242 
243 	struct amdgpu_bo		*pfp_fw_data_obj;
244 	uint64_t			pfp_fw_data_gpu_addr;
245 	uint32_t			*pfp_fw_data_ptr;
246 };
247 
248 struct amdgpu_ce {
249 	struct amdgpu_bo		*ce_fw_obj;
250 	uint64_t			ce_fw_gpu_addr;
251 	uint32_t			*ce_fw_ptr;
252 };
253 
254 struct amdgpu_me {
255 	struct amdgpu_bo		*me_fw_obj;
256 	uint64_t			me_fw_gpu_addr;
257 	uint32_t			*me_fw_ptr;
258 
259 	struct amdgpu_bo		*me_fw_data_obj;
260 	uint64_t			me_fw_data_gpu_addr;
261 	uint32_t			*me_fw_data_ptr;
262 
263 	uint32_t			num_me;
264 	uint32_t			num_pipe_per_me;
265 	uint32_t			num_queue_per_pipe;
266 	void				*mqd_backup[AMDGPU_MAX_GFX_RINGS];
267 
268 	/* These are the resources for which amdgpu takes ownership */
269 	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
270 };
271 
272 struct amdgpu_gfx {
273 	struct mutex			gpu_clock_mutex;
274 	struct amdgpu_gfx_config	config;
275 	struct amdgpu_rlc		rlc;
276 	struct amdgpu_pfp		pfp;
277 	struct amdgpu_ce		ce;
278 	struct amdgpu_me		me;
279 	struct amdgpu_mec		mec;
280 	struct amdgpu_kiq		kiq;
281 	struct amdgpu_imu		imu;
282 	bool				rs64_enable; /* firmware format */
283 	const struct firmware		*me_fw;	/* ME firmware */
284 	uint32_t			me_fw_version;
285 	const struct firmware		*pfp_fw; /* PFP firmware */
286 	uint32_t			pfp_fw_version;
287 	const struct firmware		*ce_fw;	/* CE firmware */
288 	uint32_t			ce_fw_version;
289 	const struct firmware		*rlc_fw; /* RLC firmware */
290 	uint32_t			rlc_fw_version;
291 	const struct firmware		*mec_fw; /* MEC firmware */
292 	uint32_t			mec_fw_version;
293 	const struct firmware		*mec2_fw; /* MEC2 firmware */
294 	uint32_t			mec2_fw_version;
295 	const struct firmware		*imu_fw; /* IMU firmware */
296 	uint32_t			imu_fw_version;
297 	uint32_t			me_feature_version;
298 	uint32_t			ce_feature_version;
299 	uint32_t			pfp_feature_version;
300 	uint32_t			rlc_feature_version;
301 	uint32_t			rlc_srlc_fw_version;
302 	uint32_t			rlc_srlc_feature_version;
303 	uint32_t			rlc_srlg_fw_version;
304 	uint32_t			rlc_srlg_feature_version;
305 	uint32_t			rlc_srls_fw_version;
306 	uint32_t			rlc_srls_feature_version;
307 	uint32_t			rlcp_ucode_version;
308 	uint32_t			rlcp_ucode_feature_version;
309 	uint32_t			rlcv_ucode_version;
310 	uint32_t			rlcv_ucode_feature_version;
311 	uint32_t			mec_feature_version;
312 	uint32_t			mec2_feature_version;
313 	bool				mec_fw_write_wait;
314 	bool				me_fw_write_wait;
315 	bool				cp_fw_write_wait;
316 	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
317 	unsigned			num_gfx_rings;
318 	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
319 	unsigned			num_compute_rings;
320 	struct amdgpu_irq_src		eop_irq;
321 	struct amdgpu_irq_src		priv_reg_irq;
322 	struct amdgpu_irq_src		priv_inst_irq;
323 	struct amdgpu_irq_src		cp_ecc_error_irq;
324 	struct amdgpu_irq_src		sq_irq;
325 	struct sq_work			sq_work;
326 
327 	/* gfx status */
328 	uint32_t			gfx_current_status;
329 	/* ce ram size*/
330 	unsigned			ce_ram_size;
331 	struct amdgpu_cu_info		cu_info;
332 	const struct amdgpu_gfx_funcs	*funcs;
333 
334 	/* reset mask */
335 	uint32_t                        grbm_soft_reset;
336 	uint32_t                        srbm_soft_reset;
337 
338 	/* gfx off */
339 	bool                            gfx_off_state;      /* true: enabled, false: disabled */
340 	struct mutex                    gfx_off_mutex;      /* mutex to change gfxoff state */
341 	uint32_t                        gfx_off_req_count;  /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */
342 	struct delayed_work             gfx_off_delay_work; /* async work to set gfx block off */
343 	uint32_t                        gfx_off_residency;  /* last logged residency */
344 	uint64_t                        gfx_off_entrycount; /* count of times GPU has get into GFXOFF state */
345 
346 	/* pipe reservation */
347 	struct mutex			pipe_reserve_mutex;
348 	DECLARE_BITMAP			(pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
349 
350 	/*ras */
351 	struct ras_common_if		*ras_if;
352 	struct amdgpu_gfx_ras		*ras;
353 
354 	bool				is_poweron;
355 };
356 
357 #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))
358 #define amdgpu_gfx_select_se_sh(adev, se, sh, instance) (adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance))
359 #define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid) (adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid))
360 #define amdgpu_gfx_init_spm_golden(adev) (adev)->gfx.funcs->init_spm_golden((adev))
361 
362 /**
363  * amdgpu_gfx_create_bitmask - create a bitmask
364  *
365  * @bit_width: length of the mask
366  *
367  * create a variable length bit mask.
368  * Returns the bitmask.
369  */
370 static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width)
371 {
372 	return (u32)((1ULL << bit_width) - 1);
373 }
374 
375 void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se,
376 				 unsigned max_sh);
377 
378 int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev,
379 			     struct amdgpu_ring *ring,
380 			     struct amdgpu_irq_src *irq);
381 
382 void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring);
383 
384 void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev);
385 int amdgpu_gfx_kiq_init(struct amdgpu_device *adev,
386 			unsigned hpd_size);
387 
388 int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
389 			   unsigned mqd_size);
390 void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev);
391 int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev);
392 int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev);
393 
394 void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev);
395 void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev);
396 
397 int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec,
398 				int pipe, int queue);
399 void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
400 				 int *mec, int *pipe, int *queue);
401 bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int mec,
402 				     int pipe, int queue);
403 bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev,
404 					       struct amdgpu_ring *ring);
405 bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev,
406 						struct amdgpu_ring *ring);
407 int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev, int me,
408 			       int pipe, int queue);
409 void amdgpu_gfx_bit_to_me_queue(struct amdgpu_device *adev, int bit,
410 				int *me, int *pipe, int *queue);
411 bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me,
412 				    int pipe, int queue);
413 void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable);
414 int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value);
415 int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block);
416 void amdgpu_gfx_ras_fini(struct amdgpu_device *adev);
417 int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value);
418 int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *residency);
419 int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value);
420 int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev,
421 		void *err_data,
422 		struct amdgpu_iv_entry *entry);
423 int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
424 				  struct amdgpu_irq_src *source,
425 				  struct amdgpu_iv_entry *entry);
426 uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg);
427 void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v);
428 int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev);
429 void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id);
430 
431 #endif
432