xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h (revision 570f7e331f5febb30f1384817463c7e42b65ca7d)
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 #include "amdgpu_ring_mux.h"
37 #include "amdgpu_xcp.h"
38 
39 struct amdgpu_usermode_queue;
40 
41 /* GFX current status */
42 #define AMDGPU_GFX_NORMAL_MODE			0x00000000L
43 #define AMDGPU_GFX_SAFE_MODE			0x00000001L
44 #define AMDGPU_GFX_PG_DISABLED_MODE		0x00000002L
45 #define AMDGPU_GFX_CG_DISABLED_MODE		0x00000004L
46 #define AMDGPU_GFX_LBPW_DISABLED_MODE		0x00000008L
47 
48 #define AMDGPU_MAX_GC_INSTANCES		8
49 #define AMDGPU_MAX_QUEUES		128
50 
51 #define AMDGPU_MAX_GFX_QUEUES AMDGPU_MAX_QUEUES
52 #define AMDGPU_MAX_COMPUTE_QUEUES AMDGPU_MAX_QUEUES
53 
54 enum amdgpu_gfx_pipe_priority {
55 	AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1,
56 	AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2
57 };
58 
59 #define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM  0
60 #define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM  15
61 
62 /* 1 second timeout */
63 #define GFX_PROFILE_IDLE_TIMEOUT	msecs_to_jiffies(1000)
64 
65 enum amdgpu_gfx_partition {
66 	AMDGPU_SPX_PARTITION_MODE = 0,
67 	AMDGPU_DPX_PARTITION_MODE = 1,
68 	AMDGPU_TPX_PARTITION_MODE = 2,
69 	AMDGPU_QPX_PARTITION_MODE = 3,
70 	AMDGPU_CPX_PARTITION_MODE = 4,
71 	AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE = -1,
72 	/* Automatically choose the right mode */
73 	AMDGPU_AUTO_COMPUTE_PARTITION_MODE = -2,
74 };
75 
76 enum amdgpu_gfx_partition_mem_alloc_mode {
77 	AMDGPU_PARTITION_MEM_CAPPING_EVEN = 0,
78 	AMDGPU_PARTITION_MEM_ALLOC_ALL  = 1,
79 };
80 
81 #define NUM_XCC(x) hweight16(x)
82 
83 enum amdgpu_gfx_ras_mem_id_type {
84 	AMDGPU_GFX_CP_MEM = 0,
85 	AMDGPU_GFX_GCEA_MEM,
86 	AMDGPU_GFX_GC_CANE_MEM,
87 	AMDGPU_GFX_GCUTCL2_MEM,
88 	AMDGPU_GFX_GDS_MEM,
89 	AMDGPU_GFX_LDS_MEM,
90 	AMDGPU_GFX_RLC_MEM,
91 	AMDGPU_GFX_SP_MEM,
92 	AMDGPU_GFX_SPI_MEM,
93 	AMDGPU_GFX_SQC_MEM,
94 	AMDGPU_GFX_SQ_MEM,
95 	AMDGPU_GFX_TA_MEM,
96 	AMDGPU_GFX_TCC_MEM,
97 	AMDGPU_GFX_TCA_MEM,
98 	AMDGPU_GFX_TCI_MEM,
99 	AMDGPU_GFX_TCP_MEM,
100 	AMDGPU_GFX_TD_MEM,
101 	AMDGPU_GFX_TCX_MEM,
102 	AMDGPU_GFX_ATC_L2_MEM,
103 	AMDGPU_GFX_UTCL2_MEM,
104 	AMDGPU_GFX_VML2_MEM,
105 	AMDGPU_GFX_VML2_WALKER_MEM,
106 	AMDGPU_GFX_MEM_TYPE_NUM
107 };
108 
109 struct amdgpu_mec {
110 	struct amdgpu_bo	*hpd_eop_obj;
111 	u64			hpd_eop_gpu_addr;
112 	struct amdgpu_bo	*mec_fw_obj;
113 	u64			mec_fw_gpu_addr;
114 	struct amdgpu_bo	*mec_fw_data_obj;
115 	u64			mec_fw_data_gpu_addr;
116 
117 	u32 num_mec;
118 	u32 num_pipe_per_mec;
119 	u32 num_queue_per_pipe;
120 	void			*mqd_backup[AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES];
121 	bool use_mmio_for_reset;
122 	u32 *mes_hung_db_array;
123 	struct mutex		reset_mutex;
124 };
125 
126 struct amdgpu_mec_bitmap {
127 	/* These are the resources for which amdgpu takes ownership */
128 	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
129 };
130 
131 enum amdgpu_unmap_queues_action {
132 	PREEMPT_QUEUES = 0,
133 	RESET_QUEUES,
134 	DISABLE_PROCESS_QUEUES,
135 	PREEMPT_QUEUES_NO_UNMAP,
136 };
137 
138 struct kiq_pm4_funcs {
139 	/* Support ASIC-specific kiq pm4 packets*/
140 	void (*kiq_set_resources)(struct amdgpu_ring *kiq_ring,
141 					uint64_t queue_mask);
142 	void (*kiq_map_queues)(struct amdgpu_ring *kiq_ring,
143 					struct amdgpu_ring *ring);
144 	void (*kiq_unmap_queues)(struct amdgpu_ring *kiq_ring,
145 				 struct amdgpu_ring *ring,
146 				 enum amdgpu_unmap_queues_action action,
147 				 u64 gpu_addr, u64 seq);
148 	void (*kiq_query_status)(struct amdgpu_ring *kiq_ring,
149 					struct amdgpu_ring *ring,
150 					u64 addr,
151 					u64 seq);
152 	void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
153 				uint16_t pasid, uint32_t flush_type,
154 				bool all_hub);
155 	void (*kiq_reset_hw_queue)(struct amdgpu_ring *kiq_ring,
156 				   uint32_t queue_type, uint32_t me_id,
157 				   uint32_t pipe_id, uint32_t queue_id,
158 				   uint32_t xcc_id, uint32_t vmid);
159 	/* Packet sizes */
160 	int set_resources_size;
161 	int map_queues_size;
162 	int unmap_queues_size;
163 	int query_status_size;
164 	int invalidate_tlbs_size;
165 };
166 
167 struct amdgpu_kiq {
168 	u64			eop_gpu_addr;
169 	struct amdgpu_bo	*eop_obj;
170 	spinlock_t              ring_lock;
171 	struct amdgpu_ring	ring;
172 	struct amdgpu_irq_src	irq;
173 	const struct kiq_pm4_funcs *pmf;
174 	void			*mqd_backup;
175 };
176 
177 /*
178  * GFX configurations
179  */
180 #define AMDGPU_GFX_MAX_SE 4
181 #define AMDGPU_GFX_MAX_SH_PER_SE 2
182 
183 /**
184  * amdgpu_rb_config - Configure a single Render Backend (RB)
185  *
186  * Bad RBs are fused off and there is a harvest register the driver reads to
187  * determine which RB(s) are fused off so that the driver can configure the
188  * hardware state so that nothing gets sent to them. There are also user
189  * harvest registers that the driver can program to disable additional RBs,
190  * etc., for testing purposes.
191  */
192 struct amdgpu_rb_config {
193 	/**
194 	 * @rb_backend_disable:
195 	 *
196 	 * The value captured from register RB_BACKEND_DISABLE indicates if the
197 	 * RB backend is disabled or not.
198 	 */
199 	uint32_t rb_backend_disable;
200 
201 	/**
202 	 * @user_rb_backend_disable:
203 	 *
204 	 * The value captured from register USER_RB_BACKEND_DISABLE indicates
205 	 * if the User RB backend is disabled or not.
206 	 */
207 	uint32_t user_rb_backend_disable;
208 
209 	/**
210 	 * @raster_config:
211 	 *
212 	 * To set up all of the states, it is necessary to have two registers
213 	 * to keep all of the states. This field holds the first register.
214 	 */
215 	uint32_t raster_config;
216 
217 	/**
218 	 * @raster_config_1:
219 	 *
220 	 * To set up all of the states, it is necessary to have two registers
221 	 * to keep all of the states. This field holds the second register.
222 	 */
223 	uint32_t raster_config_1;
224 };
225 
226 struct gb_addr_config {
227 	uint16_t pipe_interleave_size;
228 	uint8_t num_pipes;
229 	uint8_t max_compress_frags;
230 	uint8_t num_banks;
231 	uint8_t num_se;
232 	uint8_t num_rb_per_se;
233 	uint8_t num_pkrs;
234 };
235 
236 struct amdgpu_gfx_config {
237 	unsigned max_shader_engines;
238 	unsigned max_tile_pipes;
239 	unsigned max_cu_per_sh;
240 	unsigned max_sh_per_se;
241 	unsigned max_backends_per_se;
242 	unsigned max_texture_channel_caches;
243 	unsigned max_gprs;
244 	unsigned max_gs_threads;
245 	unsigned max_hw_contexts;
246 	unsigned sc_prim_fifo_size_frontend;
247 	unsigned sc_prim_fifo_size_backend;
248 	unsigned sc_hiz_tile_fifo_size;
249 	unsigned sc_earlyz_tile_fifo_size;
250 
251 	unsigned num_tile_pipes;
252 	unsigned backend_enable_mask;
253 	unsigned mem_max_burst_length_bytes;
254 	unsigned mem_row_size_in_kb;
255 	unsigned shader_engine_tile_size;
256 	unsigned num_gpus;
257 	unsigned multi_gpu_tile_size;
258 	unsigned mc_arb_ramcfg;
259 	unsigned num_banks;
260 	unsigned num_ranks;
261 	unsigned gb_addr_config;
262 	unsigned num_rbs;
263 	unsigned gs_vgt_table_depth;
264 	unsigned gs_prim_buffer_depth;
265 
266 	uint32_t tile_mode_array[32];
267 	uint32_t macrotile_mode_array[16];
268 
269 	struct gb_addr_config gb_addr_config_fields;
270 
271 	/**
272 	 * @rb_config:
273 	 *
274 	 * Matrix that keeps all the Render Backend (color and depth buffer
275 	 * handling) configuration on the 3D engine.
276 	 */
277 	struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
278 
279 	/* gfx configure feature */
280 	uint32_t double_offchip_lds_buf;
281 	/* cached value of DB_DEBUG2 */
282 	uint32_t db_debug2;
283 	/* gfx10 specific config */
284 	uint32_t num_sc_per_sh;
285 	uint32_t num_packer_per_sc;
286 	uint32_t pa_sc_tile_steering_override;
287 	/* Whether texture coordinate truncation is conformant. */
288 	bool ta_cntl2_truncate_coord_mode;
289 	uint64_t tcc_disabled_mask;
290 	uint32_t gc_num_tcp_per_sa;
291 	uint32_t gc_num_sdp_interface;
292 	uint32_t gc_num_tcps;
293 	uint32_t gc_num_tcp_per_wpg;
294 	uint32_t gc_tcp_l1_size;
295 	uint32_t gc_num_sqc_per_wgp;
296 	uint32_t gc_l1_instruction_cache_size_per_sqc;
297 	uint32_t gc_l1_data_cache_size_per_sqc;
298 	uint32_t gc_gl1c_per_sa;
299 	uint32_t gc_gl1c_size_per_instance;
300 	uint32_t gc_gl2c_per_gpu;
301 	uint32_t gc_tcp_size_per_cu;
302 	uint32_t gc_num_cu_per_sqc;
303 	uint32_t gc_tcc_size;
304 	uint32_t gc_tcp_cache_line_size;
305 	uint32_t gc_instruction_cache_size_per_sqc;
306 	uint32_t gc_instruction_cache_line_size;
307 	uint32_t gc_scalar_data_cache_size_per_sqc;
308 	uint32_t gc_scalar_data_cache_line_size;
309 	uint32_t gc_tcc_cache_line_size;
310 };
311 
312 struct amdgpu_cu_info {
313 	uint32_t simd_per_cu;
314 	uint32_t max_waves_per_simd;
315 	uint32_t wave_front_size;
316 	uint32_t max_scratch_slots_per_cu;
317 	uint32_t lds_size;
318 
319 	/* total active CU number */
320 	uint32_t number;
321 	uint32_t ao_cu_mask;
322 	uint32_t ao_cu_bitmap[4][4];
323 	uint32_t bitmap[AMDGPU_MAX_GC_INSTANCES][4][4];
324 };
325 
326 struct amdgpu_gfx_ras {
327 	struct amdgpu_ras_block_object  ras_block;
328 	void (*enable_watchdog_timer)(struct amdgpu_device *adev);
329 	int (*rlc_gc_fed_irq)(struct amdgpu_device *adev,
330 				struct amdgpu_irq_src *source,
331 				struct amdgpu_iv_entry *entry);
332 	int (*poison_consumption_handler)(struct amdgpu_device *adev,
333 						struct amdgpu_iv_entry *entry);
334 };
335 
336 struct amdgpu_gfx_shadow_info {
337 	u32 shadow_size;
338 	u32 shadow_alignment;
339 	u32 csa_size;
340 	u32 csa_alignment;
341 	u32 eop_size;
342 	u32 eop_alignment;
343 };
344 
345 struct amdgpu_gfx_funcs {
346 	/* get the gpu clock counter */
347 	uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev);
348 	void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num,
349 			     u32 sh_num, u32 instance, int xcc_id);
350 	void (*read_wave_data)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
351 			       uint32_t wave, uint32_t *dst, int *no_fields);
352 	void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
353 				uint32_t wave, uint32_t thread, uint32_t start,
354 				uint32_t size, uint32_t *dst);
355 	void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
356 				uint32_t wave, uint32_t start, uint32_t size,
357 				uint32_t *dst);
358 	void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe,
359 				 u32 queue, u32 vmid, u32 xcc_id);
360 	void (*init_spm_golden)(struct amdgpu_device *adev);
361 	void (*update_perfmon_mgcg)(struct amdgpu_device *adev, bool enable);
362 	int (*get_gfx_shadow_info)(struct amdgpu_device *adev,
363 				   struct amdgpu_gfx_shadow_info *shadow_info,
364 				   bool skip_check);
365 	enum amdgpu_gfx_partition
366 			(*query_partition_mode)(struct amdgpu_device *adev);
367 	int (*switch_partition_mode)(struct amdgpu_device *adev,
368 				     int num_xccs_per_xcp);
369 	int (*ih_node_to_logical_xcc)(struct amdgpu_device *adev, int ih_node);
370 	int (*get_xccs_per_xcp)(struct amdgpu_device *adev);
371 	void (*get_hdp_flush_mask)(struct amdgpu_ring *ring,
372 				uint32_t *ref_and_mask, uint32_t *reg_mem_engine);
373 };
374 
375 struct sq_work {
376 	struct work_struct	work;
377 	unsigned ih_data;
378 };
379 
380 struct amdgpu_pfp {
381 	struct amdgpu_bo		*pfp_fw_obj;
382 	uint64_t			pfp_fw_gpu_addr;
383 	uint32_t			*pfp_fw_ptr;
384 
385 	struct amdgpu_bo		*pfp_fw_data_obj;
386 	uint64_t			pfp_fw_data_gpu_addr;
387 	uint32_t			*pfp_fw_data_ptr;
388 };
389 
390 struct amdgpu_ce {
391 	struct amdgpu_bo		*ce_fw_obj;
392 	uint64_t			ce_fw_gpu_addr;
393 	uint32_t			*ce_fw_ptr;
394 };
395 
396 struct amdgpu_me {
397 	struct amdgpu_bo		*me_fw_obj;
398 	uint64_t			me_fw_gpu_addr;
399 	uint32_t			*me_fw_ptr;
400 
401 	struct amdgpu_bo		*me_fw_data_obj;
402 	uint64_t			me_fw_data_gpu_addr;
403 	uint32_t			*me_fw_data_ptr;
404 
405 	uint32_t			num_me;
406 	uint32_t			num_pipe_per_me;
407 	uint32_t			num_queue_per_pipe;
408 	void				*mqd_backup[AMDGPU_MAX_GFX_RINGS];
409 	bool				use_mmio_for_reset;
410 
411 	/* These are the resources for which amdgpu takes ownership */
412 	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
413 };
414 
415 struct amdgpu_isolation_work {
416 	struct amdgpu_device		*adev;
417 	u32				xcp_id;
418 	struct delayed_work		work;
419 };
420 
421 struct amdgpu_gfx {
422 	struct mutex			gpu_clock_mutex;
423 	struct amdgpu_gfx_config	config;
424 	struct amdgpu_rlc		rlc;
425 	struct amdgpu_pfp		pfp;
426 	struct amdgpu_ce		ce;
427 	struct amdgpu_me		me;
428 	struct amdgpu_mec		mec;
429 	struct amdgpu_mec_bitmap	mec_bitmap[AMDGPU_MAX_GC_INSTANCES];
430 	struct amdgpu_kiq		kiq[AMDGPU_MAX_GC_INSTANCES];
431 	struct amdgpu_imu		imu;
432 	bool				rs64_enable; /* firmware format */
433 	const struct firmware		*me_fw;	/* ME firmware */
434 	uint32_t			me_fw_version;
435 	const struct firmware		*pfp_fw; /* PFP firmware */
436 	uint32_t			pfp_fw_version;
437 	const struct firmware		*ce_fw;	/* CE firmware */
438 	uint32_t			ce_fw_version;
439 	const struct firmware		*rlc_fw; /* RLC firmware */
440 	uint32_t			rlc_fw_version;
441 	const struct firmware		*mec_fw; /* MEC firmware */
442 	uint32_t			mec_fw_version;
443 	const struct firmware		*mec2_fw; /* MEC2 firmware */
444 	uint32_t			mec2_fw_version;
445 	const struct firmware		*imu_fw; /* IMU firmware */
446 	uint32_t			imu_fw_version;
447 	uint32_t			me_feature_version;
448 	uint32_t			ce_feature_version;
449 	uint32_t			pfp_feature_version;
450 	uint32_t			rlc_feature_version;
451 	uint32_t			rlc_srlc_fw_version;
452 	uint32_t			rlc_srlc_feature_version;
453 	uint32_t			rlc_srlg_fw_version;
454 	uint32_t			rlc_srlg_feature_version;
455 	uint32_t			rlc_srls_fw_version;
456 	uint32_t			rlc_srls_feature_version;
457 	uint32_t			rlcp_ucode_version;
458 	uint32_t			rlcp_ucode_feature_version;
459 	uint32_t			rlcv_ucode_version;
460 	uint32_t			rlcv_ucode_feature_version;
461 	uint32_t			mec_feature_version;
462 	uint32_t			mec2_feature_version;
463 	bool				mec_fw_write_wait;
464 	bool				me_fw_write_wait;
465 	bool				cp_fw_write_wait;
466 	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
467 	unsigned			num_gfx_rings;
468 	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES];
469 	unsigned			num_compute_rings;
470 	struct amdgpu_irq_src		eop_irq;
471 	struct amdgpu_irq_src		priv_reg_irq;
472 	struct amdgpu_irq_src		priv_inst_irq;
473 	struct amdgpu_irq_src		bad_op_irq;
474 	struct amdgpu_irq_src		cp_ecc_error_irq;
475 	struct amdgpu_irq_src		sq_irq;
476 	struct amdgpu_irq_src		rlc_gc_fed_irq;
477 	struct amdgpu_irq_src		rlc_poison_irq;
478 	struct sq_work			sq_work;
479 
480 	/* gfx status */
481 	uint32_t			gfx_current_status;
482 	/* ce ram size*/
483 	unsigned			ce_ram_size;
484 	struct amdgpu_cu_info		cu_info;
485 	const struct amdgpu_gfx_funcs	*funcs;
486 
487 	/* reset mask */
488 	uint32_t 			gfx_supported_reset;
489 	uint32_t 			compute_supported_reset;
490 
491 	/* gfx off */
492 	bool                            gfx_off_state;      /* true: enabled, false: disabled */
493 	struct mutex                    gfx_off_mutex;      /* mutex to change gfxoff state */
494 	uint32_t                        gfx_off_req_count;  /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */
495 	struct delayed_work             gfx_off_delay_work; /* async work to set gfx block off */
496 	uint32_t                        gfx_off_residency;  /* last logged residency */
497 	uint64_t                        gfx_off_entrycount; /* count of times GPU has get into GFXOFF state */
498 
499 	/* pipe reservation */
500 	struct mutex			pipe_reserve_mutex;
501 	DECLARE_BITMAP			(pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
502 
503 	/*ras */
504 	struct ras_common_if		*ras_if;
505 	struct amdgpu_gfx_ras		*ras;
506 
507 	bool				is_poweron;
508 
509 	struct amdgpu_ring		sw_gfx_ring[AMDGPU_MAX_SW_GFX_RINGS];
510 	struct amdgpu_ring_mux          muxer;
511 
512 	bool				cp_gfx_shadow; /* for gfx11 */
513 
514 	uint16_t 			xcc_mask;
515 	uint32_t			num_xcc_per_xcp;
516 	struct mutex			partition_mutex;
517 	bool				mcbp; /* mid command buffer preemption */
518 
519 	/* IP reg dump */
520 	uint32_t			*ip_dump_core;
521 	uint32_t			*ip_dump_compute_queues;
522 	uint32_t			*ip_dump_gfx_queues;
523 
524 	struct mutex			reset_sem_mutex;
525 
526 	/* cleaner shader */
527 	struct amdgpu_bo		*cleaner_shader_obj;
528 	unsigned int                    cleaner_shader_size;
529 	u64				cleaner_shader_gpu_addr;
530 	void				*cleaner_shader_cpu_ptr;
531 	const void			*cleaner_shader_ptr;
532 	bool				enable_cleaner_shader;
533 	struct amdgpu_isolation_work	enforce_isolation[MAX_XCP];
534 	/* Mutex for synchronizing KFD scheduler operations */
535 	struct mutex                    userq_sch_mutex;
536 	u64				userq_sch_req_count[MAX_XCP];
537 	bool				userq_sch_inactive[MAX_XCP];
538 	/* atomic bitmap of faulted gfx UQ slots (index = pipe | queue << 2) */
539 	unsigned long			userq_priv_fault_slots;
540 	struct work_struct		userq_priv_fault_work;
541 	unsigned long			enforce_isolation_jiffies[MAX_XCP];
542 	unsigned long			enforce_isolation_time[MAX_XCP];
543 
544 	atomic_t			total_submission_cnt;
545 	struct delayed_work		idle_work;
546 	bool				workload_profile_active;
547 	struct mutex                    workload_profile_mutex;
548 
549 	bool				disable_kq;
550 	bool				disable_uq;
551 };
552 
553 struct amdgpu_gfx_deferred_entry {
554 	struct amdgpu_ring	*ring;
555 	struct amdgpu_fence	*fence;
556 };
557 
558 struct amdgpu_gfx_ras_reg_entry {
559 	struct amdgpu_ras_err_status_reg_entry reg_entry;
560 	enum amdgpu_gfx_ras_mem_id_type mem_id_type;
561 	uint32_t se_num;
562 };
563 
564 struct amdgpu_gfx_ras_mem_id_entry {
565 	const struct amdgpu_ras_memory_id_entry *mem_id_ent;
566 	uint32_t size;
567 };
568 
569 #define AMDGPU_GFX_MEMID_ENT(x) {(x), ARRAY_SIZE(x)},
570 
571 #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))
572 #define amdgpu_gfx_select_se_sh(adev, se, sh, instance, xcc_id) ((adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance), (xcc_id)))
573 #define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid, xcc_id) ((adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid), (xcc_id)))
574 #define amdgpu_gfx_init_spm_golden(adev) (adev)->gfx.funcs->init_spm_golden((adev))
575 #define amdgpu_gfx_get_gfx_shadow_info(adev, si) ((adev)->gfx.funcs->get_gfx_shadow_info((adev), (si), false))
576 
577 /**
578  * amdgpu_gfx_create_bitmask - create a bitmask
579  *
580  * @bit_width: length of the mask
581  *
582  * create a variable length bit mask.
583  * Returns the bitmask.
584  */
585 static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width)
586 {
587 	return (u32)((1ULL << bit_width) - 1);
588 }
589 
590 void amdgpu_gfx_parse_disable_cu(struct amdgpu_device *adev, unsigned int *mask,
591 				 unsigned int max_se, unsigned int max_sh);
592 
593 int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, int xcc_id);
594 
595 void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring);
596 
597 void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev, int xcc_id);
598 int amdgpu_gfx_kiq_init(struct amdgpu_device *adev,
599 			unsigned hpd_size, int xcc_id);
600 
601 int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
602 			   unsigned mqd_size, int xcc_id);
603 void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev, int xcc_id);
604 void amdgpu_gfx_mqd_symmetrically_map_cu_mask(struct amdgpu_device *adev, const uint32_t *cu_mask,
605 					      uint32_t cu_mask_count, uint32_t *se_mask);
606 int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id);
607 int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev, int xcc_id);
608 int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id);
609 int amdgpu_gfx_enable_kgq(struct amdgpu_device *adev, int xcc_id);
610 
611 void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev);
612 void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev);
613 
614 int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec,
615 				int pipe, int queue);
616 void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
617 				 int *mec, int *pipe, int *queue);
618 bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int xcc_id,
619 				     int mec, int pipe, int queue);
620 bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev,
621 					       struct amdgpu_ring *ring);
622 bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev,
623 						struct amdgpu_ring *ring);
624 bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me,
625 				    int pipe, int queue);
626 void amdgpu_gfx_handle_priv_fault(struct amdgpu_device *adev,
627 					struct amdgpu_iv_entry *entry,
628 					u8 me_id, u8 pipe_id, u8 queue_id);
629 void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable);
630 void amdgpu_gfx_off_ctrl_immediate(struct amdgpu_device *adev, bool enable);
631 int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value);
632 int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block);
633 void amdgpu_gfx_ras_suspend(struct amdgpu_device *adev, struct ras_common_if *ras_block);
634 void amdgpu_gfx_ras_fini(struct amdgpu_device *adev, struct ras_common_if *ras_block);
635 int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value);
636 int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *residency);
637 int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value);
638 int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev,
639 		void *err_data,
640 		struct amdgpu_iv_entry *entry);
641 int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
642 				  struct amdgpu_irq_src *source,
643 				  struct amdgpu_iv_entry *entry);
644 uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg, uint32_t xcc_id);
645 void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, uint32_t xcc_id);
646 void amdgpu_gfx_get_hdp_flush_mask(struct amdgpu_ring *ring,
647 		uint32_t *ref_and_mask, uint32_t *reg_mem_engine);
648 int amdgpu_kiq_hdp_flush(struct amdgpu_device *adev);
649 int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev);
650 void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id);
651 
652 int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev);
653 int amdgpu_gfx_poison_consumption_handler(struct amdgpu_device *adev,
654 						struct amdgpu_iv_entry *entry);
655 
656 bool amdgpu_gfx_is_master_xcc(struct amdgpu_device *adev, int xcc_id);
657 int amdgpu_gfx_sysfs_init(struct amdgpu_device *adev);
658 void amdgpu_gfx_sysfs_fini(struct amdgpu_device *adev);
659 int amdgpu_gfx_reset_mes_compute(struct amdgpu_device *adev,
660 				 struct amdgpu_ring *ring,
661 				 struct amdgpu_fence *guilty_fence,
662 				 struct amdgpu_usermode_queue *uq,
663 				 unsigned int *hung_queue_count,
664 				 void *faulty_queue_input);
665 void amdgpu_gfx_ras_error_func(struct amdgpu_device *adev,
666 		void *ras_error_status,
667 		void (*func)(struct amdgpu_device *adev, void *ras_error_status,
668 				int xcc_id));
669 int amdgpu_gfx_cleaner_shader_sw_init(struct amdgpu_device *adev,
670 				      unsigned int cleaner_shader_size);
671 void amdgpu_gfx_cleaner_shader_sw_fini(struct amdgpu_device *adev);
672 void amdgpu_gfx_cleaner_shader_init(struct amdgpu_device *adev,
673 				    unsigned int cleaner_shader_size,
674 				    const void *cleaner_shader_ptr);
675 void amdgpu_gfx_enforce_isolation_handler(struct work_struct *work);
676 void amdgpu_gfx_enforce_isolation_ring_begin_use(struct amdgpu_ring *ring);
677 void amdgpu_gfx_enforce_isolation_ring_end_use(struct amdgpu_ring *ring);
678 
679 void amdgpu_gfx_profile_idle_work_handler(struct work_struct *work);
680 void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring);
681 void amdgpu_gfx_profile_ring_end_use(struct amdgpu_ring *ring);
682 u32 amdgpu_gfx_csb_preamble_start(u32 *buffer);
683 u32 amdgpu_gfx_csb_data_parser(struct amdgpu_device *adev, u32 *buffer, u32 count);
684 void amdgpu_gfx_csb_preamble_end(u32 *buffer, u32 count);
685 
686 void amdgpu_debugfs_gfx_sched_mask_init(struct amdgpu_device *adev);
687 void amdgpu_debugfs_compute_sched_mask_init(struct amdgpu_device *adev);
688 
689 int amdgpu_gfx_ring_preempt_ib(struct amdgpu_ring *ring);
690 
691 int amdgpu_gfx_mes_reset_queue(struct amdgpu_ring *ring,
692 			       unsigned int vmid,
693 			       struct amdgpu_fence *timedout_fence,
694 			       bool use_mmio);
695 
696 static inline const char *amdgpu_gfx_compute_mode_desc(int mode)
697 {
698 	switch (mode) {
699 	case AMDGPU_SPX_PARTITION_MODE:
700 		return "SPX";
701 	case AMDGPU_DPX_PARTITION_MODE:
702 		return "DPX";
703 	case AMDGPU_TPX_PARTITION_MODE:
704 		return "TPX";
705 	case AMDGPU_QPX_PARTITION_MODE:
706 		return "QPX";
707 	case AMDGPU_CPX_PARTITION_MODE:
708 		return "CPX";
709 	default:
710 		return "UNKNOWN";
711 	}
712 }
713 
714 static inline const char *amdgpu_gfx_compute_mem_alloc_mode_desc(int mode)
715 {
716 	switch (mode) {
717 	case AMDGPU_PARTITION_MEM_CAPPING_EVEN:
718 			return "CAPPING";
719 	case AMDGPU_PARTITION_MEM_ALLOC_ALL:
720 		return "ALL";
721 	default:
722 		return "UNKNOWN";
723 	}
724 }
725 
726 #endif
727