1 /*	$NetBSD: amdgpu_gfx.h,v 1.3 2021/12/18 23:44:58 riastradh Exp $	*/
2 
3 /*
4  * Copyright 2014 Advanced Micro Devices, Inc.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  */
25 
26 #ifndef __AMDGPU_GFX_H__
27 #define __AMDGPU_GFX_H__
28 
29 /*
30  * GFX stuff
31  */
32 #include "clearstate_defs.h"
33 #include "amdgpu_ring.h"
34 #include "amdgpu_rlc.h"
35 
36 /* GFX current status */
37 #define AMDGPU_GFX_NORMAL_MODE			0x00000000L
38 #define AMDGPU_GFX_SAFE_MODE			0x00000001L
39 #define AMDGPU_GFX_PG_DISABLED_MODE		0x00000002L
40 #define AMDGPU_GFX_CG_DISABLED_MODE		0x00000004L
41 #define AMDGPU_GFX_LBPW_DISABLED_MODE		0x00000008L
42 
43 #define AMDGPU_MAX_GFX_QUEUES KGD_MAX_QUEUES
44 #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES
45 
46 struct amdgpu_mec {
47 	struct amdgpu_bo	*hpd_eop_obj;
48 	u64			hpd_eop_gpu_addr;
49 	struct amdgpu_bo	*mec_fw_obj;
50 	u64			mec_fw_gpu_addr;
51 	u32 num_mec;
52 	u32 num_pipe_per_mec;
53 	u32 num_queue_per_pipe;
54 	void			*mqd_backup[AMDGPU_MAX_COMPUTE_RINGS + 1];
55 
56 	/* These are the resources for which amdgpu takes ownership */
57 	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
58 };
59 
60 enum amdgpu_unmap_queues_action {
61 	PREEMPT_QUEUES = 0,
62 	RESET_QUEUES,
63 	DISABLE_PROCESS_QUEUES,
64 	PREEMPT_QUEUES_NO_UNMAP,
65 };
66 
67 struct kiq_pm4_funcs {
68 	/* Support ASIC-specific kiq pm4 packets*/
69 	void (*kiq_set_resources)(struct amdgpu_ring *kiq_ring,
70 					uint64_t queue_mask);
71 	void (*kiq_map_queues)(struct amdgpu_ring *kiq_ring,
72 					struct amdgpu_ring *ring);
73 	void (*kiq_unmap_queues)(struct amdgpu_ring *kiq_ring,
74 				 struct amdgpu_ring *ring,
75 				 enum amdgpu_unmap_queues_action action,
76 				 u64 gpu_addr, u64 seq);
77 	void (*kiq_query_status)(struct amdgpu_ring *kiq_ring,
78 					struct amdgpu_ring *ring,
79 					u64 addr,
80 					u64 seq);
81 	void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
82 				uint16_t pasid, uint32_t flush_type,
83 				bool all_hub);
84 	/* Packet sizes */
85 	int set_resources_size;
86 	int map_queues_size;
87 	int unmap_queues_size;
88 	int query_status_size;
89 	int invalidate_tlbs_size;
90 };
91 
92 struct amdgpu_kiq {
93 	u64			eop_gpu_addr;
94 	struct amdgpu_bo	*eop_obj;
95 	spinlock_t              ring_lock;
96 	struct amdgpu_ring	ring;
97 	struct amdgpu_irq_src	irq;
98 	const struct kiq_pm4_funcs *pmf;
99 	uint32_t			reg_val_offs;
100 };
101 
102 /*
103  * GPU scratch registers structures, functions & helpers
104  */
105 struct amdgpu_scratch {
106 	unsigned		num_reg;
107 	uint32_t                reg_base;
108 	uint32_t		free_mask;
109 };
110 
111 /*
112  * GFX configurations
113  */
114 #define AMDGPU_GFX_MAX_SE 4
115 #define AMDGPU_GFX_MAX_SH_PER_SE 2
116 
117 struct amdgpu_rb_config {
118 	uint32_t rb_backend_disable;
119 	uint32_t user_rb_backend_disable;
120 	uint32_t raster_config;
121 	uint32_t raster_config_1;
122 };
123 
124 struct gb_addr_config {
125 	uint16_t pipe_interleave_size;
126 	uint8_t num_pipes;
127 	uint8_t max_compress_frags;
128 	uint8_t num_banks;
129 	uint8_t num_se;
130 	uint8_t num_rb_per_se;
131 };
132 
133 struct amdgpu_gfx_config {
134 	unsigned max_shader_engines;
135 	unsigned max_tile_pipes;
136 	unsigned max_cu_per_sh;
137 	unsigned max_sh_per_se;
138 	unsigned max_backends_per_se;
139 	unsigned max_texture_channel_caches;
140 	unsigned max_gprs;
141 	unsigned max_gs_threads;
142 	unsigned max_hw_contexts;
143 	unsigned sc_prim_fifo_size_frontend;
144 	unsigned sc_prim_fifo_size_backend;
145 	unsigned sc_hiz_tile_fifo_size;
146 	unsigned sc_earlyz_tile_fifo_size;
147 
148 	unsigned num_tile_pipes;
149 	unsigned backend_enable_mask;
150 	unsigned mem_max_burst_length_bytes;
151 	unsigned mem_row_size_in_kb;
152 	unsigned shader_engine_tile_size;
153 	unsigned num_gpus;
154 	unsigned multi_gpu_tile_size;
155 	unsigned mc_arb_ramcfg;
156 	unsigned gb_addr_config;
157 	unsigned num_rbs;
158 	unsigned gs_vgt_table_depth;
159 	unsigned gs_prim_buffer_depth;
160 
161 	uint32_t tile_mode_array[32];
162 	uint32_t macrotile_mode_array[16];
163 
164 	struct gb_addr_config gb_addr_config_fields;
165 	struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
166 
167 	/* gfx configure feature */
168 	uint32_t double_offchip_lds_buf;
169 	/* cached value of DB_DEBUG2 */
170 	uint32_t db_debug2;
171 	/* gfx10 specific config */
172 	uint32_t num_sc_per_sh;
173 	uint32_t num_packer_per_sc;
174 	uint32_t pa_sc_tile_steering_override;
175 	uint64_t tcc_disabled_mask;
176 };
177 
178 struct amdgpu_cu_info {
179 	uint32_t simd_per_cu;
180 	uint32_t max_waves_per_simd;
181 	uint32_t wave_front_size;
182 	uint32_t max_scratch_slots_per_cu;
183 	uint32_t lds_size;
184 
185 	/* total active CU number */
186 	uint32_t number;
187 	uint32_t ao_cu_mask;
188 	uint32_t ao_cu_bitmap[4][4];
189 	uint32_t bitmap[4][4];
190 };
191 
192 struct amdgpu_gfx_funcs {
193 	/* get the gpu clock counter */
194 	uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev);
195 	void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num,
196 			     u32 sh_num, u32 instance);
197 	void (*read_wave_data)(struct amdgpu_device *adev, uint32_t simd,
198 			       uint32_t wave, uint32_t *dst, int *no_fields);
199 	void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t simd,
200 				uint32_t wave, uint32_t thread, uint32_t start,
201 				uint32_t size, uint32_t *dst);
202 	void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t simd,
203 				uint32_t wave, uint32_t start, uint32_t size,
204 				uint32_t *dst);
205 	void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe,
206 				 u32 queue, u32 vmid);
207 	int (*ras_error_inject)(struct amdgpu_device *adev, void *inject_if);
208 	int (*query_ras_error_count) (struct amdgpu_device *adev, void *ras_error_status);
209 };
210 
211 struct sq_work {
212 	struct work_struct	work;
213 	unsigned ih_data;
214 };
215 
216 struct amdgpu_pfp {
217 	struct amdgpu_bo		*pfp_fw_obj;
218 	uint64_t			pfp_fw_gpu_addr;
219 	uint32_t			*pfp_fw_ptr;
220 };
221 
222 struct amdgpu_ce {
223 	struct amdgpu_bo		*ce_fw_obj;
224 	uint64_t			ce_fw_gpu_addr;
225 	uint32_t			*ce_fw_ptr;
226 };
227 
228 struct amdgpu_me {
229 	struct amdgpu_bo		*me_fw_obj;
230 	uint64_t			me_fw_gpu_addr;
231 	uint32_t			*me_fw_ptr;
232 	uint32_t			num_me;
233 	uint32_t			num_pipe_per_me;
234 	uint32_t			num_queue_per_pipe;
235 	void				*mqd_backup[AMDGPU_MAX_GFX_RINGS];
236 
237 	/* These are the resources for which amdgpu takes ownership */
238 	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
239 };
240 
241 struct amdgpu_gfx {
242 	struct mutex			gpu_clock_mutex;
243 	struct amdgpu_gfx_config	config;
244 	struct amdgpu_rlc		rlc;
245 	struct amdgpu_pfp		pfp;
246 	struct amdgpu_ce		ce;
247 	struct amdgpu_me		me;
248 	struct amdgpu_mec		mec;
249 	struct amdgpu_kiq		kiq;
250 	struct amdgpu_scratch		scratch;
251 	const struct firmware		*me_fw;	/* ME firmware */
252 	uint32_t			me_fw_version;
253 	const struct firmware		*pfp_fw; /* PFP firmware */
254 	uint32_t			pfp_fw_version;
255 	const struct firmware		*ce_fw;	/* CE firmware */
256 	uint32_t			ce_fw_version;
257 	const struct firmware		*rlc_fw; /* RLC firmware */
258 	uint32_t			rlc_fw_version;
259 	const struct firmware		*mec_fw; /* MEC firmware */
260 	uint32_t			mec_fw_version;
261 	const struct firmware		*mec2_fw; /* MEC2 firmware */
262 	uint32_t			mec2_fw_version;
263 	uint32_t			me_feature_version;
264 	uint32_t			ce_feature_version;
265 	uint32_t			pfp_feature_version;
266 	uint32_t			rlc_feature_version;
267 	uint32_t			rlc_srlc_fw_version;
268 	uint32_t			rlc_srlc_feature_version;
269 	uint32_t			rlc_srlg_fw_version;
270 	uint32_t			rlc_srlg_feature_version;
271 	uint32_t			rlc_srls_fw_version;
272 	uint32_t			rlc_srls_feature_version;
273 	uint32_t			mec_feature_version;
274 	uint32_t			mec2_feature_version;
275 	bool				mec_fw_write_wait;
276 	bool				me_fw_write_wait;
277 	bool				cp_fw_write_wait;
278 	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
279 	struct drm_gpu_scheduler	*gfx_sched[AMDGPU_MAX_GFX_RINGS];
280 	uint32_t			num_gfx_sched;
281 	unsigned			num_gfx_rings;
282 	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
283 	struct drm_gpu_scheduler	*compute_sched[AMDGPU_MAX_COMPUTE_RINGS];
284 	uint32_t			num_compute_sched;
285 	unsigned			num_compute_rings;
286 	struct amdgpu_irq_src		eop_irq;
287 	struct amdgpu_irq_src		priv_reg_irq;
288 	struct amdgpu_irq_src		priv_inst_irq;
289 	struct amdgpu_irq_src		cp_ecc_error_irq;
290 	struct amdgpu_irq_src		sq_irq;
291 	struct sq_work			sq_work;
292 
293 	/* gfx status */
294 	uint32_t			gfx_current_status;
295 	/* ce ram size*/
296 	unsigned			ce_ram_size;
297 	struct amdgpu_cu_info		cu_info;
298 	const struct amdgpu_gfx_funcs	*funcs;
299 
300 	/* reset mask */
301 	uint32_t                        grbm_soft_reset;
302 	uint32_t                        srbm_soft_reset;
303 
304 	/* gfx off */
305 	bool                            gfx_off_state; /* true: enabled, false: disabled */
306 	struct mutex                    gfx_off_mutex;
307 	uint32_t                        gfx_off_req_count; /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */
308 	struct delayed_work             gfx_off_delay_work;
309 
310 	/* pipe reservation */
311 	struct mutex			pipe_reserve_mutex;
312 	DECLARE_BITMAP			(pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
313 
314 	/*ras */
315 	struct ras_common_if		*ras_if;
316 };
317 
318 #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))
319 #define amdgpu_gfx_select_se_sh(adev, se, sh, instance) (adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance))
320 #define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid) (adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid))
321 
322 /**
323  * amdgpu_gfx_create_bitmask - create a bitmask
324  *
325  * @bit_width: length of the mask
326  *
327  * create a variable length bit mask.
328  * Returns the bitmask.
329  */
amdgpu_gfx_create_bitmask(u32 bit_width)330 static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width)
331 {
332 	return (u32)((1ULL << bit_width) - 1);
333 }
334 
335 int amdgpu_gfx_scratch_get(struct amdgpu_device *adev, uint32_t *reg);
336 void amdgpu_gfx_scratch_free(struct amdgpu_device *adev, uint32_t reg);
337 
338 void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se,
339 				 unsigned max_sh);
340 
341 int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev,
342 			     struct amdgpu_ring *ring,
343 			     struct amdgpu_irq_src *irq);
344 
345 void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring);
346 
347 void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev);
348 int amdgpu_gfx_kiq_init(struct amdgpu_device *adev,
349 			unsigned hpd_size);
350 
351 int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
352 			   unsigned mqd_size);
353 void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev);
354 int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev);
355 int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev);
356 
357 void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev);
358 void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev);
359 
360 int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec,
361 				int pipe, int queue);
362 void amdgpu_gfx_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
363 				 int *mec, int *pipe, int *queue);
364 bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int mec,
365 				     int pipe, int queue);
366 int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev, int me,
367 			       int pipe, int queue);
368 void amdgpu_gfx_bit_to_me_queue(struct amdgpu_device *adev, int bit,
369 				int *me, int *pipe, int *queue);
370 bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me,
371 				    int pipe, int queue);
372 void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable);
373 int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev);
374 void amdgpu_gfx_ras_fini(struct amdgpu_device *adev);
375 int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev,
376 		void *err_data,
377 		struct amdgpu_iv_entry *entry);
378 int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
379 				  struct amdgpu_irq_src *source,
380 				  struct amdgpu_iv_entry *entry);
381 uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg);
382 void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v);
383 #endif
384