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