Loading...
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
27int amdgpu_gfx_scratch_get(struct amdgpu_device *adev, uint32_t *reg);
28void amdgpu_gfx_scratch_free(struct amdgpu_device *adev, uint32_t reg);
29
30#endif
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
38/* GFX current status */
39#define AMDGPU_GFX_NORMAL_MODE 0x00000000L
40#define AMDGPU_GFX_SAFE_MODE 0x00000001L
41#define AMDGPU_GFX_PG_DISABLED_MODE 0x00000002L
42#define AMDGPU_GFX_CG_DISABLED_MODE 0x00000004L
43#define AMDGPU_GFX_LBPW_DISABLED_MODE 0x00000008L
44
45#define AMDGPU_MAX_GFX_QUEUES KGD_MAX_QUEUES
46#define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES
47
48enum amdgpu_gfx_pipe_priority {
49 AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1,
50 AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2
51};
52
53#define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM 0
54#define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM 15
55
56struct amdgpu_mec {
57 struct amdgpu_bo *hpd_eop_obj;
58 u64 hpd_eop_gpu_addr;
59 struct amdgpu_bo *mec_fw_obj;
60 u64 mec_fw_gpu_addr;
61 struct amdgpu_bo *mec_fw_data_obj;
62 u64 mec_fw_data_gpu_addr;
63
64 u32 num_mec;
65 u32 num_pipe_per_mec;
66 u32 num_queue_per_pipe;
67 void *mqd_backup[AMDGPU_MAX_COMPUTE_RINGS + 1];
68
69 /* These are the resources for which amdgpu takes ownership */
70 DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
71};
72
73enum amdgpu_unmap_queues_action {
74 PREEMPT_QUEUES = 0,
75 RESET_QUEUES,
76 DISABLE_PROCESS_QUEUES,
77 PREEMPT_QUEUES_NO_UNMAP,
78};
79
80struct kiq_pm4_funcs {
81 /* Support ASIC-specific kiq pm4 packets*/
82 void (*kiq_set_resources)(struct amdgpu_ring *kiq_ring,
83 uint64_t queue_mask);
84 void (*kiq_map_queues)(struct amdgpu_ring *kiq_ring,
85 struct amdgpu_ring *ring);
86 void (*kiq_unmap_queues)(struct amdgpu_ring *kiq_ring,
87 struct amdgpu_ring *ring,
88 enum amdgpu_unmap_queues_action action,
89 u64 gpu_addr, u64 seq);
90 void (*kiq_query_status)(struct amdgpu_ring *kiq_ring,
91 struct amdgpu_ring *ring,
92 u64 addr,
93 u64 seq);
94 void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
95 uint16_t pasid, uint32_t flush_type,
96 bool all_hub);
97 /* Packet sizes */
98 int set_resources_size;
99 int map_queues_size;
100 int unmap_queues_size;
101 int query_status_size;
102 int invalidate_tlbs_size;
103};
104
105struct amdgpu_kiq {
106 u64 eop_gpu_addr;
107 struct amdgpu_bo *eop_obj;
108 spinlock_t ring_lock;
109 struct amdgpu_ring ring;
110 struct amdgpu_irq_src irq;
111 const struct kiq_pm4_funcs *pmf;
112};
113
114/*
115 * GFX configurations
116 */
117#define AMDGPU_GFX_MAX_SE 4
118#define AMDGPU_GFX_MAX_SH_PER_SE 2
119
120struct amdgpu_rb_config {
121 uint32_t rb_backend_disable;
122 uint32_t user_rb_backend_disable;
123 uint32_t raster_config;
124 uint32_t raster_config_1;
125};
126
127struct gb_addr_config {
128 uint16_t pipe_interleave_size;
129 uint8_t num_pipes;
130 uint8_t max_compress_frags;
131 uint8_t num_banks;
132 uint8_t num_se;
133 uint8_t num_rb_per_se;
134 uint8_t num_pkrs;
135};
136
137struct amdgpu_gfx_config {
138 unsigned max_shader_engines;
139 unsigned max_tile_pipes;
140 unsigned max_cu_per_sh;
141 unsigned max_sh_per_se;
142 unsigned max_backends_per_se;
143 unsigned max_texture_channel_caches;
144 unsigned max_gprs;
145 unsigned max_gs_threads;
146 unsigned max_hw_contexts;
147 unsigned sc_prim_fifo_size_frontend;
148 unsigned sc_prim_fifo_size_backend;
149 unsigned sc_hiz_tile_fifo_size;
150 unsigned sc_earlyz_tile_fifo_size;
151
152 unsigned num_tile_pipes;
153 unsigned backend_enable_mask;
154 unsigned mem_max_burst_length_bytes;
155 unsigned mem_row_size_in_kb;
156 unsigned shader_engine_tile_size;
157 unsigned num_gpus;
158 unsigned multi_gpu_tile_size;
159 unsigned mc_arb_ramcfg;
160 unsigned num_banks;
161 unsigned num_ranks;
162 unsigned gb_addr_config;
163 unsigned num_rbs;
164 unsigned gs_vgt_table_depth;
165 unsigned gs_prim_buffer_depth;
166
167 uint32_t tile_mode_array[32];
168 uint32_t macrotile_mode_array[16];
169
170 struct gb_addr_config gb_addr_config_fields;
171 struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
172
173 /* gfx configure feature */
174 uint32_t double_offchip_lds_buf;
175 /* cached value of DB_DEBUG2 */
176 uint32_t db_debug2;
177 /* gfx10 specific config */
178 uint32_t num_sc_per_sh;
179 uint32_t num_packer_per_sc;
180 uint32_t pa_sc_tile_steering_override;
181 uint64_t tcc_disabled_mask;
182 uint32_t gc_num_tcp_per_sa;
183 uint32_t gc_num_sdp_interface;
184 uint32_t gc_num_tcps;
185 uint32_t gc_num_tcp_per_wpg;
186 uint32_t gc_tcp_l1_size;
187 uint32_t gc_num_sqc_per_wgp;
188 uint32_t gc_l1_instruction_cache_size_per_sqc;
189 uint32_t gc_l1_data_cache_size_per_sqc;
190 uint32_t gc_gl1c_per_sa;
191 uint32_t gc_gl1c_size_per_instance;
192 uint32_t gc_gl2c_per_gpu;
193};
194
195struct amdgpu_cu_info {
196 uint32_t simd_per_cu;
197 uint32_t max_waves_per_simd;
198 uint32_t wave_front_size;
199 uint32_t max_scratch_slots_per_cu;
200 uint32_t lds_size;
201
202 /* total active CU number */
203 uint32_t number;
204 uint32_t ao_cu_mask;
205 uint32_t ao_cu_bitmap[4][4];
206 uint32_t bitmap[4][4];
207};
208
209struct amdgpu_gfx_ras {
210 struct amdgpu_ras_block_object ras_block;
211 void (*enable_watchdog_timer)(struct amdgpu_device *adev);
212 bool (*query_utcl2_poison_status)(struct amdgpu_device *adev);
213};
214
215struct amdgpu_gfx_funcs {
216 /* get the gpu clock counter */
217 uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev);
218 void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num,
219 u32 sh_num, u32 instance);
220 void (*read_wave_data)(struct amdgpu_device *adev, uint32_t simd,
221 uint32_t wave, uint32_t *dst, int *no_fields);
222 void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t simd,
223 uint32_t wave, uint32_t thread, uint32_t start,
224 uint32_t size, uint32_t *dst);
225 void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t simd,
226 uint32_t wave, uint32_t start, uint32_t size,
227 uint32_t *dst);
228 void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe,
229 u32 queue, u32 vmid);
230 void (*init_spm_golden)(struct amdgpu_device *adev);
231 void (*update_perfmon_mgcg)(struct amdgpu_device *adev, bool enable);
232};
233
234struct sq_work {
235 struct work_struct work;
236 unsigned ih_data;
237};
238
239struct amdgpu_pfp {
240 struct amdgpu_bo *pfp_fw_obj;
241 uint64_t pfp_fw_gpu_addr;
242 uint32_t *pfp_fw_ptr;
243
244 struct amdgpu_bo *pfp_fw_data_obj;
245 uint64_t pfp_fw_data_gpu_addr;
246 uint32_t *pfp_fw_data_ptr;
247};
248
249struct amdgpu_ce {
250 struct amdgpu_bo *ce_fw_obj;
251 uint64_t ce_fw_gpu_addr;
252 uint32_t *ce_fw_ptr;
253};
254
255struct amdgpu_me {
256 struct amdgpu_bo *me_fw_obj;
257 uint64_t me_fw_gpu_addr;
258 uint32_t *me_fw_ptr;
259
260 struct amdgpu_bo *me_fw_data_obj;
261 uint64_t me_fw_data_gpu_addr;
262 uint32_t *me_fw_data_ptr;
263
264 uint32_t num_me;
265 uint32_t num_pipe_per_me;
266 uint32_t num_queue_per_pipe;
267 void *mqd_backup[AMDGPU_MAX_GFX_RINGS];
268
269 /* These are the resources for which amdgpu takes ownership */
270 DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
271};
272
273struct amdgpu_gfx {
274 struct mutex gpu_clock_mutex;
275 struct amdgpu_gfx_config config;
276 struct amdgpu_rlc rlc;
277 struct amdgpu_pfp pfp;
278 struct amdgpu_ce ce;
279 struct amdgpu_me me;
280 struct amdgpu_mec mec;
281 struct amdgpu_kiq kiq;
282 struct amdgpu_imu imu;
283 bool rs64_enable; /* firmware format */
284 const struct firmware *me_fw; /* ME firmware */
285 uint32_t me_fw_version;
286 const struct firmware *pfp_fw; /* PFP firmware */
287 uint32_t pfp_fw_version;
288 const struct firmware *ce_fw; /* CE firmware */
289 uint32_t ce_fw_version;
290 const struct firmware *rlc_fw; /* RLC firmware */
291 uint32_t rlc_fw_version;
292 const struct firmware *mec_fw; /* MEC firmware */
293 uint32_t mec_fw_version;
294 const struct firmware *mec2_fw; /* MEC2 firmware */
295 uint32_t mec2_fw_version;
296 const struct firmware *imu_fw; /* IMU firmware */
297 uint32_t imu_fw_version;
298 uint32_t me_feature_version;
299 uint32_t ce_feature_version;
300 uint32_t pfp_feature_version;
301 uint32_t rlc_feature_version;
302 uint32_t rlc_srlc_fw_version;
303 uint32_t rlc_srlc_feature_version;
304 uint32_t rlc_srlg_fw_version;
305 uint32_t rlc_srlg_feature_version;
306 uint32_t rlc_srls_fw_version;
307 uint32_t rlc_srls_feature_version;
308 uint32_t rlcp_ucode_version;
309 uint32_t rlcp_ucode_feature_version;
310 uint32_t rlcv_ucode_version;
311 uint32_t rlcv_ucode_feature_version;
312 uint32_t mec_feature_version;
313 uint32_t mec2_feature_version;
314 bool mec_fw_write_wait;
315 bool me_fw_write_wait;
316 bool cp_fw_write_wait;
317 struct amdgpu_ring gfx_ring[AMDGPU_MAX_GFX_RINGS];
318 unsigned num_gfx_rings;
319 struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
320 unsigned num_compute_rings;
321 struct amdgpu_irq_src eop_irq;
322 struct amdgpu_irq_src priv_reg_irq;
323 struct amdgpu_irq_src priv_inst_irq;
324 struct amdgpu_irq_src cp_ecc_error_irq;
325 struct amdgpu_irq_src sq_irq;
326 struct sq_work sq_work;
327
328 /* gfx status */
329 uint32_t gfx_current_status;
330 /* ce ram size*/
331 unsigned ce_ram_size;
332 struct amdgpu_cu_info cu_info;
333 const struct amdgpu_gfx_funcs *funcs;
334
335 /* reset mask */
336 uint32_t grbm_soft_reset;
337 uint32_t srbm_soft_reset;
338
339 /* gfx off */
340 bool gfx_off_state; /* true: enabled, false: disabled */
341 struct mutex gfx_off_mutex; /* mutex to change gfxoff state */
342 uint32_t gfx_off_req_count; /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */
343 struct delayed_work gfx_off_delay_work; /* async work to set gfx block off */
344 uint32_t gfx_off_residency; /* last logged residency */
345 uint64_t gfx_off_entrycount; /* count of times GPU has get into GFXOFF state */
346
347 /* pipe reservation */
348 struct mutex pipe_reserve_mutex;
349 DECLARE_BITMAP (pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
350
351 /*ras */
352 struct ras_common_if *ras_if;
353 struct amdgpu_gfx_ras *ras;
354
355 bool is_poweron;
356
357 struct amdgpu_ring sw_gfx_ring[AMDGPU_MAX_SW_GFX_RINGS];
358 struct amdgpu_ring_mux muxer;
359};
360
361#define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))
362#define amdgpu_gfx_select_se_sh(adev, se, sh, instance) (adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance))
363#define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid) (adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid))
364#define amdgpu_gfx_init_spm_golden(adev) (adev)->gfx.funcs->init_spm_golden((adev))
365
366/**
367 * amdgpu_gfx_create_bitmask - create a bitmask
368 *
369 * @bit_width: length of the mask
370 *
371 * create a variable length bit mask.
372 * Returns the bitmask.
373 */
374static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width)
375{
376 return (u32)((1ULL << bit_width) - 1);
377}
378
379void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se,
380 unsigned max_sh);
381
382int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev,
383 struct amdgpu_ring *ring,
384 struct amdgpu_irq_src *irq);
385
386void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring);
387
388void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev);
389int amdgpu_gfx_kiq_init(struct amdgpu_device *adev,
390 unsigned hpd_size);
391
392int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
393 unsigned mqd_size);
394void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev);
395int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev);
396int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev);
397
398void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev);
399void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev);
400
401int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec,
402 int pipe, int queue);
403void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
404 int *mec, int *pipe, int *queue);
405bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int mec,
406 int pipe, int queue);
407bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev,
408 struct amdgpu_ring *ring);
409bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev,
410 struct amdgpu_ring *ring);
411int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev, int me,
412 int pipe, int queue);
413void amdgpu_gfx_bit_to_me_queue(struct amdgpu_device *adev, int bit,
414 int *me, int *pipe, int *queue);
415bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me,
416 int pipe, int queue);
417void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable);
418int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value);
419int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block);
420void amdgpu_gfx_ras_fini(struct amdgpu_device *adev);
421int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value);
422int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *residency);
423int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value);
424int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev,
425 void *err_data,
426 struct amdgpu_iv_entry *entry);
427int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
428 struct amdgpu_irq_src *source,
429 struct amdgpu_iv_entry *entry);
430uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg);
431void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v);
432int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev);
433void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id);
434
435#endif