xref: /linux/drivers/gpu/drm/i915/i915_gpu_error.h (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca) 
1 /*
2  * SPDX-License-Identifier: MIT
3  *
4  * Copyright © 2008-2018 Intel Corporation
5  */
6 
7 #ifndef _I915_GPU_ERROR_H_
8 #define _I915_GPU_ERROR_H_
9 
10 #include <linux/atomic.h>
11 #include <linux/kref.h>
12 #include <linux/ktime.h>
13 #include <linux/sched.h>
14 
15 #include <drm/drm_mm.h>
16 
17 #include "display/intel_display_device.h"
18 #include "display/intel_display_params.h"
19 #include "gt/intel_engine.h"
20 #include "gt/intel_engine_types.h"
21 #include "gt/intel_gt_types.h"
22 #include "gt/uc/intel_uc_fw.h"
23 
24 #include "intel_device_info.h"
25 
26 #include "i915_gem.h"
27 #include "i915_gem_gtt.h"
28 #include "i915_params.h"
29 #include "i915_scheduler.h"
30 
31 struct drm_i915_private;
32 struct i915_vma_compress;
33 struct intel_engine_capture_vma;
34 struct intel_overlay_error_state;
35 
36 struct i915_vma_coredump {
37 	struct i915_vma_coredump *next;
38 
39 	char name[20];
40 
41 	u64 gtt_offset;
42 	u64 gtt_size;
43 	u32 gtt_page_sizes;
44 
45 	int unused;
46 	struct list_head page_list;
47 };
48 
49 struct i915_request_coredump {
50 	unsigned long flags;
51 	pid_t pid;
52 	u32 context;
53 	u32 seqno;
54 	u32 head;
55 	u32 tail;
56 	struct i915_sched_attr sched_attr;
57 };
58 
59 struct __guc_capture_parsed_output;
60 
61 struct intel_engine_coredump {
62 	const struct intel_engine_cs *engine;
63 
64 	bool hung;
65 	bool simulated;
66 	u32 reset_count;
67 
68 	/* position of active request inside the ring */
69 	u32 rq_head, rq_post, rq_tail;
70 
71 	/* Register state */
72 	u32 ccid;
73 	u32 start;
74 	u32 tail;
75 	u32 head;
76 	u32 ctl;
77 	u32 mode;
78 	u32 hws;
79 	u32 ipeir;
80 	u32 ipehr;
81 	u32 esr;
82 	u32 bbstate;
83 	u32 instpm;
84 	u32 instps;
85 	u64 bbaddr;
86 	u64 acthd;
87 	u32 fault_reg;
88 	u64 faddr;
89 	u32 rc_psmi; /* sleep state */
90 	u32 nopid;
91 	u32 excc;
92 	u32 cmd_cctl;
93 	u32 cscmdop;
94 	u32 ctx_sr_ctl;
95 	u32 dma_faddr_hi;
96 	u32 dma_faddr_lo;
97 	struct intel_instdone instdone;
98 
99 	/* GuC matched capture-lists info */
100 	struct intel_guc_state_capture *guc_capture;
101 	struct __guc_capture_parsed_output *guc_capture_node;
102 
103 	struct i915_gem_context_coredump {
104 		char comm[TASK_COMM_LEN];
105 
106 		u64 total_runtime;
107 		u64 avg_runtime;
108 
109 		pid_t pid;
110 		int active;
111 		int guilty;
112 		struct i915_sched_attr sched_attr;
113 		u32 hwsp_seqno;
114 	} context;
115 
116 	struct i915_vma_coredump *vma;
117 
118 	struct i915_request_coredump execlist[EXECLIST_MAX_PORTS];
119 	unsigned int num_ports;
120 
121 	struct {
122 		u32 gfx_mode;
123 		union {
124 			u64 pdp[4];
125 			u32 pp_dir_base;
126 		};
127 	} vm_info;
128 
129 	struct intel_engine_coredump *next;
130 };
131 
132 struct intel_ctb_coredump {
133 	u32 raw_head, head;
134 	u32 raw_tail, tail;
135 	u32 raw_status;
136 	u32 desc_offset;
137 	u32 cmds_offset;
138 	u32 size;
139 };
140 
141 struct intel_gt_coredump {
142 	const struct intel_gt *_gt;
143 	bool awake;
144 	bool simulated;
145 
146 	struct intel_gt_info info;
147 
148 	/* Generic register state */
149 	u32 eir;
150 	u32 pgtbl_er;
151 	u32 ier;
152 	u32 gtier[6], ngtier;
153 	u32 forcewake;
154 	u32 error; /* gen6+ */
155 	u32 err_int; /* gen7 */
156 	u32 fault_data0; /* gen8, gen9 */
157 	u32 fault_data1; /* gen8, gen9 */
158 	u32 done_reg;
159 	u32 gac_eco;
160 	u32 gam_ecochk;
161 	u32 gab_ctl;
162 	u32 gfx_mode;
163 	u32 gtt_cache;
164 	u32 aux_err; /* gen12 */
165 	u32 gam_done; /* gen12 */
166 	u32 clock_frequency;
167 	u32 clock_period_ns;
168 
169 	/* Display related */
170 	u32 derrmr;
171 	u32 sfc_done[I915_MAX_SFC]; /* gen12 */
172 
173 	u32 nfence;
174 	u64 fence[I915_MAX_NUM_FENCES];
175 
176 	struct intel_engine_coredump *engine;
177 
178 	struct intel_uc_coredump {
179 		struct intel_uc_fw guc_fw;
180 		struct intel_uc_fw huc_fw;
181 		struct guc_info {
182 			struct intel_ctb_coredump ctb[2];
183 			struct i915_vma_coredump *vma_ctb;
184 			struct i915_vma_coredump *vma_log;
185 			u32 timestamp;
186 			u16 last_fence;
187 			bool is_guc_capture;
188 		} guc;
189 	} *uc;
190 
191 	struct intel_gt_coredump *next;
192 };
193 
194 struct i915_gpu_coredump {
195 	struct kref ref;
196 	ktime_t time;
197 	ktime_t boottime;
198 	ktime_t uptime;
199 	unsigned long capture;
200 
201 	struct drm_i915_private *i915;
202 
203 	struct intel_gt_coredump *gt;
204 
205 	char error_msg[128];
206 	bool simulated;
207 	bool wakelock;
208 	bool suspended;
209 	int iommu;
210 	u32 reset_count;
211 	u32 suspend_count;
212 
213 	struct intel_device_info device_info;
214 	struct intel_runtime_info runtime_info;
215 	struct intel_display_device_info display_device_info;
216 	struct intel_display_runtime_info display_runtime_info;
217 	struct intel_driver_caps driver_caps;
218 	struct i915_params params;
219 	struct intel_display_params display_params;
220 
221 	struct intel_overlay_error_state *overlay;
222 
223 	struct scatterlist *sgl, *fit;
224 };
225 
226 struct i915_gpu_error {
227 	/* For reset and error_state handling. */
228 	spinlock_t lock;
229 	/* Protected by the above dev->gpu_error.lock. */
230 	struct i915_gpu_coredump *first_error;
231 
232 	atomic_t pending_fb_pin;
233 
234 	/** Number of times the device has been reset (global) */
235 	atomic_t reset_count;
236 
237 	/** Number of times an engine has been reset */
238 	atomic_t reset_engine_count[MAX_ENGINE_CLASS];
239 };
240 
241 struct drm_i915_error_state_buf {
242 	struct drm_i915_private *i915;
243 	struct scatterlist *sgl, *cur, *end;
244 
245 	char *buf;
246 	size_t bytes;
247 	size_t size;
248 	loff_t iter;
249 
250 	int err;
251 };
252 
253 static inline u32 i915_reset_count(struct i915_gpu_error *error)
254 {
255 	return atomic_read(&error->reset_count);
256 }
257 
258 static inline u32 i915_reset_engine_count(struct i915_gpu_error *error,
259 					  const struct intel_engine_cs *engine)
260 {
261 	return atomic_read(&error->reset_engine_count[engine->class]);
262 }
263 
264 static inline void
265 i915_increase_reset_engine_count(struct i915_gpu_error *error,
266 				 const struct intel_engine_cs *engine)
267 {
268 	atomic_inc(&error->reset_engine_count[engine->class]);
269 }
270 
271 #define CORE_DUMP_FLAG_NONE           0x0
272 #define CORE_DUMP_FLAG_IS_GUC_CAPTURE BIT(0)
273 
274 #if IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR) && IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)
275 void intel_klog_error_capture(struct intel_gt *gt,
276 			      intel_engine_mask_t engine_mask);
277 #else
278 static inline void intel_klog_error_capture(struct intel_gt *gt,
279 					    intel_engine_mask_t engine_mask)
280 {
281 }
282 #endif
283 
284 #if IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR)
285 
286 __printf(2, 3)
287 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...);
288 
289 void i915_capture_error_state(struct intel_gt *gt,
290 			      intel_engine_mask_t engine_mask, u32 dump_flags);
291 
292 struct i915_gpu_coredump *
293 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp);
294 
295 struct intel_gt_coredump *
296 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp, u32 dump_flags);
297 
298 struct intel_engine_coredump *
299 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp, u32 dump_flags);
300 
301 struct intel_engine_capture_vma *
302 intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
303 				  struct i915_request *rq,
304 				  gfp_t gfp);
305 
306 void intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
307 				   struct intel_engine_capture_vma *capture,
308 				   struct i915_vma_compress *compress);
309 
310 struct i915_vma_compress *
311 i915_vma_capture_prepare(struct intel_gt_coredump *gt);
312 
313 void i915_vma_capture_finish(struct intel_gt_coredump *gt,
314 			     struct i915_vma_compress *compress);
315 
316 void i915_error_state_store(struct i915_gpu_coredump *error);
317 
318 static inline struct i915_gpu_coredump *
319 i915_gpu_coredump_get(struct i915_gpu_coredump *gpu)
320 {
321 	kref_get(&gpu->ref);
322 	return gpu;
323 }
324 
325 ssize_t
326 i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
327 				 char *buf, loff_t offset, size_t count);
328 
329 void __i915_gpu_coredump_free(struct kref *kref);
330 static inline void i915_gpu_coredump_put(struct i915_gpu_coredump *gpu)
331 {
332 	if (gpu)
333 		kref_put(&gpu->ref, __i915_gpu_coredump_free);
334 }
335 
336 void i915_reset_error_state(struct drm_i915_private *i915);
337 void i915_disable_error_state(struct drm_i915_private *i915, int err);
338 
339 void i915_gpu_error_debugfs_register(struct drm_i915_private *i915);
340 void i915_gpu_error_sysfs_setup(struct drm_i915_private *i915);
341 void i915_gpu_error_sysfs_teardown(struct drm_i915_private *i915);
342 
343 #else
344 
345 __printf(2, 3)
346 static inline void
347 i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
348 {
349 }
350 
351 static inline void
352 i915_capture_error_state(struct intel_gt *gt, intel_engine_mask_t engine_mask, u32 dump_flags)
353 {
354 }
355 
356 static inline struct i915_gpu_coredump *
357 i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
358 {
359 	return NULL;
360 }
361 
362 static inline struct intel_gt_coredump *
363 intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp, u32 dump_flags)
364 {
365 	return NULL;
366 }
367 
368 static inline struct intel_engine_coredump *
369 intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp, u32 dump_flags)
370 {
371 	return NULL;
372 }
373 
374 static inline struct intel_engine_capture_vma *
375 intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
376 				  struct i915_request *rq,
377 				  gfp_t gfp)
378 {
379 	return NULL;
380 }
381 
382 static inline void
383 intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
384 			      struct intel_engine_capture_vma *capture,
385 			      struct i915_vma_compress *compress)
386 {
387 }
388 
389 static inline struct i915_vma_compress *
390 i915_vma_capture_prepare(struct intel_gt_coredump *gt)
391 {
392 	return NULL;
393 }
394 
395 static inline void
396 i915_vma_capture_finish(struct intel_gt_coredump *gt,
397 			struct i915_vma_compress *compress)
398 {
399 }
400 
401 static inline void
402 i915_error_state_store(struct i915_gpu_coredump *error)
403 {
404 }
405 
406 static inline void i915_gpu_coredump_put(struct i915_gpu_coredump *gpu)
407 {
408 }
409 
410 static inline void i915_reset_error_state(struct drm_i915_private *i915)
411 {
412 }
413 
414 static inline void i915_disable_error_state(struct drm_i915_private *i915,
415 					    int err)
416 {
417 }
418 
419 static inline void i915_gpu_error_debugfs_register(struct drm_i915_private *i915)
420 {
421 }
422 
423 static inline void i915_gpu_error_sysfs_setup(struct drm_i915_private *i915)
424 {
425 }
426 
427 static inline void i915_gpu_error_sysfs_teardown(struct drm_i915_private *i915)
428 {
429 }
430 
431 #endif /* IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR) */
432 
433 #endif /* _I915_GPU_ERROR_H_ */
434