forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-05-13 9d77db3c730780c8ef5ccd4b66403ff5675cfe4e 
 kernel/drivers/gpu/drm/i915/i915_gpu_error.h
....@@ -7,15 +7,18 @@
77 #ifndef _I915_GPU_ERROR_H_
88 #define _I915_GPU_ERROR_H_
99 
10
+#include <linux/atomic.h>
1011 #include <linux/kref.h>
1112 #include <linux/ktime.h>
1213 #include <linux/sched.h>
1314 
1415 #include <drm/drm_mm.h>
1516 
17
+#include "gt/intel_engine.h"
18
+#include "gt/intel_gt_types.h"
19
+#include "gt/uc/intel_uc_fw.h"
20
+
1621 #include "intel_device_info.h"
17
-#include "intel_ringbuffer.h"
18
-#include "intel_uc_fw.h"
1922 
2023 #include "i915_gem.h"
2124 #include "i915_gem_gtt.h"
....@@ -23,43 +26,106 @@
2326 #include "i915_scheduler.h"
2427 
2528 struct drm_i915_private;
29
+struct i915_vma_compress;
30
+struct intel_engine_capture_vma;
2631 struct intel_overlay_error_state;
2732 struct intel_display_error_state;
2833 
29
-struct i915_gpu_state {
30
-	struct kref ref;
31
-	ktime_t time;
32
-	ktime_t boottime;
33
-	ktime_t uptime;
34
-	unsigned long capture;
35
-	unsigned long epoch;
34
+struct i915_vma_coredump {
35
+	struct i915_vma_coredump *next;
3636 
37
-	struct drm_i915_private *i915;
37
+	char name[20];
3838 
39
-	char error_msg[128];
39
+	u64 gtt_offset;
40
+	u64 gtt_size;
41
+	u32 gtt_page_sizes;
42
+
43
+	int num_pages;
44
+	int page_count;
45
+	int unused;
46
+	u32 *pages[];
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 intel_engine_coredump {
60
+	const struct intel_engine_cs *engine;
61
+
4062 	bool simulated;
41
-	bool awake;
42
-	bool wakelock;
43
-	bool suspended;
44
-	int iommu;
4563 	u32 reset_count;
46
-	u32 suspend_count;
47
-	struct intel_device_info device_info;
48
-	struct intel_driver_caps driver_caps;
49
-	struct i915_params params;
5064 
51
-	struct i915_error_uc {
52
-		struct intel_uc_fw guc_fw;
53
-		struct intel_uc_fw huc_fw;
54
-		struct drm_i915_error_object *guc_log;
55
-	} uc;
65
+	/* position of active request inside the ring */
66
+	u32 rq_head, rq_post, rq_tail;
67
+
68
+	/* Register state */
69
+	u32 ccid;
70
+	u32 start;
71
+	u32 tail;
72
+	u32 head;
73
+	u32 ctl;
74
+	u32 mode;
75
+	u32 hws;
76
+	u32 ipeir;
77
+	u32 ipehr;
78
+	u32 esr;
79
+	u32 bbstate;
80
+	u32 instpm;
81
+	u32 instps;
82
+	u64 bbaddr;
83
+	u64 acthd;
84
+	u32 fault_reg;
85
+	u64 faddr;
86
+	u32 rc_psmi; /* sleep state */
87
+	struct intel_instdone instdone;
88
+
89
+	struct i915_gem_context_coredump {
90
+		char comm[TASK_COMM_LEN];
91
+
92
+		u64 total_runtime;
93
+		u32 avg_runtime;
94
+
95
+		pid_t pid;
96
+		int active;
97
+		int guilty;
98
+		struct i915_sched_attr sched_attr;
99
+	} context;
100
+
101
+	struct i915_vma_coredump *vma;
102
+
103
+	struct i915_request_coredump execlist[EXECLIST_MAX_PORTS];
104
+	unsigned int num_ports;
105
+
106
+	struct {
107
+		u32 gfx_mode;
108
+		union {
109
+			u64 pdp[4];
110
+			u32 pp_dir_base;
111
+		};
112
+	} vm_info;
113
+
114
+	struct intel_engine_coredump *next;
115
+};
116
+
117
+struct intel_gt_coredump {
118
+	const struct intel_gt *_gt;
119
+	bool awake;
120
+	bool simulated;
121
+
122
+	struct intel_gt_info info;
56123 
57124 	/* Generic register state */
58125 	u32 eir;
59126 	u32 pgtbl_er;
60127 	u32 ier;
61128 	u32 gtier[6], ngtier;
62
-	u32 ccid;
63129 	u32 derrmr;
64130 	u32 forcewake;
65131 	u32 error; /* gen6+ */
....@@ -71,297 +137,213 @@
71137 	u32 gam_ecochk;
72138 	u32 gab_ctl;
73139 	u32 gfx_mode;
140
+	u32 gtt_cache;
141
+	u32 aux_err; /* gen12 */
142
+	u32 sfc_done[GEN12_SFC_DONE_MAX]; /* gen12 */
143
+	u32 gam_done; /* gen12 */
74144 
75145 	u32 nfence;
76146 	u64 fence[I915_MAX_NUM_FENCES];
147
+
148
+	struct intel_engine_coredump *engine;
149
+
150
+	struct intel_uc_coredump {
151
+		struct intel_uc_fw guc_fw;
152
+		struct intel_uc_fw huc_fw;
153
+		struct i915_vma_coredump *guc_log;
154
+	} *uc;
155
+
156
+	struct intel_gt_coredump *next;
157
+};
158
+
159
+struct i915_gpu_coredump {
160
+	struct kref ref;
161
+	ktime_t time;
162
+	ktime_t boottime;
163
+	ktime_t uptime;
164
+	unsigned long capture;
165
+
166
+	struct drm_i915_private *i915;
167
+
168
+	struct intel_gt_coredump *gt;
169
+
170
+	char error_msg[128];
171
+	bool simulated;
172
+	bool wakelock;
173
+	bool suspended;
174
+	int iommu;
175
+	u32 reset_count;
176
+	u32 suspend_count;
177
+
178
+	struct intel_device_info device_info;
179
+	struct intel_runtime_info runtime_info;
180
+	struct intel_driver_caps driver_caps;
181
+	struct i915_params params;
182
+
77183 	struct intel_overlay_error_state *overlay;
78184 	struct intel_display_error_state *display;
79185 
80
-	struct drm_i915_error_engine {
81
-		int engine_id;
82
-		/* Software tracked state */
83
-		bool idle;
84
-		bool waiting;
85
-		int num_waiters;
86
-		unsigned long hangcheck_timestamp;
87
-		bool hangcheck_stalled;
88
-		enum intel_engine_hangcheck_action hangcheck_action;
89
-		struct i915_address_space *vm;
90
-		int num_requests;
91
-		u32 reset_count;
92
-
93
-		/* position of active request inside the ring */
94
-		u32 rq_head, rq_post, rq_tail;
95
-
96
-		/* our own tracking of ring head and tail */
97
-		u32 cpu_ring_head;
98
-		u32 cpu_ring_tail;
99
-
100
-		u32 last_seqno;
101
-
102
-		/* Register state */
103
-		u32 start;
104
-		u32 tail;
105
-		u32 head;
106
-		u32 ctl;
107
-		u32 mode;
108
-		u32 hws;
109
-		u32 ipeir;
110
-		u32 ipehr;
111
-		u32 bbstate;
112
-		u32 instpm;
113
-		u32 instps;
114
-		u32 seqno;
115
-		u64 bbaddr;
116
-		u64 acthd;
117
-		u32 fault_reg;
118
-		u64 faddr;
119
-		u32 rc_psmi; /* sleep state */
120
-		u32 semaphore_mboxes[I915_NUM_ENGINES - 1];
121
-		struct intel_instdone instdone;
122
-
123
-		struct drm_i915_error_context {
124
-			char comm[TASK_COMM_LEN];
125
-			pid_t pid;
126
-			u32 handle;
127
-			u32 hw_id;
128
-			int ban_score;
129
-			int active;
130
-			int guilty;
131
-			bool bannable;
132
-			struct i915_sched_attr sched_attr;
133
-		} context;
134
-
135
-		struct drm_i915_error_object {
136
-			u64 gtt_offset;
137
-			u64 gtt_size;
138
-			int num_pages;
139
-			int page_count;
140
-			int unused;
141
-			u32 *pages[0];
142
-		} *ringbuffer, *batchbuffer, *wa_batchbuffer, *ctx, *hws_page;
143
-
144
-		struct drm_i915_error_object **user_bo;
145
-		long user_bo_count;
146
-
147
-		struct drm_i915_error_object *wa_ctx;
148
-		struct drm_i915_error_object *default_state;
149
-
150
-		struct drm_i915_error_request {
151
-			long jiffies;
152
-			pid_t pid;
153
-			u32 context;
154
-			int ban_score;
155
-			u32 seqno;
156
-			u32 start;
157
-			u32 head;
158
-			u32 tail;
159
-			struct i915_sched_attr sched_attr;
160
-		} *requests, execlist[EXECLIST_MAX_PORTS];
161
-		unsigned int num_ports;
162
-
163
-		struct drm_i915_error_waiter {
164
-			char comm[TASK_COMM_LEN];
165
-			pid_t pid;
166
-			u32 seqno;
167
-		} *waiters;
168
-
169
-		struct {
170
-			u32 gfx_mode;
171
-			union {
172
-				u64 pdp[4];
173
-				u32 pp_dir_base;
174
-			};
175
-		} vm_info;
176
-	} engine[I915_NUM_ENGINES];
177
-
178
-	struct drm_i915_error_buffer {
179
-		u32 size;
180
-		u32 name;
181
-		u32 wseqno;
182
-		u64 gtt_offset;
183
-		u32 read_domains;
184
-		u32 write_domain;
185
-		s32 fence_reg:I915_MAX_NUM_FENCE_BITS;
186
-		u32 tiling:2;
187
-		u32 dirty:1;
188
-		u32 purgeable:1;
189
-		u32 userptr:1;
190
-		s32 engine:4;
191
-		u32 cache_level:3;
192
-	} *active_bo[I915_NUM_ENGINES], *pinned_bo;
193
-	u32 active_bo_count[I915_NUM_ENGINES], pinned_bo_count;
194
-	struct i915_address_space *active_vm[I915_NUM_ENGINES];
186
+	struct scatterlist *sgl, *fit;
195187 };
196188 
197189 struct i915_gpu_error {
198
-	/* For hangcheck timer */
199
-#define DRM_I915_HANGCHECK_PERIOD 1500 /* in ms */
200
-#define DRM_I915_HANGCHECK_JIFFIES msecs_to_jiffies(DRM_I915_HANGCHECK_PERIOD)
201
-
202
-	struct delayed_work hangcheck_work;
203
-
204190 	/* For reset and error_state handling. */
205191 	spinlock_t lock;
206192 	/* Protected by the above dev->gpu_error.lock. */
207
-	struct i915_gpu_state *first_error;
193
+	struct i915_gpu_coredump *first_error;
208194 
209195 	atomic_t pending_fb_pin;
210196 
211
-	unsigned long missed_irq_rings;
212
-
213
-	/**
214
-	 * State variable controlling the reset flow and count
215
-	 *
216
-	 * This is a counter which gets incremented when reset is triggered,
217
-	 *
218
-	 * Before the reset commences, the I915_RESET_BACKOFF bit is set
219
-	 * meaning that any waiters holding onto the struct_mutex should
220
-	 * relinquish the lock immediately in order for the reset to start.
221
-	 *
222
-	 * If reset is not completed successfully, the I915_WEDGE bit is
223
-	 * set meaning that hardware is terminally sour and there is no
224
-	 * recovery. All waiters on the reset_queue will be woken when
225
-	 * that happens.
226
-	 *
227
-	 * This counter is used by the wait_seqno code to notice that reset
228
-	 * event happened and it needs to restart the entire ioctl (since most
229
-	 * likely the seqno it waited for won't ever signal anytime soon).
230
-	 *
231
-	 * This is important for lock-free wait paths, where no contended lock
232
-	 * naturally enforces the correct ordering between the bail-out of the
233
-	 * waiter and the gpu reset work code.
234
-	 */
235
-	unsigned long reset_count;
236
-
237
-	/**
238
-	 * flags: Control various stages of the GPU reset
239
-	 *
240
-	 * #I915_RESET_BACKOFF - When we start a reset, we want to stop any
241
-	 * other users acquiring the struct_mutex. To do this we set the
242
-	 * #I915_RESET_BACKOFF bit in the error flags when we detect a reset
243
-	 * and then check for that bit before acquiring the struct_mutex (in
244
-	 * i915_mutex_lock_interruptible()?). I915_RESET_BACKOFF serves a
245
-	 * secondary role in preventing two concurrent global reset attempts.
246
-	 *
247
-	 * #I915_RESET_HANDOFF - To perform the actual GPU reset, we need the
248
-	 * struct_mutex. We try to acquire the struct_mutex in the reset worker,
249
-	 * but it may be held by some long running waiter (that we cannot
250
-	 * interrupt without causing trouble). Once we are ready to do the GPU
251
-	 * reset, we set the I915_RESET_HANDOFF bit and wakeup any waiters. If
252
-	 * they already hold the struct_mutex and want to participate they can
253
-	 * inspect the bit and do the reset directly, otherwise the worker
254
-	 * waits for the struct_mutex.
255
-	 *
256
-	 * #I915_RESET_ENGINE[num_engines] - Since the driver doesn't need to
257
-	 * acquire the struct_mutex to reset an engine, we need an explicit
258
-	 * flag to prevent two concurrent reset attempts in the same engine.
259
-	 * As the number of engines continues to grow, allocate the flags from
260
-	 * the most significant bits.
261
-	 *
262
-	 * #I915_WEDGED - If reset fails and we can no longer use the GPU,
263
-	 * we set the #I915_WEDGED bit. Prior to command submission, e.g.
264
-	 * i915_request_alloc(), this bit is checked and the sequence
265
-	 * aborted (with -EIO reported to userspace) if set.
266
-	 */
267
-	unsigned long flags;
268
-#define I915_RESET_BACKOFF	0
269
-#define I915_RESET_HANDOFF	1
270
-#define I915_RESET_MODESET	2
271
-#define I915_WEDGED		(BITS_PER_LONG - 1)
272
-#define I915_RESET_ENGINE	(I915_WEDGED - I915_NUM_ENGINES)
197
+	/** Number of times the device has been reset (global) */
198
+	atomic_t reset_count;
273199 
274200 	/** Number of times an engine has been reset */
275
-	u32 reset_engine_count[I915_NUM_ENGINES];
276
-
277
-	/** Set of stalled engines with guilty requests, in the current reset */
278
-	u32 stalled_mask;
279
-
280
-	/** Reason for the current *global* reset */
281
-	const char *reason;
282
-
283
-	/**
284
-	 * Waitqueue to signal when a hang is detected. Used to for waiters
285
-	 * to release the struct_mutex for the reset to procede.
286
-	 */
287
-	wait_queue_head_t wait_queue;
288
-
289
-	/**
290
-	 * Waitqueue to signal when the reset has completed. Used by clients
291
-	 * that wait for dev_priv->mm.wedged to settle.
292
-	 */
293
-	wait_queue_head_t reset_queue;
294
-
295
-	/* For missed irq/seqno simulation. */
296
-	unsigned long test_irq_rings;
201
+	atomic_t reset_engine_count[I915_NUM_ENGINES];
297202 };
298203 
299204 struct drm_i915_error_state_buf {
300205 	struct drm_i915_private *i915;
301
-	unsigned int bytes;
302
-	unsigned int size;
206
+	struct scatterlist *sgl, *cur, *end;
207
+
208
+	char *buf;
209
+	size_t bytes;
210
+	size_t size;
211
+	loff_t iter;
212
+
303213 	int err;
304
-	u8 *buf;
305
-	loff_t start;
306
-	loff_t pos;
307214 };
308215 
309216 #if IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR)
310217 
311218 __printf(2, 3)
312219 void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...);
313
-int i915_error_state_to_str(struct drm_i915_error_state_buf *estr,
314
-			    const struct i915_gpu_state *gpu);
315
-int i915_error_state_buf_init(struct drm_i915_error_state_buf *eb,
316
-			      struct drm_i915_private *i915,
317
-			      size_t count, loff_t pos);
318220 
319
-static inline void
320
-i915_error_state_buf_release(struct drm_i915_error_state_buf *eb)
321
-{
322
-	kfree(eb->buf);
323
-}
221
+struct i915_gpu_coredump *i915_gpu_coredump(struct drm_i915_private *i915);
222
+void i915_capture_error_state(struct drm_i915_private *i915);
324223 
325
-struct i915_gpu_state *i915_capture_gpu_state(struct drm_i915_private *i915);
326
-void i915_capture_error_state(struct drm_i915_private *dev_priv,
327
-			      u32 engine_mask,
328
-			      const char *error_msg);
224
+struct i915_gpu_coredump *
225
+i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp);
329226 
330
-static inline struct i915_gpu_state *
331
-i915_gpu_state_get(struct i915_gpu_state *gpu)
227
+struct intel_gt_coredump *
228
+intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp);
229
+
230
+struct intel_engine_coredump *
231
+intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp);
232
+
233
+struct intel_engine_capture_vma *
234
+intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
235
+				  struct i915_request *rq,
236
+				  gfp_t gfp);
237
+
238
+void intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
239
+				   struct intel_engine_capture_vma *capture,
240
+				   struct i915_vma_compress *compress);
241
+
242
+struct i915_vma_compress *
243
+i915_vma_capture_prepare(struct intel_gt_coredump *gt);
244
+
245
+void i915_vma_capture_finish(struct intel_gt_coredump *gt,
246
+			     struct i915_vma_compress *compress);
247
+
248
+void i915_error_state_store(struct i915_gpu_coredump *error);
249
+
250
+static inline struct i915_gpu_coredump *
251
+i915_gpu_coredump_get(struct i915_gpu_coredump *gpu)
332252 {
333253 	kref_get(&gpu->ref);
334254 	return gpu;
335255 }
336256 
337
-void __i915_gpu_state_free(struct kref *kref);
338
-static inline void i915_gpu_state_put(struct i915_gpu_state *gpu)
257
+ssize_t
258
+i915_gpu_coredump_copy_to_buffer(struct i915_gpu_coredump *error,
259
+				 char *buf, loff_t offset, size_t count);
260
+
261
+void __i915_gpu_coredump_free(struct kref *kref);
262
+static inline void i915_gpu_coredump_put(struct i915_gpu_coredump *gpu)
339263 {
340264 	if (gpu)
341
-		kref_put(&gpu->ref, __i915_gpu_state_free);
265
+		kref_put(&gpu->ref, __i915_gpu_coredump_free);
342266 }
343267 
344
-struct i915_gpu_state *i915_first_error_state(struct drm_i915_private *i915);
268
+struct i915_gpu_coredump *i915_first_error_state(struct drm_i915_private *i915);
345269 void i915_reset_error_state(struct drm_i915_private *i915);
270
+void i915_disable_error_state(struct drm_i915_private *i915, int err);
346271 
347272 #else
348273 
349
-static inline void i915_capture_error_state(struct drm_i915_private *dev_priv,
350
-					    u32 engine_mask,
351
-					    const char *error_msg)
274
+static inline void i915_capture_error_state(struct drm_i915_private *i915)
352275 {
353276 }
354277 
355
-static inline struct i915_gpu_state *
356
-i915_first_error_state(struct drm_i915_private *i915)
278
+static inline struct i915_gpu_coredump *
279
+i915_gpu_coredump_alloc(struct drm_i915_private *i915, gfp_t gfp)
357280 {
358281 	return NULL;
359282 }
360283 
284
+static inline struct intel_gt_coredump *
285
+intel_gt_coredump_alloc(struct intel_gt *gt, gfp_t gfp)
286
+{
287
+	return NULL;
288
+}
289
+
290
+static inline struct intel_engine_coredump *
291
+intel_engine_coredump_alloc(struct intel_engine_cs *engine, gfp_t gfp)
292
+{
293
+	return NULL;
294
+}
295
+
296
+static inline struct intel_engine_capture_vma *
297
+intel_engine_coredump_add_request(struct intel_engine_coredump *ee,
298
+				  struct i915_request *rq,
299
+				  gfp_t gfp)
300
+{
301
+	return NULL;
302
+}
303
+
304
+static inline void
305
+intel_engine_coredump_add_vma(struct intel_engine_coredump *ee,
306
+			      struct intel_engine_capture_vma *capture,
307
+			      struct i915_vma_compress *compress)
308
+{
309
+}
310
+
311
+static inline struct i915_vma_compress *
312
+i915_vma_capture_prepare(struct intel_gt_coredump *gt)
313
+{
314
+	return NULL;
315
+}
316
+
317
+static inline void
318
+i915_vma_capture_finish(struct intel_gt_coredump *gt,
319
+			struct i915_vma_compress *compress)
320
+{
321
+}
322
+
323
+static inline void
324
+i915_error_state_store(struct i915_gpu_coredump *error)
325
+{
326
+}
327
+
328
+static inline void i915_gpu_coredump_put(struct i915_gpu_coredump *gpu)
329
+{
330
+}
331
+
332
+static inline struct i915_gpu_coredump *
333
+i915_first_error_state(struct drm_i915_private *i915)
334
+{
335
+	return ERR_PTR(-ENODEV);
336
+}
337
+
361338 static inline void i915_reset_error_state(struct drm_i915_private *i915)
362339 {
363340 }
364341 
342
+static inline void i915_disable_error_state(struct drm_i915_private *i915,
343
+					    int err)
344
+{
345
+}
346
+
365347 #endif /* IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR) */
366348 
367349 #endif /* _I915_GPU_ERROR_H_ */