1 /* SPDX-License-Identifier: MIT */
3 * Copyright © 2019 Intel Corporation
6 #ifndef _I915_PERF_TYPES_H_
7 #define _I915_PERF_TYPES_H_
9 #include <linux/atomic.h>
10 #include <linux/device.h>
11 #include <linux/hrtimer.h>
12 #include <linux/llist.h>
13 #include <linux/poll.h>
14 #include <linux/sysfs.h>
15 #include <linux/types.h>
16 #include <linux/uuid.h>
17 #include <linux/wait.h>
19 #include "gt/intel_sseu.h"
21 #include "intel_wakeref.h"
23 struct drm_i915_private;
26 struct i915_gem_context;
30 struct intel_engine_cs;
32 struct i915_oa_format {
42 struct i915_oa_config {
43 struct i915_perf *perf;
45 char uuid[UUID_STRING_LEN + 1];
48 const struct i915_oa_reg *mux_regs;
50 const struct i915_oa_reg *b_counter_regs;
51 u32 b_counter_regs_len;
52 const struct i915_oa_reg *flex_regs;
55 struct attribute_group sysfs_metric;
56 struct attribute *attrs[2];
57 struct device_attribute sysfs_metric_id;
63 struct i915_perf_stream;
66 * struct i915_perf_stream_ops - the OPs to support a specific stream type
68 struct i915_perf_stream_ops {
70 * @enable: Enables the collection of HW samples, either in response to
71 * `I915_PERF_IOCTL_ENABLE` or implicitly called when stream is opened
72 * without `I915_PERF_FLAG_DISABLED`.
74 void (*enable)(struct i915_perf_stream *stream);
77 * @disable: Disables the collection of HW samples, either in response
78 * to `I915_PERF_IOCTL_DISABLE` or implicitly called before destroying
81 void (*disable)(struct i915_perf_stream *stream);
84 * @poll_wait: Call poll_wait, passing a wait queue that will be woken
85 * once there is something ready to read() for the stream
87 void (*poll_wait)(struct i915_perf_stream *stream,
92 * @wait_unlocked: For handling a blocking read, wait until there is
93 * something to ready to read() for the stream. E.g. wait on the same
94 * wait queue that would be passed to poll_wait().
96 int (*wait_unlocked)(struct i915_perf_stream *stream);
99 * @read: Copy buffered metrics as records to userspace
100 * **buf**: the userspace, destination buffer
101 * **count**: the number of bytes to copy, requested by userspace
102 * **offset**: zero at the start of the read, updated as the read
103 * proceeds, it represents how many bytes have been copied so far and
104 * the buffer offset for copying the next record.
106 * Copy as many buffered i915 perf samples and records for this stream
107 * to userspace as will fit in the given buffer.
109 * Only write complete records; returning -%ENOSPC if there isn't room
110 * for a complete record.
112 * Return any error condition that results in a short read such as
113 * -%ENOSPC or -%EFAULT, even though these may be squashed before
114 * returning to userspace.
116 int (*read)(struct i915_perf_stream *stream,
122 * @destroy: Cleanup any stream specific resources.
124 * The stream will always be disabled before this is called.
126 void (*destroy)(struct i915_perf_stream *stream);
130 * struct i915_perf_stream - state for a single open stream FD
132 struct i915_perf_stream {
134 * @perf: i915_perf backpointer
136 struct i915_perf *perf;
139 * @uncore: mmio access path
141 struct intel_uncore *uncore;
144 * @engine: Engine associated with this performance stream.
146 struct intel_engine_cs *engine;
149 * @sample_flags: Flags representing the `DRM_I915_PERF_PROP_SAMPLE_*`
150 * properties given when opening a stream, representing the contents
151 * of a single sample as read() by userspace.
156 * @sample_size: Considering the configured contents of a sample
157 * combined with the required header size, this is the total size
158 * of a single sample record.
163 * @ctx: %NULL if measuring system-wide across all contexts or a
164 * specific context that is being monitored.
166 struct i915_gem_context *ctx;
169 * @enabled: Whether the stream is currently enabled, considering
170 * whether the stream was opened in a disabled state and based
171 * on `I915_PERF_IOCTL_ENABLE` and `I915_PERF_IOCTL_DISABLE` calls.
176 * @hold_preemption: Whether preemption is put on hold for command
177 * submissions done on the @ctx. This is useful for some drivers that
178 * cannot easily post process the OA buffer context to subtract delta
179 * of performance counters not associated with @ctx.
181 bool hold_preemption;
184 * @ops: The callbacks providing the implementation of this specific
185 * type of configured stream.
187 const struct i915_perf_stream_ops *ops;
190 * @oa_config: The OA configuration used by the stream.
192 struct i915_oa_config *oa_config;
195 * @oa_config_bos: A list of struct i915_oa_config_bo allocated lazily
196 * each time @oa_config changes.
198 struct llist_head oa_config_bos;
201 * @pinned_ctx: The OA context specific information.
203 struct intel_context *pinned_ctx;
206 * @specific_ctx_id: The id of the specific context.
211 * @specific_ctx_id_mask: The mask used to masking specific_ctx_id bits.
213 u32 specific_ctx_id_mask;
216 * @poll_check_timer: High resolution timer that will periodically
217 * check for data in the circular OA buffer for notifying userspace
218 * (e.g. during a read() or poll()).
220 struct hrtimer poll_check_timer;
223 * @poll_wq: The wait queue that hrtimer callback wakes when it
224 * sees data ready to read in the circular OA buffer.
226 wait_queue_head_t poll_wq;
229 * @pollin: Whether there is data available to read.
234 * @periodic: Whether periodic sampling is currently enabled.
239 * @period_exponent: The OA unit sampling frequency is derived from this.
244 * @oa_buffer: State of the OA buffer.
247 struct i915_vma *vma;
255 * @ptr_lock: Locks reads and writes to all head/tail state
257 * Consider: the head and tail pointer state needs to be read
258 * consistently from a hrtimer callback (atomic context) and
259 * read() fop (user context) with tail pointer updates happening
260 * in atomic context and head updates in user context and the
261 * (unlikely) possibility of read() errors needing to reset all
264 * Note: Contention/performance aren't currently a significant
265 * concern here considering the relatively low frequency of
266 * hrtimer callbacks (5ms period) and that reads typically only
267 * happen in response to a hrtimer event and likely complete
268 * before the next callback.
270 * Note: This lock is not held *while* reading and copying data
271 * to userspace so the value of head observed in htrimer
272 * callbacks won't represent any partial consumption of data.
277 * @aging_tail: The last HW tail reported by HW. The data
278 * might not have made it to memory yet though.
283 * @aging_timestamp: A monotonic timestamp for when the current aging tail pointer
284 * was read; used to determine when it is old enough to trust.
289 * @head: Although we can always read back the head pointer register,
290 * we prefer to avoid trusting the HW state, just to avoid any
291 * risk that some hardware condition could * somehow bump the
292 * head pointer unpredictably and cause us to forward the wrong
293 * OA buffer data to userspace.
298 * @tail: The last verified tail that can be read by userspace.
304 * @noa_wait: A batch buffer doing a wait on the GPU for the NOA logic to be
307 struct i915_vma *noa_wait;
310 * @poll_oa_period: The period in nanoseconds at which the OA
311 * buffer should be checked for available data.
317 * struct i915_oa_ops - Gen specific implementation of an OA unit stream
321 * @is_valid_b_counter_reg: Validates register's address for
322 * programming boolean counters for a particular platform.
324 bool (*is_valid_b_counter_reg)(struct i915_perf *perf, u32 addr);
327 * @is_valid_mux_reg: Validates register's address for programming mux
328 * for a particular platform.
330 bool (*is_valid_mux_reg)(struct i915_perf *perf, u32 addr);
333 * @is_valid_flex_reg: Validates register's address for programming
334 * flex EU filtering for a particular platform.
336 bool (*is_valid_flex_reg)(struct i915_perf *perf, u32 addr);
339 * @enable_metric_set: Selects and applies any MUX configuration to set
340 * up the Boolean and Custom (B/C) counters that are part of the
341 * counter reports being sampled. May apply system constraints such as
342 * disabling EU clock gating as required.
344 int (*enable_metric_set)(struct i915_perf_stream *stream,
345 struct i915_active *active);
348 * @disable_metric_set: Remove system constraints associated with using
351 void (*disable_metric_set)(struct i915_perf_stream *stream);
354 * @oa_enable: Enable periodic sampling
356 void (*oa_enable)(struct i915_perf_stream *stream);
359 * @oa_disable: Disable periodic sampling
361 void (*oa_disable)(struct i915_perf_stream *stream);
364 * @read: Copy data from the circular OA buffer into a given userspace
367 int (*read)(struct i915_perf_stream *stream,
373 * @oa_hw_tail_read: read the OA tail pointer register
375 * In particular this enables us to share all the fiddly code for
376 * handling the OA unit tail pointer race that affects multiple
379 u32 (*oa_hw_tail_read)(struct i915_perf_stream *stream);
383 struct drm_i915_private *i915;
385 struct kobject *metrics_kobj;
388 * Lock associated with adding/modifying/removing OA configs
389 * in perf->metrics_idr.
391 struct mutex metrics_lock;
394 * List of dynamic configurations (struct i915_oa_config), you
395 * need to hold perf->metrics_lock to access it.
397 struct idr metrics_idr;
400 * Lock associated with anything below within this structure
401 * except exclusive_stream.
406 * The stream currently using the OA unit. If accessed
407 * outside a syscall associated to its file
410 struct i915_perf_stream *exclusive_stream;
413 * @sseu: sseu configuration selected to run while perf is active,
414 * applies to all contexts.
416 struct intel_sseu sseu;
419 * For rate limiting any notifications of spurious
422 struct ratelimit_state spurious_report_rs;
425 * For rate limiting any notifications of tail pointer
428 struct ratelimit_state tail_pointer_race;
430 u32 gen7_latched_oastatus1;
431 u32 ctx_oactxctrl_offset;
432 u32 ctx_flexeu0_offset;
435 * The RPT_ID/reason field for Gen8+ includes a bit
436 * to determine if the CTX ID in the report is valid
437 * but the specific bit differs between Gen 8 and 9
439 u32 gen8_valid_ctx_bit;
441 struct i915_oa_ops ops;
442 const struct i915_oa_format *oa_formats;
444 atomic64_t noa_programming_delay;
447 #endif /* _I915_PERF_TYPES_H_ */