xref: /linux/tools/perf/Documentation/perf-arm-spe.txt (revision d30c1683aaecb93d2ab95685dc4300a33d3cea7a)
1perf-arm-spe(1)
2================
3
4NAME
5----
6perf-arm-spe - Support for Arm Statistical Profiling Extension within Perf tools
7
8SYNOPSIS
9--------
10[verse]
11'perf record' -e arm_spe//
12
13DESCRIPTION
14-----------
15
16The SPE (Statistical Profiling Extension) feature provides accurate attribution of latencies and
17 events down to individual instructions. Rather than being interrupt-driven, it picks an
18instruction to sample and then captures data for it during execution. Data includes execution time
19in cycles. For loads and stores it also includes data address, cache miss events, and data origin.
20
21The sampling has 5 stages:
22
23  1. Choose an operation
24  2. Collect data about the operation
25  3. Optionally discard the record based on a filter
26  4. Write the record to memory
27  5. Interrupt when the buffer is full
28
29Choose an operation
30~~~~~~~~~~~~~~~~~~~
31
32This is chosen from a sample population, for SPE this is an IMPLEMENTATION DEFINED choice of all
33architectural instructions or all micro-ops. Sampling happens at a programmable interval. The
34architecture provides a mechanism for the SPE driver to infer the minimum interval at which it should
35sample. This minimum interval is used by the driver if no interval is specified. A pseudo-random
36perturbation is also added to the sampling interval by default.
37
38Collect data about the operation
39~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
40
41Program counter, PMU events, timings and data addresses related to the operation are recorded.
42Sampling ensures there is only one sampled operation is in flight.
43
44Optionally discard the record based on a filter
45~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
46
47Based on programmable criteria, choose whether to keep the record or discard it. If the record is
48discarded then the flow stops here for this sample.
49
50Write the record to memory
51~~~~~~~~~~~~~~~~~~~~~~~~~~
52
53The record is appended to a memory buffer
54
55Interrupt when the buffer is full
56~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
57
58When the buffer fills, an interrupt is sent and the driver signals Perf to collect the records.
59Perf saves the raw data in the perf.data file.
60
61Opening the file
62----------------
63
64Up until this point no decoding of the SPE data was done by either the kernel or Perf. Only when the
65recorded file is opened with 'perf report' or 'perf script' does the decoding happen. When decoding
66the data, Perf generates "synthetic samples" as if these were generated at the time of the
67recording. These samples are the same as if normal sampling was done by Perf without using SPE,
68although they may have more attributes associated with them. For example a normal sample may have
69just the instruction pointer, but an SPE sample can have data addresses and latency attributes.
70
71Why Sampling?
72-------------
73
74 - Sampling, rather than tracing, cuts down the profiling problem to something more manageable for
75 hardware. Only one sampled operation is in flight at a time.
76
77 - Allows precise attribution data, including: Full PC of instruction, data virtual and physical
78 addresses.
79
80 - Allows correlation between an instruction and events, such as TLB and cache miss. (Data source
81 indicates which particular cache was hit, but the meaning is implementation defined because
82 different implementations can have different cache configurations.)
83
84However, SPE does not provide any call-graph information, and relies on statistical methods.
85
86Collisions
87----------
88
89When an operation is sampled while a previous sampled operation has not finished, a collision
90occurs. The new sample is dropped. Collisions affect the integrity of the data, so the sample rate
91should be set to avoid collisions.
92
93The 'sample_collision' PMU event can be used to determine the number of lost samples. Although this
94count is based on collisions _before_ filtering occurs. Therefore this can not be used as an exact
95number for samples dropped that would have made it through the filter, but can be a rough
96guide.
97
98The effect of microarchitectural sampling
99-----------------------------------------
100
101If an implementation samples micro-operations instead of instructions, the results of sampling must
102be weighted accordingly.
103
104For example, if a given instruction A is always converted into two micro-operations, A0 and A1, it
105becomes twice as likely to appear in the sample population.
106
107The coarse effect of conversions, and, if applicable, sampling of speculative operations, can be
108estimated from the 'sample_pop' and 'inst_retired' PMU events.
109
110Kernel Requirements
111-------------------
112
113The ARM_SPE_PMU config must be set to build as either a module or statically.
114
115Depending on CPU model, the kernel may need to be booted with page table isolation disabled
116(kpti=off). If KPTI needs to be disabled, this will fail with a console message "profiling buffer
117inaccessible. Try passing 'kpti=off' on the kernel command line".
118
119For the full criteria that determine whether KPTI needs to be forced off or not, see function
120unmap_kernel_at_el0() in the kernel sources. Common cases where it's not required
121are on the CPUs in kpti_safe_list, or on Arm v8.5+ where FEAT_E0PD is mandatory.
122
123The SPE interrupt must also be described by the firmware. If the module is loaded and KPTI is
124disabled (or isn't required to be disabled) but the SPE PMU still doesn't show in
125/sys/bus/event_source/devices/, then it's possible that the SPE interrupt isn't described by
126ACPI or DT. In this case no warning will be printed by the driver.
127
128Capturing SPE with perf command-line tools
129------------------------------------------
130
131You can record a session with SPE samples:
132
133  perf record -e arm_spe// -- ./mybench
134
135The sample period is set from the -c option, and because the minimum interval is used by default
136it's recommended to set this to a higher value. The value is written to PMSIRR.INTERVAL.
137
138Config parameters
139~~~~~~~~~~~~~~~~~
140
141These are placed between the // in the event and comma separated. For example '-e
142arm_spe/load_filter=1,min_latency=10/'
143
144  event_filter=<mask> - logical AND filter on specific events (PMSEVFR) - see bitfield description below
145  inv_event_filter=<mask> - logical OR to filter out specific events (PMSNEVFR, FEAT_SPEv1p2) - see bitfield description below
146  jitter=1            - use jitter to avoid resonance when sampling (PMSIRR.RND)
147  min_latency=<n>     - collect only samples with this latency or higher* (PMSLATFR)
148  pa_enable=1         - collect physical address (as well as VA) of loads/stores (PMSCR.PA) - requires privilege
149  pct_enable=1        - collect physical timestamp instead of virtual timestamp (PMSCR.PCT) - requires privilege
150  ts_enable=1         - enable timestamping with value of generic timer (PMSCR.TS)
151  discard=1           - enable SPE PMU events but don't collect sample data - see 'Discard mode' (PMBLIMITR.FM = DISCARD)
152  inv_data_src_filter=<mask> - mask to filter from 0-63 possible data sources (PMSDSFR, FEAT_SPE_FDS) - See 'Data source filtering'
153
154+++*+++ Latency is the total latency from the point at which sampling started on that instruction, rather
155than only the execution latency.
156
157Only some events can be filtered on using 'event_filter' bits. The overall
158filter is the logical AND of these bits, for example if bits 3 and 5 are set
159only samples that have both 'L1D cache refill' AND 'TLB walk' are recorded. When
160FEAT_SPEv1p2 is implemented 'inv_event_filter' can also be used to exclude
161events that have any (OR) of the filter's bits set. For example setting bits 3
162and 5 in 'inv_event_filter' will exclude any events that are either L1D cache
163refill OR TLB walk. If the same bit is set in both filters it's UNPREDICTABLE
164whether the sample is included or excluded. Filter bits for both event_filter
165and inv_event_filter are:
166
167  bit 1     - Instruction retired (i.e. omit speculative instructions)
168  bit 2     - L1D access (FEAT_SPEv1p4)
169  bit 3     - L1D refill
170  bit 4     - TLB access (FEAT_SPEv1p4)
171  bit 5     - TLB refill
172  bit 6     - Not taken event (FEAT_SPEv1p2)
173  bit 7     - Mispredict
174  bit 8     - Last level cache access (FEAT_SPEv1p4)
175  bit 9     - Last level cache miss (FEAT_SPEv1p4)
176  bit 10    - Remote access (FEAT_SPEv1p4)
177  bit 11    - Misaligned access (FEAT_SPEv1p1)
178  bit 12-15 - IMPLEMENTATION DEFINED events (when implemented)
179  bit 16    - Transaction (FEAT_TME)
180  bit 17    - Partial or empty SME or SVE predicate (FEAT_SPEv1p1)
181  bit 18    - Empty SME or SVE predicate (FEAT_SPEv1p1)
182  bit 19    - L2D access (FEAT_SPEv1p4)
183  bit 20    - L2D miss (FEAT_SPEv1p4)
184  bit 21    - Cache data modified (FEAT_SPEv1p4)
185  bit 22    - Recently fetched (FEAT_SPEv1p4)
186  bit 23    - Data snooped (FEAT_SPEv1p4)
187  bit 24    - Streaming SVE mode event (when FEAT_SPE_SME is implemented), or
188              IMPLEMENTATION DEFINED event 24 (when implemented, only versions
189              less than FEAT_SPEv1p4)
190  bit 25    - SMCU or external coprocessor operation event when FEAT_SPE_SME is
191              implemented, or IMPLEMENTATION DEFINED event 25 (when implemented,
192              only versions less than FEAT_SPEv1p4)
193  bit 26-31 - IMPLEMENTATION DEFINED events (only versions less than FEAT_SPEv1p4)
194  bit 48-63 - IMPLEMENTATION DEFINED events (when implemented)
195
196For IMPLEMENTATION DEFINED bits, refer to the CPU TRM if these bits are
197implemented.
198
199The driver will reject events if requested filter bits require unimplemented SPE
200versions, but will not reject filter bits for unimplemented IMPDEF bits or when
201their related feature is not present (e.g. SME). For example, if FEAT_SPEv1p2 is
202not implemented, filtering on "Not taken event" (bit 6) will be rejected.
203
204So to sample just retired instructions:
205
206  perf record -e arm_spe/event_filter=2/ -- ./mybench
207
208or just mispredicted branches:
209
210  perf record -e arm_spe/event_filter=0x80/ -- ./mybench
211
212When set, the following filters can be used to select samples that match any of
213the operation types (OR filtering). If only one is set then only samples of that
214type are collected:
215
216  branch_filter=1     - Collect branches (PMSFCR.B)
217  load_filter=1       - Collect loads (PMSFCR.LD)
218  store_filter=1      - Collect stores (PMSFCR.ST)
219
220When extended filtering is supported (FEAT_SPE_EFT), SIMD and float
221pointer operations can also be selected:
222
223  simd_filter=1         - Collect SIMD loads, stores and operations (PMSFCR.SIMD)
224  float_filter=1        - Collect floating point loads, stores and operations (PMSFCR.FP)
225
226When extended filtering is supported (FEAT_SPE_EFT), operation type filters can
227be changed to AND using _mask fields. For example samples could be selected if
228they are store AND SIMD by setting 'store_filter=1,simd_filter=1,
229store_filter_mask=1,simd_filter_mask=1'. The new masks are as follows:
230
231  branch_filter_mask=1  - Change branch filter behavior from OR to AND (PMSFCR.Bm)
232  load_filter_mask=1    - Change load filter behavior from OR to AND (PMSFCR.LDm)
233  store_filter_mask=1   - Change store filter behavior from OR to AND (PMSFCR.STm)
234  simd_filter_mask=1    - Change SIMD filter behavior from OR to AND (PMSFCR.SIMDm)
235  float_filter_mask=1   - Change floating point filter behavior from OR to AND (PMSFCR.FPm)
236
237Viewing the data
238~~~~~~~~~~~~~~~~~
239
240By default perf report and perf script will assign samples to separate groups depending on the
241attributes/events of the SPE record. Because instructions can have multiple events associated with
242them, the samples in these groups are not necessarily unique. For example perf report shows these
243groups:
244
245  Available samples
246  0 arm_spe//
247  0 dummy:u
248  21 l1d-miss
249  897 l1d-access
250  5 llc-miss
251  7 llc-access
252  2 tlb-miss
253  1K tlb-access
254  36 branch
255  0 remote-access
256  900 memory
257  1800 instructions
258
259The arm_spe// and dummy:u events are implementation details and are expected to be empty.
260
261The instructions group contains the full list of unique samples that are not
262sorted into other groups. To generate only this group use --itrace=i1i.
263
2641i (1 instruction interval) signifies no further downsampling. Rather than an
265instruction interval, this generates a sample every n SPE samples. For example
266to generate the default set of events for every 100 SPE samples:
267
268  perf report --itrace==bxofmtMai100i
269
270Other period types, for example nanoseconds (ns) are not currently supported.
271
272Memory access details are also stored on the samples and this can be viewed with:
273
274  perf report --mem-mode
275
276The latency value from the SPE sample is stored in the 'weight' field of the
277Perf samples and can be displayed in Perf script and report outputs by enabling
278its display from the command line.
279
280Common errors
281~~~~~~~~~~~~~
282
283 - "Cannot find PMU `arm_spe'. Missing kernel support?"
284
285   Module not built or loaded, KPTI not disabled, interrupt not described by firmware,
286   or running on a VM. See 'Kernel Requirements' above.
287
288 - "Arm SPE CONTEXT packets not found in the traces."
289
290   Root privilege is required to collect context packets. But these only increase the accuracy of
291   assigning PIDs to kernel samples. For userspace sampling this can be ignored.
292
293 - Excessively large perf.data file size
294
295   Increase sampling interval (see above)
296
297PMU events
298~~~~~~~~~~
299
300SPE has events that can be counted on core PMUs. These are prefixed with
301SAMPLE_, for example SAMPLE_POP, SAMPLE_FEED, SAMPLE_COLLISION and
302SAMPLE_FEED_BR.
303
304These events will only count when an SPE event is running on the same core that
305the PMU event is opened on, otherwise they read as 0. There are various ways to
306ensure that the PMU event and SPE event are scheduled together depending on the
307way the event is opened. For example opening both events as per-process events
308on the same process, although it's not guaranteed that the PMU event is enabled
309first when context switching. For that reason it may be better to open the PMU
310event as a systemwide event and then open SPE on the process of interest.
311
312Discard mode
313~~~~~~~~~~~~
314
315SPE related (SAMPLE_* etc) core PMU events can be used without the overhead of
316collecting sample data if discard mode is supported (optional from Armv8.6).
317First run a system wide SPE session (or on the core of interest) using options
318to minimize output. Then run perf stat:
319
320  perf record -e arm_spe/discard/ -a -N -B --no-bpf-event -o - > /dev/null &
321  perf stat -e SAMPLE_FEED_LD
322
323Data source filtering
324~~~~~~~~~~~~~~~~~~~~~
325
326When FEAT_SPE_FDS is present, 'inv_data_src_filter' can be used as a mask to
327filter on a subset (0 - 63) of possible data source IDs. The full range of data
328sources is 0 - 65535 although these are unlikely to be used in practice. Data
329sources are IMPDEF so refer to the TRM for the mappings. Each bit N of the
330filter maps to data source N. The filter is an OR of all the bits, and the value
331provided inv_data_src_filter is inverted before writing to PMSDSFR_EL1 so that
332set bits exclude that data source and cleared bits include that data source.
333Therefore the default value of 0 is equivalent to no filtering (all data sources
334included).
335
336For example, to include only data sources 0 and 3, clear bits 0 and 3
337(0xFFFFFFFFFFFFFFF6)
338
339When 'inv_data_src_filter' is set to 0xFFFFFFFFFFFFFFFF, any samples with any
340data source set are excluded.
341
342SEE ALSO
343--------
344
345linkperf:perf-record[1], linkperf:perf-script[1], linkperf:perf-report[1],
346linkperf:perf-inject[1]
347