xref: /linux/Documentation/admin-guide/perf/hisi-pcie-pmu.rst (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
1================================================
2HiSilicon PCIe Performance Monitoring Unit (PMU)
3================================================
4
5On Hip09, HiSilicon PCIe Performance Monitoring Unit (PMU) could monitor
6bandwidth, latency, bus utilization and buffer occupancy data of PCIe.
7
8Each PCIe Core has a PMU to monitor multi Root Ports of this PCIe Core and
9all Endpoints downstream these Root Ports.
10
11
12HiSilicon PCIe PMU driver
13=========================
14
15The PCIe PMU driver registers a perf PMU with the name of its sicl-id and PCIe
16Core id.::
17
18  /sys/bus/event_source/hisi_pcie<sicl>_core<core>
19
20PMU driver provides description of available events and filter options in sysfs,
21see /sys/bus/event_source/devices/hisi_pcie<sicl>_core<core>.
22
23The "format" directory describes all formats of the config (events) and config1
24(filter options) fields of the perf_event_attr structure. The "events" directory
25describes all documented events shown in perf list.
26
27The "identifier" sysfs file allows users to identify the version of the
28PMU hardware device.
29
30The "bus" sysfs file allows users to get the bus number of Root Ports
31monitored by PMU.
32
33Example usage of perf::
34
35  $# perf list
36  hisi_pcie0_core0/rx_mwr_latency/ [kernel PMU event]
37  hisi_pcie0_core0/rx_mwr_cnt/ [kernel PMU event]
38  ------------------------------------------
39
40  $# perf stat -e hisi_pcie0_core0/rx_mwr_latency,port=0xffff/
41  $# perf stat -e hisi_pcie0_core0/rx_mwr_cnt,port=0xffff/
42
43The related events usually used to calculate the bandwidth, latency or others.
44They need to start and end counting at the same time, therefore related events
45are best used in the same event group to get the expected value. There are two
46ways to know if they are related events:
47
48a) By event name, such as the latency events "xxx_latency, xxx_cnt" or
49   bandwidth events "xxx_flux, xxx_time".
50b) By event type, such as "event=0xXXXX, event=0x1XXXX".
51
52Example usage of perf group::
53
54  $# perf stat -e "{hisi_pcie0_core0/rx_mwr_latency,port=0xffff/,hisi_pcie0_core0/rx_mwr_cnt,port=0xffff/}"
55
56The current driver does not support sampling. So "perf record" is unsupported.
57Also attach to a task is unsupported for PCIe PMU.
58
59Filter options
60--------------
61
621. Target filter
63
64   PMU could only monitor the performance of traffic downstream target Root
65   Ports or downstream target Endpoint. PCIe PMU driver support "port" and
66   "bdf" interfaces for users.
67   Please notice that, one of these two interfaces must be set, and these two
68   interfaces aren't supported at the same time. If they are both set, only
69   "port" filter is valid.
70   If "port" filter not being set or is set explicitly to zero (default), the
71   "bdf" filter will be in effect, because "bdf=0" meaning 0000:000:00.0.
72
73   - port
74
75     "port" filter can be used in all PCIe PMU events, target Root Port can be
76     selected by configuring the 16-bits-bitmap "port". Multi ports can be
77     selected for AP-layer-events, and only one port can be selected for
78     TL/DL-layer-events.
79
80     For example, if target Root Port is 0000:00:00.0 (x8 lanes), bit0 of
81     bitmap should be set, port=0x1; if target Root Port is 0000:00:04.0 (x4
82     lanes), bit8 is set, port=0x100; if these two Root Ports are both
83     monitored, port=0x101.
84
85     Example usage of perf::
86
87       $# perf stat -e hisi_pcie0_core0/rx_mwr_latency,port=0x1/ sleep 5
88
89   - bdf
90
91     "bdf" filter can only be used in bandwidth events, target Endpoint is
92     selected by configuring BDF to "bdf". Counter only counts the bandwidth of
93     message requested by target Endpoint.
94
95     For example, "bdf=0x3900" means BDF of target Endpoint is 0000:39:00.0.
96
97     Example usage of perf::
98
99       $# perf stat -e hisi_pcie0_core0/rx_mrd_flux,bdf=0x3900/ sleep 5
100
1012. Trigger filter
102
103   Event statistics start when the first time TLP length is greater/smaller
104   than trigger condition. You can set the trigger condition by writing
105   "trig_len", and set the trigger mode by writing "trig_mode". This filter can
106   only be used in bandwidth events.
107
108   For example, "trig_len=4" means trigger condition is 2^4 DW, "trig_mode=0"
109   means statistics start when TLP length > trigger condition, "trig_mode=1"
110   means start when TLP length < condition.
111
112   Example usage of perf::
113
114     $# perf stat -e hisi_pcie0_core0/rx_mrd_flux,port=0xffff,trig_len=0x4,trig_mode=1/ sleep 5
115
1163. Threshold filter
117
118   Counter counts when TLP length within the specified range. You can set the
119   threshold by writing "thr_len", and set the threshold mode by writing
120   "thr_mode". This filter can only be used in bandwidth events.
121
122   For example, "thr_len=4" means threshold is 2^4 DW, "thr_mode=0" means
123   counter counts when TLP length >= threshold, and "thr_mode=1" means counts
124   when TLP length < threshold.
125
126   Example usage of perf::
127
128     $# perf stat -e hisi_pcie0_core0/rx_mrd_flux,port=0xffff,thr_len=0x4,thr_mode=1/ sleep 5
129
1304. TLP Length filter
131
132   When counting bandwidth, the data can be composed of certain parts of TLP
133   packets. You can specify it through "len_mode":
134
135   - 2'b00: Reserved (Do not use this since the behaviour is undefined)
136   - 2'b01: Bandwidth of TLP payloads
137   - 2'b10: Bandwidth of TLP headers
138   - 2'b11: Bandwidth of both TLP payloads and headers
139
140   For example, "len_mode=2" means only counting the bandwidth of TLP headers
141   and "len_mode=3" means the final bandwidth data is composed of both TLP
142   headers and payloads. Default value if not specified is 2'b11.
143
144   Example usage of perf::
145
146     $# perf stat -e hisi_pcie0_core0/rx_mrd_flux,port=0xffff,len_mode=0x1/ sleep 5
147