blob: 73cd163dbfb84e4808f50454af02ac891a34b114 [file] [log] [blame]
Michael Ellerman330a1eb2013-06-28 18:15:16 +10001PMU Event Based Branches
2========================
3
4Event Based Branches (EBBs) are a feature which allows the hardware to
5branch directly to a specified user space address when certain events occur.
6
7The full specification is available in Power ISA v2.07:
8
9 https://www.power.org/documentation/power-isa-version-2-07/
10
11One type of event for which EBBs can be configured is PMU exceptions. This
12document describes the API for configuring the Power PMU to generate EBBs,
13using the Linux perf_events API.
14
15
16Terminology
17-----------
18
19Throughout this document we will refer to an "EBB event" or "EBB events". This
20just refers to a struct perf_event which has set the "EBB" flag in its
21attr.config. All events which can be configured on the hardware PMU are
22possible "EBB events".
23
24
25Background
26----------
27
28When a PMU EBB occurs it is delivered to the currently running process. As such
29EBBs can only sensibly be used by programs for self-monitoring.
30
31It is a feature of the perf_events API that events can be created on other
32processes, subject to standard permission checks. This is also true of EBB
33events, however unless the target process enables EBBs (via mtspr(BESCR)) no
34EBBs will ever be delivered.
35
36This makes it possible for a process to enable EBBs for itself, but not
37actually configure any events. At a later time another process can come along
38and attach an EBB event to the process, which will then cause EBBs to be
39delivered to the first process. It's not clear if this is actually useful.
40
41
42When the PMU is configured for EBBs, all PMU interrupts are delivered to the
43user process. This means once an EBB event is scheduled on the PMU, no non-EBB
44events can be configured. This means that EBB events can not be run
45concurrently with regular 'perf' commands, or any other perf events.
46
47It is however safe to run 'perf' commands on a process which is using EBBs. The
48kernel will in general schedule the EBB event, and perf will be notified that
49its events could not run.
50
51The exclusion between EBB events and regular events is implemented using the
52existing "pinned" and "exclusive" attributes of perf_events. This means EBB
53events will be given priority over other events, unless they are also pinned.
54If an EBB event and a regular event are both pinned, then whichever is enabled
55first will be scheduled and the other will be put in error state. See the
56section below titled "Enabling an EBB event" for more information.
57
58
59Creating an EBB event
60---------------------
61
62To request that an event is counted using EBB, the event code should have bit
6363 set.
64
65EBB events must be created with a particular, and restrictive, set of
66attributes - this is so that they interoperate correctly with the rest of the
67perf_events subsystem.
68
69An EBB event must be created with the "pinned" and "exclusive" attributes set.
70Note that if you are creating a group of EBB events, only the leader can have
71these attributes set.
72
73An EBB event must NOT set any of the "inherit", "sample_period", "freq" or
74"enable_on_exec" attributes.
75
76An EBB event must be attached to a task. This is specified to perf_event_open()
77by passing a pid value, typically 0 indicating the current task.
78
79All events in a group must agree on whether they want EBB. That is all events
80must request EBB, or none may request EBB.
81
82EBB events must specify the PMC they are to be counted on. This ensures
83userspace is able to reliably determine which PMC the event is scheduled on.
84
85
86Enabling an EBB event
87---------------------
88
89Once an EBB event has been successfully opened, it must be enabled with the
90perf_events API. This can be achieved either via the ioctl() interface, or the
91prctl() interface.
92
93However, due to the design of the perf_events API, enabling an event does not
94guarantee that it has been scheduled on the PMU. To ensure that the EBB event
95has been scheduled on the PMU, you must perform a read() on the event. If the
96read() returns EOF, then the event has not been scheduled and EBBs are not
97enabled.
98
99This behaviour occurs because the EBB event is pinned and exclusive. When the
100EBB event is enabled it will force all other non-pinned events off the PMU. In
101this case the enable will be successful. However if there is already an event
102pinned on the PMU then the enable will not be successful.
103
104
105Reading an EBB event
106--------------------
107
108It is possible to read() from an EBB event. However the results are
109meaningless. Because interrupts are being delivered to the user process the
110kernel is not able to count the event, and so will return a junk value.
111
112
113Closing an EBB event
114--------------------
115
116When an EBB event is finished with, you can close it using close() as for any
117regular event. If this is the last EBB event the PMU will be deconfigured and
118no further PMU EBBs will be delivered.
119
120
121EBB Handler
122-----------
123
124The EBB handler is just regular userspace code, however it must be written in
125the style of an interrupt handler. When the handler is entered all registers
126are live (possibly) and so must be saved somehow before the handler can invoke
127other code.
128
129It's up to the program how to handle this. For C programs a relatively simple
130option is to create an interrupt frame on the stack and save registers there.
131
132Fork
133----
134
135EBB events are not inherited across fork. If the child process wishes to use
136EBBs it should open a new event for itself. Similarly the EBB state in
137BESCR/EBBHR/EBBRR is cleared across fork().