• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2016 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #ifndef _CHRE_H_
18 #define _CHRE_H_
19 
20 /**
21  * @file
22  * This header file includes all the headers which combine to fully define the
23  * interface for the Context Hub Runtime Environment (CHRE).  This interface is
24  * of interest to both implementers of CHREs and authors of nanoapps.  The API
25  * documentation attempts to address concerns of both.
26  *
27  * See individual header files for API details, and general comments below
28  * for overall platform information.
29  */
30 
31 #include <chre/audio.h>
32 #include <chre/common.h>
33 #include <chre/event.h>
34 #include <chre/gnss.h>
35 #include <chre/nanoapp.h>
36 #include <chre/re.h>
37 #include <chre/sensor.h>
38 #include <chre/toolchain.h>
39 #include <chre/version.h>
40 #include <chre/wifi.h>
41 #include <chre/wwan.h>
42 
43 /**
44  * @mainpage
45  * CHRE is the Context Hub Runtime Environment.  CHRE is used in Android to run
46  * contextual applications, called nanoapps, in a low-power processing domain
47  * other than the applications processor that runs Android itself.  The CHRE
48  * API, documented herein, is the common interface exposed to nanoapps for any
49  * compatible CHRE implementation.  The CHRE API provides the ability for
50  * creating nanoapps that are code-compatible across different CHRE
51  * implementations and underlying platforms. Refer to the following sections for
52  * a discussion on some important details of CHRE that aren't explicitly exposed
53  * in the API itself.
54  *
55  * @section entry_points Entry points
56  *
57  * The following entry points are used to bind a nanoapp to the CHRE system, and
58  * all three must be implemented by any nanoapp (see chre/nanoapp.h):
59  * - nanoappStart: initialization
60  * - nanoappHandleEvent: hook for event-driven processing
61  * - nanoappEnd: graceful teardown
62  *
63  * The CHRE implementation must also ensure that it performs these functions
64  * prior to invoking nanoappStart, or after nanoappEnd returns:
65  * - bss section zeroed out (prior to nanoappStart)
66  * - static variables initialized (prior to nanoappStart)
67  * - global C++ constructors called (prior to nanoappStart)
68  * - global C++ destructors called (after nanoappEnd)
69  *
70  * @section threading Threading model
71  *
72  * A CHRE implementation is free to choose among many different
73  * threading models, including a single-threaded system or a multi-threaded
74  * system with preemption.  The current platform definition is agnostic to this
75  * underlying choice.  However, the CHRE implementation must ensure that time
76  * spent executing within a nanoapp does not significantly degrade or otherwise
77  * interfere with other functions of the system in which CHRE is implemented,
78  * especially latency-sensitive tasks such as sensor event delivery to the AP.
79  * In other words, it must ensure that these functions can either occur in
80  * parallel or preempt a nanoapp's execution.  The current version of the API
81  * does not specify whether the implementation allows for CPU sharing between
82  * nanoapps on a more granular level than the handling of individual events [1].
83  * In any case, event ordering from the perspective of an individual nanoapp
84  * must be FIFO, but the CHRE implementation may choose to violate total
85  * ordering of events across all nanoapps to achieve more fair resource sharing,
86  * but this is not required.
87  *
88  * This version of the CHRE API does require that all nanoapps are treated as
89  * non-reentrant, meaning that only one instance of program flow can be inside
90  * an individual nanoapp at any given time.  That is, any of the functions of
91  * the nanoapp, including the entry points and all other callbacks, cannot be
92  * invoked if a previous invocation to the same or any other function in the
93  * nanoapp has not completed yet.
94  *
95  * For example, if a nanoapp is currently in nanoappHandleEvent(), the CHRE is
96  * not allowed to call nanoappHandleEvent() again, or to call a memory freeing
97  * callback.  Similarly, if a nanoapp is currently in a memory freeing
98  * callback, the CHRE is not allowed to call nanoappHandleEvent(), or invoke
99  * another memory freeing callback.
100  *
101  * There are two exceptions to this rule: If an invocation of chreSendEvent()
102  * fails (returns 'false'), it is allowed to immediately invoke the memory
103  * freeing callback passed into that function.  This is a rare case, and one
104  * where otherwise a CHRE implementation is likely to leak memory. Similarly,
105  * chreSendMessageToHost() is allowed to invoke the memory freeing callback
106  * directly, whether it returns 'true' or 'false'.  This is because the CHRE
107  * implementation may copy the message data to its own buffer, and therefore
108  * wouldn't need the nanoapp-supplied buffer after chreSendMessageToHost()
109  * returns.
110  *
111  * For a nanoapp author, this means no thought needs to be given to
112  * synchronization issues with global objects, as they will, by definition,
113  * only be accessed by a single thread at once.
114  *
115  * [1]: Note to CHRE implementers: A future version of the CHRE platform may
116  * require multi-threading with preemption.  This is mentioned as a heads up,
117  * and to allow implementors deciding between implementation approaches to
118  * make the most informed choice.
119  *
120  * @section timing Timing
121  *
122  * Nanoapps should expect to be running on a highly constrained system, with
123  * little memory and little CPU.  Any single nanoapp should expect to
124  * be one of several nanoapps on the system, which also share the CPU with the
125  * CHRE and possibly other services as well.
126  *
127  * Thus, a nanoapp needs to be efficient in its memory and CPU usage.
128  * Also, as noted in the Threading Model section, a CHRE implementation may
129  * be single threaded.  As a result, all methods invoked in a nanoapp
130  * (like nanoappStart, nanoappHandleEvent, memory free callbacks, etc.)
131  * must run "quickly".  "Quickly" is difficult to define, as there is a
132  * diversity of Context Hub hardware.  Nanoapp authors are strongly recommended
133  * to limit their application to consuming no more than 1 second of CPU time
134  * prior to returning control to the CHRE implementation.  A CHRE implementation
135  * may consider a nanoapp as unresponsive if it spends more time than this to
136  * process a single event, and take corrective action.
137  *
138  * A nanoapp may have the need to occasionally perform a large block of
139  * calculations that exceeds the 1 second guidance.  The recommended approach in
140  * this case is to split up the large block of calculations into smaller
141  * batches.  In one call into the nanoapp, the nanoapp can perform the first
142  * batch, and then set a timer or send an event (chreSendEvent()) to itself
143  * indicating which batch should be done next. This will allow the nanoapp to
144  * perform the entire calculation over time, without monopolizing system
145  * resources.
146  *
147  * @section floats Floating point support
148  *
149  * The C type 'float' is used in this API, and thus a CHRE implementation
150  * is required to support 'float's.
151  *
152  * Support of the C types 'double' and 'long double' is optional for a
153  * CHRE implementation.  Note that if a CHRE decides to support them, unlike
154  * 'float' support, there is no requirement that this support is particularly
155  * efficient.  So nanoapp authors should be aware this may be inefficient.
156  *
157  * If a CHRE implementation chooses not to support 'double' or
158  * 'long double', then the build toolchain setup provided needs to set
159  * the preprocessor define CHRE_NO_DOUBLE_SUPPORT.
160  *
161  * @section compat CHRE and Nanoapp compatibility
162  *
163  * CHRE implementations must make affordances to maintain binary compatibility
164  * across minor revisions of the API version (e.g. v1.1 to v1.2).  This applies
165  * to both running a nanoapp compiled for a newer version of the API on a CHRE
166  * implementation built against an older version (backwards compatibility), and
167  * vice versa (forwards compatibility).  API changes that are acceptable in
168  * minor version changes that may require special measures to ensure binary
169  * compatibility include: addition of new functions; addition of arguments to
170  * existing functions when the default value used for nanoapps compiled against
171  * the old version is well-defined and does not affect existing functionality;
172  * and addition of fields to existing structures, even when this induces a
173  * binary layout change (this should be made rare via judicious use of reserved
174  * fields).  API changes that must only occur alongside a major version change
175  * and are therefore not compatible include: removal of any function, argument,
176  * field in a data structure, or mandatory functional behavior that a nanoapp
177  * may depend on; any change in the interpretation of an existing data structure
178  * field that alters the way it was defined previously (changing the units of a
179  * field would fall under this, but appropriating a previously reserved field
180  * for some new functionality would not); and any change in functionality or
181  * expected behavior that conflicts with the previous definition.
182  *
183  * Note that the CHRE API only specifies the software interface between a
184  * nanoapp and the CHRE system - the binary interface (ABI) between nanoapp and
185  * CHRE is necessarily implementation-dependent.  Therefore, the recommended
186  * approach to accomplish binary compatibility is to build a Nanoapp Support
187  * Library (NSL) that is specific to the CHRE implementation into the nanoapp
188  * binary, and use it to handle ABI details in a way that ensures compatibility.
189  * In addition, to accomplish forwards compatibility, the CHRE implementation is
190  * expected to recognize the CHRE API version that a nanoapp is targeting and
191  * engage compatibility behaviors where necessary.
192  *
193  * By definition, major API version changes (e.g. v1.1 to v2.0) break
194  * compatibility.  Therefore, a CHRE implementation must not attempt to load a
195  * nanoapp that is targeting a newer major API version.
196  */
197 
198 #endif  /* _CHRE_H_ */
199 
200