/* * Copyright (C) 2016 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #ifndef _CHRE_H_ #define _CHRE_H_ /** * @file * This header file includes all the headers which combine to fully define the * interface for the Context Hub Runtime Environment (CHRE). This interface is * of interest to both implementers of CHREs and authors of nanoapps. The API * documentation attempts to address concerns of both. * * See individual header files for API details, and general comments below * for overall platform information. */ #include <chre/audio.h> #include <chre/common.h> #include <chre/event.h> #include <chre/gnss.h> #include <chre/nanoapp.h> #include <chre/re.h> #include <chre/sensor.h> #include <chre/version.h> #include <chre/wifi.h> #include <chre/wwan.h> /** * @mainpage * CHRE is the Context Hub Runtime Environment. CHRE is used in Android to run * contextual applications, called nanoapps, in a low-power processing domain * other than the applications processor that runs Android itself. The CHRE * API, documented herein, is the common interface exposed to nanoapps for any * compatible CHRE implementation. The CHRE API provides the ability for * creating nanoapps that are code-compatible across different CHRE * implementations and underlying platforms. Refer to the following sections for * a discussion on some important details of CHRE that aren't explicitly exposed * in the API itself. * * @section entry_points Entry points * * The following entry points are used to bind a nanoapp to the CHRE system, and * all three must be implemented by any nanoapp (see chre/nanoapp.h): * - nanoappStart: initialization * - nanoappHandleEvent: hook for event-driven processing * - nanoappEnd: graceful teardown * * The CHRE implementation must also ensure that it performs these functions * prior to invoking nanoappStart, or after nanoappEnd returns: * - bss section zeroed out (prior to nanoappStart) * - static variables initialized (prior to nanoappStart) * - global C++ constructors called (prior to nanoappStart) * - global C++ destructors called (after nanoappEnd) * * @section threading Threading model * * A CHRE implementation is free to choose among many different * threading models, including a single-threaded system or a multi-threaded * system with preemption. The current platform definition is agnostic to this * underlying choice. However, the CHRE implementation must ensure that time * spent executing within a nanoapp does not significantly degrade or otherwise * interfere with other functions of the system in which CHRE is implemented, * especially latency-sensitive tasks such as sensor event delivery to the AP. * In other words, it must ensure that these functions can either occur in * parallel or preempt a nanoapp's execution. The current version of the API * does not specify whether the implementation allows for CPU sharing between * nanoapps on a more granular level than the handling of individual events [1]. * In any case, event ordering from the perspective of an individual nanoapp * must be FIFO, but the CHRE implementation may choose to violate total * ordering of events across all nanoapps to achieve more fair resource sharing, * but this is not required. * * This version of the CHRE API does require that all nanoapps are treated as * non-reentrant, meaning that only one instance of program flow can be inside * an individual nanoapp at any given time. That is, any of the functions of * the nanoapp, including the entry points and all other callbacks, cannot be * invoked if a previous invocation to the same or any other function in the * nanoapp has not completed yet. * * For example, if a nanoapp is currently in nanoappHandleEvent(), the CHRE is * not allowed to call nanoappHandleEvent() again, or to call a memory freeing * callback. Similarly, if a nanoapp is currently in a memory freeing * callback, the CHRE is not allowed to call nanoappHandleEvent(), or invoke * another memory freeing callback. * * There are two exceptions to this rule: If an invocation of chreSendEvent() * fails (returns 'false'), it is allowed to immediately invoke the memory * freeing callback passed into that function. This is a rare case, and one * where otherwise a CHRE implementation is likely to leak memory. Similarly, * chreSendMessageToHost() is allowed to invoke the memory freeing callback * directly, whether it returns 'true' or 'false'. This is because the CHRE * implementation may copy the message data to its own buffer, and therefore * wouldn't need the nanoapp-supplied buffer after chreSendMessageToHost() * returns. * * For a nanoapp author, this means no thought needs to be given to * synchronization issues with global objects, as they will, by definition, * only be accessed by a single thread at once. * * [1]: Note to CHRE implementers: A future version of the CHRE platform may * require multi-threading with preemption. This is mentioned as a heads up, * and to allow implementors deciding between implementation approaches to * make the most informed choice. * * @section timing Timing * * Nanoapps should expect to be running on a highly constrained system, with * little memory and little CPU. Any single nanoapp should expect to * be one of several nanoapps on the system, which also share the CPU with the * CHRE and possibly other services as well. * * Thus, a nanoapp needs to be efficient in its memory and CPU usage. * Also, as noted in the Threading Model section, a CHRE implementation may * be single threaded. As a result, all methods invoked in a nanoapp * (like nanoappStart, nanoappHandleEvent, memory free callbacks, etc.) * must run "quickly". "Quickly" is difficult to define, as there is a * diversity of Context Hub hardware. Nanoapp authors are strongly recommended * to limit their application to consuming no more than 1 second of CPU time * prior to returning control to the CHRE implementation. A CHRE implementation * may consider a nanoapp as unresponsive if it spends more time than this to * process a single event, and take corrective action. * * A nanoapp may have the need to occasionally perform a large block of * calculations that exceeds the 1 second guidance. The recommended approach in * this case is to split up the large block of calculations into smaller * batches. In one call into the nanoapp, the nanoapp can perform the first * batch, and then set a timer or send an event (chreSendEvent()) to itself * indicating which batch should be done next. This will allow the nanoapp to * perform the entire calculation over time, without monopolizing system * resources. * * @section floats Floating point support * * The C type 'float' is used in this API, and thus a CHRE implementation * is required to support 'float's. * * Support of the C types 'double' and 'long double' is optional for a * CHRE implementation. Note that if a CHRE decides to support them, unlike * 'float' support, there is no requirement that this support is particularly * efficient. So nanoapp authors should be aware this may be inefficient. * * If a CHRE implementation choses not to support 'double' or * 'long double', then the build toolchain setup provided needs to set * the preprocessor define CHRE_NO_DOUBLE_SUPPORT. * * @section compat CHRE and Nanoapp compatibility * * CHRE implementations must make affordances to maintain binary compatibility * across minor revisions of the API version (e.g. v1.1 to v1.2). This applies * to both running a nanoapp compiled for a newer version of the API on a CHRE * implementation built against an older version (backwards compatibility), and * vice versa (forwards compatibility). API changes that are acceptable in * minor version changes that may require special measures to ensure binary * compatibility include: addition of new functions; addition of arguments to * existing functions when the default value used for nanoapps compiled against * the old version is well-defined and does not affect existing functionality; * and addition of fields to existing structures, even when this induces a * binary layout change (this should be made rare via judicious use of reserved * fields). API changes that must only occur alongside a major version change * and are therefore not compatible include: removal of any function, argument, * field in a data structure, or mandatory functional behavior that a nanoapp * may depend on; any change in the interpretation of an existing data structure * field that alters the way it was defined previously (changing the units of a * field would fall under this, but appropriating a previously reserved field * for some new functionality would not); and any change in functionality or * expected behavior that conflicts with the previous definition. * * Note that the CHRE API only specifies the software interface between a * nanoapp and the CHRE system - the binary interface (ABI) between nanoapp and * CHRE is necessarily implementation-dependent. Therefore, the recommended * approach to accomplish binary compatibility is to build a Nanoapp Support * Library (NSL) that is specific to the CHRE implementation into the nanoapp * binary, and use it to handle ABI details in a way that ensures compatibility. * In addition, to accomplish forwards compatibility, the CHRE implementation is * expected to recognize the CHRE API version that a nanoapp is targeting and * engage compatibility behaviors where necessary. * * By definition, major API version changes (e.g. v1.1 to v2.0) break * compatibility. Therefore, a CHRE implementation must not attempt to load a * nanoapp that is targeting a newer major API version. */ #endif /* _CHRE_H_ */