1 Using bionic 2 ============ 3 4 See the [additional documentation](docs/). 5 6 Working on bionic 7 ================= 8 9 What are the big pieces of bionic? 10 ---------------------------------- 11 12 #### libc/ --- libc.so, libc.a 13 14 The C library. Stuff like `fopen(3)` and `kill(2)`. 15 16 #### libm/ --- libm.so, libm.a 17 18 The math library. Traditionally Unix systems kept stuff like `sin(3)` and 19 `cos(3)` in a separate library to save space in the days before shared 20 libraries. 21 22 #### libdl/ --- libdl.so 23 24 The dynamic linker interface library. This is actually just a bunch of stubs 25 that the dynamic linker replaces with pointers to its own implementation at 26 runtime. This is where stuff like `dlopen(3)` lives. 27 28 #### libstdc++/ --- libstdc++.so 29 30 The C++ ABI support functions. The C++ compiler doesn't know how to implement 31 thread-safe static initialization and the like, so it just calls functions that 32 are supplied by the system. Stuff like `__cxa_guard_acquire` and 33 `__cxa_pure_virtual` live here. 34 35 #### linker/ --- /system/bin/linker and /system/bin/linker64 36 37 The dynamic linker. When you run a dynamically-linked executable, its ELF file 38 has a `DT_INTERP` entry that says "use the following program to start me". On 39 Android, that's either `linker` or `linker64` (depending on whether it's a 40 32-bit or 64-bit executable). It's responsible for loading the ELF executable 41 into memory and resolving references to symbols (so that when your code tries to 42 jump to `fopen(3)`, say, it lands in the right place). 43 44 #### tests/ --- unit tests 45 46 The `tests/` directory contains unit tests. Roughly arranged as one file per 47 publicly-exported header file. 48 49 #### benchmarks/ --- benchmarks 50 51 The `benchmarks/` directory contains benchmarks, with its own [documentation](benchmarks/README.md). 52 53 54 What's in libc/? 55 ---------------- 56 57 <pre> 58 libc/ 59 arch-arm/ 60 arch-arm64/ 61 arch-common/ 62 arch-mips/ 63 arch-mips64/ 64 arch-x86/ 65 arch-x86_64/ 66 # Each architecture has its own subdirectory for stuff that isn't shared 67 # because it's architecture-specific. There will be a .mk file in here that 68 # drags in all the architecture-specific files. 69 bionic/ 70 # Every architecture needs a handful of machine-specific assembler files. 71 # They live here. 72 include/ 73 machine/ 74 # The majority of header files are actually in libc/include/, but many 75 # of them pull in a <machine/something.h> for things like limits, 76 # endianness, and how floating point numbers are represented. Those 77 # headers live here. 78 string/ 79 # Most architectures have a handful of optional assembler files 80 # implementing optimized versions of various routines. The <string.h> 81 # functions are particular favorites. 82 syscalls/ 83 # The syscalls directories contain script-generated assembler files. 84 # See 'Adding system calls' later. 85 86 include/ 87 # The public header files on everyone's include path. These are a mixture of 88 # files written by us and files taken from BSD. 89 90 kernel/ 91 # The kernel uapi header files. These are scrubbed copies of the originals 92 # in external/kernel-headers/. These files must not be edited directly. The 93 # generate_uapi_headers.sh script should be used to go from a kernel tree to 94 # external/kernel-headers/ --- this takes care of the architecture-specific 95 # details. The update_all.py script should be used to regenerate bionic's 96 # scrubbed headers from external/kernel-headers/. 97 98 private/ 99 # These are private header files meant for use within bionic itself. 100 101 dns/ 102 # Contains the DNS resolver (originates from NetBSD code). 103 104 upstream-freebsd/ 105 upstream-netbsd/ 106 upstream-openbsd/ 107 # These directories contain unmolested upstream source. Any time we can 108 # just use a BSD implementation of something unmodified, we should. 109 # The structure under these directories mimics the upstream tree, 110 # but there's also... 111 android/ 112 include/ 113 # This is where we keep the hacks necessary to build BSD source 114 # in our world. The *-compat.h files are automatically included 115 # using -include, but we also provide equivalents for missing 116 # header/source files needed by the BSD implementation. 117 118 bionic/ 119 # This is the biggest mess. The C++ files are files we own, typically 120 # because the Linux kernel interface is sufficiently different that we 121 # can't use any of the BSD implementations. The C files are usually 122 # legacy mess that needs to be sorted out, either by replacing it with 123 # current upstream source in one of the upstream directories or by 124 # switching the file to C++ and cleaning it up. 125 126 malloc_debug/ 127 # The code that implements the functionality to enable debugging of 128 # native allocation problems. 129 130 stdio/ 131 # These are legacy files of dubious provenance. We're working to clean 132 # this mess up, and this directory should disappear. 133 134 tools/ 135 # Various tools used to maintain bionic. 136 137 tzcode/ 138 # A modified superset of the IANA tzcode. Most of the modifications relate 139 # to Android's use of a single file (with corresponding index) to contain 140 # time zone data. 141 zoneinfo/ 142 # Android-format time zone data. 143 # See 'Updating tzdata' later. 144 </pre> 145 146 147 Adding libc wrappers for system calls 148 ------------------------------------- 149 150 The first question you should ask is "should I add a libc wrapper for 151 this system call?". The answer is usually "no". 152 153 The answer is "yes" if the system call is part of the POSIX standard. 154 155 The answer is probably "yes" if the system call has a wrapper in at 156 least one other C library. 157 158 The answer may be "yes" if the system call has three/four distinct 159 users in different projects, and there isn't a more specific library 160 that would make more sense as the place to add the wrapper. 161 162 In all other cases, you should use 163 [syscall(3)](http://man7.org/linux/man-pages/man2/syscall.2.html) instead. 164 165 Adding a system call usually involves: 166 167 1. Add entries to SYSCALLS.TXT. 168 See SYSCALLS.TXT itself for documentation on the format. 169 2. Run the gensyscalls.py script. 170 3. Add constants (and perhaps types) to the appropriate header file. 171 Note that you should check to see whether the constants are already in 172 kernel uapi header files, in which case you just need to make sure that 173 the appropriate POSIX header file in libc/include/ includes the 174 relevant file or files. 175 4. Add function declarations to the appropriate header file. Don't forget 176 to include the appropriate `__INTRODUCED_IN()`. 177 5. Add the function name to the correct section in libc/libc.map.txt and 178 run `./libc/tools/genversion-scripts.py`. 179 6. Add at least basic tests. Even a test that deliberately supplies 180 an invalid argument helps check that we're generating the right symbol 181 and have the right declaration in the header file, and that you correctly 182 updated the maps in step 5. (You can use strace(1) to confirm that the 183 correct system call is being made.) 184 185 186 Updating kernel header files 187 ---------------------------- 188 189 As mentioned above, this is currently a two-step process: 190 191 1. Use generate_uapi_headers.sh to go from a Linux source tree to appropriate 192 contents for external/kernel-headers/. 193 2. Run update_all.py to scrub those headers and import them into bionic. 194 195 Note that if you're actually just trying to expose device-specific headers to 196 build your device drivers, you shouldn't modify bionic. Instead use 197 `TARGET_DEVICE_KERNEL_HEADERS` and friends described in [config.mk](https://android.googlesource.com/platform/build/+/master/core/config.mk#186). 198 199 200 Updating tzdata 201 --------------- 202 203 This is fully automated (and these days handled by the libcore team, because 204 they own icu, and that needs to be updated in sync with bionic): 205 206 1. Run update-tzdata.py in external/icu/tools/. 207 208 209 Verifying changes 210 ----------------- 211 212 If you make a change that is likely to have a wide effect on the tree (such as a 213 libc header change), you should run `make checkbuild`. A regular `make` will 214 _not_ build the entire tree; just the minimum number of projects that are 215 required for the device. Tests, additional developer tools, and various other 216 modules will not be built. Note that `make checkbuild` will not be complete 217 either, as `make tests` covers a few additional modules, but generally speaking 218 `make checkbuild` is enough. 219 220 221 Running the tests 222 ----------------- 223 224 The tests are all built from the tests/ directory. 225 226 ### Device tests 227 228 $ mma # In $ANDROID_ROOT/bionic. 229 $ adb root && adb remount && adb sync 230 $ adb shell /data/nativetest/bionic-unit-tests/bionic-unit-tests 231 $ adb shell \ 232 /data/nativetest/bionic-unit-tests-static/bionic-unit-tests-static 233 # Only for 64-bit targets 234 $ adb shell /data/nativetest64/bionic-unit-tests/bionic-unit-tests 235 $ adb shell \ 236 /data/nativetest64/bionic-unit-tests-static/bionic-unit-tests-static 237 238 Note that we use our own custom gtest runner that offers a superset of the 239 options documented at 240 <https://github.com/google/googletest/blob/master/googletest/docs/AdvancedGuide.md#running-test-programs-advanced-options>, 241 in particular for test isolation and parallelism (both on by default). 242 243 ### Device tests via CTS 244 245 Most of the unit tests are executed by CTS. By default, CTS runs as 246 a non-root user, so the unit tests must also pass when not run as root. 247 Some tests cannot do any useful work unless run as root. In this case, 248 the test should check `getuid() == 0` and do nothing otherwise (typically 249 we log in this case to prevent accidents!). Obviously, if the test can be 250 rewritten to not require root, that's an even better solution. 251 252 Currently, the list of bionic CTS tests is generated at build time by 253 running a host version of the test executable and dumping the list of 254 all tests. In order for this to continue to work, all architectures must 255 have the same number of tests, and the host version of the executable 256 must also have the same number of tests. 257 258 Running the gtests directly is orders of magnitude faster than using CTS, 259 but in cases where you really have to run CTS: 260 261 $ make cts # In $ANDROID_ROOT. 262 $ adb unroot # Because real CTS doesn't run as root. 263 # This will sync any *test* changes, but not *code* changes: 264 $ cts-tradefed \ 265 run singleCommand cts --skip-preconditions -m CtsBionicTestCases 266 267 ### Host tests 268 269 The host tests require that you have `lunch`ed either an x86 or x86_64 target. 270 Note that due to ABI limitations (specifically, the size of pthread_mutex_t), 271 32-bit bionic requires PIDs less than 65536. To enforce this, set /proc/sys/kernel/pid_max 272 to 65536. 273 274 $ ./tests/run-on-host.sh 32 275 $ ./tests/run-on-host.sh 64 # For x86_64-bit *targets* only. 276 277 You can supply gtest flags as extra arguments to this script. 278 279 ### Against glibc 280 281 As a way to check that our tests do in fact test the correct behavior (and not 282 just the behavior we think is correct), it is possible to run the tests against 283 the host's glibc. 284 285 $ ./tests/run-on-host.sh glibc 286 287 288 Gathering test coverage 289 ----------------------- 290 291 For either host or target coverage, you must first: 292 293 * `$ export NATIVE_COVERAGE=true` 294 * Note that the build system is ignorant to this flag being toggled, i.e. if 295 you change this flag, you will have to manually rebuild bionic. 296 * Set `bionic_coverage=true` in `libc/Android.mk` and `libm/Android.mk`. 297 298 ### Coverage from device tests 299 300 $ mma 301 $ adb sync 302 $ adb shell \ 303 GCOV_PREFIX=/data/local/tmp/gcov \ 304 GCOV_PREFIX_STRIP=`echo $ANDROID_BUILD_TOP | grep -o / | wc -l` \ 305 /data/nativetest/bionic-unit-tests/bionic-unit-tests 306 $ acov 307 308 `acov` will pull all coverage information from the device, push it to the right 309 directories, run `lcov`, and open the coverage report in your browser. 310 311 ### Coverage from host tests 312 313 First, build and run the host tests as usual (see above). 314 315 $ croot 316 $ lcov -c -d $ANDROID_PRODUCT_OUT -o coverage.info 317 $ genhtml -o covreport coverage.info # or lcov --list coverage.info 318 319 The coverage report is now available at `covreport/index.html`. 320 321 322 Attaching GDB to the tests 323 -------------------------- 324 325 Bionic's test runner will run each test in its own process by default to prevent 326 tests failures from impacting other tests. This also has the added benefit of 327 running them in parallel, so they are much faster. 328 329 However, this also makes it difficult to run the tests under GDB. To prevent 330 each test from being forked, run the tests with the flag `--no-isolate`. 331 332 333 32-bit ABI bugs 334 --------------- 335 336 See [32-bit ABI bugs](docs/32-bit-abi.md). 337
