1 <html> 2 <head> 3 <title>Dalvik Porting Guide</title> 4 </head> 5 6 <body> 7 <h1>Dalvik Porting Guide</h1> 8 9 <p> 10 The Dalvik virtual machine is intended to run on a variety of platforms. 11 The baseline system is expected to be a variant of UNIX (Linux, BSD, Mac 12 OS X) running the GNU C compiler. Little-endian CPUs have been exercised 13 the most heavily, but big-endian systems are explicitly supported. 14 </p><p> 15 There are two general categories of work: porting to a Linux system 16 with a previously unseen CPU architecture, and porting to a different 17 operating system. This document covers the former. 18 </p><p> 19 Basic familiarity with the Android platform, source code structure, and 20 build system is assumed. 21 </p> 22 23 24 <h2>Core Libraries</h2> 25 26 <p> 27 The native code in the core libraries (chiefly <code>libcore</code>, 28 but also <code>dalvik/vm/native</code>) is written in C/C++ and is expected 29 to work without modification in a Linux environment. 30 </p><p> 31 The core libraries pull in code from many other projects, including 32 OpenSSL, zlib, and ICU. These will also need to be ported before the VM 33 can be used. 34 </p> 35 36 37 <h2>JNI Call Bridge</h2> 38 39 <p> 40 Most of the Dalvik VM runtime is written in portable C. The one 41 non-portable component of the runtime is the JNI call bridge. Simply put, 42 this converts an array of integers into function arguments of various 43 types, and calls a function. This must be done according to the C calling 44 conventions for the platform. The task could be as simple as pushing all 45 of the arguments onto the stack, or involve complex rules for register 46 assignment and stack alignment. 47 </p><p> 48 To ease porting to new platforms, the <a href="http://sourceware.org/libffi/"> 49 open-source FFI library</a> (Foreign Function Interface) is used when a 50 custom bridge is unavailable. FFI is not as fast as a native implementation, 51 and the optional performance improvements it does offer are not used, so 52 writing a replacement is a good first step. 53 </p><p> 54 The code lives in <code>dalvik/vm/arch/*</code>, with the FFI-based version 55 in the "generic" directory. There are two source files for each architecture. 56 One defines the call bridge itself: 57 </p><p><blockquote> 58 <code>void dvmPlatformInvoke(void* pEnv, ClassObject* clazz, int argInfo, 59 int argc, const u4* argv, const char* signature, void* func, 60 JValue* pReturn)</code> 61 </blockquote></p><p> 62 This will invoke a C/C++ function declared: 63 </p><p><blockquote> 64 <code>return_type func(JNIEnv* pEnv, Object* this [, <i>args</i>])<br></code> 65 </blockquote>or (for a "static" method):<blockquote> 66 <code>return_type func(JNIEnv* pEnv, ClassObject* clazz [, <i>args</i>])</code> 67 </blockquote></p><p> 68 The role of <code>dvmPlatformInvoke</code> is to convert the values in 69 <code>argv</code> into C-style calling conventions, call the method, and 70 then place the return type into <code>pReturn</code> (a union that holds 71 all of the basic JNI types). The code may use the method signature 72 (a DEX "shorty" signature, with one character for the return type and one 73 per argument) to determine how to handle the values. 74 </p><p> 75 The other source file involved here defines a 32-bit "hint". The hint 76 is computed when the method's class is loaded, and passed in as the 77 "argInfo" argument. The hint can be used to avoid scanning the ASCII 78 method signature for things like the return value, total argument size, 79 or inter-argument 64-bit alignment restrictions. 80 81 82 <h2>Interpreter</h2> 83 84 <p> 85 The Dalvik runtime includes two interpreters, labeled "portable" and "fast". 86 The portable interpreter is largely contained within a single C function, 87 and should compile on any system that supports gcc. (If you don't have gcc, 88 you may need to disable the "threaded" execution model, which relies on 89 gcc's "goto table" implementation; look for the THREADED_INTERP define.) 90 </p><p> 91 The fast interpreter uses hand-coded assembly fragments. If none are 92 available for the current architecture, the build system will create an 93 interpreter out of C "stubs". The resulting "all stubs" interpreter is 94 quite a bit slower than the portable interpreter, making "fast" something 95 of a misnomer. 96 </p><p> 97 The fast interpreter is enabled by default. On platforms without native 98 support, you may want to switch to the portable interpreter. This can 99 be controlled with the <code>dalvik.vm.execution-mode</code> system 100 property. For example, if you: 101 </p><p><blockquote> 102 <code>adb shell "echo dalvik.vm.execution-mode = int:portable >> /data/local.prop"</code> 103 </blockquote></p><p> 104 and reboot, the Android app framework will start the VM with the portable 105 interpreter enabled. 106 </p> 107 108 109 <h3>Mterp Interpreter Structure</h3> 110 111 <p> 112 There may be significant performance advantages to rewriting the 113 interpreter core in assembly language, using architecture-specific 114 optimizations. In Dalvik this can be done one instruction at a time. 115 </p><p> 116 The simplest way to implement an interpreter is to have a large "switch" 117 statement. After each instruction is handled, the interpreter returns to 118 the top of the loop, fetches the next instruction, and jumps to the 119 appropriate label. 120 </p><p> 121 An improvement on this is called "threaded" execution. The instruction 122 fetch and dispatch are included at the end of every instruction handler. 123 This makes the interpreter a little larger overall, but you get to avoid 124 the (potentially expensive) branch back to the top of the switch statement. 125 </p><p> 126 Dalvik mterp goes one step further, using a computed goto instead of a goto 127 table. Instead of looking up the address in a table, which requires an 128 extra memory fetch on every instruction, mterp multiplies the opcode number 129 by a fixed value. By default, each handler is allowed 64 bytes of space. 130 </p><p> 131 Not all handlers fit in 64 bytes. Those that don't can have subroutines 132 or simply continue on to additional code outside the basic space. Some of 133 this is handled automatically by Dalvik, but there's no portable way to detect 134 overflow of a 64-byte handler until the VM starts executing. 135 </p><p> 136 The choice of 64 bytes is somewhat arbitrary, but has worked out well for 137 ARM and x86. 138 </p><p> 139 In the course of development it's useful to have C and assembly 140 implementations of each handler, and be able to flip back and forth 141 between them when hunting problems down. In mterp this is relatively 142 straightforward. You can always see the files being fed to the compiler 143 and assembler for your platform by looking in the 144 <code>dalvik/vm/mterp/out</code> directory. 145 </p><p> 146 The interpreter sources live in <code>dalvik/vm/mterp</code>. If you 147 haven't yet, you should read <code>dalvik/vm/mterp/README.txt</code> now. 148 </p> 149 150 151 <h3>Getting Started With Mterp</h3> 152 153 </p><p> 154 Getting started: 155 <ol> 156 <li>Decide on the name of your architecture. For the sake of discussion, 157 let's call it <code>myarch</code>. 158 <li>Make a copy of <code>dalvik/vm/mterp/config-allstubs</code> to 159 <code>dalvik/vm/mterp/config-myarch</code>. 160 <li>Create a <code>dalvik/vm/mterp/myarch</code> directory to hold your 161 source files. 162 <li>Add <code>myarch</code> to the list in 163 <code>dalvik/vm/mterp/rebuild.sh</code>. 164 <li>Make sure <code>dalvik/vm/Android.mk</code> will find the files for 165 your architecture. If <code>$(TARGET_ARCH)</code> is configured this 166 will happen automatically. 167 <li>Disable the Dalvik JIT. You can do this in the general device 168 configuration, or by editing the initialization of WITH_JIT in 169 <code>dalvik/vm/Dvm.mk</code> to always be <code>false</code>. 170 </ol> 171 </p><p> 172 You now have the basic framework in place. Whenever you make a change, you 173 need to perform two steps: regenerate the mterp output, and build the 174 core VM library. (It's two steps because we didn't want the build system 175 to require Python 2.5. Which, incidentally, you need to have.) 176 <ol> 177 <li>In the <code>dalvik/vm/mterp</code> directory, regenerate the contents 178 of the files in <code>dalvik/vm/mterp/out</code> by executing 179 <code>./rebuild.sh</code>. Note there are two files, one in C and one 180 in assembly. 181 <li>In the <code>dalvik</code> directory, regenerate the 182 <code>libdvm.so</code> library with <code>mm</code>. You can also use 183 <code>mmm dalvik/vm</code> from the top of the tree. 184 </ol> 185 </p><p> 186 This will leave you with an updated libdvm.so, which can be pushed out to 187 a device with <code>adb sync</code> or <code>adb push</code>. If you're 188 using the emulator, you need to add <code>make snod</code> (System image, 189 NO Dependency check) to rebuild the system image file. You should not 190 need to do a top-level "make" and rebuild the dependent binaries. 191 </p><p> 192 At this point you have an "all stubs" interpreter. You can see how it 193 works by examining <code>dalvik/vm/mterp/cstubs/entry.c</code>. The 194 code runs in a loop, pulling out the next opcode, and invoking the 195 handler through a function pointer. Each handler takes a "glue" argument 196 that contains all of the useful state. 197 </p><p> 198 Your goal is to replace the entry method, exit method, and each individual 199 instruction with custom implementations. The first thing you need to do 200 is create an entry function that calls the handler for the first instruction. 201 After that, the instructions chain together, so you don't need a loop. 202 (Look at the ARM or x86 implementation to see how they work.) 203 </p><p> 204 Once you have that, you need something to jump to. You can't branch 205 directly to the C stub because it's expecting to be called with a "glue" 206 argument and then return. We need a C stub "wrapper" that does the 207 setup and jumps directly to the next handler. We write this in assembly 208 and then add it to the config file definition. 209 </p><p> 210 To see how this works, create a file called 211 <code>dalvik/vm/mterp/myarch/stub.S</code> that contains one line: 212 <pre> 213 /* stub for ${opcode} */ 214 </pre> 215 Then, in <code>dalvik/vm/mterp/config-myarch</code>, add this below the 216 <code>handler-size</code> directive: 217 <pre> 218 # source for the instruction table stub 219 asm-stub myarch/stub.S 220 </pre> 221 </p><p> 222 Regenerate the sources with <code>./rebuild.sh</code>, and take a look 223 inside <code>dalvik/vm/mterp/out/InterpAsm-myarch.S</code>. You should 224 see 256 copies of the stub function in a single large block after the 225 <code>dvmAsmInstructionStart</code> label. The <code>stub.S</code> 226 code will be used anywhere you don't provide an assembly implementation. 227 </p><p> 228 Note that each block begins with a <code>.balign 64</code> directive. 229 This is what pads each handler out to 64 bytes. Note also that the 230 <code>${opcode}</code> text changed into an opcode name, which should 231 be used to call the C implementation (<code>dvmMterp_${opcode}</code>). 232 </p><p> 233 The actual contents of <code>stub.S</code> are up to you to define. 234 See <code>entry.S</code> and <code>stub.S</code> in the <code>armv5te</code> 235 or <code>x86</code> directories for working examples. 236 </p><p> 237 If you're working on a variation of an existing architecture, you may be 238 able to use most of the existing code and just provide replacements for 239 a few instructions. Look at the <code>vm/mterp/config-*</code> files 240 for examples. 241 </p> 242 243 244 <h3>Replacing Stubs</h3> 245 246 <p> 247 There are roughly 250 Dalvik opcodes, including some that are inserted by 248 <a href="dexopt.html">dexopt</a> and aren't described in the 249 <a href="dalvik-bytecode.html">Dalvik bytecode</a> documentation. Each 250 one must perform the appropriate actions, fetch the next opcode, and 251 branch to the next handler. The actions performed by the assembly version 252 must exactly match those performed by the C version (in 253 <code>dalvik/vm/mterp/c/OP_*</code>). 254 </p><p> 255 It is possible to customize the set of "optimized" instructions for your 256 platform. This is possible because optimized DEX files are not expected 257 to work on multiple devices. Adding, removing, or redefining instructions 258 is beyond the scope of this document, and for simplicity it's best to stick 259 with the basic set defined by the portable interpreter. 260 </p><p> 261 Once you have written a handler that looks like it should work, add 262 it to the config file. For example, suppose we have a working version 263 of <code>OP_NOP</code>. For demonstration purposes, fake it for now by 264 putting this into <code>dalvik/vm/mterp/myarch/OP_NOP.S</code>: 265 <pre> 266 /* This is my NOP handler */ 267 </pre> 268 </p><p> 269 Then, in the <code>op-start</code> section of <code>config-myarch</code>, add: 270 <pre> 271 op OP_NOP myarch 272 </pre> 273 </p><p> 274 This tells the generation script to use the assembly version from the 275 <code>myarch</code> directory instead of the C version from the <code>c</code> 276 directory. 277 </p><p> 278 Execute <code>./rebuild.sh</code>. Look at <code>InterpAsm-myarch.S</code> 279 and <code>InterpC-myarch.c</code> in the <code>out</code> directory. You 280 will see that the <code>OP_NOP</code> stub wrapper has been replaced with our 281 new code in the assembly file, and the C stub implementation is no longer 282 included. 283 </p><p> 284 As you implement instructions, the C version and corresponding stub wrapper 285 will disappear from the output files. Eventually you will have a 100% 286 assembly interpreter. You may find it saves a little time to examine 287 the output of your compiler for some of the operations. The 288 <a href="porting-proto.c.txt">porting-proto.c</a> sample code can be 289 helpful here. 290 </p> 291 292 293 <h3>Interpreter Switching</h3> 294 295 <p> 296 The Dalvik VM actually includes a third interpreter implementation: the debug 297 interpreter. This is a variation of the portable interpreter that includes 298 support for debugging and profiling. 299 </p><p> 300 When a debugger attaches, or a profiling feature is enabled, the VM 301 will switch interpreters at a convenient point. This is done at the 302 same time as the GC safe point check: on a backward branch, a method 303 return, or an exception throw. Similarly, when the debugger detaches 304 or profiling is discontinued, execution transfers back to the "fast" or 305 "portable" interpreter. 306 </p><p> 307 Your entry function needs to test the "entryPoint" value in the "glue" 308 pointer to determine where execution should begin. Your exit function 309 will need to return a boolean that indicates whether the interpreter is 310 exiting (because we reached the "bottom" of a thread stack) or wants to 311 switch to the other implementation. 312 </p><p> 313 See the <code>entry.S</code> file in <code>x86</code> or <code>armv5te</code> 314 for examples. 315 </p> 316 317 318 <h3>Testing</h3> 319 320 <p> 321 A number of VM tests can be found in <code>dalvik/tests</code>. The most 322 useful during interpreter development is <code>003-omnibus-opcodes</code>, 323 which tests many different instructions. 324 </p><p> 325 The basic invocation is: 326 <pre> 327 $ cd dalvik/tests 328 $ ./run-test 003 329 </pre> 330 </p><p> 331 This will run test 003 on an attached device or emulator. You can run 332 the test against your desktop VM by specifying <code>--reference</code> 333 if you suspect the test may be faulty. You can also use 334 <code>--portable</code> and <code>--fast</code> to explictly specify 335 one Dalvik interpreter or the other. 336 </p><p> 337 Some instructions are replaced by <code>dexopt</code>, notably when 338 "quickening" field accesses and method invocations. To ensure 339 that you are testing the basic form of the instruction, add the 340 <code>--no-optimize</code> option. 341 </p><p> 342 There is no in-built instruction tracing mechanism. If you want 343 to know for sure that your implementation of an opcode handler 344 is being used, the easiest approach is to insert a "printf" 345 call. For an example, look at <code>common_squeak</code> in 346 <code>dalvik/vm/mterp/armv5te/footer.S</code>. 347 </p><p> 348 At some point you need to ensure that debuggers and profiling work with 349 your interpreter. The easiest way to do this is to simply connect a 350 debugger or toggle profiling. (A future test suite may include some 351 tests for this.) 352 </p> 353 354 355 <h2>Other Performance Issues</h2> 356 357 <p> 358 The <code>System.arraycopy()</code> function is heavily used. The 359 implementation relies on the bionic C library to provide a fast, 360 platform-optimized data copy function for arrays with elements wider 361 than one byte. If you're not using bionic, or your platform does not 362 have an implementation of this method, Dalvik will use correct but 363 sub-optimal algorithms instead. For best performance you will want 364 to provide your own version. 365 </p><p> 366 See the comments in <code>dalvik/vm/native/java_lang_System.c</code> 367 for details. 368 </p> 369 370 <p> 371 <address>Copyright © 2009 The Android Open Source Project</address> 372 373 </body> 374 </html> 375
