Changeset efcb393 for chapter05/toolchaintechnotes.xml
- Timestamp:
- 05/03/2020 09:02:51 PM (4 years ago)
- Branches:
- 10.0, 10.0-rc1, 10.1, 10.1-rc1, 11.0, 11.0-rc1, 11.0-rc2, 11.0-rc3, 11.1, 11.1-rc1, 11.2, 11.2-rc1, 11.3, 11.3-rc1, 12.0, 12.0-rc1, 12.1, 12.1-rc1, arm, bdubbs/gcc13, ml-11.0, multilib, renodr/libudev-from-systemd, s6-init, trunk, xry111/arm64, xry111/arm64-12.0, xry111/clfs-ng, xry111/lfs-next, xry111/loongarch, xry111/loongarch-12.0, xry111/loongarch-12.1, xry111/mips64el, xry111/pip3, xry111/rust-wip-20221008, xry111/update-glibc
- Children:
- a715dec
- Parents:
- 9d719e2
- File:
-
- 1 edited
-
chapter05/toolchaintechnotes.xml (modified) (1 diff)
Legend:
- Unmodified
- Added
- Removed
-
chapter05/toolchaintechnotes.xml
r9d719e2 refcb393 25 25 the most educational value at the same time.</para> 26 26 27 <note> 28 <para>Before continuing, be aware of the name of the working platform, 29 often referred to as the target triplet. A simple way to determine the 30 name of the target triplet is to run the <command>config.guess</command> 31 script that comes with the source for many packages. Unpack the Binutils 32 sources and run the script: <userinput>./config.guess</userinput> and note 33 the output. For example, for a 32-bit Intel processor the 34 output will be <emphasis>i686-pc-linux-gnu</emphasis>. On a 64-bit 35 system it will be <emphasis>x86_64-pc-linux-gnu</emphasis>.</para> 36 37 <para>Also be aware of the name of the platform's dynamic linker, often 38 referred to as the dynamic loader (not to be confused with the standard 39 linker <command>ld</command> that is part of Binutils). The dynamic linker 40 provided by Glibc finds and loads the shared libraries needed by a program, 41 prepares the program to run, and then runs it. The name of the dynamic 42 linker for a 32-bit Intel machine will be <filename 43 class="libraryfile">ld-linux.so.2</filename> (<filename 44 class="libraryfile">ld-linux-x86-64.so.2</filename> for 64-bit systems). A 45 sure-fire way to determine the name of the dynamic linker is to inspect a 46 random binary from the host system by running: <userinput>readelf -l 47 <name of binary> | grep interpreter</userinput> and noting the 48 output. The authoritative reference covering all platforms is in the 49 <filename>shlib-versions</filename> file in the root of the Glibc source 50 tree.</para> 51 </note> 52 53 <para>Some key technical points of how the <xref 54 linkend="chapter-temporary-tools"/> build method works:</para> 55 56 <itemizedlist> 57 <listitem> 58 <para>Slightly adjusting the name of the working platform, by changing the 59 "vendor" field target triplet by way of the 60 <envar>LFS_TGT</envar> variable, ensures that the first build of Binutils 61 and GCC produces a compatible cross-linker and cross-compiler. Instead of 62 producing binaries for another architecture, the cross-linker and 63 cross-compiler will produce binaries compatible with the current 64 hardware.</para> 65 </listitem> 66 <listitem> 67 <para> The temporary libraries are cross-compiled. Because a 68 cross-compiler by its nature cannot rely on anything from its host 69 system, this method removes potential contamination of the target 70 system by lessening the chance of headers or libraries from the host 71 being incorporated into the new tools. Cross-compilation also allows for 72 the possibility of building both 32-bit and 64-bit libraries on 64-bit 73 capable hardware.</para> 74 </listitem> 75 <listitem> 76 <para>Careful manipulation of the GCC source tells the compiler which target 77 dynamic linker will be used.</para> 78 </listitem> 79 </itemizedlist> 80 81 <para>Binutils is installed first because the <command>configure</command> 82 runs of both GCC and Glibc perform various feature tests on the assembler 83 and linker to determine which software features to enable or disable. This 84 is more important than one might first realize. An incorrectly configured 85 GCC or Glibc can result in a subtly broken toolchain, where the impact of 86 such breakage might not show up until near the end of the build of an 87 entire distribution. A test suite failure will usually highlight this error 88 before too much additional work is performed.</para> 89 90 <para>Binutils installs its assembler and linker in two locations, 91 <filename class="directory">/tools/bin</filename> and <filename 92 class="directory">/tools/$LFS_TGT/bin</filename>. The tools in one 93 location are hard linked to the other. An important facet of the linker is 94 its library search order. Detailed information can be obtained from 95 <command>ld</command> by passing it the <parameter>--verbose</parameter> 96 flag. For example, an <userinput>ld --verbose | grep SEARCH</userinput> 97 will illustrate the current search paths and their order. It shows which 98 files are linked by <command>ld</command> by compiling a dummy program and 99 passing the <parameter>--verbose</parameter> switch to the linker. For example, 100 <userinput>gcc dummy.c -Wl,--verbose 2>&1 | grep succeeded</userinput> 101 will show all the files successfully opened during the linking.</para> 102 103 <para>The next package installed is GCC. An example of what can be 104 seen during its run of <command>configure</command> is:</para> 105 106 <screen><computeroutput>checking what assembler to use... /tools/i686-lfs-linux-gnu/bin/as 107 checking what linker to use... /tools/i686-lfs-linux-gnu/bin/ld</computeroutput></screen> 108 109 <para>This is important for the reasons mentioned above. It also demonstrates 110 that GCC's configure script does not search the PATH directories to find which 111 tools to use. However, during the actual operation of <command>gcc</command> 112 itself, the same search paths are not necessarily used. To find out which 113 standard linker <command>gcc</command> will use, run: 114 <userinput>gcc -print-prog-name=ld</userinput>.</para> 115 116 <para>Detailed information can be obtained from <command>gcc</command> by 117 passing it the <parameter>-v</parameter> command line option while compiling 118 a dummy program. For example, <userinput>gcc -v dummy.c</userinput> will show 119 detailed information about the preprocessor, compilation, and assembly stages, 120 including <command>gcc</command>'s included search paths and their order.</para> 121 122 <para>Next installed are sanitized Linux API headers. These allow the standard 123 C library (Glibc) to interface with features that the Linux kernel will 124 provide.</para> 125 126 <para>The next package installed is Glibc. The most important considerations 127 for building Glibc are the compiler, binary tools, and kernel headers. The 128 compiler is generally not an issue since Glibc will always use the compiler 129 relating to the <parameter>--host</parameter> parameter passed to its 130 configure script; e.g. in our case, the compiler will be 131 <command>i686-lfs-linux-gnu-gcc</command>. The binary tools and kernel 132 headers can be a bit more complicated. Therefore, take no risks and use the 133 available configure switches to enforce the correct selections. After the run 134 of <command>configure</command>, check the contents of the 135 <filename>config.make</filename> file in the <filename 136 class="directory">glibc-build</filename> directory for all important details. 137 Note the use of <parameter>CC="i686-lfs-gnu-gcc"</parameter> to control which 138 binary tools are used and the use of the <parameter>-nostdinc</parameter> and 139 <parameter>-isystem</parameter> flags to control the compiler's include 140 search path. These items highlight an important aspect of the Glibc 141 package—it is very self-sufficient in terms of its build machinery and 142 generally does not rely on toolchain defaults.</para> 143 144 <para>During the second pass of Binutils, we are able to utilize the 145 <parameter>--with-lib-path</parameter> configure switch to control 146 <command>ld</command>'s library search path.</para> 147 148 <para>For the second pass of GCC, its sources also need to be modified to 149 tell GCC to use the new dynamic linker. Failure to do so will result in the 150 GCC programs themselves having the name of the dynamic linker from the host 151 system's <filename class="directory">/lib</filename> directory embedded into 152 them, which would defeat the goal of getting away from the host. From this 153 point onwards, the core toolchain is self-contained and self-hosted. The 154 remainder of the <xref linkend="chapter-temporary-tools"/> packages all build 155 against the new Glibc in <filename 156 class="directory">/tools</filename>.</para> 157 158 <para>Upon entering the chroot environment in <xref 159 linkend="chapter-building-system"/>, the first major package to be 160 installed is Glibc, due to its self-sufficient nature mentioned above. 161 Once this Glibc is installed into <filename 162 class="directory">/usr</filename>, we will perform a quick changeover of the 163 toolchain defaults, and then proceed in building the rest of the target 164 LFS system.</para> 27 <para>The build process is based on the process of 28 <emphasis>cross-compilation</emphasis>. Cross-compilation is normally used 29 for building a compiler and its toolchain for a machine different from 30 the one that is used for the build. This is not strictly needed for LFS, 31 since the machine where the new system will run is the same as the one 32 used for the build. But cross-compilation has the great advantage that 33 anything that is cross-compiled cannot depend on the host environment.</para> 34 35 <sect2 id="cross-compile" xreflabel="About Cross-Compilation"> 36 37 <title>About Cross-Compilation</title> 38 39 <para>Cross-compilation involves some concepts that deserve a section on 40 their own. Although this section may be omitted in a first reading, it 41 is strongly suggested to come back to it later in order to get a full 42 grasp of the build process.</para> 43 44 <para>Let us first define some terms used in this context:</para> 45 46 <variablelist> 47 <varlistentry><term>build</term><listitem> 48 <para>is the machine where we build programs. Note that this machine 49 is referred to as the <quote>host</quote> in other 50 sections.</para></listitem> 51 </varlistentry> 52 53 <varlistentry><term>host</term><listitem> 54 <para>is the machine/system where the built programs will run. Note 55 that this use of <quote>host</quote> is not the same as in other 56 sections.</para></listitem> 57 </varlistentry> 58 59 <varlistentry><term>target</term><listitem> 60 <para>is only used for compilers. It is the machine the compiler 61 produces code for. It may be different from both build and 62 host.</para></listitem> 63 </varlistentry> 64 65 </variablelist> 66 67 <para>As an example, let us imagine the following scenario: we may have a 68 compiler on a slow machine only, let's call the machine A, and the compiler 69 ccA. We may have also a fast machine (B), but with no compiler, and we may 70 want to produce code for a another slow machine (C). Then, to build a 71 compiler for machine C, we would have three stages:</para> 72 73 <informaltable align="center"> 74 <tgroup cols="5"> 75 <colspec colnum="1" align="center"/> 76 <colspec colnum="2" align="center"/> 77 <colspec colnum="3" align="center"/> 78 <colspec colnum="4" align="center"/> 79 <colspec colnum="5" align="left"/> 80 <thead> 81 <row><entry>Stage</entry><entry>Build</entry><entry>Host</entry> 82 <entry>Target</entry><entry>Action</entry></row> 83 </thead> 84 <tbody> 85 <row> 86 <entry>1</entry><entry>A</entry><entry>A</entry><entry>B</entry> 87 <entry>build cross-compiler cc1 using ccA on machine A</entry> 88 </row> 89 <row> 90 <entry>2</entry><entry>A</entry><entry>B</entry><entry>B</entry> 91 <entry>build cross-compiler cc2 using cc1 on machine A</entry> 92 </row> 93 <row> 94 <entry>3</entry><entry>B</entry><entry>C</entry><entry>C</entry> 95 <entry>build compiler ccC using cc2 on machine B</entry> 96 </row> 97 </tbody> 98 </tgroup> 99 </informaltable> 100 101 <para>Then, all the other programs needed by machine C can be compiled 102 using cc2 on the fast machine B. Note that unless B can run programs 103 produced for C, there is no way to test the built programs until machine 104 C itself is running. For example, for testing ccC, we may want to add a 105 fourth stage:</para> 106 107 <informaltable align="center"> 108 <tgroup cols="5"> 109 <colspec colnum="1" align="center"/> 110 <colspec colnum="2" align="center"/> 111 <colspec colnum="3" align="center"/> 112 <colspec colnum="4" align="center"/> 113 <colspec colnum="5" align="left"/> 114 <thead> 115 <row><entry>Stage</entry><entry>Build</entry><entry>Host</entry> 116 <entry>Target</entry><entry>Action</entry></row> 117 </thead> 118 <tbody> 119 <row> 120 <entry>4</entry><entry>C</entry><entry>C</entry><entry>C</entry> 121 <entry>rebuild and test ccC using itself on machine C</entry> 122 </row> 123 </tbody> 124 </tgroup> 125 </informaltable> 126 127 <para>In the example above, only cc1 and cc2 are cross-compilers, that is, 128 they produce code for a machine different from the one they are run on. 129 The other compilers ccA and ccC produce code for the machine they are run 130 on. Such compilers are called <emphasis>native</emphasis> compilers.</para> 131 132 </sect2> 133 134 <sect2 id="lfs-cross"> 135 <title>Implementation of Cross-Compilation for LFS</title> 136 137 <note> 138 <para>Almost all the build systems use names of the form 139 cpu-vendor-kernel-os referred to as the machine triplet. An astute 140 reader may wonder why a <quote>triplet</quote> refers to a four component 141 name. The reason is history: initially, three component names were enough 142 to designate unambiguously a machine, but with new machines and systems 143 appearing, that proved insufficient. The word <quote>triplet</quote> 144 remained. A simple way to determine your machine triplet is to run 145 the <command>config.guess</command> 146 script that comes with the source for many packages. Unpack the Binutils 147 sources and run the script: <userinput>./config.guess</userinput> and note 148 the output. For example, for a 32-bit Intel processor the 149 output will be <emphasis>i686-pc-linux-gnu</emphasis>. On a 64-bit 150 system it will be <emphasis>x86_64-pc-linux-gnu</emphasis>.</para> 151 152 <para>Also be aware of the name of the platform's dynamic linker, often 153 referred to as the dynamic loader (not to be confused with the standard 154 linker <command>ld</command> that is part of Binutils). The dynamic linker 155 provided by Glibc finds and loads the shared libraries needed by a 156 program, prepares the program to run, and then runs it. The name of the 157 dynamic linker for a 32-bit Intel machine will be <filename 158 class="libraryfile">ld-linux.so.2</filename> (<filename 159 class="libraryfile">ld-linux-x86-64.so.2</filename> for 64-bit systems). A 160 sure-fire way to determine the name of the dynamic linker is to inspect a 161 random binary from the host system by running: <userinput>readelf -l 162 <name of binary> | grep interpreter</userinput> and noting the 163 output. The authoritative reference covering all platforms is in the 164 <filename>shlib-versions</filename> file in the root of the Glibc source 165 tree.</para> 166 </note> 167 168 <para>In order to fake a cross compilation, the name of the host triplet 169 is slightly adjusted by changing the "vendor" field in the 170 <envar>LFS_TGT</envar> variable. We also use the 171 <parameter>--with-sysroot</parameter> when building the cross linker and 172 cross compiler, to tell them where to find the needed host files. This 173 ensures none of the other programs built in <xref 174 linkend="chapter-temporary-tools"/> can link to libraries on the build 175 machine. Only two stages are mandatory, and one more for tests:</para> 176 177 <informaltable align="center"> 178 <tgroup cols="5"> 179 <colspec colnum="1" align="center"/> 180 <colspec colnum="2" align="center"/> 181 <colspec colnum="3" align="center"/> 182 <colspec colnum="4" align="center"/> 183 <colspec colnum="5" align="left"/> 184 <thead> 185 <row><entry>Stage</entry><entry>Build</entry><entry>Host</entry> 186 <entry>Target</entry><entry>Action</entry></row> 187 </thead> 188 <tbody> 189 <row> 190 <entry>1</entry><entry>pc</entry><entry>pc</entry><entry>lfs</entry> 191 <entry>build cross-compiler cc1 using cc-pc on pc</entry> 192 </row> 193 <row> 194 <entry>2</entry><entry>pc</entry><entry>lfs</entry><entry>lfs</entry> 195 <entry>build compiler cc-lfs using cc1 on pc</entry> 196 </row> 197 <row> 198 <entry>3</entry><entry>lfs</entry><entry>lfs</entry><entry>lfs</entry> 199 <entry>rebuild and test cc-lfs using itself on lfs</entry> 200 </row> 201 </tbody> 202 </tgroup> 203 </informaltable> 204 205 <para>In the above table, <quote>on pc</quote> means the commands are run 206 on a machine using the already installed distribution. <quote>On 207 lfs</quote> means the commands are run in a chrooted environment.</para> 208 209 <para>Now, there is more about cross-compiling: the C language is not 210 just a compiler, but also defines a standard library. In this book, the 211 GNU C library, named glibc, is used. This library must 212 be compiled for the lfs machine, that is, using the cross compiler cc1. 213 But the compiler itself uses an internal library implementing complex 214 instructions not available in the assembler instruction set. This 215 internal library is named libgcc, and must be linked to the glibc 216 library to be fully functional! Furthermore, the standard library for 217 C++ (libstdc++) also needs being linked to glibc. The solution 218 to this chicken and egg problem is to first build a degraded cc1+libgcc, 219 lacking some fuctionalities such as threads and exception handling, then 220 build glibc using this degraded compiler (glibc itself is not 221 degraded), then build libstdc++. But this last library will lack the 222 same functionalities as libgcc.</para> 223 224 <para>This is not the end of the story: the conclusion of the preceding 225 paragraph is that cc1 is unable to build a fully functional libstdc++, but 226 this is the only compiler available for building the C/C++ libraries 227 during stage 2! Of course, the compiler built during stage 2, cc-lfs, 228 would be able to build those libraries, but (i) the build system of 229 gcc does not know that it is usable on pc, and (ii) using it on pc 230 would be at risk of linking to the pc libraries, since cc-lfs is a native 231 compiler. So we have to build libstdc++ later, in chroot.</para> 232 233 </sect2> 234 235 <sect2 id="other-details"> 236 237 <title>Other procedural details</title> 238 239 <para>The cross-compiler will be installed in a separate <filename 240 class="directory">$LFS/tools</filename> directory, since it will not 241 be part of the final system.</para> 242 243 <para>Binutils is installed first because the <command>configure</command> 244 runs of both GCC and Glibc perform various feature tests on the assembler 245 and linker to determine which software features to enable or disable. This 246 is more important than one might first realize. An incorrectly configured 247 GCC or Glibc can result in a subtly broken toolchain, where the impact of 248 such breakage might not show up until near the end of the build of an 249 entire distribution. A test suite failure will usually highlight this error 250 before too much additional work is performed.</para> 251 252 <para>Binutils installs its assembler and linker in two locations, 253 <filename class="directory">$LFS/tools/bin</filename> and <filename 254 class="directory">$LFS/tools/$LFS_TGT/bin</filename>. The tools in one 255 location are hard linked to the other. An important facet of the linker is 256 its library search order. Detailed information can be obtained from 257 <command>ld</command> by passing it the <parameter>--verbose</parameter> 258 flag. For example, <command>$LFS_TGT-ld --verbose | grep SEARCH</command> 259 will illustrate the current search paths and their order. It shows which 260 files are linked by <command>ld</command> by compiling a dummy program and 261 passing the <parameter>--verbose</parameter> switch to the linker. For 262 example, 263 <command>$LFS_TGT-gcc dummy.c -Wl,--verbose 2>&1 | grep succeeded</command> 264 will show all the files successfully opened during the linking.</para> 265 266 <para>The next package installed is GCC. An example of what can be 267 seen during its run of <command>configure</command> is:</para> 268 269 <screen><computeroutput>checking what assembler to use... /mnt/lfs/tools/i686-lfs-linux-gnu/bin/as 270 checking what linker to use... /mnt/lfs/tools/i686-lfs-linux-gnu/bin/ld</computeroutput></screen> 271 272 <para>This is important for the reasons mentioned above. It also 273 demonstrates that GCC's configure script does not search the PATH 274 directories to find which tools to use. However, during the actual 275 operation of <command>gcc</command> itself, the same search paths are not 276 necessarily used. To find out which standard linker <command>gcc</command> 277 will use, run: <command>$LFS_TGT-gcc -print-prog-name=ld</command>.</para> 278 279 <para>Detailed information can be obtained from <command>gcc</command> by 280 passing it the <parameter>-v</parameter> command line option while compiling 281 a dummy program. For example, <command>gcc -v dummy.c</command> will show 282 detailed information about the preprocessor, compilation, and assembly 283 stages, including <command>gcc</command>'s included search paths and their 284 order.</para> 285 286 <para>Next installed are sanitized Linux API headers. These allow the 287 standard C library (Glibc) to interface with features that the Linux 288 kernel will provide.</para> 289 290 <para>The next package installed is Glibc. The most important 291 considerations for building Glibc are the compiler, binary tools, and 292 kernel headers. The compiler is generally not an issue since Glibc will 293 always use the compiler relating to the <parameter>--host</parameter> 294 parameter passed to its configure script; e.g. in our case, the compiler 295 will be <command>$LFS_TGT-gcc</command>. The binary tools and kernel 296 headers can be a bit more complicated. Therefore, take no risks and use 297 the available configure switches to enforce the correct selections. After 298 the run of <command>configure</command>, check the contents of the 299 <filename>config.make</filename> file in the <filename 300 class="directory">build</filename> directory for all important details. 301 Note the use of <parameter>CC="$LFS_TGT-gcc"</parameter> (with 302 <envar>$LFS_TGT</envar> expanded) to control which binary tools are used 303 and the use of the <parameter>-nostdinc</parameter> and 304 <parameter>-isystem</parameter> flags to control the compiler's include 305 search path. These items highlight an important aspect of the Glibc 306 package—it is very self-sufficient in terms of its build machinery 307 and generally does not rely on toolchain defaults.</para> 308 309 <para>As said above, the standard C++ library is compiled next, followed 310 by all the programs that need themselves to be built. The install step 311 uses the <envar>DESTDIR</envar> variable to have the programs land into 312 the LFS filesystem.</para> 313 314 <para>Then the native lfs compiler is built. First Binutils Pass 2, with 315 the same <envar>DESTDIR</envar> install as the other programs, then the 316 second pass of GCC, omitting libstdc++ and other non-important libraries. 317 Due to some weird logic in GCC's configure script, 318 <envar>CC_FOR_TARGET</envar> ends up as <command>cc</command> when host 319 is the same as target, but is different from build. This is why 320 <parameter>CC_FOR_TARGET=$LFS_TGT-gcc</parameter> is put explicitely into 321 the configure options.</para> 322 323 <para>Upon entering the chroot environment in <xref 324 linkend="chapter-building-system"/>, the first task is to install 325 libstdc++. Then temporary installations of programs needed for the proper 326 operation of the toolchain are performed. Programs needed for testing 327 other programs are also built. From this point onwards, the 328 core toolchain is self-contained and self-hosted. In the remainder of 329 the <xref linkend="chapter-building-system"/>, final versions of all the 330 packages needed for a fully functional system are built, tested and 331 installed.</para> 332 333 </sect2> 165 334 166 335 </sect1>
Note:
See TracChangeset
for help on using the changeset viewer.
