[arch-dev-public] [signoff] binutils-2.21 toolchain rebuild
Here is a summary of the changes: binutils-2.21-1 - major upstream update - provide gold linker as ld.gold (default remains the usual bfd linker) glibc-2.12.2-1 - *minimum kernel version required = 2.6.27* - bump to latest upstream - build completely outside source directory - remove patches included upstream - update patch for origin privilege exploit from Fedora - keep scsi.h - provide gai.conf (FS#21742) linux-api-headers-2.6.36.2-1 - bump to latest upstream - remove scsi.h (back in glibc) gcc-4.5.2-1 - upstream update - removed de_DE check as - fix/clarify some pkgdesc values libtool-2.4-2 - gcc verbump rebuild You can see the current toolchain testsuite status at https://wiki.archlinux.org/index.php/User:Allan/Toolchain . There are some failures there that I am still looking into, but nothing to be concerned about... One binutils issues that people may want to keep an eye out for is http://sourceware.org/bugzilla/show_bug.cgi?id=12327 . This only affects people building kernels with CONFIG_RELOCATABLE=y (not used in the Arch kernels) who will get a boot panic. Other than that, the binutils update seems relatively issue free. I have been running it on my system for the last week while waiting on the glibc and gcc releases without noticing any issues. Allan
On Fri, 2010-12-17 at 16:02 +1000, Allan McRae wrote:
- removed de_DE check as
What's the reason for that? Doesn't gcc check for locale support using de_DE anymore?
On 17/12/10 20:37, Jan de Groot wrote:
On Fri, 2010-12-17 at 16:02 +1000, Allan McRae wrote:
- removed de_DE check as
What's the reason for that? Doesn't gcc check for locale support using de_DE anymore?
It has not for ages provided you have --enable-clocale=gnu in the configure line. Allan
On Fri, Dec 17, 2010 at 7:02 AM, Allan McRae <allan@archlinux.org> wrote:
- provide gold linker as ld.gold (default remains the usual bfd linker)
Is this gold built with plugin support? This is one of the really nice features.
On 18/12/10 04:55, Jan Steffens wrote:
On Fri, Dec 17, 2010 at 7:02 AM, Allan McRae<allan@archlinux.org> wrote:
- provide gold linker as ld.gold (default remains the usual bfd linker)
Is this gold built with plugin support? This is one of the really nice features.
Umm... no actually. Long story: I had it there and then disabled it as I got confused with enabling LTO plugin support for the BFD linker which requires gcc to be built with --enable-gold, which I do not want to do yet. But that is unneeded for plugin support in the gold linker and the BFD LTO plugin is only actually available in binutils git so none of that mattered anyway... So I should re-enable it. I also noticed I did not enable threading support for the gold linker. I will do a rebuild of binutils enabling both of those tomorrow. Allan
On 17/12/10 16:02, Allan McRae wrote:
Here is a summary of the changes:
binutils-2.21-1 - major upstream update - provide gold linker as ld.gold (default remains the usual bfd linker)
glibc-2.12.2-1 - *minimum kernel version required = 2.6.27* - bump to latest upstream - build completely outside source directory - remove patches included upstream - update patch for origin privilege exploit from Fedora - keep scsi.h - provide gai.conf (FS#21742)
linux-api-headers-2.6.36.2-1 - bump to latest upstream - remove scsi.h (back in glibc)
gcc-4.5.2-1 - upstream update - removed de_DE check as - fix/clarify some pkgdesc values
libtool-2.4-2 - gcc verbump rebuild
Updates: binutils-2.21-2 - enable plugin and threading support in gold linker gcc-4.5.2-2 - enable linker plugin support. gcc will use ld(.bfd) linker by default but switch to the ld.gold linker when -fuse-linker-plugin is specified. Hopefully this has now hit the right balance where all the features of the gold linker can be used but the usual bfd linker still serves as our default linker. Allan
On Sun, Dec 19, 2010 at 6:14 AM, Allan McRae <allan@archlinux.org> wrote:
gcc-4.5.2-2 - enable linker plugin support. gcc will use ld(.bfd) linker by default but switch to the ld.gold linker when -fuse-linker-plugin is specified.
Where do you have this from? I can't find anything about this "with-plugin-linker" option (not even in the source). gcc still seems to use /usr/bin/ld when I call it via "gcc -v -flto" or "gcc -v -flto -fuse-linker-plugin". It doesn't care when I remove /usr/bin/ld.gold.
On 20/12/10 07:23, Jan Steffens wrote:
On Sun, Dec 19, 2010 at 6:14 AM, Allan McRae<allan@archlinux.org> wrote:
gcc-4.5.2-2 - enable linker plugin support. gcc will use ld(.bfd) linker by default but switch to the ld.gold linker when -fuse-linker-plugin is specified.
Where do you have this from? I can't find anything about this "with-plugin-linker" option (not even in the source). gcc still seems to use /usr/bin/ld when I call it via "gcc -v -flto" or "gcc -v -flto -fuse-linker-plugin". It doesn't care when I remove /usr/bin/ld.gold.
Bah... This took me a while to figure out as everything is working with the toolchain on my system (I am currently testing glibc and gcc snapshots). It turns out the option is --with-plugin-ld and I copied it wrong from my local PKGBUILDs. Fixed in gcc-4.5.2-3.
gcc -fuse-linker-plugin -v -o test test.c ... GNU assembler version 2.21.0 (i686-pc-linux-gnu) using BFD version (GNU Binutils) 2.21.0.20101217 ... /usr/bin/ld.gold -plugin /usr/lib/gcc/i686-pc-linux-gnu/4.5.2/liblto_plugin.so -plugin-opt=/usr/lib/gcc/i686-pc-linux-gnu/4.5.2/lto-wrapper ...
Allan
[2010-12-20 10:47:43 +1000] Allan McRae:
Fixed in gcc-4.5.2-3.
If you want signoffs for this and the rest of the toolchain in [testing], here is one for i686; it has all been working perfectly well for me. -- Gaetan
On Fri, Dec 17, 2010 at 7:02 AM, Allan McRae <allan@archlinux.org> wrote:
Here is a summary of the changes:
binutils-2.21-1 - major upstream update - provide gold linker as ld.gold (default remains the usual bfd linker)
glibc-2.12.2-1 - *minimum kernel version required = 2.6.27* - bump to latest upstream - build completely outside source directory - remove patches included upstream - update patch for origin privilege exploit from Fedora - keep scsi.h - provide gai.conf (FS#21742)
linux-api-headers-2.6.36.2-1 - bump to latest upstream - remove scsi.h (back in glibc)
gcc-4.5.2-1 - upstream update - removed de_DE check as - fix/clarify some pkgdesc values
libtool-2.4-2 - gcc verbump rebuild
You can see the current toolchain testsuite status at https://wiki.archlinux.org/index.php/User:Allan/Toolchain . There are some failures there that I am still looking into, but nothing to be concerned about...
One binutils issues that people may want to keep an eye out for is http://sourceware.org/bugzilla/show_bug.cgi?id=12327 . This only affects people building kernels with CONFIG_RELOCATABLE=y (not used in the Arch kernels) who will get a boot panic.
Other than that, the binutils update seems relatively issue free. I have been running it on my system for the last week while waiting on the glibc and gcc releases without noticing any issues.
Allan
Seems to work fine, signoff both.
participants (4)
-
Allan McRae
-
Gaetan Bisson
-
Jan de Groot
-
Jan Steffens