Add linux-csky list and originally cced people and replace the linux-mm with linux-kbuild list, which match the discussion better and avoid email noise for mm people. On Tue, Jun 22, 2021 at 05:48:18PM +0800, Tang, Feng wrote: > On Tue, Jun 22, 2021 at 09:49:05AM +0800, kernel test robot wrote: > > tree: https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git master > > head: 889bab4c367a0ef58208fd80fafa74bb6e2dca26 > > commit: cf536e185869d4815d506e777bcca6edd9966a6e [3665/11714] Makefile: extend 32B aligned debug option to 64B aligned > > config: csky-randconfig-c024-20210622 (attached as .config) > > compiler: csky-linux-gcc (GCC) 9.3.0 > > reproduce (this is a W=1 build): > > wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross > > chmod +x ~/bin/make.cross > > # https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commi... > > git remote add linux-next https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git > > git fetch --no-tags linux-next master > > git checkout cf536e185869d4815d506e777bcca6edd9966a6e > > # save the attached .config to linux build tree > > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-9.3.0 make.cross ARCH=csky > Thanks for the report. > > I just reproduced the issue locally, with similar log: > > CC drivers/target/target_core_sbc.o > {standard input}: Assembler messages: > {standard input}:2644: Error: pcrel offset for branch to .LS0015 too far (0x3e) > {standard input}:2653: Error: pcrel offset for branch to .LS0015 too far (0x34) > {standard input}:2659: Error: pcrel offset for branch to .LS0015 too far (0x2c) > make[2]: *** [scripts/Makefile.build:272: drivers/target/target_core_sbc.o] Error 1 > m > > And when I changed the function align option from 64 to 32, the > compilation did pass, so this looks to be related with the alignment > option. > > I'm not very familiar with compiler, and will try to check more. I did some check and found the error info comes from the assembler for 'csky' in binutils' gas/config/tc-csky.c, and I could hardly dive deeper into the code as limited by my knowledge. > I know it works on x86_64, but don't know how about other > architectures, and if 'csky' is the only not working one, one > workaround I can think of is to add kconfig dependency for !csky I reused the 0day's reproduce process, and tried on arm64, powerpc64, and arc, the kernel compilation all succeeded. So maybe we can add some dependency for this debug option like: diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug index 3cf48998a374..eb035b31657f 100644 --- a/lib/Kconfig.debug +++ b/lib/Kconfig.debug @@ -415,7 +415,8 @@ config SECTION_MISMATCH_WARN_ONLY If unsure, say Y. config DEBUG_FORCE_FUNCTION_ALIGN_64B - bool "Force all function address 64B aligned" if EXPERT + bool "Force all function address 64B aligned" + depends on EXPERT && (X86_64 || ARM64 || PPC32 || PPC64 || ARC) help There are cases that a commit from one domain changes the function address alignment of other domains, and cause magic performance Any comments? thanks!

- Feng > Thanks, > Feng > >

On Thu, Jun 24, 2021 at 04:06:51PM +0800, Guo Ren wrote: > On Wed, Jun 23, 2021 at 4:59 PM Feng Tang <feng.tang(a)intel.com&gt; wrote: > > > > Add linux-csky list and originally cced people and replace the linux-mm > > with linux-kbuild list, which match the discussion better and avoid > > email noise for mm people. > > > > On Tue, Jun 22, 2021 at 05:48:18PM +0800, Tang, Feng wrote: > > > On Tue, Jun 22, 2021 at 09:49:05AM +0800, kernel test robot wrote: > > > > tree: https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git master > > > > head: 889bab4c367a0ef58208fd80fafa74bb6e2dca26 > > > > commit: cf536e185869d4815d506e777bcca6edd9966a6e [3665/11714] Makefile: extend 32B aligned debug option to 64B aligned > > > > config: csky-randconfig-c024-20210622 (attached as .config) > > > > compiler: csky-linux-gcc (GCC) 9.3.0 > > > > reproduce (this is a W=1 build): > > > > wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross > > > > chmod +x ~/bin/make.cross > > > > # https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commi... > > > > git remote add linux-next https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git > > > > git fetch --no-tags linux-next master > > > > git checkout cf536e185869d4815d506e777bcca6edd9966a6e > > > > # save the attached .config to linux build tree > > > > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-9.3.0 make.cross ARCH=csky > > > > > Thanks for the report. > > > > > > I just reproduced the issue locally, with similar log: > > > > > > CC drivers/target/target_core_sbc.o > > > {standard input}: Assembler messages: > > > {standard input}:2644: Error: pcrel offset for branch to .LS0015 too far (0x3e) > > > {standard input}:2653: Error: pcrel offset for branch to .LS0015 too far (0x34) > > > {standard input}:2659: Error: pcrel offset for branch to .LS0015 too far (0x2c) > > > make[2]: *** [scripts/Makefile.build:272: drivers/target/target_core_sbc.o] Error 1 > > > m > > > > > > And when I changed the function align option from 64 to 32, the > > > compilation did pass, so this looks to be related with the alignment > > > option. > > > > > > I'm not very familiar with compiler, and will try to check more. > > > > I did some check and found the error info comes from the assembler > > for 'csky' in binutils' gas/config/tc-csky.c, and I could hardly > > dive deeper into the code as limited by my knowledge. > > > > > > > I know it works on x86_64, but don't know how about other > > > architectures, and if 'csky' is the only not working one, one > > > workaround I can think of is to add kconfig dependency for !csky > > > > I reused the 0day's reproduce process, and tried on arm64, powerpc64, > > and arc, the kernel compilation all succeeded. So maybe we can > > add some dependency for this debug option like: > > > > diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug > > index 3cf48998a374..eb035b31657f 100644 > > --- a/lib/Kconfig.debug > > +++ b/lib/Kconfig.debug > > @@ -415,7 +415,8 @@ config SECTION_MISMATCH_WARN_ONLY > > If unsure, say Y. > > > > config DEBUG_FORCE_FUNCTION_ALIGN_64B > > - bool "Force all function address 64B aligned" if EXPERT > > + bool "Force all function address 64B aligned" > > + depends on EXPERT && (X86_64 || ARM64 || PPC32 || PPC64 || ARC) > > help > > There are cases that a commit from one domain changes the function > > address alignment of other domains, and cause magic performance > > > > Any comments? thanks! > Have tried riscv32? I don't think csky is the only problem. > > Assembler just follows gcc backend result and also depends on ISA > design. Our condition branch seems not far enough. Thanks for the info. I just used 0day's cross compile env to compile 'riscv32', and it succeeded with 64B align option. (the kernel config is attached)

Anyway, this debug option is for debug only, and nice to have. If it doesn't work on one architecture, I dont' think it means a severe problem. Thanks, Feng > Cc: xianmiao_qu(a)c-sky.com > > > > > - Feng > > > > > Thanks, > > > Feng > > > > > > > > > > -- > Best Regards > Guo Ren > > ML: https://lore.kernel.org/linux-csky/