Le 11/11/17 à 17:34, Fengguang Wu a écrit : > On Sun, Nov 12, 2017 at 09:23:52AM +0800, Alexei Starovoitov wrote: >> On 11/12/17 9:18 AM, Fengguang Wu wrote: >>> On Sun, Nov 12, 2017 at 09:14:14AM +0800, Alexei Starovoitov wrote: >>>> On 11/12/17 8:23 AM, kbuild test robot wrote: >>>>> tree: >>>>> https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git >>>>> master >>>>> head: 7c5556decd0a629e9ee02e93653f75ba7b7da03c >>>>> commit: 638f5b90d46016372a8e3e0a434f199cc5e12b8c [488/665] bpf: >>>>> reduce verifier memory consumption >>>>> config: mips-64r6el_defconfig (attached as .config) >>>>> compiler: mips64el-linux-gnuabi64-gcc (Debian 7.2.0-11) 7.2.0 >>>>> reproduce: >>>>> git checkout 638f5b90d46016372a8e3e0a434f199cc5e12b8c >>>>> # save the attached .config to linux build tree >>>>> make.cross ARCH=mips >>>>> >>>>> All errors (new ones prefixed by >>): >>>>> >>>>> kernel/bpf/verifier.o: In function >>>>> `realloc_verifier_state.isra.19': >>>>>>> verifier.c:(.text+0x36fc): undefined reference to `__multi3' >>>> >>>> that's a known issue with gcc 7 on mips that is "optimizing" >>>> normal 64-bit multiply into 128-bit variant. >>>> Nothing to fix on the kernel side. >>> >>> Good to know that! Do you think it a good idea to blacklist __multi3 >>> errors in mips builds? >> >> I would do so. yes. > > OK. > >> Though digging further this function was added to >> arch/sparc/lib/multi3.S >> since gcc doing the same "optimization" there. >> Adding asm code doesn't look right to me. I'd rather push >> gcc folks to avoid such codegen. > > Sure, I just forwarded the original report to GCC list.
Thomas encountered a similar problem, reported on linux-mips here: https://www.linux-mips.org/archives/linux-mips/2017-08/msg00041.html -- Florian
