| Summary: | Hyper Threading slowdown | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Alex Efros <powerman-asdf> |
| Component: | [OLD] Core system | Assignee: | Gentoo Kernel Bug Wranglers and Kernel Maintainers <kernel> |
| Status: | RESOLVED UPSTREAM | ||
| Severity: | major | ||
| Priority: | High | ||
| Version: | unspecified | ||
| Hardware: | x86 | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
*gentoo-sources-2.6.14-r5 (15 Dec 2005) 15 Dec 2005; Daniel Drake <dsd@gentoo.org> +gentoo-sources-2.6.14-r5.ebuild: Update to Linux 2.6.14.4. Really fix the hyperthreading slowdown. Fix NUMA boot problem. Reopen it if still does not work when with latest gentoo-sources. No, it doesn't work with sys-kernel/gentoo-sources-2.6.14-r5. :-( Don't sure is it important, but I use sys-devel/gcc-3.4.4-r1 (HARDENED) for compiling kernel. And, just for the case, here is my `emerge info`: ---cut--- Portage 2.0.51.22-r3 (hardened/x86/2.6, gcc-3.4.4, glibc-2.3.5-r2, 2.6.14- gentoo-r5 i686) ================================================================= System uname: 2.6.14-gentoo-r5 i686 Intel(R) Pentium(R) 4 CPU 3.00GHz Gentoo Base System version 1.6.13 dev-lang/python: 2.3.5-r2, 2.4.2 sys-apps/sandbox: 1.2.12 sys-devel/autoconf: 2.13, 2.59-r6 sys-devel/automake: 1.4_p6, 1.5, 1.6.3, 1.7.9-r1, 1.8.5-r3, 1.9.6-r1 sys-devel/binutils: 2.16.1 sys-devel/libtool: 1.4.3-r4, 1.5.20 virtual/os-headers: 2.6.11-r2 ACCEPT_KEYWORDS="x86" AUTOCLEAN="yes" CBUILD="i686-pc-linux-gnu" CFLAGS="-O2 -march=pentium4 -pipe" CHOST="i686-pc-linux-gnu" CONFIG_PROTECT="/etc /usr/kde/2/share/config /usr/kde/3/share/config /usr/share/ config /var/qmail/alias /var/qmail/control" CONFIG_PROTECT_MASK="/etc/gconf /etc/terminfo /etc/env.d" CXXFLAGS="-O2 -march=pentium4 -pipe" DISTDIR="/usr/portage-distfiles" FEATURES="autoconfig buildpkg ccache distlocks sandbox sfperms strict userpriv usersandbox" GENTOO_MIRRORS="http://pandemonium.tiscali.de/pub/gentoo/ ftp://pandemonium. tiscali.de/pub/gentoo/ http://ftp-stud.fht-esslingen.de/pub/Mirrors/gentoo/ http://ftp.belnet.be/mirror/rsync.gentoo.org/gentoo/" MAKEOPTS="-j3" PKGDIR="/usr/portage-packages" PORTAGE_TMPDIR="/var/tmp" PORTDIR="/usr/portage" PORTDIR_OVERLAY="/usr/local/portage /usr/local/portage-java-experimental /usr/ local/portage-power" SYNC="rsync://rsync.gentoo.org/gentoo-portage" USE="acpi aim apache2 berkdb bzip2 crypt curl dlloader expat gd gdbm gif hardened icq idn imagemagick imap irc jabber java javascript jpeg junit lm_sensors mailbox mbox mmx msn mysql ncurses nls nptl nptlonly pam pcre perl pic png pwdb python readline rss slang sse sse2 ssl sysfs tcpd tiff truetype udev userlocales x86 xinetd xml2 yahoo zlib userland_GNU kernel_linux elibc_glibc" Unset: ASFLAGS, CTARGET, LANG, LC_ALL, LDFLAGS, LINGUAS ---cut--- Please test vanilla-sources-2.6.15_rc6 with this additional patch applied: http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff_plain;h=1e483969930a82e16767884449f3a121a817ef00;hp=4c0335526c95d90a1d958e0059f40a5745fc7c5d (In reply to comment #3) > Please test vanilla-sources-2.6.15_rc6 with this additional patch applied: > http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff_plain;h=1e483969930a82e16767884449f3a121a817ef00;hp=4c0335526c95d90a1d958e0059f40a5745fc7c5d Bad idea. :-( With this patch kernel don't boot. This is dedicated server on good enough hosting which provide 'serial console' and 'rescue network boot'. I've checked serial console, but there was no output from booting kernel, so it's probably hang before serial console driver was loaded. Then I've used 'rescue network boot' mode to login to my server and change grub.conf to load another kernel - same vanilla-sources-2.6.15_rc6, but without that patch. It's boot ok, but "hyper threading slowdown" bug still exists in this version. I notice some part of this patch in '#ifdef CONFIG_HOTPLUG_CPU', so I've enabled in my kernel CONFIG_HOTPLUG_CPU and tried again to boot kernel with this patch - without success - it hangs again. Have you tried 2.6.15-rc6 without the patch? Also, maybe you could try 2.6.14-r5 with CONFIG_HOTPLUG_CPU enabled and see if that helps. Yeah, I have tried 2.6.15-rc6 without the patch (see comment #4). I've just tried 2.6.14-r5 with CONFIG_HOTPLUG_CPU enabled. This don't fix the bug. Looks like an upstream bug - if you still get this with 2.6.15 vanilla please post to the LKML and reference this bug. Thanks! 2.6.15 vanilla hangs if CONFIG_HOTPLUG_CPU enabled. If CONFIG_HOTPLUG_CPU disabled - it boot ok, but 'hyper threading slowdown' issue still exists. :-( Enabling/disabling CONFIG_SCHED_SMT has no effect. |
Executing simple CPU test on Intel(R) Pentium(R) 4 CPU 3.00GHz (with Hyper Threading enabled) show about 3 times slowdown when 2 processes use CPU simultaneously. This slowdown exists in: - sys-kernel/hardened-sources-2.6.14-r1 - sys-kernel/gentoo-sources-2.6.14-r4 - sys-kernel/vanilla-sources-2.6.14.2 but there no such slowdown in kernel supplied with Fedora Core 2 "Linux version 2.6.14-051115a (somebody@somewhere) (gcc version 3.3.5 (Debian 1:3.3.5-13)) #1 SMP Tue Nov 15 13:06:41 CET 2005" (probably this kernel isn't really from Fedora, maybe it was compiled by my ISP for my dedicated server, don't sure). My CPU test is simple. Here results from Fedora: ---cut--- # time perl -e '$pid=fork if @ARGV; print "$$ started\n"; $i++ for 1..100000000; print "$$ finished\n"; print("waiting for $pid\n"), wait() if $pid' 1001 started 1001 finished real 0m11.118s user 0m11.117s sys 0m0.000s # time perl -e '$pid=fork if @ARGV; print "$$ started\n"; $i++ for 1..100000000; print "$$ finished\n"; print("waiting for $pid\n"), wait() if $pid' 1 1003 started 1002 started 1003 finished 1002 finished waiting for 1003 real 0m23.818s ### ATTENTION HERE !!! user 0m47.627s sys 0m0.000s ---cut--- and here results from Gentoo (with any *-sources listed above): ---cut--- # time perl -e '$pid=fork if @ARGV; print "$$ started\n"; $i++ for 1..100000000; print "$$ finished\n"; print("waiting for $pid\n"), wait() if $pid' 4493 started 4493 finished real 0m12.490s user 0m12.485s sys 0m0.004s # time perl -e '$pid=fork if @ARGV; print "$$ started\n"; $i++ for 1..100000000; print "$$ finished\n"; print("waiting for $pid\n"), wait() if $pid' 1 4495 started 4494 started 4494 finished waiting for 4495 4495 finished real 1m7.266s ### ATTENTION HERE !!! user 2m14.432s sys 0m0.000s ---cut--- All these kernels was compiled with exactly same .config (which I take from Fedora's /proc/config.gz). All these kernels show nearly same information in / proc/cpuinfo, dmesg and lsmod. Probably some patch (from Fedora?) should be added to gentoo-sources & hardened- sources to fix this issue...