Date: Sun, 13 Aug 2017 15:54:42 +0200 From: Andre Albsmeier <andre@fbsd.e4m.org> To: Konstantin Belousov <kostikbel@gmail.com> Cc: Andre Albsmeier <andre@fbsd.e4m.org>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-11@freebsd.org Subject: Re: svn commit: r322345 - stable/11/lib/libc/x86/sys Message-ID: <20170813135442.GA28728@voyager> In-Reply-To: <20170813133736.GO1700@kib.kiev.ua> References: <201708100900.v7A90FjI094474@repo.freebsd.org> <20170813131311.GA24473@voyager> <20170813133736.GO1700@kib.kiev.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 13-Aug-2017 at 16:37:36 +0300, Konstantin Belousov wrote: > On Sun, Aug 13, 2017 at 03:13:11PM +0200, Andre Albsmeier wrote: > > On Thu, 10-Aug-2017 at 09:00:15 +0000, Konstantin Belousov wrote: > > > Author: kib > > > Date: Thu Aug 10 09:00:15 2017 > > > New Revision: 322345 > > > URL: https://svnweb.freebsd.org/changeset/base/322345 > > > > > > Log: > > > MFC r321608: > > > Use MFENCE to serialize RDTSC on non-Intel CPUs. > > > > This broke libc on my > > > > FreeBSD clang version 4.0.0 (tags/RELEASE_400/final 297347) (based on LLVM 4.0.0) > > VT(vga): text 80x25 > > CPU: AMD-K6tm w/ multimedia extensions (199.44-MHz 586-class CPU) > > Origin="AuthenticAMD" Id=0x562 Family=0x5 Model=0x6 Stepping=2 > > Features=0x8001bf<FPU,VME,DE,PSE,TSC,MSR,MCE,CX8,MMX> > > AMD Features=0x400<<b10>> > > > > On system start, it actually makes various programmes hang until > > Ctrl-C is pressed. Reboot is not possible (endlessly hanging with > > 'zillions of defunct processes). > > > > Going back to r322042 fixes it... > > Indeed code is broken on i386 machines which have CPUID but no SSE. Wonder how many of these still exist... > Try this. I just compiled and installed libc.so.7 with this patch on the machine in question and the test I used for hunting down the reason (ls -l /lib) was OK. I then rebooted the whole box and it looked good as well. I did not compile stuff which links libc statically and hung before (devd) but I am pretty sure this will be OK as well... Thanks, -Andre > > diff --git a/lib/libc/x86/sys/__vdso_gettc.c b/lib/libc/x86/sys/__vdso_gettc.c > index 4ada5e7eac1..ef2ca69edf7 100644 > --- a/lib/libc/x86/sys/__vdso_gettc.c > +++ b/lib/libc/x86/sys/__vdso_gettc.c > @@ -101,6 +101,7 @@ init_fence(void) > #if defined(__i386__) > u_int cpuid_supported, p[4]; > > + lfence_works = LMB_NONE; > __asm __volatile( > " pushfl\n" > " popl %%eax\n" > @@ -121,8 +122,7 @@ init_fence(void) > cpuidp(0x1, p); > if ((p[3] & CPUID_SSE2) != 0) > lfence_works = select_lmb(); > - } else > - lfence_works = LMB_NONE; > + } > #elif defined(__amd64__) > lfence_works = select_lmb(); > #else -- My other computer is your windows box.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20170813135442.GA28728>