From: Paul H. Hargrove (PHHargrove_at_lbl_dot_gov)
Date: Fri Jun 05 2009 - 12:03:27 PDT
Peter,
The CFLAGS you quote from our configure script are obtained by running
the kernel kbuild infrastructure on a dummy module:
make V=1 -C ${LINUX_SRCDIR} builddir=${TOPBUILDDIR}/conftestdir
where conftestdir contains a Makefile and conftest.c for a dummy kernel
module. The CFLAGS are then extracted from the output of that command.
So, it is the kernel's own build infrastructure (perhaps w/ gentoo
specific changes or additions) that is providing these CFLAGS.
I doubt that "-msoft-float" could be a factor. Along with "-mno-sse
-mno-mmx -mno-sse2 -mno-3dnow", these come from arch/x86/Makefile and
are included for all 32-bit x86 kernel builds to ensure that there are
no floating-point instructions in the kernel code other than inline
assembly which is guarded by "kernel_fpu_begin()" and
"kernel_fpu_end()". This is required because the kernel runs with fpu
register state "borrowed" from the previously running user process.
Similar flags are passed on other architectures for the same reason.
While I have no gentoo installations to test, I can run BLCR with a
vanilla 2.6.28 kernel on a Pentium4 with no problems, and have confirmed
that the CFLAGS there are very similar to yours:
-nostdinc -isystem
/usr/local/pkg/gcc-4.1.2/lib/gcc/i686-pc-linux-gnu/4.1.2/include
-D__KERNEL__ -I/lib/modules/2.6.28/build/include
-I/lib/modules/2.6.28/build/include2
-I/home/data1/phargrov/kernel-src/linux-2.6.28/include
-I/home/data1/phargrov/kernel-src/linux-2.6.28/arch/x86/include -include
/lib/modules/2.6.28/build/include/linux/autoconf.h
-I/data/blcr/BUILD-2.6.28/conftestdir -Wall -Wundef -Wstrict-prototypes
-Wno-trigraphs -fno-strict-aliasing -fno-common
-Werror-implicit-function-declaration -m32 -msoft-float -mregparm=3
-freg-struct-return -mpreferred-stack-boundary=2 -march=i686
-mtune=pentium4 -ffreestanding -pipe -Wno-sign-compare
-fno-asynchronous-unwind-tables -mno-sse -mno-mmx -mno-sse2 -mno-3dnow
-I/home/data1/phargrov/kernel-src/linux-2.6.28/arch/x86/include/asm/mach-default
-I/lib/modules/2.6.28/build/arch/x86/include/asm/mach-default
-fno-stack-protector -fomit-frame-pointer -Wdeclaration-after-statement
-Wno-pointer-sign -DMODULE -DKBUILD_STR(s)=#s
-DKBUILD_BASENAME=KBUILD_STR(conftest) -DKBUILD_MODNAME=KBUILD_STR(conftest)
So, I am afraid I still have no good guesses as to the source of your
problem.
-Paul
Peter Elias wrote:
> Hello,
> I have succesfully fixed my CHOST (and recompiled everything, to be
> sure). As one may expect, it did not help. But I have noticed the
> following when running configure:
>
> checking for flags to compile Linux kernel probes... -nostdinc
> -isystem /usr/lib/gcc/i686-pc-linux-gnu/4.3.2/include -D__KERNEL__
> -I/lib/modules/2.6.28-gentoo-r5/build/include
> -I/usr/src/linux-2.6.28-gentoo-r5/arch/x86/include -include
> /lib/modules/2.6.28-gentoo-r5/build/include/linux/autoconf.h -Wall
> -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing
> -fno-common -Werror-implicit-function-declaration -m32 -msoft-float
> -mregparm=3 -freg-struct-return -mpreferred-stack-boundary=2
> -march=athlon -Wa,-mtune=generic32 -ffreestanding -DCONFIG_AS_CFI=1
> -DCONFIG_AS_CFI_SIGNAL_FRAME=1 -pipe -Wno-sign-compare
> -fno-asynchronous-unwind-tables -mno-sse -mno-mmx -mno-sse2 -mno-3dnow
> -I/lib/modules/2.6.28-gentoo-r5/build/arch/x86/include/asm/mach-default
> -fno-stack-protector -fomit-frame-pointer
> -Wdeclaration-after-statement -Wno-pointer-sign -fwrapv -DMODULE
> -DKBUILD_STR(s)=#s -DKBUILD_BASENAME=KBUILD_STR(conftest)
> -DKBUILD_MODNAME=KBUILD_STR(conftest)
>
> Is that what it should be? Especially notice the usage of -msoft-float.
> How the compiler flags are guessed? I am now afraid that Gentoo
> manages CFLAGS slightly non-standard way. Could this cause the problem?
> Peter
>
>
> Peter Elias wrote:
>> Hello,
>> I apologize for my late answer; I had some troubles running X after
>> upgrading to Xorg 1.5 ...
>> I am running the latest gentoo-patched kernel
>> (linux-2.6.28-gentoo-r5), and NOT running Xen (nor any other
>> virtualization software).
>> Is there any chance that the failure of math.ct test signalizes
>> some hardware problem?
>> There is one thing I should probably try: when I have installed
>> Gentoo, I have downloaded wrong stage3 tarball, so now I have CHOST
>> set to i486-pc-linux-gnu (instead of i686-pc-linux-gnu). Anyway, I
>> have then compiled everything with -march=athlon option. As I have
>> learned, a change of CHOST leads to some (possibly lot of)
>> recompiling. I do not know what exactly will the change of CHOST do,
>> but who knows...
>> Peter
>>
>> Eric Roman wrote:
>>> Can you please let us know which kernel you're running?
>>>
>>> Eric
>>>
>>> On Thu, May 14, 2009 at 07:43:02AM -0700, Paul Hargrove wrote:
>>>> Peter,
>>>> The list address is checkpoint_at_lbl_dot_gov (I've cc:ed this reply and
>>>> set Reply-To)
>>>>
>>>> If you are seeing this failure, then I would NOT proceed with
>>>> using BLCR. We should figure out the source of the failure.
>>>> Given the things we've been through in the past, my only guess is
>>>> that you are using Xen with a version older than 3.1.2. If that is
>>>> the case, the bug is actaully in Xen nor BLCR and you will need to
>>>> upgrade Xen. If that is not the case, let us know and we'll see if
>>>> we can figure out the source of the problem.
>>>>
>>>> -Paul
>>>>
>>>> Peter Elias wrote:
>>>>> Dear Paul H. Hargrove!
>>>>>
>>>>> I apologize for sending this e-mail to your personal address. I
>>>>> planned to send it to the checkpoint mailing list (archives are at
>>>>> www.nersc.gov), but I couldn't find list's address.
>>>>> I am trying to compile and run blcr-0.8.1 on my machine (Gentoo
>>>>> Linux, AMD Athlon CPU). Compilation went without errors and from
>>>>> all tests only one has failed:
>>>>>
>>>>> .../blcr-0.8.1/builddir/tests/.libs/lt-math[31579]: file "
>>>>> ../../tests/math.c", line 54, in math_callback: FP restore failure
>>>>> nan != 38842.2
>>>>>
>>>>> cr_poll_request() failed: Restart aborted: restart cancelled and
>>>>> job killed
>>>>> FAIL: math.ct
>>>>>
>>>>> Shall I install and use blcr? I want to save and then repeatedly
>>>>> restore some ocaml sessions.
>>>>>
>>>>> With best regards,
>>>>>
>>>>> Peter Elias
>>>>>
>>>>
>>>> --
>>>> Paul H. Hargrove PHHargrove_at_lbl_dot_gov
>>>> Future Technologies Group Tel: +1-510-495-2352
>>>> HPC Research Department Fax: +1-510-486-6900
>>>> Lawrence Berkeley National Laboratory
>>
>>
>
--
Paul H. Hargrove PHHargrove_at_lbl_dot_gov
Future Technologies Group Tel: +1-510-495-2352
HPC Research Department Fax: +1-510-486-6900
Lawrence Berkeley National Laboratory