2.5.44-mm2 compile error using gcc 3.2 (gcc 2.96 works fine).

2.5.44-mm2 compile error using gcc 3.2 (gcc 2.96 works fine).

[Steven Cole]

I got the following error compiling 2.5.44-mm2 with gcc 3.2
as shipped with Mandrake 9.0 (3.2-1mdk).

kernel/softirq.c:353: cpu_nfb causes a section type conflict
make[1]: *** [kernel/softirq.o] Error 1

I was able to compile 2.5.44-mm2 with the same .config on another
machine using gcc 2.96 as shipped with RedHat 7.3 without this error.
Plain 2.5.44 built OK using gcc 3.2.

FWIW, I booted that 2.5.44-mm2 with CONFIG_SHAREPTE=y built with gcc
2.96 and I have run KDE 3.0.3 without any problems so far.

Steven


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: 2.5.44-mm2 compile error using gcc 3.2 (gcc 2.96 works fine).

[Andrew Morton]

Steven Cole wrote:
> 
> I got the following error compiling 2.5.44-mm2 with gcc 3.2
> as shipped with Mandrake 9.0 (3.2-1mdk).
> 
> kernel/softirq.c:353: cpu_nfb causes a section type conflict
> make[1]: *** [kernel/softirq.o] Error 1
> 
> I was able to compile 2.5.44-mm2 with the same .config on another
> machine using gcc 2.96 as shipped with RedHat 7.3 without this error.
> Plain 2.5.44 built OK using gcc 3.2.
> 
> FWIW, I booted that 2.5.44-mm2 with CONFIG_SHAREPTE=y built with gcc
> 2.96 and I have run KDE 3.0.3 without any problems so far.
> 

Well gosh, you made me build gcc-3.2.  Couldn't resist benchmarking it:

time make -j3 bzImage ; size vmlinux:

gcc-3.2:
	781.26s user 62.84s system 185% cpu 7:34.41 total
        text     data    bss    dec      hex    filename
	3395957  448896  419476 4264329  411189 vmlinux-3.2

2.95.3:
	454.57s user 52.45s system 188% cpu 4:29.31 total
        text     data    bss    dec      hex    filename
	3055661  445064  419476 3920201  3bd149 vmlinux-2.95.3

2.91.66:
	420.78s user 51.87s system 188% cpu 4:11.09 total
        text     data    bss    dec      hex    filename
	3125069  338536  526100 3989705  3ce0c9 vmlinux-2.91.66

Kinda makes you wonder why we're bothering, but ho hum.

I didn't see the error to which you refer.  What binutils
are you using?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: 2.5.44-mm2 compile error using gcc 3.2 (gcc 2.96 works fine).

[Steven Cole]

On Mon, 2002-10-21 at 15:05, Andrew Morton wrote:
> Steven Cole wrote:
> > 
> > I got the following error compiling 2.5.44-mm2 with gcc 3.2
> > as shipped with Mandrake 9.0 (3.2-1mdk).
> > 
> > kernel/softirq.c:353: cpu_nfb causes a section type conflict
> > make[1]: *** [kernel/softirq.o] Error 1
> > 
> > I was able to compile 2.5.44-mm2 with the same .config on another
> > machine using gcc 2.96 as shipped with RedHat 7.3 without this error.
> > Plain 2.5.44 built OK using gcc 3.2.
> > 
> > FWIW, I booted that 2.5.44-mm2 with CONFIG_SHAREPTE=y built with gcc
> > 2.96 and I have run KDE 3.0.3 without any problems so far.
> > 
> 
> Well gosh, you made me build gcc-3.2.  Couldn't resist benchmarking it:
> 
> time make -j3 bzImage ; size vmlinux:
> 
> gcc-3.2:
> 	781.26s user 62.84s system 185% cpu 7:34.41 total
>         text     data    bss    dec      hex    filename
> 	3395957  448896  419476 4264329  411189 vmlinux-3.2
> 
> 2.95.3:
> 	454.57s user 52.45s system 188% cpu 4:29.31 total
>         text     data    bss    dec      hex    filename
> 	3055661  445064  419476 3920201  3bd149 vmlinux-2.95.3
> 
> 2.91.66:
> 	420.78s user 51.87s system 188% cpu 4:11.09 total
>         text     data    bss    dec      hex    filename
> 	3125069  338536  526100 3989705  3ce0c9 vmlinux-2.91.66
> 
> Kinda makes you wonder why we're bothering, but ho hum.

Yeah, I know.  Ahh, for the days of 2.7.2.3.

> 
> I didn't see the error to which you refer.  What binutils
> are you using?
> 

binutils version is 2.12.90.0.15 for Mandrake 9.0.

BTW, I did a make mrproper on the gcc 3.2 box, retrieved the .config,
recompiled, and got the very same "section type conflict" error as
before.

After running the gcc 2.96 2.5.44-mm2 for a while longer, I started up
dbench and ran some an increasing client load up to 24 clients.  I
started a new Konsole in KDE and the system hung, not even responding to
pings. That failure was repeatable once, but after those two hangs which
required a hard reset, the system was able to run dbench 32 and launch
new Konsoles without hanging.  Non-deterministic behavior is so much
fun.

I'll do more testing tomorrow.

Steven

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: 2.5.44-mm2 compile error using gcc 3.2 (gcc 2.96 works fine).

[Andrew Morton]

Steven Cole wrote:
> 
> ..
> binutils version is 2.12.90.0.15 for Mandrake 9.0.

Thanks.
 
> BTW, I did a make mrproper on the gcc 3.2 box, retrieved the .config,
> recompiled, and got the very same "section type conflict" error as
> before.
> 
> After running the gcc 2.96 2.5.44-mm2 for a while longer, I started up
> dbench and ran some an increasing client load up to 24 clients.  I
> started a new Konsole in KDE and the system hung, not even responding to
> pings. That failure was repeatable once, but after those two hangs which
> required a hard reset, the system was able to run dbench 32 and launch
> new Konsoles without hanging.  Non-deterministic behavior is so much
> fun.

You're on SMP, yes?  Please test with Hugh's "[PATCH] mm mremap freeze"
patch applied.

But it should have responded to pings even if deadlocked there.

Are you using "nmi_watchdog=1"?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: 2.5.44-mm2 compile error using gcc 3.2 (gcc 2.96 works fine).

[Steven Cole]

On Mon, 2002-10-21 at 17:21, Andrew Morton wrote:
> Steven Cole wrote:
> > 
> > ..
> > binutils version is 2.12.90.0.15 for Mandrake 9.0.
> 
> Thanks.
>  
> > BTW, I did a make mrproper on the gcc 3.2 box, retrieved the .config,
> > recompiled, and got the very same "section type conflict" error as
> > before.
> > 
> > After running the gcc 2.96 2.5.44-mm2 for a while longer, I started up
> > dbench and ran some an increasing client load up to 24 clients.  I
> > started a new Konsole in KDE and the system hung, not even responding to
> > pings. That failure was repeatable once, but after those two hangs which
> > required a hard reset, the system was able to run dbench 32 and launch
> > new Konsoles without hanging.  Non-deterministic behavior is so much
> > fun.
> 
> You're on SMP, yes?  Please test with Hugh's "[PATCH] mm mremap freeze"
> patch applied.

I have two machines for testing:

1) SMP box is dual PIII, SCSI, 1GB, RH 7.3 base distro, X not installed.
2) UP  box is PIII, IDE, 256MB, LM 9.0 base distro, X and KDE 3.0.3.

The build failure and lockups were on the UP box.  
The SMP box also ran 2.5.44-mm2 for several hours under load with no
failures.  It was used to successfully build the UP 2.5.44-mm2 kernel.

> 
> But it should have responded to pings even if deadlocked there

Yep, that was weird.  I'm only certain about the first hang when it did
not respond to pings.
.
> 
> Are you using "nmi_watchdog=1"?
> 

Nope.  My understanding is that is only applicable for SMP systems, but
I can add that to the lilo.conf append line if it would do any good on
this troublesome UP box.

Steven

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: 2.5.44-mm2 compile error using gcc 3.2 (gcc 2.96 works fine).

[Andrew Morton]

Steven Cole wrote:
> 
> > Are you using "nmi_watchdog=1"?
> >
> 
> Nope.  My understanding is that is only applicable for SMP systems, but
> I can add that to the lilo.conf append line if it would do any good on
> this troublesome UP box.

It should still work OK if you have the IO-APIC enabled.  

Running an SMP kernel on UP sometimes will catch deadlocks
which a UP build would miss.  But in this case, it might
make it go away - your SMP box is unaffected.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/