Official thread: "zen-sources" - Part 7

[kernelOfTruth]

Well, at least I know it works on this machine. I just finished an overnight emerge -aev system, and I woke to a finished job, and a computer still running with BFS. It seems to like single cores...a lot.

Blessed be!
Pappy

sweet !

so at least it's stable on Uniprocessor systems now

Pappy,

could you please post the .config section for that box - do you have

CONFIG_SCHED_SMT or any of the SMP (e.g. CONFIG_SMP) enabled ?

thanks

[aTan]

I've tried 209 with 2.6.31-rc9 in virtualbox uniprocessor environment and it works fine. kernel was SMP PREEMPT.

[kernelOfTruth]

Well, at least I know it works on this machine. I just finished an overnight emerge -aev system, and I woke to a finished job, and a computer still running with BFS. It seems to like single cores...a lot.

Blessed be!
Pappy

was this a 32bit / i686 box or also and x86_64 kernel ?

I've tried 209 with 2.6.31-rc9 in virtualbox uniprocessor environment and it works fine. kernel was SMP PREEMPT.

interesting

it was a x86_64 kernel, right ?

if yes it's not 64bit-specific (I just had an idea that it might be specific to 64bit) - but rather multi-processor, etc. caused

[xmixahlx]

k3b is locking up with several bfs releases (2.6.30 & 2.6.31) and processes are halting/unkillable.... going back to cfs for a bit.

amd64, debian unstable, monolithic kernel.

[cheater1034]

2.6.31-rc9-zen1 is hot for me, did 3 boots (didn't extensively test this, but every boot was lightning fast) - later i shall do 6-10 boots to see if the locking bug is still there, but it appears early that it is not

also wrote/ported the cfq drain async i/o operations for bfq, and i'm using that right now as well

will port the bfq drain patch to 2.6.30 as well

[darklegion]

sweet !

so at least it's stable on Uniprocessor systems now

I have a dual core e5200, and provided it gets past the init stage, BFS has always been stable for me.Usually specific BFS releases will boot 10/10 times, while other releases will boot intermittently, and others again will not boot at all.I don't think I've ever got a 2.6.31 release to boot at all, for that matter.

[pappy_mcfae]

Well, at least I know it works on this machine. I just finished an overnight emerge -aev system, and I woke to a finished job, and a computer still running with BFS. It seems to like single cores...a lot.

Blessed be!
Pappy

sweet !

so at least it's stable on Uniprocessor systems now

Pappy,

could you please post the .config section for that box - do you have

CONFIG_SCHED_SMT or any of the SMP (e.g. CONFIG_SMP) enabled ?

thanks

No, it's a Celeron running as i686. I'm now past two/thirds of the way though the emerge -aev world. I am also typing this at the same time. BFS is really running smoothly.

The .config that did it is here. I've set it up to stay forever. It is made with the 208 patch as well as the race condition patch.

Blessed be!
Pappy

[aTan]

it was a x86_64 kernel, right ?

No, it was i686.

[tranquilcool]

i am going cfs for a while.
lockups, no sound in video card,
hald starts once every couple of boot up with bfs etc.
kernel is 2.6.31-rc9-zen1.

[pappy_mcfae]

There are less than 260 packages left to emerge. This machine has been running a full day (plus an hour) with 2.6.30-bfs. It has survived lots of acid tests.

I just set up my other single-processor machine with the latest patch, version 209. I am going to subject it to the same tests. Presently it is running emerge -aev system --keep-going. While it is currently only into package 8 of 124, I have a feeling from the way it's acting that this system is probably going to be stable as well with bfs. I will report back tomorrow.

Blessed be!
Pappy

[rahulthewall]

New Gentoo installation with BFS + Latest Zen Sources + ext4 + ~x86 (Dual Core - T2300) and have been compiling packages like crazy since last night.

Everything rock solid and fast.

[darklegion]

Well, I pulled the latest release again with 2.6.31-rc9 + BFS, and it failed to boot again.So I tried with CFS, and no issues booting so it's not an issue with the 2.6.31 kernel.Also, the problems I had with CFS are somewhat reduced with 2.6.31.So either CFS is improving or some other part of the kernel is getting faster.Maybe this http://kernelnewbies.org/Linux_2_6_31#h ... 414f366ef2 is related.

[kernelOfTruth]

New Gentoo installation with BFS + Latest Zen Sources + ext4 + ~x86 (Dual Core - T2300) and have been compiling packages like crazy since last night.

Everything rock solid and fast.

that was with SMP enabled, right ?

so it's more of a 64bit-problem with the kernel than a 32bit-problem ?!


edit:

2.6.31-rc9-zen1 keeps on failing with:

--
CC init/version.o
LD init/built-in.o
LD .tmp_vmlinux1
kernel/built-in.o: In function `rcu_batches_completed':
(.text+0x4b065): undefined reference to `rcu_batches_completed_sched'
make: *** [.tmp_vmlinux1] Error 1

it doesn't matter whether I choose classic rcu or tree-based rcu

[rahulthewall]

New Gentoo installation with BFS + Latest Zen Sources + ext4 + ~x86 (Dual Core - T2300) and have been compiling packages like crazy since last night.

Everything rock solid and fast.

that was with SMP enabled, right ?

so it's more of a 64bit-problem with the kernel than a 32bit-problem ?!


edit:

2.6.31-rc9-zen1 keeps on failing with:

--
CC init/version.o
LD init/built-in.o
LD .tmp_vmlinux1
kernel/built-in.o: In function `rcu_batches_completed':
(.text+0x4b065): undefined reference to `rcu_batches_completed_sched'
make: *** [.tmp_vmlinux1] Error 1

it doesn't matter whether I choose classic rcu or tree-based rcu

SMP is enabled, and I have preemtible RCU.

[cheater1034]

New Gentoo installation with BFS + Latest Zen Sources + ext4 + ~x86 (Dual Core - T2300) and have been compiling packages like crazy since last night.

Everything rock solid and fast.

that was with SMP enabled, right ?

so it's more of a 64bit-problem with the kernel than a 32bit-problem ?!


edit:

2.6.31-rc9-zen1 keeps on failing with:

--
CC init/version.o
LD init/built-in.o
LD .tmp_vmlinux1
kernel/built-in.o: In function `rcu_batches_completed':
(.text+0x4b065): undefined reference to `rcu_batches_completed_sched'
make: *** [.tmp_vmlinux1] Error 1

it doesn't matter whether I choose classic rcu or tree-based rcu

it's fixed, but why not pick preemptible rcu

[xmixahlx]

can 2.6.31-rc9 be tagged?

[kernelOfTruth]

[snip] remove unneeded double quote [snip]

it's fixed, but why not pick preemptible rcu

it would be nice to be able to do so but I can't choose it

RCU Implementation ────────────────────────┐
│ Use the arrow keys to navigate this window or press the hotkey of │
│ the item you wish to select followed by the <SPACE BAR>. Press │
│ <?> for additional information about this option. │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ ( ) Classic RCU │ │
│ │ (X) Tree-based hierarchical RCU │ │
│ │ │ │
│ │ │ │
│ │ │ │
│ │ │ │
│ └────────────────────────────────────────────────────────────────┘ │
├────────────────────────────────────────────────────────────────────┤
│ <Select> < Help > │
└──────────────────────────────────────────────────────

BFS seems to be broken for me with 2.6.31-rc9-zen1 -> error message during boot

[rahulthewall]

Here is my config for you:
http://pastebin.com/m31e12d50

This has PREEMP_RCU enabled. Edit it to suit yours.

[kernelOfTruth]

Here is my config for you:
http://pastebin.com/m31e12d50

This has PREEMP_RCU enabled. Edit it to suit yours.

thanks !

right now it always resets back to tree-based rcu during compile

seems that kernel-configuration stuff hates me

[cheater1034]

the only thing preempt_rcu depends on is CONFIG_PREEMPT, so enable that and then preempt rcu

[pappy_mcfae]

My second uniprocessor system has finished emerge -aev system without any issues. I am about to start emerge -aev world as well.

I can't wait for the fix for multi-processor systems.

Blessed be!
Pappy

[albright]

trying to get the BFS working (just for fun) ...

I downloaded the zen kernel with this

Code: Select all

git clone git://zen-sources.org/zen/zen.git linux-2.6-zen

But when I try to run make menuconfig I get errors, like this:

Code: Select all

  HOSTLD  scripts/kconfig/mconf
collect2: ld terminated with signal 11 [Segmentation fault]
/usr/lib/gcc/i686-pc-linux-gnu/4.3.2/../../../../i686-pc-linux-gnu/bin/ld: i386:x86-64 architecture of input file `scripts/kconfig/mconf.o' is incompatible with i386 output

This is an intel dual core (thinkpad x300), with x86 architecture
(CHOST="i686-pc-linux-gnu")[/code]

I should add, the gentoo and tuxonice kernels from portage compile
perfectly.

Why would this be happening? TIA,

[ponciarello]

tried new stuff (can't resist )

Code: Select all

kernel/rcutree.c:142: error: redefinition of 'rcu_batches_completed'
kernel/rcutree_plugin.h:56: error: previous definition of 'rcu_batches_completed' was here
kernel/rcutree.c:146: error: redefinition of '__kstrtab_rcu_batches_completed'
kernel/rcutree_plugin.h:59: error: previous definition of '__kstrtab_rcu_batches_completed' was here
kernel/rcutree.c:146: error: redefinition of '__ksymtab_rcu_batches_completed'
kernel/rcutree_plugin.h:59: error: previous definition of '__ksymtab_rcu_batches_completed' was here
make[1]: *** [kernel/rcutree.o] Error 1

compiling now with preemptive rcu

EDIT: up an runnin'

[Vash63]

Hmm, rc8-zen1 was working fine, but for some reason the same config (make oldconfig'd) on rc9 crashes almost immediately with NX-bit or other assorted errors less than a half-second into the kernel booting process. It's on an i7 if that matters.

Edit: Fixed it. It was an 'early panic', still not sure what caused it but I redid my config, guess something in the old config was screwing it up with all the new options added.

[albright]

about my earlier post re ld segfaulting

It was some kind of problem with cross compiling;
removing /usr/lib/distcc/bin from $PATH fixed it.

sorry for the noise

BTW, BFS seems to work fine (but the real benefit
with 2.6.31 for me seems to be intel graphics)