2.6.12.1 problems I meet (please CC: me)

2.6.12.1 problems I meet (please CC: me)

[Wang Jian]

Hi,

I use a customized kernel to do packets analysis. The analysis code is
linked into kernel. It will vmalloc() nearly 128M (a little less) when
initialized.

The original code runs on 2.6.10 and works fine. The platform is a
general P4 with 100M ethernet. The user space system is a 8M compressed
ramdisk image which is a 32M filesystem.

Now I want to make it work on 2.6.12+ and on Athlon64 platform, for
better driver and better CPU/NIC performance.

I have a P4 box (compilation bed, CB), a 2-way Athlon64 box (test bed,
TB).

The problems are:

1. I port the code directly to 2.6.12.1 on CB, and it compiles ok. But
during boot, the kernel boot with error "unknown bus type 0" and freeze.
Especially, it can't detect harddisk's partition table. I use "quiet" to
strip non-error message and hand copy error messages

unknown bus type 0.
...
<repeat for several times>
...
unknown bus type 0.
irq 15: nobody cared!
handlers:
[<c0252927>]  (ide_intr+0x0/0x146)
Disabling IRQ #15
irq 1: nobody cared!
handlers:
[<c021a79a>]  (i8042_interrupt+0x0/0x256)
Disabling IRQ #1

But the same configuration works fine for 2.6.10.


2. I compile kernel 2.6.12.1 for K7 on CB. Boot it on TB, the system
boot up execept that the analysis code can't vmalloc() the needed memory.

"allocation failed: out of vmalloc space - use vmalloc=<size> to increase size."

If I use vmalloc=256m in boot command line, then

initrd extends beyond end of memory (0x37fef716 > 0x30000000)
initrd extends beyond end of memory (0x37fef716 > 0x30000000)
Kernel panic - unsyncing: VFS: Unable to mount root fs on unknown-block
(1,0)

BTW: Is there any way that a 512M memory chunk can be alloc? The system
has a lot of memory (1G+) and can't be used. It's better to provide a
mechanism to alloc big chunk of memory for special purpose (they will
never be freed and will be used as a single big chunk)


-- 
Wang Jian

Re: 2.6.12.1 problems I meet (please CC: me)

[Wang Jian]

Sorry,

I am not intended to do that. gmail's pop server has problem of not
sending response after final "." is sent. And my email client keeps trying
to send it out in thinking that sending failed.

Actually, I manually stopped it before you complain coz I was afraid it
sent duplicate copies, and check gmail account's outbox via web. It
seems that only one copy is sent out and I am relieved. But it is not
true, it did send 3 copies.

Sorry again.


On Wed, 29 Jun 2005 23:00:20 -0700
randy_dunlap <rdunlap@xenotime.net> wrote:

> On Thu, 30 Jun 2005 13:52:18 +0800 Wang Jian wrote:
> 
> | Hi,
> | 
> | I use a customized kernel to do packets analysis. The analysis code is
> | linked into kernel. It will vmalloc() nearly 128M (a little less) when
> | initialized.
> | 
> 
> OK, we have your message 3 times now, so please stop sending it.
> 
> ---
> ~Randy

-- 
Wang Jian

Re: 2.6.12.1 problems I meet (please CC: me)

[Jens Axboe]

On Thu, Jun 30 2005, Wang Jian wrote:
> Hi,
> 
> I use a customized kernel to do packets analysis. The analysis code is
> linked into kernel. It will vmalloc() nearly 128M (a little less) when
> initialized.
> 
> The original code runs on 2.6.10 and works fine. The platform is a
> general P4 with 100M ethernet. The user space system is a 8M compressed
> ramdisk image which is a 32M filesystem.
> 
> Now I want to make it work on 2.6.12+ and on Athlon64 platform, for
> better driver and better CPU/NIC performance.
> 
> I have a P4 box (compilation bed, CB), a 2-way Athlon64 box (test bed,
> TB).
> 
> The problems are:
> 
> 1. I port the code directly to 2.6.12.1 on CB, and it compiles ok. But
> during boot, the kernel boot with error "unknown bus type 0" and freeze.
> Especially, it can't detect harddisk's partition table. I use "quiet" to
> strip non-error message and hand copy error messages

Which compiler? 2.6.12.2 should work for you, looks like you are hit my
the memcpy reordering bug.

-- 
Jens Axboe

Re: 2.6.12.1 problems I meet (please CC: me)

[Wang Jian]

[root@devel root]# gcc --version
gcc (GCC) 3.3.2 (Mandrake Linux 10.0 3.3.2-6mdk)

I will try 2.6.12.2 and report back.

On 6/30/05, Jens Axboe <axboe@suse.de> wrote:
> On Thu, Jun 30 2005, Wang Jian wrote:
> > Hi,
> >
> > I use a customized kernel to do packets analysis. The analysis code is
> > linked into kernel. It will vmalloc() nearly 128M (a little less) when
> > initialized.
> >
> > The original code runs on 2.6.10 and works fine. The platform is a
> > general P4 with 100M ethernet. The user space system is a 8M compressed
> > ramdisk image which is a 32M filesystem.
> >
> > Now I want to make it work on 2.6.12+ and on Athlon64 platform, for
> > better driver and better CPU/NIC performance.
> >
> > I have a P4 box (compilation bed, CB), a 2-way Athlon64 box (test bed,
> > TB).
> >
> > The problems are:
> >
> > 1. I port the code directly to 2.6.12.1 on CB, and it compiles ok. But
> > during boot, the kernel boot with error "unknown bus type 0" and freeze.
> > Especially, it can't detect harddisk's partition table. I use "quiet" to
> > strip non-error message and hand copy error messages
> 
> Which compiler? 2.6.12.2 should work for you, looks like you are hit my
> the memcpy reordering bug.
> 
> --
> Jens Axboe
> 
>

Re: 2.6.12.1 problems I meet (please CC: me)

[Wang Jian]

The problem #1 is gone for 2.6.12.2. Thanks.

2005/6/30, Wang Jian <larkwang@gmail.com>:
> [root@devel root]# gcc --version
> gcc (GCC) 3.3.2 (Mandrake Linux 10.0 3.3.2-6mdk)
> 
> I will try 2.6.12.2 and report back.
> 
> On 6/30/05, Jens Axboe <axboe@suse.de> wrote:
> > On Thu, Jun 30 2005, Wang Jian wrote:
> > > Hi,
> > >
> > > I use a customized kernel to do packets analysis. The analysis code is
> > > linked into kernel. It will vmalloc() nearly 128M (a little less) when
> > > initialized.
> > >
> > > The original code runs on 2.6.10 and works fine. The platform is a
> > > general P4 with 100M ethernet. The user space system is a 8M compressed
> > > ramdisk image which is a 32M filesystem.
> > >
> > > Now I want to make it work on 2.6.12+ and on Athlon64 platform, for
> > > better driver and better CPU/NIC performance.
> > >
> > > I have a P4 box (compilation bed, CB), a 2-way Athlon64 box (test bed,
> > > TB).
> > >
> > > The problems are:
> > >
> > > 1. I port the code directly to 2.6.12.1 on CB, and it compiles ok. But
> > > during boot, the kernel boot with error "unknown bus type 0" and freeze.
> > > Especially, it can't detect harddisk's partition table. I use "quiet" to
> > > strip non-error message and hand copy error messages
> >
> > Which compiler? 2.6.12.2 should work for you, looks like you are hit my
> > the memcpy reordering bug.
> >
> > --
> > Jens Axboe
> >
> >
>

Re: 2.6.12.1 problems I meet (please CC: me)

[Wang Jian]

2005/6/30, Wang Jian <larkwang@gmail.com>:
> Hi,
> 
> I use a customized kernel to do packets analysis. The analysis code is
> linked into kernel. It will vmalloc() nearly 128M (a little less) when
> initialized.
> 
> The original code runs on 2.6.10 and works fine. The platform is a
> general P4 with 100M ethernet. The user space system is a 8M compressed
> ramdisk image which is a 32M filesystem.
> 
> Now I want to make it work on 2.6.12+ and on Athlon64 platform, for
> better driver and better CPU/NIC performance.
> 
> I have a P4 box (compilation bed, CB), a 2-way Athlon64 box (test bed,
> TB).
> 
> The problems are:
> 
> 2. I compile kernel 2.6.12.1 for K7 on CB. Boot it on TB, the system
> boot up execept that the analysis code can't vmalloc() the needed memory.
> 
> "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size."
> 
> If I use vmalloc=256m in boot command line, then
> 
> initrd extends beyond end of memory (0x37fef716 > 0x30000000)
> initrd extends beyond end of memory (0x37fef716 > 0x30000000)
> Kernel panic - unsyncing: VFS: Unable to mount root fs on unknown-block
> (1,0)
> 

This problem also presents itself in 2.6.10.

I remove 512M physic RAM from this TB, so it has 512M RAM left. The
problem is  then gone. The needed memory can be correctly vmalloc()
during boot without vmalloc=256m specified in boot command line.

I am curious why this happens?

1. With 512M physic RAM, vmalloc(16776989 * 8) succeeds.
2. With 1G physic RAM, vmalloc(16776989 * 8) fails.
3. With 1G physic RAM, vmalloc=256m boot option will cause kernel fail
to expand a 8M initrd image which is of 32M filesystem.