2.4.5-ac1 won't boot with 4GB bigmem option

2.4.5-ac1 won't boot with 4GB bigmem option

[Ben Twijnstra]

Hi,

I compiled and booted the 2.4.5-ac1 kernel with the CONFIG_HIGHMEM4G=y option 
and got an oops in __alloc_pages() (called by alloc_bounce() called by 
schedule()). Everything works fine if I turn the 4GB mode off.

Machine is a Dell Precision with 2 Xeons and 2GB of RAM.

2.4.5 works fine with the 4GB. Any idea what changed between the two?

Grtz,



Ben

Re: 2.4.5-ac1 won't boot with 4GB bigmem option

[Fabio Riccardi]

Same here, I have a dual 1GHz PIII with 4G, I don't get an oops but an infinite
loop of:

 > mm: critical shortage of bounce buffers.

Indeed this message has been pestering me in all the recent .4-acx kernels when
the machine is under heavy FS pressure.

In these kernels I observe a significative (5-10%) performance degradation as
soon as the FS cache fills up all the available memory, at this moment "kswapd"
starts to take lots of CPU time (10-20%) and I keep getting plenty of the above
messages.

I'm running SpecWeb with the X15 webserver, which uses sendfile to send its
content, and a very large file set (8-9G, more than twice as much as the
physical RAM).

2.4.2-acx and early 2.4.3-acx kernles were much better in this respect and a lot
more stable.

 - Fabio

Ben Twijnstra wrote:

> Hi,
>
> I compiled and booted the 2.4.5-ac1 kernel with the CONFIG_HIGHMEM4G=y option
> and got an oops in __alloc_pages() (called by alloc_bounce() called by
> schedule()). Everything works fine if I turn the 4GB mode off.
>
> Machine is a Dell Precision with 2 Xeons and 2GB of RAM.
>
> 2.4.5 works fine with the 4GB. Any idea what changed between the two?
>
> Grtz,
>
> Ben
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: 2.4.5-ac1 won't boot with 4GB bigmem option

[Alan Cox]

>  > mm: critical shortage of bounce buffers.
> 
> Indeed this message has been pestering me in all the recent .4-acx kernels when
> the machine is under heavy FS pressure.
> 
> In these kernels I observe a significative (5-10%) performance degradation as
> soon as the FS cache fills up all the available memory, at this moment "kswapd"

Its there to prove we had a problem

> 2.4.2-acx and early 2.4.3-acx kernles were much better in this respect and a lot
> more stable.

Hit any 2.4 kernel pre 2.4.5 vanilla [maybe fixed] and you will break bigmem
that way.

Re: 2.4.5-ac1 won't boot with 4GB bigmem option

[Fabio Riccardi]

Alan Cox wrote:

> >  > mm: critical shortage of bounce buffers.
> >
> > Indeed this message has been pestering me in all the recent .4-acx kernels when
> > the machine is under heavy FS pressure.
> >
> > In these kernels I observe a significative (5-10%) performance degradation as
> > soon as the FS cache fills up all the available memory, at this moment "kswapd"
>
> Its there to prove we had a problem

granted

> > 2.4.2-acx and early 2.4.3-acx kernles were much better in this respect and a lot
> > more stable.
>
> Hit any 2.4 kernel pre 2.4.5 vanilla [maybe fixed] and you will break bigmem
> that way.

I've been using many kernels (I upgrade every week or so) and as far as I can
recollect I started experiencing serious performance problems with the 2.4.4 series.
I'm double checking my data right now with older kernels.

 - Fabio

Re: 2.4.5-ac1 won't boot with 4GB bigmem option

[Alan Cox]

> I compiled and booted the 2.4.5-ac1 kernel with the CONFIG_HIGHMEM4G=y option 
> and got an oops in __alloc_pages() (called by alloc_bounce() called by 
> schedule()). Everything works fine if I turn the 4GB mode off.
> Machine is a Dell Precision with 2 Xeons and 2GB of RAM.
> 
> 2.4.5 works fine with the 4GB. Any idea what changed between the two?

The -ac tree has some VM differences for bigmem that really need to be
cleaned up and/or removed now the Linus tree is looking solid. I'll probably
drop those diffs for -ac2 so that folks are working against one set of VM

Re: 2.4.5-ac1 won't boot with 4GB bigmem option

[Ben Twijnstra]

Alan,

2.4.5-ac2 works fine with 4GB on. Thanks! Will try a 2.4.5-aa2 later 
today, just for fun.

Grtz,


Ben

On 27 May 2001, at 22:21, Alan Cox wrote:

> > I compiled and booted the 2.4.5-ac1 kernel with the CONFIG_HIGHMEM4G=y option 
> > and got an oops in __alloc_pages() (called by alloc_bounce() called by 
> > schedule()). Everything works fine if I turn the 4GB mode off.
> > Machine is a Dell Precision with 2 Xeons and 2GB of RAM.
> > 
> > 2.4.5 works fine with the 4GB. Any idea what changed between the two?
> 
> The -ac tree has some VM differences for bigmem that really need to be
> cleaned up and/or removed now the Linus tree is looking solid. I'll probably
> drop those diffs for -ac2 so that folks are working against one set of VM
>