Re: Memory management woes - order 1 allocation failures

[Greg KH <gregkh@suse.de>]

On Tue, Mar 02, 2010 at 07:11:10PM +0000, Mel Gorman wrote:
> On Tue, Mar 02, 2010 at 06:34:51PM +0000, Alan Cox wrote:
> > > For reasons that are not particularly clear to me, tty_buffer_alloc() is
> > > called far more frequently in 2.6.33 than in 2.6.24. I instrumented the
> > > function to print out the size of the buffers allocated, booted under
> > > qemu and would just "cat /bin/ls" to see what buffers were allocated.
> > > 2.6.33 allocates loads, including high-order allocations. 2.6.24
> > > appeared to allocate once and keep silent.
> > 
> > The pty layer is using them now and didn't before. That will massively
> > distort your numhers.
> > 
> 
> That makes perfect sense. It explains why only one allocation showed up
> because it must belong to the tty attached to the serial console.
> 
> Thanks Alan.
> 
> > > While there have been snags recently with respect to high-order
> > > allocation failures in recent kernels, this might be one of the cases
> > > where it's due to subsystems requesting high-order allocations more.
> > 
> > The pty code certainly triggered more such allocations. I've sent Greg
> > patches to make the tty buffering layer allocate sensible sizes as it
> > doesn't need multiple page allocations in the first place.
> > 
> 
> Greg, what's the story with these patches?

They are in -next and will go to Linus later on today for .34.

thanks,

greg k-h