Re: Non-e1000, 2.6.9 page allocation failures with OSS/audio.

[Nick Piggin <nickpiggin@yahoo.com.au>]

Justin Piszcz wrote:
> The other page allocation failures were due to the e1000+TSO 
> segmentation offload issue.
> 
> I use OSS sound drivers with 2.6.9.
> 
> Here are the options I use and have been using for quite some time 
> without error:
> 
> 1] <*> Open Sound System (DEPRECATED)
> 2] <*>   OSS sound modules
> 3] [*]     Verbose initialisation
> 4] [*]     Persistent DMA buffers
> 5] <*>     Crystal CS4232 based (PnP) cards
> 
> My LILO append line as is follows:
> append="cs4232=0x530,5,1,0,388,5 mce video=atyfb:1600x1200-16@60"
> 
> What happened to the page allocation / memory subsystem in 2.6.9?
> I do not recall getting these with 2.6.4, 2.6.5, 2.6.6, 2.6.7, 2.6.8 or 
> 2.6.8.1.
> 

These failures are actually harmless and not entirely unexpected.
That path should tell the allocator not to produce warnings on
failure.

The issue is being actively worked on in the -mm kernels, and the
important stuff should get into 2.6.10.

However, earlier kernels had a bug that prevented a large amount
of ZONE_DMA memory from being allocated by GFP_KERNEL allocations
which is probably why this particular path hasn't been a problem
before. This actually won't be completely reverted in 2.6.10, however
the correct way to get this behaviour is with "lower zone protection"
which Andrea had been working on... so things may happen soon.