[openib-general] Re: page allocation failure with kdapl/mthca

[Grant Grundler]

On Thu, Jun 23, 2005 at 08:51:46AM -0700, Roland Dreier wrote:
>     Grant> On ia64 I expect both functions to land in the same bit of
>     Grant> code.  For ZX1 platforms, that is sba_alloc_coherent().
>     Grant> See arch/ia64/hp/common/sba_iommu.c since the behavior is
>     Grant> different depending on CONFIG_NUMA setting.
> 
> Sure, but follow the code all the way through.  ia64 uses
> <asm-generic/pci-dma-compat.h>, which defines pci_alloc_consistent()
> to be dma_alloc_coherent(..., GFP_ATOMIC).

Ah ok - that's the bit I skipped over last night.

> You're right that both
> will end up in sba_alloc_coherent(), but the difference is the flags
> parameter.  GFP_ATOMIC allocations can easily fail even when the
> system has plenty of pages available for GFP_KERNEL allocations.

Ah! I had forgotten dma_alloc_coherent() takes a GFP_* flag parameter.
nevermind...

thanks,
grant