arm_vmregion_alloc call

arm_vmregion_alloc call

[Robert Beckett]

Hello all,

I just debugged a problem with a USB driver where it was calling
dma_alloc_coherent with a GFP_DMA32 flag.
It turned out that in __dma_alloc_remap, it clears the GFP_DMA and
GFP_HIGHMEM flags when calling arm_vmregion_alloc so that it does not clash
with the slab allocator flag requirements, but it does not clear GFP_DMA32.
The GFP_SLAB_BUG_MASK includes GFP_DMA32, so this looks like a bug to me.
Should it be clearing GFP_DMA32 as well as GFP_DMA and GFP_HIGHMEM ?

I checked in the latest kernel.org code, and the issue is still there.

Is this a bug, or should it just be calling with GFP_DMA instead of
GFP_DMA32?

For now, I am working around it by using GFP_DMA in the dma_alloc_coherent
call as we have no GFP_DMA lower limit, so GFP_DMA and GFP_DMA32 are the
same on my architecture.

Any advice appreciated,

Thanks

Bob
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20120702/2b683745/attachment-0001.html>

arm_vmregion_alloc call

[Robert Beckett]

Hello all,

I just debugged a problem with a USB driver where it was calling
dma_alloc_coherent with a GFP_DMA32 flag.
It turned out that in __dma_alloc_remap, it clears the GFP_DMA and
GFP_HIGHMEM flags when calling arm_vmregion_alloc so that it does not
clash with the slab allocator flag requirements, but it does not clear
GFP_DMA32. The GFP_SLAB_BUG_MASK includes GFP_DMA32, so this looks
like a bug to me. Should it be clearing GFP_DMA32 as well as GFP_DMA
and GFP_HIGHMEM ?

I checked in the latest kernel.org code, and the issue is still there.

Is this a bug, or should it just be calling with GFP_DMA instead of GFP_DMA32?

For now, I am working around it by using GFP_DMA in the
dma_alloc_coherent call as we have no GFP_DMA lower limit, so GFP_DMA
and GFP_DMA32 are the same on my architecture.

Any advice appreciated,

Thanks

Bob

p.s. apologies if this sent multiple times, Im trying to work around
the mailing list "suspicious header" moderation...

arm_vmregion_alloc call

[Robert Beckett]

Hello all,

I just debugged a problem with a USB driver where it was calling
dma_alloc_coherent with a GFP_DMA32 flag.
It turned out that in __dma_alloc_remap, it clears the GFP_DMA and
GFP_HIGHMEM flags when calling arm_vmregion_alloc so that it does not clash
with the slab allocator flag requirements, but it does not clear GFP_DMA32.
The GFP_SLAB_BUG_MASK includes GFP_DMA32, so this looks like a bug to me.
Should it be clearing GFP_DMA32 as well as GFP_DMA and GFP_HIGHMEM ?

I checked in the latest kernel.org code, and the issue is still there.

Is this a bug, or should it just be calling with GFP_DMA instead of
GFP_DMA32?

For now, I am working around it by using GFP_DMA in the dma_alloc_coherent
call as we have no GFP_DMA lower limit, so GFP_DMA and GFP_DMA32 are the
same on my architecture.

Any advice appreciated,

Thanks

Bob
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20120702/a710cc1c/attachment-0001.html>