On Thu, 2019-10-31 at 16:57 +0100, Christoph Hellwig wrote:
> On Thu, Oct 31, 2019 at 04:53:13PM +0100, Nicolas Saenz Julienne wrote:
> > > > +#define ARM64_ZONE_DMA_BITS	30
> > > > +
> > > >  /*
> > > >   * We need to be able to catch inadvertent references to memstart_addr
> > > >   * that occur (potentially in generic code) before
> > > > arm64_memblock_init()
> > > > @@ -424,6 +427,8 @@ void __init arm64_memblock_init(void)
> > > >  	else
> > > >  		arm64_dma_phys_limit = PHYS_MASK + 1;
> > > >  
> > > > +	zone_dma_bits = ARM64_ZONE_DMA_BITS;
> > > > +
> > > >  	reserve_crashkernel();
> > > 
> > > This actually adds a new limit, as there wasn't one before for arm64.
> > 
> > Well, as zone_dma_bits is only relevant in dma/direct when ZONE_DMA is
> > defined
> > I figured it doesn't matter if the variable is set conditionally to ZONE_DMA
> > or
> > not.
> 
> I'd much prefer that to do separately.

OK, I see what you mean now. It's wrong indeed.

The trouble is the ZONE_DMA series[1] in arm64, also due for v5.5, will be
affected by this patch. I don't know the right way to approach this problem
since depending on the merge order, this patch should be updated or the arm64
ZONE_DMA series fixed.

Maybe it's easier to just wait for v5.6.

Regards,
Nicolas

[1] https://lkml.org/lkml/2019/9/11/734