Hi Catalin, sorry for the late reply.

On Fri, 2020-10-02 at 12:55 +0100, Catalin Marinas wrote:
> On Thu, Oct 01, 2020 at 07:31:19PM +0200, Nicolas Saenz Julienne wrote:
> > On Thu, 2020-10-01 at 18:23 +0100, Catalin Marinas wrote:
> > > On Thu, Oct 01, 2020 at 06:15:01PM +0100, Catalin Marinas wrote:
> > > > On Thu, Oct 01, 2020 at 06:17:37PM +0200, Nicolas Saenz Julienne wrote:
> > > > > diff --git a/drivers/of/fdt.c b/drivers/of/fdt.c
> > > > > index 4602e467ca8b..cd0d115ef329 100644
> > > > > --- a/drivers/of/fdt.c
> > > > > +++ b/drivers/of/fdt.c
> > > > > @@ -25,6 +25,7 @@
> > > > >  #include <linux/serial_core.h>
> > > > >  #include <linux/sysfs.h>
> > > > >  #include <linux/random.h>
> > > > > +#include <linux/dma-direct.h>	/* for zone_dma_bits */
> > > > >  
> > > > >  #include <asm/setup.h>  /* for COMMAND_LINE_SIZE */
> > > > >  #include <asm/page.h>
> > > > > @@ -1198,6 +1199,14 @@ void __init early_init_dt_scan_nodes(void)
> > > > >  	of_scan_flat_dt(early_init_dt_scan_memory, NULL);
> > > > >  }
> > > > >  
> > > > > +void __init early_init_dt_update_zone_dma_bits(void)
> > > > > +{
> > > > > +	unsigned long dt_root = of_get_flat_dt_root();
> > > > > +
> > > > > +	if (of_flat_dt_is_compatible(dt_root, "brcm,bcm2711"))
> > > > > +		zone_dma_bits = 30;
> > > > > +}
> > > > 
> > > > I think we could keep this entirely in the arm64 setup_machine_fdt() and
> > > > not pollute the core code with RPi4-specific code.
> > > 
> > > Actually, even better, could we not move the check to
> > > arm64_memblock_init() when we initialise zone_dma_bits?
> > 
> > I did it this way as I vaguely remembered Rob saying he wanted to centralise
> > all early boot fdt code in one place. But I'll be happy to move it there.
> 
> I can see Rob replied and I'm fine if that's his preference. However,
> what I don't particularly like is that in the arm64 code, if
> zone_dma_bits == 24, we set it to 32 assuming that it wasn't touched by
> the early_init_dt_update_zone_dma_bits(). What if at some point we'll
> get a platform that actually needs 24 here (I truly hope not, but just
> the principle of relying on magic values)?
> 
> So rather than guessing, I'd prefer if the arch code can override
> ZONE_DMA_BITS_DEFAULT. Then, in arm64, we'll just set it to 32 and no
> need to explicitly touch the zone_dma_bits variable.

Yes, sonds like the way to go. TBH I wasn't happy with that solution either,
but couldn't think of a nicer alternative.

Sadly I just realised that the series is incomplete, we have RPi4 users that
want to boot unsing ACPI, and this series would break things for them. I'll
have a word with them to see what we can do for their use-case.

Regards,
Nicolas