Re: [Xen-devel] [PATCH] arm: xen: mm: use __GPF_DMA32 for arm64

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Wed, 28 Aug 2019, Peng Fan wrote:
> Hi Robin,
> 
> > Subject: Re: [PATCH] arm: xen: mm: use __GPF_DMA32 for arm64
> > 
> > On 09/07/2019 09:22, Peng Fan wrote:
> > > arm64 shares some code under arch/arm/xen, including mm.c.
> > > However ZONE_DMA is removed by commit
> > > ad67f5a6545("arm64: replace ZONE_DMA with ZONE_DMA32").
> > > So to ARM64, need use __GFP_DMA32.

Hi Peng,

Sorry for being so late in replying, this email got lost in the noise.


> > > Signed-off-by: Peng Fan <peng.fan@xxxxxxx>
> > > ---
> > >   arch/arm/xen/mm.c | 2 +-
> > >   1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/arch/arm/xen/mm.c b/arch/arm/xen/mm.c index
> > > e1d44b903dfc..a95e76d18bf9 100644
> > > --- a/arch/arm/xen/mm.c
> > > +++ b/arch/arm/xen/mm.c
> > > @@ -27,7 +27,7 @@ unsigned long xen_get_swiotlb_free_pages(unsigned
> > > int order)
> > >
> > >           for_each_memblock(memory, reg) {
> > >                   if (reg->base < (phys_addr_t)0xffffffff) {
> > > -                 flags |= __GFP_DMA;
> > > +                 flags |= __GFP_DMA | __GFP_DMA32;
> > 
> > Given the definition of GFP_ZONE_BAD, I'm not sure this combination of flags
> > is strictly valid, but rather is implicitly reliant on only one of those 
> > zones ever
> > actually existing. As such, it seems liable to blow up if the plans to add
> > ZONE_DMA to arm64[1] go ahead.
> 
> How about this, or do you have any suggestions?
> diff --git a/arch/arm/xen/mm.c b/arch/arm/xen/mm.c
> index d33b77e9add3..f61c29a4430f 100644
> --- a/arch/arm/xen/mm.c
> +++ b/arch/arm/xen/mm.c
> @@ -28,7 +28,11 @@ unsigned long xen_get_swiotlb_free_pages(unsigned int 
> order)
> 
>         for_each_memblock(memory, reg) {
>                 if (reg->base < (phys_addr_t)0xffffffff) {
> +#ifdef CONFIG_ARM64
> +                       flags |= __GFP_DMA32;
> +#else
>                         flags |= __GFP_DMA;
> +#endif
>                         break;
>                 }
>         }

Yes I think this is the way to go, but we are trying not to add any
#ifdef CONFIG_ARM64 under arch/arm. Maybe you could introduce a static
inline function to set GFP_DMA:

  static inline void xen_set_gfp_dma(gfp_t *flags)

it could be implemented under arch/arm/include/asm/xen/page.h for arm
and under arch/arm64/include/asm/xen/page.h for arm64 using __GFP_DMA
for the former and __GFP_DMA32 for the latter.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel