[PATCH] arm64: Revert L1_CACHE_SHIFT back to 6 (64-byte cache line size)

[Catalin Marinas]

On Thu, Feb 22, 2018 at 05:51:15PM +0000, Robin Murphy wrote:
> On 22/02/18 16:06, Catalin Marinas wrote:
> [...]
> > diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
> > index a96ec0181818..36deb34dd956 100644
> > --- a/arch/arm64/mm/dma-mapping.c
> > +++ b/arch/arm64/mm/dma-mapping.c
> > @@ -33,6 +33,7 @@
> >   #include <asm/cacheflush.h>
> >   static int swiotlb __ro_after_init;
> > +DEFINE_STATIC_KEY_FALSE(swiotlb_noncoherent_bounce);
> >   static pgprot_t __get_dma_pgprot(unsigned long attrs, pgprot_t prot,
> >   				 bool coherent)
> > @@ -882,6 +883,20 @@ static void __iommu_setup_dma_ops(struct device *dev, u64 dma_base, u64 size,
> >   void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size,
> >   			const struct iommu_ops *iommu, bool coherent)
> >   {
> > +	/*
> > +	 * Enable swiotlb for buffer bouncing if ARCH_DMA_MINALIGN < CWG.
> > +	 * dma_capable() forces the actual bounce if the device is
> > +	 * non-coherent.
> > +	 */
> > +	if (WARN_TAINT_ONCE(!coherent && ARCH_DMA_MINALIGN < cache_line_size(),
> > +			    TAINT_CPU_OUT_OF_SPEC,
> > +			    "ARCH_DMA_MINALIGN smaller than CTR_EL0.CWG (%d < %d)",
> > +			    ARCH_DMA_MINALIGN, cache_line_size())) {
> > +		static_branch_enable(&swiotlb_noncoherent_bounce);
> > +		swiotlb = 1;
> 
> I think it's possible (if a little contrived) for the first non-coherent
> device to only appear after loading a module from userspace, wherein this
> would fall foul of __ro_after_init. That said, it would be nice to keep the
> logic in arm64_dma_init() consistent with its mirror in mem_init() (i.e. add
> the GWC check there as below), at which point the aforementioned problem no
> longer applies anyway.

That's a good point actually. As discovered, with DT we will get at
least one call to arch_setup_dma_ops() with coherent == false even when
the device does not intend to do any DMA at all. So moving it to
arm64_dma_init() wouldn't make any functional difference (but fixing
__ro_after_init).

We still force bouncing according to whether the device is coherent or
not.

-- 
Catalin