[RFT PATCH v3 11/27] arm64: Implement ioremap_np() to map MMIO as nGnRnE

Will Deacon will at kernel.org
Wed Mar 24 18:18:04 GMT 2021


On Fri, Mar 05, 2021 at 06:38:46AM +0900, Hector Martin wrote:
> This is used on Apple ARM platforms, which require most MMIO
> (except PCI devices) to be mapped as nGnRnE.
> 
> Signed-off-by: Hector Martin <marcan at marcan.st>
> ---
>  arch/arm64/include/asm/io.h | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/arm64/include/asm/io.h b/arch/arm64/include/asm/io.h
> index 5ea8656a2030..953b8703af60 100644
> --- a/arch/arm64/include/asm/io.h
> +++ b/arch/arm64/include/asm/io.h
> @@ -169,6 +169,7 @@ extern void __iomem *ioremap_cache(phys_addr_t phys_addr, size_t size);
>  
>  #define ioremap(addr, size)		__ioremap((addr), (size), __pgprot(PROT_DEVICE_nGnRE))
>  #define ioremap_wc(addr, size)		__ioremap((addr), (size), __pgprot(PROT_NORMAL_NC))
> +#define ioremap_np(addr, size)		__ioremap((addr), (size), __pgprot(PROT_DEVICE_nGnRnE))

Probably worth noting that whether or not this actually results in a
non-posted mapping depends on the system architecture, but this is the
best we can do, so:

Acked-by: Will Deacon <will at kernel.org>

I would personally prefer that drivers didn't have to care about this, and
ioremap on arm64 figured out the right attributes based on the region being
mapped, but I haven't followed the discussion closely so I won't die on that
hill.

Will



More information about the linux-arm-kernel mailing list