[BISECTED REGRESSION] b43legacy broken on G4 PowerBook

[Larry Finger]

On 6/6/19 6:43 AM, Christoph Hellwig wrote:
> On Thu, Jun 06, 2019 at 08:57:49PM +1000, Benjamin Herrenschmidt wrote:
>>> Wow... that's an odd amount. One thing we could possibly do is add code
>>> to limit the amount of RAM when we detect that device....
>>
>> Sent too quickly... I mean that *or* force swiotlb at 30-bits on those systems based
>> on detecting the presence of that device in the device-tree.
> 
> swiotlb doesn't really help you, as these days swiotlb on matters for
> the dma_map* case.  What would help is a ZONE_DMA that covers these
> devices.  No need to do the 24-bit x86 does, but 30-bit would do it.
> 
> WIP patch for testing below:
> 
> diff --git a/arch/powerpc/include/asm/page.h b/arch/powerpc/include/asm/page.h
> index b8286a2013b4..7a367ce87c41 100644
> --- a/arch/powerpc/include/asm/page.h
> +++ b/arch/powerpc/include/asm/page.h
> @@ -319,6 +319,10 @@ struct vm_area_struct;
>   #endif /* __ASSEMBLY__ */
>   #include <asm/slice.h>
>   
> +#if 1 /* XXX: pmac?  dynamic discovery? */
> +#define ARCH_ZONE_DMA_BITS 30
> +#else
>   #define ARCH_ZONE_DMA_BITS 31
> +#endif
>   
>   #endif /* _ASM_POWERPC_PAGE_H */
> diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
> index cba29131bccc..2540d3b2588c 100644
> --- a/arch/powerpc/mm/mem.c
> +++ b/arch/powerpc/mm/mem.c
> @@ -248,7 +248,8 @@ void __init paging_init(void)
>   	       (long int)((top_of_ram - total_ram) >> 20));
>   
>   #ifdef CONFIG_ZONE_DMA
> -	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn, 0x7fffffffUL >> PAGE_SHIFT);
> +	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
> +			((1UL << ARCH_ZONE_DMA_BITS) - 1) >> PAGE_SHIFT);
>   #endif
>   	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
>   #ifdef CONFIG_HIGHMEM
> 

I am generating a test kernel with this patch.

FYI, the "free" command on my machine shows 1.5+ G of memory. That likely means 
I have 2G installed.

I have tested a patched kernel in which b43legacy falls back to a 31-bit DMA 
mask when the 32-bit one failed. That worked, but would likely kill the x86 
version. Let me know if think a fix in the driver rather than the kernel would 
be better. I still need to understand why the same setup works in b43 and fails 
in b43legacy. :(

Larry