Re: [PATCH] ARC: io.h: Implement reads{x}()/writes{x}()

[Jose Abreu <jose.abreu@synopsys.com>]

On 29-11-2018 13:03, David Laight wrote:
> From: Jose Abreu [mailto:jose.abreu@synopsys.com]
>> On 29-11-2018 12:47, David Laight wrote:
>>> From: Jose Abreu
>>>> Sent: 29 November 2018 12:42
>>>>
>>>> Some ARC CPU's do not support unaligned loads/stores. Currently, generic
>>>> implementation of reads{b/w/l}()/writes{b/w/l}() is being used with ARC.
>>>> This can lead to misfunction of some drivers as generic functions do a
>>>> plain dereference of a pointer that can be unaligned.
>>>>
>>>> Let's use {get/put}_unaligned() helper instead of plain dereference of
>>>> pointer in order to fix this.
>>> Is it worth adding a check for the pointer being aligned?
>> We could but then we would need to know which CPU version is
>> currently running because some ARC processors support unaligned
>> accesses.
> Eh?
> If the CPU supports unaligned accesses you could patch the code
> to do unaligned accesses.

*Some* ARC CPU versions support unaligned memory access. The one
we tested this on does not support unaligned accesses.

>
> I was thinking of the (probably likely) case where the pointer is
> actually aligned.
> An extra check for ((pointer) & 3) is almost certainly a 'win'
> over the byte accesses and shift/mask/or use by get/put_unaligned().
>
> The IO accesses probably dominate making more complex optimisations
> less likely to have any benefit.
>
> 	David
>
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)