Re: [RFC/PATCH] map drivers. DCACHE option for physmap and mphysmap drivers

["Alexey, Korolev" <alexey.korolev@intel.com>]

Josh,

Thanks a lot for your feedback.


Josh Boyer wrote:

> On 1/26/06, Korolev, Alexey <alexey.korolev@intel.com> wrote:
> > Hi All,
> >
> > Some mapping drivers of linux-mtd has feature of cached mapping. This
> > can improve read performance on FLASH significantly .
> > I think adding the option of data cache mapping would also be helpful
> > for physmap and mphysmap drivers.
>
> Is this a read-only caching?  I know several architectures and
> hardware configurations that would quickly throw a machine check if
> trying to flush the cache back to flash since they don't support
> bursted writes.
>
In general case not.  I've never heard about the such issue. Is there 
any way to define has platform flash devices with bursted writes or not?
What architectures have this problem?

> I would recommend adding a dependency on ARM for all of this for now.
> As other architectures are proven to be working, they can be added.
>
Oh. It's a good point. I'll add ARM dependency to Kconfig file. Thanks.

> > +#ifdef CONFIG_MTD_MULTI_PHYSMAP_DCACHE
> > +        if (map->inval_cache)
> > +        {
> > +#ifdef CONFIG_ARM
> > +           map->cached = ioremap_cached(map->phys, map->size);
> > +#else
> > +           map->cached = __ioremap(map->phys, map->size,
> > L_PTE_CACHEABLE);
> > +#endif
>
> Gah.  The above doesn't work.  L_PTE_CACHEABLE seems to be an ARM only
> flag.  And it wouldn't work on PPC for example.  That version of
> __ioremap makes everything implicitly cached unless _PAGE_NO_CACHED is
> passed in.
>
> As I said before, I would make this an ARM only option for now.
>
> > +#ifdef CONFIG_MTD_PHYSMAP_DCACHE
> > +#ifdef CONFIG_ARM
> > + physmap_map.cached = ioremap_cached(physmap_map.phys,
> > physmap_map.size);
> > +#else
> > + physmap_map.cached = __ioremap(physmap_map.phys, physmap_map.size,
> > L_PTE_CACHEABLE);
> > +#endif
>
> Same issues as in mphysmap.
>
> josh
>
Thanks,
Alexey