On Fri, 13 Jul 2007, James Bottomley wrote:
> On Fri, 2007-07-13 at 17:10 +0200, Geert Uytterhoeven wrote:
> > On Fri, 13 Jul 2007, Arnd Bergmann wrote:
> > > On Friday 13 July 2007, James Bottomley wrote:
> > > > > IC.
> > > > >
> > > > > - flush_kernel_dcache_page() is a no-op on ppc64
> > > > > (ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE is defined on parisc only).
> > > > >
> > > > > - For reference, drivers/scsi/ipr.c (another ppc64 driver) just uses a plain
> > > > > kmap/memcpy/kunmap sequence
> > > > >
> > > > > So what should I do?
> > > >
> > > > Ask someone who knows the architecture ... Anton, Paulus or Benh ... I'm
> > > > fairly certain PPC is VIPT and will need some kind of alias
> > > > resolution ... perhaps its associative enough not to let the aliases be
> > > > a problem.
> > >
> > > I'm pretty sure that no ppc64 machine needs alias resolution in the kernel,
> > > although some are VIPT. Last time we discussed this, Segher explained it
> > > to me, but I don't remember which way Cell does it. IIRC, it automatically
> > > flushes cache lines that are accessed through aliases.
> >
> > Thanks for confirming!
> >
> > > It's probably a good idea to have the flush_kernel_dcache_page() in there
> > > anyway, if only to serve as an example for people that copy it into
> > > architecture-independent drivers, same as what we do for the
> > > k{,un}map_atomic() that is also not required on ppc64.
> >
> > Now my next question: why should I add it, if currently no single driver in
> > mainline calls flush_kernel_dcache_page()?
> >
> > `git grep' finds it in the following files only:
> > Documentation/cachetlb.txt
> > arch/parisc/kernel/cache.c
> > arch/parisc/kernel/pacache.S
> > include/asm-parisc/cacheflush.h
> > include/linux/highmem.h
>
> It's a recent addition to the API ... the previous way of doing it was
> flush_dcache_page() but that's expensive and flushes all the user
> aliases. Since you know in this case there are no current user aliases
> and you only need to flush the kernel alias, flush_kernel_dcache_page()
> was introduced as the cheaper alternative.
Ah, that explains it. flush_dcache_page() is used in some drivers.
I'll update my patches. Thanks for the comments!
With kind regards,
Geert Uytterhoeven
Software Architect
Sony Network and Software Technology Center Europe
The Corporate Village · Da Vincilaan 7-D1 · B-1935 Zaventem · Belgium
Phone: +32 (0)2 700 8453
Fax: +32 (0)2 700 8622
E-mail: Geert.Uytterhoeven@xxxxxxxxxxx
Internet: http://www.sony-europe.com/
Sony Network and Software Technology Center Europe
A division of Sony Service Centre (Europe) N.V.
Registered office: Technologielaan 7 · B-1840 Londerzeel · Belgium
VAT BE 0413.825.160 · RPR Brussels
Fortis Bank Zaventem · Swift GEBABEBB08A · IBAN BE39001382358619
