11:20 p.m.
On Mon, May 18, 2020 at 2:57 PM Arnd Bergmann <arnd(a)arndb.de> wrote:
On Mon, May 18, 2020 at 1:43 PM Andy Shevchenko
<andy.shevchenko(a)gmail.com> wrote:
> On Mon, May 18, 2020 at 2:39 PM Ramuthevar, Vadivel MuruganX
> <vadivel.muruganx.ramuthevar(a)linux.intel.com> wrote:
> > On 15/5/2020 10:30 pm, Arnd Bergmann wrote:
> > > On Fri, May 15, 2020 at 4:25 PM Andy Shevchenko
> > > <andy.shevchenko(a)gmail.com> wrote:
> > >> On Fri, May 15, 2020 at 4:48 PM kbuild test robot
<lkp(a)intel.com> wrote:
>
> > > iowrite_be32() is the correct way to store word into a big-endian mmio
register,
> > > if that is the intention here.
> > Thank you for suggestions to use iowrite32be(), it suits exactly.
>
> Can you before doing this comment what is the real intention here?
>
> And note, if you are going to use iowrite*() / ioread*() in one place,
> you will probably need to replace all of the read*() / write*() to
> respective io* API.
The way that ioread/iowrite are defined, they are required to be a superset
of what readl/writel do and can take __iomem pointers from either
ioremap() or ioport_map()/pci_iomap() style mappings, while readl/writel
are only required to work with ioremap().
There is no technical requirement to stick to one set or the other for
ioremap(), but the overhead of ioread/iowrite is also small enough
that it generally does not hurt.
Right, my suggestion is solely for consistency. It would be a bit
weird to see readl() along with ioread32() in the same driver (in case
there are no differentiated callbacks specifically for different type
of IP).
--
With Best Regards,
Andy Shevchenko