Re: [PATCH 02/10] sdhci: Add support for bus-specific IO memory accessors

[Anton Vorontsov <avorontsov@ru.mvista.com>]

On Thu, Jan 22, 2009 at 12:55:48PM +0100, Arnd Bergmann wrote:
> On Thursday 22 January 2009, Anton Vorontsov wrote:
> > +       /*
> > +        * These accessors duplicate sdhci_ops, but there are two reasons for
> > +        * this:
> > +        * 1. sdhci_ops are const, so the sdhci driver won't able to assign
> > +        *    default ops;
> 
> You could assign the pointer to a const default_sdhci_ops structure,
> which IMHO would be cleaner than copying the function pointers
> separately.

This won't work because ops also specify non-memory accessors ops
(i.e. enable_dma, set_clock, ...), so we can't just assign
default_ops ptr.

> > +        * 2. Using host->X instead of host->ops->X saves us one dereference.
> > +        *    This can be useful in PIO mode. (Though the benefit of this
> > +        *    is negligibly small).
> > +        */
> 
> I doubt that this is even measurable.

Yeah, I didn't even bother w/ measuring. ;-)

> If it was, you could still use a copy of that structure, like
> 
> struct sdhci_host {
> 	...
> 	struct sdhci_ops ops; /* not struct sdhci_ops *ops */
> 	...
> };
> 
> and do an assignment of the structure, like
> 
> static void assign_ops(struct sdhci_host *host, struct sdhci_ops *ops)
> {
> 	host->ops = *ops;
> }

This will work. Pierre, what do you think? That way drivers
will call this routine instead of "host->ops = driver_ops_ptr".

The other option would be to remove const from the ops. Or implement
another non-const ops struct (struct sdhci_io_ops?).

In any way, the type-checked variant will be even longer to type:

host->mem_ops->writel(val, host->ioaddr + reg);
host->ops.writel(val, host->ioaddr + reg);
host->writel(val, host->ioaddr + reg);


Thanks,

-- 
Anton Vorontsov
email: cbouatmailru@gmail.com
irc://irc.freenode.net/bd2