Re: [PATCH] ahci_platform: Provide for vendor specific init

[Anton Vorontsov <cbouatmailru@gmail.com>]

On Fri, Jun 25, 2010 at 03:26:58AM -0400, Jeff Garzik wrote:
> On 06/25/2010 12:57 AM, Jassi Brar wrote:
> >From: Jassi Brar<jassi.brar@samsung.com>
> >
> >Some AHCI implementations may use Vendor Specific HBA[A0h, FFh]
> >and/or Port[70h, 7Fh] registers to 'prepare' for initialization.
> >For that, the platform needs memory mapped address of AHCI registers.
> >
> >This patch adds the 'mmio' argument and reorders the call to
> >platform init function.
> >
> >Signed-off-by: Jassi Brar<jassi.brar@samsung.com>
> >---
> >  drivers/ata/ahci_platform.c   |   23 +++++++++++++----------
> >  include/linux/ahci_platform.h |    2 +-
> >  2 files changed, 14 insertions(+), 11 deletions(-)
> 
> It is also possible that platforms may need the init to prep the
> MMIO area for use?  Anton, does this patch work for you, or
> introduce such a problem?

Currently it doesn't introduce such a problem, but you're right,
platforms might need to, for example, turn on the clocks before
the MMIO area becomes accessible (otherwise machine check
exception might arrive).

There are two options: introduce a ->post_init() hook, or just
make sure that ahci_platform driver won't start accessing the
mmio region between ioremap() and pdata->init() calls. I think
a simple comment in the code will suffice, so no need for the
additional hook.

And if we ever need some hook before ioremap(), we can name it
pre_init().

Thanks,

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