Re: 2.6.17-rc1-mm1

[Roger Luethi <rl@hellgate.ch>]

On Wed, 05 Apr 2006 00:29:09 -0700, Andrew Morton wrote:
> Roger Luethi <rl@hellgate.ch> wrote:
> >
> > Any suggestions for an elegant solution?
> 
> Move the locking down lower, so it just locks the stuff which needs locking?
> 
> It all depends on what the lock's role is, and so often that's a big secret.

Indeed.

Over a dozen drivers now use calls like mii_ethtool_gset or
generic_mii_ioctl, and only one (sis190) makes the call without grabbing
the lock to the driver private data.

Pushing that lock down is not something I'd do lightly.

> If the intention is to prevent concurrent execution of mdio_read()
> (reasonable) and we really need that 1 msec delay between writing the

I'm afraid that delay is not negotiable -- unless we want to go back to
polling the chip for link state changes.

Roger