Re: [PATCH 1/1] regmap: regcache-rbtree: Use GFP_ATOMIC when using spinlocks

[Mark Brown <broonie@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 1666 bytes --]

On Tue, Jul 21, 2015 at 08:14:32AM +0200, Alexander Stein wrote:
> On Monday 20 July 2015 18:26:27, Mark Brown wrote:

> > Well, it's *better* to provide defaults since otherwise everything
> > defaults to 0 but it does avoid the whole allocation during fast path
> > issue since it allocates the cache on init and perhaps that's OK.

> There is another reason for using REGCACHE_FLAT: Using regcache_cache_only 
> (e.g. during suspend) which is not possible with REGCACHE_NONE.

Sure, if you want a cache you need to use a cache.

> > > So how to handle this properly? Bail out, if fast_io is available and
> > > cache_type != (REGCACHE_NONE || REGCACHE_FLAT)?

> > Or perhaps just if we have to do an allocation?  I can see that someone
> > might want to use an rbtree and would be careful enough to do the init,
> > though I *am* a bit dubious about it.

> I'm feeling uncomfortable this warning occured only when (at least) 
> CONFIG_LOCKDEP is enabled. It warns right ahead but only if you begged for 
> it...

OTOH it'll splat with lockdep enabled just as soon as someone tries to
access a register that they didn't provide a default for.  An explicitly
added error and bomb out like I suggested above would do the same thing.

> Even if defaults are provided an extension to the register set (e.g. a more 
> recent IP revision with more features) might not be synchronized with the 
> defaults. Nobody might noticed until CONFIG_LOCKDEP is enabled and the 
> register without defaults gets written.

Sure, it's got sharp edges but the whole point with my suggestion above
is that detection wouldn't depend on CONFIG_LOCKDEP.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]