Re: [2.6 patch] fs/block_dev.c:I_BDEV() mustn't be inline

[Adrian Bunk <bunk@kernel.org>]

On Tue, Apr 29, 2008 at 10:17:05AM +0200, Sam Ravnborg wrote:
> On Tue, Apr 29, 2008 at 11:06:31AM +0300, Adrian Bunk wrote:
> > On Mon, Apr 28, 2008 at 05:41:11PM -0700, Linus Torvalds wrote:
> > >...
> > > > -inline struct block_device *I_BDEV(struct inode *inode)
> > > > +struct block_device *I_BDEV(struct inode *inode)
> > > >  {
> > > >  	return &BDEV_I(inode)->bdev;
> > > >  }
> > > 
> > > There is NO WAY IN HELL this function shouldn't be inlined,
> > >...
> > 
> > Then CONFIG_OPTIMIZE_INLINING already broke it in your tree.
> 
> Can you please help me here.
> Previously we said: gcc shall inline
> Now we say: hi gcc - it is a good idea to inline this

Previosly we said:
"inline" means "must inline"

With CONFIG_OPTIMIZE_INLINING=y we say:
"inline" means "try a bit harder to inline"

> But even without any inline annotation we would expect gcc
> to inline it based on normal heuristics simply due to the
> small size/footprint of the function.

gcc anyway has to emit one global copy of this function.

And with the common CONFIG_CC_OPTIMIZE_FOR_SIZE=y we tell gcc to focus 
on getting the smallest possible code.

> So I cannot see how the CONFIG_OPTIMIZE_INLINING have any
> effect here.
> It all smells like a gcc bug to me.

The compile error is a gcc bug.

> 	Sam

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed