On Thu, 2003-11-06 at 14:39, Andreas Dilger wrote:
> On Nov 06, 2003 13:58 -0500, Martin Hicks wrote:
> > +static int ext3_setup_sbi (struct ext3_sb_info *sbi)
> > +{
> > + sbi->s_freeblocks_counter = kmalloc(sizeof(struct percpu_counter), GFP_KERNEL);
> > + if (!sbi->s_freeblocks_counter)
> > + return -ENOMEM;
> > + sbi->s_freeinodes_counter = kmalloc(sizeof(struct percpu_counter), GFP_KERNEL);
> > + if (!sbi->s_freeinodes_counter)
> > + goto out_freeblocks;
> > + sbi->s_dirs_counter = kmalloc(sizeof(struct percpu_counter), GFP_KERNEL);
>
> Another poster mentioned alloc_percpu() which I wasn't aware of. That might
> be preferrable if it does clever things like allocating out of per-cpu pages
> or something. Sorry, I don't know the details of how to use it.
I think it's possible to use alloc_percpu(). The best way would be to
overhaul the percpu_counter structure to use alloc_percpu(). With just
a quick inspection it looks like this would solve the kmalloc size
problems for a long time, although it would effect performance because
you do one kmalloc() per processor when you do alloc_percpu()
Let me take a look at that...
mh
--
Martin Hicks Wild Open Source Inc.
mort@xxxxxxxxxxxxxxxxxx 613-266-2296
_______________________________________________
Ext3-users@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/ext3-users
