Re: [PATCH v3] z3fold: use per-page read/write lock

[Andrew Morton]

On Wed, 9 Nov 2016 15:01:17 -0800 Andi Kleen <andi@xxxxxxxxxxxxxx> wrote:

> On Wed, Nov 09, 2016 at 02:33:04PM -0800, Andrew Morton wrote:
> > On Wed, 9 Nov 2016 11:55:31 +0100 Vitaly Wool <vitalywool@xxxxxxxxx> wrote:
> > 
> > > Subject: [PATCH v3] z3fold: use per-page read/write lock
> > 
> > I've rewritten the title to "mm/z3fold.c: use per-page spinlock"
> > 
> > (I prefer to have "mm" in the title to easily identify it as an MM
> > patch, and using "mm: z3fold: ..." seems odd when the actual pathname
> > conveys the same info.)
> 
> Still think it needs to be raw_spinlock_t, otherwise the build bug on
> on the header size will break again. 
> 
> Better would be to fix that build bug though

Yeah, that triggers for me immediately.  We could suppress it with
something silly like

--- a/mm/z3fold.c~z3fold-use-per-page-read-write-lock-fix
+++ a/mm/z3fold.c
@@ -872,7 +872,7 @@ MODULE_ALIAS("zpool-z3fold");
 static int __init init_z3fold(void)
 {
 	/* Make sure the z3fold header will fit in one chunk */
-	BUILD_BUG_ON(sizeof(struct z3fold_header) > ZHDR_SIZE_ALIGNED);
+	BUILD_BUG_ON(sizeof(struct z3fold_header) - sizeof(spinlock_t) > ZHDR_SIZE_ALIGNED);
 	zpool_register_driver(&z3fold_zpool_driver);
 
 	return 0;

but that doesn't fix anything - the header is just too large with
lockdep enabled.

I'll drop the patch for now.