Re: [PATCH] dax: kill uml support

From: Ross Zwisler <ross.zwisler@linux.intel.com>
To: Jan Kara <jack@suse.cz>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>,
	Dan Williams <dan.j.williams@intel.com>,
	akpm@linux-foundation.org,
	Matthew Wilcox <mawilcox@microsoft.com>,
	Dave Chinner <david@fromorbit.com>,
	linux-kernel@vger.kernel.org, Dave Hansen <dave.hansen@intel.com>,
	Alexander Viro <viro@zeniv.linux.org.uk>,
	linux-fsdevel@vger.kernel.org, Christoph Hellwig <hch@lst.de>
Subject: Re: [PATCH] dax: kill uml support
Date: Thu, 22 Dec 2016 09:40:42 -0700	[thread overview]
Message-ID: <20161222164042.GA18407@linux.intel.com> (raw)
In-Reply-To: <20161222100055.GB28510@quack2.suse.cz>

On Thu, Dec 22, 2016 at 11:00:55AM +0100, Jan Kara wrote:
> On Wed 21-12-16 10:51:18, Ross Zwisler wrote:
> > On Wed, Dec 21, 2016 at 09:53:47AM +0100, Jan Kara wrote:
> > > On Tue 20-12-16 17:37:40, Dan Williams wrote:
> > > > The lack of common transparent-huge-page helpers for UML is becoming
> > > > increasingly painful for fs/dax.c now that it is growing more pmd
> > > > functionality. Add UML to the list of unsupported architectures, and
> > > > clean up no-longer-necessary ifdef as a result.
> > > ...
> > > > diff --git a/fs/dax.c b/fs/dax.c
> > > > index ddcddfeaa03b..86df835783ea 100644
> > > > --- a/fs/dax.c
> > > > +++ b/fs/dax.c
> > > > @@ -710,8 +710,7 @@ static void dax_mapping_entry_mkclean(struct address_space *mapping,
> > > >  		if (follow_pte_pmd(vma->vm_mm, address, &ptep, &pmdp, &ptl))
> > > >  			continue;
> > > >  
> > > > -		if (pmdp) {
> > > > -#ifdef CONFIG_FS_DAX_PMD
> > > > +		if (pmdp && IS_ENABLED(CONFIG_FS_DAX_PMD)) {
> > > >  			pmd_t pmd;
> > > 
> > > So I was under the impression that pmdp can never be != NULL when
> > > CONFIG_FS_DAX_PMD is disabled. Otherwise Ross' patch would leave ptl locked
> > > in that case... Did I miss something or we can just remove IS_ENABLED()
> > > check?
> > 
> > We need the IS_ENABLED() check to prevent a different build error.
> > 
> > The #ifdef was there to prevent compile time errors where pmd_pfn(),
> > pmd_write(), etc were all undefined symbols because they aren't defined in
> > the arch/um headers.
> > 
> > For x86_64 we can get linker errors if CONFIG_TRANSPARENT_HUGEPAGE isn't set:
> > 
> > fs/built-in.o: In function `dax_writeback_mapping_range.part.27':
> > dax.c:(.text+0x4ce28): undefined reference to `pmdp_huge_clear_flush'
> > 
> > In this config we do have a prototype for pmdp_huge_clear_flush() in
> > include/asm-generic/pgtable.h so it doesn't show up as an undefined symbol,
> > but the implementation in mm/pgtable-generic.c is in a
> > #ifdef CONFIG_TRANSPARENT_HUGEPAGE block.
> > 
> > The IS_ENABLED() lets the compiler optimize out the code that calls
> > pmdp_huge_clear_flush() so it doesn't try and resolve that symbol at link
> > time.
> 
> Ah, OK. Won't it be cleaner to just have empty stub for
> pmdp_huge_clear_flush() if !CONFIG_TRANSPARENT_HUGEPAGE? I like that more
> but I can live with IS_ENABLED check as well but please add a comment that
> it is there only to make compiler happy.

Yea, making an empty stub is cleaner.  That has the added bonus that we don't
have to rely on the assumption that 'pmdp' will always be NULL in if
CONFIG_TRANSPARENT_HUGEPAGE isn't set.

I'll reflow & send out v2.