Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 20 Jan 2023 04:54:51 +0000
From:   Matthew Wilcox <willy@infradead.org>
To:     Al Viro <viro@zeniv.linux.org.uk>
Cc:     "Fabio M. De Francesco" <fmdefrancesco@gmail.com>,
        Christoph Hellwig <hch@infradead.org>,
        linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        Ira Weiny <ira.weiny@intel.com>
Subject: Re: [PATCH v3 4/4] fs/sysv: Replace kmap() with kmap_local_page()
Message-ID: <Y8oem+z9SN487MIm@casper.infradead.org>
References: <20230119153232.29750-1-fmdefrancesco@gmail.com>
 <20230119153232.29750-5-fmdefrancesco@gmail.com>
 <Y8oWsiNWSXlDNn5i@ZenIV>
 <Y8oYXEjunDDAzSbe@casper.infradead.org>
 <Y8ocXbztTPbxu3bq@ZenIV>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y8ocXbztTPbxu3bq@ZenIV>
Precedence: bulk

On Fri, Jan 20, 2023 at 04:45:17AM +0000, Al Viro wrote:
> On Fri, Jan 20, 2023 at 04:28:12AM +0000, Matthew Wilcox wrote:
> > On Fri, Jan 20, 2023 at 04:21:06AM +0000, Al Viro wrote:
> > > On Thu, Jan 19, 2023 at 04:32:32PM +0100, Fabio M. De Francesco wrote:
> > > 
> > > > -inline void dir_put_page(struct page *page)
> > > > +inline void dir_put_page(struct page *page, void *page_addr)
> > > >  {
> > > > -	kunmap(page);
> > > > +	kunmap_local(page_addr);
> > > 
> > > ... and that needed to be fixed - at some point "round down to beginning of
> > > page" got lost in rebasing...
> > 
> > You don't need to round down in kunmap().  See:
> > 
> > void kunmap_local_indexed(const void *vaddr)
> > {
> >         unsigned long addr = (unsigned long) vaddr & PAGE_MASK;
> > 
> 
> Sure, but... there's also this:
> 
> static inline void __kunmap_local(const void *addr)
> {
> #ifdef ARCH_HAS_FLUSH_ON_KUNMAP
>         kunmap_flush_on_unmap(addr);
> #endif
> }
> 
> Are you sure that the guts of that thing will be happy with address that is not
> page-aligned?  I've looked there at some point, got scared of parisc (IIRC)
> MMU details and decided not to rely upon that...

Ugh, PA-RISC (the only implementor) definitely will flush the wrong
addresses.  I think we should do this, as having bugs that only manifest
on one not-well-tested architecture seems Bad.

 static inline void __kunmap_local(const void *addr)
 {
 #ifdef ARCH_HAS_FLUSH_ON_KUNMAP
-       kunmap_flush_on_unmap(addr);
+       kunmap_flush_on_unmap(PAGE_ALIGN_DOWN(addr));
 #endif
 }

Thoughts?