Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Tue, 12 Nov 2019 22:24:23 +0000
From:   Luis Chamberlain <mcgrof@kernel.org>
To:     Daniel Vetter <daniel.vetter@ffwll.ch>,
        Juergen Gross <jgross@suse.com>
Cc:     Christoph Hellwig <hch@lst.de>, Tuowen Zhao <ztuowen@gmail.com>,
        AceLan Kao <acelan.kao@canonical.com>,
        Mika Westerberg <mika.westerberg@linux.intel.com>,
        Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
        Roman Gilg <subdiff@gmail.com>,
        Lee Jones <lee.jones@linaro.org>,
        "Luis R. Rodriguez" <mcgrof@suse.com>,
        Arnd Bergmann <arnd@arndb.de>,
        Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>,
        X86 ML <x86@kernel.org>, Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        "H. Peter Anvin" <hpa@zytor.com>, linux-ia64@vger.kernel.org,
        Tony Luck <tony.luck@intel.com>,
        Fenghua Yu <fenghua.yu@intel.com>,
        Maarten Lankhorst <maarten.lankhorst@linux.intel.com>,
        Souptick Joarder <jrdr.linux@gmail.com>,
        dri-devel <dri-devel@lists.freedesktop.org>,
        Linux Fbdev development list <linux-fbdev@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] video: fbdev: atyfb: only use ioremap_uc() on i386 and
 ia64
Message-ID: <20191112222423.GO11244@42.do-not-panic.com>
References: <20191111192258.2234502-1-arnd@arndb.de>
 <20191112105507.GA7122@lst.de>
 <CAKMK7uEEz1n+zuTs29rbPHU74Dspaib=prpMge63L_-rUk_o4A@mail.gmail.com>
 <20191112140631.GA10922@lst.de>
 <CAKMK7uFaA607rOS6x_FWjXQ2+Qdm8dQ1dQ+Oi-9if_Qh_wHWPg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAKMK7uFaA607rOS6x_FWjXQ2+Qdm8dQ1dQ+Oi-9if_Qh_wHWPg@mail.gmail.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Nov 12, 2019 at 03:26:35PM +0100, Daniel Vetter wrote:
> On Tue, Nov 12, 2019 at 3:06 PM Christoph Hellwig <hch@lst.de> wrote:
> > On Tue, Nov 12, 2019 at 02:04:16PM +0100, Daniel Vetter wrote:
> > > Wut ... Maybe I'm missing something, but from how we use mtrr in other
> > > gpu drivers it's a) either you use MTRR because that's all you got or
> > > b) you use pat. Mixing both sounds like a pretty bad idea, since if
> > > you need MTRR for performance (because you dont have PAT) then you
> > > can't fix the wc with the PAT-based ioremap_uc. And if you have PAT,
> > > then you don't really need an MTRR to get wc.
> > >
> > > So I'd revert this patch from Luis and ...
> >
> > Sounds great to me..
> >
> > > ... apply this one. Since the same reasoning should apply to anything
> > > that's running on any cpu with PAT.
> >
> > Can you take a look at "mfd: intel-lpss: Use devm_ioremap_uc for MMIO"
> > in linux-next, which also looks rather fishy to me?  Can't we use
> > the MTRR APIs to override the broken BIOS MTRR setup there as well?
> 
> Hm so that's way out of my knowledge, but I think mtrr_cleanup() was
> supposed to fix up messy/broken MTRR setups by the bios. So maybe they
> simply didn't enable that in their .config with CONFIG_MTRR_SANITIZER.

I had originally suggested to just make the driver build on x86, but an
atlternative was to provide the call for the missing architecture.

> An explicit cleanup is currently not possible for drivers, since the
> only interface exported to drivers is arch_phys_wc_add/del (which
> short-circuits if pat works since you don't need mtrr in that case).

Right, the goal was to not call MTRR directly.

> Adding everyone from that commit, plus Luis. Drivers really shouldn't
> assume/work around the bios setting up superflous/wrong MTRR.

Such things are needed, otherwise some systems may not boot...

> > With that we could kill ioremap_uc entirely.
> 
> So yeah removing that seems definitely like the right thing.

I think this would be possible if we could flop ioremap_nocache() to UC
instead of UC- on x86. Otherwise, I can't see how we can remove this by
still not allowing direct MTRR calls.

  Luis