Date: Tue, 7 Aug 2012 17:48:22 +0100
From: Dave Martin <dave.martin@linaro.org>
To: Matt Sealey <matt@genesi-usa.com>
Cc: Anton Vorontsov <anton.vorontsov@linaro.org>,
        linaro-kernel@lists.linaro.org, Russell King <linux@arm.linux.org.uk>,
        Mark Brown <broonie@opensource.wolfsonmicro.com>,
        linux-kernel@vger.kernel.org, John Stultz <john.stultz@linaro.org>,
        Sascha Hauer <kernel@pengutronix.de>,
        linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH 4/9] ASoC: imx: Don't use {en,dis}able_fiq() calls
Message-ID: <20120807164822.GA2135@linaro.org>
References: <20120805230238.GA1663@lizard>
 <1344207819-3415-4-git-send-email-anton.vorontsov@linaro.org>
 <CAKGA1bkiHj=zMJXSzoGp3fpLS3POnFXSV5y7sYYFAikqoUT9tQ@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAKGA1bkiHj=zMJXSzoGp3fpLS3POnFXSV5y7sYYFAikqoUT9tQ@mail.gmail.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 4292
Lines: 78

On Mon, Aug 06, 2012 at 10:19:50AM -0500, Matt Sealey wrote:
> On Sun, Aug 5, 2012 at 6:03 PM, Anton Vorontsov
> <anton.vorontsov@linaro.org> wrote:
> > The driver uses platform-specific mxc_set_irq_fiq() with the VIRQ cookie
> > passed to it, so it's pretty clear that the driver is absolutely sure
> > that the FIQ is routed via platform-specific IC, and that the cookie can
> > be used to mask/unmask FIQs. So, let's switch to the genirq routines,
> > since we're about to remove FIQ-specific variants.
> 
> I have a semi-related question about this;
> 
> Firstly, I was planning on (re-)submitting a patch for the
> arch/arm/plat-mxc/ssi-fiq.S code which made it build in ARM mode
> (since the code isn't Thumb compatible for various reasons) as it was
> a blocker for a Thumb2-compiled kernel. Since the code was only needed
> on ARM-capable processors it wouldn't cause a problem. Sascha signed
> off on this a long while back and I've been testing it on all my
> internal kernel versions, and I don't see any ill effects (that said I
> don't have an i.MX28 or so to really verify it, I can't see why it
> would not work). I realise this is redundant right now, anyway, since
> it's only really enabled on imx_v4_v5 configs and they don't support
> Thumb2 kernels anyway. What might be worth submitting is a switch to
> add the ".arm" directive anyway simply for correctness since it could
> never be compiled for Thumb anyway. We all know what gnu as is like.
> 
> Looking at it again on the back of these patches, I noticed the
> ssi-fiq.S code is compiled in when SND_IMX_SOC is enabled - of course,
> it's only needed in the kernel if SND_SOC_IMX_PCM_FIQ is enabled (the
> code that uses the FIQ stuff is only compiled then) but here on the
> Efika MX builds it's being built, and I noticed it when it broke my
> build because of the above. I'm therefore going to submit a patch
> which changes the ifdef SND_SOC_IMX to ifdef SND_SOC_IMX_PCM_FIQ so
> it's not compiled in unless absolutely necessary. However, there was a
> rumble that this code may disappear or be reworked in the future
> making this also quite redundant. Since it's not in the
> imx_v6_v7_defconfig anyway, I'm sure this only didn't get noticed
> because nobody's building Thumb2 kernels and nobody is trying configs
> with audio enabled anyway..
> 
> This begs the question, could there be something somewhere hidden deep
> in the kernel that is enabled by default or in some config somewhere
> that requires "select FIQ" in it's Kconfig entry, but isn't being
> enabled? On i.MX the only thing turning it on is that code, but other
> ARM arch enable it by default. Since things have been moved to more
> generic routines I can't in my mind guarantee that such a patch would
> be well tested here since I would be relying on symbols missing or
> defines not there anymore.. I have no real way to ensure that it would
> work on boards I don't have.
> 
> So, is the first patch (ssi-fiq.S .arm) worth it? I think the
> SND_SOC_IMX_PCM_FIQ patch is worth it for v6_v7 systems anyway, but
> maybe this should have been caught sooner, so should I update the
> defconfig to enable some kind of audio bus support (Babbage has it in
> the DT so is a needful thing for testing, I figure..)? And does anyone
> forsee any problems with that option changing and the only "user" of
> "FIQ" in the Kconfigs going away now all the FIQ-specific symbols went
> away outside of the generic irq subsystem?

I hit this issue some time ago when I was trying to do a Thumb2 build
of this kernel.

I don't remember having to fix the generic FIQ code just for this,
but I had a patch somewhere to swap r13 with another register in
ssi-fiq.S.  I think that use of r13 in ways not permitted for Thumb
was the only problem I found.  I can try to dig it out if you want.

In any case, it didn't seem worth pushing at the time, because it
seemed that the SSI FIQ code was not relevant to any v7 or later
platform -- so building that code for Thumb presumably doesn't make
sense.

Cheers
---Dave
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/