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;
MIME-Version: 1.0
References: <20220805205321.19452-1-andriy.shevchenko@linux.intel.com>
 <20220805205321.19452-5-andriy.shevchenko@linux.intel.com> <YvENHSRQuwAzfQJW@sirena.org.uk>
In-Reply-To: <YvENHSRQuwAzfQJW@sirena.org.uk>
From:   Andy Shevchenko <andy.shevchenko@gmail.com>
Date:   Mon, 8 Aug 2022 16:40:26 +0200
Message-ID: <CAHp75VeA8tA-m_gHu5iuhriSfm_qo1PRnUWe7kQ68CCm+UCgHA@mail.gmail.com>
Subject: Re: [PATCH v1 5/5] regmap: mmio: Introduce IO accessors that can talk
 to IO port
To:     Mark Brown <broonie@kernel.org>
Cc:     Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
        Aidan MacDonald <aidanmacdonald.0x0@gmail.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        "Rafael J. Wysocki" <rafael@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, Aug 8, 2022 at 3:31 PM Mark Brown <broonie@kernel.org> wrote:
> On Fri, Aug 05, 2022 at 11:53:21PM +0300, Andy Shevchenko wrote:
>
> > Currently regmap MMIO is inconsistent with IO accessors. I.e.
> > the Big Endian counterparts are using ioreadXXbe() / iowriteXXbe()
> > which are not clean implementations of readXXbe(). Besides that
> > some users may use regmap MMIO for IO ports, and this can be done
> > by assigning ioreadXX()/iowriteXX() and their Big Endian counterparts
> > to the regmap context.
>
> Have you validated that nothing is relying on whatever the problem is
> with using the io versions?

I have cross-checked 1) the architectures that are BE and have IO port
capability, and 2) the drivers that are using regmap MMIO with a
big-endian setting. I found no driver is mapping IO ports and uses
regmap MMIO at the same time. The architecture wise the x86 and ia64
are not in question, I think. And alpha is more academical nowadays.
Did I miss anything?

That said, I'm 99.999% sure there is no problem with that.

...

> > That said, reimplement current Big Endian MMIO accessors by replacing
> > ioread()/iowrite() with respective read()/write() and swab() calls.
> > While at it, add IO port support with a corresponding flag added.
>
> This should be a separate patch.

OK! Then we remove some code and (re-)add it later. Do we need this churn?
Another way is to add IO port accessors and then fix the MMIO.

...

> > +                     if (config->io_port) {
> > +                             ctx->reg_read = regmap_mmio_ioread8;
> > +                             ctx->reg_write = regmap_mmio_iowrite8;
> > +                     } else if (config->use_relaxed_mmio) {
>
> If these options are mutually exclusive we should validate that they are
> not simultaneously set.

Yes, the validation is missed. I will add it.

-- 
With Best Regards,
Andy Shevchenko