Re: [PATCH 14/23] memory: ti-emif-pm: Fix cast to iomem pointer

From: Arnd Bergmann <arnd@arndb.de>
To: Krzysztof Kozlowski <krzk@kernel.org>
Cc: Olof Johansson <olof@lixom.net>, arm-soc <arm@kernel.org>,
	SoC Team <soc@kernel.org>, Markus Mayer <mmayer@broadcom.com>,
	bcm-kernel-feedback-list <bcm-kernel-feedback-list@broadcom.com>,
	Florian Fainelli <f.fainelli@gmail.com>,
	Santosh Shilimkar <ssantosh@kernel.org>,
	Matthias Brugger <matthias.bgg@gmail.com>,
	Roger Quadros <rogerq@ti.com>, Tony Lindgren <tony@atomide.com>,
	Vladimir Zapolskiy <vz@mleia.com>,
	Thierry Reding <thierry.reding@gmail.com>,
	Jonathan Hunter <jonathanh@nvidia.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Linux ARM <linux-arm-kernel@lists.infradead.org>,
	"moderated list:ARM/Mediatek SoC..." 
	<linux-mediatek@lists.infradead.org>,
	linux-omap <linux-omap@vger.kernel.org>,
	"open list:TEGRA ARCHITECTURE SUPPORT"
	<linux-tegra@vger.kernel.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Subject: Re: [PATCH 14/23] memory: ti-emif-pm: Fix cast to iomem pointer
Date: Thu, 23 Jul 2020 11:14:02 +0200	[thread overview]
Message-ID: <CAK8P3a2O4bCb_Bcpc+Y4BuXmrO6mbVwgo67aGF5E=uxHQpeTnw@mail.gmail.com> (raw)
In-Reply-To: <20200723090230.GB25935@kozik-lap>

On Thu, Jul 23, 2020 at 11:02 AM Krzysztof Kozlowski <krzk@kernel.org> wrote:
> On Thu, Jul 23, 2020 at 10:48:19AM +0200, Arnd Bergmann wrote:
> > On Thu, Jul 23, 2020 at 9:39 AM Krzysztof Kozlowski <krzk@kernel.org> wrote:
> > >
> > > Cast pointer to iomem memory properly to fix sparse warning:
> > >
> > >     drivers/memory/ti-emif-pm.c:251:38: warning: incorrect type in argument 1 (different address spaces)
> > >     drivers/memory/ti-emif-pm.c:251:38:    expected void const volatile [noderef] __iomem *addr
> > >     drivers/memory/ti-emif-pm.c:251:38:    got void *
> > >
> > > Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
> > > ---
> > >  drivers/memory/ti-emif-pm.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/memory/ti-emif-pm.c b/drivers/memory/ti-emif-pm.c
> > > index 9c90f815ad3a..6c747c1e98cb 100644
> > > --- a/drivers/memory/ti-emif-pm.c
> > > +++ b/drivers/memory/ti-emif-pm.c
> > > @@ -248,7 +248,7 @@ MODULE_DEVICE_TABLE(of, ti_emif_of_match);
> > >  static int ti_emif_resume(struct device *dev)
> > >  {
> > >         unsigned long tmp =
> > > -                       __raw_readl((void *)emif_instance->ti_emif_sram_virt);
> > > +                       __raw_readl((void __iomem *)emif_instance->ti_emif_sram_virt);
> > >
> >
> > Maybe this shouldn't even be __raw_readl(), but instead READ_ONCE()?
>
> Won't readl() be enough? Indeed it looks problematic.

readl() won't work on big-endian kernels, since this is a byte comparison.

> > The other accesses in this file don't use MMIO wrappers either but just treat
> > it as a pointer. The effect would be the same though.
>
> I think all the reads and writes are with readl() and writel().

I actually see only one other access:

        copy_addr = sram_exec_copy(emif_data->sram_pool_code,
                                   (void *)emif_data->ti_emif_sram_virt,
                                   &ti_emif_sram, ti_emif_sram_sz);

and this one ends up in a memcpy() that does not perform any byte
swapping or barriers.

     Arnd