Re: [PATCH 17/24] ARM: OMAP: use __iomem pointers for MMIO

From: Arnd Bergmann <arnd@arndb.de>
To: Tony Lindgren <tony@atomide.com>
Cc: linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, Will Deacon <will.deacon@arm.com>,
	Russell King <rmk+kernel@arm.linux.org.uk>,
	Nicolas Pitre <nico@linaro.org>,
	Felipe Balbi <felipe.balbi@nokia.com>,
	Lokesh Vutla <lokeshvutla@ti.com>,
	Santosh Shilimkar <santosh.shilimkar@ti.com>,
	Shubhrajyoti D <shubhrajyoti@ti.com>,
	sricharan <r.sricharan@ti.com>
Subject: Re: [PATCH 17/24] ARM: OMAP: use __iomem pointers for MMIO
Date: Wed, 19 Sep 2012 13:35:47 +0000	[thread overview]
Message-ID: <201209191335.48051.arnd@arndb.de> (raw)
In-Reply-To: <20120917212506.GB11762@atomide.com>

On Monday 17 September 2012, Tony Lindgren wrote:
> * Tony Lindgren <tony@atomide.com> [120916 13:39]:
> > * Arnd Bergmann <arnd@arndb.de> [120915 13:15]:
> > > On Saturday 15 September 2012, Tony Lindgren wrote:
> > > > With my patches, this is now all omap1 specific and
> > > > moved to arch/arm/mach-omap1/include/mach/hardware.h.
> > > > It's probably easiest to just update this patch on
> > > > top of the hardware.h changes I've done.
> > > 
> > > Yes, sounds good. Do you want to send a patch for that
> > > and let me drop this one then?
> > 
> > Yes I can pick this one and update it against one of my
> > branches to avoid merge conflicts.
> 
> This applies against mach-omap1/include/mach/hardware.h
> with some fuzz so no issues there.
> 
> But I think we should not apply it as these are physical
> addresses, not virtual addresses for omap1.

Right, I misread what is actually going on here because the
only driver I looked at treated the address as a virtual
address pointer.

> We have IOMEM already in use for omap_read/write because of:
> 
> #define OMAP1_IO_ADDRESS(pa)    IOMEM((pa) - OMAP1_IO_OFFSET)
> 
> I think the right solution is to eventually get rid of
> omap_read/write for omap1 also and replace them with ioremap
> + readl/writel.

Agreed.

> Or am I missing something?

I did not see any new warnings for omap1, but I did see this
on omap2plus_defconfig:

drivers/watchdog/omap_wdt.c: In function 'omap_wdt_ioctl':
drivers/watchdog/omap_wdt.c:222:4: error: passing argument 1 of '__raw_readw' makes pointer from integer without a cast [-Werror]
arch/arm/include/asm/io.h:71:90: note: expected 'const volatile void *' but argument is of type 'unsigned int'

It seems I misinterpreted this, and it's actually a bug in the watchdog
driver that should be fixed using this patch instead (and backport it
to stable)

diff --git a/drivers/watchdog/omap_wdt.c b/drivers/watchdog/omap_wdt.c
index fceec4f..7b45802 100644
--- a/drivers/watchdog/omap_wdt.c
+++ b/drivers/watchdog/omap_wdt.c
@@ -218,9 +218,11 @@ static long omap_wdt_ioctl(struct file *file, unsigned int cmd,
 	case WDIOC_GETSTATUS:
 		return put_user(0, (int __user *)arg);
 	case WDIOC_GETBOOTSTATUS:
+#ifdef CONFIG_ARCH_OMAP1
 		if (cpu_is_omap16xx())
-			return put_user(__raw_readw(ARM_SYSST),
+			return put_user(omap_readw(ARM_SYSST),
 					(int __user *)arg);
+#endif
 		if (cpu_is_omap24xx())
 			return put_user(omap_prcm_get_reset_sources(),
 					(int __user *)arg);

This bug seems to have been introduced in 2008 by 9f69e3b0c "[WATCHDOG]
omap_wdt.c: another ioremap() fix" without anyone ever noticing and now
got caught. Of course it should be replaced by something better when
omap_read/write is finally getting removed.

I'll drop my omap patch for now, because it's obviously wrong, and let
you guys figure out what to do about the watchdog driver.

	Arnd
