linux-3.4-rc1 PXA2xx PCMCIA possible regression

linux-3.4-rc1 PXA2xx PCMCIA possible regression

[Paul Parsons]

On linux-3.3 the PCMCIA/CF on my hx4700 worked OK.

On linux-3.4-rc1 the PCMCIA/CF has stopped working:

pcmcia_socket pcmcia_socket0: pccard: PCMCIA card inserted into slot 0
pcmcia 0.0: pcmcia: registering new device pcmcia0.0 (IRQ: 156)
pata_pcmcia: probe of 0.0 failed with error -12

I've traced the -ENOMEM error back to pcmcia_init_one() in
drivers/ata/pata_pcmcia.c:

   237		/* iomap */
   238		ret = -ENOMEM;
   239		io_addr = devm_ioport_map(&pdev->dev, io_base, 8);
   240		ctl_addr = devm_ioport_map(&pdev->dev, ctl_base, 1);
   241		if (!io_addr || !ctl_addr)
   242			goto failed;

Both calls to devm_ioport_map() return NULL.

The ultimate source of both NULL values appears to be __io() and
thence __typesafe_io(), both defined in arch/arm/include/asm/io.h

linux-3.4-rc1 PXA2xx PCMCIA possible regression

[Rob Herring]

On 04/03/2012 08:57 PM, Paul Parsons wrote:
> On linux-3.3 the PCMCIA/CF on my hx4700 worked OK.
> 
> On linux-3.4-rc1 the PCMCIA/CF has stopped working:
> 
> pcmcia_socket pcmcia_socket0: pccard: PCMCIA card inserted into slot 0
> pcmcia 0.0: pcmcia: registering new device pcmcia0.0 (IRQ: 156)
> pata_pcmcia: probe of 0.0 failed with error -12
> 
> I've traced the -ENOMEM error back to pcmcia_init_one() in
> drivers/ata/pata_pcmcia.c:
> 
>    237		/* iomap */
>    238		ret = -ENOMEM;
>    239		io_addr = devm_ioport_map(&pdev->dev, io_base, 8);
>    240		ctl_addr = devm_ioport_map(&pdev->dev, ctl_base, 1);
>    241		if (!io_addr || !ctl_addr)
>    242			goto failed;
> 
> Both calls to devm_ioport_map() return NULL.
> 
> The ultimate source of both NULL values appears to be __io() and
> thence __typesafe_io(), both defined in arch/arm/include/asm/io.h

I broke it with the io.h clean-up. It seems PCMCIA drivers rely on
__io() just being a cast. The easy fix is just restore that behavior for
PCMCIA:

diff --git a/arch/arm/include/asm/io.h b/arch/arm/include/asm/io.h
index df0ac0b..1b50c28 100644
--- a/arch/arm/include/asm/io.h
+++ b/arch/arm/include/asm/io.h
@@ -118,6 +118,8 @@ static inline void __iomem *__typesafe_io(unsigned
long addr)
  */
 #ifdef CONFIG_NEED_MACH_IO_H
 #include <mach/io.h>
+#elif defined(CONFIG_PCMCIA)
+#define __io(a)                __typesafe_io(a)
 #else
 #define __io(a)                ({ (void)(a); __typesafe_io(0); })
 #endif


Can you please test this change. This is what we'll have to do for 3.4,
but this is not really the right solution. io_base should not be a
virtual address here, but an i/o address in the range of 0-64K. We need
to move PCMCIA to the fixed i/o mapping I've posted:

http://www.spinics.net/lists/arm-kernel/msg162547.html

I've been trying to get PCMCIA to work on spitz in QEMU to test out
fixes, but haven't had much luck. I'm not getting a card detected even
though QEMU thinks one is present.

Rob

> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

linux-3.4-rc1 PXA2xx PCMCIA possible regression

[Paul Parsons]

Hello Rob,

--- On Wed, 4/4/12, Rob Herring <robherring2@gmail.com> wrote:
> On 04/03/2012 08:57 PM, Paul Parsons
> wrote:
> > On linux-3.3 the PCMCIA/CF on my hx4700 worked OK.
> > 
> > On linux-3.4-rc1 the PCMCIA/CF has stopped working:
> > 
> > pcmcia_socket pcmcia_socket0: pccard: PCMCIA card
> inserted into slot 0
> > pcmcia 0.0: pcmcia: registering new device pcmcia0.0
> (IRQ: 156)
> > pata_pcmcia: probe of 0.0 failed with error -12
> > 
> > I've traced the -ENOMEM error back to pcmcia_init_one()
> in
> > drivers/ata/pata_pcmcia.c:
> > 
> >? ? 237??? ???
> /* iomap */
> >? ? 238??? ???
> ret = -ENOMEM;
> >? ? 239??? ???
> io_addr = devm_ioport_map(&pdev->dev, io_base, 8);
> >? ? 240??? ???
> ctl_addr = devm_ioport_map(&pdev->dev, ctl_base, 1);
> >? ? 241??? ???
> if (!io_addr || !ctl_addr)
> >? ? 242??? ???
> ??? goto failed;
> > 
> > Both calls to devm_ioport_map() return NULL.
> > 
> > The ultimate source of both NULL values appears to be
> __io() and
> > thence __typesafe_io(), both defined in
> arch/arm/include/asm/io.h
> 
> I broke it with the io.h clean-up. It seems PCMCIA drivers
> rely on
> __io() just being a cast. The easy fix is just restore that
> behavior for
> PCMCIA:
> 
> diff --git a/arch/arm/include/asm/io.h
> b/arch/arm/include/asm/io.h
> index df0ac0b..1b50c28 100644
> --- a/arch/arm/include/asm/io.h
> +++ b/arch/arm/include/asm/io.h
> @@ -118,6 +118,8 @@ static inline void __iomem
> *__typesafe_io(unsigned
> long addr)
> ? */
>  #ifdef CONFIG_NEED_MACH_IO_H
>  #include <mach/io.h>
> +#elif defined(CONFIG_PCMCIA)
> +#define __io(a)? ? ? ? ? ?
> ? ? __typesafe_io(a)
>  #else
>  #define __io(a)? ? ? ? ? ?
> ? ? ({ (void)(a); __typesafe_io(0); })
>  #endif
> 
> 
> Can you please test this change. This is what we'll have to
> do for 3.4,
> but this is not really the right solution. io_base should
> not be a
> virtual address here, but an i/o address in the range of
> 0-64K. We need
> to move PCMCIA to the fixed i/o mapping I've posted:
> 
> http://www.spinics.net/lists/arm-kernel/msg162547.html
> 
> I've been trying to get PCMCIA to work on spitz in QEMU to
> test out
> fixes, but haven't had much luck. I'm not getting a card
> detected even
> though QEMU thinks one is present.

Yes, this change fixes the PCMCIA/CF on my hx4700 (PXA270).

Thanks,
Paul

linux-3.4-rc1 PXA2xx PCMCIA possible regression

[Russell King - ARM Linux]

On Wed, Apr 04, 2012 at 02:57:22AM +0100, Paul Parsons wrote:
> On linux-3.3 the PCMCIA/CF on my hx4700 worked OK.
> 
> On linux-3.4-rc1 the PCMCIA/CF has stopped working:
> 
> pcmcia_socket pcmcia_socket0: pccard: PCMCIA card inserted into slot 0
> pcmcia 0.0: pcmcia: registering new device pcmcia0.0 (IRQ: 156)
> pata_pcmcia: probe of 0.0 failed with error -12

I see no PCMCIA support for HX4700 in mainline.  Where is it?

linux-3.4-rc1 PXA2xx PCMCIA possible regression

[Paul Parsons]

Hello Russell,

--- On Wed, 4/4/12, Russell King - ARM Linux <linux@arm.linux.org.uk> wrote:
> On Wed, Apr 04, 2012 at 02:57:22AM
> +0100, Paul Parsons wrote:
> > On linux-3.3 the PCMCIA/CF on my hx4700 worked OK.
> > 
> > On linux-3.4-rc1 the PCMCIA/CF has stopped working:
> > 
> > pcmcia_socket pcmcia_socket0: pccard: PCMCIA card
> inserted into slot 0
> > pcmcia 0.0: pcmcia: registering new device pcmcia0.0
> (IRQ: 156)
> > pata_pcmcia: probe of 0.0 failed with error -12
> 
> I see no PCMCIA support for HX4700 in mainline.? Where
> is it?

It is languishing in pending patch land:

http://lists.infradead.org/pipermail/linux-arm-kernel/2012-March/089266.html
http://lists.infradead.org/pipermail/linux-arm-kernel/2012-March/088988.html

Regards,
Paul

linux-3.4-rc1 PXA2xx PCMCIA possible regression

[Russell King - ARM Linux]

On Wed, Apr 04, 2012 at 02:55:21PM +0100, Paul Parsons wrote:
> It is languishing in pending patch land:
> 
> http://lists.infradead.org/pipermail/linux-arm-kernel/2012-March/089266.html
> http://lists.infradead.org/pipermail/linux-arm-kernel/2012-March/088988.html

These links aren't working for me at the moment.  I found the patches
in my mailbox nevertheless.

As you're using the PXA stuff, that selects PCMCIA_SOC_COMMON, which means
this part of Rob's patch is rubbish, because you'll get the 4GB IO space
precisely because of PCMCIA_SOC_COMMON being enabled.

 #ifndef IO_SPACE_LIMIT
-#if defined(CONFIG_PCMCIA_SOC_COMMON) || defined(CONFIG_PCMCIA_SOC_COMMON_MODULE)
+#if defined(CONFIG_PCCARD)
 #define IO_SPACE_LIMIT ((resource_size_t)0xffffffff)
-#elif defined(CONFIG_PCI) || defined(CONFIG_ISA) || defined(CONFIG_PCCARD)
+#elif defined(CONFIG_PCI) || defined(CONFIG_ISA)

So it all hinges around what we supply for the __io() macro.