2008-10-28 22:41:56

by Manuel Traut

Subject: UIO: don't check irq_enabled flag in uio_cif irq handler

Below patch ignores INT1_ENABLED flag in the uio_cif irq handler.

On a Hilscher DeviceNet Slave card, the INT_ENABLE flag is (probably) reseted
by the firmware. Without this patch, every interrupt produced by this card is
dropped.

This patch is tested and works with a DeviceNet Slave and a Profibus Slave card.

Signed-off-by: Manuel Traut <[email protected]>

--- linux-git/drivers/uio/uio_cif.c 2008-10-28 23:42:18.000000000 +0100
+++ uio/drivers/uio/uio_cif.c 2008-10-28 23:41:43.000000000 +0100
@@ -18,7 +18,6 @@
#define PLX9030_INTCSR 0x4C
#define INTSCR_INT1_ENABLE 0x01
#define INTSCR_INT1_STATUS 0x04
-#define INT1_ENABLED_AND_ACTIVE (INTSCR_INT1_ENABLE | INTSCR_INT1_STATUS)

#define PCI_SUBVENDOR_ID_PEP 0x1518
#define CIF_SUBDEVICE_PROFIBUS 0x430
@@ -30,8 +29,8 @@
void __iomem *plx_intscr = dev_info->mem[0].internal_addr
+ PLX9030_INTCSR;

- if ((ioread8(plx_intscr) & INT1_ENABLED_AND_ACTIVE)
- != INT1_ENABLED_AND_ACTIVE)
+ if ((ioread8(plx_intscr) & INT1_STATUS)
+ != INT1_STATUS)
return IRQ_NONE;

/* Disable interrupt */

2008-10-29 20:11:19

by Hans J. Koch

Subject: Re: UIO: don't check irq_enabled flag in uio_cif irq handler

On Tue, Oct 28, 2008 at 11:41:37PM +0100, Manuel Traut wrote:
> Below patch ignores INT1_ENABLED flag in the uio_cif irq handler.
>
> On a Hilscher DeviceNet Slave card, the INT_ENABLE flag is (probably) reseted
> by the firmware.

If this is actually the case, then this definetly a firmware bug. If
firmware clears the interrupt enable bit, you have no chance of knowing
if the irq was enabled or not. As a consequence, you cannot find out
whether your card caused the interrupt or not.
A hardware with such a behaviour is simply not a decent PCI card, which
should support shared interrupts.

Send a bug report to the firmware author.

> Without this patch, every interrupt produced by this card is
> dropped.

Yes, probably you'll need that patch if you want to make a card with
such broken behaviour work. But _with_ this patch, you'll run into
problems if you have two such PCI cards that share the same interrupt.

I consider this a workaround that might be OK for your usecase, but it's
not something that should go to mainline.

So, NAK to this.

Thanks,
Hans

>
> This patch is tested and works with a DeviceNet Slave and a Profibus Slave card.
>
> Signed-off-by: Manuel Traut <[email protected]>
>
> --- linux-git/drivers/uio/uio_cif.c 2008-10-28 23:42:18.000000000 +0100
> +++ uio/drivers/uio/uio_cif.c 2008-10-28 23:41:43.000000000 +0100
> @@ -18,7 +18,6 @@
> #define PLX9030_INTCSR 0x4C
> #define INTSCR_INT1_ENABLE 0x01
> #define INTSCR_INT1_STATUS 0x04
> -#define INT1_ENABLED_AND_ACTIVE (INTSCR_INT1_ENABLE | INTSCR_INT1_STATUS)
>
> #define PCI_SUBVENDOR_ID_PEP 0x1518
> #define CIF_SUBDEVICE_PROFIBUS 0x430
> @@ -30,8 +29,8 @@
> void __iomem *plx_intscr = dev_info->mem[0].internal_addr
> + PLX9030_INTCSR;
>
> - if ((ioread8(plx_intscr) & INT1_ENABLED_AND_ACTIVE)
> - != INT1_ENABLED_AND_ACTIVE)
> + if ((ioread8(plx_intscr) & INT1_STATUS)
> + != INT1_STATUS)
> return IRQ_NONE;
>
> /* Disable interrupt */