[PATCH] Making TI bridges work with kernel PCMCIA

[PATCH] Making TI bridges work with kernel PCMCIA

[Pavel Roskin]

Hello!

The driver for Cardbus/PCMCIA bridges (yenta_socket) doesn't initialize 
the irqmux register of TI bridges.  In fact, the driver doesn't even 
access that register.

My experience shows that relying in the initial values of that register is
not a good idea.  It is set to use ISA interrupts, but they don't work
with either of two motherboards I tested the bridge with.

Bridge:
CardBus bridge: Texas Instruments PCI1410 PC card Cardbus Controller (rev 01)

Fisrt system:
Athlon 1GHz, motherboard AOpen KT-133.

Second system:
Compaq AP500, 2 x Pentium II, motherboard with Intel 440BX chipset.

This patch makes the same thing as "irq_mode=0" does in the standalone
PCMCIA driver (i82365).  It means that PCMCIA cards inserted to the socket
have their interrupt redirected to the same PCI interrupt that is used for
the Card Status Change (csc) notification.  I believe that's the safest
behavior, as it avoids probing ISA interupts.

Since the patch changes ti_open(), it affects all TI bridges.  This should 
be the right thing to do according to the pcmcia-cs sources.  I'm a bit 
surprized that the driver doesn't access those registers at all except 
some old TI113x models.

The code with TI122X_SCR_INTRTIE has been copied from pcmcia-cs without
changes.  It should only affect two-socket bridges.

The patch is against 2.4.17-rc2.

=========================
--- linux.orig/drivers/pcmcia/ti113x.h
+++ linux/drivers/pcmcia/ti113x.h
@@ -150,11 +150,27 @@
  */
 static int ti_open(pci_socket_t *socket)
 {
+	u32 irqmux;
+	u8 devctl, sysctl;
 	u8 new, reg = exca_readb(socket, I365_INTCTL);
 
 	new = reg & ~I365_INTR_ENA;
 	if (new != reg)
 		exca_writeb(socket, I365_INTCTL, new);
+
+	/* Disable ISA interrupt routing ... */
+	devctl = config_readb(socket, TI113X_DEVICE_CONTROL);
+	devctl &= ~TI113X_DCR_IMODE_MASK;
+	config_writeb(socket, TI113X_DEVICE_CONTROL, devctl);
+	/* ... and enable PCI routing instead */
+	sysctl = config_readb(socket, TI113X_SYSTEM_CONTROL);
+	irqmux = config_readl(socket, TI122X_IRQMUX);
+	irqmux = (irqmux & ~0x0f) | 0x02; /* route INTA */
+	if (!(sysctl & TI122X_SCR_INTRTIE)) {
+		irqmux = (irqmux & ~0xf0) | 0x20; /* route INTB */
+	}
+	config_writel(socket, TI122X_IRQMUX, irqmux);
+
 	return 0;
 }
 
=========================

-- 
Regards,
Pavel Roskin

Re: [PATCH] Making TI bridges work with kernel PCMCIA

[Benjamin Herrenschmidt]

>Hello!
>
>The driver for Cardbus/PCMCIA bridges (yenta_socket) doesn't initialize 
>the irqmux register of TI bridges.  In fact, the driver doesn't even 
>access that register.
>
>My experience shows that relying in the initial values of that register is
>not a good idea.  It is set to use ISA interrupts, but they don't work
>with either of two motherboards I tested the bridge with.
>
>Bridge:
>CardBus bridge: Texas Instruments PCI1410 PC card Cardbus Controller (rev 01)
>
>Fisrt system:
>Athlon 1GHz, motherboard AOpen KT-133.
>
>Second system:
>Compaq AP500, 2 x Pentium II, motherboard with Intel 440BX chipset.
>
>This patch makes the same thing as "irq_mode=0" does in the standalone
>PCMCIA driver (i82365).  It means that PCMCIA cards inserted to the socket
>have their interrupt redirected to the same PCI interrupt that is used for
>the Card Status Change (csc) notification.  I believe that's the safest
>behavior, as it avoids probing ISA interupts.
>
>Since the patch changes ti_open(), it affects all TI bridges.  This should 
>be the right thing to do according to the pcmcia-cs sources.  I'm a bit 
>surprized that the driver doesn't access those registers at all except 
>some old TI113x models.
>
>The code with TI122X_SCR_INTRTIE has been copied from pcmcia-cs without
>changes.  It should only affect two-socket bridges.
>
>The patch is against 2.4.17-rc2.

I have to use a similar workaround on pmac laptops. However, I do that
in the platform specific quirks.

Ben.

>=========================
>--- linux.orig/drivers/pcmcia/ti113x.h
>+++ linux/drivers/pcmcia/ti113x.h
>@@ -150,11 +150,27 @@
>  */
> static int ti_open(pci_socket_t *socket)
> {
>+	u32 irqmux;
>+	u8 devctl, sysctl;
> 	u8 new, reg = exca_readb(socket, I365_INTCTL);
> 
> 	new = reg & ~I365_INTR_ENA;
> 	if (new != reg)
> 		exca_writeb(socket, I365_INTCTL, new);
>+
>+	/* Disable ISA interrupt routing ... */
>+	devctl = config_readb(socket, TI113X_DEVICE_CONTROL);
>+	devctl &= ~TI113X_DCR_IMODE_MASK;
>+	config_writeb(socket, TI113X_DEVICE_CONTROL, devctl);
>+	/* ... and enable PCI routing instead */
>+	sysctl = config_readb(socket, TI113X_SYSTEM_CONTROL);
>+	irqmux = config_readl(socket, TI122X_IRQMUX);
>+	irqmux = (irqmux & ~0x0f) | 0x02; /* route INTA */
>+	if (!(sysctl & TI122X_SCR_INTRTIE)) {
>+		irqmux = (irqmux & ~0xf0) | 0x20; /* route INTB */
>+	}
>+	config_writel(socket, TI122X_IRQMUX, irqmux);
>+
> 	return 0;
> }

Re: [PATCH] Making TI bridges work with kernel PCMCIA

[Pavel Roskin]

Hi, Benjamin!

> I have to use a similar workaround on pmac laptops. However, I do that
> in the platform specific quirks.

Yes, I found it in arch/ppc/kernel/pci.c, function pcibios_fixup_cardbus.

I believe it should be safe to remove that "fixup" once my patch is
applied.  There is no reason to believe that BIOS or OpenFirmware know
better than the Linux kernel how to route interrupts on PCMCIA bridges.

I believe that relying on proprietary firmware (i.e BIOS) or on power-on
defaults is only justified when we have insufficient information how to
initialize the chip.  That's not the case for TI bridges.

By the way, I'll appreciate if somebody applies my patch, removes the PPC
fixup and tests it on PPC.  I'm 99% sure that it will work, but still it
will be good to know for sure.

-- 
Regards,
Pavel Roskin