Skip to contentSkip to conversation

bc47xx : fix ssb irq setup

PATCH In review matthieu castet 3 participants 6 messages May 18, 2009 – Jun 11, 2009

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bc47xx : fix ssb irq setup

Hi,

the current ssb irq setup (in ssb_mipscore_init) have some problem :
it configure some device on some irq without checking that the irq is not taken by an other device.

For example in my case PCI host is on irq 0 and IPSEC on irq 3. The current code :
- store in dev->irq that IPSEC irq is 3+2
- do a set_irq 0->3 on PCI host

But now IPSEC irq is not routed anymore to the mips code and dev->irq is wrong. This cause problem described in [1].

This patch try to solve the problem by making set_irq configure the device we want to take the irq on the shared irq0. The previous example become :
- store in dev->irq that IPSEC irq is 3+2
- do a set_irq 0->3 on PCI host :
- irq 3 is already taken by IPSEC. do a set_irq 3->0 on IPSEC

I also added some code to print the irq configuration before and after irq setup to allow easier debugging. And I add extra checking in ssb_mips_irq to report device without irq or device with not routed irq.

[1] http://www.danm.de/files/src/bcm5365p/REPORTED_DEVICES

drivers/ssb/driver_mipscore.c +76 -10
index 3fd3e3b..6acf43f 100644
@@ -49,29 +49,54 @@ static const u32 ipsflag_irq_shift[] = {
 
 static inline u32 ssb_irqflag(struct ssb_device *dev)
 {
-	return ssb_read32(dev, SSB_TPSFLAG) & SSB_TPSFLAG_BPFLAG;
+	u32 tpsflag = ssb_read32(dev, SSB_TPSFLAG);
+	if (tpsflag)
+		return ssb_read32(dev, SSB_TPSFLAG) & SSB_TPSFLAG_BPFLAG;
+	else
+		/* not irq supported */
+		return 0x3f;
+}
+
+static struct ssb_device *find_device(struct ssb_device *rdev, int irqflag)
+{
+	struct ssb_bus *bus = rdev->bus;
+	int i;
+	for (i = 0; i < bus->nr_devices; i++) {
+		struct ssb_device *dev;
+		dev = &(bus->devices[i]);
+		if (ssb_irqflag(dev) == irqflag)
+			return dev;
+	}
+	return NULL;
 }
 
 /* Get the MIPS IRQ assignment for a specified device.
  * If unassigned, 0 is returned.
+ * If disabled, 5 is returned.
+ * If not supported, 6 is returned.
  */
 unsigned int ssb_mips_irq(struct ssb_device *dev)
 {
 	struct ssb_bus *bus = dev->bus;
+	struct ssb_device *mdev = bus->mipscore.dev;
 	u32 irqflag;
 	u32 ipsflag;
 	u32 tmp;
 	unsigned int irq;
 
 	irqflag = ssb_irqflag(dev);
+	if (irqflag == 0x3f)
+		return 6;
 	ipsflag = ssb_read32(bus->mipscore.dev, SSB_IPSFLAG);
 	for (irq = 1; irq <= 4; irq++) {
 		tmp = ((ipsflag & ipsflag_irq_mask[irq]) >> ipsflag_irq_shift[irq]);
 		if (tmp == irqflag)
 			break;
 	}
-	if (irq	== 5)
-		irq = 0;
+	if (irq	== 5) {
+		if ((1 << irqflag) & ssb_read32(mdev, SSB_INTVEC))
+			irq = 0;
+	}
 
 	return irq;
 }
@@ -97,25 +122,56 @@ static void set_irq(struct ssb_device *dev, unsigned int irq)
 	struct ssb_device *mdev = bus->mipscore.dev;
 	u32 irqflag = ssb_irqflag(dev);
 
+	BUG_ON(oldirq == 6);
+
 	dev->irq = irq + 2;
 
-	ssb_dprintk(KERN_INFO PFX
-		    "set_irq: core 0x%04x, irq %d => %d\n",
-		    dev->id.coreid, oldirq, irq);
 	/* clear the old irq */
 	if (oldirq == 0)
 		ssb_write32(mdev, SSB_INTVEC, (~(1 << irqflag) & ssb_read32(mdev, SSB_INTVEC)));
-	else
+	else if (oldirq != 5)
 		clear_irq(bus, oldirq);
 
 	/* assign the new one */
 	if (irq == 0) {
 		ssb_write32(mdev, SSB_INTVEC, ((1 << irqflag) | ssb_read32(mdev, SSB_INTVEC)));
 	} else {
+		u32 ipsflag = ssb_read32(mdev, SSB_IPSFLAG);
+		if ((ipsflag & ipsflag_irq_mask[irq]) != ipsflag_irq_mask[irq]) {
+			u32 oldipsflag = (ipsflag & ipsflag_irq_mask[irq]) >> ipsflag_irq_shift[irq];
+			struct ssb_device *olddev = find_device(dev, oldipsflag);
+			if (olddev)
+				set_irq(olddev, 0);
+		}
 		irqflag <<= ipsflag_irq_shift[irq];
-		irqflag |= (ssb_read32(mdev, SSB_IPSFLAG) & ~ipsflag_irq_mask[irq]);
+		irqflag |= (ipsflag & ~ipsflag_irq_mask[irq]);
 		ssb_write32(mdev, SSB_IPSFLAG, irqflag);
 	}
+	ssb_dprintk(KERN_INFO PFX
+		    "set_irq: core 0x%04x, irq %d => %d\n",
+		    dev->id.coreid, oldirq+2, irq+2);
+}
+
+static void print_irq(struct ssb_device *dev, unsigned int irq)
+{
+	int i;
+	static const char *irq_name[] = {"2(S)", "3", "4", "5", "6", "D", "I"};
+	ssb_dprintk(KERN_INFO PFX
+		"core 0x%04x, irq :", dev->id.coreid);
+	for (i = 0; i <= 6; i++) {
+		ssb_dprintk(" %s%s", irq_name[i], i==irq?"*":" ");
+	}
+	ssb_dprintk("\n");
+}
+
+static void dump_irq(struct ssb_bus *bus)
+{
+	int i;
+	for (i = 0; i < bus->nr_devices; i++) {
+		struct ssb_device *dev;
+		dev = &(bus->devices[i]);
+		print_irq(dev, ssb_mips_irq(dev));
+	}
 }
 
 static void ssb_mips_serial_init(struct ssb_mipscore *mcore)
@@ -195,18 +251,26 @@ void ssb_mipscore_init(struct ssb_mipscore *mcore)
 	else if (bus->chipco.dev)
 		ssb_chipco_timing_init(&bus->chipco, ns);
 
+	dump_irq(bus);
 	/* Assign IRQs to all cores on the bus, start with irq line 2, because serial usually takes 1 */
 	for (irq = 2, i = 0; i < bus->nr_devices; i++) {
+		int mips_irq;
 		dev = &(bus->devices[i]);
-		dev->irq = ssb_mips_irq(dev) + 2;
+		mips_irq = ssb_mips_irq(dev);
+		if (mips_irq > 4)
+			dev->irq = 0;
+		else
+			dev->irq = mips_irq + 2;
+		if (dev->irq > 5)
+			continue;
 		switch (dev->id.coreid) {
 		case SSB_DEV_USB11_HOST:
 			/* shouldn't need a separate irq line for non-4710, most of them have a proper
 			 * external usb controller on the pci */
 			if ((bus->chip_id == 0x4710) && (irq <= 4)) {
 				set_irq(dev, irq++);
-				break;
 			}
+			break;
 			/* fallthrough */
 		case SSB_DEV_PCI:
 		case SSB_DEV_ETHERNET:
@@ -220,6 +284,8 @@ void ssb_mipscore_init(struct ssb_mipscore *mcore)
 			}
 		}
 	}
+	ssb_dprintk(KERN_INFO PFX "after irq reconfiguration\n");
+	dump_irq(bus);
 
 	ssb_mips_serial_init(mcore);
 	ssb_mips_flash_detect(mcore);
#1 lore raw
Quoted 22 lines · On Tuesday 19 May 2009 00:06:12 matthieu castet wrote:

Hi,

the current ssb irq setup (in ssb_mipscore_init) have some problem :
it configure some device on some irq without checking that the irq is not taken by an other device.

For example in my case PCI host is on irq 0 and IPSEC on irq 3. The current code :
- store in dev->irq that IPSEC irq is 3+2
- do a set_irq 0->3 on PCI host

But now IPSEC irq is not routed anymore to the mips code and dev->irq is wrong. This cause problem described in [1].

This patch try to solve the problem by making set_irq configure the device we want to take the irq on the shared irq0.
The previous example become :
- store in dev->irq that IPSEC irq is 3+2
- do a set_irq 0->3 on PCI host :
- irq 3 is already taken by IPSEC. do a set_irq 3->0 on IPSEC

Quoted 9 lines

I also added some code to print the irq configuration before and after irq setup to allow easier debugging. And I add extra checking in ssb_mips_irq to report device without irq or device with not routed irq.

[1] http://www.danm.de/files/src/bcm5365p/REPORTED_DEVICES

If this works on all devices, I'm OK with this. Please submit to [email protected] You can add my ack.

Signature · 1 line
Greetings, Michael.
#2 lore raw
Quoted 4 lines · Michael Buesch
Quoted 3 lines · On Tuesday 19 May 2009 00:06:12 matthieu castet wrote:

Hi,

Quoted 8 lines
Quoted 4 lines

[1] http://www.danm.de/files/src/bcm5365p/REPORTED_DEVICES

If this works on all devices, I'm OK with this. Please submit to [email protected] You can add my ack.

Well I have only a wl500gd.
I have submit it on openwrt project in order to test in more devices.

Matthieu

#3 lore raw

Hi Matthieu, Michael,

Le Monday 25 May 2009 19:55:58 matthieu castet, vous avez écrit :

Quoted 7 lines
Quoted 6 lines · Michael Buesch
Quoted 5 lines · On Tuesday 19 May 2009 00:06:12 matthieu castet wrote:

Hi,

[1] http://www.danm.de/files/src/bcm5365p/REPORTED_DEVICES

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

If this works on all devices, I'm OK with this. Please submit to [email protected] You can add my ack.

Well I have only a wl500gd.
I have submit it on openwrt project in order to test in more devices.

It makes the IPsec core work on my Netgear WGT634U and I did not see any regression on a Linksys WRT54GS.

#4 lore raw

Le Thursday 11 June 2009 09:59:36 Florian Fainelli, vous avez écrit :

Quoted 15 lines

Hi Matthieu, Michael,

Le Monday 25 May 2009 19:55:58 matthieu castet, vous avez écrit :

Quoted 12 lines
Quoted 10 lines · Michael Buesch
Quoted 6 lines · On Tuesday 19 May 2009 00:06:12 matthieu castet wrote:

Hi,

[1] http://www.danm.de/files/src/bcm5365p/REPORTED_DEVICES

If this works on all devices, I'm OK with this. Please submit to [email protected] You can add my ack.

Quoted 7 lines
Quoted 2 lines

Well I have only a wl500gd.
I have submit it on openwrt project in order to test in more devices.

It makes the IPsec core work on my Netgear WGT634U and I did not see any regression on a Linksys WRT54GS.

One minor thing, please remove the dump_irqs call, it is convenient for debugging to print the IRQ routing, but I find it a little too verbose for production. This can be a follow-up patch if you prefer not to respin it.

Signature · 4 lines
Best regards, Florian Fainelli
Email : [email protected]
http://openwrt.org
-------------------------------
#5 lore raw

Quoting Florian Fainelli <[email protected]>:

Quoted 19 lines

Le Thursday 11 June 2009 09:59:36 Florian Fainelli, vous avez écrit :

Quoted 18 lines

Hi Matthieu, Michael,

Le Monday 25 May 2009 19:55:58 matthieu castet, vous avez écrit :

Quoted 14 lines
Quoted 10 lines · Michael Buesch
Quoted 6 lines · On Tuesday 19 May 2009 00:06:12 matthieu castet wrote:

Hi,

[1] http://www.danm.de/files/src/bcm5365p/REPORTED_DEVICES

If this works on all devices, I'm OK with this. Please submit to [email protected] You can add my ack.

Well I have only a wl500gd.
I have submit it on openwrt project in order to test in more devices.

Quoted 8 lines
Quoted 4 lines

It makes the IPsec core work on my Netgear WGT634U and I did not see any regression on a Linksys WRT54GS.

One minor thing, please remove the dump_irqs call, it is convenient for debugging to print the IRQ routing, but I find it a little too verbose for production. This can be a follow-up patch if you prefer not to respin it.

Well all x86 does something similar, with acpi dumping the pci interrupt Routing Table even in production kernel.
But it can removed if people don't want it.

Matthieu