Skip to contentSkip to conversation

fix MVEBU GPIO driver bug causing kernel hang

PATCH New Gerlando Falauto 3 participants 9 messages Oct 2, 2013 – Oct 11, 2013 GPIO

lore
Share & tools
lore
#0 lore raw

cover fix MVEBU GPIO driver bug causing kernel hang

The MVEBU GPIO driver (like the old orion-gpio) is affected by a subtle bug, where requesting (and triggering) both EDGE- and LEVEL- based IRQs causes the kernel to hang.
Part of the reworking necessary to fix this bug (i.e. handling two separate mask registers, one per chip type) was first submitted by me, then reworked by Thomas Gleixner, then mainlined, finally landing in 3.11.

The final bits to actually fix this are still missing though. This patch will fix this.
I also marked the necessary prerequisites to apply this to linux-stable.

SIDE NOTE: If I understand correctly, a brand new driver from Sebastian Hesselbarth (drivers/irqchip/irq-orion.c) may replace this driver in the future, but it still misses the IRQ_GC_MASK_CACHE_PER_TYPE bit:

static int __init orion_bridge_irq_init(struct device_node *np,
struct device_node *parent)
{
[...]

       ret = irq_alloc_domain_generic_chips(domain, nrirqs, 1, np->name,
                            handle_level_irq, clr, 0, IRQ_GC_INIT_MASK_CACHE);

Cc: Thomas Petazzoni <[email protected]>
Cc: Grant Likely <[email protected]>
Cc: Linus Walleij <[email protected]>
Cc: Andrew Lunn <[email protected]>
Cc: Jason Cooper <[email protected]>
Cc: Gregory Clement <[email protected]>
Cc: Sebastian Hesselbarth <[email protected]>
Cc: Linus Walleij <[email protected]>
Cc: Andrew Lunn <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Cc: linux-arm-kernel at lists.infradead.org
#Cc: <[email protected]> # 3.7.x cfeaa93 genirq: Generic chip: Remove the local cur_regs() function #Cc: <[email protected]> # 3.7.x 899f0e6 genirq: Generic chip: Add support for per chip type mask cache #Cc: <[email protected]> # 3.7.x af80b0f genirq: Generic chip: Handle separate mask registers

Gerlando Falauto (1):
gpio: mvebu: enable and use IRQ_GC_MASK_CACHE_PER_TYPE

 drivers/gpio/gpio-mvebu.c | 23 ++++++++++++++---------
 1 file changed, 14 insertions(+), 9 deletions(-)
Signature · 1 line
1.8.0.1
#1 lore raw

PATCH 1/1 gpio: mvebu: enable and use IRQ_GC_MASK_CACHE_PER_TYPE

Since we have now introduced mask_cache within irq_chip_type to also handle per-chip-type mask registers, convert gpio-mvebu driver to use this new pointer.

Also enable IRQ_GC_MASK_CACHE_PER_TYPE to actually handle separate mask registers for all three SoC variants handled by this driver.

This wll fix a bug where requesting (and triggering) both EDGE- and LEVEL- based IRQs causes the kernel to hang.

Cc Simon Guinot <[email protected]>
Cc Thomas Petazzoni <[email protected]>
Cc Grant Likely <[email protected]>
Cc Linus Walleij <[email protected]>
Cc Andrew Lunn <[email protected]>
Cc Jason Cooper <[email protected]>
Cc Gregory Clement <[email protected]>
Cc Sebastian Hesselbarth <[email protected]>
Cc Linus Walleij <[email protected]>
Cc Andrew Lunn <[email protected]>
Cc Arnd Bergmann <[email protected]>
Cc linux-arm-kernel at lists.infradead.org
---
 drivers/gpio/gpio-mvebu.c | 23 ++++++++++++++---------
 1 file changed, 14 insertions(+), 9 deletions(-)
drivers/gpio/gpio-mvebu.c +14 -9
index 3a4816a..ce425b3 100644
@@ -303,48 +303,52 @@ static void mvebu_gpio_irq_ack(struct irq_data *d)
 static void mvebu_gpio_edge_irq_mask(struct irq_data *d)
 {
 	struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d);
+	struct irq_chip_type *ct = irq_data_get_chip_type(d);
 	struct mvebu_gpio_chip *mvchip = gc->private;
 	u32 mask = 1 << (d->irq - gc->irq_base);
 
 	irq_gc_lock(gc);
-	gc->mask_cache &= ~mask;
-	writel_relaxed(gc->mask_cache, mvebu_gpioreg_edge_mask(mvchip));
+	*ct->mask_cache &= ~mask;
+	writel_relaxed(*ct->mask_cache, mvebu_gpioreg_edge_mask(mvchip));
 	irq_gc_unlock(gc);
 }
 
 static void mvebu_gpio_edge_irq_unmask(struct irq_data *d)
 {
 	struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d);
+	struct irq_chip_type *ct = irq_data_get_chip_type(d);
 	struct mvebu_gpio_chip *mvchip = gc->private;
 	u32 mask = 1 << (d->irq - gc->irq_base);
 
 	irq_gc_lock(gc);
-	gc->mask_cache |= mask;
-	writel_relaxed(gc->mask_cache, mvebu_gpioreg_edge_mask(mvchip));
+	*ct->mask_cache |= mask;
+	writel_relaxed(*ct->mask_cache, mvebu_gpioreg_edge_mask(mvchip));
 	irq_gc_unlock(gc);
 }
 
 static void mvebu_gpio_level_irq_mask(struct irq_data *d)
 {
 	struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d);
+	struct irq_chip_type *ct = irq_data_get_chip_type(d);
 	struct mvebu_gpio_chip *mvchip = gc->private;
 	u32 mask = 1 << (d->irq - gc->irq_base);
 
 	irq_gc_lock(gc);
-	gc->mask_cache &= ~mask;
-	writel_relaxed(gc->mask_cache, mvebu_gpioreg_level_mask(mvchip));
+	*ct->mask_cache &= ~mask;
+	writel_relaxed(*ct->mask_cache, mvebu_gpioreg_level_mask(mvchip));
 	irq_gc_unlock(gc);
 }
 
 static void mvebu_gpio_level_irq_unmask(struct irq_data *d)
 {
 	struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d);
+	struct irq_chip_type *ct = irq_data_get_chip_type(d);
 	struct mvebu_gpio_chip *mvchip = gc->private;
 	u32 mask = 1 << (d->irq - gc->irq_base);
 
 	irq_gc_lock(gc);
-	gc->mask_cache |= mask;
-	writel_relaxed(gc->mask_cache, mvebu_gpioreg_level_mask(mvchip));
+	*ct->mask_cache |= mask;
+	writel_relaxed(*ct->mask_cache, mvebu_gpioreg_level_mask(mvchip));
 	irq_gc_unlock(gc);
 }
 
@@ -708,7 +712,8 @@ static int mvebu_gpio_probe(struct platform_device *pdev)
 	ct->handler = handle_edge_irq;
 	ct->chip.name = mvchip->chip.label;
 
-	irq_setup_generic_chip(gc, IRQ_MSK(ngpios), 0,
+	irq_setup_generic_chip(gc, IRQ_MSK(ngpios),
+			       IRQ_GC_MASK_CACHE_PER_TYPE,
 			       IRQ_NOREQUEST, IRQ_LEVEL | IRQ_NOPROBE);
 
 	/* Setup irq domain on top of the generic chip. */
Signature · 1 line
1.8.0.1
#2 lore raw

PATCH 1/1 gpio: mvebu: enable and use IRQ_GC_MASK_CACHE_PER_TYPE

Quoted 10 lines · Gerlando Falauto

Since we have now introduced mask_cache within irq_chip_type to also handle per-chip-type mask registers, convert gpio-mvebu driver to use this new pointer.

Also enable IRQ_GC_MASK_CACHE_PER_TYPE to actually handle separate mask registers for all three SoC variants handled by this driver.

This wll fix a bug where requesting (and triggering) both EDGE- and LEVEL- based IRQs causes the kernel to hang.

Quoted 13 lines
Cc Simon Guinot <[email protected]>
Cc Thomas Petazzoni <[email protected]>
Cc Grant Likely <[email protected]>
Cc Linus Walleij <[email protected]>
Cc Andrew Lunn <[email protected]>
Cc Jason Cooper <[email protected]>
Cc Gregory Clement <[email protected]>
Cc Sebastian Hesselbarth <[email protected]>
Cc Linus Walleij <[email protected]>
Cc Andrew Lunn <[email protected]>
Cc Arnd Bergmann <[email protected]>
Cc linux-arm-kernel at lists.infradead.org

Thomas/Gregory: can you comment on this patch?

Yours,
Linus Walleij

#3 lore raw

cover fix MVEBU GPIO driver bug causing kernel hang

In reply to Gerlando Falauto

Dear Gerlando Falauto,

Quoted 3 lines · Gerlando Falauto

SIDE NOTE: If I understand correctly, a brand new driver from Sebastian Hesselbarth (drivers/irqchip/irq-orion.c) may replace this driver in the future, but it still misses the IRQ_GC_MASK_CACHE_PER_TYPE bit:

Hum? The driver you're touching is a GPIO driver, while drivers/irqchip/irq-orion.c is an interrupt controller driver. The gpio-mvebu driver can be used on all mvebu platforms that are DT capable, but the main interrupt controller is different from one platform to another: we already have irq-armada-370-xp.c for Armada 370/XP, while Sebastian's irq-orion.c is for older families of SoCs.

Or maybe I'm missing something?

Thanks!

Thomas

Signature · 3 lines
Thomas Petazzoni, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#4 lore raw

cover fix MVEBU GPIO driver bug causing kernel hang

Hi Thomas,

Quoted 8 lines · Thomas Petazzoni

Dear Gerlando Falauto,

Quoted 3 lines · Gerlando Falauto

SIDE NOTE: If I understand correctly, a brand new driver from Sebastian Hesselbarth (drivers/irqchip/irq-orion.c) may replace this driver in the future, but it still misses the IRQ_GC_MASK_CACHE_PER_TYPE bit:

Quoted 6 lines

Hum? The driver you're touching is a GPIO driver, while drivers/irqchip/irq-orion.c is an interrupt controller driver. The gpio-mvebu driver can be used on all mvebu platforms that are DT capable, but the main interrupt controller is different from one platform to another: we already have irq-armada-370-xp.c for Armada 370/XP, while Sebastian's irq-orion.c is for older families of SoCs.

My foreword was "if I understand correctly", which I clearly didn't. ;-)

OK, so Sebastian's rework irq-orion.c is about the MAIN interrupt controller (replaces arch/arm/plat-orion/irq.c), which also cascades interrupts from GPIOs.
These are then either handled by gpio-mvebu.c (for all platforms which are DT capable) or by arch/arm/plat-orion/gpio.c (legacy). Is that right?

Quoted 1 line

Or maybe I'm missing something?

Sorry, I'm the one missing something here... perhaps too much.

Thanks!
Gerlando

#5 lore raw

cover fix MVEBU GPIO driver bug causing kernel hang

Dear Gerlando Falauto,

Quoted 7 lines · Gerlando Falauto
Quoted 6 lines

Hum? The driver you're touching is a GPIO driver, while drivers/irqchip/irq-orion.c is an interrupt controller driver. The gpio-mvebu driver can be used on all mvebu platforms that are DT capable, but the main interrupt controller is different from one platform to another: we already have irq-armada-370-xp.c for Armada 370/XP, while Sebastian's irq-orion.c is for older families of SoCs.

Quoted 8 lines

My foreword was "if I understand correctly", which I clearly didn't. ;-)

OK, so Sebastian's rework irq-orion.c is about the MAIN interrupt controller (replaces arch/arm/plat-orion/irq.c), which also cascades interrupts from GPIOs.
These are then either handled by gpio-mvebu.c (for all platforms which are DT capable) or by arch/arm/plat-orion/gpio.c (legacy). Is that right?

Yes, that's right.

Thomas

Signature · 3 lines
Thomas Petazzoni, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#6 lore raw

cover fix MVEBU GPIO driver bug causing kernel hang

In reply to Gerlando Falauto

OK I see these comments on patch [0/1] now (how confusing) forget what I just said. Patch is dropped.

Yours,
Linus Walleij

#7 lore raw

cover fix MVEBU GPIO driver bug causing kernel hang

Hi LinusW,

what do you mean?
My statement on patch [0/1] contained speculation about a driver which is clearly unrelated (the main, generic, orion IRQ controller, which is *NOT* what this patch is about). Our discussion with Thomas was about this and him explaining how I got the two things confused in my mind. But the patch, in its real context (the GPIO chip and its interrupt handling) should still be sound and needed...
[Patch 0/1] will not be in the commit message anyway...

Thanks,
Gerlando

Quoted 6 lines · Linus Walleij

OK I see these comments on patch [0/1] now (how confusing) forget what I just said. Patch is dropped.

Yours,
Linus Walleij

#8 lore raw

cover fix MVEBU GPIO driver bug causing kernel hang

Quoted 3 lines · Gerlando Falauto

My statement on patch [0/1] contained speculation about a driver which is clearly unrelated (the main, generic, orion IRQ controller, which is *NOT* what this patch is about).

OK I got it all wrong then, I'll wait for an ACK on patch 1/1 before applying though.

Yours,
Linus Walleij