Prev: [tip:x86/urgent] x86, pat: Fix memory leak in free_memtype
Next: massive polling problems with udevd and other processes
From: adharmap on 26 May 2010 16:40
From: Abhijeet Dharmapurikar <adharmap(a)codeaurora.org>

The current code fails to register a handler for the same irq
without taking in to account that it could be a per cpu interrupt.
If the IRQF_PERCPU flag is set, enable the interrupt on that cpu
and return success.

Change-Id: I748b3aa08d794342ad74cbd0bb900cc599f883a6
Signed-off-by: Abhijeet Dharmapurikar <adharmap(a)codeaurora.org>
---

On systems with an interrupt controller that supports
private interrupts per core, it is not possible to call
request_irq/setup_irq from multiple cores for the same irq. This is because
the second+ invocation of __setup_irq checks if the previous
hndler had a IRQ_SHARED flag set and errors out if not.

The current irq handling code doesnt take in to account what cpu it
is executing on. Usually the local interrupt controller registers are banked
per cpu a.k.a. a cpu can enable its local interrupt by writing to its banked
registers.

One way to get around this problem is to call the setup_irq on a single cpu
while other cpus simply enable their private interrupts by writing to their
banked registers

For eg. code in arch/arm/time/smp_twd.c
/* Make sure our local interrupt controller has this enabled */
local_irq_save(flags);
get_irq_chip(clk->irq)->unmask(clk->irq);
local_irq_restore(flags);

This looks like a hacky way to get local interrupts working on
multiple cores.

The patch adds a check for PERCPU flag in __setup_irq - if an handler is
present it simply enables that interrupt for that core and returns 0.

arch/arm/Kconfig | 5 +++++
kernel/irq/manage.c | 38 +++++++++++++++++++++++++++++++-------
2 files changed, 36 insertions(+), 7 deletions(-)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index fcb79c9..ceefffd 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -979,6 +979,11 @@ config SMP

If you don't know what to do here, say N.

+config IRQ_PER_CPU
+ bool
+ depends on SMP
+ default y
+
config HAVE_ARM_SCU
bool
depends on SMP
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index bde4c66..9592a2d 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -683,6 +683,37 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
old_ptr = &desc->action;
old = *old_ptr;
if (old) {
+#if defined(CONFIG_IRQ_PER_CPU)
+ /* All handlers must agree on per-cpuness */
+ if ((old->flags & IRQF_PERCPU) !=
+ (new->flags & IRQF_PERCPU))
+ goto mismatch;
+
+ if (old->flags & IRQF_PERCPU) {
+ /* the chip must have been set for this interrupt*/
+ if (!(desc->status & IRQ_NOAUTOEN)) {
+ desc->depth = 0;
+ desc->status &= ~IRQ_DISABLED;
+ desc->chip->startup(irq);
+ } else
+ /* Undo nested disables: */
+ desc->depth = 1;
+
+ spin_unlock_irqrestore(&desc->lock, flags);
+ if (new->thread)
+ wake_up_process(new->thread);
+ return 0;
+ }
+#endif
+
+ /* they are the same types and same handler
+ * perhaps it is a private cpu interrupt
+ */
+ if (old->flags == new->flags
+ && old->handler == new->handler)
+ setup_affinity(irq, desc);
+ return 0;
+
/*
* Can't share interrupts unless both agree to and are
* the same type (level, edge, polarity). So both flag
@@ -695,13 +726,6 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
goto mismatch;
}

-#if defined(CONFIG_IRQ_PER_CPU)
- /* All handlers must agree on per-cpuness */
- if ((old->flags & IRQF_PERCPU) !=
- (new->flags & IRQF_PERCPU))
- goto mismatch;
-#endif
-
/* add new interrupt at end of irq queue */
do {
old_ptr = &old->next;
--
1.5.6.3
--
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
 | 
Pages: 1
Prev: [tip:x86/urgent] x86, pat: Fix memory leak in free_memtype
Next: massive polling problems with udevd and other processes