Skip to contentSkip to conversation

request_irq() and setup_irq()

DISCUSSION Christian Hildner 4 participants 5 messages Oct 21, 2004 – Oct 22, 2004

lore
Share & tools
lore
#0 lore raw

Hi,

looking through the sources of 2.4.21 there come up one question. If a device wants to register its interrupt it may call request_irq() that then calls setup_irq(). Now since SA_PERCPU_IRQ is not set by the device setup_irq() does not initialize desc->handler. Can that be fixed by adding SA_PERCPU_IRQ to irqflags in request_irq() or does that introduce problems at another place?

Christian

#1 lore raw
Quoted 3 lines · Christian Hildner

Hi,

looking through the sources of 2.4.21 there come up one question. If a

... why on earth are you looking at 2.4.21?

Signature · 6 lines
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain
#2 lore raw
In reply to Christian Hildner
Quoted 6 lines · Christian Hildner

Hi,

looking through the sources of 2.4.21 there come up one question. If a device wants to register its interrupt it may call request_irq() that then calls setup_irq(). Now since SA_PERCPU_IRQ is not set by the device setup_irq() does not initialize desc->handler.

Sorry - can you explain why this is a problem?
Are you trying to avoid the spinlock in do_IRQ() or something else?

It's been a while since I've last played with 2.4.21, but it mostly worked for the things I was testing at the time.

Quoted 3 lines

Can that be fixed by
adding SA_PERCPU_IRQ to irqflags in request_irq() or does that introduce problems at another place?

For IO SAPIC generated interrupts, that should work fine on platforms that don't implement XPR (IIRC, e.g. HP ZX1). Systems which implement XPR could redirect the interrupt to a different CPU (e.g Intel). IIRC, one can "pin" the IO SAPIC generated interrupt to a CPU despite XPR but I don't recall details or if that's already being done. You'll need to investigate this a bit more.

hth,
grant

#3 lore raw
In reply to Christian Hildner
Quoted 2 lines

-----Original Message-----
From: [email protected]

[mailto:[email protected]]

Quoted 7 lines

On Behalf Of Christian Hildner
Sent: Thursday, October 21, 2004 11:12 PM
Hi,

looking through the sources of 2.4.21 there come up one question. If a device wants to register its interrupt it may call request_irq() that then calls setup_irq(). Now since SA_PERCPU_IRQ is not set by the

device

Quoted 2 lines

setup_irq() does not initialize desc->handler. Can that be fixed by adding SA_PERCPU_IRQ to irqflags in request_irq() or does that

introduce

Quoted 3 lines

problems at another place?

Christian

ACPI can provide MADT table for lsapic and iosapic info in local system. From that table, linux will configure interrupt controller info for each interrupt line when bootup. You can look at iosapic_init(), which constructs per-line info related to iosapic. Say, by trigger mode of specific line, register_intr() will decide whether irq_type_iosapic_edge or irq_type_iosapic_level should be used.

Thanks,
Kevin

#4 lore raw
In reply to Christian Hildner
Quoted 2 lines · Kevin

ACPI can provide MADT table for lsapic and iosapic info in local system.

From that table, linux will configure interrupt controller info for each

Quoted 5 lines

interrupt line when bootup. You can look at iosapic_init(), which constructs per-line info related to iosapic. Say, by trigger mode of specific line, register_intr() will decide whether irq_type_iosapic_edge or irq_type_iosapic_level should be used.

Kevin,
thanks for your good pointers. Indeed iosapic_init() and register_intr() are the choice since they install a handler as needed. But before I had to walk through the hell of acpi.

Thanks a lot

Christian