Re: [PATCH 2/2] PowerPC: make 4xx uic use generic edge and level irq handlers
Benjamin Herrenschmidt wrote: > On Wed, 2007-11-14 at 13:13 +1100, David Gibson wrote: >> Hrm. I *think* I'm convinced this is safe, although acking in a >> callback which doesn't say it acks is rather yucky. Essentially this >> code is trading flow readability (because just reading >> handle_level_irq will tell you something other than what it does in >> our case) for smaller code size. I'm not sure if this is a good trade >> or not. >> It's not just code size. Actually, I was having problems with Ingo's real-time patch, that works on all platforms that use generic hard irq handlers and doesn't work just on 4xx since we use a custom one here. And I thought that using generic handlers would be easier to maintain. I agree that ack'ing in a callback which doesn't say it ack's looks odd, but ack'ing level-triggered interrupts is quirky on UIC itself. So, I just thought that adding a couple of quirks to mask_ack and unmask callbacks was not that bad. >> There's also one definite problem: according to the discussions I had >> with Thomas Gleixner when I wrote uic.c, handle_edge_irq is not what >> we want for edge interrupts. >> >> Apparently handle_edge_irq is only for edge interrupts on "broken" >> PICs which won't latch new interrupts while the irq is masked. UIC is >> not in this category, so handle_level_irq is actually what we want, >> even for an edge irq. >> >> Yes, I thought the naming was more than a little confusing, too. > > Hrm... handle_edge_irq works for both and you have a small performance > benefit in not masking, and thus using handle_edge_irq, so I don't > totally agree here. Basically, what handle_edge_irq() does is lazy > masking. Now there -is- an issue here is that if you do lazy masking, > you need to be able to re-emit in some convenient way. With the ack quirks added we can use handle_level_irq for edge-triggered interrupts. I'll test and resubmit the patch. > > Ben. > > Thanks, Valentine. ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
[PATCH 2/2] PowerPC: make 4xx uic use generic edge and level irq handlers
This patch makes PowerPC 4xx UIC use generic edge and level irq handlers
instead of a custom handle_uic_irq() function. Acking a level irq on UIC
has no effect if the interrupt is still asserted by the device, even if
the interrupt is already masked. So, to really de-assert the interrupt
we need to de-assert the external source first *and* ack it on UIC then.
The handle_level_irq() function masks and ack's the interrupt with mask_ack
callback prior to calling the actual ISR and unmasks it at the end.
So, to use it with UIC level interrupts we need to ack in the unmask
callback instead, after the ISR has de-asserted the external interrupt source.
Even if we ack the interrupt that we didn't handle (unmask/ack it at
the end of the handler, while next irq is already pending) it will not
de-assert the irq, untill we de-assert its exteral source.
Signed-off-by: Valentine Barshak <[EMAIL PROTECTED]>
---
arch/powerpc/sysdev/uic.c | 89 --
1 files changed, 24 insertions(+), 65 deletions(-)
--- linux-2.6/arch/powerpc/sysdev/uic.c 2007-11-13 22:28:01.0 +0300
+++ linux-2.6.o/arch/powerpc/sysdev/uic.c 2007-11-13 22:27:20.0
+0300
@@ -60,14 +60,19 @@ struct uic {
static void uic_unmask_irq(unsigned int virq)
{
+ struct irq_desc *desc = get_irq_desc(virq);
struct uic *uic = get_irq_chip_data(virq);
unsigned int src = uic_irq_to_hw(virq);
unsigned long flags;
- u32 er;
+ u32 er, sr;
+ sr = 1 << (31-src);
spin_lock_irqsave(&uic->lock, flags);
+ /* ack level interrupts here */
+ if (desc->status & IRQ_LEVEL)
+ mtdcr(uic->dcrbase + UIC_SR, sr);
er = mfdcr(uic->dcrbase + UIC_ER);
- er |= 1 << (31 - src);
+ er |= sr;
mtdcr(uic->dcrbase + UIC_ER, er);
spin_unlock_irqrestore(&uic->lock, flags);
}
@@ -99,6 +104,7 @@ static void uic_ack_irq(unsigned int vir
static void uic_mask_ack_irq(unsigned int virq)
{
+ struct irq_desc *desc = get_irq_desc(virq);
struct uic *uic = get_irq_chip_data(virq);
unsigned int src = uic_irq_to_hw(virq);
unsigned long flags;
@@ -109,7 +115,16 @@ static void uic_mask_ack_irq(unsigned in
er = mfdcr(uic->dcrbase + UIC_ER);
er &= ~sr;
mtdcr(uic->dcrbase + UIC_ER, er);
- mtdcr(uic->dcrbase + UIC_SR, sr);
+ /* on the uic, acking (i.e. clearing the sr bit)
+* a level irq will have no effect if the interrupt
+* is still asserted by the device, even if
+* the interrupt is already masked. therefore
+* we only ack the egde interrupts here, while
+* level interrupts are ack'ed after the actual
+* isr call in the uic_unmask_irq()
+*/
+ if (!(desc->status & IRQ_LEVEL))
+ mtdcr(uic->dcrbase + UIC_SR, sr);
spin_unlock_irqrestore(&uic->lock, flags);
}
@@ -156,8 +171,11 @@ static int uic_set_irq_type(unsigned int
desc->status &= ~(IRQ_TYPE_SENSE_MASK | IRQ_LEVEL);
desc->status |= flow_type & IRQ_TYPE_SENSE_MASK;
- if (!trigger)
+ if (!trigger) {
desc->status |= IRQ_LEVEL;
+ set_irq_handler(virq, handle_level_irq);
+ } else
+ set_irq_handler(virq, handle_edge_irq);
spin_unlock_irqrestore(&uic->lock, flags);
@@ -173,73 +191,14 @@ static struct irq_chip uic_irq_chip = {
.set_type = uic_set_irq_type,
};
-/**
- * handle_uic_irq - irq flow handler for UIC
- * @irq: the interrupt number
- * @desc: the interrupt description structure for this irq
- *
- * This is modified version of the generic handle_level_irq() suitable
- * for the UIC. On the UIC, acking (i.e. clearing the SR bit) a level
- * irq will have no effect if the interrupt is still asserted by the
- * device, even if the interrupt is already masked. Therefore, unlike
- * the standard handle_level_irq(), we must ack the interrupt *after*
- * invoking the ISR (which should have de-asserted the interrupt in
- * the external source). For edge interrupts we ack at the beginning
- * instead of the end, to keep the window in which we can miss an
- * interrupt as small as possible.
- */
-void fastcall handle_uic_irq(unsigned int irq, struct irq_desc *desc)
-{
- unsigned int cpu = smp_processor_id();
- struct irqaction *action;
- irqreturn_t action_ret;
-
- spin_lock(&desc->lock);
- if (desc->status & IRQ_LEVEL)
- desc->chip->mask(irq);
- else
- desc->chip->mask_ack(irq);
-
- if (unlikely(desc->status & IRQ_INPROGRESS))
- goto out_unlock;
- desc->status &= ~(IRQ_REPLAY | IRQ_WAITING);
- kstat_cpu(cpu).irqs[irq]++;
-
- /*
-* If its disabled or no action available
-* keep it masked and get out of here
-*/
- action = desc->action;
- if (unlikely(!action || (desc->status &
Re: [PATCH 2/2] PowerPC: make 4xx uic use generic edge and level irq handlers
On Wed, 2007-11-14 at 13:13 +1100, David Gibson wrote: > Hrm. I *think* I'm convinced this is safe, although acking in a > callback which doesn't say it acks is rather yucky. Essentially this > code is trading flow readability (because just reading > handle_level_irq will tell you something other than what it does in > our case) for smaller code size. I'm not sure if this is a good trade > or not. > > There's also one definite problem: according to the discussions I had > with Thomas Gleixner when I wrote uic.c, handle_edge_irq is not what > we want for edge interrupts. > > Apparently handle_edge_irq is only for edge interrupts on "broken" > PICs which won't latch new interrupts while the irq is masked. UIC is > not in this category, so handle_level_irq is actually what we want, > even for an edge irq. > > Yes, I thought the naming was more than a little confusing, too. Hrm... handle_edge_irq works for both and you have a small performance benefit in not masking, and thus using handle_edge_irq, so I don't totally agree here. Basically, what handle_edge_irq() does is lazy masking. Now there -is- an issue here is that if you do lazy masking, you need to be able to re-emit in some convenient way. Ben. ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
Re: [PATCH 2/2] PowerPC: make 4xx uic use generic edge and level irq handlers
On Tue, Nov 13, 2007 at 11:27:31PM +0300, Valentine Barshak wrote: > This patch makes PowerPC 4xx UIC use generic edge and level irq handlers > instead of a custom handle_uic_irq() function. Acking a level irq on UIC > has no effect if the interrupt is still asserted by the device, even if > the interrupt is already masked. So, to really de-assert the interrupt > we need to de-assert the external source first *and* ack it on UIC then. > The handle_level_irq() function masks and ack's the interrupt with mask_ack > callback prior to calling the actual ISR and unmasks it at the end. > So, to use it with UIC level interrupts we need to ack in the unmask > callback instead, after the ISR has de-asserted the external interrupt source. > Even if we ack the interrupt that we didn't handle (unmask/ack it at > the end of the handler, while next irq is already pending) it will not > de-assert the irq, untill we de-assert its exteral source. Hrm. I *think* I'm convinced this is safe, although acking in a callback which doesn't say it acks is rather yucky. Essentially this code is trading flow readability (because just reading handle_level_irq will tell you something other than what it does in our case) for smaller code size. I'm not sure if this is a good trade or not. There's also one definite problem: according to the discussions I had with Thomas Gleixner when I wrote uic.c, handle_edge_irq is not what we want for edge interrupts. Apparently handle_edge_irq is only for edge interrupts on "broken" PICs which won't latch new interrupts while the irq is masked. UIC is not in this category, so handle_level_irq is actually what we want, even for an edge irq. Yes, I thought the naming was more than a little confusing, too. -- David Gibson| I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson ___ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
