* [PATCH 1/2] genirq: Abstract access to irq_chip flags
2015-01-08 17:32 [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains Marc Zyngier
@ 2015-01-08 17:32 ` Marc Zyngier
2015-01-09 3:00 ` Jiang Liu
2015-01-08 17:32 ` [PATCH 2/2] genirq: Allow irq_desc to carry the union of stacked " Marc Zyngier
2015-01-09 3:07 ` [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains Jiang Liu
2 siblings, 1 reply; 9+ messages in thread
From: Marc Zyngier @ 2015-01-08 17:32 UTC (permalink / raw)
To: Thomas Gleixner, Jiang Liu; +Cc: linux-kernel
In order to safely migrate to a cumulative set of flags, start by
abstracting the way we look at these flags. There is otherwise no
change in semantics here.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
---
include/linux/irqdesc.h | 5 +++++
kernel/irq/chip.c | 10 +++++-----
kernel/irq/manage.c | 6 +++---
kernel/irq/pm.c | 2 +-
4 files changed, 14 insertions(+), 9 deletions(-)
diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
index faf433a..32d9fff 100644
--- a/include/linux/irqdesc.h
+++ b/include/linux/irqdesc.h
@@ -107,6 +107,11 @@ static inline void *irq_desc_get_chip_data(struct irq_desc *desc)
return desc->irq_data.chip_data;
}
+static inline unsigned long irq_desc_get_chip_flags(struct irq_desc *desc)
+{
+ return desc->irq_data.chip->flags;
+}
+
static inline void *irq_desc_get_handler_data(struct irq_desc *desc)
{
return desc->irq_data.handler_data;
diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
index 6f1c7a5..d8ffb36 100644
--- a/kernel/irq/chip.c
+++ b/kernel/irq/chip.c
@@ -287,7 +287,7 @@ void unmask_threaded_irq(struct irq_desc *desc)
{
struct irq_chip *chip = desc->irq_data.chip;
- if (chip->flags & IRQCHIP_EOI_THREADED)
+ if (irq_desc_get_chip_flags(desc) & IRQCHIP_EOI_THREADED)
chip->irq_eoi(&desc->irq_data);
if (chip->irq_unmask) {
@@ -489,7 +489,7 @@ static void cond_unmask_eoi_irq(struct irq_desc *desc, struct irq_chip *chip)
irqd_irq_masked(&desc->irq_data) && !desc->threads_oneshot) {
chip->irq_eoi(&desc->irq_data);
unmask_irq(desc);
- } else if (!(chip->flags & IRQCHIP_EOI_THREADED)) {
+ } else if (!(irq_desc_get_chip_flags(desc) & IRQCHIP_EOI_THREADED)) {
chip->irq_eoi(&desc->irq_data);
}
}
@@ -538,7 +538,7 @@ handle_fasteoi_irq(unsigned int irq, struct irq_desc *desc)
raw_spin_unlock(&desc->lock);
return;
out:
- if (!(chip->flags & IRQCHIP_EOI_IF_HANDLED))
+ if (!(irq_desc_get_chip_flags(desc) & IRQCHIP_EOI_IF_HANDLED))
chip->irq_eoi(&desc->irq_data);
raw_spin_unlock(&desc->lock);
}
@@ -836,7 +836,7 @@ void irq_cpu_online(void)
chip = irq_data_get_irq_chip(&desc->irq_data);
if (chip && chip->irq_cpu_online &&
- (!(chip->flags & IRQCHIP_ONOFFLINE_ENABLED) ||
+ (!(irq_desc_get_chip_flags(desc) & IRQCHIP_ONOFFLINE_ENABLED) ||
!irqd_irq_disabled(&desc->irq_data)))
chip->irq_cpu_online(&desc->irq_data);
@@ -866,7 +866,7 @@ void irq_cpu_offline(void)
chip = irq_data_get_irq_chip(&desc->irq_data);
if (chip && chip->irq_cpu_offline &&
- (!(chip->flags & IRQCHIP_ONOFFLINE_ENABLED) ||
+ (!(irq_desc_get_chip_flags(desc) & IRQCHIP_ONOFFLINE_ENABLED) ||
!irqd_irq_disabled(&desc->irq_data)))
chip->irq_cpu_offline(&desc->irq_data);
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index 8069237..b2a43e0 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -491,7 +491,7 @@ static int set_irq_wake_real(unsigned int irq, unsigned int on)
struct irq_desc *desc = irq_to_desc(irq);
int ret = -ENXIO;
- if (irq_desc_get_chip(desc)->flags & IRQCHIP_SKIP_SET_WAKE)
+ if (irq_desc_get_chip_flags(desc) & IRQCHIP_SKIP_SET_WAKE)
return 0;
if (desc->irq_data.chip->irq_set_wake)
@@ -589,7 +589,7 @@ int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
flags &= IRQ_TYPE_SENSE_MASK;
- if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
+ if (irq_desc_get_chip_flags(desc) & IRQCHIP_SET_TYPE_MASKED) {
if (!irqd_irq_masked(&desc->irq_data))
mask_irq(desc);
if (!irqd_irq_disabled(&desc->irq_data))
@@ -1043,7 +1043,7 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
* chip flags, so we can avoid the unmask dance at the end of
* the threaded handler for those.
*/
- if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
+ if (irq_desc_get_chip_flags(desc) & IRQCHIP_ONESHOT_SAFE)
new->flags &= ~IRQF_ONESHOT;
/*
diff --git a/kernel/irq/pm.c b/kernel/irq/pm.c
index 3ca5325..8ed029d 100644
--- a/kernel/irq/pm.c
+++ b/kernel/irq/pm.c
@@ -88,7 +88,7 @@ static bool suspend_device_irq(struct irq_desc *desc, int irq)
* chip level. The chip implementation indicates that with
* IRQCHIP_MASK_ON_SUSPEND.
*/
- if (irq_desc_get_chip(desc)->flags & IRQCHIP_MASK_ON_SUSPEND)
+ if (irq_desc_get_chip_flags(desc) & IRQCHIP_MASK_ON_SUSPEND)
mask_irq(desc);
return true;
}
--
2.1.4
^ permalink raw reply related [flat|nested] 9+ messages in thread* Re: [PATCH 1/2] genirq: Abstract access to irq_chip flags
2015-01-08 17:32 ` [PATCH 1/2] genirq: Abstract access to irq_chip flags Marc Zyngier
@ 2015-01-09 3:00 ` Jiang Liu
2015-01-09 9:43 ` Marc Zyngier
0 siblings, 1 reply; 9+ messages in thread
From: Jiang Liu @ 2015-01-09 3:00 UTC (permalink / raw)
To: Marc Zyngier, Thomas Gleixner; +Cc: linux-kernel
On 2015/1/9 1:32, Marc Zyngier wrote:
> In order to safely migrate to a cumulative set of flags, start by
> abstracting the way we look at these flags. There is otherwise no
> change in semantics here.
<snit>
> diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> index 8069237..b2a43e0 100644
> --- a/kernel/irq/manage.c
> +++ b/kernel/irq/manage.c
> @@ -491,7 +491,7 @@ static int set_irq_wake_real(unsigned int irq, unsigned int on)
> struct irq_desc *desc = irq_to_desc(irq);
> int ret = -ENXIO;
>
> - if (irq_desc_get_chip(desc)->flags & IRQCHIP_SKIP_SET_WAKE)
> + if (irq_desc_get_chip_flags(desc) & IRQCHIP_SKIP_SET_WAKE)
> return 0;
>
> if (desc->irq_data.chip->irq_set_wake)
> @@ -589,7 +589,7 @@ int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
>
> flags &= IRQ_TYPE_SENSE_MASK;
>
> - if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
> + if (irq_desc_get_chip_flags(desc) & IRQCHIP_SET_TYPE_MASKED) {
> if (!irqd_irq_masked(&desc->irq_data))
> mask_irq(desc);
> if (!irqd_irq_disabled(&desc->irq_data))
> @@ -1043,7 +1043,7 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
> * chip flags, so we can avoid the unmask dance at the end of
> * the threaded handler for those.
> */
> - if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
> + if (irq_desc_get_chip_flags(desc) & IRQCHIP_ONESHOT_SAFE)
> new->flags &= ~IRQF_ONESHOT;
>
> /*
Hi Mark,
Seems you missed on instance of IRQCHIP_ONESHOT_SAFE in
__setup_irq() as below.
Thanks!
Gerry
-----------------------------------------------------------------
} else if (new->handler == irq_default_primary_handler &&
!(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
------------------------------------------------------------------
> diff --git a/kernel/irq/pm.c b/kernel/irq/pm.c
> index 3ca5325..8ed029d 100644
> --- a/kernel/irq/pm.c
> +++ b/kernel/irq/pm.c
> @@ -88,7 +88,7 @@ static bool suspend_device_irq(struct irq_desc *desc, int irq)
> * chip level. The chip implementation indicates that with
> * IRQCHIP_MASK_ON_SUSPEND.
> */
> - if (irq_desc_get_chip(desc)->flags & IRQCHIP_MASK_ON_SUSPEND)
> + if (irq_desc_get_chip_flags(desc) & IRQCHIP_MASK_ON_SUSPEND)
> mask_irq(desc);
> return true;
> }
>
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH 1/2] genirq: Abstract access to irq_chip flags
2015-01-09 3:00 ` Jiang Liu
@ 2015-01-09 9:43 ` Marc Zyngier
0 siblings, 0 replies; 9+ messages in thread
From: Marc Zyngier @ 2015-01-09 9:43 UTC (permalink / raw)
To: Jiang Liu, Thomas Gleixner; +Cc: linux-kernel@vger.kernel.org
On 09/01/15 03:00, Jiang Liu wrote:
> On 2015/1/9 1:32, Marc Zyngier wrote:
>> In order to safely migrate to a cumulative set of flags, start by
>> abstracting the way we look at these flags. There is otherwise no
>> change in semantics here.
> <snit>
>> diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
>> index 8069237..b2a43e0 100644
>> --- a/kernel/irq/manage.c
>> +++ b/kernel/irq/manage.c
>> @@ -491,7 +491,7 @@ static int set_irq_wake_real(unsigned int irq, unsigned int on)
>> struct irq_desc *desc = irq_to_desc(irq);
>> int ret = -ENXIO;
>>
>> - if (irq_desc_get_chip(desc)->flags & IRQCHIP_SKIP_SET_WAKE)
>> + if (irq_desc_get_chip_flags(desc) & IRQCHIP_SKIP_SET_WAKE)
>> return 0;
>>
>> if (desc->irq_data.chip->irq_set_wake)
>> @@ -589,7 +589,7 @@ int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
>>
>> flags &= IRQ_TYPE_SENSE_MASK;
>>
>> - if (chip->flags & IRQCHIP_SET_TYPE_MASKED) {
>> + if (irq_desc_get_chip_flags(desc) & IRQCHIP_SET_TYPE_MASKED) {
>> if (!irqd_irq_masked(&desc->irq_data))
>> mask_irq(desc);
>> if (!irqd_irq_disabled(&desc->irq_data))
>> @@ -1043,7 +1043,7 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
>> * chip flags, so we can avoid the unmask dance at the end of
>> * the threaded handler for those.
>> */
>> - if (desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)
>> + if (irq_desc_get_chip_flags(desc) & IRQCHIP_ONESHOT_SAFE)
>> new->flags &= ~IRQF_ONESHOT;
>>
>> /*
> Hi Mark,
> Seems you missed on instance of IRQCHIP_ONESHOT_SAFE in
> __setup_irq() as below.
> Thanks!
> Gerry
> -----------------------------------------------------------------
> } else if (new->handler == irq_default_primary_handler &&
> !(desc->irq_data.chip->flags & IRQCHIP_ONESHOT_SAFE)) {
> ------------------------------------------------------------------
Ah, nice catch.
Thanks,
M.
--
Jazz is not dead. It just smells funny...
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH 2/2] genirq: Allow irq_desc to carry the union of stacked irq_chip flags
2015-01-08 17:32 [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains Marc Zyngier
2015-01-08 17:32 ` [PATCH 1/2] genirq: Abstract access to irq_chip flags Marc Zyngier
@ 2015-01-08 17:32 ` Marc Zyngier
2015-01-09 3:07 ` [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains Jiang Liu
2 siblings, 0 replies; 9+ messages in thread
From: Marc Zyngier @ 2015-01-08 17:32 UTC (permalink / raw)
To: Thomas Gleixner, Jiang Liu; +Cc: linux-kernel
The current infrastructure for stacked domains doesn't propagate
irq_chip flags, and as we only look at the top-level irq_chip,
we may miss a number of critical flags.
This patch accumulates the flags into a new set, stored at the
irq_desc level, with an additional flag to indicate that this is
a stack of irqchip. The accessor is updated to return the right one.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
---
include/linux/irq.h | 4 ++++
include/linux/irqdesc.h | 7 +++++++
kernel/irq/irqdesc.c | 3 +++
kernel/irq/irqdomain.c | 20 +++++++++++++++++++-
4 files changed, 33 insertions(+), 1 deletion(-)
diff --git a/include/linux/irq.h b/include/linux/irq.h
index d09ec7a..cb956f6 100644
--- a/include/linux/irq.h
+++ b/include/linux/irq.h
@@ -377,6 +377,7 @@ struct irq_chip {
* IRQCHIP_SKIP_SET_WAKE: Skip chip.irq_set_wake(), for this irq chip
* IRQCHIP_ONESHOT_SAFE: One shot does not require mask/unmask
* IRQCHIP_EOI_THREADED: Chip requires eoi() on unmask in threaded mode
+ * IRQCHIP_STACKED_CHIPS: Pseudo flag for stacked irq chips
*/
enum {
IRQCHIP_SET_TYPE_MASKED = (1 << 0),
@@ -386,6 +387,9 @@ enum {
IRQCHIP_SKIP_SET_WAKE = (1 << 4),
IRQCHIP_ONESHOT_SAFE = (1 << 5),
IRQCHIP_EOI_THREADED = (1 << 6),
+
+ /* The following is only valid as part of irq_desc.chip_flags */
+ IRQCHIP_STACKED_CHIPS = (1 << 31),
};
/* This include will go away once we isolated irq_desc usage to core code */
diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
index 32d9fff..f35f192 100644
--- a/include/linux/irqdesc.h
+++ b/include/linux/irqdesc.h
@@ -50,6 +50,9 @@ struct irq_desc {
struct irq_data irq_data;
unsigned int __percpu *kstat_irqs;
irq_flow_handler_t handle_irq;
+#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
+ unsigned long chip_flags; /* Union of all the irqchips */
+#endif
#ifdef CONFIG_IRQ_PREFLOW_FASTEOI
irq_preflow_handler_t preflow_handler;
#endif
@@ -109,6 +112,10 @@ static inline void *irq_desc_get_chip_data(struct irq_desc *desc)
static inline unsigned long irq_desc_get_chip_flags(struct irq_desc *desc)
{
+#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
+ if (desc->chip_flags & IRQCHIP_STACKED_CHIPS)
+ return desc->chip_flags;
+#endif
return desc->irq_data.chip->flags;
}
diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
index 99793b9..b916ea4 100644
--- a/kernel/irq/irqdesc.c
+++ b/kernel/irq/irqdesc.c
@@ -84,6 +84,9 @@ static void desc_set_defaults(unsigned int irq, struct irq_desc *desc, int node,
irq_settings_clr_and_set(desc, ~0, _IRQ_DEFAULT_INIT_FLAGS);
irqd_set(&desc->irq_data, IRQD_IRQ_DISABLED);
desc->handle_irq = handle_bad_irq;
+#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
+ desc->chip_flags = 0;
+#endif
desc->depth = 1;
desc->irq_count = 0;
desc->irqs_unhandled = 0;
diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
index 7fac311..3d0e366 100644
--- a/kernel/irq/irqdomain.c
+++ b/kernel/irq/irqdomain.c
@@ -1013,6 +1013,17 @@ static void irq_domain_free_irqs_recursive(struct irq_domain *domain,
}
}
+static void irq_desc_update_chip_flags(struct irq_domain *domain,
+ unsigned int virq)
+{
+#ifdef CONFIG_IRQ_DOMAIN_HIERARCHY
+ struct irq_desc *desc = irq_to_desc(virq);
+ struct irq_data *data = irq_domain_get_irq_data(domain, virq);
+
+ desc->chip_flags |= data->chip->flags | IRQCHIP_STACKED_CHIPS;
+#endif
+}
+
static int irq_domain_alloc_irqs_recursive(struct irq_domain *domain,
unsigned int irq_base,
unsigned int nr_irqs, void *arg)
@@ -1025,8 +1036,15 @@ static int irq_domain_alloc_irqs_recursive(struct irq_domain *domain,
if (recursive)
ret = irq_domain_alloc_irqs_recursive(parent, irq_base,
nr_irqs, arg);
- if (ret >= 0)
+ if (ret >= 0) {
ret = domain->ops->alloc(domain, irq_base, nr_irqs, arg);
+ if (ret >= 0) {
+ int i;
+
+ for (i = 0; i < nr_irqs; i++)
+ irq_desc_update_chip_flags(domain, irq_base + i);
+ }
+ }
if (ret < 0 && recursive)
irq_domain_free_irqs_recursive(parent, irq_base, nr_irqs);
--
2.1.4
^ permalink raw reply related [flat|nested] 9+ messages in thread* Re: [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains
2015-01-08 17:32 [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains Marc Zyngier
2015-01-08 17:32 ` [PATCH 1/2] genirq: Abstract access to irq_chip flags Marc Zyngier
2015-01-08 17:32 ` [PATCH 2/2] genirq: Allow irq_desc to carry the union of stacked " Marc Zyngier
@ 2015-01-09 3:07 ` Jiang Liu
2015-01-09 9:52 ` Marc Zyngier
2 siblings, 1 reply; 9+ messages in thread
From: Jiang Liu @ 2015-01-09 3:07 UTC (permalink / raw)
To: Marc Zyngier, Thomas Gleixner; +Cc: linux-kernel
On 2015/1/9 1:32, Marc Zyngier wrote:
> With the landing of stacked irq domains in 3.19, we have ended up in a
> situation where we have a stack of IRQ controllers, each with their
> set of flags, but the core code is only able to look at the top-most,
> which is not very helpful. This small series is trying to fix this.
>
> The first patch converts all access to desc->irq_data.chip->flags to
> using an accessor, without changing anything else. The second patch
> adds some logic to combine these flags as we allocate the interrupts,
> ultimately storing the resulting set as part of the irq_desc
> structure.
>
> We end-up with a configuration where the flags can either be located
> in the irq_chip structure (non stacked case), or in the irq_desc
> (stacked case). While this isn't really ideal, this gives at least the
> right level of information to the rest of the IRQ framework.
Hi Mark,
By this way, we need to aggregate irq_chip flags for every
irq. How about changing irq_desc_get_chip_flags(struct irq_desc *desc)
to irq_desc_check_chip_flags(struct irq_desc *desc, unsigned int flags)
which dynamically walks the stacked irqchips?
Thanks!
Gerry
>
> Based on 3.19-rc3, and available at:
> git://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git irq/stacked-irqchip-flags
>
> Marc Zyngier (2):
> genirq: Abstract access to irq_chip flags
> genirq: Allow irq_desc to carry the union of stacked irq_chip flags
>
> include/linux/irq.h | 4 ++++
> include/linux/irqdesc.h | 12 ++++++++++++
> kernel/irq/chip.c | 10 +++++-----
> kernel/irq/irqdesc.c | 3 +++
> kernel/irq/irqdomain.c | 20 +++++++++++++++++++-
> kernel/irq/manage.c | 6 +++---
> kernel/irq/pm.c | 2 +-
> 7 files changed, 47 insertions(+), 10 deletions(-)
>
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains
2015-01-09 3:07 ` [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains Jiang Liu
@ 2015-01-09 9:52 ` Marc Zyngier
2015-01-13 10:41 ` Thomas Gleixner
0 siblings, 1 reply; 9+ messages in thread
From: Marc Zyngier @ 2015-01-09 9:52 UTC (permalink / raw)
To: Jiang Liu, Thomas Gleixner; +Cc: linux-kernel@vger.kernel.org
Hi Jiang,
On 09/01/15 03:07, Jiang Liu wrote:
> On 2015/1/9 1:32, Marc Zyngier wrote:
>> With the landing of stacked irq domains in 3.19, we have ended up in a
>> situation where we have a stack of IRQ controllers, each with their
>> set of flags, but the core code is only able to look at the top-most,
>> which is not very helpful. This small series is trying to fix this.
>>
>> The first patch converts all access to desc->irq_data.chip->flags to
>> using an accessor, without changing anything else. The second patch
>> adds some logic to combine these flags as we allocate the interrupts,
>> ultimately storing the resulting set as part of the irq_desc
>> structure.
>>
>> We end-up with a configuration where the flags can either be located
>> in the irq_chip structure (non stacked case), or in the irq_desc
>> (stacked case). While this isn't really ideal, this gives at least the
>> right level of information to the rest of the IRQ framework.
> Hi Mark,
> By this way, we need to aggregate irq_chip flags for every
> irq. How about changing irq_desc_get_chip_flags(struct irq_desc *desc)
> to irq_desc_check_chip_flags(struct irq_desc *desc, unsigned int flags)
> which dynamically walks the stacked irqchips?
That was the other option, but I felt that going through the list of
irqchips was not a very nice idea performance wise. This is used is some
moderately hot paths (cond_unmask_eoi_irq, for example), and I wondered
if we wanted to pay this price at runtime...
Thoughts?
Thanks,
M.
--
Jazz is not dead. It just smells funny...
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains
2015-01-09 9:52 ` Marc Zyngier
@ 2015-01-13 10:41 ` Thomas Gleixner
2015-01-13 10:49 ` Marc Zyngier
0 siblings, 1 reply; 9+ messages in thread
From: Thomas Gleixner @ 2015-01-13 10:41 UTC (permalink / raw)
To: Marc Zyngier; +Cc: Jiang Liu, linux-kernel@vger.kernel.org
On Fri, 9 Jan 2015, Marc Zyngier wrote:
> Hi Jiang,
>
> On 09/01/15 03:07, Jiang Liu wrote:
> > On 2015/1/9 1:32, Marc Zyngier wrote:
> >> With the landing of stacked irq domains in 3.19, we have ended up in a
> >> situation where we have a stack of IRQ controllers, each with their
> >> set of flags, but the core code is only able to look at the top-most,
> >> which is not very helpful. This small series is trying to fix this.
> >>
> >> The first patch converts all access to desc->irq_data.chip->flags to
> >> using an accessor, without changing anything else. The second patch
> >> adds some logic to combine these flags as we allocate the interrupts,
> >> ultimately storing the resulting set as part of the irq_desc
> >> structure.
> >>
> >> We end-up with a configuration where the flags can either be located
> >> in the irq_chip structure (non stacked case), or in the irq_desc
> >> (stacked case). While this isn't really ideal, this gives at least the
> >> right level of information to the rest of the IRQ framework.
> > Hi Mark,
> > By this way, we need to aggregate irq_chip flags for every
> > irq. How about changing irq_desc_get_chip_flags(struct irq_desc *desc)
> > to irq_desc_check_chip_flags(struct irq_desc *desc, unsigned int flags)
> > which dynamically walks the stacked irqchips?
>
> That was the other option, but I felt that going through the list of
> irqchips was not a very nice idea performance wise. This is used is some
> moderately hot paths (cond_unmask_eoi_irq, for example), and I wondered
> if we wanted to pay this price at runtime...
We rather do it at setup time.
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains
2015-01-13 10:41 ` Thomas Gleixner
@ 2015-01-13 10:49 ` Marc Zyngier
0 siblings, 0 replies; 9+ messages in thread
From: Marc Zyngier @ 2015-01-13 10:49 UTC (permalink / raw)
To: Thomas Gleixner; +Cc: Jiang Liu, linux-kernel@vger.kernel.org
On 13/01/15 10:41, Thomas Gleixner wrote:
> On Fri, 9 Jan 2015, Marc Zyngier wrote:
>> Hi Jiang,
>>
>> On 09/01/15 03:07, Jiang Liu wrote:
>>> On 2015/1/9 1:32, Marc Zyngier wrote:
>>>> With the landing of stacked irq domains in 3.19, we have ended up in a
>>>> situation where we have a stack of IRQ controllers, each with their
>>>> set of flags, but the core code is only able to look at the top-most,
>>>> which is not very helpful. This small series is trying to fix this.
>>>>
>>>> The first patch converts all access to desc->irq_data.chip->flags to
>>>> using an accessor, without changing anything else. The second patch
>>>> adds some logic to combine these flags as we allocate the interrupts,
>>>> ultimately storing the resulting set as part of the irq_desc
>>>> structure.
>>>>
>>>> We end-up with a configuration where the flags can either be located
>>>> in the irq_chip structure (non stacked case), or in the irq_desc
>>>> (stacked case). While this isn't really ideal, this gives at least the
>>>> right level of information to the rest of the IRQ framework.
>>> Hi Mark,
>>> By this way, we need to aggregate irq_chip flags for every
>>> irq. How about changing irq_desc_get_chip_flags(struct irq_desc *desc)
>>> to irq_desc_check_chip_flags(struct irq_desc *desc, unsigned int flags)
>>> which dynamically walks the stacked irqchips?
>>
>> That was the other option, but I felt that going through the list of
>> irqchips was not a very nice idea performance wise. This is used is some
>> moderately hot paths (cond_unmask_eoi_irq, for example), and I wondered
>> if we wanted to pay this price at runtime...
>
> We rather do it at setup time.
That was my thought as well. I'll respin the series with Jiang's fix.
Thanks,
M.
--
Jazz is not dead. It just smells funny...
^ permalink raw reply [flat|nested] 9+ messages in thread