[PATCH v1 0/2] genirq: support multiple IRQ notifier.

[PATCH v1 0/2] genirq: support multiple IRQ notifier.

[Weongyo Jeong]

Each irq_desc only supports one IRQ affinity notifier at current
implementation so when we try to register another notifier, it silently
unregister previous entry and register new one.

However the problem is that if CONFIG_RFS_ACCEL is set, at current
implementation no way to set additional IRQ affinity notifier for
some NIC cards RFS enabled because it already used for RFS.
With this patch we can register multiple IRQ affinity notifiers.

Weongyo Jeong (2):
  genirq: clean up for irq_set_affinity_notifier().
  genirq: support multiple IRQ notifier.

 drivers/infiniband/hw/qib/qib_iba7322.c | 10 ++--
 drivers/scsi/qla2xxx/qla_isr.c          |  4 +-
 include/linux/interrupt.h               | 13 +++++-
 include/linux/irqdesc.h                 |  2 +-
 kernel/irq/irqdesc.c                    |  1 +
 kernel/irq/manage.c                     | 82 ++++++++++++++++++++++-----------
 lib/cpu_rmap.c                          |  4 +-
 7 files changed, 78 insertions(+), 38 deletions(-)

-- 
2.1.3

[PATCH v1 1/2] genirq: clean up for irq_set_affinity_notifier().

[Weongyo Jeong]

At current irq_set_affinity_notifier() implementation, it has two meaning;
set and clear notify.  To be more clear, separate it into two;
irq_set_affinity_notifier() and irq_del_affinity_notifier().

At irq_set_affinity_notifier() function, no longer might_sleep() function
is called unnecessarily and -EEXIST code is returned.

Signed-off-by: Weongyo Jeong <weongyo.linux@gmail.com>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |  8 ++---
 drivers/scsi/qla2xxx/qla_isr.c          |  2 +-
 include/linux/interrupt.h               |  2 ++
 kernel/irq/manage.c                     | 60 ++++++++++++++++++++++++---------
 lib/cpu_rmap.c                          |  2 +-
 5 files changed, 52 insertions(+), 22 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 6c8ff10..58c482a 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -3355,10 +3355,10 @@ static void reset_dca_notifier(struct qib_devdata *dd, struct qib_msix_entry *m)
 		"Disabling notifier on HCA %d irq %d\n",
 		dd->unit,
 		m->msix.vector);
-	irq_set_affinity_notifier(
-		m->msix.vector,
-		NULL);
-	m->notifier = NULL;
+	if (m->notifier != NULL) {
+		irq_del_affinity_notifier(&m->notifier->notify);
+		m->notifier = NULL;
+	}
 }
 
 static void setup_dca_notifier(struct qib_devdata *dd, struct qib_msix_entry *m)
diff --git a/drivers/scsi/qla2xxx/qla_isr.c b/drivers/scsi/qla2xxx/qla_isr.c
index 5649c20..0a652fa 100644
--- a/drivers/scsi/qla2xxx/qla_isr.c
+++ b/drivers/scsi/qla2xxx/qla_isr.c
@@ -3013,7 +3013,7 @@ qla24xx_disable_msix(struct qla_hw_data *ha)
 		qentry = &ha->msix_entries[i];
 		if (qentry->have_irq) {
 			/* un-register irq cpu affinity notification */
-			irq_set_affinity_notifier(qentry->vector, NULL);
+			irq_del_affinity_notifier(&qentry->irq_notify);
 			free_irq(qentry->vector, qentry->rsp);
 		}
 	}
diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
index 358076e..fc54356 100644
--- a/include/linux/interrupt.h
+++ b/include/linux/interrupt.h
@@ -277,6 +277,8 @@ extern int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m);
 
 extern int
 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify);
+extern int
+irq_del_affinity_notifier(struct irq_affinity_notify *notify);
 
 #else /* CONFIG_SMP */
 
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index 64731e8..6fb1414 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -282,39 +282,67 @@ out:
 }
 
 /**
- *	irq_set_affinity_notifier - control notification of IRQ affinity changes
- *	@irq:		Interrupt for which to enable/disable notification
- *	@notify:	Context for notification, or %NULL to disable
- *			notification.  Function pointers must be initialised;
+ *	irq_set_affinity_notifier - set notification of IRQ affinity changes
+ *	@irq:		Interrupt for which to enable notification
+ *	@notify:	Context for notification.
+ *			Function pointers must be initialised;
  *			the other fields will be initialised by this function.
  *
- *	Must be called in process context.  Notification may only be enabled
- *	after the IRQ is allocated and must be disabled before the IRQ is
- *	freed using free_irq().
+ *	Notification may only be enabled after the IRQ is allocated.
  */
 int
 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 {
 	struct irq_desc *desc = irq_to_desc(irq);
+	unsigned long flags;
+
+	if (!desc)
+		return -EINVAL;
+	if (!notify)
+		return -EINVAL;
+
+	raw_spin_lock_irqsave(&desc->lock, flags);
+	if (desc->affinity_notify != NULL) {
+		raw_spin_unlock_irqrestore(&desc->lock, flags);
+		return -EEXIST;
+	}
+	notify->irq = irq;
+	kref_init(&notify->kref);
+	INIT_WORK(&notify->work, irq_affinity_notify);
+
+	desc->affinity_notify = notify;
+	raw_spin_unlock_irqrestore(&desc->lock, flags);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
+
+/**
+ *	irq_del_affinity_notifier - delete notification of IRQ affinity changes
+ *	@notify:	Context for notification.
+ *
+ *	Must be called in process context.  Notification must be disabled
+ *	before the IRQ is freed using free_irq().
+ */
+int
+irq_del_affinity_notifier(struct irq_affinity_notify *notify)
+{
+	struct irq_desc *desc;
 	struct irq_affinity_notify *old_notify;
 	unsigned long flags;
 
 	/* The release function is promised process context */
 	might_sleep();
 
+	if (!notify)
+		return -EINVAL;
+	desc = irq_to_desc(notify->irq);
 	if (!desc)
 		return -EINVAL;
 
-	/* Complete initialisation of *notify */
-	if (notify) {
-		notify->irq = irq;
-		kref_init(&notify->kref);
-		INIT_WORK(&notify->work, irq_affinity_notify);
-	}
-
 	raw_spin_lock_irqsave(&desc->lock, flags);
 	old_notify = desc->affinity_notify;
-	desc->affinity_notify = notify;
+	desc->affinity_notify = NULL;
 	raw_spin_unlock_irqrestore(&desc->lock, flags);
 
 	if (old_notify)
@@ -322,7 +350,7 @@ irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 
 	return 0;
 }
-EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
+EXPORT_SYMBOL_GPL(irq_del_affinity_notifier);
 
 #ifndef CONFIG_AUTO_IRQ_AFFINITY
 /*
diff --git a/lib/cpu_rmap.c b/lib/cpu_rmap.c
index f610b2a..0412a51 100644
--- a/lib/cpu_rmap.c
+++ b/lib/cpu_rmap.c
@@ -235,7 +235,7 @@ void free_irq_cpu_rmap(struct cpu_rmap *rmap)
 
 	for (index = 0; index < rmap->used; index++) {
 		glue = rmap->obj[index];
-		irq_set_affinity_notifier(glue->notify.irq, NULL);
+		irq_del_affinity_notifier(&glue->notify);
 	}
 
 	cpu_rmap_put(rmap);
-- 
2.1.3

[PATCH v1 2/2] genirq: support multiple IRQ notifier.

[Weongyo Jeong]

At current implementation, it only supports one IRQ notifier for irq_desc.
When if we try to register another IRQ notifier, it automatically
un-register previous one and register new one.  With this patch, multiple
IRQ notifier is supported.

Signed-off-by: Weongyo Jeong <weongyo.linux@gmail.com>
---
 drivers/infiniband/hw/qib/qib_iba7322.c |  2 +-
 drivers/scsi/qla2xxx/qla_isr.c          |  2 +-
 include/linux/interrupt.h               | 11 ++++++++--
 include/linux/irqdesc.h                 |  2 +-
 kernel/irq/irqdesc.c                    |  1 +
 kernel/irq/manage.c                     | 38 +++++++++++++++++----------------
 lib/cpu_rmap.c                          |  2 +-
 7 files changed, 34 insertions(+), 24 deletions(-)

diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 58c482a..5376592 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -3380,7 +3380,7 @@ static void setup_dca_notifier(struct qib_devdata *dd, struct qib_msix_entry *m)
 		qib_devinfo(dd->pcidev,
 			"set notifier irq %d rcv %d notify %p\n",
 			n->notify.irq, n->rcv, &n->notify);
-		ret = irq_set_affinity_notifier(
+		ret = irq_add_affinity_notifier(
 				n->notify.irq,
 				&n->notify);
 		if (ret) {
diff --git a/drivers/scsi/qla2xxx/qla_isr.c b/drivers/scsi/qla2xxx/qla_isr.c
index 0a652fa..13318c0 100644
--- a/drivers/scsi/qla2xxx/qla_isr.c
+++ b/drivers/scsi/qla2xxx/qla_isr.c
@@ -3101,7 +3101,7 @@ qla24xx_enable_msix(struct qla_hw_data *ha, struct rsp_que *rsp)
 		rsp->msix = qentry;
 
 		/* Register for CPU affinity notification. */
-		irq_set_affinity_notifier(qentry->vector, &qentry->irq_notify);
+		irq_add_affinity_notifier(qentry->vector, &qentry->irq_notify);
 
 		/* Schedule work (ie. trigger a notification) to read cpu
 		 * mask for this specific irq.
diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
index fc54356..d7368ab 100644
--- a/include/linux/interrupt.h
+++ b/include/linux/interrupt.h
@@ -230,6 +230,7 @@ struct irq_affinity_notify {
 	struct work_struct work;
 	void (*notify)(struct irq_affinity_notify *, const cpumask_t *mask);
 	void (*release)(struct kref *ref);
+	struct list_head list;
 };
 
 #if defined(CONFIG_SMP)
@@ -276,7 +277,7 @@ extern int irq_select_affinity(unsigned int irq);
 extern int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m);
 
 extern int
-irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify);
+irq_add_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify);
 extern int
 irq_del_affinity_notifier(struct irq_affinity_notify *notify);
 
@@ -306,7 +307,13 @@ static inline int irq_set_affinity_hint(unsigned int irq,
 }
 
 static inline int
-irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
+irq_add_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
+{
+	return 0;
+}
+
+static inline int
+irq_del_affinity_notifier(struct irq_affinity_notify *notify)
 {
 	return 0;
 }
diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
index dcca77c..51cc163 100644
--- a/include/linux/irqdesc.h
+++ b/include/linux/irqdesc.h
@@ -68,7 +68,7 @@ struct irq_desc {
 	struct cpumask		*percpu_enabled;
 #ifdef CONFIG_SMP
 	const struct cpumask	*affinity_hint;
-	struct irq_affinity_notify *affinity_notify;
+	struct list_head	affinity_notify_list;
 #ifdef CONFIG_GENERIC_PENDING_IRQ
 	cpumask_var_t		pending_mask;
 #endif
diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
index 0ccd028..4f99fb3 100644
--- a/kernel/irq/irqdesc.c
+++ b/kernel/irq/irqdesc.c
@@ -70,6 +70,7 @@ static int alloc_masks(struct irq_desc *desc, gfp_t gfp, int node)
 
 static void desc_smp_init(struct irq_desc *desc, int node)
 {
+	INIT_LIST_HEAD(&desc->affinity_notify_list);
 	cpumask_copy(desc->irq_common_data.affinity, irq_default_affinity);
 #ifdef CONFIG_GENERIC_PENDING_IRQ
 	cpumask_clear(desc->pending_mask);
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index 6fb1414..bfd7673 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -202,6 +202,7 @@ int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
 int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
 			    bool force)
 {
+	struct irq_affinity_notify *notify;
 	struct irq_chip *chip = irq_data_get_irq_chip(data);
 	struct irq_desc *desc = irq_data_to_desc(data);
 	int ret = 0;
@@ -216,9 +217,9 @@ int irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask,
 		irq_copy_pending(desc, mask);
 	}
 
-	if (desc->affinity_notify) {
-		kref_get(&desc->affinity_notify->kref);
-		schedule_work(&desc->affinity_notify->work);
+	list_for_each_entry(notify, &desc->affinity_notify_list, list) {
+		kref_get(&notify->kref);
+		schedule_work(&notify->work);
 	}
 	irqd_set(data, IRQD_AFFINITY_SET);
 
@@ -282,7 +283,7 @@ out:
 }
 
 /**
- *	irq_set_affinity_notifier - set notification of IRQ affinity changes
+ *	irq_add_affinity_notifier - set notification of IRQ affinity changes
  *	@irq:		Interrupt for which to enable notification
  *	@notify:	Context for notification.
  *			Function pointers must be initialised;
@@ -291,7 +292,7 @@ out:
  *	Notification may only be enabled after the IRQ is allocated.
  */
 int
-irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
+irq_add_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 {
 	struct irq_desc *desc = irq_to_desc(irq);
 	unsigned long flags;
@@ -302,20 +303,16 @@ irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify)
 		return -EINVAL;
 
 	raw_spin_lock_irqsave(&desc->lock, flags);
-	if (desc->affinity_notify != NULL) {
-		raw_spin_unlock_irqrestore(&desc->lock, flags);
-		return -EEXIST;
-	}
 	notify->irq = irq;
 	kref_init(&notify->kref);
 	INIT_WORK(&notify->work, irq_affinity_notify);
 
-	desc->affinity_notify = notify;
+	list_add(&notify->list, &desc->affinity_notify_list);
 	raw_spin_unlock_irqrestore(&desc->lock, flags);
 
 	return 0;
 }
-EXPORT_SYMBOL_GPL(irq_set_affinity_notifier);
+EXPORT_SYMBOL_GPL(irq_add_affinity_notifier);
 
 /**
  *	irq_del_affinity_notifier - delete notification of IRQ affinity changes
@@ -328,7 +325,6 @@ int
 irq_del_affinity_notifier(struct irq_affinity_notify *notify)
 {
 	struct irq_desc *desc;
-	struct irq_affinity_notify *old_notify;
 	unsigned long flags;
 
 	/* The release function is promised process context */
@@ -341,12 +337,10 @@ irq_del_affinity_notifier(struct irq_affinity_notify *notify)
 		return -EINVAL;
 
 	raw_spin_lock_irqsave(&desc->lock, flags);
-	old_notify = desc->affinity_notify;
-	desc->affinity_notify = NULL;
+	list_del(&notify->list);
 	raw_spin_unlock_irqrestore(&desc->lock, flags);
 
-	if (old_notify)
-		kref_put(&old_notify->kref, old_notify->release);
+	kref_put(&notify->kref, notify->release);
 
 	return 0;
 }
@@ -1571,14 +1565,22 @@ EXPORT_SYMBOL_GPL(remove_irq);
  */
 void free_irq(unsigned int irq, void *dev_id)
 {
+#ifdef CONFIG_SMP
+	struct irq_affinity_notify *notify, *notifytmp;
+#endif
 	struct irq_desc *desc = irq_to_desc(irq);
 
 	if (!desc || WARN_ON(irq_settings_is_per_cpu_devid(desc)))
 		return;
 
 #ifdef CONFIG_SMP
-	if (WARN_ON(desc->affinity_notify))
-		desc->affinity_notify = NULL;
+	list_for_each_entry_safe(notify, notifytmp, &desc->affinity_notify_list,
+				 list) {
+		if (notify->irq != irq)
+			continue;
+		__WARN();
+		list_del(&notify->list);
+	}
 #endif
 
 	kfree(__free_irq(irq, dev_id));
diff --git a/lib/cpu_rmap.c b/lib/cpu_rmap.c
index 0412a51..fb3a8e4 100644
--- a/lib/cpu_rmap.c
+++ b/lib/cpu_rmap.c
@@ -297,7 +297,7 @@ int irq_cpu_rmap_add(struct cpu_rmap *rmap, int irq)
 	glue->rmap = rmap;
 	cpu_rmap_get(rmap);
 	glue->index = cpu_rmap_add(rmap, glue);
-	rc = irq_set_affinity_notifier(irq, &glue->notify);
+	rc = irq_add_affinity_notifier(irq, &glue->notify);
 	if (rc) {
 		cpu_rmap_put(glue->rmap);
 		kfree(glue);
-- 
2.1.3

Re: [PATCH v1 2/2] genirq: support multiple IRQ notifier.

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 2975 bytes --]

Hi Weongyo,

[auto build test WARNING on tip/irq/core]
[also build test WARNING on v4.5 next-20160324]
[if your patch is applied to the wrong git tree, please drop us a note to help improving the system]

url:    https://github.com/0day-ci/linux/commits/Weongyo-Jeong/genirq-support-multiple-IRQ-notifier/20160325-235708
reproduce: make htmldocs

All warnings (new ones prefixed by >>):

>> include/linux/interrupt.h:224: warning: No description found for parameter 'list'
   kernel/irq/handle.c:1: warning: no structured comments found

vim +/list +224 include/linux/interrupt.h

f0ba3d05c Eyal Perry         2014-05-20  208   * @kref:		Reference count, for internal use
f0ba3d05c Eyal Perry         2014-05-20  209   * @work:		Work item, for internal use
f0ba3d05c Eyal Perry         2014-05-20  210   * @notify:		Function to be called on change.  This will be
f0ba3d05c Eyal Perry         2014-05-20  211   *			called in process context.
f0ba3d05c Eyal Perry         2014-05-20  212   * @release:		Function to be called on release.  This will be
f0ba3d05c Eyal Perry         2014-05-20  213   *			called in process context.  Once registered, the
f0ba3d05c Eyal Perry         2014-05-20  214   *			structure must only be freed when this function is
f0ba3d05c Eyal Perry         2014-05-20  215   *			called or later.
f0ba3d05c Eyal Perry         2014-05-20  216   */
f0ba3d05c Eyal Perry         2014-05-20  217  struct irq_affinity_notify {
f0ba3d05c Eyal Perry         2014-05-20  218  	unsigned int irq;
f0ba3d05c Eyal Perry         2014-05-20  219  	struct kref kref;
f0ba3d05c Eyal Perry         2014-05-20  220  	struct work_struct work;
f0ba3d05c Eyal Perry         2014-05-20  221  	void (*notify)(struct irq_affinity_notify *, const cpumask_t *mask);
f0ba3d05c Eyal Perry         2014-05-20  222  	void (*release)(struct kref *ref);
3fb671e48 Weongyo Jeong      2016-03-25  223  	struct list_head list;
f0ba3d05c Eyal Perry         2014-05-20 @224  };
f0ba3d05c Eyal Perry         2014-05-20  225  
0244ad004 Martin Schwidefsky 2013-08-30  226  #if defined(CONFIG_SMP)
d7b906897 Russell King       2008-04-17  227  
d036e67b4 Rusty Russell      2009-01-01  228  extern cpumask_var_t irq_default_affinity;
184047567 Max Krasnyansky    2008-05-29  229  
01f8fa4f0 Thomas Gleixner    2014-04-16  230  /* Internal implementation. Use the helpers below */
01f8fa4f0 Thomas Gleixner    2014-04-16  231  extern int __irq_set_affinity(unsigned int irq, const struct cpumask *cpumask,
01f8fa4f0 Thomas Gleixner    2014-04-16  232  			      bool force);

:::::: The code at line 224 was first introduced by commit
:::::: f0ba3d05c9c647ab42ed6a0dbdfdeae42bfbd6de genirq: Provide !SMP stub for irq_set_affinity_notifier()

:::::: TO: Eyal Perry <eyalpe@mellanox.com>
:::::: CC: Thomas Gleixner <tglx@linutronix.de>

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/octet-stream, Size: 6226 bytes --]

Re: [PATCH v1 2/2] genirq: support multiple IRQ notifier.

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 1359 bytes --]

Hi Weongyo,

[auto build test ERROR on tip/irq/core]
[also build test ERROR on v4.5 next-20160324]
[if your patch is applied to the wrong git tree, please drop us a note to help improving the system]

url:    https://github.com/0day-ci/linux/commits/Weongyo-Jeong/genirq-support-multiple-IRQ-notifier/20160325-235708
config: arm64-allnoconfig (attached as .config)
reproduce:
        wget https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/plain/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # save the attached .config to linux build tree
        make.cross ARCH=arm64 

All errors (new ones prefixed by >>):

   kernel/irq/manage.c: In function 'free_irq':
>> kernel/irq/manage.c:1581:3: error: implicit declaration of function '__WARN' [-Werror=implicit-function-declaration]
      __WARN();
      ^
   cc1: some warnings being treated as errors

vim +/__WARN +1581 kernel/irq/manage.c

  1575	
  1576	#ifdef CONFIG_SMP
  1577		list_for_each_entry_safe(notify, notifytmp, &desc->affinity_notify_list,
  1578					 list) {
  1579			if (notify->irq != irq)
  1580				continue;
> 1581			__WARN();
  1582			list_del(&notify->list);
  1583		}
  1584	#endif

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/octet-stream, Size: 5758 bytes --]

Re: [PATCH v1 0/2] genirq: support multiple IRQ notifier.

[Christoph Hellwig]

On Fri, Mar 25, 2016 at 08:51:51AM -0700, Weongyo Jeong wrote:
> Each irq_desc only supports one IRQ affinity notifier at current
> implementation so when we try to register another notifier, it silently
> unregister previous entry and register new one.
> 
> However the problem is that if CONFIG_RFS_ACCEL is set, at current
> implementation no way to set additional IRQ affinity notifier for
> some NIC cards RFS enabled because it already used for RFS.
> With this patch we can register multiple IRQ affinity notifiers.

The whole concept of these irq affinity notifiers seems wrong to me.

If a device supports MSI-X it should simply request per-cpu or per-node
vectors and we should prevent affinity changes for them.

Re: [PATCH v1 0/2] genirq: support multiple IRQ notifier.

[Thomas Gleixner]

On Fri, 25 Mar 2016, Christoph Hellwig wrote:
> On Fri, Mar 25, 2016 at 08:51:51AM -0700, Weongyo Jeong wrote:
> > Each irq_desc only supports one IRQ affinity notifier at current
> > implementation so when we try to register another notifier, it silently
> > unregister previous entry and register new one.
> > 
> > However the problem is that if CONFIG_RFS_ACCEL is set, at current
> > implementation no way to set additional IRQ affinity notifier for
> > some NIC cards RFS enabled because it already used for RFS.
> > With this patch we can register multiple IRQ affinity notifiers.
> 
> The whole concept of these irq affinity notifiers seems wrong to me.
> 
> If a device supports MSI-X it should simply request per-cpu or per-node
> vectors and we should prevent affinity changes for them.

You beat me to it. I'm having a half baken prototype to implement this at the
core level. Will post next week.

Thanks,

	tglx

Re: [PATCH v1 0/2] genirq: support multiple IRQ notifier.

[Weongyo Jeong]

On Fri, Mar 25, 2016 at 12:32:43PM -0700, Christoph Hellwig wrote:
> On Fri, Mar 25, 2016 at 08:51:51AM -0700, Weongyo Jeong wrote:
> > Each irq_desc only supports one IRQ affinity notifier at current
> > implementation so when we try to register another notifier, it silently
> > unregister previous entry and register new one.
> > 
> > However the problem is that if CONFIG_RFS_ACCEL is set, at current
> > implementation no way to set additional IRQ affinity notifier for
> > some NIC cards RFS enabled because it already used for RFS.
> > With this patch we can register multiple IRQ affinity notifiers.
> 
> The whole concept of these irq affinity notifiers seems wrong to me.
> 
> If a device supports MSI-X it should simply request per-cpu or per-node
> vectors and we should prevent affinity changes for them.

This could be a silly question.  Are you meaning that we should remove
feature of IRQ affinity notifiers and device writer should explicitly set
CPU affinity with masking when it requests IRQ?

And then some CPU affinity of IRQ are still changable via
/proc/irq/<n>/smp_affinity and some aren't for device drivers, right?

Regards,
Weongyo Jeong