Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1436298 > unrolled thread

automatic interrupt affinity for MSI/MSI-X capable devices V3

Started byChristoph Hellwig <hch@lst.de>
First post2016-07-04 10:40 +0200
Last post2016-07-04 12:40 +0200
Articles 4 on this page of 24 — 5 participants

Back to article view | Back to linux.kernel

Contents

  automatic interrupt affinity for MSI/MSI-X capable devices V3 Christoph Hellwig <hch@lst.de> - 2016-07-04 10:40 +0200
    [PATCH 04/13] irq: Use affinity hint in irqdesc allocation Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      [tip:irq/core] genirq: Use affinity hint in irqdesc allocation tip-bot for Thomas Gleixner <tipbot@zytor.com> - 2016-07-04 12:40 +0200
    [PATCH 02/13] irq: Introduce IRQD_AFFINITY_MANAGED flag Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      [tip:irq/core] genirq: Introduce IRQD_AFFINITY_MANAGED flag tip-bot for Thomas Gleixner <tipbot@zytor.com> - 2016-07-04 12:40 +0200
    [PATCH 05/13] irq/msi: Make use of affinity aware allocations Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      [tip:irq/core] genirq/msi: Make use of affinity aware allocations tip-bot for Thomas Gleixner <tipbot@zytor.com> - 2016-07-04 12:40 +0200
    [PATCH 11/13] blk-mq: allow the driver to pass in an affinity mask Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
    [PATCH 13/13] nvme: remove the post_scan callout Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
    [PATCH 08/13] pci: spread interrupt vectors in pci_alloc_irq_vectors Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      Re: [PATCH 08/13] pci: spread interrupt vectors in  pci_alloc_irq_vectors Alexander Gordeev <agordeev@redhat.com> - 2016-07-07 13:10 +0200
    [PATCH 01/13] irq/msi: Remove unused MSI_FLAG_IDENTITY_MAP Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      [tip:irq/core] genirq/msi: Remove unused MSI_FLAG_IDENTITY_MAP tip-bot for Thomas Gleixner <tipbot@zytor.com> - 2016-07-04 12:40 +0200
    [PATCH 12/13] nvme: switch to use pci_alloc_irq_vectors Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      Re: [PATCH 12/13] nvme: switch to use pci_alloc_irq_vectors Alexander Gordeev <agordeev@redhat.com> - 2016-07-07 21:40 +0200
    [PATCH 10/13] blk-mq: only allocate a single mq_map per tag_set Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
    [PATCH 09/13] blk-mq: don't redistribute hardware queues on a CPU hotplug event Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
    [PATCH 06/13] irq: add a helper spread an affinity mask for MSI/MSI-X vectors Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      [tip:irq/core] genirq: Add a helper to spread an affinity mask for  MSI/MSI-X vectors tip-bot for Christoph Hellwig <tipbot@zytor.com> - 2016-07-04 12:40 +0200
    [PATCH 07/13] pci: Provide sensible irq vector alloc/free routines Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      Re: [PATCH 07/13] pci: Provide sensible irq vector alloc/free  routines Alexander Gordeev <agordeev@redhat.com> - 2016-07-06 10:10 +0200
    [PATCH 03/13] irq: Add affinity hint to irq allocation Christoph Hellwig <hch@lst.de> - 2016-07-04 10:50 +0200
      [tip:irq/core] genirq: Add affinity hint to irq allocation tip-bot for Thomas Gleixner <tipbot@zytor.com> - 2016-07-04 12:40 +0200
    Re: automatic interrupt affinity for MSI/MSI-X capable devices V3 Thomas Gleixner <tglx@linutronix.de> - 2016-07-04 12:40 +0200

Page 2 of 2 — ← Prev page 1 [2]

#1437527 — Re: [PATCH 07/13] pci: Provide sensible irq vector alloc/free routines

FromAlexander Gordeev <agordeev@redhat.com>
Date2016-07-06 10:10 +0200
SubjectRe: [PATCH 07/13] pci: Provide sensible irq vector alloc/free routines
Message-ID<rRQ49-6FO-17@gated-at.bofh.it>
In reply to#1436328
On Mon, Jul 04, 2016 at 05:39:28PM +0900, Christoph Hellwig wrote:
> Add a function to allocate a range of interrupt vectors, which will
> transparently use MSI-X and MSI if available or fallback to legacy
> vectors.  The interrupts are available in a core managed array in the
> pci_dev structure, and can also be released using a similar function.
> A new helper, __pci_enable_msix_range, is introduced to allow allocating
> the array of msix descriptors in the core PCIe code at the exact number
> of vectors supported by the PCI host complex, and to also help with the
> automatic IRQ affinity assignment that will be added in the next commit.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  Documentation/PCI/MSI-HOWTO.txt | 437 ++++------------------------------------

[...]

> + pci_enable_msi, pci_enable_msi_range, pci_enable_msi_exact, pci_disable_msi,
> + pci_msi_vec_count, pci_enable_msix_range, pci_enable_msix_exact,
> + pci_disable_msix, pci_msix_vec_count

Description of these functions can be removed when all drivers migrated
to the new API. Also implementation descriptions + examples would still
be needed AFAICT.

[...]

> --- a/drivers/pci/msi.c
> +++ b/drivers/pci/msi.c
> @@ -4,6 +4,7 @@
>   *
>   * Copyright (C) 2003-2004 Intel
>   * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
> + * Copyright (C) 2016 Christoph Hellwig.
>   */
>  
>  #include <linux/err.h>
> @@ -1120,6 +1121,98 @@ int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
>  }
>  EXPORT_SYMBOL(pci_enable_msix_range);
>  
> +static int __pci_enable_msix_range(struct pci_dev *dev, unsigned int min_vecs,
> +		unsigned int max_vecs)
> +{
> +	int vecs = max_vecs, ret, i;
> +
> +retry:
> +	if (vecs < min_vecs)
> +		return -ENOSPC;
> +
> +	dev->msix_vectors = kmalloc_array(vecs, sizeof(struct msix_entry),
> +				GFP_KERNEL);
> +	if (!dev->msix_vectors)
> +		return -ENOMEM;
> +
> +	for (i = 0; i < vecs; i++)
> +		dev->msix_vectors[i].entry = i;
> +
> +	ret = pci_enable_msix(dev, dev->msix_vectors, vecs);
> +	if (ret)
> +		goto out_fail;
> +
> +	return vecs;
> +
> +out_fail:
> +	kfree(dev->msix_vectors);
> +	dev->msix_vectors = NULL;
> +
> +	if (ret >= 0) {
> +		/* retry with the actually supported number of vectors */
> +		vecs = ret;
> +		goto retry;
> +	}
> +
> +	return ret;
> +}

This function's code almost matches the existing pci_enable_msix_range()
so pci_enable_msix_range() should be reworked instead IMHO.

I took a look at the existing code and the rework does not appear
necessary as __pci_enable_msix_range() could look like this:

static int __pci_enable_msix_range(struct pci_dev *dev, unsigned int min_vecs,
		unsigned int max_vecs)
{
	int ret, i;
	struct msix_entry *entries;

	entries = kmalloc_array(max_vecs, sizeof(entries[0]), GFP_KERNEL);
	if (!entries)
		return -ENOMEM;

	for (i = 0; i < max_vecs; i++)
		entries[i].entry = i;

	ret = pci_enable_msix_range(dev, entries, min_vecs, max_vecs);

	kfree(entries);

	return ret;
}


We do not need to keep msix_entry array, since it only needed for
pci_irq_vector() function. But the same info could be retrieved from
msi_desc::irq.

> +/**
> + * pci_alloc_irq_vectors - allocate multiple IRQs for a device
> + * @dev:		PCI device to operate on
> + * @min_vecs:		minimum number of vectors required (must be >= 1)
> + * @max_vecs:		maximum (desired) number of vectors
> + * @flags:		flags or quirks for the allocation
> + *
> + * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
> + * vectors if available, and fall back to a single legacy vector
> + * if neither is available.  Return the number of vectors allocated,
> + * (which might be smaller than @max_vecs) if successful, or a negative
> + * error code on error. If less than @min_vecs interrupt vectors are
> + * available for @dev the function will fail with -ENOSPC.
> + *
> + * To get the Linux irq number used for a vector that can be passed to
> + * request_irq use the pci_irq_vector() helper.
> + */
> +int pci_alloc_irq_vectors(struct pci_dev *dev, unsigned int min_vecs,
> +		unsigned int max_vecs, unsigned int flags)
> +{
> +	int vecs;
> +
> +	if (!(flags & PCI_IRQ_NOMSIX)) {
> +		vecs = __pci_enable_msix_range(dev, min_vecs, max_vecs);
> +		if (vecs > 0)
> +			return vecs;
> +	}
> +
> +	if (!(flags & PCI_IRQ_NOMSI)) {
> +		vecs = pci_enable_msi_range(dev, min_vecs, max_vecs);
> +		if (vecs > 0)
> +			return vecs;
> +	}
> +
> +	/* use legacy irq if allowed */
> +	if (min_vecs == 1)
> +		return 1;
> +	return -ENOSPC;

The original error code (in vecs) would be overridden with -ENOSPC here.

> +}
> +EXPORT_SYMBOL(pci_alloc_irq_vectors);
> +
> +/**
> + * pci_free_irq_vectors - free previously allocated IRQs for a device
> + * @dev:		PCI device to operate on
> + *
> + * Undoes the allocations and enabling in pci_alloc_irq_vectors().
> + */
> +void pci_free_irq_vectors(struct pci_dev *dev)
> +{
> +	if (dev->msix_enabled)
> +		kfree(dev->msix_vectors);
> +	pci_disable_msix(dev);
> +	pci_disable_msi(dev);
> +}
> +EXPORT_SYMBOL(pci_free_irq_vectors);
> +
>  struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
>  {
>  	return to_pci_dev(desc->dev);
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index b67e4df..129871f 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -320,6 +320,7 @@ struct pci_dev {
>  	 * directly, use the values stored here. They might be different!
>  	 */
>  	unsigned int	irq;
> +	struct msix_entry *msix_vectors;

As suggested above, it is not needed.

>  	struct resource resource[DEVICE_COUNT_RESOURCE]; /* I/O and memory regions + expansion ROMs */
>  
>  	bool match_driver;		/* Skip attaching driver */
> @@ -1237,6 +1238,9 @@ resource_size_t pcibios_iov_resource_alignment(struct pci_dev *dev, int resno);
>  int pci_set_vga_state(struct pci_dev *pdev, bool decode,
>  		      unsigned int command_bits, u32 flags);
>  
> +#define PCI_IRQ_NOMSI		(1 << 0) /* don't try to use MSI interrupts */
> +#define PCI_IRQ_NOMSIX		(1 << 1) /* don't try to use MSI-X interrupts */
> +
>  /* kmem_cache style wrapper around pci_alloc_consistent() */
>  
>  #include <linux/pci-dma.h>
> @@ -1284,6 +1288,24 @@ static inline int pci_enable_msix_exact(struct pci_dev *dev,
>  		return rc;
>  	return 0;
>  }
> +int pci_alloc_irq_vectors(struct pci_dev *dev, unsigned int min_vecs,
> +		unsigned int max_vecs, unsigned int flags);
> +void pci_free_irq_vectors(struct pci_dev *dev);
> +
> +
> +/**
> + * pci_irq_vector - return Linux IRQ number of a device vector
> + * @dev: PCI device to operate on
> + * @nr: device-relative interrupt vector index (0-based).
> + */
> +static inline unsigned int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
> +{
> +	if (dev->msix_enabled)
> +		return dev->msix_vectors[nr].vector;

So here you could search with for_each_pci_msi_entry() instead.


> +	WARN_ON_ONCE(!dev->msi_enabled && nr > 0);
> +	return dev->irq + nr;

I think this function should check irq number existence and return the
vector number or -EINVAL;

> +}
>  #else
>  static inline int pci_msi_vec_count(struct pci_dev *dev) { return -ENOSYS; }
>  static inline void pci_msi_shutdown(struct pci_dev *dev) { }
> @@ -1307,6 +1329,23 @@ static inline int pci_enable_msix_range(struct pci_dev *dev,
>  static inline int pci_enable_msix_exact(struct pci_dev *dev,
>  		      struct msix_entry *entries, int nvec)
>  { return -ENOSYS; }
> +static inline int pci_alloc_irq_vectors(struct pci_dev *dev,
> +		unsigned int min_vecs, unsigned int max_vecs,
> +		unsigned int flags)
> +{
> +	if (min_vecs > 1)
> +		return -ENOSPC;

In case CONFIG_PCI_MSI is unset min_vecs > 1 is -EINVAL;

> +	return 1;
> +}
> +static inline void pci_free_irq_vectors(struct pci_dev *dev)
> +{
> +}
> +
> +static inline unsigned int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
> +{
> +	WARN_ON_ONCE(nr > 0);

As suggested above, it would rather -EINVAL;

> +	return dev->irq;
> +}
>  #endif
>  
>  #ifdef CONFIG_PCIEPORTBUS
> -- 
> 2.1.4
>

[toc] | [prev] | [next] | [standalone]

#1436329 — [PATCH 03/13] irq: Add affinity hint to irq allocation

FromChristoph Hellwig <hch@lst.de>
Date2016-07-04 10:50 +0200
Subject[PATCH 03/13] irq: Add affinity hint to irq allocation
Message-ID<rR7JN-3ot-41@gated-at.bofh.it>
In reply to#1436298
From: Thomas Gleixner <tglx@linutronix.de>

Add an extra argument to the irq(domain) allocation functions, so we can hand
down affinity hints to the allocator. Thats necessary to implement proper
support for multiqueue devices.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 arch/sparc/kernel/irq_64.c     |  2 +-
 arch/x86/kernel/apic/io_apic.c |  5 +++--
 include/linux/irq.h            |  4 ++--
 include/linux/irqdomain.h      |  9 ++++++---
 kernel/irq/ipi.c               |  2 +-
 kernel/irq/irqdesc.c           | 12 ++++++++----
 kernel/irq/irqdomain.c         | 22 ++++++++++++++--------
 kernel/irq/manage.c            |  7 ++++---
 kernel/irq/msi.c               |  3 ++-
 9 files changed, 41 insertions(+), 25 deletions(-)

diff --git a/arch/sparc/kernel/irq_64.c b/arch/sparc/kernel/irq_64.c
index e22416c..34a7930 100644
--- a/arch/sparc/kernel/irq_64.c
+++ b/arch/sparc/kernel/irq_64.c
@@ -242,7 +242,7 @@ unsigned int irq_alloc(unsigned int dev_handle, unsigned int dev_ino)
 {
 	int irq;
 
-	irq = __irq_alloc_descs(-1, 1, 1, numa_node_id(), NULL);
+	irq = __irq_alloc_descs(-1, 1, 1, numa_node_id(), NULL, NULL);
 	if (irq <= 0)
 		goto out;
 
diff --git a/arch/x86/kernel/apic/io_apic.c b/arch/x86/kernel/apic/io_apic.c
index 84e33ff..bca0c81 100644
--- a/arch/x86/kernel/apic/io_apic.c
+++ b/arch/x86/kernel/apic/io_apic.c
@@ -981,7 +981,7 @@ static int alloc_irq_from_domain(struct irq_domain *domain, int ioapic, u32 gsi,
 
 	return __irq_domain_alloc_irqs(domain, irq, 1,
 				       ioapic_alloc_attr_node(info),
-				       info, legacy);
+				       info, legacy, NULL);
 }
 
 /*
@@ -1014,7 +1014,8 @@ static int alloc_isa_irq_from_domain(struct irq_domain *domain,
 					  info->ioapic_pin))
 			return -ENOMEM;
 	} else {
-		irq = __irq_domain_alloc_irqs(domain, irq, 1, node, info, true);
+		irq = __irq_domain_alloc_irqs(domain, irq, 1, node, info, true,
+					      NULL);
 		if (irq >= 0) {
 			irq_data = irq_domain_get_irq_data(domain, irq);
 			data = irq_data->chip_data;
diff --git a/include/linux/irq.h b/include/linux/irq.h
index f607481..39ce46a 100644
--- a/include/linux/irq.h
+++ b/include/linux/irq.h
@@ -708,11 +708,11 @@ static inline struct cpumask *irq_data_get_affinity_mask(struct irq_data *d)
 unsigned int arch_dynirq_lower_bound(unsigned int from);
 
 int __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
-		struct module *owner);
+		      struct module *owner, const struct cpumask *affinity);
 
 /* use macros to avoid needing export.h for THIS_MODULE */
 #define irq_alloc_descs(irq, from, cnt, node)	\
-	__irq_alloc_descs(irq, from, cnt, node, THIS_MODULE)
+	__irq_alloc_descs(irq, from, cnt, node, THIS_MODULE, NULL)
 
 #define irq_alloc_desc(node)			\
 	irq_alloc_descs(-1, 0, 1, node)
diff --git a/include/linux/irqdomain.h b/include/linux/irqdomain.h
index f1f36e0..1aee0fb 100644
--- a/include/linux/irqdomain.h
+++ b/include/linux/irqdomain.h
@@ -39,6 +39,7 @@ struct irq_domain;
 struct of_device_id;
 struct irq_chip;
 struct irq_data;
+struct cpumask;
 
 /* Number of irqs reserved for a legacy isa controller */
 #define NUM_ISA_INTERRUPTS	16
@@ -217,7 +218,8 @@ extern struct irq_domain *irq_find_matching_fwspec(struct irq_fwspec *fwspec,
 						   enum irq_domain_bus_token bus_token);
 extern void irq_set_default_host(struct irq_domain *host);
 extern int irq_domain_alloc_descs(int virq, unsigned int nr_irqs,
-				  irq_hw_number_t hwirq, int node);
+				  irq_hw_number_t hwirq, int node,
+				  const struct cpumask *affinity);
 
 static inline struct fwnode_handle *of_node_to_fwnode(struct device_node *node)
 {
@@ -389,7 +391,7 @@ static inline struct irq_domain *irq_domain_add_hierarchy(struct irq_domain *par
 
 extern int __irq_domain_alloc_irqs(struct irq_domain *domain, int irq_base,
 				   unsigned int nr_irqs, int node, void *arg,
-				   bool realloc);
+				   bool realloc, const struct cpumask *affinity);
 extern void irq_domain_free_irqs(unsigned int virq, unsigned int nr_irqs);
 extern void irq_domain_activate_irq(struct irq_data *irq_data);
 extern void irq_domain_deactivate_irq(struct irq_data *irq_data);
@@ -397,7 +399,8 @@ extern void irq_domain_deactivate_irq(struct irq_data *irq_data);
 static inline int irq_domain_alloc_irqs(struct irq_domain *domain,
 			unsigned int nr_irqs, int node, void *arg)
 {
-	return __irq_domain_alloc_irqs(domain, -1, nr_irqs, node, arg, false);
+	return __irq_domain_alloc_irqs(domain, -1, nr_irqs, node, arg, false,
+				       NULL);
 }
 
 extern int irq_domain_alloc_irqs_recursive(struct irq_domain *domain,
diff --git a/kernel/irq/ipi.c b/kernel/irq/ipi.c
index 89b49f6..4fd2351 100644
--- a/kernel/irq/ipi.c
+++ b/kernel/irq/ipi.c
@@ -76,7 +76,7 @@ int irq_reserve_ipi(struct irq_domain *domain,
 		}
 	}
 
-	virq = irq_domain_alloc_descs(-1, nr_irqs, 0, NUMA_NO_NODE);
+	virq = irq_domain_alloc_descs(-1, nr_irqs, 0, NUMA_NO_NODE, NULL);
 	if (virq <= 0) {
 		pr_warn("Can't reserve IPI, failed to alloc descs\n");
 		return -ENOMEM;
diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
index 8731e1c..b8df4fc 100644
--- a/kernel/irq/irqdesc.c
+++ b/kernel/irq/irqdesc.c
@@ -223,7 +223,7 @@ static void free_desc(unsigned int irq)
 }
 
 static int alloc_descs(unsigned int start, unsigned int cnt, int node,
-		       struct module *owner)
+		       const struct cpumask *affinity, struct module *owner)
 {
 	struct irq_desc *desc;
 	int i;
@@ -333,6 +333,7 @@ static void free_desc(unsigned int irq)
 }
 
 static inline int alloc_descs(unsigned int start, unsigned int cnt, int node,
+			      const struct cpumask *affinity,
 			      struct module *owner)
 {
 	u32 i;
@@ -453,12 +454,15 @@ EXPORT_SYMBOL_GPL(irq_free_descs);
  * @cnt:	Number of consecutive irqs to allocate.
  * @node:	Preferred node on which the irq descriptor should be allocated
  * @owner:	Owning module (can be NULL)
+ * @affinity:	Optional pointer to an affinity mask which hints where the
+ *		irq descriptors should be allocated and which default
+ *		affinities to use
  *
  * Returns the first irq number or error code
  */
 int __ref
 __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
-		  struct module *owner)
+		  struct module *owner, const struct cpumask *affinity)
 {
 	int start, ret;
 
@@ -494,7 +498,7 @@ __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
 
 	bitmap_set(allocated_irqs, start, cnt);
 	mutex_unlock(&sparse_irq_lock);
-	return alloc_descs(start, cnt, node, owner);
+	return alloc_descs(start, cnt, node, affinity, owner);
 
 err:
 	mutex_unlock(&sparse_irq_lock);
@@ -512,7 +516,7 @@ EXPORT_SYMBOL_GPL(__irq_alloc_descs);
  */
 unsigned int irq_alloc_hwirqs(int cnt, int node)
 {
-	int i, irq = __irq_alloc_descs(-1, 0, cnt, node, NULL);
+	int i, irq = __irq_alloc_descs(-1, 0, cnt, node, NULL, NULL);
 
 	if (irq < 0)
 		return 0;
diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
index 8798b6c..79459b7 100644
--- a/kernel/irq/irqdomain.c
+++ b/kernel/irq/irqdomain.c
@@ -481,7 +481,7 @@ unsigned int irq_create_mapping(struct irq_domain *domain,
 	}
 
 	/* Allocate a virtual interrupt number */
-	virq = irq_domain_alloc_descs(-1, 1, hwirq, of_node_to_nid(of_node));
+	virq = irq_domain_alloc_descs(-1, 1, hwirq, of_node_to_nid(of_node), NULL);
 	if (virq <= 0) {
 		pr_debug("-> virq allocation failed\n");
 		return 0;
@@ -835,19 +835,23 @@ const struct irq_domain_ops irq_domain_simple_ops = {
 EXPORT_SYMBOL_GPL(irq_domain_simple_ops);
 
 int irq_domain_alloc_descs(int virq, unsigned int cnt, irq_hw_number_t hwirq,
-			   int node)
+			   int node, const struct cpumask *affinity)
 {
 	unsigned int hint;
 
 	if (virq >= 0) {
-		virq = irq_alloc_descs(virq, virq, cnt, node);
+		virq = __irq_alloc_descs(virq, virq, cnt, node, THIS_MODULE,
+					 affinity);
 	} else {
 		hint = hwirq % nr_irqs;
 		if (hint == 0)
 			hint++;
-		virq = irq_alloc_descs_from(hint, cnt, node);
-		if (virq <= 0 && hint > 1)
-			virq = irq_alloc_descs_from(1, cnt, node);
+		virq = __irq_alloc_descs(-1, hint, cnt, node, THIS_MODULE,
+					 affinity);
+		if (virq <= 0 && hint > 1) {
+			virq = __irq_alloc_descs(-1, 1, cnt, node, THIS_MODULE,
+						 affinity);
+		}
 	}
 
 	return virq;
@@ -1160,6 +1164,7 @@ int irq_domain_alloc_irqs_recursive(struct irq_domain *domain,
  * @node:	NUMA node id for memory allocation
  * @arg:	domain specific argument
  * @realloc:	IRQ descriptors have already been allocated if true
+ * @affinity:	Optional irq affinity mask for multiqueue devices
  *
  * Allocate IRQ numbers and initialized all data structures to support
  * hierarchy IRQ domains.
@@ -1175,7 +1180,7 @@ int irq_domain_alloc_irqs_recursive(struct irq_domain *domain,
  */
 int __irq_domain_alloc_irqs(struct irq_domain *domain, int irq_base,
 			    unsigned int nr_irqs, int node, void *arg,
-			    bool realloc)
+			    bool realloc, const struct cpumask *affinity)
 {
 	int i, ret, virq;
 
@@ -1193,7 +1198,8 @@ int __irq_domain_alloc_irqs(struct irq_domain *domain, int irq_base,
 	if (realloc && irq_base >= 0) {
 		virq = irq_base;
 	} else {
-		virq = irq_domain_alloc_descs(irq_base, nr_irqs, 0, node);
+		virq = irq_domain_alloc_descs(irq_base, nr_irqs, 0, node,
+					      affinity);
 		if (virq < 0) {
 			pr_debug("cannot allocate IRQ(base %d, count %d)\n",
 				 irq_base, nr_irqs);
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index 30658e9..ad0aac6 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -353,10 +353,11 @@ static int setup_affinity(struct irq_desc *desc, struct cpumask *mask)
 		return 0;
 
 	/*
-	 * Preserve an userspace affinity setup, but make sure that
-	 * one of the targets is online.
+	 * Preserve the managed affinity setting and an userspace affinity
+	 * setup, but make sure that one of the targets is online.
 	 */
-	if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
+	if (irqd_affinity_is_managed(&desc->irq_data) ||
+	    irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
 		if (cpumask_intersects(desc->irq_common_data.affinity,
 				       cpu_online_mask))
 			set = desc->irq_common_data.affinity;
diff --git a/kernel/irq/msi.c b/kernel/irq/msi.c
index eb5bf2b..58dbbac 100644
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -334,7 +334,8 @@ int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
 		ops->set_desc(&arg, desc);
 
 		virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
-					       dev_to_node(dev), &arg, false);
+					       dev_to_node(dev), &arg, false,
+					       NULL);
 		if (virq < 0) {
 			ret = -ENOSPC;
 			if (ops->handle_error)
-- 
2.1.4

[toc] | [prev] | [next] | [standalone]

#1436336 — [tip:irq/core] genirq: Add affinity hint to irq allocation

Fromtip-bot for Thomas Gleixner <tipbot@zytor.com>
Date2016-07-04 12:40 +0200
Subject[tip:irq/core] genirq: Add affinity hint to irq allocation
Message-ID<rR9sd-4tf-25@gated-at.bofh.it>
In reply to#1436329
Commit-ID:  06ee6d571f0e350253a8fc3492316b2be007fae2
Gitweb:     http://git.kernel.org/tip/06ee6d571f0e350253a8fc3492316b2be007fae2
Author:     Thomas Gleixner <tglx@linutronix.de>
AuthorDate: Mon, 4 Jul 2016 17:39:24 +0900
Committer:  Thomas Gleixner <tglx@linutronix.de>
CommitDate: Mon, 4 Jul 2016 12:25:13 +0200

genirq: Add affinity hint to irq allocation

Add an extra argument to the irq(domain) allocation functions, so we can hand
down affinity hints to the allocator. Thats necessary to implement proper
support for multiqueue devices.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: linux-block@vger.kernel.org
Cc: linux-pci@vger.kernel.org
Cc: linux-nvme@lists.infradead.org
Cc: axboe@fb.com
Cc: agordeev@redhat.com
Link: http://lkml.kernel.org/r/1467621574-8277-4-git-send-email-hch@lst.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---
 arch/sparc/kernel/irq_64.c     |  2 +-
 arch/x86/kernel/apic/io_apic.c |  5 +++--
 include/linux/irq.h            |  4 ++--
 include/linux/irqdomain.h      |  9 ++++++---
 kernel/irq/ipi.c               |  2 +-
 kernel/irq/irqdesc.c           | 12 ++++++++----
 kernel/irq/irqdomain.c         | 22 ++++++++++++++--------
 kernel/irq/manage.c            |  7 ++++---
 kernel/irq/msi.c               |  3 ++-
 9 files changed, 41 insertions(+), 25 deletions(-)

diff --git a/arch/sparc/kernel/irq_64.c b/arch/sparc/kernel/irq_64.c
index e22416c..34a7930 100644
--- a/arch/sparc/kernel/irq_64.c
+++ b/arch/sparc/kernel/irq_64.c
@@ -242,7 +242,7 @@ unsigned int irq_alloc(unsigned int dev_handle, unsigned int dev_ino)
 {
 	int irq;
 
-	irq = __irq_alloc_descs(-1, 1, 1, numa_node_id(), NULL);
+	irq = __irq_alloc_descs(-1, 1, 1, numa_node_id(), NULL, NULL);
 	if (irq <= 0)
 		goto out;
 
diff --git a/arch/x86/kernel/apic/io_apic.c b/arch/x86/kernel/apic/io_apic.c
index 446702e..7c4f90d 100644
--- a/arch/x86/kernel/apic/io_apic.c
+++ b/arch/x86/kernel/apic/io_apic.c
@@ -981,7 +981,7 @@ static int alloc_irq_from_domain(struct irq_domain *domain, int ioapic, u32 gsi,
 
 	return __irq_domain_alloc_irqs(domain, irq, 1,
 				       ioapic_alloc_attr_node(info),
-				       info, legacy);
+				       info, legacy, NULL);
 }
 
 /*
@@ -1014,7 +1014,8 @@ static int alloc_isa_irq_from_domain(struct irq_domain *domain,
 					  info->ioapic_pin))
 			return -ENOMEM;
 	} else {
-		irq = __irq_domain_alloc_irqs(domain, irq, 1, node, info, true);
+		irq = __irq_domain_alloc_irqs(domain, irq, 1, node, info, true,
+					      NULL);
 		if (irq >= 0) {
 			irq_data = irq_domain_get_irq_data(domain, irq);
 			data = irq_data->chip_data;
diff --git a/include/linux/irq.h b/include/linux/irq.h
index f607481..39ce46a 100644
--- a/include/linux/irq.h
+++ b/include/linux/irq.h
@@ -708,11 +708,11 @@ static inline struct cpumask *irq_data_get_affinity_mask(struct irq_data *d)
 unsigned int arch_dynirq_lower_bound(unsigned int from);
 
 int __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
-		struct module *owner);
+		      struct module *owner, const struct cpumask *affinity);
 
 /* use macros to avoid needing export.h for THIS_MODULE */
 #define irq_alloc_descs(irq, from, cnt, node)	\
-	__irq_alloc_descs(irq, from, cnt, node, THIS_MODULE)
+	__irq_alloc_descs(irq, from, cnt, node, THIS_MODULE, NULL)
 
 #define irq_alloc_desc(node)			\
 	irq_alloc_descs(-1, 0, 1, node)
diff --git a/include/linux/irqdomain.h b/include/linux/irqdomain.h
index f1f36e0..1aee0fb 100644
--- a/include/linux/irqdomain.h
+++ b/include/linux/irqdomain.h
@@ -39,6 +39,7 @@ struct irq_domain;
 struct of_device_id;
 struct irq_chip;
 struct irq_data;
+struct cpumask;
 
 /* Number of irqs reserved for a legacy isa controller */
 #define NUM_ISA_INTERRUPTS	16
@@ -217,7 +218,8 @@ extern struct irq_domain *irq_find_matching_fwspec(struct irq_fwspec *fwspec,
 						   enum irq_domain_bus_token bus_token);
 extern void irq_set_default_host(struct irq_domain *host);
 extern int irq_domain_alloc_descs(int virq, unsigned int nr_irqs,
-				  irq_hw_number_t hwirq, int node);
+				  irq_hw_number_t hwirq, int node,
+				  const struct cpumask *affinity);
 
 static inline struct fwnode_handle *of_node_to_fwnode(struct device_node *node)
 {
@@ -389,7 +391,7 @@ static inline struct irq_domain *irq_domain_add_hierarchy(struct irq_domain *par
 
 extern int __irq_domain_alloc_irqs(struct irq_domain *domain, int irq_base,
 				   unsigned int nr_irqs, int node, void *arg,
-				   bool realloc);
+				   bool realloc, const struct cpumask *affinity);
 extern void irq_domain_free_irqs(unsigned int virq, unsigned int nr_irqs);
 extern void irq_domain_activate_irq(struct irq_data *irq_data);
 extern void irq_domain_deactivate_irq(struct irq_data *irq_data);
@@ -397,7 +399,8 @@ extern void irq_domain_deactivate_irq(struct irq_data *irq_data);
 static inline int irq_domain_alloc_irqs(struct irq_domain *domain,
 			unsigned int nr_irqs, int node, void *arg)
 {
-	return __irq_domain_alloc_irqs(domain, -1, nr_irqs, node, arg, false);
+	return __irq_domain_alloc_irqs(domain, -1, nr_irqs, node, arg, false,
+				       NULL);
 }
 
 extern int irq_domain_alloc_irqs_recursive(struct irq_domain *domain,
diff --git a/kernel/irq/ipi.c b/kernel/irq/ipi.c
index 89b49f6..4fd2351 100644
--- a/kernel/irq/ipi.c
+++ b/kernel/irq/ipi.c
@@ -76,7 +76,7 @@ int irq_reserve_ipi(struct irq_domain *domain,
 		}
 	}
 
-	virq = irq_domain_alloc_descs(-1, nr_irqs, 0, NUMA_NO_NODE);
+	virq = irq_domain_alloc_descs(-1, nr_irqs, 0, NUMA_NO_NODE, NULL);
 	if (virq <= 0) {
 		pr_warn("Can't reserve IPI, failed to alloc descs\n");
 		return -ENOMEM;
diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
index 8731e1c..b8df4fc 100644
--- a/kernel/irq/irqdesc.c
+++ b/kernel/irq/irqdesc.c
@@ -223,7 +223,7 @@ static void free_desc(unsigned int irq)
 }
 
 static int alloc_descs(unsigned int start, unsigned int cnt, int node,
-		       struct module *owner)
+		       const struct cpumask *affinity, struct module *owner)
 {
 	struct irq_desc *desc;
 	int i;
@@ -333,6 +333,7 @@ static void free_desc(unsigned int irq)
 }
 
 static inline int alloc_descs(unsigned int start, unsigned int cnt, int node,
+			      const struct cpumask *affinity,
 			      struct module *owner)
 {
 	u32 i;
@@ -453,12 +454,15 @@ EXPORT_SYMBOL_GPL(irq_free_descs);
  * @cnt:	Number of consecutive irqs to allocate.
  * @node:	Preferred node on which the irq descriptor should be allocated
  * @owner:	Owning module (can be NULL)
+ * @affinity:	Optional pointer to an affinity mask which hints where the
+ *		irq descriptors should be allocated and which default
+ *		affinities to use
  *
  * Returns the first irq number or error code
  */
 int __ref
 __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
-		  struct module *owner)
+		  struct module *owner, const struct cpumask *affinity)
 {
 	int start, ret;
 
@@ -494,7 +498,7 @@ __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
 
 	bitmap_set(allocated_irqs, start, cnt);
 	mutex_unlock(&sparse_irq_lock);
-	return alloc_descs(start, cnt, node, owner);
+	return alloc_descs(start, cnt, node, affinity, owner);
 
 err:
 	mutex_unlock(&sparse_irq_lock);
@@ -512,7 +516,7 @@ EXPORT_SYMBOL_GPL(__irq_alloc_descs);
  */
 unsigned int irq_alloc_hwirqs(int cnt, int node)
 {
-	int i, irq = __irq_alloc_descs(-1, 0, cnt, node, NULL);
+	int i, irq = __irq_alloc_descs(-1, 0, cnt, node, NULL, NULL);
 
 	if (irq < 0)
 		return 0;
diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
index 8798b6c..79459b7 100644
--- a/kernel/irq/irqdomain.c
+++ b/kernel/irq/irqdomain.c
@@ -481,7 +481,7 @@ unsigned int irq_create_mapping(struct irq_domain *domain,
 	}
 
 	/* Allocate a virtual interrupt number */
-	virq = irq_domain_alloc_descs(-1, 1, hwirq, of_node_to_nid(of_node));
+	virq = irq_domain_alloc_descs(-1, 1, hwirq, of_node_to_nid(of_node), NULL);
 	if (virq <= 0) {
 		pr_debug("-> virq allocation failed\n");
 		return 0;
@@ -835,19 +835,23 @@ const struct irq_domain_ops irq_domain_simple_ops = {
 EXPORT_SYMBOL_GPL(irq_domain_simple_ops);
 
 int irq_domain_alloc_descs(int virq, unsigned int cnt, irq_hw_number_t hwirq,
-			   int node)
+			   int node, const struct cpumask *affinity)
 {
 	unsigned int hint;
 
 	if (virq >= 0) {
-		virq = irq_alloc_descs(virq, virq, cnt, node);
+		virq = __irq_alloc_descs(virq, virq, cnt, node, THIS_MODULE,
+					 affinity);
 	} else {
 		hint = hwirq % nr_irqs;
 		if (hint == 0)
 			hint++;
-		virq = irq_alloc_descs_from(hint, cnt, node);
-		if (virq <= 0 && hint > 1)
-			virq = irq_alloc_descs_from(1, cnt, node);
+		virq = __irq_alloc_descs(-1, hint, cnt, node, THIS_MODULE,
+					 affinity);
+		if (virq <= 0 && hint > 1) {
+			virq = __irq_alloc_descs(-1, 1, cnt, node, THIS_MODULE,
+						 affinity);
+		}
 	}
 
 	return virq;
@@ -1160,6 +1164,7 @@ int irq_domain_alloc_irqs_recursive(struct irq_domain *domain,
  * @node:	NUMA node id for memory allocation
  * @arg:	domain specific argument
  * @realloc:	IRQ descriptors have already been allocated if true
+ * @affinity:	Optional irq affinity mask for multiqueue devices
  *
  * Allocate IRQ numbers and initialized all data structures to support
  * hierarchy IRQ domains.
@@ -1175,7 +1180,7 @@ int irq_domain_alloc_irqs_recursive(struct irq_domain *domain,
  */
 int __irq_domain_alloc_irqs(struct irq_domain *domain, int irq_base,
 			    unsigned int nr_irqs, int node, void *arg,
-			    bool realloc)
+			    bool realloc, const struct cpumask *affinity)
 {
 	int i, ret, virq;
 
@@ -1193,7 +1198,8 @@ int __irq_domain_alloc_irqs(struct irq_domain *domain, int irq_base,
 	if (realloc && irq_base >= 0) {
 		virq = irq_base;
 	} else {
-		virq = irq_domain_alloc_descs(irq_base, nr_irqs, 0, node);
+		virq = irq_domain_alloc_descs(irq_base, nr_irqs, 0, node,
+					      affinity);
 		if (virq < 0) {
 			pr_debug("cannot allocate IRQ(base %d, count %d)\n",
 				 irq_base, nr_irqs);
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index 30658e9..ad0aac6 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -353,10 +353,11 @@ static int setup_affinity(struct irq_desc *desc, struct cpumask *mask)
 		return 0;
 
 	/*
-	 * Preserve an userspace affinity setup, but make sure that
-	 * one of the targets is online.
+	 * Preserve the managed affinity setting and an userspace affinity
+	 * setup, but make sure that one of the targets is online.
 	 */
-	if (irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
+	if (irqd_affinity_is_managed(&desc->irq_data) ||
+	    irqd_has_set(&desc->irq_data, IRQD_AFFINITY_SET)) {
 		if (cpumask_intersects(desc->irq_common_data.affinity,
 				       cpu_online_mask))
 			set = desc->irq_common_data.affinity;
diff --git a/kernel/irq/msi.c b/kernel/irq/msi.c
index eb5bf2b..58dbbac 100644
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -334,7 +334,8 @@ int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
 		ops->set_desc(&arg, desc);
 
 		virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
-					       dev_to_node(dev), &arg, false);
+					       dev_to_node(dev), &arg, false,
+					       NULL);
 		if (virq < 0) {
 			ret = -ENOSPC;
 			if (ops->handle_error)

[toc] | [prev] | [next] | [standalone]

#1436345

FromThomas Gleixner <tglx@linutronix.de>
Date2016-07-04 12:40 +0200
Message-ID<rR9sd-4tf-13@gated-at.bofh.it>
In reply to#1436298
On Mon, 4 Jul 2016, Christoph Hellwig wrote:

> This series enhances the irq and PCI code to allow spreading around MSI and
> MSI-X vectors so that they have per-cpu affinity if possible, or at least
> per-node.  For that it takes the algorithm from blk-mq, moves it to
> a common place, and makes it available through a vastly simplified PCI
> interrupt allocation API.  It then switches blk-mq to be able to pick up
> the queue mapping from the device if available, and demonstrates all this
> using the NVMe driver.
> 
> Compared to the last posting the core IRQ changes are stable and it would
> be great to get them merged int the tip tree.  The two PCI patches have
> been completely rewritten after feedback from Alexander, while the block
> changes have also been stable.

I applied the first 6 patches. They are available in a seperate branch:

   git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git irq/for-block

and integrated into the irq/core branch. irq/for-block can be pulled from the
block folks if required.

Thanks,

	tglx

[toc] | [prev] | [standalone]

Page 2 of 2 — ← Prev page 1 [2]

Back to top | Article view | linux.kernel

csiph-web