Skip to contentSkip to conversation

genirq/affinity: add .calc_sets for improving IRQ allocation & spread

PATCH Superseded Ming Lei 6 participants 24 messages Feb 13, 2019 – Feb 19, 2019

Subsystems (4)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Ming Lei

author
#0 lore raw

v3 cover genirq/affinity: add .calc_sets for improving IRQ allocation & spread

Hi,

Currently pre-caculated set vectors are provided by driver for allocating & spread vectors. This way only works when drivers passes same 'max_vecs' and 'min_vecs' to pci_alloc_irq_vectors_affinity(), also requires driver to retry the allocating & spread.

As Bjorn and Keith mentioned, the current usage & interface for irq sets is a bit awkward because the retrying should have been avoided by providing one resonable 'min_vecs'. However, if 'min_vecs' isn't same with 'max_vecs', number of the allocated vectors is unknown before calling pci_alloc_irq_vectors_affinity(), then each set's vectors can't be pre-caculated.

Add a new callback of .calc_sets into 'struct irq_affinity' so that driver can caculate set vectors after IRQ vector is allocated and before spread IRQ vectors.

V3:
- don't mark 'affd' as const in related functions
- add sanity check on affd->nr_sets
- remove the local variable of 'nr_sets' in irq_create_affinity_masks
- set .nr_sets as 2 in nvme
- update comment in msi.c

V2:
- add .calc_sets instead of .setup_affinity() which is easy to be abused by drivers

Ming Lei (5):
genirq/affinity: don't mark 'affd' as const genirq/affinity: store irq set vectors in 'struct irq_affinity' genirq/affinity: add new callback for caculating set vectors
nvme-pci: avoid irq allocation retrying via .calc_sets genirq/affinity: Document .calc_sets as required in case of multiple
sets

 drivers/nvme/host/pci.c   | 63 ++++++++++++-----------------------------------
 drivers/pci/msi.c         | 34 ++++++++++++++-----------
 include/linux/interrupt.h | 13 +++++++---
 include/linux/pci.h       |  4 +--
 kernel/irq/affinity.c     | 26 ++++++++++++-------
 5 files changed, 64 insertions(+), 76 deletions(-)
Signature · 1 line
2.9.5
author
#1 lore raw

PATCH v3 1/5 genirq/affinity: don't mark 'affd' as const

Currently all parameters in 'affd' are read-only, so 'affd' is marked as const in both pci_alloc_irq_vectors_affinity() and irq_create_affinity_masks().

We have to ask driver to re-caculate set vectors after the whole IRQ vectors are allocated later, and the result needs to be stored in 'affd'. Also both the two interfaces are core APIs, which should be trusted.

So don't mark 'affd' as const both pci_alloc_irq_vectors_affinity() and irq_create_affinity_masks().

Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/pci/msi.c         | 18 +++++++++---------
 include/linux/interrupt.h |  2 +-
 include/linux/pci.h       |  4 ++--
 kernel/irq/affinity.c     |  2 +-
 4 files changed, 13 insertions(+), 13 deletions(-)
drivers/pci/msi.c +9 -9
index 4c0b47867258..96978459e2a0 100644
@@ -532,7 +532,7 @@ static int populate_msi_sysfs(struct pci_dev *pdev)
 }
 
 static struct msi_desc *
-msi_setup_entry(struct pci_dev *dev, int nvec, const struct irq_affinity *affd)
+msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
 {
 	struct irq_affinity_desc *masks = NULL;
 	struct msi_desc *entry;
@@ -597,7 +597,7 @@ static int msi_verify_entries(struct pci_dev *dev)
  * which could have been allocated.
  */
 static int msi_capability_init(struct pci_dev *dev, int nvec,
-			       const struct irq_affinity *affd)
+			       struct irq_affinity *affd)
 {
 	struct msi_desc *entry;
 	int ret;
@@ -669,7 +669,7 @@ static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
 
 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
 			      struct msix_entry *entries, int nvec,
-			      const struct irq_affinity *affd)
+			      struct irq_affinity *affd)
 {
 	struct irq_affinity_desc *curmsk, *masks = NULL;
 	struct msi_desc *entry;
@@ -736,7 +736,7 @@ static void msix_program_entries(struct pci_dev *dev,
  * requested MSI-X entries with allocated irqs or non-zero for otherwise.
  **/
 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
-				int nvec, const struct irq_affinity *affd)
+				int nvec, struct irq_affinity *affd)
 {
 	int ret;
 	u16 control;
@@ -932,7 +932,7 @@ int pci_msix_vec_count(struct pci_dev *dev)
 EXPORT_SYMBOL(pci_msix_vec_count);
 
 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
-			     int nvec, const struct irq_affinity *affd)
+			     int nvec, struct irq_affinity *affd)
 {
 	int nr_entries;
 	int i, j;
@@ -1018,7 +1018,7 @@ int pci_msi_enabled(void)
 EXPORT_SYMBOL(pci_msi_enabled);
 
 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
-				  const struct irq_affinity *affd)
+				  struct irq_affinity *affd)
 {
 	int nvec;
 	int rc;
@@ -1086,7 +1086,7 @@ EXPORT_SYMBOL(pci_enable_msi);
 
 static int __pci_enable_msix_range(struct pci_dev *dev,
 				   struct msix_entry *entries, int minvec,
-				   int maxvec, const struct irq_affinity *affd)
+				   int maxvec, struct irq_affinity *affd)
 {
 	int rc, nvec = maxvec;
 
@@ -1165,9 +1165,9 @@ EXPORT_SYMBOL(pci_enable_msix_range);
  */
 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
 				   unsigned int max_vecs, unsigned int flags,
-				   const struct irq_affinity *affd)
+				   struct irq_affinity *affd)
 {
-	static const struct irq_affinity msi_default_affd;
+	struct irq_affinity msi_default_affd = {0};
 	int msix_vecs = -ENOSPC;
 	int msi_vecs = -ENOSPC;
 
include/linux/interrupt.h +1 -1
index 7c9434652f36..1ed1014c9684 100644
@@ -332,7 +332,7 @@ extern int
 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify);
 
 struct irq_affinity_desc *
-irq_create_affinity_masks(int nvec, const struct irq_affinity *affd);
+irq_create_affinity_masks(int nvec, struct irq_affinity *affd);
 
 int irq_calc_affinity_vectors(int minvec, int maxvec, const struct irq_affinity *affd);
 
include/linux/pci.h +2 -2
index 40b327b814aa..4eca42cf611b 100644
@@ -1396,7 +1396,7 @@ static inline int pci_enable_msix_exact(struct pci_dev *dev,
 }
 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
 				   unsigned int max_vecs, unsigned int flags,
-				   const struct irq_affinity *affd);
+				   struct irq_affinity *affd);
 
 void pci_free_irq_vectors(struct pci_dev *dev);
 int pci_irq_vector(struct pci_dev *dev, unsigned int nr);
@@ -1422,7 +1422,7 @@ static inline int pci_enable_msix_exact(struct pci_dev *dev,
 static inline int
 pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
 			       unsigned int max_vecs, unsigned int flags,
-			       const struct irq_affinity *aff_desc)
+			       struct irq_affinity *aff_desc)
 {
 	if ((flags & PCI_IRQ_LEGACY) && min_vecs == 1 && dev->irq)
 		return 1;
kernel/irq/affinity.c +1 -1
index 118b66d64a53..9200d3b26f7d 100644
@@ -239,7 +239,7 @@ static int irq_build_affinity_masks(const struct irq_affinity *affd,
  * Returns the irq_affinity_desc pointer or NULL if allocation failed.
  */
 struct irq_affinity_desc *
-irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
+irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 {
 	int affvecs = nvecs - affd->pre_vectors - affd->post_vectors;
 	int curvec, usedvecs;
Signature · 1 line
2.9.5
#2 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

Quoted 2 lines · Ming Lei

Currently all parameters in 'affd' are read-only, so 'affd' is marked as const in both pci_alloc_irq_vectors_affinity() and irq_create_affinity_masks().

s/all parameters in 'affd'/the contents of '*affd'/

Quoted 3 lines

We have to ask driver to re-caculate set vectors after the whole IRQ vectors are allocated later, and the result needs to be stored in 'affd'. Also both the two interfaces are core APIs, which should be trusted.

s/re-caculate/recalculate/
s/stored in 'affd'/stored in '*affd'/
s/both the two/both/

This is a little confusing because you're talking about both "IRQ vectors" and these other "set vectors", which I think are different things. I assume the "set vectors" are cpumasks showing the affinity of the IRQ vectors with some CPUs?

AFAICT, *this* patch doesn't add anything that writes to *affd. I think the removal of "const" should be in the same patch that makes the removal necessary.

Quoted 141 lines

So don't mark 'affd' as const both pci_alloc_irq_vectors_affinity() and irq_create_affinity_masks().

Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/pci/msi.c         | 18 +++++++++---------
 include/linux/interrupt.h |  2 +-
 include/linux/pci.h       |  4 ++--
 kernel/irq/affinity.c     |  2 +-
 4 files changed, 13 insertions(+), 13 deletions(-)
drivers/pci/msi.c +9 -9
index 4c0b47867258..96978459e2a0 100644
@@ -532,7 +532,7 @@ static int populate_msi_sysfs(struct pci_dev *pdev)
 }
 
 static struct msi_desc *
-msi_setup_entry(struct pci_dev *dev, int nvec, const struct irq_affinity *affd)
+msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
 {
 	struct irq_affinity_desc *masks = NULL;
 	struct msi_desc *entry;
@@ -597,7 +597,7 @@ static int msi_verify_entries(struct pci_dev *dev)
  * which could have been allocated.
  */
 static int msi_capability_init(struct pci_dev *dev, int nvec,
-			       const struct irq_affinity *affd)
+			       struct irq_affinity *affd)
 {
 	struct msi_desc *entry;
 	int ret;
@@ -669,7 +669,7 @@ static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
 
 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
 			      struct msix_entry *entries, int nvec,
-			      const struct irq_affinity *affd)
+			      struct irq_affinity *affd)
 {
 	struct irq_affinity_desc *curmsk, *masks = NULL;
 	struct msi_desc *entry;
@@ -736,7 +736,7 @@ static void msix_program_entries(struct pci_dev *dev,
  * requested MSI-X entries with allocated irqs or non-zero for otherwise.
  **/
 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
-				int nvec, const struct irq_affinity *affd)
+				int nvec, struct irq_affinity *affd)
 {
 	int ret;
 	u16 control;
@@ -932,7 +932,7 @@ int pci_msix_vec_count(struct pci_dev *dev)
 EXPORT_SYMBOL(pci_msix_vec_count);
 
 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
-			     int nvec, const struct irq_affinity *affd)
+			     int nvec, struct irq_affinity *affd)
 {
 	int nr_entries;
 	int i, j;
@@ -1018,7 +1018,7 @@ int pci_msi_enabled(void)
 EXPORT_SYMBOL(pci_msi_enabled);
 
 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
-				  const struct irq_affinity *affd)
+				  struct irq_affinity *affd)
 {
 	int nvec;
 	int rc;
@@ -1086,7 +1086,7 @@ EXPORT_SYMBOL(pci_enable_msi);
 
 static int __pci_enable_msix_range(struct pci_dev *dev,
 				   struct msix_entry *entries, int minvec,
-				   int maxvec, const struct irq_affinity *affd)
+				   int maxvec, struct irq_affinity *affd)
 {
 	int rc, nvec = maxvec;
 
@@ -1165,9 +1165,9 @@ EXPORT_SYMBOL(pci_enable_msix_range);
  */
 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
 				   unsigned int max_vecs, unsigned int flags,
-				   const struct irq_affinity *affd)
+				   struct irq_affinity *affd)
 {
-	static const struct irq_affinity msi_default_affd;
+	struct irq_affinity msi_default_affd = {0};
 	int msix_vecs = -ENOSPC;
 	int msi_vecs = -ENOSPC;
 
include/linux/interrupt.h +1 -1
index 7c9434652f36..1ed1014c9684 100644
@@ -332,7 +332,7 @@ extern int
 irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify);
 
 struct irq_affinity_desc *
-irq_create_affinity_masks(int nvec, const struct irq_affinity *affd);
+irq_create_affinity_masks(int nvec, struct irq_affinity *affd);
 
 int irq_calc_affinity_vectors(int minvec, int maxvec, const struct irq_affinity *affd);
 
include/linux/pci.h +2 -2
index 40b327b814aa..4eca42cf611b 100644
@@ -1396,7 +1396,7 @@ static inline int pci_enable_msix_exact(struct pci_dev *dev,
 }
 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
 				   unsigned int max_vecs, unsigned int flags,
-				   const struct irq_affinity *affd);
+				   struct irq_affinity *affd);
 
 void pci_free_irq_vectors(struct pci_dev *dev);
 int pci_irq_vector(struct pci_dev *dev, unsigned int nr);
@@ -1422,7 +1422,7 @@ static inline int pci_enable_msix_exact(struct pci_dev *dev,
 static inline int
 pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
 			       unsigned int max_vecs, unsigned int flags,
-			       const struct irq_affinity *aff_desc)
+			       struct irq_affinity *aff_desc)
 {
 	if ((flags & PCI_IRQ_LEGACY) && min_vecs == 1 && dev->irq)
 		return 1;
kernel/irq/affinity.c +1 -1
index 118b66d64a53..9200d3b26f7d 100644
@@ -239,7 +239,7 @@ static int irq_build_affinity_masks(const struct irq_affinity *affd,
  * Returns the irq_affinity_desc pointer or NULL if allocation failed.
  */
 struct irq_affinity_desc *
-irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd)
+irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 {
 	int affvecs = nvecs - affd->pre_vectors - affd->post_vectors;
 	int curvec, usedvecs;
Signature · 1 line
2.9.5
#3 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

Quoted 10 lines · Bjorn Helgaas
Quoted 2 lines · Ming Lei

Currently all parameters in 'affd' are read-only, so 'affd' is marked as const in both pci_alloc_irq_vectors_affinity() and irq_create_affinity_masks().

s/all parameters in 'affd'/the contents of '*affd'/

Quoted 3 lines

We have to ask driver to re-caculate set vectors after the whole IRQ vectors are allocated later, and the result needs to be stored in 'affd'. Also both the two interfaces are core APIs, which should be trusted.

Quoted 8 lines

s/re-caculate/recalculate/
s/stored in 'affd'/stored in '*affd'/
s/both the two/both/

This is a little confusing because you're talking about both "IRQ vectors" and these other "set vectors", which I think are different things. I assume the "set vectors" are cpumasks showing the affinity of the IRQ vectors with some CPUs?

I think we should drop the whole vector wording completely.

The driver does not care about vectors, it only cares about a block of interrupt numbers. These numbers are kernel managed and the interrupts just happen to have a CPU vector assigned at some point. Depending on the CPU architecture the underlying mechanism might not even be named vector.

Quoted 3 lines

AFAICT, *this* patch doesn't add anything that writes to *affd. I think the removal of "const" should be in the same patch that makes the removal necessary.

So this should be:

The interrupt affinity spreading mechanism supports to spread out affinities for one or more interrupt sets. A interrupt set contains one or more interrupts. Each set is mapped to a specific functionality of a device, e.g. general I/O queues and read I/O queus of multiqueue block devices.

The number of interrupts per set is defined by the driver. It depends on the total number of available interrupts for the device, which is determined by the PCI capabilites and the availability of underlying CPU resources, and the number of queues which the device provides and the driver wants to instantiate.

The driver passes initial configuration for the interrupt allocation via a pointer to struct affinity_desc.

Right now the allocation mechanism is complex as it requires to have a loop in the driver to determine the maximum number of interrupts which are provided by the PCI capabilities and the underlying CPU resources. This loop would have to be replicated in every driver which wants to utilize this mechanism. That's unwanted code duplication and error prone.

In order to move this into generic facilities it is required to have a mechanism, which allows the recalculation of the interrupt sets and their size, in the core code. As the core code does not have any knowledge about the underlying device, a driver specific callback will be added to struct affinity_desc, which will be invoked by the core code. The callback will get the number of available interupts as an argument, so the driver can calculate the corresponding number and size of interrupt sets.

To support this, two modifications for the handling of struct affinity_desc are required:

1) The (optional) interrupt sets size information is contained in a separate array of integers and struct affinity_desc contains a pointer to it.

This is cumbersome and as the maximum number of interrupt sets is small, there is no reason to have separate storage. Moving the size array into struct affinity_desc avoids indirections makes the code simpler.

2) At the moment the struct affinity_desc pointer which is handed in from the driver and passed through to several core functions is marked 'const'.

With the upcoming callback to recalculate the number and size of interrupt sets, it's necessary to remove the 'const' qualifier. Otherwise the callback would not be able to update the data.

Move the set size array into struct affinity_desc as a first preparatory step. The removal of the 'const' qualifier will be done when adding the callback.

IOW, The first patch moves the set array into the struct itself.

The second patch introduces the callback and removes the 'const' qualifier. I wouldn't mind to have the same changelog duplicated (+/- the last two paragraphs which need some update of course).

Thanks,

tglx

#4 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

Quoted 15 lines · Thomas Gleixner
Quoted 5 lines · Bjorn Helgaas
Quoted 3 lines · Ming Lei

We have to ask driver to re-caculate set vectors after the whole IRQ vectors are allocated later, and the result needs to be stored in 'affd'. Also both the two interfaces are core APIs, which should be trusted.

Quoted 8 lines

s/re-caculate/recalculate/
s/stored in 'affd'/stored in '*affd'/
s/both the two/both/

This is a little confusing because you're talking about both "IRQ vectors" and these other "set vectors", which I think are different things. I assume the "set vectors" are cpumasks showing the affinity of the IRQ vectors with some CPUs?

Quoted 6 lines

I think we should drop the whole vector wording completely.

The driver does not care about vectors, it only cares about a block of interrupt numbers. These numbers are kernel managed and the interrupts just happen to have a CPU vector assigned at some point. Depending on the CPU architecture the underlying mechanism might not even be named vector.

Perhaps longer term we could move affinity mask creation from the irq subsystem into a more generic library. Interrupts aren't the only resource that want to spread across CPUs. For example, blk-mq has it's own implementation to for polled queues, so I think a non-irq specific implementation would be a nice addition to the kernel lib.

#5 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

Quoted 23 lines · Keith Busch
Quoted 15 lines · Thomas Gleixner
Quoted 5 lines · Bjorn Helgaas
Quoted 3 lines · Ming Lei

We have to ask driver to re-caculate set vectors after the whole IRQ vectors are allocated later, and the result needs to be stored in 'affd'. Also both the two interfaces are core APIs, which should be trusted.

Quoted 8 lines

s/re-caculate/recalculate/
s/stored in 'affd'/stored in '*affd'/
s/both the two/both/

This is a little confusing because you're talking about both "IRQ vectors" and these other "set vectors", which I think are different things. I assume the "set vectors" are cpumasks showing the affinity of the IRQ vectors with some CPUs?

Quoted 6 lines

I think we should drop the whole vector wording completely.

The driver does not care about vectors, it only cares about a block of interrupt numbers. These numbers are kernel managed and the interrupts just happen to have a CPU vector assigned at some point. Depending on the CPU architecture the underlying mechanism might not even be named vector.

Quoted 5 lines

Perhaps longer term we could move affinity mask creation from the irq subsystem into a more generic library. Interrupts aren't the only resource that want to spread across CPUs. For example, blk-mq has it's own implementation to for polled queues, so I think a non-irq specific implementation would be a nice addition to the kernel lib.

Agreed. There is nothing interrupt specific in that code aside of some name choices.

Btw, while I have your attention. There popped up an issue recently related to that affinity logic.

The current implementation fails when:

        /*
         * If there aren't any vectors left after applying the pre/post
         * vectors don't bother with assigning affinity.
	 */
	if (nvecs == affd->pre_vectors + affd->post_vectors)
    		return NULL;

Now the discussion arised, that in that case the affinity sets are not allocated and filled in for the pre/post vectors, but somehow the underlying device still works and later on triggers the warning in the blk-mq code because the MSI entries do not have affinity information attached.

Sure, we could make that work, but there are several issues:

1) irq_create_affinity_masks() has another reason to return NULL:
memory allocation fails.

2) Does it make sense at all.

Right now the PCI allocator ignores the NULL return and proceeds without setting any affinities. As a consequence nothing is managed and everything happens to work.

But that happens to work is more by chance than by design and the warning is bogus if this is an expected mode of operation.

We should address these points in some way.

Thanks,

tglx

#6 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

Quoted 25 lines · Thomas Gleixner

Btw, while I have your attention. There popped up an issue recently related to that affinity logic.

The current implementation fails when:

        /*
         * If there aren't any vectors left after applying the pre/post
         * vectors don't bother with assigning affinity.
	 */
	if (nvecs == affd->pre_vectors + affd->post_vectors)
    		return NULL;

Now the discussion arised, that in that case the affinity sets are not allocated and filled in for the pre/post vectors, but somehow the underlying device still works and later on triggers the warning in the blk-mq code because the MSI entries do not have affinity information attached.

Sure, we could make that work, but there are several issues:

1) irq_create_affinity_masks() has another reason to return NULL:
memory allocation fails.

2) Does it make sense at all.

Quoted 8 lines

Right now the PCI allocator ignores the NULL return and proceeds without setting any affinities. As a consequence nothing is managed and everything happens to work.

But that happens to work is more by chance than by design and the warning is bogus if this is an expected mode of operation.

We should address these points in some way.

Ah, yes, that's a mistake in the nvme driver. It is assuming IO queues are always on managed interrupts, but that's not true if when only 1 vector could be allocated. This should be an appropriate fix to the warning:

---

drivers/nvme/host/pci.c +1 -1
index 022ea1ee63f8..f2ccebe1c926 100644
@@ -506,7 +506,7 @@ static int nvme_pci_map_queues(struct blk_mq_tag_set *set)
 		 * affinity), so use the regular blk-mq cpu mapping
 		 */
 		map->queue_offset = qoff;
-		if (i != HCTX_TYPE_POLL)
+		if (i != HCTX_TYPE_POLL && dev->num_vecs > 1)
 			blk_mq_pci_map_queues(map, to_pci_dev(dev->dev), offset);
 		else
 			blk_mq_map_queues(map);

--

#7 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

On Wed, 13 Feb 2019, Keith Busch wrote:

Cc+ Huacai Chen

Quoted 30 lines · Thomas Gleixner
Quoted 29 lines

Btw, while I have your attention. There popped up an issue recently related to that affinity logic.

The current implementation fails when:

        /*
         * If there aren't any vectors left after applying the pre/post
         * vectors don't bother with assigning affinity.
	 */
	if (nvecs == affd->pre_vectors + affd->post_vectors)
    		return NULL;

Now the discussion arised, that in that case the affinity sets are not allocated and filled in for the pre/post vectors, but somehow the underlying device still works and later on triggers the warning in the blk-mq code because the MSI entries do not have affinity information attached.

Sure, we could make that work, but there are several issues:

1) irq_create_affinity_masks() has another reason to return NULL:
memory allocation fails.

2) Does it make sense at all.

Right now the PCI allocator ignores the NULL return and proceeds without setting any affinities. As a consequence nothing is managed and everything happens to work.

Quoted 8 lines
Quoted 4 lines

But that happens to work is more by chance than by design and the warning is bogus if this is an expected mode of operation.

We should address these points in some way.

Ah, yes, that's a mistake in the nvme driver. It is assuming IO queues are always on managed interrupts, but that's not true if when only 1 vector could be allocated. This should be an appropriate fix to the warning:

Looks correct. Chen, can you please test that?

Quoted 16 lines

---

drivers/nvme/host/pci.c +1 -2
index 022ea1ee63f8..f2ccebe1c926 100644
@@ -506,7 +506,7 @@ static int nvme_pci_map_queues(struct blk_mq_tag_set *set)
 		 * affinity), so use the regular blk-mq cpu mapping
 		 */
 		map->queue_offset = qoff;
-		if (i != HCTX_TYPE_POLL)
+		if (i != HCTX_TYPE_POLL && dev->num_vecs > 1)
 			blk_mq_pci_map_queues(map, to_pci_dev(dev->dev), offset);
 		else
 			blk_mq_map_queues(map);
--
#8 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

I'll test next week, but 4.19 has the same problem, how to fix that for 4.19?

Huacai

------------------ Original ------------------

From:  "Thomas Gleixner"<[email protected]>;
Date:  Thu, Feb 14, 2019 04:50 PM
To:  "Keith Busch"<keith.busch at intel.com>;

Cc: "Bjorn Helgaas"<helgaas at kernel.org>; "Jens Axboe"<axboe at kernel.dk>; "Sagi Grimberg"<sagi at grimberg.me>; "linux-pci"<linux-pci at vger.kernel.org>; "LKML"<linux-kernel at vger.kernel.org>; "linux-nvme"<linux-nvme at lists.infradead.org>; "Ming Lei"<ming.lei at redhat.com>; "linux-block"<linux-block at vger.kernel.org>; "Christoph Hellwig"<hch at lst.de>; "Huacai Chen"<chenhc at lemote.com>;
Subject: Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

On Wed, 13 Feb 2019, Keith Busch wrote:

Cc+ Huacai Chen

Quoted 30 lines · Thomas Gleixner
Quoted 29 lines

Btw, while I have your attention. There popped up an issue recently related to that affinity logic.

The current implementation fails when:

/*
* If there aren't any vectors left after applying the pre/post
* vectors don't bother with assigning affinity.
*/
if (nvecs == affd->pre_vectors + affd->post_vectors)

    return NULL;

Now the discussion arised, that in that case the affinity sets are not allocated and filled in for the pre/post vectors, but somehow the underlying device still works and later on triggers the warning in the blk-mq code because the MSI entries do not have affinity information attached.

Sure, we could make that work, but there are several issues:

1) irq_create_affinity_masks() has another reason to return NULL:
memory allocation fails.

2) Does it make sense at all.

Right now the PCI allocator ignores the NULL return and proceeds without setting any affinities. As a consequence nothing is managed and everything happens to work.

Quoted 8 lines
Quoted 4 lines

But that happens to work is more by chance than by design and the warning is bogus if this is an expected mode of operation.

We should address these points in some way.

Ah, yes, that's a mistake in the nvme driver. It is assuming IO queues are always on managed interrupts, but that's not true if when only 1 vector could be allocated. This should be an appropriate fix to the warning:

Looks correct. Chen, can you please test that?

Quoted 16 lines

---

drivers/nvme/host/pci.c +1 -2
index 022ea1ee63f8..f2ccebe1c926 100644
@@ -506,7 +506,7 @@ static int nvme_pci_map_queues(struct blk_mq_tag_set *set)
 * affinity), so use the regular blk-mq cpu mapping
 */
 map->queue_offset = qoff;
-	if (i != HCTX_TYPE_POLL)
+	if (i != HCTX_TYPE_POLL && dev->num_vecs > 1)
 blk_mq_pci_map_queues(map, to_pci_dev(dev->dev), offset);
 else
 blk_mq_map_queues(map);
--
#9 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

On Thu, 14 Feb 2019, ??? wrote:

Please do not top post....

Quoted 1 line

I'll test next week, but 4.19 has the same problem, how to fix that for 4.19?

By applying the very same patch perhaps?

Thanks,

tglx

#10 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

My lemote.com email is too weak, I'll stop top post because this time I've cc-ed my gmail.

I've tested, this patch can fix the nvme problem, but it can't be applied to 4.19 because of different context. And, I still think my original solution (genirq/affinity: Assign default affinity to pre/post vectors) is correct. There may be similar problems except nvme.

Huacai

------------------ Original ------------------

From:  "Thomas Gleixner"<[email protected]>;
Date:  Thu, Feb 14, 2019 09:31 PM
To:  "???"<chenhc at lemote.com>;

Cc: "Keith Busch"<keith.busch at intel.com>; "Bjorn Helgaas"<helgaas at kernel.org>; "Jens Axboe"<axboe at kernel.dk>; "Sagi Grimberg"<sagi at grimberg.me>; "linux-pci"<linux-pci at vger.kernel.org>; "LKML"<linux-kernel at vger.kernel.org>; "linux-nvme"<linux-nvme at lists.infradead.org>; "Ming Lei"<ming.lei at redhat.com>; "linux-block"<linux-block at vger.kernel.org>; "Christoph Hellwig"<hch at lst.de>;
Subject: Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

On Thu, 14 Feb 2019, ??? wrote:

Please do not top post....

Quoted 1 line

I'll test next week, but 4.19 has the same problem, how to fix that for 4.19?

By applying the very same patch perhaps?

Thanks,

tglx

#11 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

Quoted 5 lines · ???

I've tested, this patch can fix the nvme problem, but it can't be applied to 4.19 because of different context. And, I still think my original solution (genirq/affinity: Assign default affinity to pre/post vectors) is correct. There may be similar problems except nvme.

As I explained you in great length, it is not correct because it's just papering over the underlying problem.

Please stop advertising semantically broken duct tape solutions.

Thanks,

tglx

#12 lore raw

Re: [PATCH V3 1/5] genirq/affinity: don't mark 'affd' as const

In reply to 陈华才
Quoted 4 lines · ???

I've tested, this patch can fix the nvme problem, but it can't be applied to 4.19 because of different context. And, I still think my original solution (genirq/affinity: Assign default affinity to pre/post vectors) is correct. There may be similar problems except nvme.

I'll send a proper patch to fix the warning. It's always a driver bug if they try to extract affinity from a non-managed irq, but I don't think this particular nvme instance is very urgent as the end result is the same, just without the warning.

author
#13 lore raw

PATCH v3 2/5 genirq/affinity: store irq set vectors in 'struct irq_affinity'

In reply to Ming Lei

Currently the array of irq set vectors is provided by driver.

irq_create_affinity_masks() can be simplied a bit by treating the non-irq-set case as single irq set.

So move this array into 'struct irq_affinity', and pre-define the max set number as 4, which should be enough for normal cases.

Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/nvme/host/pci.c   |  5 ++---
 include/linux/interrupt.h |  6 ++++--
 kernel/irq/affinity.c     | 18 +++++++++++-------
 3 files changed, 17 insertions(+), 12 deletions(-)
drivers/nvme/host/pci.c +2 -3
index 022ea1ee63f8..0086bdf80ea1 100644
@@ -2081,12 +2081,11 @@ static void nvme_calc_io_queues(struct nvme_dev *dev, unsigned int irq_queues)
 static int nvme_setup_irqs(struct nvme_dev *dev, unsigned int nr_io_queues)
 {
 	struct pci_dev *pdev = to_pci_dev(dev->dev);
-	int irq_sets[2];
 	struct irq_affinity affd = {
 		.pre_vectors = 1,
-		.nr_sets = ARRAY_SIZE(irq_sets),
-		.sets = irq_sets,
+		.nr_sets = 2,
 	};
+	int *irq_sets = affd.set_vectors;
 	int result = 0;
 	unsigned int irq_queues, this_p_queues;
 
include/linux/interrupt.h +4 -2
index 1ed1014c9684..a20150627a32 100644
@@ -259,6 +259,8 @@ struct irq_affinity_notify {
 	void (*release)(struct kref *ref);
 };
 
+#define	IRQ_MAX_SETS  4
+
 /**
  * struct irq_affinity - Description for automatic irq affinity assignements
  * @pre_vectors:	Don't apply affinity to @pre_vectors at beginning of
@@ -266,13 +268,13 @@ struct irq_affinity_notify {
  * @post_vectors:	Don't apply affinity to @post_vectors at end of
  *			the MSI(-X) vector space
  * @nr_sets:		Length of passed in *sets array
- * @sets:		Number of affinitized sets
+ * @set_vectors:	Number of affinitized sets
  */
 struct irq_affinity {
 	int	pre_vectors;
 	int	post_vectors;
 	int	nr_sets;
-	int	*sets;
+	int	set_vectors[IRQ_MAX_SETS];
 };
 
 /**
kernel/irq/affinity.c +11 -7
index 9200d3b26f7d..b868b9d3df7f 100644
@@ -244,7 +244,7 @@ irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 	int affvecs = nvecs - affd->pre_vectors - affd->post_vectors;
 	int curvec, usedvecs;
 	struct irq_affinity_desc *masks = NULL;
-	int i, nr_sets;
+	int i;
 
 	/*
 	 * If there aren't any vectors left after applying the pre/post
@@ -253,6 +253,9 @@ irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 	if (nvecs == affd->pre_vectors + affd->post_vectors)
 		return NULL;
 
+	if (affd->nr_sets > IRQ_MAX_SETS)
+		return NULL;
+
 	masks = kcalloc(nvecs, sizeof(*masks), GFP_KERNEL);
 	if (!masks)
 		return NULL;
@@ -264,12 +267,13 @@ irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 	 * Spread on present CPUs starting from affd->pre_vectors. If we
 	 * have multiple sets, build each sets affinity mask separately.
 	 */
-	nr_sets = affd->nr_sets;
-	if (!nr_sets)
-		nr_sets = 1;
+	if (!affd->nr_sets) {
+		affd->nr_sets = 1;
+		affd->set_vectors[0] = affvecs;
+	}
 
-	for (i = 0, usedvecs = 0; i < nr_sets; i++) {
-		int this_vecs = affd->sets ? affd->sets[i] : affvecs;
+	for (i = 0, usedvecs = 0; i < affd->nr_sets; i++) {
+		int this_vecs = affd->set_vectors[i];
 		int ret;
 
 		ret = irq_build_affinity_masks(affd, curvec, this_vecs,
@@ -316,7 +320,7 @@ int irq_calc_affinity_vectors(int minvec, int maxvec, const struct irq_affinity
 		int i;
 
 		for (i = 0, set_vecs = 0;  i < affd->nr_sets; i++)
-			set_vecs += affd->sets[i];
+			set_vecs += affd->set_vectors[i];
 	} else {
 		get_online_cpus();
 		set_vecs = cpumask_weight(cpu_possible_mask);
Signature · 1 line
2.9.5
#14 lore raw

Re: [PATCH V3 2/5] genirq/affinity: store irq set vectors in 'struct irq_affinity'

Quoted 4 lines · Ming Lei

Currently the array of irq set vectors is provided by driver.

irq_create_affinity_masks() can be simplied a bit by treating the non-irq-set case as single irq set.

s/simplied a bit/simplified/

Quoted 2 lines

So move this array into 'struct irq_affinity', and pre-define the max set number as 4, which should be enough for normal cases.

s/irq/IRQ/
You have a real mixture of capitalization across the changelogs.

Quoted 31 lines
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/nvme/host/pci.c   |  5 ++---
 include/linux/interrupt.h |  6 ++++--
 kernel/irq/affinity.c     | 18 +++++++++++-------
 3 files changed, 17 insertions(+), 12 deletions(-)
drivers/nvme/host/pci.c +2 -3
index 022ea1ee63f8..0086bdf80ea1 100644
@@ -2081,12 +2081,11 @@ static void nvme_calc_io_queues(struct nvme_dev *dev, unsigned int irq_queues)
 static int nvme_setup_irqs(struct nvme_dev *dev, unsigned int nr_io_queues)
 {
 	struct pci_dev *pdev = to_pci_dev(dev->dev);
-	int irq_sets[2];
 	struct irq_affinity affd = {
 		.pre_vectors = 1,
-		.nr_sets = ARRAY_SIZE(irq_sets),
-		.sets = irq_sets,
+		.nr_sets = 2,
 	};
+	int *irq_sets = affd.set_vectors;
 	int result = 0;
 	unsigned int irq_queues, this_p_queues;
 
include/linux/interrupt.h +0 -0
index 1ed1014c9684..a20150627a32 100644

include/linux/interrupt.h struct irq_affinity_notify { · line 262

Quoted 4 lines
include/linux/interrupt.h +1 -0
@@ -259,6 +259,8 @@ struct irq_affinity_notify {
 	void (*release)(struct kref *ref);
 };
 
+#define	IRQ_MAX_SETS  4

This is a pretty generic name. Maybe it should include a hint that it's related to affinity?

Quoted 73 lines
include/linux/interrupt.h +2 -2
 /**
  * struct irq_affinity - Description for automatic irq affinity assignements
  * @pre_vectors:	Don't apply affinity to @pre_vectors at beginning of
@@ -266,13 +268,13 @@ struct irq_affinity_notify {
  * @post_vectors:	Don't apply affinity to @post_vectors at end of
  *			the MSI(-X) vector space
  * @nr_sets:		Length of passed in *sets array
- * @sets:		Number of affinitized sets
+ * @set_vectors:	Number of affinitized sets
  */
 struct irq_affinity {
 	int	pre_vectors;
 	int	post_vectors;
 	int	nr_sets;
-	int	*sets;
+	int	set_vectors[IRQ_MAX_SETS];
 };
 
 /**
kernel/irq/affinity.c +11 -7
index 9200d3b26f7d..b868b9d3df7f 100644
@@ -244,7 +244,7 @@ irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 	int affvecs = nvecs - affd->pre_vectors - affd->post_vectors;
 	int curvec, usedvecs;
 	struct irq_affinity_desc *masks = NULL;
-	int i, nr_sets;
+	int i;
 
 	/*
 	 * If there aren't any vectors left after applying the pre/post
@@ -253,6 +253,9 @@ irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 	if (nvecs == affd->pre_vectors + affd->post_vectors)
 		return NULL;
 
+	if (affd->nr_sets > IRQ_MAX_SETS)
+		return NULL;
+
 	masks = kcalloc(nvecs, sizeof(*masks), GFP_KERNEL);
 	if (!masks)
 		return NULL;
@@ -264,12 +267,13 @@ irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 	 * Spread on present CPUs starting from affd->pre_vectors. If we
 	 * have multiple sets, build each sets affinity mask separately.
 	 */
-	nr_sets = affd->nr_sets;
-	if (!nr_sets)
-		nr_sets = 1;
+	if (!affd->nr_sets) {
+		affd->nr_sets = 1;
+		affd->set_vectors[0] = affvecs;
+	}
 
-	for (i = 0, usedvecs = 0; i < nr_sets; i++) {
-		int this_vecs = affd->sets ? affd->sets[i] : affvecs;
+	for (i = 0, usedvecs = 0; i < affd->nr_sets; i++) {
+		int this_vecs = affd->set_vectors[i];
 		int ret;
 
 		ret = irq_build_affinity_masks(affd, curvec, this_vecs,
@@ -316,7 +320,7 @@ int irq_calc_affinity_vectors(int minvec, int maxvec, const struct irq_affinity
 		int i;
 
 		for (i = 0, set_vecs = 0;  i < affd->nr_sets; i++)
-			set_vecs += affd->sets[i];
+			set_vecs += affd->set_vectors[i];
 	} else {
 		get_online_cpus();
 		set_vecs = cpumask_weight(cpu_possible_mask);
Signature · 1 line
2.9.5
author
#15 lore raw

PATCH v3 3/5 genirq/affinity: add new callback for caculating set vectors

In reply to Ming Lei

Currently pre-caculated set vectors are provided by driver for allocating & spread vectors. This way only works when drivers passes same 'max_vecs' and 'min_vecs' to pci_alloc_irq_vectors_affinity(), also requires driver to retry the allocating & spread.

As Bjorn and Keith mentioned, the current usage & interface for irq sets is a bit awkward because the retrying should have been avoided by providing one resonable 'min_vecs'. However, if 'min_vecs' isn't same with 'max_vecs', number of the allocated vectors is unknown before calling pci_alloc_irq_vectors_affinity(), then each set's vectors can't be pre-caculated.

Add a new callback of .calc_sets into 'struct irq_affinity' so that driver can caculate set vectors after IRQ vector is allocated and before spread IRQ vectors. Add 'priv' so that driver may retrieve its private data via the 'struct irq_affinity'.

Suggested-by Thomas Gleixner <tglx at linutronix.de>
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 include/linux/interrupt.h | 4 ++++
 kernel/irq/affinity.c     | 8 ++++++--
 2 files changed, 10 insertions(+), 2 deletions(-)
include/linux/interrupt.h +4 -0
index a20150627a32..7a27f6ba1f2f 100644
@@ -269,12 +269,16 @@ struct irq_affinity_notify {
  *			the MSI(-X) vector space
  * @nr_sets:		Length of passed in *sets array
  * @set_vectors:	Number of affinitized sets
+ * @calc_sets:		Callback for caculating set vectors
+ * @priv:		Private data of @calc_sets
  */
 struct irq_affinity {
 	int	pre_vectors;
 	int	post_vectors;
 	int	nr_sets;
 	int	set_vectors[IRQ_MAX_SETS];
+	void	(*calc_sets)(struct irq_affinity *, int nvecs);
+	void	*priv;
 };
 
 /**
kernel/irq/affinity.c +6 -2
index b868b9d3df7f..2341b1f005fa 100644
@@ -267,7 +267,9 @@ irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 	 * Spread on present CPUs starting from affd->pre_vectors. If we
 	 * have multiple sets, build each sets affinity mask separately.
 	 */
-	if (!affd->nr_sets) {
+	if (affd->calc_sets) {
+		affd->calc_sets(affd, nvecs);
+	} else if (!affd->nr_sets) {
 		affd->nr_sets = 1;
 		affd->set_vectors[0] = affvecs;
 	}
@@ -316,7 +318,9 @@ int irq_calc_affinity_vectors(int minvec, int maxvec, const struct irq_affinity
 	if (resv > minvec)
 		return 0;
 
-	if (affd->nr_sets) {
+	if (affd->calc_sets) {
+		set_vecs = vecs;
+	} else if (affd->nr_sets) {
 		int i;
 
 		for (i = 0, set_vecs = 0;  i < affd->nr_sets; i++)
Signature · 1 line
2.9.5
#16 lore raw

Re: [PATCH V3 3/5] genirq/affinity: add new callback for caculating set vectors

Quoted 4 lines · Ming Lei

Currently pre-caculated set vectors are provided by driver for allocating & spread vectors. This way only works when drivers passes same 'max_vecs' and 'min_vecs' to pci_alloc_irq_vectors_affinity(), also requires driver to retry the allocating & spread.

s/pre-caculated/precalculated/
s/drivers/a driver/
s/also requires/which also requires the/

Quoted 6 lines

As Bjorn and Keith mentioned, the current usage & interface for irq sets is a bit awkward because the retrying should have been avoided by providing one resonable 'min_vecs'. However, if 'min_vecs' isn't same with 'max_vecs', number of the allocated vectors is unknown before calling pci_alloc_irq_vectors_affinity(), then each set's vectors can't be pre-caculated.

s/resonable/reasonable/
s/isn't same with/isn't the same as/
s/number of/the number of/
s/pre-caculated/precalculated/
s/then each/and the/
s/irq/IRQ/

Quoted 4 lines

Add a new callback of .calc_sets into 'struct irq_affinity' so that driver can caculate set vectors after IRQ vector is allocated and before spread IRQ vectors. Add 'priv' so that driver may retrieve its private data via the 'struct irq_affinity'.

s/caculate/calculate/

I *think* "set vectors" here has something to do with an affinity cpumask? "Set vectors" doesn't seem like quite the right terminology.

Quoted 12 lines
Suggested-by Thomas Gleixner <tglx at linutronix.de>
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 include/linux/interrupt.h | 4 ++++
 kernel/irq/affinity.c     | 8 ++++++--
 2 files changed, 10 insertions(+), 2 deletions(-)
include/linux/interrupt.h +0 -0
index a20150627a32..7a27f6ba1f2f 100644

include/linux/interrupt.h struct irq_affinity_notify { · line 272

Quoted 4 lines
include/linux/interrupt.h +1 -0
@@ -269,12 +269,16 @@ struct irq_affinity_notify {
  *			the MSI(-X) vector space
  * @nr_sets:		Length of passed in *sets array
  * @set_vectors:	Number of affinitized sets
+ * @calc_sets:		Callback for caculating set vectors

Possible s/set vectors/<something else>/ ?

Quoted 41 lines
include/linux/interrupt.h +3 -0
+ * @priv:		Private data of @calc_sets
  */
 struct irq_affinity {
 	int	pre_vectors;
 	int	post_vectors;
 	int	nr_sets;
 	int	set_vectors[IRQ_MAX_SETS];
+	void	(*calc_sets)(struct irq_affinity *, int nvecs);
+	void	*priv;
 };
 
 /**
kernel/irq/affinity.c +6 -2
index b868b9d3df7f..2341b1f005fa 100644
@@ -267,7 +267,9 @@ irq_create_affinity_masks(int nvecs, struct irq_affinity *affd)
 	 * Spread on present CPUs starting from affd->pre_vectors. If we
 	 * have multiple sets, build each sets affinity mask separately.
 	 */
-	if (!affd->nr_sets) {
+	if (affd->calc_sets) {
+		affd->calc_sets(affd, nvecs);
+	} else if (!affd->nr_sets) {
 		affd->nr_sets = 1;
 		affd->set_vectors[0] = affvecs;
 	}
@@ -316,7 +318,9 @@ int irq_calc_affinity_vectors(int minvec, int maxvec, const struct irq_affinity
 	if (resv > minvec)
 		return 0;
 
-	if (affd->nr_sets) {
+	if (affd->calc_sets) {
+		set_vecs = vecs;
+	} else if (affd->nr_sets) {
 		int i;
 
 		for (i = 0, set_vecs = 0;  i < affd->nr_sets; i++)
Signature · 1 line
2.9.5
#17 lore raw

Re: [PATCH V3 3/5] genirq/affinity: add new callback for caculating set vectors

Quoted 9 lines · Bjorn Helgaas
Quoted 4 lines

Add a new callback of .calc_sets into 'struct irq_affinity' so that driver can caculate set vectors after IRQ vector is allocated and before spread IRQ vectors. Add 'priv' so that driver may retrieve its private data via the 'struct irq_affinity'.

s/caculate/calculate/

I *think* "set vectors" here has something to do with an affinity cpumask? "Set vectors" doesn't seem like quite the right terminology.

See previous mail. Let's stay with 'interrupt sets' and get rid of the vector naming completely.

Thanks,

tglx

author
#18 lore raw

PATCH v3 4/5 nvme-pci: avoid irq allocation retrying via .calc_sets

In reply to Ming Lei

Currently pre-caculate each set vectors, and this way requires same 'max_vecs' and 'min_vecs' passed to pci_alloc_irq_vectors_affinity(), then nvme_setup_irqs() has to retry in case of allocation failure.

This usage & interface is a bit awkward because the retry should have been avoided by providing one reasonable 'min_vecs'.

Implement the callback of .calc_sets, so that pci_alloc_irq_vectors_affinity() can calculate each set's vector after IRQ vectors is allocated and before spread IRQ, then NVMe's retry in case of irq allocation failure can be removed.

Reviewed-by Keith Busch <keith.busch at intel.com>
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/nvme/host/pci.c | 62 +++++++++++++------------------------------------
 1 file changed, 16 insertions(+), 46 deletions(-)
drivers/nvme/host/pci.c +16 -46
index 0086bdf80ea1..8c51252a897e 100644
@@ -2078,14 +2078,25 @@ static void nvme_calc_io_queues(struct nvme_dev *dev, unsigned int irq_queues)
 	}
 }
 
+static void nvme_calc_irq_sets(struct irq_affinity *affd, int nvecs)
+{
+	struct nvme_dev *dev = affd->priv;
+
+	nvme_calc_io_queues(dev, nvecs);
+
+	affd->set_vectors[HCTX_TYPE_DEFAULT] = dev->io_queues[HCTX_TYPE_DEFAULT];
+	affd->set_vectors[HCTX_TYPE_READ] = dev->io_queues[HCTX_TYPE_READ];
+	affd->nr_sets = 2;
+}
+
 static int nvme_setup_irqs(struct nvme_dev *dev, unsigned int nr_io_queues)
 {
 	struct pci_dev *pdev = to_pci_dev(dev->dev);
 	struct irq_affinity affd = {
 		.pre_vectors = 1,
-		.nr_sets = 2,
+		.calc_sets = nvme_calc_irq_sets,
+		.priv = dev,
 	};
-	int *irq_sets = affd.set_vectors;
 	int result = 0;
 	unsigned int irq_queues, this_p_queues;
 
@@ -2102,50 +2113,8 @@ static int nvme_setup_irqs(struct nvme_dev *dev, unsigned int nr_io_queues)
 	}
 	dev->io_queues[HCTX_TYPE_POLL] = this_p_queues;
 
-	/*
-	 * For irq sets, we have to ask for minvec == maxvec. This passes
-	 * any reduction back to us, so we can adjust our queue counts and
-	 * IRQ vector needs.
-	 */
-	do {
-		nvme_calc_io_queues(dev, irq_queues);
-		irq_sets[0] = dev->io_queues[HCTX_TYPE_DEFAULT];
-		irq_sets[1] = dev->io_queues[HCTX_TYPE_READ];
-		if (!irq_sets[1])
-			affd.nr_sets = 1;
-
-		/*
-		 * If we got a failure and we're down to asking for just
-		 * 1 + 1 queues, just ask for a single vector. We'll share
-		 * that between the single IO queue and the admin queue.
-		 * Otherwise, we assign one independent vector to admin queue.
-		 */
-		if (irq_queues > 1)
-			irq_queues = irq_sets[0] + irq_sets[1] + 1;
-
-		result = pci_alloc_irq_vectors_affinity(pdev, irq_queues,
-				irq_queues,
-				PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
-
-		/*
-		 * Need to reduce our vec counts. If we get ENOSPC, the
-		 * platform should support mulitple vecs, we just need
-		 * to decrease our ask. If we get EINVAL, the platform
-		 * likely does not. Back down to ask for just one vector.
-		 */
-		if (result == -ENOSPC) {
-			irq_queues--;
-			if (!irq_queues)
-				return result;
-			continue;
-		} else if (result == -EINVAL) {
-			irq_queues = 1;
-			continue;
-		} else if (result <= 0)
-			return -EIO;
-		break;
-	} while (1);
-
+	result = pci_alloc_irq_vectors_affinity(pdev, 1, irq_queues,
+			PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
 	return result;
 }
 
@@ -3021,6 +2990,7 @@ static struct pci_driver nvme_driver = {
 
 static int __init nvme_init(void)
 {
+	BUILD_BUG_ON(2 > IRQ_MAX_SETS);
 	return pci_register_driver(&nvme_driver);
 }
 
Signature · 1 line
2.9.5
#19 lore raw

Re: [PATCH V3 4/5] nvme-pci: avoid irq allocation retrying via .calc_sets

Quoted 3 lines · Ming Lei

Currently pre-caculate each set vectors, and this way requires same 'max_vecs' and 'min_vecs' passed to pci_alloc_irq_vectors_affinity(), then nvme_setup_irqs() has to retry in case of allocation failure.

s/pre-caculate/precalculate/
My usual "set vectors" question as on other patches.

Quoted 7 lines

This usage & interface is a bit awkward because the retry should have been avoided by providing one reasonable 'min_vecs'.

Implement the callback of .calc_sets, so that pci_alloc_irq_vectors_affinity() can calculate each set's vector after IRQ vectors is allocated and before spread IRQ, then NVMe's retry in case of irq allocation failure can be removed.

s/irq/IRQ/

Quoted 92 lines
Reviewed-by Keith Busch <keith.busch at intel.com>
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/nvme/host/pci.c | 62 +++++++++++++------------------------------------
 1 file changed, 16 insertions(+), 46 deletions(-)
drivers/nvme/host/pci.c +15 -46
index 0086bdf80ea1..8c51252a897e 100644
@@ -2078,14 +2078,25 @@ static void nvme_calc_io_queues(struct nvme_dev *dev, unsigned int irq_queues)
 	}
 }
 
+static void nvme_calc_irq_sets(struct irq_affinity *affd, int nvecs)
+{
+	struct nvme_dev *dev = affd->priv;
+
+	nvme_calc_io_queues(dev, nvecs);
+
+	affd->set_vectors[HCTX_TYPE_DEFAULT] = dev->io_queues[HCTX_TYPE_DEFAULT];
+	affd->set_vectors[HCTX_TYPE_READ] = dev->io_queues[HCTX_TYPE_READ];
+	affd->nr_sets = 2;
+}
+
 static int nvme_setup_irqs(struct nvme_dev *dev, unsigned int nr_io_queues)
 {
 	struct pci_dev *pdev = to_pci_dev(dev->dev);
 	struct irq_affinity affd = {
 		.pre_vectors = 1,
-		.nr_sets = 2,
+		.calc_sets = nvme_calc_irq_sets,
+		.priv = dev,
 	};
-	int *irq_sets = affd.set_vectors;
 	int result = 0;
 	unsigned int irq_queues, this_p_queues;
 
@@ -2102,50 +2113,8 @@ static int nvme_setup_irqs(struct nvme_dev *dev, unsigned int nr_io_queues)
 	}
 	dev->io_queues[HCTX_TYPE_POLL] = this_p_queues;
 
-	/*
-	 * For irq sets, we have to ask for minvec == maxvec. This passes
-	 * any reduction back to us, so we can adjust our queue counts and
-	 * IRQ vector needs.
-	 */
-	do {
-		nvme_calc_io_queues(dev, irq_queues);
-		irq_sets[0] = dev->io_queues[HCTX_TYPE_DEFAULT];
-		irq_sets[1] = dev->io_queues[HCTX_TYPE_READ];
-		if (!irq_sets[1])
-			affd.nr_sets = 1;
-
-		/*
-		 * If we got a failure and we're down to asking for just
-		 * 1 + 1 queues, just ask for a single vector. We'll share
-		 * that between the single IO queue and the admin queue.
-		 * Otherwise, we assign one independent vector to admin queue.
-		 */
-		if (irq_queues > 1)
-			irq_queues = irq_sets[0] + irq_sets[1] + 1;
-
-		result = pci_alloc_irq_vectors_affinity(pdev, irq_queues,
-				irq_queues,
-				PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
-
-		/*
-		 * Need to reduce our vec counts. If we get ENOSPC, the
-		 * platform should support mulitple vecs, we just need
-		 * to decrease our ask. If we get EINVAL, the platform
-		 * likely does not. Back down to ask for just one vector.
-		 */
-		if (result == -ENOSPC) {
-			irq_queues--;
-			if (!irq_queues)
-				return result;
-			continue;
-		} else if (result == -EINVAL) {
-			irq_queues = 1;
-			continue;
-		} else if (result <= 0)
-			return -EIO;
-		break;
-	} while (1);
-
+	result = pci_alloc_irq_vectors_affinity(pdev, 1, irq_queues,
+			PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd);
 	return result;
 }
 

drivers/nvme/host/pci.c static struct pci_driver nvme_driver = { · line 2993

Quoted 4 lines
drivers/nvme/host/pci.c +1 -0
@@ -3021,6 +2990,7 @@ static struct pci_driver nvme_driver = {
 
 static int __init nvme_init(void)
 {
+	BUILD_BUG_ON(2 > IRQ_MAX_SETS);

"IRQ_MAX_SETS < 2" would read more naturally; is there a reason to have it reversed?

Quoted 6 lines
drivers/nvme/host/pci.c +0 -0
 	return pci_register_driver(&nvme_driver);
 }
 
Signature · 1 line
2.9.5
#20 lore raw

Re: [PATCH V3 4/5] nvme-pci: avoid irq allocation retrying via .calc_sets

Quoted 11 lines · Bjorn Helgaas
Quoted 3 lines · Ming Lei

Currently pre-caculate each set vectors, and this way requires same 'max_vecs' and 'min_vecs' passed to pci_alloc_irq_vectors_affinity(), then nvme_setup_irqs() has to retry in case of allocation failure.

s/pre-caculate/precalculate/
My usual "set vectors" question as on other patches.

Quoted 3 lines

This usage & interface is a bit awkward because the retry should have been avoided by providing one reasonable 'min_vecs'.

Quoted 6 lines
Quoted 4 lines

Implement the callback of .calc_sets, so that pci_alloc_irq_vectors_affinity() can calculate each set's vector after IRQ vectors is allocated and before spread IRQ, then NVMe's retry in case of irq allocation failure can be removed.

s/irq/IRQ/

Let me rephrase that thing as well

Subject: nvme-pci: Simplify interrupt allocation

The NVME PCI driver contains a tedious mechanism for interrupt allocation, which is necessary to adjust the number and size of interrupt sets to the maximum available number of interrupts which depends on the underlying PCI capabilities and the available CPU resources.

It works around the former short comings of the PCI and core interrupt allocation mechanims in combination with interrupt sets.

The PCI interrupt allocation function allows to provide a maximum and a minimum number of interrupts to be allocated and tries to allocate as many as possible. This worked without driver interaction as long as there was only a single set of interrupts to handle.

With the addition of support for multiple interrupt sets in the generic affinity spreading logic, which is invoked from the PCI interrupt allocation, the adaptive loop in the PCI interrupt allocation did not work for multiple interrupt sets. The reason is that depending on the total number of interrupts which the PCI allocation adaptive loop tries to allocate in each step, the number and the size of the interrupt sets need to be adapted as well. Due to the way the interrupt sets support was implemented there was no way for the PCI interrupt allocation code or the core affinity spreading mechanism to invoke a driver specific function for adapting the interrupt sets configuration.

As a consequence the driver had to implement another adaptive loop around the PCI interrupt allocation function and calling that with maximum and minimum interrupts set to the same value. This ensured that the allocation either succeeded or immediately failed without any attempt to adjust the number of interrupts in the PCI code.

The core code now allows drivers to provide a callback to recalculate the number and the size of interrupt sets during PCI interrupt allocation, which in turn allows the PCI interrupt allocation function to be called in the same way as with a single set of interrupts. The PCI code handles the adaptive loop and the interrupt affinity spreading mechanism invokes the driver callback to adapt the interrupt set configuration to the current loop value. This replaces the adaptive loop in the driver completely.

Implement the NVME specific callback which adjusts the interrupt sets configuration and remove the adaptive allocation loop.

Thanks,

tglx

author
#21 lore raw

PATCH v3 5/5 genirq/affinity: Document .calc_sets as required in case of multiple sets

In reply to Ming Lei

Now NVMe has implemented the .calc_sets callback for caculating each set's vectors.

For other cases of multiple IRQ sets, pre-caculating each set's vectors before allocating IRQ vectors can't work because the whole vectors number is unknow at that time.

So document .calc_sets as required explicitly for multiple sets.

Acked-by Bjorn Helgaas <bhelgaas at google.com>
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/pci/msi.c         | 16 ++++++++++------
 include/linux/interrupt.h |  3 ++-
 2 files changed, 12 insertions(+), 7 deletions(-)
drivers/pci/msi.c +10 -6
index 96978459e2a0..199d708b4099 100644
@@ -1036,10 +1036,12 @@ static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
 		return -ERANGE;
 
 	/*
-	 * If the caller is passing in sets, we can't support a range of
-	 * vectors. The caller needs to handle that.
+	 * If the caller requests multiple sets of IRQs where each set
+	 * requires different affinity, it must also supply a ->calc_sets()
+	 * callback to compute vectors for each set after whole vectors are
+	 * allocated.
 	 */
-	if (affd && affd->nr_sets && minvec != maxvec)
+	if (affd && affd->nr_sets > 1 && !affd->calc_sets)
 		return -EINVAL;
 
 	if (WARN_ON_ONCE(dev->msi_enabled))
@@ -1094,10 +1096,12 @@ static int __pci_enable_msix_range(struct pci_dev *dev,
 		return -ERANGE;
 
 	/*
-	 * If the caller is passing in sets, we can't support a range of
-	 * supported vectors. The caller needs to handle that.
+	 * If the caller requests multiple sets of IRQs where each set
+	 * requires different affinity, it must also supply a ->calc_sets()
+	 * callback to compute vectors for each set after whole vectors are
+	 * allocated.
 	 */
-	if (affd && affd->nr_sets && minvec != maxvec)
+	if (affd && affd->nr_sets > 1 && !affd->calc_sets)
 		return -EINVAL;
 
 	if (WARN_ON_ONCE(dev->msix_enabled))
include/linux/interrupt.h +2 -1
index 7a27f6ba1f2f..a053f7fb0ff1 100644
@@ -269,7 +269,8 @@ struct irq_affinity_notify {
  *			the MSI(-X) vector space
  * @nr_sets:		Length of passed in *sets array
  * @set_vectors:	Number of affinitized sets
- * @calc_sets:		Callback for caculating set vectors
+ * @calc_sets:		Callback for caculating set vectors, required for
+ * 			multiple irq sets.
  * @priv:		Private data of @calc_sets
  */
 struct irq_affinity {
Signature · 1 line
2.9.5
#22 lore raw

Re: [PATCH V3 5/5] genirq/affinity: Document .calc_sets as required in case of multiple sets

Quoted 2 lines · Ming Lei

Now NVMe has implemented the .calc_sets callback for caculating each set's vectors.

s/caculating/calculating/

Quoted 3 lines

For other cases of multiple IRQ sets, pre-caculating each set's vectors before allocating IRQ vectors can't work because the whole vectors number is unknow at that time.

s/unknow/unknown/
Maybe spell check could be helpful?

Quoted 14 lines

So document .calc_sets as required explicitly for multiple sets.

Acked-by Bjorn Helgaas <bhelgaas at google.com>
Signed-off-by Ming Lei <ming.lei at redhat.com>
---
 drivers/pci/msi.c         | 16 ++++++++++------
 include/linux/interrupt.h |  3 ++-
 2 files changed, 12 insertions(+), 7 deletions(-)
drivers/pci/msi.c +0 -0
index 96978459e2a0..199d708b4099 100644

drivers/pci/msi.c static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec, · line 1042

Quoted 9 lines
drivers/pci/msi.c +4 -2
@@ -1036,10 +1036,12 @@ static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
 		return -ERANGE;
 
 	/*
-	 * If the caller is passing in sets, we can't support a range of
-	 * vectors. The caller needs to handle that.
+	 * If the caller requests multiple sets of IRQs where each set
+	 * requires different affinity, it must also supply a ->calc_sets()
+	 * callback to compute vectors for each set after whole vectors are
+	 * allocated.

"compute vectors for each set after whole vectors are allocated" doesn't read smoothly. The caller is requesting "multiple sets of IRQs". So each set contains several IRQ vectors. IIUC, the vectors calc_sets() are computing are something other than IRQ vectors.

Quoted 39 lines
drivers/pci/msi.c +6 -4
 	 */
-	if (affd && affd->nr_sets && minvec != maxvec)
+	if (affd && affd->nr_sets > 1 && !affd->calc_sets)
 		return -EINVAL;
 
 	if (WARN_ON_ONCE(dev->msi_enabled))
@@ -1094,10 +1096,12 @@ static int __pci_enable_msix_range(struct pci_dev *dev,
 		return -ERANGE;
 
 	/*
-	 * If the caller is passing in sets, we can't support a range of
-	 * supported vectors. The caller needs to handle that.
+	 * If the caller requests multiple sets of IRQs where each set
+	 * requires different affinity, it must also supply a ->calc_sets()
+	 * callback to compute vectors for each set after whole vectors are
+	 * allocated.
 	 */
-	if (affd && affd->nr_sets && minvec != maxvec)
+	if (affd && affd->nr_sets > 1 && !affd->calc_sets)
 		return -EINVAL;
 
 	if (WARN_ON_ONCE(dev->msix_enabled))
include/linux/interrupt.h +2 -1
index 7a27f6ba1f2f..a053f7fb0ff1 100644
@@ -269,7 +269,8 @@ struct irq_affinity_notify {
  *			the MSI(-X) vector space
  * @nr_sets:		Length of passed in *sets array
  * @set_vectors:	Number of affinitized sets
- * @calc_sets:		Callback for caculating set vectors
+ * @calc_sets:		Callback for caculating set vectors, required for
+ * 			multiple irq sets.
  * @priv:		Private data of @calc_sets
  */
 struct irq_affinity {
Signature · 1 line
2.9.5
maintainer Reviewed-by on the v3 series
#23 lore raw
In reply to Ming Lei
Quoted 14 lines · Ming Lei

Hi,

Currently pre-caculated set vectors are provided by driver for allocating & spread vectors. This way only works when drivers passes same 'max_vecs' and 'min_vecs' to pci_alloc_irq_vectors_affinity(), also requires driver to retry the allocating & spread.

As Bjorn and Keith mentioned, the current usage & interface for irq sets is a bit awkward because the retrying should have been avoided by providing one resonable 'min_vecs'. However, if 'min_vecs' isn't same with 'max_vecs', number of the allocated vectors is unknown before calling pci_alloc_irq_vectors_affinity(), then each set's vectors can't be pre-caculated.

Quoted 3 lines

Add a new callback of .calc_sets into 'struct irq_affinity' so that driver can caculate set vectors after IRQ vector is allocated and before spread IRQ vectors.

Much improved, thanks for doing this work. You can add

Reviewed-by Jens Axboe <axboe at kernel.dk>

if you want.

Signature · 1 line
Jens Axboe