From 20e407e195b29a4f5a18d713a61f54a75f992bd5 Mon Sep 17 00:00:00 2001 From: Christoph Hellwig Date: Tue, 8 Nov 2016 17:15:01 -0800 Subject: genirq/affinity: Introduce struct irq_affinity Some drivers (various network and RDMA adapter for example) have a MSI-X vector layout where most of the vectors are used for I/O queues and should have CPU affinity assigned to them, but some (usually 1 but sometimes more) at the beginning or end are used for low-performance admin or configuration work and should not have any explicit affinity assigned to them. Add a new irq_affinity structure, which will be passed through a variant of pci_irq_alloc_vectors that allows to specify these requirements (and is extensible to any future quirks in that area) so that the core IRQ affinity algorithm can take this quirks into account. Signed-off-by: Christoph Hellwig Reviewed-by: Johannes Thumshirn Reviewed-by: Hannes Reinecke Acked-by: Jens Axboe Cc: linux-block@vger.kernel.org Cc: linux-pci@vger.kernel.org Link: http://lkml.kernel.org/r/1478654107-7384-2-git-send-email-hch@lst.de Signed-off-by: Thomas Gleixner --- include/linux/interrupt.h | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'include/linux/interrupt.h') diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h index 72f0721f75e7..6b5268688a81 100644 --- a/include/linux/interrupt.h +++ b/include/linux/interrupt.h @@ -232,6 +232,18 @@ struct irq_affinity_notify { void (*release)(struct kref *ref); }; +/** + * struct irq_affinity - Description for automatic irq affinity assignements + * @pre_vectors: Don't apply affinity to @pre_vectors at beginning of + * the MSI(-X) vector space + * @post_vectors: Don't apply affinity to @post_vectors at end of + * the MSI(-X) vector space + */ +struct irq_affinity { + int pre_vectors; + int post_vectors; +}; + #if defined(CONFIG_SMP) extern cpumask_var_t irq_default_affinity; -- cgit v1.2.3 From 212bd846223c718b6577d4df16fd8d05a55ad914 Mon Sep 17 00:00:00 2001 From: Christoph Hellwig Date: Tue, 8 Nov 2016 17:15:02 -0800 Subject: genirq/affinity: Handle pre/post vectors in irq_calc_affinity_vectors() Only calculate the affinity for the main I/O vectors, and skip the pre or post vectors specified by struct irq_affinity. Also remove the irq_affinity cpumask argument that has never been used. If we ever need it in the future we can pass it through struct irq_affinity. Signed-off-by: Christoph Hellwig Reviewed-by: Hannes Reinecke Acked-by: Jens Axboe Cc: linux-block@vger.kernel.org Cc: linux-pci@vger.kernel.org Link: http://lkml.kernel.org/r/1478654107-7384-3-git-send-email-hch@lst.de Signed-off-by: Thomas Gleixner --- drivers/pci/msi.c | 8 ++++---- include/linux/interrupt.h | 4 ++-- kernel/irq/affinity.c | 24 ++++++++++-------------- 3 files changed, 16 insertions(+), 20 deletions(-) (limited to 'include/linux/interrupt.h') diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c index ad70507cfb56..dad2da7cf80e 100644 --- a/drivers/pci/msi.c +++ b/drivers/pci/msi.c @@ -1061,6 +1061,7 @@ EXPORT_SYMBOL(pci_msi_enabled); static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec, unsigned int flags) { + static const struct irq_affinity default_affd; bool affinity = flags & PCI_IRQ_AFFINITY; int nvec; int rc; @@ -1091,8 +1092,7 @@ static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec, for (;;) { if (affinity) { - nvec = irq_calc_affinity_vectors(dev->irq_affinity, - nvec); + nvec = irq_calc_affinity_vectors(nvec, &default_affd); if (nvec < minvec) return -ENOSPC; } @@ -1132,6 +1132,7 @@ static int __pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries, int minvec, int maxvec, unsigned int flags) { + static const struct irq_affinity default_affd; bool affinity = flags & PCI_IRQ_AFFINITY; int rc, nvec = maxvec; @@ -1140,8 +1141,7 @@ static int __pci_enable_msix_range(struct pci_dev *dev, for (;;) { if (affinity) { - nvec = irq_calc_affinity_vectors(dev->irq_affinity, - nvec); + nvec = irq_calc_affinity_vectors(nvec, &default_affd); if (nvec < minvec) return -ENOSPC; } diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h index 6b5268688a81..9081f23bc0ff 100644 --- a/include/linux/interrupt.h +++ b/include/linux/interrupt.h @@ -291,7 +291,7 @@ extern int irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify); struct cpumask *irq_create_affinity_masks(const struct cpumask *affinity, int nvec); -int irq_calc_affinity_vectors(const struct cpumask *affinity, int maxvec); +int irq_calc_affinity_vectors(int maxvec, const struct irq_affinity *affd); #else /* CONFIG_SMP */ @@ -331,7 +331,7 @@ irq_create_affinity_masks(const struct cpumask *affinity, int nvec) } static inline int -irq_calc_affinity_vectors(const struct cpumask *affinity, int maxvec) +irq_calc_affinity_vectors(int maxvec, const struct irq_affinity *affd) { return maxvec; } diff --git a/kernel/irq/affinity.c b/kernel/irq/affinity.c index 17f51d63da56..8d9259727cb4 100644 --- a/kernel/irq/affinity.c +++ b/kernel/irq/affinity.c @@ -131,24 +131,20 @@ out: } /** - * irq_calc_affinity_vectors - Calculate to optimal number of vectors for a given affinity mask - * @affinity: The affinity mask to spread. If NULL cpu_online_mask - * is used - * @maxvec: The maximum number of vectors available + * irq_calc_affinity_vectors - Calculate the optimal number of vectors + * @maxvec: The maximum number of vectors available + * @affd: Description of the affinity requirements */ -int irq_calc_affinity_vectors(const struct cpumask *affinity, int maxvec) +int irq_calc_affinity_vectors(int maxvec, const struct irq_affinity *affd) { - int cpus, ret; + int resv = affd->pre_vectors + affd->post_vectors; + int vecs = maxvec - resv; + int cpus; /* Stabilize the cpumasks */ get_online_cpus(); - /* If the supplied affinity mask is NULL, use cpu online mask */ - if (!affinity) - affinity = cpu_online_mask; - - cpus = cpumask_weight(affinity); - ret = (cpus < maxvec) ? cpus : maxvec; - + cpus = cpumask_weight(cpu_online_mask); put_online_cpus(); - return ret; + + return min(cpus, vecs) + resv; } -- cgit v1.2.3 From 67c93c218dc5d1b45d547771f1fdb44a381e1faf Mon Sep 17 00:00:00 2001 From: Christoph Hellwig Date: Tue, 8 Nov 2016 17:15:03 -0800 Subject: genirq/affinity: Handle pre/post vectors in irq_create_affinity_masks() Only calculate the affinity for the main I/O vectors, and skip the pre or post vectors specified by struct irq_affinity. Also remove the irq_affinity cpumask argument that has never been used. If we ever need it in the future we can pass it through struct irq_affinity. Signed-off-by: Christoph Hellwig Reviewed-by: Hannes Reinecke Acked-by: Bjorn Helgaas Acked-by: Jens Axboe Cc: linux-block@vger.kernel.org Cc: linux-pci@vger.kernel.org Link: http://lkml.kernel.org/r/1478654107-7384-4-git-send-email-hch@lst.de Signed-off-by: Thomas Gleixner --- drivers/pci/msi.c | 6 ++++-- include/linux/interrupt.h | 4 ++-- kernel/irq/affinity.c | 46 +++++++++++++++++++++++++--------------------- 3 files changed, 31 insertions(+), 25 deletions(-) (limited to 'include/linux/interrupt.h') diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c index dad2da7cf80e..f4a108b59336 100644 --- a/drivers/pci/msi.c +++ b/drivers/pci/msi.c @@ -553,12 +553,13 @@ error_attrs: static struct msi_desc * msi_setup_entry(struct pci_dev *dev, int nvec, bool affinity) { + static const struct irq_affinity default_affd; struct cpumask *masks = NULL; struct msi_desc *entry; u16 control; if (affinity) { - masks = irq_create_affinity_masks(dev->irq_affinity, nvec); + masks = irq_create_affinity_masks(nvec, &default_affd); if (!masks) pr_err("Unable to allocate affinity masks, ignoring\n"); } @@ -692,12 +693,13 @@ static int msix_setup_entries(struct pci_dev *dev, void __iomem *base, struct msix_entry *entries, int nvec, bool affinity) { + static const struct irq_affinity default_affd; struct cpumask *curmsk, *masks = NULL; struct msi_desc *entry; int ret, i; if (affinity) { - masks = irq_create_affinity_masks(dev->irq_affinity, nvec); + masks = irq_create_affinity_masks(nvec, &default_affd); if (!masks) pr_err("Unable to allocate affinity masks, ignoring\n"); } diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h index 9081f23bc0ff..53144e78a369 100644 --- a/include/linux/interrupt.h +++ b/include/linux/interrupt.h @@ -290,7 +290,7 @@ extern int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m); extern int irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify); -struct cpumask *irq_create_affinity_masks(const struct cpumask *affinity, int nvec); +struct cpumask *irq_create_affinity_masks(int nvec, const struct irq_affinity *affd); int irq_calc_affinity_vectors(int maxvec, const struct irq_affinity *affd); #else /* CONFIG_SMP */ @@ -325,7 +325,7 @@ irq_set_affinity_notifier(unsigned int irq, struct irq_affinity_notify *notify) } static inline struct cpumask * -irq_create_affinity_masks(const struct cpumask *affinity, int nvec) +irq_create_affinity_masks(int nvec, const struct irq_affinity *affd) { return NULL; } diff --git a/kernel/irq/affinity.c b/kernel/irq/affinity.c index 8d9259727cb4..17360bd9619b 100644 --- a/kernel/irq/affinity.c +++ b/kernel/irq/affinity.c @@ -51,16 +51,16 @@ static int get_nodes_in_cpumask(const struct cpumask *mask, nodemask_t *nodemsk) /** * irq_create_affinity_masks - Create affinity masks for multiqueue spreading - * @affinity: The affinity mask to spread. If NULL cpu_online_mask - * is used - * @nvecs: The number of vectors + * @nvecs: The total number of vectors + * @affd: Description of the affinity requirements * * Returns the masks pointer or NULL if allocation failed. */ -struct cpumask *irq_create_affinity_masks(const struct cpumask *affinity, - int nvec) +struct cpumask * +irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd) { - int n, nodes, vecs_per_node, cpus_per_vec, extra_vecs, curvec = 0; + int n, nodes, vecs_per_node, cpus_per_vec, extra_vecs, curvec; + int affv = nvecs - affd->pre_vectors - affd->post_vectors; nodemask_t nodemsk = NODE_MASK_NONE; struct cpumask *masks; cpumask_var_t nmsk; @@ -68,46 +68,46 @@ struct cpumask *irq_create_affinity_masks(const struct cpumask *affinity, if (!zalloc_cpumask_var(&nmsk, GFP_KERNEL)) return NULL; - masks = kzalloc(nvec * sizeof(*masks), GFP_KERNEL); + masks = kcalloc(nvecs, sizeof(*masks), GFP_KERNEL); if (!masks) goto out; + /* Fill out vectors at the beginning that don't need affinity */ + for (curvec = 0; curvec < affd->pre_vectors; curvec++) + cpumask_copy(masks + curvec, cpu_possible_mask); + /* Stabilize the cpumasks */ get_online_cpus(); - /* If the supplied affinity mask is NULL, use cpu online mask */ - if (!affinity) - affinity = cpu_online_mask; - - nodes = get_nodes_in_cpumask(affinity, &nodemsk); + nodes = get_nodes_in_cpumask(cpu_online_mask, &nodemsk); /* * If the number of nodes in the mask is less than or equal the * number of vectors we just spread the vectors across the nodes. */ - if (nvec <= nodes) { + if (affv <= nodes) { for_each_node_mask(n, nodemsk) { cpumask_copy(masks + curvec, cpumask_of_node(n)); - if (++curvec == nvec) + if (++curvec == affv) break; } - goto outonl; + goto done; } /* Spread the vectors per node */ - vecs_per_node = nvec / nodes; + vecs_per_node = affv / nodes; /* Account for rounding errors */ - extra_vecs = nvec - (nodes * vecs_per_node); + extra_vecs = affv - (nodes * vecs_per_node); for_each_node_mask(n, nodemsk) { int ncpus, v, vecs_to_assign = vecs_per_node; /* Get the cpus on this node which are in the mask */ - cpumask_and(nmsk, affinity, cpumask_of_node(n)); + cpumask_and(nmsk, cpu_online_mask, cpumask_of_node(n)); /* Calculate the number of cpus per vector */ ncpus = cpumask_weight(nmsk); - for (v = 0; curvec < nvec && v < vecs_to_assign; curvec++, v++) { + for (v = 0; curvec < affv && v < vecs_to_assign; curvec++, v++) { cpus_per_vec = ncpus / vecs_to_assign; /* Account for extra vectors to compensate rounding errors */ @@ -119,12 +119,16 @@ struct cpumask *irq_create_affinity_masks(const struct cpumask *affinity, irq_spread_init_one(masks + curvec, nmsk, cpus_per_vec); } - if (curvec >= nvec) + if (curvec >= affv) break; } -outonl: +done: put_online_cpus(); + + /* Fill out vectors at the end that don't need affinity */ + for (; curvec < nvecs; curvec++) + cpumask_copy(masks + curvec, cpu_possible_mask); out: free_cpumask_var(nmsk); return masks; -- cgit v1.2.3