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


Groups > linux.kernel > #1571262 > unrolled thread

[PATCH 2/2] nvme: allocate nvme_queue in correct node

Started byShaohua Li <shli@fb.com>
First post2017-02-01 05:30 +0100
Last post2017-02-01 10:40 +0100
Articles 2 — 2 participants

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.


Contents

  [PATCH 2/2] nvme: allocate nvme_queue in correct node Shaohua Li <shli@fb.com> - 2017-02-01 05:30 +0100
    Re: [PATCH 2/2] nvme: allocate nvme_queue in correct node Christoph Hellwig <hch@lst.de> - 2017-02-01 10:40 +0100

#1571262 — [PATCH 2/2] nvme: allocate nvme_queue in correct node

FromShaohua Li <shli@fb.com>
Date2017-02-01 05:30 +0100
Subject[PATCH 2/2] nvme: allocate nvme_queue in correct node
Message-ID<t5UIp-1aG-3@gated-at.bofh.it>
nvme_queue is per-cpu queue (mostly). Allocating it in node where blk-mq
will use it.

Signed-off-by: Shaohua Li <shli@fb.com>
---
 drivers/nvme/host/pci.c | 19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 3faefab..f81c0ed 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1041,9 +1041,10 @@ static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq,
 }
 
 static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
-							int depth)
+							int depth, int node)
 {
-	struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL);
+	struct nvme_queue *nvmeq = kzalloc_node(sizeof(*nvmeq), GFP_KERNEL,
+							node);
 	if (!nvmeq)
 		return NULL;
 
@@ -1219,7 +1220,8 @@ static int nvme_configure_admin_queue(struct nvme_dev *dev)
 
 	nvmeq = dev->queues[0];
 	if (!nvmeq) {
-		nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH);
+		nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH,
+						dev_to_node(dev->dev));
 		if (!nvmeq)
 			return -ENOMEM;
 	}
@@ -1309,9 +1311,18 @@ static int nvme_create_io_queues(struct nvme_dev *dev)
 {
 	unsigned i, max;
 	int ret = 0;
+	const struct cpumask *mask;
 
 	for (i = dev->queue_count; i <= dev->max_qid; i++) {
-		if (!nvme_alloc_queue(dev, i, dev->q_depth)) {
+		int node = dev_to_node(dev->dev);
+
+		mask = pci_irq_get_affinity(to_pci_dev(dev->dev), i);
+		if (mask) {
+			node = cpu_to_node(cpumask_first(mask));
+			node = local_memory_node(node);
+		}
+
+		if (!nvme_alloc_queue(dev, i, dev->q_depth, node)) {
 			ret = -ENOMEM;
 			break;
 		}
-- 
2.9.3

[toc] | [next] | [standalone]


#1571366

FromChristoph Hellwig <hch@lst.de>
Date2017-02-01 10:40 +0100
Message-ID<t5Zyp-4a5-7@gated-at.bofh.it>
In reply to#1571262
> +		mask = pci_irq_get_affinity(to_pci_dev(dev->dev), i);
> +		if (mask) {
> +			node = cpu_to_node(cpumask_first(mask));
> +			node = local_memory_node(node);
> +		}

Can you move this to a PCI-layer helper, e.g. something like:

int pci_irq_get_node(struct pci_dev *dev, unsigned vec)
{
	const struct cpumask *mask = pci_irq_get_affinity(dev), i);
	if (mask)
		return local_memory_node(cpu_to_node(cpumask_first(mask)));
	return dev_to_node(&dev->dev);
}

Otherwise this looks fine.

[toc] | [prev] | [standalone]


Back to top | Article view | linux.kernel


csiph-web