linux-kernel - Re:Re: [PATCH] make blk_mq_map_queues more friendly for cpu topology

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 26 Mar 2019 15:55:10 +0800 (CST)
From:   luferry <luferry@....com>
To:     "Christoph Hellwig" <hch@....de>
Cc:     "Jens Axboe" <axboe@...nel.dk>,
        "Dongli Zhang" <dongli.zhang@...cle.com>,
        "Ming Lei" <ming.lei@...hat.com>, linux-block@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re:Re: [PATCH] make blk_mq_map_queues more friendly for cpu
 topology




At 2019-03-26 15:39:54, "Christoph Hellwig" <hch@....de> wrote:
>Why isn't this using the automatic PCI-level affinity assignment to
>start with?

When enable virtio-blk with multi queues but with only 2 msix-vector.
vp_dev->per_vq_vectors will be false, vp_get_vq_affintity will return NULL directly
so blk_mq_virtio_map_queues will fallback to blk_mq_map_queues.


448 const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
449 {
450     struct virtio_pci_device *vp_dev = to_vp_device(vdev);
451
452     if (!vp_dev->per_vq_vectors ||
453         vp_dev->vqs[index]->msix_vector == VIRTIO_MSI_NO_VECTOR)
454         return NULL;
455
456     return pci_irq_get_affinity(vp_dev->pci_dev,
457                     vp_dev->vqs[index]->msix_vector);
458 }

 32 int blk_mq_virtio_map_queues(struct blk_mq_queue_map *qmap,
 33         struct virtio_device *vdev, int first_vec)
 34 {
 35     const struct cpumask *mask;
 36     unsigned int queue, cpu;
 37
 38     if (!vdev->config->get_vq_affinity)
 39         goto fallback;
 40
 41     for (queue = 0; queue < qmap->nr_queues; queue++) {
 42         mask = vdev->config->get_vq_affinity(vdev, first_vec + queue); //vp_get_vq_affinity return NULL
 43         if (!mask)
 44             goto fallback;
 45
 46         for_each_cpu(cpu, mask)
 47             qmap->mq_map[cpu] = qmap->queue_offset + queue;
 48     }
 49
 50     return 0;
 51 fallback:
 52     return blk_mq_map_queues(qmap);
 53 }

here is previous discussion
https://patchwork.kernel.org/patch/10865461/