Comment 46 for bug 1670634

Revision history for this message
Juerg Haefliger (juergh) wrote :

For the record, the following commit fixes the issue for later kernels (>4.11). What we're seeing with the 4.4 kernel is most likely a different issue though.

commit ba74b6f7fcc07355d087af6939712eed4a454821 (refs/bisect/new)
Author: Christoph Hellwig <email address hidden>
Date: Thu Aug 24 18:07:02 2017 +0200

    virtio_pci: fix cpu affinity support

    Commit 0b0f9dc5 ("Revert "virtio_pci: use shared interrupts for
    virtqueues"") removed the adjustment of the pre_vectors for the virtio
    MSI-X vector allocation which was added in commit fb5e31d9 ("virtio:
    allow drivers to request IRQ affinity when creating VQs"). This will
    lead to an incorrect assignment of MSI-X vectors, and potential
    deadlocks when offlining cpus.

    Signed-off-by: Christoph Hellwig <email address hidden>
    Fixes: 0b0f9dc5 ("Revert "virtio_pci: use shared interrupts for virtqueues")
    Reported-by: YASUAKI ISHIMATSU <email address hidden>
    Cc: <email address hidden>
    Signed-off-by: Michael S. Tsirkin <email address hidden>

diff --git a/drivers/virtio/virtio_pci_common.c b/drivers/virtio/virtio_pci_common.c
index 007a4f366086..1c4797e53f68 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -107,6 +107,7 @@ static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
 {
        struct virtio_pci_device *vp_dev = to_vp_device(vdev);
        const char *name = dev_name(&vp_dev->vdev.dev);
+ unsigned flags = PCI_IRQ_MSIX;
        unsigned i, v;
        int err = -ENOMEM;

@@ -126,10 +127,13 @@ static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
                                        GFP_KERNEL))
                        goto error;

+ if (desc) {
+ flags |= PCI_IRQ_AFFINITY;
+ desc->pre_vectors++; /* virtio config vector */
+ }
+
        err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
- nvectors, PCI_IRQ_MSIX |
- (desc ? PCI_IRQ_AFFINITY : 0),
- desc);
+ nvectors, flags, desc);
        if (err < 0)
                goto error;
        vp_dev->msix_enabled = 1;