On Sun, 2017-01-29 at 10:07 +0100, h...@lst.de wrote:
> Hi Bart,
>
> thanks for the analysis. I think the main issue here is that we shouldn't
> try to assign affinity masks if there are not affinity vectors left to
> assign.
>
> Does the patch below fix the issue for you?
>
> diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
> index 50c5003..96c0713 100644
> --- a/drivers/pci/msi.c
> +++ b/drivers/pci/msi.c
> @@ -1206,6 +1206,11 @@ int pci_alloc_irq_vectors_affinity(struct pci_dev
> *dev, unsigned int min_vecs,
> if (flags & PCI_IRQ_AFFINITY) {
> if (!affd)
> affd = &msi_default_affd;
> +
> + if (affd->pre_vectors + affd->post_vectors > min_vecs)
> + return -EINVAL;
> + if (affd->pre_vectors + affd->post_vectors == min_vecs)
> + flags &=~ PCI_IRQ_AFFINITY;
> } else {
> if (WARN_ON(affd))
> affd = NULL;
Hello Christoph,
Thanks for the quick reply. Sorry but the above patch didn't help - the
PCI_IRQ_AFFINITY isn't tested after the above code clears it. But after I
changed "flags &=~ PCI_IRQ_AFFINITY" into "affd = NULL" the memory
corruption complaint was gone and I/O through the qla2xxx adapter
was still working fine. If you want you can add the following to the above
patch with the aforementioned change:
Tested-by: Bart Van Assche <bart.vanass...@sandisk.com>
Bart.
