On Fri, 15 Nov 2024 14:04:11 -0600 Wei Huang wrote:
> +static void bnxt_irq_affinity_release(struct kref __always_unused *ref)
unused? you're using it now
> +{
> + struct irq_affinity_notify *notify =
> + (struct irq_affinity_notify *)
> + container_of(ref, struct irq_affinity_notify, kref);
this is ugly, and cast is unnecessary.
> + struct bnxt_irq *irq;
> +
> + irq = container_of(notify, struct bnxt_irq, affinity_notify);
since you init irq out of line you can as well init notify here
> + if (pcie_tph_set_st_entry(irq->bp->pdev, irq->msix_nr, 0)) {
You checked this function can sleep, right? Because rtnl_lock()
will sleep.
Bjorn, do you have a strong preference to have a user of the TPH code
merged as part of 6.13? We're very close to the merge window, I'm not
sure build bots etc. will have enough time to hammer this code.
My weak preference would be to punt these driver changes to 6.14
avoid all the conflicts and risks (unless Linus gives us another week.)
--
pw-bot: nap
