On 05/05/2016 05:24 AM, Andre Przywara wrote:
Hi Tom, thanks for looking at the patches!
You're quite welcome. It's a great way to spin-up.
The idea isn't too bad indeed. I see that we could use that in vgic_queue_irq_unlock() and vgic_mmio_write_sactive(). But as Marc mentioned in a conversation yesterday we will have a mixture of wrapped locks and open coded lock sequences. See for instance vgic_prune_ap_list(), where we have the sequence, but we can't use vgic_lock_aplist_irq() because the IRQ lock is taken inside the loop while the ap_list_lock is taken once outside of it.
Agreed. I looked at a few places where it wouldn't work. Definitely not a blanket solution but every little bit helps, especially if it helps avoid a bug such as a deadlock that may take hours to track down. Been there, done that. :-) _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
