On 2014/11/2 2:59, Thomas Gleixner wrote: > On Mon, 27 Oct 2014, Jiang Liu wrote: >> We are going to support ACPI based IOAPIC hotplug, so introduce a rwsem >> to protect IOAPIC data structures from IOAPIC hotplug. We choose to >> serialize in ACPI instead of in the IOAPIC core because: >> 1) currently we are only plan to support ACPI based IOAPIC hotplug >> 2) it's much more cleaner and easier >> 3) It does't affect IOAPIC discovered by devicetree, SFI and mpparse. > > I had a last intensive look at this series as I was about to merge > it. So I looked at the locking rules here again > >> +/* >> + * Locks related to IOAPIC hotplug >> + * Hotplug side: >> + * ->lock_device_hotplug() //device_hotplug_lock >> + * ->acpi_ioapic_rwsem >> + * ->ioapic_lock >> + * Interrupt mapping side: >> + * ->acpi_ioapic_rwsem >> + * ->ioapic_mutex >> + * ->ioapic_lock >> + */ > > This looks sane, but I cannot figure out at all why this needs to be a > rwsem. > >> +static DECLARE_RWSEM(acpi_ioapic_rwsem); > > I think it should be a simple mutex because the rwsem does not protect > against concurrent execution what taken for read. > > And the site which takes it for write is in the early boot process > where nothing runs in parallel AFAICT. Hi Thomas, You are right. It's not on hot path, so a mutex is better than a rwsem here. I will send out an updated version soon. Regards! Gerry > > Thanks, > > tglx > -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
