On 07/26/2012 04:21 PM, liu ping fan wrote: > On Thu, Jul 26, 2012 at 9:15 PM, Avi Kivity <a...@redhat.com> wrote: >> On 07/26/2012 04:14 PM, liu ping fan wrote: >>>> >>>> From the description above, I don't see why it can't be a mutex. >>>> >>> Searching in the device tree (or MemoryRegion view) can be often in >>> parallel, especially in mmio-dispatch code path >> >> In mmio dispatch we have a pointer to the object, we don't need to >> search anything. Is device tree search a hot path? >> > I think, we need lock to protect searching --phys_page_find() from > deleter--DeviceClass:unmap, so rwlock?
Better a lock on phys_map (because it is easily replaced by rcu, later). I think phys_map is also better isolated, so it will be easier to find all the placed that need protection and to avoid deadlock. -- error compiling committee.c: too many arguments to function
