On Mon, Jun 29, 2009 at 12:53:48PM +0300, Avi Kivity wrote: > On 06/29/2009 12:41 PM, Michael S. Tsirkin wrote: >> On Mon, Jun 29, 2009 at 11:37:00AM +0300, Avi Kivity wrote: >> >>> On 06/28/2009 10:34 PM, Michael S. Tsirkin wrote: >>> >>>> This changes bus accesses to use high-level kvm_io_bus_read/kvm_io_bus_write >>>> functions, which utilize read/write semaphore intead of mutex. in_range now >>>> becomes unused so it is removed from device ops in favor of read/write >>>> callbacks performing range checks internally. >>>> >>>> This allows aliasing (mostly for in-kernel virtio), as well as better error >>>> handling by making it possible to pass errors up to userspace. And it's enough >>>> to look at the diffstat to see that it's a better API anyway. >>>> >>>> While we are at it, document locking rules for kvm_io_device_ops. >>>> >>>> Note: since the use of the new bus_lock is localized to a small number of >>>> places, it will be easy to switch to srcu in the future if we so desire. >>>> >>>> >>> Looks good. But please split into a locking change patch and an API >>> change patch (in whatever order makes more sense). >>> >> >> This is harder than it seems. Is this really important? >> >> The locking change itself is about 6 lines, but >> 1. if I do it after in_range removal I get deadlocks >> as after marcelo's change kvm->lock is taken internally by writers. >> > > slots_lock is an outer lock to kvm->lock. > >> 2. if I do it before in_range removal it's a lot of churn: >> one of the reasons for code reorg is so that there are less >> places to change locking. >> >> >> > > I don't think you really need to change anything. slots_lock is already > taken (except where you modify the list). Are you sure about this? I don't understand the code well enough, so this reuse of an apparently unrelated lock just makes me nervious. For example what about emulate_instruction? It is sometimes called from svm/vmx without slot lock ... > How about this: > > 1. add slots_lock for write when modifying the list > 2. change the api > 3. drop kvm->lock > > ? Looks like I will just have to bite the bullet and switch to RCU. -- MST -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
