On Fri, Oct 09, 2015 at 08:40:27AM +0200, Thibaut Collet wrote: > Hi Michael, > > this patch is not correct. If we apply QEMU enters in a dead lock if > we have several queue pairs for virttio/vhost and vhost backend > crashes during the initialisation sequence > I have sent some comments yesterday to Marc to correct theses issues. > The attached fixup solves the issues (already sent to Marc Andre on > QEMU mailing list): > - The two first corrections solve the vhost backend crashes (the > message id used for some messages are the kernel IOCTL and not the > VHOST_USER id) > - the third patchs solves the QEMU dead lock in case of multiqueue > (VHOST_USER_GET_QUEUE_NUM is a one time message. This message is sent > through the new vhost_user_get_u64 function but except for the first > queue a blocking read is sent whereas no write has been sent before > that causes the lock) > > If you (and Marc Andre) agree this correction could you update this > patch with the fixup and resend the pull request ? > > Thanks > > Best regards. > > Thibaut.
I dropped this and dependent patches from the tree for now. Thanks! -- MST