On Fri, 27 Jun 2014 11:43:20 +0200 Stefan Hajnoczi <stefa...@redhat.com> wrote:
> On Tue, Jun 03, 2014 at 06:39:05PM +0200, Paolo Bonzini wrote: > > Even though virtio-blk-dataplane mostly synchronizes with the block layer > > by means of the AioContext, we still need to introduce mutexes for other > > QEMU subsystems that the dataplane thread might encounter on its way. > > Adding rerror/werror support, for example, means that the dataplane > > thread will have to generate QMP events. > > > > monitor_puts is the entry point for generating QMP responses and events. > > Making it thread-safe lets virtio-blk-dataplane threads generate QMP > > events; because the same entry point is also used for responses, a > > response and an event will never be intertwined. > > > > Protection is inserted at both the qemu-char and monitor levels. > > A generic mutex is necessary in qemu_fe_chr_write so that > > qemu_chr_fe_write_all does not break its output; we reuse that > > mutex in some of the character devices. > > > > There is no need to protect against removal of the monitor's backend, > > since the monitor itself cannot be removed. > > > > Paolo Bonzini (6): > > qemu-char: introduce qemu_chr_alloc > > qemu-char: do not call chr_write directly > > qemu-char: move pty_chr_update_read_handler around > > qemu-char: make writes thread-safe > > monitor: protect outbuf with mutex > > monitor: protect event emission > > > > backends/baum.c | 2 +- > > backends/msmouse.c | 2 +- > > include/sysemu/char.h | 20 ++++++-- > > monitor.c | 55 ++++++++++++++++++---- > > qemu-char.c | 125 > > +++++++++++++++++++++++++++++++++----------------- > > spice-qemu-char.c | 2 +- > > ui/console.c | 2 +- > > 7 files changed, 149 insertions(+), 59 deletions(-) > > Modulo Fam's missing unlock comment: > > Reviewed-by: Stefan Hajnoczi <stefa...@redhat.com> I appreciate your review, but this one is already merged on master.
