Re: [Qemu-devel] [PATCH v3 4/4] hw/qxl-render: drop cursor locks, replace with pipe

[Jes Sorensen <Jes.Sorensen@redhat.com>]

On 03/17/11 10:32, Alon Levy wrote:
> On Wed, Mar 16, 2011 at 05:48:41PM +0100, Jes Sorensen wrote:
>> > On 03/16/11 16:52, Alon Levy wrote:
>>> > > +void qxl_server_request_cursor_set(PCIQXLDevice *qxl, QEMUCursor *c, int x, int y)
>>> > > +{
>>> > > +    QXLServerCursorSetRequest req;
>>> > > +    int r;
>>> > > +
>>> > > +    req.req = QXL_SERVER_CURSOR_SET;
>>> > > +    req.data.c = c;
>>> > > +    req.data.x = x;
>>> > > +    req.data.y = y;
>>> > > +    r = write(qxl->ssd.pipe[1], &req, sizeof(req));
>>> > > +    assert(r == sizeof(req));
>>> > > +}
>> > 
>> > There's a number of asserts here, which I am not sure is a good thing. I
>> > don't understand how far down the code this is, and if it is really
>> > fatal if this write fails?
> A failure there means we can't write to a pipe between the server thread
> and the iothread (main thread). That is not supposed to happen - and if
> it does it means some operation by the spice server will never complete.
> 
> Same for the asserts below, writes are from spice server thread, reads
> are in iothread.

But shouldn't this make it try to reconnect? Even if the reconnect
fails, it shouldn't kill the guest IMHO.

Cheers,
Jes