On 05/16/2013 08:49 AM, Michal Privoznik wrote:
With multiqueue network feature, we are advised to pass multiple vhost-net FDs as well. The ratio should be 1:1. Therefore we must alter the qemuOpenVhostNet function to allow that. --- src/qemu/qemu_command.c | 60 ++++++++++++++++++++++++++++++++++++------------- src/qemu/qemu_command.h | 3 ++- src/qemu/qemu_hotplug.c | 12 ++++++---- 3 files changed, 54 insertions(+), 21 deletions(-)
diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c index a053d49..ec66a33 100644 --- a/src/qemu/qemu_command.c +++ b/src/qemu/qemu_command.c @@ -405,17 +405,34 @@ cleanup: }
+/** + * qemuOpenVhostNet: + * @def: domain definition + * @net: network definition + * @qemuCaps: qemu binary capabilities + * @vhostfd: array of opened vhost-net device + * @vhostfdSize: size of @vhostfd array + * + * Open vhost-net, multiple times - if requested. + * In case, no vhost-net is needed, @vhostfdSize is set to 0 + * and 0 is returned. + * + * Returns: 0 on success + * -1 on failure + */ int qemuOpenVhostNet(virDomainDefPtr def, virDomainNetDefPtr net, virQEMUCapsPtr qemuCaps, - int *vhostfd) + int *vhostfd, + int *vhostfdSize) { - *vhostfd = -1; /* assume we won't use vhost */ + int i;
/* If the config says explicitly to not use vhost, return now */ if (net->driver.virtio.name == VIR_DOMAIN_NET_BACKEND_TYPE_QEMU) { - return 0; + *vhostfdSize = 0; + return 0; }
/* If qemu doesn't support vhost-net mode (including the -netdev command @@ -430,6 +447,7 @@ qemuOpenVhostNet(virDomainDefPtr def, "this QEMU binary")); return -1; } + *vhostfdSize = 0; return 0; }
@@ -441,23 +459,32 @@ qemuOpenVhostNet(virDomainDefPtr def, "virtio network interfaces")); return -1; } + *vhostfdSize = 0; return 0; }
- *vhostfd = open("/dev/vhost-net", O_RDWR); - virDomainAuditNetDevice(def, net, "/dev/vhost-net", *vhostfd >= 0); + for (i = 0; i < *vhostfdSize; i++) { + vhostfd[i] = open("/dev/vhost-net", O_RDWR); + virDomainAuditNetDevice(def, net, "/dev/vhost-net", vhostfd[i] >= 0);
I just realized while reviewing the next patch that we probably only want a single audit message here, rather than one for each individual fd.