Fix race in interrupt handling that could cause IO to hang up under heavy

load.
This commit is contained in:
Nathan Whitehorn 2015-02-23 20:38:00 +00:00
parent c82bcdb25e
commit 09ae596bf6

View File

@ -931,10 +931,11 @@ vscsi_check_response_queue(struct vscsi_softc *sc)
mtx_assert(&sc->io_lock, MA_OWNED);
phyp_hcall(H_VIO_SIGNAL, sc->unit, 0);
bus_dmamap_sync(sc->crq_tag, sc->crq_map, BUS_DMASYNC_POSTREAD);
while (sc->crq_queue[sc->cur_crq].valid != 0) {
/* The hypercalls at both ends of this are not optimal */
phyp_hcall(H_VIO_SIGNAL, sc->unit, 0);
bus_dmamap_sync(sc->crq_tag, sc->crq_map, BUS_DMASYNC_POSTREAD);
crq = &sc->crq_queue[sc->cur_crq];
switch (crq->valid) {
@ -983,9 +984,9 @@ vscsi_check_response_queue(struct vscsi_softc *sc)
crq->valid = 0;
sc->cur_crq = (sc->cur_crq + 1) % sc->n_crqs;
};
bus_dmamap_sync(sc->crq_tag, sc->crq_map, BUS_DMASYNC_PREWRITE);
phyp_hcall(H_VIO_SIGNAL, sc->unit, 1);
bus_dmamap_sync(sc->crq_tag, sc->crq_map, BUS_DMASYNC_PREWRITE);
phyp_hcall(H_VIO_SIGNAL, sc->unit, 1);
}
}