netmap: vtnet: fix race condition in rxsync

This change prevents a race that happens when rxsync dequeues
N-1 rx packets (with N being the size of the netmap rx ring).
In this situation, the loop exits without re-enabling the
rx interrupts, thus causing the VQ to stall.

MFC after:	1 week
This commit is contained in:
vmaffione 2020-06-03 17:46:21 +00:00
parent 07f67f4be2
commit ed8ca220e9

View File

@ -310,8 +310,11 @@ vtnet_netmap_rxsync(struct netmap_kring *kring, int flags)
/*
* First part: import newly received packets.
* Only accept our own buffers (matching the token). We should only get
* matching buffers. We may need to stop early to avoid hwtail to overrun
* hwcur.
* matching buffers. The hwtail should never overrun hwcur, because
* we publish only N-1 receive buffers (and non N).
* In any case we must not leave this routine with the interrupts
* disabled, pending packets in the VQ and hwtail == (hwcur - 1),
* otherwise the pending packets could stall.
*/
if (netmap_no_pendintr || force_update) {
uint32_t hwtail_lim = nm_prev(kring->nr_hwcur, lim);
@ -320,10 +323,17 @@ vtnet_netmap_rxsync(struct netmap_kring *kring, int flags)
vtnet_rxq_disable_intr(rxq);
nm_i = kring->nr_hwtail;
while (nm_i != hwtail_lim) {
for (;;) {
int len;
token = virtqueue_dequeue(vq, &len);
if (token == NULL) {
/*
* Enable the interrupts again and double-check
* for more work. We can go on until we win the
* race condition, since we are not replenishing
* in the meanwhile, and thus we will process at
* most N-1 slots.
*/
if (interrupts && vtnet_rxq_enable_intr(rxq)) {
vtnet_rxq_disable_intr(rxq);
continue;
@ -333,6 +343,11 @@ vtnet_netmap_rxsync(struct netmap_kring *kring, int flags)
if (unlikely(token != (void *)rxq)) {
nm_prerr("BUG: RX token mismatch");
} else {
if (nm_i == hwtail_lim) {
KASSERT(false, ("hwtail would "
"overrun hwcur"));
}
/* Skip the virtio-net header. */
len -= sc->vtnet_hdr_size;
if (unlikely(len < 0)) {