2 * xen paravirt network card backend
4 * (c) Gerd Hoffmann <kraxel@redhat.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; under version 2 of the License.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, see <http://www.gnu.org/licenses/>.
28 #include <sys/socket.h>
29 #include <sys/ioctl.h>
30 #include <sys/types.h>
37 #include <xen/io/xenbus.h>
38 #include <xen/io/netif.h>
42 #include "net/checksum.h"
44 #include "qemu-char.h"
45 #include "xen_backend.h"
47 /* ------------------------------------------------------------- */
50 struct XenDevice xendev; /* must be first */
55 struct netif_tx_sring *txs;
56 struct netif_rx_sring *rxs;
57 netif_tx_back_ring_t tx_ring;
58 netif_rx_back_ring_t rx_ring;
63 /* ------------------------------------------------------------- */
65 static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
67 RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
68 netif_tx_response_t *resp;
71 resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
76 if (txp->flags & NETTXF_extra_info) {
77 RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
81 netdev->tx_ring.rsp_prod_pvt = ++i;
82 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
84 xen_be_send_notify(&netdev->xendev);
87 if (i == netdev->tx_ring.req_cons) {
89 RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
96 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
100 * Hmm, why netback fails everything in the ring?
101 * Should we do that even when not supporting SG and TSO?
103 RING_IDX cons = netdev->tx_ring.req_cons;
106 make_tx_response(netif, txp, NETIF_RSP_ERROR);
110 txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
112 netdev->tx_ring.req_cons = cons;
113 netif_schedule_work(netif);
116 net_tx_response(netdev, txp, NETIF_RSP_ERROR);
120 static void net_tx_packets(struct XenNetDev *netdev)
122 netif_tx_request_t txreq;
128 rc = netdev->tx_ring.req_cons;
129 rp = netdev->tx_ring.sring->req_prod;
130 xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
133 if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc)) {
136 memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
137 netdev->tx_ring.req_cons = ++rc;
140 /* should not happen in theory, we don't announce the *
141 * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
142 if (txreq.flags & NETTXF_extra_info) {
143 xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
144 net_tx_error(netdev, &txreq, rc);
147 if (txreq.flags & NETTXF_more_data) {
148 xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
149 net_tx_error(netdev, &txreq, rc);
154 if (txreq.size < 14) {
155 xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
156 net_tx_error(netdev, &txreq, rc);
160 if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
161 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
162 net_tx_error(netdev, &txreq, rc);
166 xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
167 txreq.gref, txreq.offset, txreq.size, txreq.flags,
168 (txreq.flags & NETTXF_csum_blank) ? " csum_blank" : "",
169 (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
170 (txreq.flags & NETTXF_more_data) ? " more_data" : "",
171 (txreq.flags & NETTXF_extra_info) ? " extra_info" : "");
173 page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
175 txreq.gref, PROT_READ);
177 xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
179 net_tx_error(netdev, &txreq, rc);
182 if (txreq.flags & NETTXF_csum_blank) {
183 /* have read-only mapping -> can't fill checksum in-place */
185 tmpbuf = g_malloc(XC_PAGE_SIZE);
187 memcpy(tmpbuf, page + txreq.offset, txreq.size);
188 net_checksum_calculate(tmpbuf, txreq.size);
189 qemu_send_packet(&netdev->nic->nc, tmpbuf, txreq.size);
191 qemu_send_packet(&netdev->nic->nc, page + txreq.offset, txreq.size);
193 xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
194 net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
196 if (!netdev->tx_work) {
204 /* ------------------------------------------------------------- */
206 static void net_rx_response(struct XenNetDev *netdev,
207 netif_rx_request_t *req, int8_t st,
208 uint16_t offset, uint16_t size,
211 RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
212 netif_rx_response_t *resp;
215 resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
216 resp->offset = offset;
219 resp->status = (int16_t)size;
221 resp->status = (int16_t)st;
224 xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
225 i, resp->status, resp->flags);
227 netdev->rx_ring.rsp_prod_pvt = ++i;
228 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
230 xen_be_send_notify(&netdev->xendev);
234 #define NET_IP_ALIGN 2
236 static int net_rx_ok(VLANClientState *nc)
238 struct XenNetDev *netdev = DO_UPCAST(NICState, nc, nc)->opaque;
241 if (netdev->xendev.be_state != XenbusStateConnected) {
245 rc = netdev->rx_ring.req_cons;
246 rp = netdev->rx_ring.sring->req_prod;
249 if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
250 xen_be_printf(&netdev->xendev, 2, "%s: no rx buffers (%d/%d)\n",
251 __FUNCTION__, rc, rp);
257 static ssize_t net_rx_packet(VLANClientState *nc, const uint8_t *buf, size_t size)
259 struct XenNetDev *netdev = DO_UPCAST(NICState, nc, nc)->opaque;
260 netif_rx_request_t rxreq;
264 if (netdev->xendev.be_state != XenbusStateConnected) {
268 rc = netdev->rx_ring.req_cons;
269 rp = netdev->rx_ring.sring->req_prod;
270 xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
272 if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
273 xen_be_printf(&netdev->xendev, 2, "no buffer, drop packet\n");
276 if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
277 xen_be_printf(&netdev->xendev, 0, "packet too big (%lu > %ld)",
278 (unsigned long)size, XC_PAGE_SIZE - NET_IP_ALIGN);
282 memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
283 netdev->rx_ring.req_cons = ++rc;
285 page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
287 rxreq.gref, PROT_WRITE);
289 xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
291 net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
294 memcpy(page + NET_IP_ALIGN, buf, size);
295 xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
296 net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
301 /* ------------------------------------------------------------- */
303 static NetClientInfo net_xen_info = {
304 .type = NET_CLIENT_TYPE_NIC,
305 .size = sizeof(NICState),
306 .can_receive = net_rx_ok,
307 .receive = net_rx_packet,
310 static int net_init(struct XenDevice *xendev)
312 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
314 /* read xenstore entries */
315 if (netdev->mac == NULL) {
316 netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
319 /* do we have all we need? */
320 if (netdev->mac == NULL) {
324 if (net_parse_macaddr(netdev->conf.macaddr.a, netdev->mac) < 0) {
328 netdev->conf.vlan = qemu_find_vlan(netdev->xendev.dev, 1);
329 netdev->conf.peer = NULL;
331 netdev->nic = qemu_new_nic(&net_xen_info, &netdev->conf,
332 "xen", NULL, netdev);
334 snprintf(netdev->nic->nc.info_str, sizeof(netdev->nic->nc.info_str),
335 "nic: xenbus vif macaddr=%s", netdev->mac);
338 xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
339 xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
344 static int net_connect(struct XenDevice *xendev)
346 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
349 if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
350 &netdev->tx_ring_ref) == -1) {
353 if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
354 &netdev->rx_ring_ref) == -1) {
357 if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
358 &netdev->xendev.remote_port) == -1) {
362 if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1) {
366 xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
370 netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
373 PROT_READ | PROT_WRITE);
374 netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
377 PROT_READ | PROT_WRITE);
378 if (!netdev->txs || !netdev->rxs) {
381 BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
382 BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
384 xen_be_bind_evtchn(&netdev->xendev);
386 xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
387 "remote port %d, local port %d\n",
388 netdev->tx_ring_ref, netdev->rx_ring_ref,
389 netdev->xendev.remote_port, netdev->xendev.local_port);
391 net_tx_packets(netdev);
395 static void net_disconnect(struct XenDevice *xendev)
397 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
399 xen_be_unbind_evtchn(&netdev->xendev);
402 xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
406 xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
410 qemu_del_vlan_client(&netdev->nic->nc);
415 static void net_event(struct XenDevice *xendev)
417 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
418 net_tx_packets(netdev);
421 static int net_free(struct XenDevice *xendev)
423 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
429 /* ------------------------------------------------------------- */
431 struct XenDevOps xen_netdev_ops = {
432 .size = sizeof(struct XenNetDev),
433 .flags = DEVOPS_FLAG_NEED_GNTDEV,
435 .initialise = net_connect,
437 .disconnect = net_disconnect,