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, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
30 #include <sys/socket.h>
31 #include <sys/ioctl.h>
32 #include <sys/types.h>
39 #include <xen/io/xenbus.h>
40 #include <xen/io/netif.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;
62 /* ------------------------------------------------------------- */
64 static void net_tx_response(struct XenNetDev *netdev, netif_tx_request_t *txp, int8_t st)
66 RING_IDX i = netdev->tx_ring.rsp_prod_pvt;
67 netif_tx_response_t *resp;
70 resp = RING_GET_RESPONSE(&netdev->tx_ring, i);
75 if (txp->flags & NETTXF_extra_info)
76 RING_GET_RESPONSE(&netdev->tx_ring, ++i)->status = NETIF_RSP_NULL;
79 netdev->tx_ring.rsp_prod_pvt = ++i;
80 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->tx_ring, notify);
82 xen_be_send_notify(&netdev->xendev);
84 if (i == netdev->tx_ring.req_cons) {
86 RING_FINAL_CHECK_FOR_REQUESTS(&netdev->tx_ring, more_to_do);
92 static void net_tx_error(struct XenNetDev *netdev, netif_tx_request_t *txp, RING_IDX end)
96 * Hmm, why netback fails everything in the ring?
97 * Should we do that even when not supporting SG and TSO?
99 RING_IDX cons = netdev->tx_ring.req_cons;
102 make_tx_response(netif, txp, NETIF_RSP_ERROR);
105 txp = RING_GET_REQUEST(&netdev->tx_ring, cons++);
107 netdev->tx_ring.req_cons = cons;
108 netif_schedule_work(netif);
111 net_tx_response(netdev, txp, NETIF_RSP_ERROR);
115 static void net_tx_packets(struct XenNetDev *netdev)
117 netif_tx_request_t txreq;
123 rc = netdev->tx_ring.req_cons;
124 rp = netdev->tx_ring.sring->req_prod;
125 xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
128 if (RING_REQUEST_CONS_OVERFLOW(&netdev->tx_ring, rc))
130 memcpy(&txreq, RING_GET_REQUEST(&netdev->tx_ring, rc), sizeof(txreq));
131 netdev->tx_ring.req_cons = ++rc;
134 /* should not happen in theory, we don't announce the *
135 * feature-{sg,gso,whatelse} flags in xenstore (yet?) */
136 if (txreq.flags & NETTXF_extra_info) {
137 xen_be_printf(&netdev->xendev, 0, "FIXME: extra info flag\n");
138 net_tx_error(netdev, &txreq, rc);
141 if (txreq.flags & NETTXF_more_data) {
142 xen_be_printf(&netdev->xendev, 0, "FIXME: more data flag\n");
143 net_tx_error(netdev, &txreq, rc);
148 if (txreq.size < 14) {
149 xen_be_printf(&netdev->xendev, 0, "bad packet size: %d\n", txreq.size);
150 net_tx_error(netdev, &txreq, rc);
154 if ((txreq.offset + txreq.size) > XC_PAGE_SIZE) {
155 xen_be_printf(&netdev->xendev, 0, "error: page crossing\n");
156 net_tx_error(netdev, &txreq, rc);
160 xen_be_printf(&netdev->xendev, 3, "tx packet ref %d, off %d, len %d, flags 0x%x%s%s%s%s\n",
161 txreq.gref, txreq.offset, txreq.size, txreq.flags,
162 (txreq.flags & NETTXF_csum_blank) ? " csum_blank" : "",
163 (txreq.flags & NETTXF_data_validated) ? " data_validated" : "",
164 (txreq.flags & NETTXF_more_data) ? " more_data" : "",
165 (txreq.flags & NETTXF_extra_info) ? " extra_info" : "");
167 page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
169 txreq.gref, PROT_READ);
171 xen_be_printf(&netdev->xendev, 0, "error: tx gref dereference failed (%d)\n",
173 net_tx_error(netdev, &txreq, rc);
176 if (txreq.flags & NETTXF_csum_blank) {
177 /* have read-only mapping -> can't fill checksum in-place */
179 tmpbuf = malloc(PAGE_SIZE);
180 memcpy(tmpbuf, page + txreq.offset, txreq.size);
181 net_checksum_calculate(tmpbuf, txreq.size);
182 qemu_send_packet(netdev->vs, tmpbuf, txreq.size);
184 qemu_send_packet(netdev->vs, page + txreq.offset, txreq.size);
186 xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
187 net_tx_response(netdev, &txreq, NETIF_RSP_OKAY);
189 if (!netdev->tx_work)
196 /* ------------------------------------------------------------- */
198 static void net_rx_response(struct XenNetDev *netdev,
199 netif_rx_request_t *req, int8_t st,
200 uint16_t offset, uint16_t size,
203 RING_IDX i = netdev->rx_ring.rsp_prod_pvt;
204 netif_rx_response_t *resp;
207 resp = RING_GET_RESPONSE(&netdev->rx_ring, i);
208 resp->offset = offset;
211 resp->status = (int16_t)size;
213 resp->status = (int16_t)st;
215 xen_be_printf(&netdev->xendev, 3, "rx response: idx %d, status %d, flags 0x%x\n",
216 i, resp->status, resp->flags);
218 netdev->rx_ring.rsp_prod_pvt = ++i;
219 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&netdev->rx_ring, notify);
221 xen_be_send_notify(&netdev->xendev);
224 #define NET_IP_ALIGN 2
226 static int net_rx_ok(void *opaque)
228 struct XenNetDev *netdev = opaque;
231 if (netdev->xendev.be_state != XenbusStateConnected)
234 rc = netdev->rx_ring.req_cons;
235 rp = netdev->rx_ring.sring->req_prod;
238 if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
239 xen_be_printf(&netdev->xendev, 2, "%s: no rx buffers (%d/%d)\n",
240 __FUNCTION__, rc, rp);
246 static void net_rx_packet(void *opaque, const uint8_t *buf, int size)
248 struct XenNetDev *netdev = opaque;
249 netif_rx_request_t rxreq;
253 if (netdev->xendev.be_state != XenbusStateConnected)
256 rc = netdev->rx_ring.req_cons;
257 rp = netdev->rx_ring.sring->req_prod;
258 xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
260 if (rc == rp || RING_REQUEST_CONS_OVERFLOW(&netdev->rx_ring, rc)) {
261 xen_be_printf(&netdev->xendev, 2, "no buffer, drop packet\n");
264 if (size > XC_PAGE_SIZE - NET_IP_ALIGN) {
265 xen_be_printf(&netdev->xendev, 0, "packet too big (%d > %ld)",
266 size, XC_PAGE_SIZE - NET_IP_ALIGN);
270 memcpy(&rxreq, RING_GET_REQUEST(&netdev->rx_ring, rc), sizeof(rxreq));
271 netdev->rx_ring.req_cons = ++rc;
273 page = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
275 rxreq.gref, PROT_WRITE);
277 xen_be_printf(&netdev->xendev, 0, "error: rx gref dereference failed (%d)\n",
279 net_rx_response(netdev, &rxreq, NETIF_RSP_ERROR, 0, 0, 0);
282 memcpy(page + NET_IP_ALIGN, buf, size);
283 xc_gnttab_munmap(netdev->xendev.gnttabdev, page, 1);
284 net_rx_response(netdev, &rxreq, NETIF_RSP_OKAY, NET_IP_ALIGN, size, 0);
287 /* ------------------------------------------------------------- */
289 static int net_init(struct XenDevice *xendev)
291 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
294 /* read xenstore entries */
295 if (netdev->mac == NULL)
296 netdev->mac = xenstore_read_be_str(&netdev->xendev, "mac");
298 /* do we have all we need? */
299 if (netdev->mac == NULL)
302 vlan = qemu_find_vlan(netdev->xendev.dev);
303 netdev->vs = qemu_new_vlan_client(vlan, "xen", NULL,
304 net_rx_packet, net_rx_ok, NULL,
306 snprintf(netdev->vs->info_str, sizeof(netdev->vs->info_str),
307 "nic: xenbus vif macaddr=%s", netdev->mac);
310 xenstore_write_be_int(&netdev->xendev, "feature-rx-copy", 1);
311 xenstore_write_be_int(&netdev->xendev, "feature-rx-flip", 0);
316 static int net_connect(struct XenDevice *xendev)
318 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
321 if (xenstore_read_fe_int(&netdev->xendev, "tx-ring-ref",
322 &netdev->tx_ring_ref) == -1)
324 if (xenstore_read_fe_int(&netdev->xendev, "rx-ring-ref",
325 &netdev->rx_ring_ref) == -1)
327 if (xenstore_read_fe_int(&netdev->xendev, "event-channel",
328 &netdev->xendev.remote_port) == -1)
331 if (xenstore_read_fe_int(&netdev->xendev, "request-rx-copy", &rx_copy) == -1)
334 xen_be_printf(&netdev->xendev, 0, "frontend doesn't support rx-copy.\n");
338 netdev->txs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
341 PROT_READ | PROT_WRITE);
342 netdev->rxs = xc_gnttab_map_grant_ref(netdev->xendev.gnttabdev,
345 PROT_READ | PROT_WRITE);
346 if (!netdev->txs || !netdev->rxs)
348 BACK_RING_INIT(&netdev->tx_ring, netdev->txs, XC_PAGE_SIZE);
349 BACK_RING_INIT(&netdev->rx_ring, netdev->rxs, XC_PAGE_SIZE);
351 xen_be_bind_evtchn(&netdev->xendev);
353 xen_be_printf(&netdev->xendev, 1, "ok: tx-ring-ref %d, rx-ring-ref %d, "
354 "remote port %d, local port %d\n",
355 netdev->tx_ring_ref, netdev->rx_ring_ref,
356 netdev->xendev.remote_port, netdev->xendev.local_port);
360 static void net_disconnect(struct XenDevice *xendev)
362 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
364 xen_be_unbind_evtchn(&netdev->xendev);
367 xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->txs, 1);
371 xc_gnttab_munmap(netdev->xendev.gnttabdev, netdev->rxs, 1);
375 qemu_del_vlan_client(netdev->vs);
380 static void net_event(struct XenDevice *xendev)
382 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
383 net_tx_packets(netdev);
386 static int net_free(struct XenDevice *xendev)
388 struct XenNetDev *netdev = container_of(xendev, struct XenNetDev, xendev);
390 qemu_free(netdev->mac);
394 /* ------------------------------------------------------------- */
396 struct XenDevOps xen_netdev_ops = {
397 .size = sizeof(struct XenNetDev),
398 .flags = DEVOPS_FLAG_NEED_GNTDEV,
400 .connect = net_connect,
402 .disconnect = net_disconnect,