Index: stable/12/usr.sbin/bhyve/net_backends.c =================================================================== --- stable/12/usr.sbin/bhyve/net_backends.c (revision 358184) +++ stable/12/usr.sbin/bhyve/net_backends.c (revision 358185) @@ -1,873 +1,814 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019 Vincenzo Maffione * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ /* * This file implements multiple network backends (tap, netmap, ...), * to be used by network frontends such as virtio-net and e1000. * The API to access the backend (e.g. send/receive packets, negotiate * features) is exported by net_backends.h. */ #include __FBSDID("$FreeBSD$"); #include /* u_short etc */ #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include #include #include #include #define NETMAP_WITH_LIBS #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "debug.h" #include "iov.h" #include "mevent.h" #include "net_backends.h" #include /* * Each network backend registers a set of function pointers that are * used to implement the net backends API. * This might need to be exposed if we implement backends in separate files. */ struct net_backend { const char *prefix; /* prefix matching this backend */ /* * Routines used to initialize and cleanup the resources needed * by a backend. The cleanup function is used internally, * and should not be called by the frontend. */ int (*init)(struct net_backend *be, const char *devname, net_be_rxeof_t cb, void *param); void (*cleanup)(struct net_backend *be); /* * Called to serve a guest transmit request. The scatter-gather * vector provided by the caller has 'iovcnt' elements and contains * the packet to send. */ - ssize_t (*send)(struct net_backend *be, struct iovec *iov, int iovcnt); + ssize_t (*send)(struct net_backend *be, const struct iovec *iov, + int iovcnt); /* * Called to receive a packet from the backend. When the function * returns a positive value 'len', the scatter-gather vector * provided by the caller contains a packet with such length. * The function returns 0 if the backend doesn't have a new packet to * receive. */ - ssize_t (*recv)(struct net_backend *be, struct iovec *iov, int iovcnt); + ssize_t (*recv)(struct net_backend *be, const struct iovec *iov, + int iovcnt); /* * Ask the backend to enable or disable receive operation in the * backend. On return from a disable operation, it is guaranteed * that the receive callback won't be called until receive is * enabled again. Note however that it is up to the caller to make * sure that netbe_recv() is not currently being executed by another * thread. */ void (*recv_enable)(struct net_backend *be); void (*recv_disable)(struct net_backend *be); /* * Ask the backend for the virtio-net features it is able to * support. Possible features are TSO, UFO and checksum offloading * in both rx and tx direction and for both IPv4 and IPv6. */ uint64_t (*get_cap)(struct net_backend *be); /* * Tell the backend to enable/disable the specified virtio-net * features (capabilities). */ int (*set_cap)(struct net_backend *be, uint64_t features, unsigned int vnet_hdr_len); struct pci_vtnet_softc *sc; int fd; /* * Length of the virtio-net header used by the backend and the * frontend, respectively. A zero value means that the header * is not used. */ unsigned int be_vnet_hdr_len; unsigned int fe_vnet_hdr_len; /* Size of backend-specific private data. */ size_t priv_size; /* Room for backend-specific data. */ char opaque[0]; }; SET_DECLARE(net_backend_set, struct net_backend); #define VNET_HDR_LEN sizeof(struct virtio_net_rxhdr) #define WPRINTF(params) PRINTLN params /* * The tap backend */ struct tap_priv { struct mevent *mevp; }; static void tap_cleanup(struct net_backend *be) { struct tap_priv *priv = (struct tap_priv *)be->opaque; if (priv->mevp) { mevent_delete(priv->mevp); } if (be->fd != -1) { close(be->fd); be->fd = -1; } } static int tap_init(struct net_backend *be, const char *devname, net_be_rxeof_t cb, void *param) { struct tap_priv *priv = (struct tap_priv *)be->opaque; char tbuf[80]; int opt = 1; #ifndef WITHOUT_CAPSICUM cap_rights_t rights; #endif if (cb == NULL) { WPRINTF(("TAP backend requires non-NULL callback")); return (-1); } strcpy(tbuf, "/dev/"); strlcat(tbuf, devname, sizeof(tbuf)); be->fd = open(tbuf, O_RDWR); if (be->fd == -1) { WPRINTF(("open of tap device %s failed", tbuf)); goto error; } /* * Set non-blocking and register for read * notifications with the event loop */ if (ioctl(be->fd, FIONBIO, &opt) < 0) { WPRINTF(("tap device O_NONBLOCK failed")); goto error; } #ifndef WITHOUT_CAPSICUM cap_rights_init(&rights, CAP_EVENT, CAP_READ, CAP_WRITE); if (caph_rights_limit(be->fd, &rights) == -1) errx(EX_OSERR, "Unable to apply rights for sandbox"); #endif priv->mevp = mevent_add_disabled(be->fd, EVF_READ, cb, param); if (priv->mevp == NULL) { WPRINTF(("Could not register event")); goto error; } return (0); error: tap_cleanup(be); return (-1); } /* * Called to send a buffer chain out to the tap device */ static ssize_t -tap_send(struct net_backend *be, struct iovec *iov, int iovcnt) +tap_send(struct net_backend *be, const struct iovec *iov, int iovcnt) { return (writev(be->fd, iov, iovcnt)); } static ssize_t -tap_recv(struct net_backend *be, struct iovec *iov, int iovcnt) +tap_recv(struct net_backend *be, const struct iovec *iov, int iovcnt) { ssize_t ret; /* Should never be called without a valid tap fd */ assert(be->fd != -1); ret = readv(be->fd, iov, iovcnt); if (ret < 0 && errno == EWOULDBLOCK) { return (0); } return (ret); } static void tap_recv_enable(struct net_backend *be) { struct tap_priv *priv = (struct tap_priv *)be->opaque; mevent_enable(priv->mevp); } static void tap_recv_disable(struct net_backend *be) { struct tap_priv *priv = (struct tap_priv *)be->opaque; mevent_disable(priv->mevp); } static uint64_t tap_get_cap(struct net_backend *be) { return (0); /* no capabilities for now */ } static int tap_set_cap(struct net_backend *be, uint64_t features, unsigned vnet_hdr_len) { return ((features || vnet_hdr_len) ? -1 : 0); } static struct net_backend tap_backend = { .prefix = "tap", .priv_size = sizeof(struct tap_priv), .init = tap_init, .cleanup = tap_cleanup, .send = tap_send, .recv = tap_recv, .recv_enable = tap_recv_enable, .recv_disable = tap_recv_disable, .get_cap = tap_get_cap, .set_cap = tap_set_cap, }; /* A clone of the tap backend, with a different prefix. */ static struct net_backend vmnet_backend = { .prefix = "vmnet", .priv_size = sizeof(struct tap_priv), .init = tap_init, .cleanup = tap_cleanup, .send = tap_send, .recv = tap_recv, .recv_enable = tap_recv_enable, .recv_disable = tap_recv_disable, .get_cap = tap_get_cap, .set_cap = tap_set_cap, }; DATA_SET(net_backend_set, tap_backend); DATA_SET(net_backend_set, vmnet_backend); /* * The netmap backend */ /* The virtio-net features supported by netmap. */ #define NETMAP_FEATURES (VIRTIO_NET_F_CSUM | VIRTIO_NET_F_HOST_TSO4 | \ VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_UFO | \ VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 | \ VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_UFO | \ VIRTIO_NET_F_MRG_RXBUF) struct netmap_priv { char ifname[IFNAMSIZ]; struct nm_desc *nmd; uint16_t memid; struct netmap_ring *rx; struct netmap_ring *tx; struct mevent *mevp; net_be_rxeof_t cb; void *cb_param; }; static void nmreq_init(struct nmreq *req, char *ifname) { memset(req, 0, sizeof(*req)); strlcpy(req->nr_name, ifname, sizeof(req->nr_name)); req->nr_version = NETMAP_API; } static int netmap_set_vnet_hdr_len(struct net_backend *be, int vnet_hdr_len) { int err; struct nmreq req; struct netmap_priv *priv = (struct netmap_priv *)be->opaque; nmreq_init(&req, priv->ifname); req.nr_cmd = NETMAP_BDG_VNET_HDR; req.nr_arg1 = vnet_hdr_len; err = ioctl(be->fd, NIOCREGIF, &req); if (err) { WPRINTF(("Unable to set vnet header length %d", vnet_hdr_len)); return (err); } be->be_vnet_hdr_len = vnet_hdr_len; return (0); } static int netmap_has_vnet_hdr_len(struct net_backend *be, unsigned vnet_hdr_len) { int prev_hdr_len = be->be_vnet_hdr_len; int ret; if (vnet_hdr_len == prev_hdr_len) { return (1); } ret = netmap_set_vnet_hdr_len(be, vnet_hdr_len); if (ret) { return (0); } netmap_set_vnet_hdr_len(be, prev_hdr_len); return (1); } static uint64_t netmap_get_cap(struct net_backend *be) { return (netmap_has_vnet_hdr_len(be, VNET_HDR_LEN) ? NETMAP_FEATURES : 0); } static int netmap_set_cap(struct net_backend *be, uint64_t features, unsigned vnet_hdr_len) { return (netmap_set_vnet_hdr_len(be, vnet_hdr_len)); } static int netmap_init(struct net_backend *be, const char *devname, net_be_rxeof_t cb, void *param) { struct netmap_priv *priv = (struct netmap_priv *)be->opaque; strlcpy(priv->ifname, devname, sizeof(priv->ifname)); priv->ifname[sizeof(priv->ifname) - 1] = '\0'; priv->nmd = nm_open(priv->ifname, NULL, NETMAP_NO_TX_POLL, NULL); if (priv->nmd == NULL) { WPRINTF(("Unable to nm_open(): interface '%s', errno (%s)", devname, strerror(errno))); free(priv); return (-1); } priv->memid = priv->nmd->req.nr_arg2; priv->tx = NETMAP_TXRING(priv->nmd->nifp, 0); priv->rx = NETMAP_RXRING(priv->nmd->nifp, 0); priv->cb = cb; priv->cb_param = param; be->fd = priv->nmd->fd; priv->mevp = mevent_add_disabled(be->fd, EVF_READ, cb, param); if (priv->mevp == NULL) { WPRINTF(("Could not register event")); return (-1); } return (0); } static void netmap_cleanup(struct net_backend *be) { struct netmap_priv *priv = (struct netmap_priv *)be->opaque; if (priv->mevp) { mevent_delete(priv->mevp); } if (priv->nmd) { nm_close(priv->nmd); } be->fd = -1; } static ssize_t -netmap_send(struct net_backend *be, struct iovec *iov, +netmap_send(struct net_backend *be, const struct iovec *iov, int iovcnt) { struct netmap_priv *priv = (struct netmap_priv *)be->opaque; struct netmap_ring *ring; ssize_t totlen = 0; int nm_buf_size; int nm_buf_len; uint32_t head; void *nm_buf; int j; ring = priv->tx; head = ring->head; if (head == ring->tail) { WPRINTF(("No space, drop %zu bytes", count_iov(iov, iovcnt))); goto txsync; } nm_buf = NETMAP_BUF(ring, ring->slot[head].buf_idx); nm_buf_size = ring->nr_buf_size; nm_buf_len = 0; for (j = 0; j < iovcnt; j++) { int iov_frag_size = iov[j].iov_len; void *iov_frag_buf = iov[j].iov_base; totlen += iov_frag_size; /* * Split each iovec fragment over more netmap slots, if * necessary. */ for (;;) { int copylen; copylen = iov_frag_size < nm_buf_size ? iov_frag_size : nm_buf_size; memcpy(nm_buf, iov_frag_buf, copylen); iov_frag_buf += copylen; iov_frag_size -= copylen; nm_buf += copylen; nm_buf_size -= copylen; nm_buf_len += copylen; if (iov_frag_size == 0) { break; } ring->slot[head].len = nm_buf_len; ring->slot[head].flags = NS_MOREFRAG; head = nm_ring_next(ring, head); if (head == ring->tail) { /* * We ran out of netmap slots while * splitting the iovec fragments. */ WPRINTF(("No space, drop %zu bytes", count_iov(iov, iovcnt))); goto txsync; } nm_buf = NETMAP_BUF(ring, ring->slot[head].buf_idx); nm_buf_size = ring->nr_buf_size; nm_buf_len = 0; } } /* Complete the last slot, which must not have NS_MOREFRAG set. */ ring->slot[head].len = nm_buf_len; ring->slot[head].flags = 0; head = nm_ring_next(ring, head); /* Now update ring->head and ring->cur. */ ring->head = ring->cur = head; txsync: ioctl(be->fd, NIOCTXSYNC, NULL); return (totlen); } static ssize_t -netmap_recv(struct net_backend *be, struct iovec *iov, int iovcnt) +netmap_recv(struct net_backend *be, const struct iovec *iov, int iovcnt) { struct netmap_priv *priv = (struct netmap_priv *)be->opaque; struct netmap_slot *slot = NULL; struct netmap_ring *ring; void *iov_frag_buf; int iov_frag_size; ssize_t totlen = 0; uint32_t head; assert(iovcnt); ring = priv->rx; head = ring->head; iov_frag_buf = iov->iov_base; iov_frag_size = iov->iov_len; do { int nm_buf_len; void *nm_buf; if (head == ring->tail) { return (0); } slot = ring->slot + head; nm_buf = NETMAP_BUF(ring, slot->buf_idx); nm_buf_len = slot->len; for (;;) { int copylen = nm_buf_len < iov_frag_size ? nm_buf_len : iov_frag_size; memcpy(iov_frag_buf, nm_buf, copylen); nm_buf += copylen; nm_buf_len -= copylen; iov_frag_buf += copylen; iov_frag_size -= copylen; totlen += copylen; if (nm_buf_len == 0) { break; } iov++; iovcnt--; if (iovcnt == 0) { /* No space to receive. */ WPRINTF(("Short iov, drop %zd bytes", totlen)); return (-ENOSPC); } iov_frag_buf = iov->iov_base; iov_frag_size = iov->iov_len; } head = nm_ring_next(ring, head); } while (slot->flags & NS_MOREFRAG); /* Release slots to netmap. */ ring->head = ring->cur = head; return (totlen); } static void netmap_recv_enable(struct net_backend *be) { struct netmap_priv *priv = (struct netmap_priv *)be->opaque; mevent_enable(priv->mevp); } static void netmap_recv_disable(struct net_backend *be) { struct netmap_priv *priv = (struct netmap_priv *)be->opaque; mevent_disable(priv->mevp); } static struct net_backend netmap_backend = { .prefix = "netmap", .priv_size = sizeof(struct netmap_priv), .init = netmap_init, .cleanup = netmap_cleanup, .send = netmap_send, .recv = netmap_recv, .recv_enable = netmap_recv_enable, .recv_disable = netmap_recv_disable, .get_cap = netmap_get_cap, .set_cap = netmap_set_cap, }; /* A clone of the netmap backend, with a different prefix. */ static struct net_backend vale_backend = { .prefix = "vale", .priv_size = sizeof(struct netmap_priv), .init = netmap_init, .cleanup = netmap_cleanup, .send = netmap_send, .recv = netmap_recv, .recv_enable = netmap_recv_enable, .recv_disable = netmap_recv_disable, .get_cap = netmap_get_cap, .set_cap = netmap_set_cap, }; DATA_SET(net_backend_set, netmap_backend); DATA_SET(net_backend_set, vale_backend); /* * Initialize a backend and attach to the frontend. * This is called during frontend initialization. * @pbe is a pointer to the backend to be initialized * @devname is the backend-name as supplied on the command line, * e.g. -s 2:0,frontend-name,backend-name[,other-args] * @cb is the receive callback supplied by the frontend, * and it is invoked in the event loop when a receive * event is generated in the hypervisor, * @param is a pointer to the frontend, and normally used as * the argument for the callback. */ int netbe_init(struct net_backend **ret, const char *devname, net_be_rxeof_t cb, void *param) { struct net_backend **pbe, *nbe, *tbe = NULL; int err; /* * Find the network backend that matches the user-provided * device name. net_backend_set is built using a linker set. */ SET_FOREACH(pbe, net_backend_set) { if (strncmp(devname, (*pbe)->prefix, strlen((*pbe)->prefix)) == 0) { tbe = *pbe; assert(tbe->init != NULL); assert(tbe->cleanup != NULL); assert(tbe->send != NULL); assert(tbe->recv != NULL); assert(tbe->get_cap != NULL); assert(tbe->set_cap != NULL); break; } } *ret = NULL; if (tbe == NULL) return (EINVAL); nbe = calloc(1, sizeof(*nbe) + tbe->priv_size); *nbe = *tbe; /* copy the template */ nbe->fd = -1; nbe->sc = param; nbe->be_vnet_hdr_len = 0; nbe->fe_vnet_hdr_len = 0; /* Initialize the backend. */ err = nbe->init(nbe, devname, cb, param); if (err) { free(nbe); return (err); } *ret = nbe; return (0); } void netbe_cleanup(struct net_backend *be) { if (be != NULL) { be->cleanup(be); free(be); } } uint64_t netbe_get_cap(struct net_backend *be) { assert(be != NULL); return (be->get_cap(be)); } int netbe_set_cap(struct net_backend *be, uint64_t features, unsigned vnet_hdr_len) { int ret; assert(be != NULL); /* There are only three valid lengths, i.e., 0, 10 and 12. */ if (vnet_hdr_len && vnet_hdr_len != VNET_HDR_LEN && vnet_hdr_len != (VNET_HDR_LEN - sizeof(uint16_t))) return (-1); be->fe_vnet_hdr_len = vnet_hdr_len; ret = be->set_cap(be, features, vnet_hdr_len); assert(be->be_vnet_hdr_len == 0 || be->be_vnet_hdr_len == be->fe_vnet_hdr_len); return (ret); } -static __inline struct iovec * -iov_trim(struct iovec *iov, int *iovcnt, unsigned int tlen) -{ - struct iovec *riov; - - /* XXX short-cut: assume first segment is >= tlen */ - assert(iov[0].iov_len >= tlen); - - iov[0].iov_len -= tlen; - if (iov[0].iov_len == 0) { - assert(*iovcnt > 1); - *iovcnt -= 1; - riov = &iov[1]; - } else { - iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + tlen); - riov = &iov[0]; - } - - return (riov); -} - ssize_t -netbe_send(struct net_backend *be, struct iovec *iov, int iovcnt) +netbe_send(struct net_backend *be, const struct iovec *iov, int iovcnt) { - assert(be != NULL); - if (be->be_vnet_hdr_len != be->fe_vnet_hdr_len) { - /* - * The frontend uses a virtio-net header, but the backend - * does not. We ignore it (as it must be all zeroes) and - * strip it. - */ - assert(be->be_vnet_hdr_len == 0); - iov = iov_trim(iov, &iovcnt, be->fe_vnet_hdr_len); - } - return (be->send(be, iov, iovcnt)); } /* * Try to read a packet from the backend, without blocking. * If no packets are available, return 0. In case of success, return * the length of the packet just read. Return -1 in case of errors. */ ssize_t -netbe_recv(struct net_backend *be, struct iovec *iov, int iovcnt) +netbe_recv(struct net_backend *be, const struct iovec *iov, int iovcnt) { - /* Length of prepended virtio-net header. */ - unsigned int hlen = be->fe_vnet_hdr_len; - int ret; - assert(be != NULL); - - if (hlen && hlen != be->be_vnet_hdr_len) { - /* - * The frontend uses a virtio-net header, but the backend - * does not. We need to prepend a zeroed header. - */ - struct virtio_net_rxhdr *vh; - - assert(be->be_vnet_hdr_len == 0); - - /* - * Get a pointer to the rx header, and use the - * data immediately following it for the packet buffer. - */ - vh = iov[0].iov_base; - iov = iov_trim(iov, &iovcnt, hlen); - - /* - * The only valid field in the rx packet header is the - * number of buffers if merged rx bufs were negotiated. - */ - memset(vh, 0, hlen); - if (hlen == VNET_HDR_LEN) { - vh->vrh_bufs = 1; - } - } - - ret = be->recv(be, iov, iovcnt); - if (ret > 0) { - ret += hlen; - } - - return (ret); + return (be->recv(be, iov, iovcnt)); } /* * Read a packet from the backend and discard it. * Returns the size of the discarded packet or zero if no packet was available. * A negative error code is returned in case of read error. */ ssize_t netbe_rx_discard(struct net_backend *be) { /* * MP note: the dummybuf is only used to discard frames, * so there is no need for it to be per-vtnet or locked. * We only make it large enough for TSO-sized segment. */ static uint8_t dummybuf[65536 + 64]; struct iovec iov; iov.iov_base = dummybuf; iov.iov_len = sizeof(dummybuf); return netbe_recv(be, &iov, 1); } void netbe_rx_disable(struct net_backend *be) { return be->recv_disable(be); } void netbe_rx_enable(struct net_backend *be) { return be->recv_enable(be); +} + +size_t +netbe_get_vnet_hdr_len(struct net_backend *be) +{ + + return (be->be_vnet_hdr_len); } Index: stable/12/usr.sbin/bhyve/net_backends.h =================================================================== --- stable/12/usr.sbin/bhyve/net_backends.h (revision 358184) +++ stable/12/usr.sbin/bhyve/net_backends.h (revision 358185) @@ -1,91 +1,92 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019 Vincenzo Maffione * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ #ifndef __NET_BACKENDS_H__ #define __NET_BACKENDS_H__ #include /* Opaque type representing a network backend. */ typedef struct net_backend net_backend_t; /* Interface between network frontends and the network backends. */ typedef void (*net_be_rxeof_t)(int, enum ev_type, void *param); int netbe_init(net_backend_t **be, const char *devname, net_be_rxeof_t cb, void *param); void netbe_cleanup(net_backend_t *be); uint64_t netbe_get_cap(net_backend_t *be); int netbe_set_cap(net_backend_t *be, uint64_t cap, unsigned vnet_hdr_len); -ssize_t netbe_send(net_backend_t *be, struct iovec *iov, int iovcnt); -ssize_t netbe_recv(net_backend_t *be, struct iovec *iov, int iovcnt); +size_t netbe_get_vnet_hdr_len(net_backend_t *be); +ssize_t netbe_send(net_backend_t *be, const struct iovec *iov, int iovcnt); +ssize_t netbe_recv(net_backend_t *be, const struct iovec *iov, int iovcnt); ssize_t netbe_rx_discard(net_backend_t *be); void netbe_rx_disable(net_backend_t *be); void netbe_rx_enable(net_backend_t *be); /* * Network device capabilities taken from the VirtIO standard. * Despite the name, these capabilities can be used by different frontents * (virtio-net, ptnet) and supported by different backends (netmap, tap, ...). */ #define VIRTIO_NET_F_CSUM (1 << 0) /* host handles partial cksum */ #define VIRTIO_NET_F_GUEST_CSUM (1 << 1) /* guest handles partial cksum */ #define VIRTIO_NET_F_MAC (1 << 5) /* host supplies MAC */ #define VIRTIO_NET_F_GSO_DEPREC (1 << 6) /* deprecated: host handles GSO */ #define VIRTIO_NET_F_GUEST_TSO4 (1 << 7) /* guest can rcv TSOv4 */ #define VIRTIO_NET_F_GUEST_TSO6 (1 << 8) /* guest can rcv TSOv6 */ #define VIRTIO_NET_F_GUEST_ECN (1 << 9) /* guest can rcv TSO with ECN */ #define VIRTIO_NET_F_GUEST_UFO (1 << 10) /* guest can rcv UFO */ #define VIRTIO_NET_F_HOST_TSO4 (1 << 11) /* host can rcv TSOv4 */ #define VIRTIO_NET_F_HOST_TSO6 (1 << 12) /* host can rcv TSOv6 */ #define VIRTIO_NET_F_HOST_ECN (1 << 13) /* host can rcv TSO with ECN */ #define VIRTIO_NET_F_HOST_UFO (1 << 14) /* host can rcv UFO */ #define VIRTIO_NET_F_MRG_RXBUF (1 << 15) /* host can merge RX buffers */ #define VIRTIO_NET_F_STATUS (1 << 16) /* config status field available */ #define VIRTIO_NET_F_CTRL_VQ (1 << 17) /* control channel available */ #define VIRTIO_NET_F_CTRL_RX (1 << 18) /* control channel RX mode support */ #define VIRTIO_NET_F_CTRL_VLAN (1 << 19) /* control channel VLAN filtering */ #define VIRTIO_NET_F_GUEST_ANNOUNCE \ (1 << 21) /* guest can send gratuitous pkts */ /* * Fixed network header size */ struct virtio_net_rxhdr { uint8_t vrh_flags; uint8_t vrh_gso_type; uint16_t vrh_hdr_len; uint16_t vrh_gso_size; uint16_t vrh_csum_start; uint16_t vrh_csum_offset; uint16_t vrh_bufs; } __packed; #endif /* __NET_BACKENDS_H__ */ Index: stable/12/usr.sbin/bhyve/net_utils.c =================================================================== --- stable/12/usr.sbin/bhyve/net_utils.c (revision 358184) +++ stable/12/usr.sbin/bhyve/net_utils.c (revision 358185) @@ -1,89 +1,87 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011 NetApp, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ #include "net_utils.h" #include "bhyverun.h" #include #include #include #include #include #include "bhyverun.h" #include "debug.h" #include "net_utils.h" int net_parsemac(char *mac_str, uint8_t *mac_addr) { struct ether_addr *ea; - char *tmpstr; char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 }; - tmpstr = strsep(&mac_str,"="); + if (mac_str == NULL) + return (EINVAL); - if ((mac_str != NULL) && (!strcmp(tmpstr,"mac"))) { - ea = ether_aton(mac_str); + ea = ether_aton(mac_str); - if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) || - memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) { - EPRINTLN("Invalid MAC %s", mac_str); - return (EINVAL); - } else - memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN); - } + if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) || + memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) { + EPRINTLN("Invalid MAC %s", mac_str); + return (EINVAL); + } else + memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN); return (0); } void net_genmac(struct pci_devinst *pi, uint8_t *macaddr) { /* * The default MAC address is the standard NetApp OUI of 00-a0-98, * followed by an MD5 of the PCI slot/func number and dev name */ MD5_CTX mdctx; unsigned char digest[16]; char nstr[80]; snprintf(nstr, sizeof(nstr), "%d-%d-%s", pi->pi_slot, pi->pi_func, vmname); MD5Init(&mdctx); MD5Update(&mdctx, nstr, (unsigned int)strlen(nstr)); MD5Final(digest, &mdctx); macaddr[0] = 0x00; macaddr[1] = 0xa0; macaddr[2] = 0x98; macaddr[3] = digest[0]; macaddr[4] = digest[1]; macaddr[5] = digest[2]; } Index: stable/12/usr.sbin/bhyve/pci_e82545.c =================================================================== --- stable/12/usr.sbin/bhyve/pci_e82545.c (revision 358184) +++ stable/12/usr.sbin/bhyve/pci_e82545.c (revision 358185) @@ -1,2370 +1,2388 @@ /* * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2016 Alexander Motin * Copyright (c) 2015 Peter Grehan * Copyright (c) 2013 Jeremiah Lott, Avere Systems * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include #include #include #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include #include #include #include #include #include #include #include #include #include "e1000_regs.h" #include "e1000_defines.h" #include "mii.h" #include "bhyverun.h" #include "debug.h" #include "pci_emul.h" #include "mevent.h" #include "net_utils.h" #include "net_backends.h" /* Hardware/register definitions XXX: move some to common code. */ #define E82545_VENDOR_ID_INTEL 0x8086 #define E82545_DEV_ID_82545EM_COPPER 0x100F #define E82545_SUBDEV_ID 0x1008 #define E82545_REVISION_4 4 #define E82545_MDIC_DATA_MASK 0x0000FFFF #define E82545_MDIC_OP_MASK 0x0c000000 #define E82545_MDIC_IE 0x20000000 #define E82545_EECD_FWE_DIS 0x00000010 /* Flash writes disabled */ #define E82545_EECD_FWE_EN 0x00000020 /* Flash writes enabled */ #define E82545_EECD_FWE_MASK 0x00000030 /* Flash writes mask */ #define E82545_BAR_REGISTER 0 #define E82545_BAR_REGISTER_LEN (128*1024) #define E82545_BAR_FLASH 1 #define E82545_BAR_FLASH_LEN (64*1024) #define E82545_BAR_IO 2 #define E82545_BAR_IO_LEN 8 #define E82545_IOADDR 0x00000000 #define E82545_IODATA 0x00000004 #define E82545_IO_REGISTER_MAX 0x0001FFFF #define E82545_IO_FLASH_BASE 0x00080000 #define E82545_IO_FLASH_MAX 0x000FFFFF #define E82545_ARRAY_ENTRY(reg, offset) (reg + (offset<<2)) #define E82545_RAR_MAX 15 #define E82545_MTA_MAX 127 #define E82545_VFTA_MAX 127 /* Slightly modified from the driver versions, hardcoded for 3 opcode bits, * followed by 6 address bits. * TODO: make opcode bits and addr bits configurable? * NVM Commands - Microwire */ #define E82545_NVM_OPCODE_BITS 3 #define E82545_NVM_ADDR_BITS 6 #define E82545_NVM_DATA_BITS 16 #define E82545_NVM_OPADDR_BITS (E82545_NVM_OPCODE_BITS + E82545_NVM_ADDR_BITS) #define E82545_NVM_ADDR_MASK ((1 << E82545_NVM_ADDR_BITS)-1) #define E82545_NVM_OPCODE_MASK \ (((1 << E82545_NVM_OPCODE_BITS) - 1) << E82545_NVM_ADDR_BITS) #define E82545_NVM_OPCODE_READ (0x6 << E82545_NVM_ADDR_BITS) /* read */ #define E82545_NVM_OPCODE_WRITE (0x5 << E82545_NVM_ADDR_BITS) /* write */ #define E82545_NVM_OPCODE_ERASE (0x7 << E82545_NVM_ADDR_BITS) /* erase */ #define E82545_NVM_OPCODE_EWEN (0x4 << E82545_NVM_ADDR_BITS) /* wr-enable */ #define E82545_NVM_EEPROM_SIZE 64 /* 64 * 16-bit values == 128K */ #define E1000_ICR_SRPD 0x00010000 /* This is an arbitrary number. There is no hard limit on the chip. */ #define I82545_MAX_TXSEGS 64 /* Legacy receive descriptor */ struct e1000_rx_desc { uint64_t buffer_addr; /* Address of the descriptor's data buffer */ uint16_t length; /* Length of data DMAed into data buffer */ uint16_t csum; /* Packet checksum */ uint8_t status; /* Descriptor status */ uint8_t errors; /* Descriptor Errors */ uint16_t special; }; /* Transmit descriptor types */ #define E1000_TXD_MASK (E1000_TXD_CMD_DEXT | 0x00F00000) #define E1000_TXD_TYP_L (0) #define E1000_TXD_TYP_C (E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_C) #define E1000_TXD_TYP_D (E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D) /* Legacy transmit descriptor */ struct e1000_tx_desc { uint64_t buffer_addr; /* Address of the descriptor's data buffer */ union { uint32_t data; struct { uint16_t length; /* Data buffer length */ uint8_t cso; /* Checksum offset */ uint8_t cmd; /* Descriptor control */ } flags; } lower; union { uint32_t data; struct { uint8_t status; /* Descriptor status */ uint8_t css; /* Checksum start */ uint16_t special; } fields; } upper; }; /* Context descriptor */ struct e1000_context_desc { union { uint32_t ip_config; struct { uint8_t ipcss; /* IP checksum start */ uint8_t ipcso; /* IP checksum offset */ uint16_t ipcse; /* IP checksum end */ } ip_fields; } lower_setup; union { uint32_t tcp_config; struct { uint8_t tucss; /* TCP checksum start */ uint8_t tucso; /* TCP checksum offset */ uint16_t tucse; /* TCP checksum end */ } tcp_fields; } upper_setup; uint32_t cmd_and_length; union { uint32_t data; struct { uint8_t status; /* Descriptor status */ uint8_t hdr_len; /* Header length */ uint16_t mss; /* Maximum segment size */ } fields; } tcp_seg_setup; }; /* Data descriptor */ struct e1000_data_desc { uint64_t buffer_addr; /* Address of the descriptor's buffer address */ union { uint32_t data; struct { uint16_t length; /* Data buffer length */ uint8_t typ_len_ext; uint8_t cmd; } flags; } lower; union { uint32_t data; struct { uint8_t status; /* Descriptor status */ uint8_t popts; /* Packet Options */ uint16_t special; } fields; } upper; }; union e1000_tx_udesc { struct e1000_tx_desc td; struct e1000_context_desc cd; struct e1000_data_desc dd; }; /* Tx checksum info for a packet. */ struct ck_info { int ck_valid; /* ck_info is valid */ uint8_t ck_start; /* start byte of cksum calcuation */ uint8_t ck_off; /* offset of cksum insertion */ uint16_t ck_len; /* length of cksum calc: 0 is to packet-end */ }; /* * Debug printf */ static int e82545_debug = 0; #define WPRINTF(msg,params...) PRINTLN("e82545: " msg, params) #define DPRINTF(msg,params...) if (e82545_debug) WPRINTF(msg, params) #define MIN(a,b) (((a)<(b))?(a):(b)) #define MAX(a,b) (((a)>(b))?(a):(b)) /* s/w representation of the RAL/RAH regs */ struct eth_uni { int eu_valid; int eu_addrsel; struct ether_addr eu_eth; }; struct e82545_softc { struct pci_devinst *esc_pi; struct vmctx *esc_ctx; struct mevent *esc_mevpitr; pthread_mutex_t esc_mtx; struct ether_addr esc_mac; net_backend_t *esc_be; /* General */ uint32_t esc_CTRL; /* x0000 device ctl */ uint32_t esc_FCAL; /* x0028 flow ctl addr lo */ uint32_t esc_FCAH; /* x002C flow ctl addr hi */ uint32_t esc_FCT; /* x0030 flow ctl type */ uint32_t esc_VET; /* x0038 VLAN eth type */ uint32_t esc_FCTTV; /* x0170 flow ctl tx timer */ uint32_t esc_LEDCTL; /* x0E00 LED control */ uint32_t esc_PBA; /* x1000 pkt buffer allocation */ /* Interrupt control */ int esc_irq_asserted; uint32_t esc_ICR; /* x00C0 cause read/clear */ uint32_t esc_ITR; /* x00C4 intr throttling */ uint32_t esc_ICS; /* x00C8 cause set */ uint32_t esc_IMS; /* x00D0 mask set/read */ uint32_t esc_IMC; /* x00D8 mask clear */ /* Transmit */ union e1000_tx_udesc *esc_txdesc; struct e1000_context_desc esc_txctx; pthread_t esc_tx_tid; pthread_cond_t esc_tx_cond; int esc_tx_enabled; int esc_tx_active; uint32_t esc_TXCW; /* x0178 transmit config */ uint32_t esc_TCTL; /* x0400 transmit ctl */ uint32_t esc_TIPG; /* x0410 inter-packet gap */ uint16_t esc_AIT; /* x0458 Adaptive Interframe Throttle */ uint64_t esc_tdba; /* verified 64-bit desc table addr */ uint32_t esc_TDBAL; /* x3800 desc table addr, low bits */ uint32_t esc_TDBAH; /* x3804 desc table addr, hi 32-bits */ uint32_t esc_TDLEN; /* x3808 # descriptors in bytes */ uint16_t esc_TDH; /* x3810 desc table head idx */ uint16_t esc_TDHr; /* internal read version of TDH */ uint16_t esc_TDT; /* x3818 desc table tail idx */ uint32_t esc_TIDV; /* x3820 intr delay */ uint32_t esc_TXDCTL; /* x3828 desc control */ uint32_t esc_TADV; /* x382C intr absolute delay */ /* L2 frame acceptance */ struct eth_uni esc_uni[16]; /* 16 x unicast MAC addresses */ uint32_t esc_fmcast[128]; /* Multicast filter bit-match */ uint32_t esc_fvlan[128]; /* VLAN 4096-bit filter */ /* Receive */ struct e1000_rx_desc *esc_rxdesc; pthread_cond_t esc_rx_cond; int esc_rx_enabled; int esc_rx_active; int esc_rx_loopback; uint32_t esc_RCTL; /* x0100 receive ctl */ uint32_t esc_FCRTL; /* x2160 flow cntl thresh, low */ uint32_t esc_FCRTH; /* x2168 flow cntl thresh, hi */ uint64_t esc_rdba; /* verified 64-bit desc table addr */ uint32_t esc_RDBAL; /* x2800 desc table addr, low bits */ uint32_t esc_RDBAH; /* x2804 desc table addr, hi 32-bits*/ uint32_t esc_RDLEN; /* x2808 #descriptors */ uint16_t esc_RDH; /* x2810 desc table head idx */ uint16_t esc_RDT; /* x2818 desc table tail idx */ uint32_t esc_RDTR; /* x2820 intr delay */ uint32_t esc_RXDCTL; /* x2828 desc control */ uint32_t esc_RADV; /* x282C intr absolute delay */ uint32_t esc_RSRPD; /* x2C00 recv small packet detect */ uint32_t esc_RXCSUM; /* x5000 receive cksum ctl */ /* IO Port register access */ uint32_t io_addr; /* Shadow copy of MDIC */ uint32_t mdi_control; /* Shadow copy of EECD */ uint32_t eeprom_control; /* Latest NVM in/out */ uint16_t nvm_data; uint16_t nvm_opaddr; /* stats */ uint32_t missed_pkt_count; /* dropped for no room in rx queue */ uint32_t pkt_rx_by_size[6]; uint32_t pkt_tx_by_size[6]; uint32_t good_pkt_rx_count; uint32_t bcast_pkt_rx_count; uint32_t mcast_pkt_rx_count; uint32_t good_pkt_tx_count; uint32_t bcast_pkt_tx_count; uint32_t mcast_pkt_tx_count; uint32_t oversize_rx_count; uint32_t tso_tx_count; uint64_t good_octets_rx; uint64_t good_octets_tx; uint64_t missed_octets; /* counts missed and oversized */ uint8_t nvm_bits:6; /* number of bits remaining in/out */ uint8_t nvm_mode:2; #define E82545_NVM_MODE_OPADDR 0x0 #define E82545_NVM_MODE_DATAIN 0x1 #define E82545_NVM_MODE_DATAOUT 0x2 /* EEPROM data */ uint16_t eeprom_data[E82545_NVM_EEPROM_SIZE]; }; static void e82545_reset(struct e82545_softc *sc, int dev); static void e82545_rx_enable(struct e82545_softc *sc); static void e82545_rx_disable(struct e82545_softc *sc); static void e82545_rx_callback(int fd, enum ev_type type, void *param); static void e82545_tx_start(struct e82545_softc *sc); static void e82545_tx_enable(struct e82545_softc *sc); static void e82545_tx_disable(struct e82545_softc *sc); static inline int e82545_size_stat_index(uint32_t size) { if (size <= 64) { return 0; } else if (size >= 1024) { return 5; } else { /* should be 1-4 */ return (ffs(size) - 6); } } static void e82545_init_eeprom(struct e82545_softc *sc) { uint16_t checksum, i; /* mac addr */ sc->eeprom_data[NVM_MAC_ADDR] = ((uint16_t)sc->esc_mac.octet[0]) | (((uint16_t)sc->esc_mac.octet[1]) << 8); sc->eeprom_data[NVM_MAC_ADDR+1] = ((uint16_t)sc->esc_mac.octet[2]) | (((uint16_t)sc->esc_mac.octet[3]) << 8); sc->eeprom_data[NVM_MAC_ADDR+2] = ((uint16_t)sc->esc_mac.octet[4]) | (((uint16_t)sc->esc_mac.octet[5]) << 8); /* pci ids */ sc->eeprom_data[NVM_SUB_DEV_ID] = E82545_SUBDEV_ID; sc->eeprom_data[NVM_SUB_VEN_ID] = E82545_VENDOR_ID_INTEL; sc->eeprom_data[NVM_DEV_ID] = E82545_DEV_ID_82545EM_COPPER; sc->eeprom_data[NVM_VEN_ID] = E82545_VENDOR_ID_INTEL; /* fill in the checksum */ checksum = 0; for (i = 0; i < NVM_CHECKSUM_REG; i++) { checksum += sc->eeprom_data[i]; } checksum = NVM_SUM - checksum; sc->eeprom_data[NVM_CHECKSUM_REG] = checksum; DPRINTF("eeprom checksum: 0x%x", checksum); } static void e82545_write_mdi(struct e82545_softc *sc, uint8_t reg_addr, uint8_t phy_addr, uint32_t data) { DPRINTF("Write mdi reg:0x%x phy:0x%x data: 0x%x", reg_addr, phy_addr, data); } static uint32_t e82545_read_mdi(struct e82545_softc *sc, uint8_t reg_addr, uint8_t phy_addr) { //DPRINTF("Read mdi reg:0x%x phy:0x%x", reg_addr, phy_addr); switch (reg_addr) { case PHY_STATUS: return (MII_SR_LINK_STATUS | MII_SR_AUTONEG_CAPS | MII_SR_AUTONEG_COMPLETE); case PHY_AUTONEG_ADV: return NWAY_AR_SELECTOR_FIELD; case PHY_LP_ABILITY: return 0; case PHY_1000T_STATUS: return (SR_1000T_LP_FD_CAPS | SR_1000T_REMOTE_RX_STATUS | SR_1000T_LOCAL_RX_STATUS); case PHY_ID1: return (M88E1011_I_PHY_ID >> 16) & 0xFFFF; case PHY_ID2: return (M88E1011_I_PHY_ID | E82545_REVISION_4) & 0xFFFF; default: DPRINTF("Unknown mdi read reg:0x%x phy:0x%x", reg_addr, phy_addr); return 0; } /* not reached */ } static void e82545_eecd_strobe(struct e82545_softc *sc) { /* Microwire state machine */ /* DPRINTF("eeprom state machine srtobe " "0x%x 0x%x 0x%x 0x%x", sc->nvm_mode, sc->nvm_bits, sc->nvm_opaddr, sc->nvm_data);*/ if (sc->nvm_bits == 0) { DPRINTF("eeprom state machine not expecting data! " "0x%x 0x%x 0x%x 0x%x", sc->nvm_mode, sc->nvm_bits, sc->nvm_opaddr, sc->nvm_data); return; } sc->nvm_bits--; if (sc->nvm_mode == E82545_NVM_MODE_DATAOUT) { /* shifting out */ if (sc->nvm_data & 0x8000) { sc->eeprom_control |= E1000_EECD_DO; } else { sc->eeprom_control &= ~E1000_EECD_DO; } sc->nvm_data <<= 1; if (sc->nvm_bits == 0) { /* read done, back to opcode mode. */ sc->nvm_opaddr = 0; sc->nvm_mode = E82545_NVM_MODE_OPADDR; sc->nvm_bits = E82545_NVM_OPADDR_BITS; } } else if (sc->nvm_mode == E82545_NVM_MODE_DATAIN) { /* shifting in */ sc->nvm_data <<= 1; if (sc->eeprom_control & E1000_EECD_DI) { sc->nvm_data |= 1; } if (sc->nvm_bits == 0) { /* eeprom write */ uint16_t op = sc->nvm_opaddr & E82545_NVM_OPCODE_MASK; uint16_t addr = sc->nvm_opaddr & E82545_NVM_ADDR_MASK; if (op != E82545_NVM_OPCODE_WRITE) { DPRINTF("Illegal eeprom write op 0x%x", sc->nvm_opaddr); } else if (addr >= E82545_NVM_EEPROM_SIZE) { DPRINTF("Illegal eeprom write addr 0x%x", sc->nvm_opaddr); } else { DPRINTF("eeprom write eeprom[0x%x] = 0x%x", addr, sc->nvm_data); sc->eeprom_data[addr] = sc->nvm_data; } /* back to opcode mode */ sc->nvm_opaddr = 0; sc->nvm_mode = E82545_NVM_MODE_OPADDR; sc->nvm_bits = E82545_NVM_OPADDR_BITS; } } else if (sc->nvm_mode == E82545_NVM_MODE_OPADDR) { sc->nvm_opaddr <<= 1; if (sc->eeprom_control & E1000_EECD_DI) { sc->nvm_opaddr |= 1; } if (sc->nvm_bits == 0) { uint16_t op = sc->nvm_opaddr & E82545_NVM_OPCODE_MASK; switch (op) { case E82545_NVM_OPCODE_EWEN: DPRINTF("eeprom write enable: 0x%x", sc->nvm_opaddr); /* back to opcode mode */ sc->nvm_opaddr = 0; sc->nvm_mode = E82545_NVM_MODE_OPADDR; sc->nvm_bits = E82545_NVM_OPADDR_BITS; break; case E82545_NVM_OPCODE_READ: { uint16_t addr = sc->nvm_opaddr & E82545_NVM_ADDR_MASK; sc->nvm_mode = E82545_NVM_MODE_DATAOUT; sc->nvm_bits = E82545_NVM_DATA_BITS; if (addr < E82545_NVM_EEPROM_SIZE) { sc->nvm_data = sc->eeprom_data[addr]; DPRINTF("eeprom read: eeprom[0x%x] = 0x%x", addr, sc->nvm_data); } else { DPRINTF("eeprom illegal read: 0x%x", sc->nvm_opaddr); sc->nvm_data = 0; } break; } case E82545_NVM_OPCODE_WRITE: sc->nvm_mode = E82545_NVM_MODE_DATAIN; sc->nvm_bits = E82545_NVM_DATA_BITS; sc->nvm_data = 0; break; default: DPRINTF("eeprom unknown op: 0x%x", sc->nvm_opaddr); /* back to opcode mode */ sc->nvm_opaddr = 0; sc->nvm_mode = E82545_NVM_MODE_OPADDR; sc->nvm_bits = E82545_NVM_OPADDR_BITS; } } } else { DPRINTF("eeprom state machine wrong state! " "0x%x 0x%x 0x%x 0x%x", sc->nvm_mode, sc->nvm_bits, sc->nvm_opaddr, sc->nvm_data); } } static void e82545_itr_callback(int fd, enum ev_type type, void *param) { uint32_t new; struct e82545_softc *sc = param; pthread_mutex_lock(&sc->esc_mtx); new = sc->esc_ICR & sc->esc_IMS; if (new && !sc->esc_irq_asserted) { DPRINTF("itr callback: lintr assert %x", new); sc->esc_irq_asserted = 1; pci_lintr_assert(sc->esc_pi); } else { mevent_delete(sc->esc_mevpitr); sc->esc_mevpitr = NULL; } pthread_mutex_unlock(&sc->esc_mtx); } static void e82545_icr_assert(struct e82545_softc *sc, uint32_t bits) { uint32_t new; DPRINTF("icr assert: 0x%x", bits); /* * An interrupt is only generated if bits are set that * aren't already in the ICR, these bits are unmasked, * and there isn't an interrupt already pending. */ new = bits & ~sc->esc_ICR & sc->esc_IMS; sc->esc_ICR |= bits; if (new == 0) { DPRINTF("icr assert: masked %x, ims %x", new, sc->esc_IMS); } else if (sc->esc_mevpitr != NULL) { DPRINTF("icr assert: throttled %x, ims %x", new, sc->esc_IMS); } else if (!sc->esc_irq_asserted) { DPRINTF("icr assert: lintr assert %x", new); sc->esc_irq_asserted = 1; pci_lintr_assert(sc->esc_pi); if (sc->esc_ITR != 0) { sc->esc_mevpitr = mevent_add( (sc->esc_ITR + 3905) / 3906, /* 256ns -> 1ms */ EVF_TIMER, e82545_itr_callback, sc); } } } static void e82545_ims_change(struct e82545_softc *sc, uint32_t bits) { uint32_t new; /* * Changing the mask may allow previously asserted * but masked interrupt requests to generate an interrupt. */ new = bits & sc->esc_ICR & ~sc->esc_IMS; sc->esc_IMS |= bits; if (new == 0) { DPRINTF("ims change: masked %x, ims %x", new, sc->esc_IMS); } else if (sc->esc_mevpitr != NULL) { DPRINTF("ims change: throttled %x, ims %x", new, sc->esc_IMS); } else if (!sc->esc_irq_asserted) { DPRINTF("ims change: lintr assert %x", new); sc->esc_irq_asserted = 1; pci_lintr_assert(sc->esc_pi); if (sc->esc_ITR != 0) { sc->esc_mevpitr = mevent_add( (sc->esc_ITR + 3905) / 3906, /* 256ns -> 1ms */ EVF_TIMER, e82545_itr_callback, sc); } } } static void e82545_icr_deassert(struct e82545_softc *sc, uint32_t bits) { DPRINTF("icr deassert: 0x%x", bits); sc->esc_ICR &= ~bits; /* * If there are no longer any interrupt sources and there * was an asserted interrupt, clear it */ if (sc->esc_irq_asserted && !(sc->esc_ICR & sc->esc_IMS)) { DPRINTF("icr deassert: lintr deassert %x", bits); pci_lintr_deassert(sc->esc_pi); sc->esc_irq_asserted = 0; } } static void e82545_intr_write(struct e82545_softc *sc, uint32_t offset, uint32_t value) { DPRINTF("intr_write: off %x, val %x", offset, value); switch (offset) { case E1000_ICR: e82545_icr_deassert(sc, value); break; case E1000_ITR: sc->esc_ITR = value; break; case E1000_ICS: sc->esc_ICS = value; /* not used: store for debug */ e82545_icr_assert(sc, value); break; case E1000_IMS: e82545_ims_change(sc, value); break; case E1000_IMC: sc->esc_IMC = value; /* for debug */ sc->esc_IMS &= ~value; // XXX clear interrupts if all ICR bits now masked // and interrupt was pending ? break; default: break; } } static uint32_t e82545_intr_read(struct e82545_softc *sc, uint32_t offset) { uint32_t retval; retval = 0; DPRINTF("intr_read: off %x", offset); switch (offset) { case E1000_ICR: retval = sc->esc_ICR; sc->esc_ICR = 0; e82545_icr_deassert(sc, ~0); break; case E1000_ITR: retval = sc->esc_ITR; break; case E1000_ICS: /* write-only register */ break; case E1000_IMS: retval = sc->esc_IMS; break; case E1000_IMC: /* write-only register */ break; default: break; } return (retval); } static void e82545_devctl(struct e82545_softc *sc, uint32_t val) { sc->esc_CTRL = val & ~E1000_CTRL_RST; if (val & E1000_CTRL_RST) { DPRINTF("e1k: s/w reset, ctl %x", val); e82545_reset(sc, 1); } /* XXX check for phy reset ? */ } static void e82545_rx_update_rdba(struct e82545_softc *sc) { /* XXX verify desc base/len within phys mem range */ sc->esc_rdba = (uint64_t)sc->esc_RDBAH << 32 | sc->esc_RDBAL; /* Cache host mapping of guest descriptor array */ sc->esc_rxdesc = paddr_guest2host(sc->esc_ctx, sc->esc_rdba, sc->esc_RDLEN); } static void e82545_rx_ctl(struct e82545_softc *sc, uint32_t val) { int on; on = ((val & E1000_RCTL_EN) == E1000_RCTL_EN); /* Save RCTL after stripping reserved bits 31:27,24,21,14,11:10,0 */ sc->esc_RCTL = val & ~0xF9204c01; DPRINTF("rx_ctl - %s RCTL %x, val %x", on ? "on" : "off", sc->esc_RCTL, val); /* state change requested */ if (on != sc->esc_rx_enabled) { if (on) { /* Catch disallowed/unimplemented settings */ //assert(!(val & E1000_RCTL_LBM_TCVR)); if (sc->esc_RCTL & E1000_RCTL_LBM_TCVR) { sc->esc_rx_loopback = 1; } else { sc->esc_rx_loopback = 0; } e82545_rx_update_rdba(sc); e82545_rx_enable(sc); } else { e82545_rx_disable(sc); sc->esc_rx_loopback = 0; sc->esc_rdba = 0; sc->esc_rxdesc = NULL; } } } static void e82545_tx_update_tdba(struct e82545_softc *sc) { /* XXX verify desc base/len within phys mem range */ sc->esc_tdba = (uint64_t)sc->esc_TDBAH << 32 | sc->esc_TDBAL; /* Cache host mapping of guest descriptor array */ sc->esc_txdesc = paddr_guest2host(sc->esc_ctx, sc->esc_tdba, sc->esc_TDLEN); } static void e82545_tx_ctl(struct e82545_softc *sc, uint32_t val) { int on; on = ((val & E1000_TCTL_EN) == E1000_TCTL_EN); /* ignore TCTL_EN settings that don't change state */ if (on == sc->esc_tx_enabled) return; if (on) { e82545_tx_update_tdba(sc); e82545_tx_enable(sc); } else { e82545_tx_disable(sc); sc->esc_tdba = 0; sc->esc_txdesc = NULL; } /* Save TCTL value after stripping reserved bits 31:25,23,2,0 */ sc->esc_TCTL = val & ~0xFE800005; } int e82545_bufsz(uint32_t rctl) { switch (rctl & (E1000_RCTL_BSEX | E1000_RCTL_SZ_256)) { case (E1000_RCTL_SZ_2048): return (2048); case (E1000_RCTL_SZ_1024): return (1024); case (E1000_RCTL_SZ_512): return (512); case (E1000_RCTL_SZ_256): return (256); case (E1000_RCTL_BSEX|E1000_RCTL_SZ_16384): return (16384); case (E1000_RCTL_BSEX|E1000_RCTL_SZ_8192): return (8192); case (E1000_RCTL_BSEX|E1000_RCTL_SZ_4096): return (4096); } return (256); /* Forbidden value. */ } /* XXX one packet at a time until this is debugged */ static void e82545_rx_callback(int fd, enum ev_type type, void *param) { struct e82545_softc *sc = param; struct e1000_rx_desc *rxd; struct iovec vec[64]; int left, len, lim, maxpktsz, maxpktdesc, bufsz, i, n, size; uint32_t cause = 0; uint16_t *tp, tag, head; pthread_mutex_lock(&sc->esc_mtx); DPRINTF("rx_run: head %x, tail %x", sc->esc_RDH, sc->esc_RDT); if (!sc->esc_rx_enabled || sc->esc_rx_loopback) { DPRINTF("rx disabled (!%d || %d) -- packet(s) dropped", sc->esc_rx_enabled, sc->esc_rx_loopback); while (netbe_rx_discard(sc->esc_be) > 0) { } goto done1; } bufsz = e82545_bufsz(sc->esc_RCTL); maxpktsz = (sc->esc_RCTL & E1000_RCTL_LPE) ? 16384 : 1522; maxpktdesc = (maxpktsz + bufsz - 1) / bufsz; size = sc->esc_RDLEN / 16; head = sc->esc_RDH; left = (size + sc->esc_RDT - head) % size; if (left < maxpktdesc) { DPRINTF("rx overflow (%d < %d) -- packet(s) dropped", left, maxpktdesc); while (netbe_rx_discard(sc->esc_be) > 0) { } goto done1; } sc->esc_rx_active = 1; pthread_mutex_unlock(&sc->esc_mtx); for (lim = size / 4; lim > 0 && left >= maxpktdesc; lim -= n) { /* Grab rx descriptor pointed to by the head pointer */ for (i = 0; i < maxpktdesc; i++) { rxd = &sc->esc_rxdesc[(head + i) % size]; vec[i].iov_base = paddr_guest2host(sc->esc_ctx, rxd->buffer_addr, bufsz); vec[i].iov_len = bufsz; } len = netbe_recv(sc->esc_be, vec, maxpktdesc); if (len <= 0) { DPRINTF("netbe_recv() returned %d", len); goto done; } /* * Adjust the packet length based on whether the CRC needs * to be stripped or if the packet is less than the minimum * eth packet size. */ if (len < ETHER_MIN_LEN - ETHER_CRC_LEN) len = ETHER_MIN_LEN - ETHER_CRC_LEN; if (!(sc->esc_RCTL & E1000_RCTL_SECRC)) len += ETHER_CRC_LEN; n = (len + bufsz - 1) / bufsz; DPRINTF("packet read %d bytes, %d segs, head %d", len, n, head); /* Apply VLAN filter. */ tp = (uint16_t *)vec[0].iov_base + 6; if ((sc->esc_RCTL & E1000_RCTL_VFE) && (ntohs(tp[0]) == sc->esc_VET)) { tag = ntohs(tp[1]) & 0x0fff; if ((sc->esc_fvlan[tag >> 5] & (1 << (tag & 0x1f))) != 0) { DPRINTF("known VLAN %d", tag); } else { DPRINTF("unknown VLAN %d", tag); n = 0; continue; } } /* Update all consumed descriptors. */ for (i = 0; i < n - 1; i++) { rxd = &sc->esc_rxdesc[(head + i) % size]; rxd->length = bufsz; rxd->csum = 0; rxd->errors = 0; rxd->special = 0; rxd->status = E1000_RXD_STAT_DD; } rxd = &sc->esc_rxdesc[(head + i) % size]; rxd->length = len % bufsz; rxd->csum = 0; rxd->errors = 0; rxd->special = 0; /* XXX signal no checksum for now */ rxd->status = E1000_RXD_STAT_PIF | E1000_RXD_STAT_IXSM | E1000_RXD_STAT_EOP | E1000_RXD_STAT_DD; /* Schedule receive interrupts. */ if (len <= sc->esc_RSRPD) { cause |= E1000_ICR_SRPD | E1000_ICR_RXT0; } else { /* XXX: RDRT and RADV timers should be here. */ cause |= E1000_ICR_RXT0; } head = (head + n) % size; left -= n; } done: pthread_mutex_lock(&sc->esc_mtx); sc->esc_rx_active = 0; if (sc->esc_rx_enabled == 0) pthread_cond_signal(&sc->esc_rx_cond); sc->esc_RDH = head; /* Respect E1000_RCTL_RDMTS */ left = (size + sc->esc_RDT - head) % size; if (left < (size >> (((sc->esc_RCTL >> 8) & 3) + 1))) cause |= E1000_ICR_RXDMT0; /* Assert all accumulated interrupts. */ if (cause != 0) e82545_icr_assert(sc, cause); done1: DPRINTF("rx_run done: head %x, tail %x", sc->esc_RDH, sc->esc_RDT); pthread_mutex_unlock(&sc->esc_mtx); } static uint16_t e82545_carry(uint32_t sum) { sum = (sum & 0xFFFF) + (sum >> 16); if (sum > 0xFFFF) sum -= 0xFFFF; return (sum); } static uint16_t e82545_buf_checksum(uint8_t *buf, int len) { int i; uint32_t sum = 0; /* Checksum all the pairs of bytes first... */ for (i = 0; i < (len & ~1U); i += 2) sum += *((u_int16_t *)(buf + i)); /* * If there's a single byte left over, checksum it, too. * Network byte order is big-endian, so the remaining byte is * the high byte. */ if (i < len) sum += htons(buf[i] << 8); return (e82545_carry(sum)); } static uint16_t e82545_iov_checksum(struct iovec *iov, int iovcnt, int off, int len) { int now, odd; uint32_t sum = 0, s; /* Skip completely unneeded vectors. */ while (iovcnt > 0 && iov->iov_len <= off && off > 0) { off -= iov->iov_len; iov++; iovcnt--; } /* Calculate checksum of requested range. */ odd = 0; while (len > 0 && iovcnt > 0) { now = MIN(len, iov->iov_len - off); s = e82545_buf_checksum(iov->iov_base + off, now); sum += odd ? (s << 8) : s; odd ^= (now & 1); len -= now; off = 0; iov++; iovcnt--; } return (e82545_carry(sum)); } /* * Return the transmit descriptor type. */ int e82545_txdesc_type(uint32_t lower) { int type; type = 0; if (lower & E1000_TXD_CMD_DEXT) type = lower & E1000_TXD_MASK; return (type); } static void e82545_transmit_checksum(struct iovec *iov, int iovcnt, struct ck_info *ck) { uint16_t cksum; int cklen; DPRINTF("tx cksum: iovcnt/s/off/len %d/%d/%d/%d", iovcnt, ck->ck_start, ck->ck_off, ck->ck_len); cklen = ck->ck_len ? ck->ck_len - ck->ck_start + 1 : INT_MAX; cksum = e82545_iov_checksum(iov, iovcnt, ck->ck_start, cklen); *(uint16_t *)((uint8_t *)iov[0].iov_base + ck->ck_off) = ~cksum; } static void e82545_transmit_backend(struct e82545_softc *sc, struct iovec *iov, int iovcnt) { if (sc->esc_be == NULL) return; (void) netbe_send(sc->esc_be, iov, iovcnt); } static void e82545_transmit_done(struct e82545_softc *sc, uint16_t head, uint16_t tail, uint16_t dsize, int *tdwb) { union e1000_tx_udesc *dsc; for ( ; head != tail; head = (head + 1) % dsize) { dsc = &sc->esc_txdesc[head]; if (dsc->td.lower.data & E1000_TXD_CMD_RS) { dsc->td.upper.data |= E1000_TXD_STAT_DD; *tdwb = 1; } } } static int e82545_transmit(struct e82545_softc *sc, uint16_t head, uint16_t tail, uint16_t dsize, uint16_t *rhead, int *tdwb) { uint8_t *hdr, *hdrp; struct iovec iovb[I82545_MAX_TXSEGS + 2]; struct iovec tiov[I82545_MAX_TXSEGS + 2]; struct e1000_context_desc *cd; struct ck_info ckinfo[2]; struct iovec *iov; union e1000_tx_udesc *dsc; int desc, dtype, len, ntype, iovcnt, tlen, tcp, tso; int mss, paylen, seg, tiovcnt, left, now, nleft, nnow, pv, pvoff; unsigned hdrlen, vlen; uint32_t tcpsum, tcpseq; uint16_t ipcs, tcpcs, ipid, ohead; ckinfo[0].ck_valid = ckinfo[1].ck_valid = 0; iovcnt = 0; tlen = 0; ntype = 0; tso = 0; ohead = head; /* iovb[0/1] may be used for writable copy of headers. */ iov = &iovb[2]; for (desc = 0; ; desc++, head = (head + 1) % dsize) { if (head == tail) { *rhead = head; return (0); } dsc = &sc->esc_txdesc[head]; dtype = e82545_txdesc_type(dsc->td.lower.data); if (desc == 0) { switch (dtype) { case E1000_TXD_TYP_C: DPRINTF("tx ctxt desc idx %d: %016jx " "%08x%08x", head, dsc->td.buffer_addr, dsc->td.upper.data, dsc->td.lower.data); /* Save context and return */ sc->esc_txctx = dsc->cd; goto done; case E1000_TXD_TYP_L: DPRINTF("tx legacy desc idx %d: %08x%08x", head, dsc->td.upper.data, dsc->td.lower.data); /* * legacy cksum start valid in first descriptor */ ntype = dtype; ckinfo[0].ck_start = dsc->td.upper.fields.css; break; case E1000_TXD_TYP_D: DPRINTF("tx data desc idx %d: %08x%08x", head, dsc->td.upper.data, dsc->td.lower.data); ntype = dtype; break; default: break; } } else { /* Descriptor type must be consistent */ assert(dtype == ntype); DPRINTF("tx next desc idx %d: %08x%08x", head, dsc->td.upper.data, dsc->td.lower.data); } len = (dtype == E1000_TXD_TYP_L) ? dsc->td.lower.flags.length : dsc->dd.lower.data & 0xFFFFF; if (len > 0) { /* Strip checksum supplied by guest. */ if ((dsc->td.lower.data & E1000_TXD_CMD_EOP) != 0 && (dsc->td.lower.data & E1000_TXD_CMD_IFCS) == 0) len -= 2; tlen += len; if (iovcnt < I82545_MAX_TXSEGS) { iov[iovcnt].iov_base = paddr_guest2host( sc->esc_ctx, dsc->td.buffer_addr, len); iov[iovcnt].iov_len = len; } iovcnt++; } /* * Pull out info that is valid in the final descriptor * and exit descriptor loop. */ if (dsc->td.lower.data & E1000_TXD_CMD_EOP) { if (dtype == E1000_TXD_TYP_L) { if (dsc->td.lower.data & E1000_TXD_CMD_IC) { ckinfo[0].ck_valid = 1; ckinfo[0].ck_off = dsc->td.lower.flags.cso; ckinfo[0].ck_len = 0; } } else { cd = &sc->esc_txctx; if (dsc->dd.lower.data & E1000_TXD_CMD_TSE) tso = 1; if (dsc->dd.upper.fields.popts & E1000_TXD_POPTS_IXSM) ckinfo[0].ck_valid = 1; if (dsc->dd.upper.fields.popts & E1000_TXD_POPTS_IXSM || tso) { ckinfo[0].ck_start = cd->lower_setup.ip_fields.ipcss; ckinfo[0].ck_off = cd->lower_setup.ip_fields.ipcso; ckinfo[0].ck_len = cd->lower_setup.ip_fields.ipcse; } if (dsc->dd.upper.fields.popts & E1000_TXD_POPTS_TXSM) ckinfo[1].ck_valid = 1; if (dsc->dd.upper.fields.popts & E1000_TXD_POPTS_TXSM || tso) { ckinfo[1].ck_start = cd->upper_setup.tcp_fields.tucss; ckinfo[1].ck_off = cd->upper_setup.tcp_fields.tucso; ckinfo[1].ck_len = cd->upper_setup.tcp_fields.tucse; } } break; } } if (iovcnt > I82545_MAX_TXSEGS) { WPRINTF("tx too many descriptors (%d > %d) -- dropped", iovcnt, I82545_MAX_TXSEGS); goto done; } hdrlen = vlen = 0; /* Estimate writable space for VLAN header insertion. */ if ((sc->esc_CTRL & E1000_CTRL_VME) && (dsc->td.lower.data & E1000_TXD_CMD_VLE)) { hdrlen = ETHER_ADDR_LEN*2; vlen = ETHER_VLAN_ENCAP_LEN; } if (!tso) { /* Estimate required writable space for checksums. */ if (ckinfo[0].ck_valid) hdrlen = MAX(hdrlen, ckinfo[0].ck_off + 2); if (ckinfo[1].ck_valid) hdrlen = MAX(hdrlen, ckinfo[1].ck_off + 2); /* Round up writable space to the first vector. */ if (hdrlen != 0 && iov[0].iov_len > hdrlen && iov[0].iov_len < hdrlen + 100) hdrlen = iov[0].iov_len; } else { /* In case of TSO header length provided by software. */ hdrlen = sc->esc_txctx.tcp_seg_setup.fields.hdr_len; /* * Cap the header length at 240 based on 7.2.4.5 of * the Intel 82576EB (Rev 2.63) datasheet. */ if (hdrlen > 240) { WPRINTF("TSO hdrlen too large: %d", hdrlen); goto done; } /* * If VLAN insertion is requested, ensure the header * at least holds the amount of data copied during * VLAN insertion below. * * XXX: Realistic packets will include a full Ethernet * header before the IP header at ckinfo[0].ck_start, * but this check is sufficient to prevent * out-of-bounds access below. */ if (vlen != 0 && hdrlen < ETHER_ADDR_LEN*2) { WPRINTF("TSO hdrlen too small for vlan insertion " "(%d vs %d) -- dropped", hdrlen, ETHER_ADDR_LEN*2); goto done; } /* * Ensure that the header length covers the used fields * in the IP and TCP headers as well as the IP and TCP * checksums. The following fields are accessed below: * * Header | Field | Offset | Length * -------+-------+--------+------- * IPv4 | len | 2 | 2 * IPv4 | ID | 4 | 2 * IPv6 | len | 4 | 2 * TCP | seq # | 4 | 4 * TCP | flags | 13 | 1 * UDP | len | 4 | 4 */ if (hdrlen < ckinfo[0].ck_start + 6 || hdrlen < ckinfo[0].ck_off + 2) { WPRINTF("TSO hdrlen too small for IP fields (%d) " "-- dropped", hdrlen); goto done; } if (sc->esc_txctx.cmd_and_length & E1000_TXD_CMD_TCP) { if (hdrlen < ckinfo[1].ck_start + 14 || (ckinfo[1].ck_valid && hdrlen < ckinfo[1].ck_off + 2)) { WPRINTF("TSO hdrlen too small for TCP fields " "(%d) -- dropped", hdrlen); goto done; } } else { if (hdrlen < ckinfo[1].ck_start + 8) { WPRINTF("TSO hdrlen too small for UDP fields " "(%d) -- dropped", hdrlen); goto done; } } } /* Allocate, fill and prepend writable header vector. */ if (hdrlen != 0) { hdr = __builtin_alloca(hdrlen + vlen); hdr += vlen; for (left = hdrlen, hdrp = hdr; left > 0; left -= now, hdrp += now) { now = MIN(left, iov->iov_len); memcpy(hdrp, iov->iov_base, now); iov->iov_base += now; iov->iov_len -= now; if (iov->iov_len == 0) { iov++; iovcnt--; } } iov--; iovcnt++; iov->iov_base = hdr; iov->iov_len = hdrlen; } else hdr = NULL; /* Insert VLAN tag. */ if (vlen != 0) { hdr -= ETHER_VLAN_ENCAP_LEN; memmove(hdr, hdr + ETHER_VLAN_ENCAP_LEN, ETHER_ADDR_LEN*2); hdrlen += ETHER_VLAN_ENCAP_LEN; hdr[ETHER_ADDR_LEN*2 + 0] = sc->esc_VET >> 8; hdr[ETHER_ADDR_LEN*2 + 1] = sc->esc_VET & 0xff; hdr[ETHER_ADDR_LEN*2 + 2] = dsc->td.upper.fields.special >> 8; hdr[ETHER_ADDR_LEN*2 + 3] = dsc->td.upper.fields.special & 0xff; iov->iov_base = hdr; iov->iov_len += ETHER_VLAN_ENCAP_LEN; /* Correct checksum offsets after VLAN tag insertion. */ ckinfo[0].ck_start += ETHER_VLAN_ENCAP_LEN; ckinfo[0].ck_off += ETHER_VLAN_ENCAP_LEN; if (ckinfo[0].ck_len != 0) ckinfo[0].ck_len += ETHER_VLAN_ENCAP_LEN; ckinfo[1].ck_start += ETHER_VLAN_ENCAP_LEN; ckinfo[1].ck_off += ETHER_VLAN_ENCAP_LEN; if (ckinfo[1].ck_len != 0) ckinfo[1].ck_len += ETHER_VLAN_ENCAP_LEN; } /* Simple non-TSO case. */ if (!tso) { /* Calculate checksums and transmit. */ if (ckinfo[0].ck_valid) e82545_transmit_checksum(iov, iovcnt, &ckinfo[0]); if (ckinfo[1].ck_valid) e82545_transmit_checksum(iov, iovcnt, &ckinfo[1]); e82545_transmit_backend(sc, iov, iovcnt); goto done; } /* Doing TSO. */ tcp = (sc->esc_txctx.cmd_and_length & E1000_TXD_CMD_TCP) != 0; mss = sc->esc_txctx.tcp_seg_setup.fields.mss; paylen = (sc->esc_txctx.cmd_and_length & 0x000fffff); DPRINTF("tx %s segmentation offload %d+%d/%d bytes %d iovs", tcp ? "TCP" : "UDP", hdrlen, paylen, mss, iovcnt); ipid = ntohs(*(uint16_t *)&hdr[ckinfo[0].ck_start + 4]); tcpseq = 0; if (tcp) tcpseq = ntohl(*(uint32_t *)&hdr[ckinfo[1].ck_start + 4]); ipcs = *(uint16_t *)&hdr[ckinfo[0].ck_off]; tcpcs = 0; if (ckinfo[1].ck_valid) /* Save partial pseudo-header checksum. */ tcpcs = *(uint16_t *)&hdr[ckinfo[1].ck_off]; pv = 1; pvoff = 0; for (seg = 0, left = paylen; left > 0; seg++, left -= now) { now = MIN(left, mss); /* Construct IOVs for the segment. */ /* Include whole original header. */ tiov[0].iov_base = hdr; tiov[0].iov_len = hdrlen; tiovcnt = 1; /* Include respective part of payload IOV. */ for (nleft = now; pv < iovcnt && nleft > 0; nleft -= nnow) { nnow = MIN(nleft, iov[pv].iov_len - pvoff); tiov[tiovcnt].iov_base = iov[pv].iov_base + pvoff; tiov[tiovcnt++].iov_len = nnow; if (pvoff + nnow == iov[pv].iov_len) { pv++; pvoff = 0; } else pvoff += nnow; } DPRINTF("tx segment %d %d+%d bytes %d iovs", seg, hdrlen, now, tiovcnt); /* Update IP header. */ if (sc->esc_txctx.cmd_and_length & E1000_TXD_CMD_IP) { /* IPv4 -- set length and ID */ *(uint16_t *)&hdr[ckinfo[0].ck_start + 2] = htons(hdrlen - ckinfo[0].ck_start + now); *(uint16_t *)&hdr[ckinfo[0].ck_start + 4] = htons(ipid + seg); } else { /* IPv6 -- set length */ *(uint16_t *)&hdr[ckinfo[0].ck_start + 4] = htons(hdrlen - ckinfo[0].ck_start - 40 + now); } /* Update pseudo-header checksum. */ tcpsum = tcpcs; tcpsum += htons(hdrlen - ckinfo[1].ck_start + now); /* Update TCP/UDP headers. */ if (tcp) { /* Update sequence number and FIN/PUSH flags. */ *(uint32_t *)&hdr[ckinfo[1].ck_start + 4] = htonl(tcpseq + paylen - left); if (now < left) { hdr[ckinfo[1].ck_start + 13] &= ~(TH_FIN | TH_PUSH); } } else { /* Update payload length. */ *(uint32_t *)&hdr[ckinfo[1].ck_start + 4] = hdrlen - ckinfo[1].ck_start + now; } /* Calculate checksums and transmit. */ if (ckinfo[0].ck_valid) { *(uint16_t *)&hdr[ckinfo[0].ck_off] = ipcs; e82545_transmit_checksum(tiov, tiovcnt, &ckinfo[0]); } if (ckinfo[1].ck_valid) { *(uint16_t *)&hdr[ckinfo[1].ck_off] = e82545_carry(tcpsum); e82545_transmit_checksum(tiov, tiovcnt, &ckinfo[1]); } e82545_transmit_backend(sc, tiov, tiovcnt); } done: head = (head + 1) % dsize; e82545_transmit_done(sc, ohead, head, dsize, tdwb); *rhead = head; return (desc + 1); } static void e82545_tx_run(struct e82545_softc *sc) { uint32_t cause; uint16_t head, rhead, tail, size; int lim, tdwb, sent; head = sc->esc_TDH; tail = sc->esc_TDT; size = sc->esc_TDLEN / 16; DPRINTF("tx_run: head %x, rhead %x, tail %x", sc->esc_TDH, sc->esc_TDHr, sc->esc_TDT); pthread_mutex_unlock(&sc->esc_mtx); rhead = head; tdwb = 0; for (lim = size / 4; sc->esc_tx_enabled && lim > 0; lim -= sent) { sent = e82545_transmit(sc, head, tail, size, &rhead, &tdwb); if (sent == 0) break; head = rhead; } pthread_mutex_lock(&sc->esc_mtx); sc->esc_TDH = head; sc->esc_TDHr = rhead; cause = 0; if (tdwb) cause |= E1000_ICR_TXDW; if (lim != size / 4 && sc->esc_TDH == sc->esc_TDT) cause |= E1000_ICR_TXQE; if (cause) e82545_icr_assert(sc, cause); DPRINTF("tx_run done: head %x, rhead %x, tail %x", sc->esc_TDH, sc->esc_TDHr, sc->esc_TDT); } static _Noreturn void * e82545_tx_thread(void *param) { struct e82545_softc *sc = param; pthread_mutex_lock(&sc->esc_mtx); for (;;) { while (!sc->esc_tx_enabled || sc->esc_TDHr == sc->esc_TDT) { if (sc->esc_tx_enabled && sc->esc_TDHr != sc->esc_TDT) break; sc->esc_tx_active = 0; if (sc->esc_tx_enabled == 0) pthread_cond_signal(&sc->esc_tx_cond); pthread_cond_wait(&sc->esc_tx_cond, &sc->esc_mtx); } sc->esc_tx_active = 1; /* Process some tx descriptors. Lock dropped inside. */ e82545_tx_run(sc); } } static void e82545_tx_start(struct e82545_softc *sc) { if (sc->esc_tx_active == 0) pthread_cond_signal(&sc->esc_tx_cond); } static void e82545_tx_enable(struct e82545_softc *sc) { sc->esc_tx_enabled = 1; } static void e82545_tx_disable(struct e82545_softc *sc) { sc->esc_tx_enabled = 0; while (sc->esc_tx_active) pthread_cond_wait(&sc->esc_tx_cond, &sc->esc_mtx); } static void e82545_rx_enable(struct e82545_softc *sc) { sc->esc_rx_enabled = 1; } static void e82545_rx_disable(struct e82545_softc *sc) { sc->esc_rx_enabled = 0; while (sc->esc_rx_active) pthread_cond_wait(&sc->esc_rx_cond, &sc->esc_mtx); } static void e82545_write_ra(struct e82545_softc *sc, int reg, uint32_t wval) { struct eth_uni *eu; int idx; idx = reg >> 1; assert(idx < 15); eu = &sc->esc_uni[idx]; if (reg & 0x1) { /* RAH */ eu->eu_valid = ((wval & E1000_RAH_AV) == E1000_RAH_AV); eu->eu_addrsel = (wval >> 16) & 0x3; eu->eu_eth.octet[5] = wval >> 8; eu->eu_eth.octet[4] = wval; } else { /* RAL */ eu->eu_eth.octet[3] = wval >> 24; eu->eu_eth.octet[2] = wval >> 16; eu->eu_eth.octet[1] = wval >> 8; eu->eu_eth.octet[0] = wval; } } static uint32_t e82545_read_ra(struct e82545_softc *sc, int reg) { struct eth_uni *eu; uint32_t retval; int idx; idx = reg >> 1; assert(idx < 15); eu = &sc->esc_uni[idx]; if (reg & 0x1) { /* RAH */ retval = (eu->eu_valid << 31) | (eu->eu_addrsel << 16) | (eu->eu_eth.octet[5] << 8) | eu->eu_eth.octet[4]; } else { /* RAL */ retval = (eu->eu_eth.octet[3] << 24) | (eu->eu_eth.octet[2] << 16) | (eu->eu_eth.octet[1] << 8) | eu->eu_eth.octet[0]; } return (retval); } static void e82545_write_register(struct e82545_softc *sc, uint32_t offset, uint32_t value) { int ridx; if (offset & 0x3) { DPRINTF("Unaligned register write offset:0x%x value:0x%x", offset, value); return; } DPRINTF("Register write: 0x%x value: 0x%x", offset, value); switch (offset) { case E1000_CTRL: case E1000_CTRL_DUP: e82545_devctl(sc, value); break; case E1000_FCAL: sc->esc_FCAL = value; break; case E1000_FCAH: sc->esc_FCAH = value & ~0xFFFF0000; break; case E1000_FCT: sc->esc_FCT = value & ~0xFFFF0000; break; case E1000_VET: sc->esc_VET = value & ~0xFFFF0000; break; case E1000_FCTTV: sc->esc_FCTTV = value & ~0xFFFF0000; break; case E1000_LEDCTL: sc->esc_LEDCTL = value & ~0x30303000; break; case E1000_PBA: sc->esc_PBA = value & 0x0000FF80; break; case E1000_ICR: case E1000_ITR: case E1000_ICS: case E1000_IMS: case E1000_IMC: e82545_intr_write(sc, offset, value); break; case E1000_RCTL: e82545_rx_ctl(sc, value); break; case E1000_FCRTL: sc->esc_FCRTL = value & ~0xFFFF0007; break; case E1000_FCRTH: sc->esc_FCRTH = value & ~0xFFFF0007; break; case E1000_RDBAL(0): sc->esc_RDBAL = value & ~0xF; if (sc->esc_rx_enabled) { /* Apparently legal: update cached address */ e82545_rx_update_rdba(sc); } break; case E1000_RDBAH(0): assert(!sc->esc_rx_enabled); sc->esc_RDBAH = value; break; case E1000_RDLEN(0): assert(!sc->esc_rx_enabled); sc->esc_RDLEN = value & ~0xFFF0007F; break; case E1000_RDH(0): /* XXX should only ever be zero ? Range check ? */ sc->esc_RDH = value; break; case E1000_RDT(0): /* XXX if this opens up the rx ring, do something ? */ sc->esc_RDT = value; break; case E1000_RDTR: /* ignore FPD bit 31 */ sc->esc_RDTR = value & ~0xFFFF0000; break; case E1000_RXDCTL(0): sc->esc_RXDCTL = value & ~0xFEC0C0C0; break; case E1000_RADV: sc->esc_RADV = value & ~0xFFFF0000; break; case E1000_RSRPD: sc->esc_RSRPD = value & ~0xFFFFF000; break; case E1000_RXCSUM: sc->esc_RXCSUM = value & ~0xFFFFF800; break; case E1000_TXCW: sc->esc_TXCW = value & ~0x3FFF0000; break; case E1000_TCTL: e82545_tx_ctl(sc, value); break; case E1000_TIPG: sc->esc_TIPG = value; break; case E1000_AIT: sc->esc_AIT = value; break; case E1000_TDBAL(0): sc->esc_TDBAL = value & ~0xF; if (sc->esc_tx_enabled) { /* Apparently legal */ e82545_tx_update_tdba(sc); } break; case E1000_TDBAH(0): //assert(!sc->esc_tx_enabled); sc->esc_TDBAH = value; break; case E1000_TDLEN(0): //assert(!sc->esc_tx_enabled); sc->esc_TDLEN = value & ~0xFFF0007F; break; case E1000_TDH(0): //assert(!sc->esc_tx_enabled); /* XXX should only ever be zero ? Range check ? */ sc->esc_TDHr = sc->esc_TDH = value; break; case E1000_TDT(0): /* XXX range check ? */ sc->esc_TDT = value; if (sc->esc_tx_enabled) e82545_tx_start(sc); break; case E1000_TIDV: sc->esc_TIDV = value & ~0xFFFF0000; break; case E1000_TXDCTL(0): //assert(!sc->esc_tx_enabled); sc->esc_TXDCTL = value & ~0xC0C0C0; break; case E1000_TADV: sc->esc_TADV = value & ~0xFFFF0000; break; case E1000_RAL(0) ... E1000_RAH(15): /* convert to u32 offset */ ridx = (offset - E1000_RAL(0)) >> 2; e82545_write_ra(sc, ridx, value); break; case E1000_MTA ... (E1000_MTA + (127*4)): sc->esc_fmcast[(offset - E1000_MTA) >> 2] = value; break; case E1000_VFTA ... (E1000_VFTA + (127*4)): sc->esc_fvlan[(offset - E1000_VFTA) >> 2] = value; break; case E1000_EECD: { //DPRINTF("EECD write 0x%x -> 0x%x", sc->eeprom_control, value); /* edge triggered low->high */ uint32_t eecd_strobe = ((sc->eeprom_control & E1000_EECD_SK) ? 0 : (value & E1000_EECD_SK)); uint32_t eecd_mask = (E1000_EECD_SK|E1000_EECD_CS| E1000_EECD_DI|E1000_EECD_REQ); sc->eeprom_control &= ~eecd_mask; sc->eeprom_control |= (value & eecd_mask); /* grant/revoke immediately */ if (value & E1000_EECD_REQ) { sc->eeprom_control |= E1000_EECD_GNT; } else { sc->eeprom_control &= ~E1000_EECD_GNT; } if (eecd_strobe && (sc->eeprom_control & E1000_EECD_CS)) { e82545_eecd_strobe(sc); } return; } case E1000_MDIC: { uint8_t reg_addr = (uint8_t)((value & E1000_MDIC_REG_MASK) >> E1000_MDIC_REG_SHIFT); uint8_t phy_addr = (uint8_t)((value & E1000_MDIC_PHY_MASK) >> E1000_MDIC_PHY_SHIFT); sc->mdi_control = (value & ~(E1000_MDIC_ERROR|E1000_MDIC_DEST)); if ((value & E1000_MDIC_READY) != 0) { DPRINTF("Incorrect MDIC ready bit: 0x%x", value); return; } switch (value & E82545_MDIC_OP_MASK) { case E1000_MDIC_OP_READ: sc->mdi_control &= ~E82545_MDIC_DATA_MASK; sc->mdi_control |= e82545_read_mdi(sc, reg_addr, phy_addr); break; case E1000_MDIC_OP_WRITE: e82545_write_mdi(sc, reg_addr, phy_addr, value & E82545_MDIC_DATA_MASK); break; default: DPRINTF("Unknown MDIC op: 0x%x", value); return; } /* TODO: barrier? */ sc->mdi_control |= E1000_MDIC_READY; if (value & E82545_MDIC_IE) { // TODO: generate interrupt } return; } case E1000_MANC: case E1000_STATUS: return; default: DPRINTF("Unknown write register: 0x%x value:%x", offset, value); return; } } static uint32_t e82545_read_register(struct e82545_softc *sc, uint32_t offset) { uint32_t retval; int ridx; if (offset & 0x3) { DPRINTF("Unaligned register read offset:0x%x", offset); return 0; } DPRINTF("Register read: 0x%x", offset); switch (offset) { case E1000_CTRL: retval = sc->esc_CTRL; break; case E1000_STATUS: retval = E1000_STATUS_FD | E1000_STATUS_LU | E1000_STATUS_SPEED_1000; break; case E1000_FCAL: retval = sc->esc_FCAL; break; case E1000_FCAH: retval = sc->esc_FCAH; break; case E1000_FCT: retval = sc->esc_FCT; break; case E1000_VET: retval = sc->esc_VET; break; case E1000_FCTTV: retval = sc->esc_FCTTV; break; case E1000_LEDCTL: retval = sc->esc_LEDCTL; break; case E1000_PBA: retval = sc->esc_PBA; break; case E1000_ICR: case E1000_ITR: case E1000_ICS: case E1000_IMS: case E1000_IMC: retval = e82545_intr_read(sc, offset); break; case E1000_RCTL: retval = sc->esc_RCTL; break; case E1000_FCRTL: retval = sc->esc_FCRTL; break; case E1000_FCRTH: retval = sc->esc_FCRTH; break; case E1000_RDBAL(0): retval = sc->esc_RDBAL; break; case E1000_RDBAH(0): retval = sc->esc_RDBAH; break; case E1000_RDLEN(0): retval = sc->esc_RDLEN; break; case E1000_RDH(0): retval = sc->esc_RDH; break; case E1000_RDT(0): retval = sc->esc_RDT; break; case E1000_RDTR: retval = sc->esc_RDTR; break; case E1000_RXDCTL(0): retval = sc->esc_RXDCTL; break; case E1000_RADV: retval = sc->esc_RADV; break; case E1000_RSRPD: retval = sc->esc_RSRPD; break; case E1000_RXCSUM: retval = sc->esc_RXCSUM; break; case E1000_TXCW: retval = sc->esc_TXCW; break; case E1000_TCTL: retval = sc->esc_TCTL; break; case E1000_TIPG: retval = sc->esc_TIPG; break; case E1000_AIT: retval = sc->esc_AIT; break; case E1000_TDBAL(0): retval = sc->esc_TDBAL; break; case E1000_TDBAH(0): retval = sc->esc_TDBAH; break; case E1000_TDLEN(0): retval = sc->esc_TDLEN; break; case E1000_TDH(0): retval = sc->esc_TDH; break; case E1000_TDT(0): retval = sc->esc_TDT; break; case E1000_TIDV: retval = sc->esc_TIDV; break; case E1000_TXDCTL(0): retval = sc->esc_TXDCTL; break; case E1000_TADV: retval = sc->esc_TADV; break; case E1000_RAL(0) ... E1000_RAH(15): /* convert to u32 offset */ ridx = (offset - E1000_RAL(0)) >> 2; retval = e82545_read_ra(sc, ridx); break; case E1000_MTA ... (E1000_MTA + (127*4)): retval = sc->esc_fmcast[(offset - E1000_MTA) >> 2]; break; case E1000_VFTA ... (E1000_VFTA + (127*4)): retval = sc->esc_fvlan[(offset - E1000_VFTA) >> 2]; break; case E1000_EECD: //DPRINTF("EECD read %x", sc->eeprom_control); retval = sc->eeprom_control; break; case E1000_MDIC: retval = sc->mdi_control; break; case E1000_MANC: retval = 0; break; /* stats that we emulate. */ case E1000_MPC: retval = sc->missed_pkt_count; break; case E1000_PRC64: retval = sc->pkt_rx_by_size[0]; break; case E1000_PRC127: retval = sc->pkt_rx_by_size[1]; break; case E1000_PRC255: retval = sc->pkt_rx_by_size[2]; break; case E1000_PRC511: retval = sc->pkt_rx_by_size[3]; break; case E1000_PRC1023: retval = sc->pkt_rx_by_size[4]; break; case E1000_PRC1522: retval = sc->pkt_rx_by_size[5]; break; case E1000_GPRC: retval = sc->good_pkt_rx_count; break; case E1000_BPRC: retval = sc->bcast_pkt_rx_count; break; case E1000_MPRC: retval = sc->mcast_pkt_rx_count; break; case E1000_GPTC: case E1000_TPT: retval = sc->good_pkt_tx_count; break; case E1000_GORCL: retval = (uint32_t)sc->good_octets_rx; break; case E1000_GORCH: retval = (uint32_t)(sc->good_octets_rx >> 32); break; case E1000_TOTL: case E1000_GOTCL: retval = (uint32_t)sc->good_octets_tx; break; case E1000_TOTH: case E1000_GOTCH: retval = (uint32_t)(sc->good_octets_tx >> 32); break; case E1000_ROC: retval = sc->oversize_rx_count; break; case E1000_TORL: retval = (uint32_t)(sc->good_octets_rx + sc->missed_octets); break; case E1000_TORH: retval = (uint32_t)((sc->good_octets_rx + sc->missed_octets) >> 32); break; case E1000_TPR: retval = sc->good_pkt_rx_count + sc->missed_pkt_count + sc->oversize_rx_count; break; case E1000_PTC64: retval = sc->pkt_tx_by_size[0]; break; case E1000_PTC127: retval = sc->pkt_tx_by_size[1]; break; case E1000_PTC255: retval = sc->pkt_tx_by_size[2]; break; case E1000_PTC511: retval = sc->pkt_tx_by_size[3]; break; case E1000_PTC1023: retval = sc->pkt_tx_by_size[4]; break; case E1000_PTC1522: retval = sc->pkt_tx_by_size[5]; break; case E1000_MPTC: retval = sc->mcast_pkt_tx_count; break; case E1000_BPTC: retval = sc->bcast_pkt_tx_count; break; case E1000_TSCTC: retval = sc->tso_tx_count; break; /* stats that are always 0. */ case E1000_CRCERRS: case E1000_ALGNERRC: case E1000_SYMERRS: case E1000_RXERRC: case E1000_SCC: case E1000_ECOL: case E1000_MCC: case E1000_LATECOL: case E1000_COLC: case E1000_DC: case E1000_TNCRS: case E1000_SEC: case E1000_CEXTERR: case E1000_RLEC: case E1000_XONRXC: case E1000_XONTXC: case E1000_XOFFRXC: case E1000_XOFFTXC: case E1000_FCRUC: case E1000_RNBC: case E1000_RUC: case E1000_RFC: case E1000_RJC: case E1000_MGTPRC: case E1000_MGTPDC: case E1000_MGTPTC: case E1000_TSCTFC: retval = 0; break; default: DPRINTF("Unknown read register: 0x%x", offset); retval = 0; break; } return (retval); } static void e82545_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx, uint64_t offset, int size, uint64_t value) { struct e82545_softc *sc; //DPRINTF("Write bar:%d offset:0x%lx value:0x%lx size:%d", baridx, offset, value, size); sc = pi->pi_arg; pthread_mutex_lock(&sc->esc_mtx); switch (baridx) { case E82545_BAR_IO: switch (offset) { case E82545_IOADDR: if (size != 4) { DPRINTF("Wrong io addr write sz:%d value:0x%lx", size, value); } else sc->io_addr = (uint32_t)value; break; case E82545_IODATA: if (size != 4) { DPRINTF("Wrong io data write size:%d value:0x%lx", size, value); } else if (sc->io_addr > E82545_IO_REGISTER_MAX) { DPRINTF("Non-register io write addr:0x%x value:0x%lx", sc->io_addr, value); } else e82545_write_register(sc, sc->io_addr, (uint32_t)value); break; default: DPRINTF("Unknown io bar write offset:0x%lx value:0x%lx size:%d", offset, value, size); break; } break; case E82545_BAR_REGISTER: if (size != 4) { DPRINTF("Wrong register write size:%d offset:0x%lx value:0x%lx", size, offset, value); } else e82545_write_register(sc, (uint32_t)offset, (uint32_t)value); break; default: DPRINTF("Unknown write bar:%d off:0x%lx val:0x%lx size:%d", baridx, offset, value, size); } pthread_mutex_unlock(&sc->esc_mtx); } static uint64_t e82545_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx, uint64_t offset, int size) { struct e82545_softc *sc; uint64_t retval; //DPRINTF("Read bar:%d offset:0x%lx size:%d", baridx, offset, size); sc = pi->pi_arg; retval = 0; pthread_mutex_lock(&sc->esc_mtx); switch (baridx) { case E82545_BAR_IO: switch (offset) { case E82545_IOADDR: if (size != 4) { DPRINTF("Wrong io addr read sz:%d", size); } else retval = sc->io_addr; break; case E82545_IODATA: if (size != 4) { DPRINTF("Wrong io data read sz:%d", size); } if (sc->io_addr > E82545_IO_REGISTER_MAX) { DPRINTF("Non-register io read addr:0x%x", sc->io_addr); } else retval = e82545_read_register(sc, sc->io_addr); break; default: DPRINTF("Unknown io bar read offset:0x%lx size:%d", offset, size); break; } break; case E82545_BAR_REGISTER: if (size != 4) { DPRINTF("Wrong register read size:%d offset:0x%lx", size, offset); } else retval = e82545_read_register(sc, (uint32_t)offset); break; default: DPRINTF("Unknown read bar:%d offset:0x%lx size:%d", baridx, offset, size); break; } pthread_mutex_unlock(&sc->esc_mtx); return (retval); } static void e82545_reset(struct e82545_softc *sc, int drvr) { int i; e82545_rx_disable(sc); e82545_tx_disable(sc); /* clear outstanding interrupts */ if (sc->esc_irq_asserted) pci_lintr_deassert(sc->esc_pi); /* misc */ if (!drvr) { sc->esc_FCAL = 0; sc->esc_FCAH = 0; sc->esc_FCT = 0; sc->esc_VET = 0; sc->esc_FCTTV = 0; } sc->esc_LEDCTL = 0x07061302; sc->esc_PBA = 0x00100030; /* start nvm in opcode mode. */ sc->nvm_opaddr = 0; sc->nvm_mode = E82545_NVM_MODE_OPADDR; sc->nvm_bits = E82545_NVM_OPADDR_BITS; sc->eeprom_control = E1000_EECD_PRES | E82545_EECD_FWE_EN; e82545_init_eeprom(sc); /* interrupt */ sc->esc_ICR = 0; sc->esc_ITR = 250; sc->esc_ICS = 0; sc->esc_IMS = 0; sc->esc_IMC = 0; /* L2 filters */ if (!drvr) { memset(sc->esc_fvlan, 0, sizeof(sc->esc_fvlan)); memset(sc->esc_fmcast, 0, sizeof(sc->esc_fmcast)); memset(sc->esc_uni, 0, sizeof(sc->esc_uni)); /* XXX not necessary on 82545 ?? */ sc->esc_uni[0].eu_valid = 1; memcpy(sc->esc_uni[0].eu_eth.octet, sc->esc_mac.octet, ETHER_ADDR_LEN); } else { /* Clear RAH valid bits */ for (i = 0; i < 16; i++) sc->esc_uni[i].eu_valid = 0; } /* receive */ if (!drvr) { sc->esc_RDBAL = 0; sc->esc_RDBAH = 0; } sc->esc_RCTL = 0; sc->esc_FCRTL = 0; sc->esc_FCRTH = 0; sc->esc_RDLEN = 0; sc->esc_RDH = 0; sc->esc_RDT = 0; sc->esc_RDTR = 0; sc->esc_RXDCTL = (1 << 24) | (1 << 16); /* default GRAN/WTHRESH */ sc->esc_RADV = 0; sc->esc_RXCSUM = 0; /* transmit */ if (!drvr) { sc->esc_TDBAL = 0; sc->esc_TDBAH = 0; sc->esc_TIPG = 0; sc->esc_AIT = 0; sc->esc_TIDV = 0; sc->esc_TADV = 0; } sc->esc_tdba = 0; sc->esc_txdesc = NULL; sc->esc_TXCW = 0; sc->esc_TCTL = 0; sc->esc_TDLEN = 0; sc->esc_TDT = 0; sc->esc_TDHr = sc->esc_TDH = 0; sc->esc_TXDCTL = 0; } static int e82545_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts) { char nstr[80]; struct e82545_softc *sc; char *devname; char *vtopts; int mac_provided; DPRINTF("Loading with options: %s", opts); /* Setup our softc */ sc = calloc(1, sizeof(*sc)); pi->pi_arg = sc; sc->esc_pi = pi; sc->esc_ctx = ctx; pthread_mutex_init(&sc->esc_mtx, NULL); pthread_cond_init(&sc->esc_rx_cond, NULL); pthread_cond_init(&sc->esc_tx_cond, NULL); pthread_create(&sc->esc_tx_tid, NULL, e82545_tx_thread, sc); snprintf(nstr, sizeof(nstr), "e82545-%d:%d tx", pi->pi_slot, pi->pi_func); pthread_set_name_np(sc->esc_tx_tid, nstr); pci_set_cfgdata16(pi, PCIR_DEVICE, E82545_DEV_ID_82545EM_COPPER); pci_set_cfgdata16(pi, PCIR_VENDOR, E82545_VENDOR_ID_INTEL); pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK); pci_set_cfgdata8(pi, PCIR_SUBCLASS, PCIS_NETWORK_ETHERNET); pci_set_cfgdata16(pi, PCIR_SUBDEV_0, E82545_SUBDEV_ID); pci_set_cfgdata16(pi, PCIR_SUBVEND_0, E82545_VENDOR_ID_INTEL); pci_set_cfgdata8(pi, PCIR_HDRTYPE, PCIM_HDRTYPE_NORMAL); pci_set_cfgdata8(pi, PCIR_INTPIN, 0x1); /* TODO: this card also supports msi, but the freebsd driver for it * does not, so I have not implemented it. */ pci_lintr_request(pi); pci_emul_alloc_bar(pi, E82545_BAR_REGISTER, PCIBAR_MEM32, E82545_BAR_REGISTER_LEN); pci_emul_alloc_bar(pi, E82545_BAR_FLASH, PCIBAR_MEM32, E82545_BAR_FLASH_LEN); pci_emul_alloc_bar(pi, E82545_BAR_IO, PCIBAR_IO, E82545_BAR_IO_LEN); /* * Attempt to open the net backend and read the MAC address * if specified. Copied from virtio-net, slightly modified. */ mac_provided = 0; sc->esc_be = NULL; if (opts != NULL) { - int err; + int err = 0; devname = vtopts = strdup(opts); (void) strsep(&vtopts, ","); - if (vtopts != NULL) { - err = net_parsemac(vtopts, sc->esc_mac.octet); - if (err != 0) { - free(devname); - return (err); + /* + * Parse the list of options in the form + * key1=value1,...,keyN=valueN. + */ + while (vtopts != NULL) { + char *value = vtopts; + char *key; + + key = strsep(&value, "="); + if (value == NULL) + break; + vtopts = value; + (void) strsep(&vtopts, ","); + + if (strcmp(key, "mac") == 0) { + err = net_parsemac(value, sc->esc_mac.octet); + if (err) + break; + mac_provided = 1; } - mac_provided = 1; + } + + if (err) { + free(devname); + return (err); } err = netbe_init(&sc->esc_be, devname, e82545_rx_callback, sc); free(devname); if (err) return (err); } if (!mac_provided) { net_genmac(pi, sc->esc_mac.octet); } netbe_rx_enable(sc->esc_be); /* H/w initiated reset */ e82545_reset(sc, 0); return (0); } struct pci_devemu pci_de_e82545 = { .pe_emu = "e1000", .pe_init = e82545_init, .pe_barwrite = e82545_write, .pe_barread = e82545_read }; PCI_EMUL_SET(pci_de_e82545); Index: stable/12/usr.sbin/bhyve/pci_virtio_net.c =================================================================== --- stable/12/usr.sbin/bhyve/pci_virtio_net.c (revision 358184) +++ stable/12/usr.sbin/bhyve/pci_virtio_net.c (revision 358185) @@ -1,603 +1,709 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include /* IFNAMSIZ */ #include #include #include #include #include #include #include #include #include #include #include #include #include "bhyverun.h" #include "debug.h" #include "pci_emul.h" #include "mevent.h" #include "virtio.h" #include "net_utils.h" #include "net_backends.h" #include "iov.h" #define VTNET_RINGSZ 1024 #define VTNET_MAXSEGS 256 #define VTNET_MAX_PKT_LEN (65536 + 64) #define VTNET_S_HOSTCAPS \ ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_STATUS | \ VIRTIO_F_NOTIFY_ON_EMPTY | VIRTIO_RING_F_INDIRECT_DESC) /* * PCI config-space "registers" */ struct virtio_net_config { uint8_t mac[6]; uint16_t status; } __packed; /* * Queue definitions. */ #define VTNET_RXQ 0 #define VTNET_TXQ 1 #define VTNET_CTLQ 2 /* NB: not yet supported */ #define VTNET_MAXQ 3 /* * Debug printf */ static int pci_vtnet_debug; #define DPRINTF(params) if (pci_vtnet_debug) PRINTLN params #define WPRINTF(params) PRINTLN params /* * Per-device softc */ struct pci_vtnet_softc { struct virtio_softc vsc_vs; struct vqueue_info vsc_queues[VTNET_MAXQ - 1]; pthread_mutex_t vsc_mtx; net_backend_t *vsc_be; int resetting; /* protected by tx_mtx */ uint64_t vsc_features; /* negotiated features */ pthread_mutex_t rx_mtx; int rx_merge; /* merged rx bufs in use */ pthread_t tx_tid; pthread_mutex_t tx_mtx; pthread_cond_t tx_cond; int tx_in_progress; + size_t vhdrlen; + size_t be_vhdrlen; + struct virtio_net_config vsc_config; struct virtio_consts vsc_consts; }; static void pci_vtnet_reset(void *); /* static void pci_vtnet_notify(void *, struct vqueue_info *); */ static int pci_vtnet_cfgread(void *, int, int, uint32_t *); static int pci_vtnet_cfgwrite(void *, int, int, uint32_t); static void pci_vtnet_neg_features(void *, uint64_t); static struct virtio_consts vtnet_vi_consts = { "vtnet", /* our name */ VTNET_MAXQ - 1, /* we currently support 2 virtqueues */ sizeof(struct virtio_net_config), /* config reg size */ pci_vtnet_reset, /* reset */ NULL, /* device-wide qnotify -- not used */ pci_vtnet_cfgread, /* read PCI config */ pci_vtnet_cfgwrite, /* write PCI config */ pci_vtnet_neg_features, /* apply negotiated features */ VTNET_S_HOSTCAPS, /* our capabilities */ }; static void pci_vtnet_reset(void *vsc) { struct pci_vtnet_softc *sc = vsc; DPRINTF(("vtnet: device reset requested !")); /* Acquire the RX lock to block RX processing. */ pthread_mutex_lock(&sc->rx_mtx); /* * Make sure receive operation is disabled at least until we * re-negotiate the features, since receive operation depends * on the value of sc->rx_merge and the header length, which * are both set in pci_vtnet_neg_features(). * Receive operation will be enabled again once the guest adds * the first receive buffers and kicks us. */ netbe_rx_disable(sc->vsc_be); /* Set sc->resetting and give a chance to the TX thread to stop. */ pthread_mutex_lock(&sc->tx_mtx); sc->resetting = 1; while (sc->tx_in_progress) { pthread_mutex_unlock(&sc->tx_mtx); usleep(10000); pthread_mutex_lock(&sc->tx_mtx); } /* * Now reset rings, MSI-X vectors, and negotiated capabilities. * Do that with the TX lock held, since we need to reset * sc->resetting. */ vi_reset_dev(&sc->vsc_vs); sc->resetting = 0; pthread_mutex_unlock(&sc->tx_mtx); pthread_mutex_unlock(&sc->rx_mtx); } +static __inline struct iovec * +iov_trim_hdr(struct iovec *iov, int *iovcnt, unsigned int hlen) +{ + struct iovec *riov; + + if (iov[0].iov_len < hlen) { + /* + * Not enough header space in the first fragment. + * That's not ok for us. + */ + return NULL; + } + + iov[0].iov_len -= hlen; + if (iov[0].iov_len == 0) { + *iovcnt -= 1; + if (*iovcnt == 0) { + /* + * Only space for the header. That's not + * enough for us. + */ + return NULL; + } + riov = &iov[1]; + } else { + iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + hlen); + riov = &iov[0]; + } + + return (riov); +} + struct virtio_mrg_rxbuf_info { uint16_t idx; uint16_t pad; uint32_t len; }; static void pci_vtnet_rx(struct pci_vtnet_softc *sc) { + int prepend_hdr_len = sc->vhdrlen - sc->be_vhdrlen; struct virtio_mrg_rxbuf_info info[VTNET_MAXSEGS]; struct iovec iov[VTNET_MAXSEGS + 1]; struct vqueue_info *vq; - uint32_t cur_iov_bytes; - struct iovec *cur_iov; - uint16_t cur_iov_len; + uint32_t riov_bytes; + struct iovec *riov; + int riov_len; uint32_t ulen; int n_chains; int len; vq = &sc->vsc_queues[VTNET_RXQ]; for (;;) { + struct virtio_net_rxhdr *hdr; + /* * Get a descriptor chain to store the next ingress * packet. In case of mergeable rx buffers, get as * many chains as necessary in order to make room * for a maximum sized LRO packet. */ - cur_iov_bytes = 0; - cur_iov_len = 0; - cur_iov = iov; + riov_bytes = 0; + riov_len = 0; + riov = iov; n_chains = 0; do { - int n = vq_getchain(vq, &info[n_chains].idx, cur_iov, - VTNET_MAXSEGS - cur_iov_len, NULL); + int n = vq_getchain(vq, &info[n_chains].idx, riov, + VTNET_MAXSEGS - riov_len, NULL); if (n == 0) { /* * No rx buffers. Enable RX kicks and double * check. */ vq_kick_enable(vq); if (!vq_has_descs(vq)) { /* * Still no buffers. Return the unused * chains (if any), interrupt if needed * (including for NOTIFY_ON_EMPTY), and * disable the backend until the next * kick. */ vq_retchains(vq, n_chains); vq_endchains(vq, /*used_all_avail=*/1); netbe_rx_disable(sc->vsc_be); return; } /* More rx buffers found, so keep going. */ vq_kick_disable(vq); continue; } - assert(n >= 1 && cur_iov_len + n <= VTNET_MAXSEGS); - cur_iov_len += n; + assert(n >= 1 && riov_len + n <= VTNET_MAXSEGS); + riov_len += n; if (!sc->rx_merge) { n_chains = 1; break; } - info[n_chains].len = (uint32_t)count_iov(cur_iov, n); - cur_iov_bytes += info[n_chains].len; - cur_iov += n; + info[n_chains].len = (uint32_t)count_iov(riov, n); + riov_bytes += info[n_chains].len; + riov += n; n_chains++; - } while (cur_iov_bytes < VTNET_MAX_PKT_LEN && - cur_iov_len < VTNET_MAXSEGS); + } while (riov_bytes < VTNET_MAX_PKT_LEN && + riov_len < VTNET_MAXSEGS); - len = netbe_recv(sc->vsc_be, iov, cur_iov_len); + riov = iov; + hdr = riov[0].iov_base; + if (prepend_hdr_len > 0) { + /* + * The frontend uses a virtio-net header, but the + * backend does not. We need to prepend a zeroed + * header. + */ + riov = iov_trim_hdr(riov, &riov_len, prepend_hdr_len); + if (riov == NULL) { + /* + * The first collected chain is nonsensical, + * as it is not even enough to store the + * virtio-net header. Just drop it. + */ + vq_relchain(vq, info[0].idx, 0); + vq_retchains(vq, n_chains - 1); + continue; + } + memset(hdr, 0, prepend_hdr_len); + } + len = netbe_recv(sc->vsc_be, riov, riov_len); + if (len <= 0) { /* * No more packets (len == 0), or backend errored * (err < 0). Return unused available buffers * and stop. */ vq_retchains(vq, n_chains); /* Interrupt if needed/appropriate and stop. */ vq_endchains(vq, /*used_all_avail=*/0); return; } - ulen = (uint32_t)len; /* avoid too many casts below */ + ulen = (uint32_t)(len + prepend_hdr_len); - /* Publish the used buffers to the guest. */ + /* + * Publish the used buffers to the guest, reporting the + * number of bytes that we wrote. + */ if (!sc->rx_merge) { vq_relchain(vq, info[0].idx, ulen); } else { - struct virtio_net_rxhdr *hdr = iov[0].iov_base; uint32_t iolen; int i = 0; - assert(iov[0].iov_len >= sizeof(*hdr)); - do { iolen = info[i].len; if (iolen > ulen) { iolen = ulen; } vq_relchain_prepare(vq, info[i].idx, iolen); ulen -= iolen; i++; assert(i <= n_chains); } while (ulen > 0); hdr->vrh_bufs = i; vq_relchain_publish(vq); vq_retchains(vq, n_chains - i); } } } /* * Called when there is read activity on the backend file descriptor. * Each buffer posted by the guest is assumed to be able to contain * an entire ethernet frame + rx header. */ static void pci_vtnet_rx_callback(int fd, enum ev_type type, void *param) { struct pci_vtnet_softc *sc = param; pthread_mutex_lock(&sc->rx_mtx); pci_vtnet_rx(sc); pthread_mutex_unlock(&sc->rx_mtx); } /* Called on RX kick. */ static void pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq) { struct pci_vtnet_softc *sc = vsc; /* * A qnotify means that the rx process can now begin. */ pthread_mutex_lock(&sc->rx_mtx); vq_kick_disable(vq); netbe_rx_enable(sc->vsc_be); pthread_mutex_unlock(&sc->rx_mtx); } /* TX virtqueue processing, called by the TX thread. */ static void pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq) { struct iovec iov[VTNET_MAXSEGS + 1]; + struct iovec *siov = iov; uint16_t idx; ssize_t len; int n; /* * Obtain chain of descriptors. The first descriptor also * contains the virtio-net header. */ n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL); assert(n >= 1 && n <= VTNET_MAXSEGS); - len = netbe_send(sc->vsc_be, iov, n); + if (sc->vhdrlen != sc->be_vhdrlen) { + /* + * The frontend uses a virtio-net header, but the backend + * does not. We simply strip the header and ignore it, as + * it should be zero-filled. + */ + siov = iov_trim_hdr(siov, &n, sc->vhdrlen); + } - /* chain is processed, release it and set len */ - vq_relchain(vq, idx, len > 0 ? len : 0); + if (siov == NULL) { + /* The chain is nonsensical. Just drop it. */ + len = 0; + } else { + len = netbe_send(sc->vsc_be, siov, n); + if (len < 0) { + /* + * If send failed, report that 0 bytes + * were read. + */ + len = 0; + } + } + + /* + * Return the processed chain to the guest, reporting + * the number of bytes that we read. + */ + vq_relchain(vq, idx, len); } /* Called on TX kick. */ static void pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq) { struct pci_vtnet_softc *sc = vsc; /* * Any ring entries to process? */ if (!vq_has_descs(vq)) return; /* Signal the tx thread for processing */ pthread_mutex_lock(&sc->tx_mtx); vq_kick_disable(vq); if (sc->tx_in_progress == 0) pthread_cond_signal(&sc->tx_cond); pthread_mutex_unlock(&sc->tx_mtx); } /* * Thread which will handle processing of TX desc */ static void * pci_vtnet_tx_thread(void *param) { struct pci_vtnet_softc *sc = param; struct vqueue_info *vq; int error; vq = &sc->vsc_queues[VTNET_TXQ]; /* * Let us wait till the tx queue pointers get initialised & * first tx signaled */ pthread_mutex_lock(&sc->tx_mtx); error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx); assert(error == 0); for (;;) { /* note - tx mutex is locked here */ while (sc->resetting || !vq_has_descs(vq)) { vq_kick_enable(vq); if (!sc->resetting && vq_has_descs(vq)) break; sc->tx_in_progress = 0; error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx); assert(error == 0); } vq_kick_disable(vq); sc->tx_in_progress = 1; pthread_mutex_unlock(&sc->tx_mtx); do { /* * Run through entries, placing them into * iovecs and sending when an end-of-packet * is found */ pci_vtnet_proctx(sc, vq); } while (vq_has_descs(vq)); /* * Generate an interrupt if needed. */ vq_endchains(vq, /*used_all_avail=*/1); pthread_mutex_lock(&sc->tx_mtx); } } #ifdef notyet static void pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq) { DPRINTF(("vtnet: control qnotify!")); } #endif static int pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts) { struct pci_vtnet_softc *sc; char tname[MAXCOMLEN + 1]; int mac_provided; /* * Allocate data structures for further virtio initializations. * sc also contains a copy of vtnet_vi_consts, since capabilities * change depending on the backend. */ sc = calloc(1, sizeof(struct pci_vtnet_softc)); sc->vsc_consts = vtnet_vi_consts; pthread_mutex_init(&sc->vsc_mtx, NULL); sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ; sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq; sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ; sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq; #ifdef notyet sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ; sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq; #endif /* * Attempt to open the backend device and read the MAC address * if specified. */ mac_provided = 0; if (opts != NULL) { char *devname; char *vtopts; - int err; + int err = 0; + /* Get the device name. */ devname = vtopts = strdup(opts); (void) strsep(&vtopts, ","); - if (vtopts != NULL) { - err = net_parsemac(vtopts, sc->vsc_config.mac); - if (err != 0) { - free(devname); - free(sc); - return (err); + /* + * Parse the list of options in the form + * key1=value1,...,keyN=valueN. + */ + while (vtopts != NULL) { + char *value = vtopts; + char *key; + + key = strsep(&value, "="); + if (value == NULL) + break; + vtopts = value; + (void) strsep(&vtopts, ","); + + if (strcmp(key, "mac") == 0) { + err = net_parsemac(value, sc->vsc_config.mac); + if (err) + break; + mac_provided = 1; } - mac_provided = 1; } + if (err) { + free(devname); + free(sc); + return (err); + } + err = netbe_init(&sc->vsc_be, devname, pci_vtnet_rx_callback, sc); free(devname); if (err) { free(sc); return (err); } sc->vsc_consts.vc_hv_caps |= netbe_get_cap(sc->vsc_be); } if (!mac_provided) { net_genmac(pi, sc->vsc_config.mac); } /* initialize config space */ pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET); pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR); pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK); pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_NET); pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR); /* Link is up if we managed to open backend device. */ sc->vsc_config.status = (opts == NULL || sc->vsc_be); vi_softc_linkup(&sc->vsc_vs, &sc->vsc_consts, sc, pi, sc->vsc_queues); sc->vsc_vs.vs_mtx = &sc->vsc_mtx; /* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */ if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix())) { free(sc); return (1); } /* use BAR 0 to map config regs in IO space */ vi_set_io_bar(&sc->vsc_vs, 0); sc->resetting = 0; sc->rx_merge = 0; + sc->vhdrlen = sizeof(struct virtio_net_rxhdr) - 2; pthread_mutex_init(&sc->rx_mtx, NULL); /* * Initialize tx semaphore & spawn TX processing thread. * As of now, only one thread for TX desc processing is * spawned. */ sc->tx_in_progress = 0; pthread_mutex_init(&sc->tx_mtx, NULL); pthread_cond_init(&sc->tx_cond, NULL); pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc); snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot, pi->pi_func); pthread_set_name_np(sc->tx_tid, tname); return (0); } static int pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value) { struct pci_vtnet_softc *sc = vsc; void *ptr; if (offset < (int)sizeof(sc->vsc_config.mac)) { assert(offset + size <= (int)sizeof(sc->vsc_config.mac)); /* * The driver is allowed to change the MAC address */ ptr = &sc->vsc_config.mac[offset]; memcpy(ptr, &value, size); } else { /* silently ignore other writes */ DPRINTF(("vtnet: write to readonly reg %d", offset)); } return (0); } static int pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval) { struct pci_vtnet_softc *sc = vsc; void *ptr; ptr = (uint8_t *)&sc->vsc_config + offset; memcpy(retval, ptr, size); return (0); } static void pci_vtnet_neg_features(void *vsc, uint64_t negotiated_features) { struct pci_vtnet_softc *sc = vsc; - unsigned int rx_vhdrlen; sc->vsc_features = negotiated_features; if (negotiated_features & VIRTIO_NET_F_MRG_RXBUF) { - rx_vhdrlen = sizeof(struct virtio_net_rxhdr); + sc->vhdrlen = sizeof(struct virtio_net_rxhdr); sc->rx_merge = 1; } else { /* * Without mergeable rx buffers, virtio-net header is 2 * bytes shorter than sizeof(struct virtio_net_rxhdr). */ - rx_vhdrlen = sizeof(struct virtio_net_rxhdr) - 2; + sc->vhdrlen = sizeof(struct virtio_net_rxhdr) - 2; sc->rx_merge = 0; } /* Tell the backend to enable some capabilities it has advertised. */ - netbe_set_cap(sc->vsc_be, negotiated_features, rx_vhdrlen); + netbe_set_cap(sc->vsc_be, negotiated_features, sc->vhdrlen); + sc->be_vhdrlen = netbe_get_vnet_hdr_len(sc->vsc_be); + assert(sc->be_vhdrlen == 0 || sc->be_vhdrlen == sc->vhdrlen); } static struct pci_devemu pci_de_vnet = { .pe_emu = "virtio-net", .pe_init = pci_vtnet_init, .pe_barwrite = vi_pci_write, .pe_barread = vi_pci_read }; PCI_EMUL_SET(pci_de_vnet);