/*
*---------------------------------------------------------------------------
*
* vmxnet3_tx_prepare_offload --
*
* Build the offload context of a msg.
*
* Results:
* 0 if everything went well.
* +n if n bytes need to be pulled up.
* -1 in case of error (not used).
*
* Side effects:
* None.
*
*---------------------------------------------------------------------------
*/
static int
vmxnet3_tx_prepare_offload(vmxnet3_softc_t *dp,
vmxnet3_offload_t *ol,
mblk_t *mp)
{
int ret = 0;
uint32_t start, stuff, value, flags, lso_flag, mss;
ol->om = VMXNET3_OM_NONE;
ol->hlen = 0;
ol->msscof = 0;
hcksum_retrieve(mp, NULL, NULL, &start, &stuff, NULL, &value, &flags);
mac_lso_get(mp, &mss, &lso_flag);
if (flags || lso_flag) {
struct ether_vlan_header *eth = (void *) mp->b_rptr;
uint8_t ethLen;
if (eth->ether_tpid == htons(ETHERTYPE_VLAN)) {
ethLen = sizeof(struct ether_vlan_header);
} else {
ethLen = sizeof(struct ether_header);
}
VMXNET3_DEBUG(dp, 4, "flags=0x%x, ethLen=%u, start=%u, stuff=%u, value=%u\n",
flags, ethLen, start, stuff, value);
if (lso_flag & HW_LSO) {
mblk_t *mblk = mp;
uint8_t *ip, *tcp;
uint8_t ipLen, tcpLen;
/*
* Copy e1000g's behavior:
* - Do not assume all the headers are in the same mblk.
* - Assume each header is always within one mblk.
* - Assume the ethernet header is in the first mblk.
*/
ip = mblk->b_rptr + ethLen;
if (ip >= mblk->b_wptr) {
mblk = mblk->b_cont;
ip = mblk->b_rptr;
}
ipLen = IPH_HDR_LENGTH((ipha_t *) ip);
tcp = ip + ipLen;
if (tcp >= mblk->b_wptr) {
mblk = mblk->b_cont;
tcp = mblk->b_rptr;
}
tcpLen = TCP_HDR_LENGTH((tcph_t *) tcp);
if (tcp + tcpLen > mblk->b_wptr) { // careful, '>' instead of '>=' here
mblk = mblk->b_cont;
}
ol->om = VMXNET3_OM_TSO;
ol->hlen = ethLen + ipLen + tcpLen;
ol->msscof = mss;
if (mblk != mp) {
ret = ol->hlen;
}
} else if (flags & HCK_PARTIALCKSUM) {
ol->om = VMXNET3_OM_CSUM;
ol->hlen = start + ethLen;
ol->msscof = stuff + ethLen;
}
}
return ret;
}
static enum proto_parse_status tcp_parse(struct parser *parser, struct proto_info *parent, unsigned way, uint8_t const *packet, size_t cap_len, size_t wire_len, struct timeval const *now, size_t tot_cap_len, uint8_t const *tot_packet)
{
struct mux_parser *mux_parser = DOWNCAST(parser, parser, mux_parser);
struct tcp_hdr const *tcphdr = (struct tcp_hdr *)packet;
// Sanity checks
if (wire_len < sizeof(*tcphdr)) {
SLOG(LOG_DEBUG, "Bogus TCP packet: too short (%zu < %zu)", wire_len, sizeof(*tcphdr));
return PROTO_PARSE_ERR;
}
if (cap_len < sizeof(*tcphdr)) return PROTO_TOO_SHORT;
size_t tcphdr_len = TCP_HDR_LENGTH(tcphdr);
if (tcphdr_len < sizeof(*tcphdr)) {
SLOG(LOG_DEBUG, "Bogus TCP packet: header size too smal (%zu < %zu)", tcphdr_len, sizeof(*tcphdr));
return PROTO_PARSE_ERR;
}
if (tcphdr_len > wire_len) {
SLOG(LOG_DEBUG, "Bogus TCP packet: wrong length %zu > %zu", tcphdr_len, wire_len);
return PROTO_PARSE_ERR;
}
if (tcphdr_len > cap_len) return PROTO_TOO_SHORT;
// TODO: move this below call to tcp_proto_info_ctor() and use info instead of reading tcphdr directly
uint16_t const sport = READ_U16N(&tcphdr->src);
uint16_t const dport = READ_U16N(&tcphdr->dst);
bool const syn = !!(READ_U8(&tcphdr->flags) & TCP_SYN_MASK);
bool const fin = !!(READ_U8(&tcphdr->flags) & TCP_FIN_MASK);
bool const ack = !!(READ_U8(&tcphdr->flags) & TCP_ACK_MASK);
bool const rst = !!(READ_U8(&tcphdr->flags) & TCP_RST_MASK);
bool const urg = !!(READ_U8(&tcphdr->flags) & TCP_URG_MASK);
bool const psh = !!(READ_U8(&tcphdr->flags) & TCP_PSH_MASK);
SLOG(LOG_DEBUG, "New TCP packet of %zu bytes (%zu captured), %zu payload, ports %"PRIu16" -> %"PRIu16" Flags: %s%s%s%s%s%s, Seq:%"PRIu32", Ack:%"PRIu32,
wire_len, cap_len, wire_len - tcphdr_len, sport, dport,
syn ? "Syn":"", fin ? "Fin":"", ack ? "Ack":"", rst ? "Rst":"", urg ? "Urg":"", psh ? "Psh":"",
READ_U32N(&tcphdr->seq_num), READ_U32N(&tcphdr->ack_seq));
// Parse
struct tcp_proto_info info;
tcp_proto_info_ctor(&info, parser, parent, tcphdr_len, wire_len - tcphdr_len, sport, dport, tcphdr);
// Parse TCP options
uint8_t const *options = (uint8_t *)(tcphdr+1);
assert(tcphdr_len >= sizeof(*tcphdr));
for (size_t rem_len = tcphdr_len - sizeof(*tcphdr); rem_len > 0; ) {
ssize_t const len = parse_next_option(&info, options, rem_len);
if (len < 0) return PROTO_PARSE_ERR;
rem_len -= len;
options += len;
}
// Search an already spawned subparser
struct port_key key;
port_key_init(&key, sport, dport, way);
struct mux_subparser *subparser = mux_subparser_lookup(mux_parser, NULL, NULL, &key, now);
if (subparser) SLOG(LOG_DEBUG, "Found subparser@%p for this cnx, for proto %s", subparser->parser, subparser->parser->proto->name);
if (! subparser) {
struct proto *requestor = NULL;
struct proto *sub_proto = NULL;
// Use connection tracking first
ASSIGN_INFO_OPT2(ip, ip6, parent);
if (! ip) ip = ip6;
if (ip) sub_proto = cnxtrack_ip_lookup(IPPROTO_TCP, ip->key.addr+0, sport, ip->key.addr+1, dport, now, &requestor);
if (! sub_proto) { // Then try predefined ports
sub_proto = port_muxer_find(&tcp_port_muxers, info.key.port[0], info.key.port[1]);
}
if (sub_proto) {
subparser = mux_subparser_and_parser_new(mux_parser, sub_proto, requestor, &key, now);
} else {
// Even if we have no child parser to send payload to, we want to submit payload in stream order to our plugins
subparser = tcp_subparser_new(mux_parser, NULL, NULL, &key, now);
}
}
if (! subparser) goto fallback;
// Keep track of TCP flags & ISN
struct tcp_subparser *tcp_sub = DOWNCAST(subparser, mux_subparser, tcp_subparser);
mutex_lock(tcp_sub->mutex);
if (
info.ack &&
(!IS_SET_FOR_WAY(way, tcp_sub->ack) || seqnum_gt(info.ack_num, tcp_sub->max_acknum[way]))
) {
SET_FOR_WAY(way, tcp_sub->ack);
tcp_sub->max_acknum[way] = info.ack_num;
}
if (info.fin) {
SET_FOR_WAY(way, tcp_sub->fin);
tcp_sub->fin_seqnum[way] = info.seq_num + info.info.payload; // The FIN is acked after the payload
}
if (info.syn && !IS_SET_FOR_WAY(way, tcp_sub->syn)) {
SET_FOR_WAY(way, tcp_sub->syn);
tcp_sub->isn[way] = info.seq_num;
}
if (!IS_SET_FOR_WAY(way, tcp_sub->origin)) {
SET_FOR_WAY(way, tcp_sub->origin);
tcp_sub->wl_origin[way] = info.seq_num;
if (! IS_SET_FOR_WAY(way, tcp_sub->syn)) SLOG(LOG_DEBUG, "Starting a WL while SYN is yet to be received!");
}
// Set relative sequence number if we know it
if (IS_SET_FOR_WAY(way, tcp_sub->syn)) info.rel_seq_num = info.seq_num - tcp_sub->isn[way];
// Set srv_way
assert(tcp_sub->srv_set < 3);
if (tcp_sub->srv_set == 0 || (tcp_sub->srv_set == 1 && info.syn)) {
if (comes_from_client(info.key.port, info.syn, info.ack)) {
// this packet comes from the client
tcp_sub->srv_way = !way;
} else {
tcp_sub->srv_way = way;
}
tcp_sub->srv_set = info.syn ? 2:1;
}
// Now patch it into tcp info
info.to_srv = tcp_sub->srv_way != way;
SLOG(LOG_DEBUG, "Subparser@%p state: >ISN:%"PRIu32"%s Fin:%"PRIu32" Ack:%"PRIu32" <ISN:%"PRIu32"%s Fin:%"PRIu32" Ack:%"PRIu32", SrvWay=%u%s",
subparser->parser,
IS_SET_FOR_WAY(0, tcp_sub->syn) ? tcp_sub->isn[0] : IS_SET_FOR_WAY(0, tcp_sub->origin) ? tcp_sub->wl_origin[0] : 0,
IS_SET_FOR_WAY(0, tcp_sub->syn) ? "" : " (approx)",
IS_SET_FOR_WAY(0, tcp_sub->fin) ? tcp_sub->fin_seqnum[0] : 0,
IS_SET_FOR_WAY(0, tcp_sub->ack) ? tcp_sub->max_acknum[0] : 0,
IS_SET_FOR_WAY(1, tcp_sub->syn) ? tcp_sub->isn[1] : IS_SET_FOR_WAY(1, tcp_sub->origin) ? tcp_sub->wl_origin[1] : 0,
IS_SET_FOR_WAY(1, tcp_sub->syn) ? "" : " (approx)",
IS_SET_FOR_WAY(1, tcp_sub->fin) ? tcp_sub->fin_seqnum[1] : 0,
IS_SET_FOR_WAY(1, tcp_sub->ack) ? tcp_sub->max_acknum[1] : 0,
tcp_sub->srv_way,
tcp_sub->srv_set == 0 ? " (unset)":
tcp_sub->srv_set == 1 ? " (unsure)":"(certain)");
enum proto_parse_status err;
/* Use the wait_list to parse this packet.
Notice that we do queue empty packets because subparser (or subscriber) want to receive all packets in order, including empty ones. */
size_t const packet_len = wire_len - tcphdr_len;
assert(IS_SET_FOR_WAY(way, tcp_sub->origin));
unsigned const offset = info.seq_num - tcp_sub->wl_origin[way];
unsigned const next_offset = offset + packet_len + info.syn + info.fin;
unsigned const sync_offset = info.ack_num - tcp_sub->wl_origin[!way]; // we must not parse this one before we parsed (or timeouted) this one from wl[!way]
// FIXME: Here the parser is chosen before we actually parse anything. If later the parser fails we cannot try another one.
// Choice of parser should be delayed until we start actual parse.
bool const do_sync = info.ack && IS_SET_FOR_WAY(!way, tcp_sub->origin);
err = pkt_wait_list_add(tcp_sub->wl+way, offset, next_offset, do_sync, sync_offset, true, &info.info, way, packet + tcphdr_len, cap_len - tcphdr_len, packet_len, now, tot_cap_len, tot_packet);
SLOG(LOG_DEBUG, "Waiting list returned %s", proto_parse_status_2_str(err));
if (err == PROTO_OK) {
// Try advancing each WL until we are stuck or met an error
pkt_wait_list_try_both(tcp_sub->wl+!way, &err, now, false);
}
bool const term = tcp_subparser_term(tcp_sub);
mutex_unlock(tcp_sub->mutex);
if (term || err == PROTO_PARSE_ERR) {
if (term) {
SLOG(LOG_DEBUG, "TCP cnx terminated (was %s)", parser_name(subparser->parser));
} else {
SLOG(LOG_DEBUG, "No suitable subparser for this payload");
}
mux_subparser_deindex(subparser);
}
mux_subparser_unref(&subparser);
if (err == PROTO_OK) return PROTO_OK;
fallback:
(void)proto_parse(NULL, &info.info, way, packet + tcphdr_len, cap_len - tcphdr_len, wire_len - tcphdr_len, now, tot_cap_len, tot_packet);
return PROTO_OK;
}
/*
*---------------------------------------------------------------------------
*
* vmxnet3_tx_prepare_offload --
*
* Build the offload context of a msg.
*
* Results:
* 0 if everything went well.
* +n if n bytes need to be pulled up.
* -1 in case of error (not used).
*
* Side effects:
* None.
*
*---------------------------------------------------------------------------
*/
static int
vmxnet3_tx_prepare_offload(vmxnet3_softc_t *dp,
vmxnet3_offload_t *ol,
mblk_t *mp)
{
int ret = 0;
uint32_t start, stuff, value, flags;
#if defined(OPEN_SOLARIS) || defined(SOL11)
uint32_t lso_flag, mss;
#endif
ol->om = VMXNET3_OM_NONE;
ol->hlen = 0;
ol->msscof = 0;
hcksum_retrieve(mp, NULL, NULL, &start, &stuff, NULL, &value, &flags);
#if defined(OPEN_SOLARIS) || defined(SOL11)
mac_lso_get(mp, &mss, &lso_flag);
if (flags || lso_flag) {
#else
if (flags) {
#endif
struct ether_vlan_header *eth = (void *) mp->b_rptr;
uint8_t ethLen;
if (eth->ether_tpid == htons(ETHERTYPE_VLAN)) {
ethLen = sizeof(struct ether_vlan_header);
} else {
ethLen = sizeof(struct ether_header);
}
VMXNET3_DEBUG(dp, 4, "flags=0x%x, ethLen=%u, start=%u, stuff=%u, value=%u\n",
flags, ethLen, start, stuff, value);
#if defined(OPEN_SOLARIS) || defined(SOL11)
if (lso_flag & HW_LSO) {
#else
if (flags & HCK_PARTIALCKSUM) {
ol->om = VMXNET3_OM_CSUM;
ol->hlen = start + ethLen;
ol->msscof = stuff + ethLen;
}
if (flags & HW_LSO) {
#endif
mblk_t *mblk = mp;
uint8_t *ip, *tcp;
uint8_t ipLen, tcpLen;
/*
* Copy e1000g's behavior:
* - Do not assume all the headers are in the same mblk.
* - Assume each header is always within one mblk.
* - Assume the ethernet header is in the first mblk.
*/
ip = mblk->b_rptr + ethLen;
if (ip >= mblk->b_wptr) {
mblk = mblk->b_cont;
ip = mblk->b_rptr;
}
ipLen = IPH_HDR_LENGTH((ipha_t *) ip);
tcp = ip + ipLen;
if (tcp >= mblk->b_wptr) {
mblk = mblk->b_cont;
tcp = mblk->b_rptr;
}
tcpLen = TCP_HDR_LENGTH((tcph_t *) tcp);
if (tcp + tcpLen > mblk->b_wptr) { // careful, '>' instead of '>=' here
mblk = mblk->b_cont;
}
ol->om = VMXNET3_OM_TSO;
ol->hlen = ethLen + ipLen + tcpLen;
#if defined(OPEN_SOLARIS) || defined(SOL11)
ol->msscof = mss;
#else
/* OpenSolaris fills 'value' with the MSS but Solaris doesn't. */
ol->msscof = DB_LSOMSS(mp);
#endif
if (mblk != mp) {
ret = ol->hlen;
}
}
#if defined(OPEN_SOLARIS) || defined(SOL11)
else if (flags & HCK_PARTIALCKSUM) {
ol->om = VMXNET3_OM_CSUM;
ol->hlen = start + ethLen;
ol->msscof = stuff + ethLen;
}
#endif
}
return ret;
}
/*
*---------------------------------------------------------------------------
*
* vmxnet3_tx_one --
*
* Map a msg into the Tx command ring of a vmxnet3 device.
*
* Results:
* VMXNET3_TX_OK if everything went well.
* VMXNET3_TX_RINGFULL if the ring is nearly full.
* VMXNET3_TX_PULLUP if the msg is overfragmented.
* VMXNET3_TX_FAILURE if there was a DMA or offload error.
*
* Side effects:
* The ring is filled if VMXNET3_TX_OK is returned.
*
*---------------------------------------------------------------------------
*/
static vmxnet3_txstatus
vmxnet3_tx_one(vmxnet3_softc_t *dp,
vmxnet3_txqueue_t *txq,
vmxnet3_offload_t *ol,
mblk_t *mp,
boolean_t retry)
{
int ret = VMXNET3_TX_OK;
unsigned int frags = 0, totLen = 0;
vmxnet3_cmdring_t *cmdRing = &txq->cmdRing;
Vmxnet3_TxQueueCtrl *txqCtrl = txq->sharedCtrl;
Vmxnet3_GenericDesc *txDesc;
uint16_t sopIdx, eopIdx;
uint8_t sopGen, curGen;
mblk_t *mblk;
mutex_enter(&dp->txLock);
sopIdx = eopIdx = cmdRing->next2fill;
sopGen = cmdRing->gen;
curGen = !cmdRing->gen;
for (mblk = mp; mblk != NULL; mblk = mblk->b_cont) {
unsigned int len = MBLKL(mblk);
ddi_dma_cookie_t cookie;
uint_t cookieCount;
if (len) {
totLen += len;
} else {
continue;
}
if (ddi_dma_addr_bind_handle(dp->txDmaHandle, NULL,
(caddr_t) mblk->b_rptr, len,
DDI_DMA_RDWR | DDI_DMA_STREAMING,
DDI_DMA_DONTWAIT, NULL,
&cookie, &cookieCount) != DDI_DMA_MAPPED) {
VMXNET3_WARN(dp, "ddi_dma_addr_bind_handle() failed\n");
ret = VMXNET3_TX_FAILURE;
goto error;
}
ASSERT(cookieCount);
do {
uint64_t addr = cookie.dmac_laddress;
size_t len = cookie.dmac_size;
do {
uint32_t dw2, dw3;
size_t chunkLen;
ASSERT(!txq->metaRing[eopIdx].mp);
ASSERT(cmdRing->avail - frags);
if (frags >= cmdRing->size - 1 ||
(ol->om != VMXNET3_OM_TSO && frags >= VMXNET3_MAX_TXD_PER_PKT)) {
if (retry) {
VMXNET3_DEBUG(dp, 2, "overfragmented, frags=%u ring=%hu om=%hu\n",
frags, cmdRing->size, ol->om);
}
ddi_dma_unbind_handle(dp->txDmaHandle);
ret = VMXNET3_TX_PULLUP;
goto error;
}
if (cmdRing->avail - frags <= 1) {
dp->txMustResched = B_TRUE;
ddi_dma_unbind_handle(dp->txDmaHandle);
ret = VMXNET3_TX_RINGFULL;
goto error;
}
if (len > VMXNET3_MAX_TX_BUF_SIZE) {
chunkLen = VMXNET3_MAX_TX_BUF_SIZE;
} else {
chunkLen = len;
}
frags++;
eopIdx = cmdRing->next2fill;
txDesc = VMXNET3_GET_DESC(cmdRing, eopIdx);
ASSERT(txDesc->txd.gen != cmdRing->gen);
// txd.addr
txDesc->txd.addr = addr;
// txd.dw2
dw2 = chunkLen == VMXNET3_MAX_TX_BUF_SIZE ? 0 : chunkLen;
dw2 |= curGen << VMXNET3_TXD_GEN_SHIFT;
txDesc->dword[2] = dw2;
ASSERT(txDesc->txd.len == len || txDesc->txd.len == 0);
// txd.dw3
dw3 = 0;
txDesc->dword[3] = dw3;
VMXNET3_INC_RING_IDX(cmdRing, cmdRing->next2fill);
curGen = cmdRing->gen;
addr += chunkLen;
len -= chunkLen;
} while (len);
if (--cookieCount) {
ddi_dma_nextcookie(dp->txDmaHandle, &cookie);
}
} while (cookieCount);
ddi_dma_unbind_handle(dp->txDmaHandle);
}
/* Update the EOP descriptor */
txDesc = VMXNET3_GET_DESC(cmdRing, eopIdx);
txDesc->dword[3] |= VMXNET3_TXD_CQ | VMXNET3_TXD_EOP;
/* Update the SOP descriptor. Must be done last */
txDesc = VMXNET3_GET_DESC(cmdRing, sopIdx);
if (ol->om == VMXNET3_OM_TSO &&
txDesc->txd.len != 0 &&
txDesc->txd.len < ol->hlen) {
ret = VMXNET3_TX_FAILURE;
goto error;
}
txDesc->txd.om = ol->om;
txDesc->txd.hlen = ol->hlen;
txDesc->txd.msscof = ol->msscof;
membar_producer();
txDesc->txd.gen = sopGen;
/* Update the meta ring & metadata */
txq->metaRing[sopIdx].mp = mp;
txq->metaRing[eopIdx].sopIdx = sopIdx;
txq->metaRing[eopIdx].frags = frags;
cmdRing->avail -= frags;
if (ol->om == VMXNET3_OM_TSO) {
txqCtrl->txNumDeferred +=
(totLen - ol->hlen + ol->msscof - 1) / ol->msscof;
} else {
txqCtrl->txNumDeferred++;
}
VMXNET3_DEBUG(dp, 3, "tx 0x%p on [%u;%u]\n", mp, sopIdx, eopIdx);
goto done;
error:
/* Reverse the generation bits */
while (sopIdx != cmdRing->next2fill) {
VMXNET3_DEC_RING_IDX(cmdRing, cmdRing->next2fill);
txDesc = VMXNET3_GET_DESC(cmdRing, cmdRing->next2fill);
txDesc->txd.gen = !cmdRing->gen;
}
done:
mutex_exit(&dp->txLock);
return ret;
}
/*
*---------------------------------------------------------------------------
*
* vmxnet3_tx --
*
* Send packets on a vmxnet3 device.
*
* Results:
* NULL in case of success or failure.
* The mps to be retransmitted later if the ring is full.
*
* Side effects:
* None.
*
*---------------------------------------------------------------------------
*/
mblk_t *
vmxnet3_tx(void *data, mblk_t *mps)
{
vmxnet3_softc_t *dp = data;
vmxnet3_txqueue_t *txq = &dp->txQueue;
vmxnet3_cmdring_t *cmdRing = &txq->cmdRing;
Vmxnet3_TxQueueCtrl *txqCtrl = txq->sharedCtrl;
vmxnet3_txstatus status = VMXNET3_TX_OK;
mblk_t *mp;
ASSERT(mps != NULL);
do {
vmxnet3_offload_t ol;
int pullup;
mp = mps;
mps = mp->b_next;
mp->b_next = NULL;
if (DB_TYPE(mp) != M_DATA) {
/*
* PR #315560: Solaris might pass M_PROTO mblks for some reason.
* Drop them because we don't understand them and because their
* contents are not Ethernet frames anyway.
*/
ASSERT(B_FALSE);
freemsg(mp);
continue;
}
/*
* Prepare the offload while we're still handling the original
* message -- msgpullup() discards the metadata afterwards.
*/
pullup = vmxnet3_tx_prepare_offload(dp, &ol, mp);
if (pullup) {
mblk_t *new_mp = msgpullup(mp, pullup);
freemsg(mp);
if (new_mp) {
mp = new_mp;
} else {
continue;
}
}
/*
* Try to map the message in the Tx ring.
* This call might fail for non-fatal reasons.
*/
status = vmxnet3_tx_one(dp, txq, &ol, mp, B_FALSE);
if (status == VMXNET3_TX_PULLUP) {
/*
* Try one more time after flattening
* the message with msgpullup().
*/
if (mp->b_cont != NULL) {
mblk_t *new_mp = msgpullup(mp, -1);
freemsg(mp);
if (new_mp) {
mp = new_mp;
status = vmxnet3_tx_one(dp, txq, &ol, mp, B_TRUE);
} else {
continue;
}
}
}
if (status != VMXNET3_TX_OK && status != VMXNET3_TX_RINGFULL) {
/* Fatal failure, drop it */
freemsg(mp);
}
} while (mps && status != VMXNET3_TX_RINGFULL);
if (status == VMXNET3_TX_RINGFULL) {
mp->b_next = mps;
mps = mp;
} else {
ASSERT(!mps);
}
/* Notify the device */
mutex_enter(&dp->txLock);
if (txqCtrl->txNumDeferred >= txqCtrl->txThreshold) {
txqCtrl->txNumDeferred = 0;
VMXNET3_BAR0_PUT32(dp, VMXNET3_REG_TXPROD, cmdRing->next2fill);
}
mutex_exit(&dp->txLock);
return mps;
}
/* ARGSUSED */
static int
vmxnet3s_tx_prepare_offload(vmxnet3s_softc_t *dp, vmxnet3s_offload_t *ol,
mblk_t *mp, int *to_copy)
{
int ret = 0;
uint32_t start;
uint32_t stuff;
uint32_t value;
uint32_t flags;
uint32_t lsoflags;
uint32_t mss;
ol->om = VMXNET3_OM_NONE;
ol->hlen = 0;
ol->msscof = 0;
hcksum_retrieve(mp, NULL, NULL, &start, &stuff, NULL, &value, &flags);
mac_lso_get(mp, &mss, &lsoflags);
if (lsoflags & HW_LSO)
flags |= HW_LSO;
if (flags) {
struct ether_vlan_header *eth = (void *) mp->b_rptr;
uint8_t ethlen;
if (eth->ether_tpid == htons(ETHERTYPE_VLAN))
ethlen = sizeof (struct ether_vlan_header);
else
ethlen = sizeof (struct ether_header);
if (flags & HCK_PARTIALCKSUM) {
ol->om = VMXNET3_OM_CSUM;
ol->hlen = start + ethlen;
ol->msscof = stuff + ethlen;
}
if (flags & HW_LSO) {
mblk_t *mblk = mp;
uint8_t *ip;
uint8_t *tcp;
uint8_t iplen;
uint8_t tcplen;
/*
* Copy e1000g's behavior:
* - Do not assume all the headers are in the same mblk.
* - Assume each header is always within one mblk.
* - Assume the ethernet header is in the first mblk.
*/
ip = mblk->b_rptr + ethlen;
if (ip >= mblk->b_wptr) {
mblk = mblk->b_cont;
ip = mblk->b_rptr;
}
iplen = IPH_HDR_LENGTH((ipha_t *)ip);
tcp = ip + iplen;
if (tcp >= mblk->b_wptr) {
mblk = mblk->b_cont;
tcp = mblk->b_rptr;
}
tcplen = TCP_HDR_LENGTH((tcph_t *)tcp);
/* Careful, '>' instead of '>=' here */
if (tcp + tcplen > mblk->b_wptr)
mblk = mblk->b_cont;
ol->om = VMXNET3_OM_TSO;
ol->hlen = ethlen + iplen + tcplen;
ol->msscof = DB_LSOMSS(mp);
if (mblk != mp)
ret = ol->hlen;
else
*to_copy = ol->hlen;
}
}
return (ret);
}
