Author: gallatin
Date: Tue Feb 26 15:51:46 2013
New Revision: 247320
URL: http://svnweb.freebsd.org/changeset/base/247320
Log:
MFC r247011:
Add support to mxge for IPv6 TX csum offload & IPv6 TSO.
Sponsored by: Myricom, Inc.
Modified:
stable/9/sys/dev/mxge/if_mxge.c
stable/9/sys/dev/mxge/if_mxge_var.h
stable/9/sys/modules/mxge/mxge/Makefile
Directory Properties:
stable/9/sys/ (props changed)
stable/9/sys/dev/ (props changed)
stable/9/sys/modules/ (props changed)
Modified: stable/9/sys/dev/mxge/if_mxge.c
==============================================================================
--- stable/9/sys/dev/mxge/if_mxge.c Tue Feb 26 15:25:43 2013
(r247319)
+++ stable/9/sys/dev/mxge/if_mxge.c Tue Feb 26 15:51:46 2013
(r247320)
@@ -62,7 +62,9 @@ __FBSDID("$FreeBSD$");
#include <netinet/in_systm.h>
#include <netinet/in.h>
#include <netinet/ip.h>
+#include <netinet/ip6.h>
#include <netinet/tcp.h>
+#include <netinet6/ip6_var.h>
#include <machine/bus.h>
#include <machine/in_cksum.h>
@@ -91,6 +93,7 @@ __FBSDID("$FreeBSD$");
#endif
#include "opt_inet.h"
+#include "opt_inet6.h"
/* tunable params */
static int mxge_nvidia_ecrc_enable = 1;
@@ -1809,21 +1812,99 @@ mxge_submit_req(mxge_tx_ring_t *tx, mcp_
wmb();
}
+static int
+mxge_parse_tx(struct mxge_slice_state *ss, struct mbuf *m,
+ struct mxge_pkt_info *pi)
+{
+ struct ether_vlan_header *eh;
+ uint16_t etype;
+ int tso = m->m_pkthdr.csum_flags & (CSUM_TSO);
+#if IFCAP_TSO6 && defined(INET6)
+ int nxt;
+#endif
+
+ eh = mtod(m, struct ether_vlan_header *);
+ if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
+ etype = ntohs(eh->evl_proto);
+ pi->ip_off = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
+ } else {
+ etype = ntohs(eh->evl_encap_proto);
+ pi->ip_off = ETHER_HDR_LEN;
+ }
+
+ switch (etype) {
+ case ETHERTYPE_IP:
+ /*
+ * ensure ip header is in first mbuf, copy it to a
+ * scratch buffer if not
+ */
+ pi->ip = (struct ip *)(m->m_data + pi->ip_off);
+ pi->ip6 = NULL;
+ if (__predict_false(m->m_len < pi->ip_off + sizeof(*pi->ip))) {
+ m_copydata(m, 0, pi->ip_off + sizeof(*pi->ip),
+ ss->scratch);
+ pi->ip = (struct ip *)(ss->scratch + pi->ip_off);
+ }
+ pi->ip_hlen = pi->ip->ip_hl << 2;
+ if (!tso)
+ return 0;
+
+ if (__predict_false(m->m_len < pi->ip_off + pi->ip_hlen +
+ sizeof(struct tcphdr))) {
+ m_copydata(m, 0, pi->ip_off + pi->ip_hlen +
+ sizeof(struct tcphdr), ss->scratch);
+ pi->ip = (struct ip *)(ss->scratch + pi->ip_off);
+ }
+ pi->tcp = (struct tcphdr *)((char *)pi->ip + pi->ip_hlen);
+ break;
+#if IFCAP_TSO6 && defined(INET6)
+ case ETHERTYPE_IPV6:
+ pi->ip6 = (struct ip6_hdr *)(m->m_data + pi->ip_off);
+ if (__predict_false(m->m_len < pi->ip_off + sizeof(*pi->ip6))) {
+ m_copydata(m, 0, pi->ip_off + sizeof(*pi->ip6),
+ ss->scratch);
+ pi->ip6 = (struct ip6_hdr *)(ss->scratch + pi->ip_off);
+ }
+ nxt = 0;
+ pi->ip_hlen = ip6_lasthdr(m, pi->ip_off, IPPROTO_IPV6, &nxt);
+ pi->ip_hlen -= pi->ip_off;
+ if (nxt != IPPROTO_TCP && nxt != IPPROTO_UDP)
+ return EINVAL;
+
+ if (!tso)
+ return 0;
+
+ if (pi->ip_off + pi->ip_hlen > ss->sc->max_tso6_hlen)
+ return EINVAL;
+
+ if (__predict_false(m->m_len < pi->ip_off + pi->ip_hlen +
+ sizeof(struct tcphdr))) {
+ m_copydata(m, 0, pi->ip_off + pi->ip_hlen +
+ sizeof(struct tcphdr), ss->scratch);
+ pi->ip6 = (struct ip6_hdr *)(ss->scratch + pi->ip_off);
+ }
+ pi->tcp = (struct tcphdr *)((char *)pi->ip6 + pi->ip_hlen);
+ break;
+#endif
+ default:
+ return EINVAL;
+ }
+ return 0;
+}
+
#if IFCAP_TSO4
static void
mxge_encap_tso(struct mxge_slice_state *ss, struct mbuf *m,
- int busdma_seg_cnt, int ip_off)
+ int busdma_seg_cnt, struct mxge_pkt_info *pi)
{
mxge_tx_ring_t *tx;
mcp_kreq_ether_send_t *req;
bus_dma_segment_t *seg;
- struct ip *ip;
- struct tcphdr *tcp;
uint32_t low, high_swapped;
int len, seglen, cum_len, cum_len_next;
int next_is_first, chop, cnt, rdma_count, small;
- uint16_t pseudo_hdr_offset, cksum_offset, mss;
+ uint16_t pseudo_hdr_offset, cksum_offset, mss, sum;
uint8_t flags, flags_next;
static int once;
@@ -1834,38 +1915,33 @@ mxge_encap_tso(struct mxge_slice_state *
* header portion of the TSO packet.
*/
- /* ensure we have the ethernet, IP and TCP
- header together in the first mbuf, copy
- it to a scratch buffer if not */
- if (__predict_false(m->m_len < ip_off + sizeof (*ip))) {
- m_copydata(m, 0, ip_off + sizeof (*ip),
- ss->scratch);
- ip = (struct ip *)(ss->scratch + ip_off);
- } else {
- ip = (struct ip *)(mtod(m, char *) + ip_off);
- }
- if (__predict_false(m->m_len < ip_off + (ip->ip_hl << 2)
- + sizeof (*tcp))) {
- m_copydata(m, 0, ip_off + (ip->ip_hl << 2)
- + sizeof (*tcp), ss->scratch);
- ip = (struct ip *)(mtod(m, char *) + ip_off);
- }
-
- tcp = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
- cum_len = -(ip_off + ((ip->ip_hl + tcp->th_off) << 2));
- cksum_offset = ip_off + (ip->ip_hl << 2);
+ cksum_offset = pi->ip_off + pi->ip_hlen;
+ cum_len = -(cksum_offset + (pi->tcp->th_off << 2));
/* TSO implies checksum offload on this hardware */
- if (__predict_false((m->m_pkthdr.csum_flags & (CSUM_TCP)) == 0)) {
+ if (__predict_false((m->m_pkthdr.csum_flags & (CSUM_TCP|CSUM_TCP_IPV6))
== 0)) {
/*
* If packet has full TCP csum, replace it with pseudo hdr
* sum that the NIC expects, otherwise the NIC will emit
* packets with bad TCP checksums.
*/
- m->m_pkthdr.csum_flags = CSUM_TCP;
m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
- tcp->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
- htons(IPPROTO_TCP + (m->m_pkthdr.len - cksum_offset)));
+ if (pi->ip6) {
+#if (CSUM_TCP_IPV6 != 0) && defined(INET6)
+ m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6;
+ sum = in6_cksum_pseudo(pi->ip6,
+ m->m_pkthdr.len - cksum_offset,
+ IPPROTO_TCP, 0);
+#endif
+ } else {
+ m->m_pkthdr.csum_flags |= CSUM_TCP;
+ sum = in_pseudo(pi->ip->ip_src.s_addr,
+ pi->ip->ip_dst.s_addr,
+ htons(IPPROTO_TCP + (m->m_pkthdr.len -
+ cksum_offset)));
+ }
+ m_copyback(m, offsetof(struct tcphdr, th_sum) +
+ cksum_offset, sizeof(sum), (caddr_t)&sum);
}
flags = MXGEFW_FLAGS_TSO_HDR | MXGEFW_FLAGS_FIRST;
@@ -1875,6 +1951,14 @@ mxge_encap_tso(struct mxge_slice_state *
* the checksum by parsing the header. */
pseudo_hdr_offset = htobe16(mss);
+ if (pi->ip6) {
+ /*
+ * for IPv6 TSO, the "checksum offset" is re-purposed
+ * to store the TCP header len
+ */
+ cksum_offset = (pi->tcp->th_off << 2);
+ }
+
tx = &ss->tx;
req = tx->req_list;
seg = tx->seg_list;
@@ -1946,10 +2030,12 @@ mxge_encap_tso(struct mxge_slice_state *
req++;
cnt++;
rdma_count++;
- if (__predict_false(cksum_offset > seglen))
- cksum_offset -= seglen;
- else
- cksum_offset = 0;
+ if (cksum_offset != 0 && !pi->ip6) {
+ if (__predict_false(cksum_offset > seglen))
+ cksum_offset -= seglen;
+ else
+ cksum_offset = 0;
+ }
if (__predict_false(cnt > tx->max_desc))
goto drop;
}
@@ -2029,14 +2115,14 @@ mxge_vlan_tag_insert(struct mbuf *m)
static void
mxge_encap(struct mxge_slice_state *ss, struct mbuf *m)
{
+ struct mxge_pkt_info pi = {0,0,0,0};
mxge_softc_t *sc;
mcp_kreq_ether_send_t *req;
bus_dma_segment_t *seg;
struct mbuf *m_tmp;
struct ifnet *ifp;
mxge_tx_ring_t *tx;
- struct ip *ip;
- int cnt, cum_len, err, i, idx, odd_flag, ip_off;
+ int cnt, cum_len, err, i, idx, odd_flag;
uint16_t pseudo_hdr_offset;
uint8_t flags, cksum_offset;
@@ -2045,15 +2131,19 @@ mxge_encap(struct mxge_slice_state *ss,
ifp = sc->ifp;
tx = &ss->tx;
- ip_off = sizeof (struct ether_header);
#ifdef MXGE_NEW_VLAN_API
if (m->m_flags & M_VLANTAG) {
m = mxge_vlan_tag_insert(m);
if (__predict_false(m == NULL))
- goto drop;
- ip_off += ETHER_VLAN_ENCAP_LEN;
+ goto drop_without_m;
}
#endif
+ if (m->m_pkthdr.csum_flags &
+ (CSUM_TSO | CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) {
+ if (mxge_parse_tx(ss, m, &pi))
+ goto drop;
+ }
+
/* (try to) map the frame for DMA */
idx = tx->req & tx->mask;
err = bus_dmamap_load_mbuf_sg(tx->dmat, tx->info[idx].map,
@@ -2085,7 +2175,7 @@ mxge_encap(struct mxge_slice_state *ss,
#if IFCAP_TSO4
/* TSO is different enough, we handle it in another routine */
if (m->m_pkthdr.csum_flags & (CSUM_TSO)) {
- mxge_encap_tso(ss, m, cnt, ip_off);
+ mxge_encap_tso(ss, m, cnt, &pi);
return;
}
#endif
@@ -2096,17 +2186,11 @@ mxge_encap(struct mxge_slice_state *ss,
flags = MXGEFW_FLAGS_NO_TSO;
/* checksum offloading? */
- if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) {
+ if (m->m_pkthdr.csum_flags &
+ (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) {
/* ensure ip header is in first mbuf, copy
it to a scratch buffer if not */
- if (__predict_false(m->m_len < ip_off + sizeof (*ip))) {
- m_copydata(m, 0, ip_off + sizeof (*ip),
- ss->scratch);
- ip = (struct ip *)(ss->scratch + ip_off);
- } else {
- ip = (struct ip *)(mtod(m, char *) + ip_off);
- }
- cksum_offset = ip_off + (ip->ip_hl << 2);
+ cksum_offset = pi.ip_off + pi.ip_hlen;
pseudo_hdr_offset = cksum_offset + m->m_pkthdr.csum_data;
pseudo_hdr_offset = htobe16(pseudo_hdr_offset);
req->cksum_offset = cksum_offset;
@@ -2189,6 +2273,7 @@ mxge_encap(struct mxge_slice_state *ss,
drop:
m_freem(m);
+drop_without_m:
ss->oerrors++;
return;
}
@@ -4125,8 +4210,7 @@ mxge_ioctl(struct ifnet *ifp, u_long com
if (mask & IFCAP_TXCSUM) {
if (IFCAP_TXCSUM & ifp->if_capenable) {
ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
- ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
- | CSUM_TSO);
+ ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP);
} else {
ifp->if_capenable |= IFCAP_TXCSUM;
ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP);
@@ -4143,7 +4227,6 @@ mxge_ioctl(struct ifnet *ifp, u_long com
if (mask & IFCAP_TSO4) {
if (IFCAP_TSO4 & ifp->if_capenable) {
ifp->if_capenable &= ~IFCAP_TSO4;
- ifp->if_hwassist &= ~CSUM_TSO;
} else if (IFCAP_TXCSUM & ifp->if_capenable) {
ifp->if_capenable |= IFCAP_TSO4;
ifp->if_hwassist |= CSUM_TSO;
@@ -4153,6 +4236,43 @@ mxge_ioctl(struct ifnet *ifp, u_long com
err = EINVAL;
}
}
+#if IFCAP_TSO6
+ if (mask & IFCAP_TXCSUM_IPV6) {
+ if (IFCAP_TXCSUM_IPV6 & ifp->if_capenable) {
+ ifp->if_capenable &= ~(IFCAP_TXCSUM_IPV6
+ | IFCAP_TSO6);
+ ifp->if_hwassist &= ~(CSUM_TCP_IPV6
+ | CSUM_UDP);
+ } else {
+ ifp->if_capenable |= IFCAP_TXCSUM_IPV6;
+ ifp->if_hwassist |= (CSUM_TCP_IPV6
+ | CSUM_UDP_IPV6);
+ }
+#ifdef NOTYET
+ } else if (mask & IFCAP_RXCSUM6) {
+ if (IFCAP_RXCSUM6 & ifp->if_capenable) {
+ ifp->if_capenable &= ~IFCAP_RXCSUM6;
+ sc->csum_flag = 0;
+ } else {
+ ifp->if_capenable |= IFCAP_RXCSUM6;
+ sc->csum_flag = 1;
+ }
+#endif
+ }
+ if (mask & IFCAP_TSO6) {
+ if (IFCAP_TSO6 & ifp->if_capenable) {
+ ifp->if_capenable &= ~IFCAP_TSO6;
+ } else if (IFCAP_TXCSUM_IPV6 & ifp->if_capenable) {
+ ifp->if_capenable |= IFCAP_TSO6;
+ ifp->if_hwassist |= CSUM_TSO;
+ } else {
+ printf("mxge requires tx checksum offload"
+ " be enabled to use TSO\n");
+ err = EINVAL;
+ }
+ }
+#endif /*IFCAP_TSO6 */
+
if (mask & IFCAP_LRO) {
if (IFCAP_LRO & ifp->if_capenable)
err = mxge_change_lro_locked(sc, 0);
@@ -4645,6 +4765,7 @@ mxge_add_irq(mxge_softc_t *sc)
static int
mxge_attach(device_t dev)
{
+ mxge_cmd_t cmd;
mxge_softc_t *sc = device_get_softc(dev);
struct ifnet *ifp;
int err, rid;
@@ -4775,7 +4896,7 @@ mxge_attach(device_t dev)
ifp->if_baudrate = IF_Gbps(10UL);
ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TXCSUM | IFCAP_TSO4 |
- IFCAP_VLAN_MTU | IFCAP_LINKSTATE;
+ IFCAP_VLAN_MTU | IFCAP_LINKSTATE | IFCAP_TXCSUM_IPV6;
#ifdef INET
ifp->if_capabilities |= IFCAP_LRO;
#endif
@@ -4788,7 +4909,6 @@ mxge_attach(device_t dev)
sc->fw_ver_tiny >= 32)
ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
#endif
-
sc->max_mtu = mxge_max_mtu(sc);
if (sc->max_mtu >= 9000)
ifp->if_capabilities |= IFCAP_JUMBO_MTU;
@@ -4797,6 +4917,14 @@ mxge_attach(device_t dev)
"latest firmware for 9000 byte jumbo support\n",
sc->max_mtu - ETHER_HDR_LEN);
ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_TSO;
+ ifp->if_hwassist |= CSUM_TCP_IPV6 | CSUM_UDP_IPV6;
+ /* check to see if f/w supports TSO for IPv6 */
+ if (!mxge_send_cmd(sc, MXGEFW_CMD_GET_MAX_TSO6_HDR_SIZE, &cmd)) {
+ if (CSUM_TCP_IPV6)
+ ifp->if_capabilities |= IFCAP_TSO6;
+ sc->max_tso6_hlen = min(cmd.data0,
+ sizeof (sc->ss[0].scratch));
+ }
ifp->if_capenable = ifp->if_capabilities;
if (sc->lro_cnt == 0)
ifp->if_capenable &= ~IFCAP_LRO;
Modified: stable/9/sys/dev/mxge/if_mxge_var.h
==============================================================================
--- stable/9/sys/dev/mxge/if_mxge_var.h Tue Feb 26 15:25:43 2013
(r247319)
+++ stable/9/sys/dev/mxge/if_mxge_var.h Tue Feb 26 15:51:46 2013
(r247320)
@@ -50,6 +50,19 @@ $FreeBSD$
#define IFNET_BUF_RING 1
#endif
+#if (__FreeBSD_version < 1000020)
+#undef IF_Kbps
+#undef IF_Mbps
+#undef IF_Gbps
+#define IF_Kbps(x) ((uintmax_t)(x) * 1000) /* kilobits/sec. */
+#define IF_Mbps(x) (IF_Kbps((x) * 1000)) /* megabits/sec. */
+#define IF_Gbps(x) (IF_Mbps((x) * 1000)) /* gigabits/sec. */
+static __inline void
+if_initbaudrate(struct ifnet *ifp, uintmax_t baud)
+{
+ ifp->if_baudrate = baud;
+}
+#endif
#ifndef VLAN_CAPABILITIES
#define VLAN_CAPABILITIES(ifp)
#define mxge_vlans_active(sc) (sc)->ifp->if_nvlans
@@ -73,10 +86,33 @@ $FreeBSD$
#define IFCAP_TSO4 0
#endif
+#ifndef IFCAP_TSO6
+#define IFCAP_TSO6 0
+#endif
+
+#ifndef IFCAP_TXCSUM_IPV6
+#define IFCAP_TXCSUM_IPV6 0
+#endif
+
+#ifndef IFCAP_RXCSUM_IPV6
+#define IFCAP_RXCSUM_IPV6 0
+#endif
+
#ifndef CSUM_TSO
#define CSUM_TSO 0
#endif
+#ifndef CSUM_TCP_IPV6
+#define CSUM_TCP_IPV6 0
+#endif
+
+#ifndef CSUM_UDP_IPV6
+#define CSUM_UDP_IPV6 0
+#endif
+
+#ifndef CSUM_DELAY_DATA_IPV6
+#define CSUM_DELAY_DATA_IPV6 0
+#endif
typedef struct {
void *addr;
@@ -270,6 +306,7 @@ struct mxge_softc {
int dying;
int connector;
int current_media;
+ int max_tso6_hlen;
mxge_dma_t dmabench_dma;
struct callout co_hdl;
struct taskqueue *tq;
@@ -312,6 +349,15 @@ struct mxge_media_type
char *name;
};
+struct mxge_pkt_info {
+ int ip_off;
+ int ip_hlen;
+ struct ip *ip;
+ struct ip6_hdr *ip6;
+ struct tcphdr *tcp;
+};
+
+
/* implement our own memory barriers, since bus_space_barrier
cannot handle write-combining regions */
Modified: stable/9/sys/modules/mxge/mxge/Makefile
==============================================================================
--- stable/9/sys/modules/mxge/mxge/Makefile Tue Feb 26 15:25:43 2013
(r247319)
+++ stable/9/sys/modules/mxge/mxge/Makefile Tue Feb 26 15:51:46 2013
(r247320)
@@ -3,6 +3,6 @@
.PATH: ${.CURDIR}/../../../dev/mxge
KMOD= if_mxge
-SRCS= if_mxge.c mxge_lro.c device_if.h bus_if.h pci_if.h opt_inet.h
+SRCS= if_mxge.c mxge_lro.c device_if.h bus_if.h pci_if.h opt_inet.h
opt_inet6.h
.include <bsd.kmod.mk>
_______________________________________________
svn-src-all@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-all
To unsubscribe, send any mail to "svn-src-all-unsubscr...@freebsd.org"
