bce: Add TSO support

Obatined-from: FreeBSD

With following modification:
These chips can handle ip.ip_len and tcphdr.th_sum, if they are setup
according to Microsoft LSO specification, so ip.ip_len should not be
cleared and tcphdr.th_sum should be left as it is.

According-to: bnx2

diff --git a/sys/dev/netif/bce/if_bce.c b/sys/dev/netif/bce/if_bce.c
index ea252de..b963e1a 100644 (file)
--- a/sys/dev/netif/bce/if_bce.c
+++ b/sys/dev/netif/bce/if_bce.c
@@ -68,6 +68,9 @@
 #include <sys/sockio.h>
 #include <sys/sysctl.h>
 
+#include <netinet/ip.h>
+#include <netinet/tcp.h>
+
 #include <net/bpf.h>
 #include <net/ethernet.h>
 #include <net/if.h>
@@ -417,6 +420,8 @@ static void bce_free_rx_chain(struct bce_softc *);
 static void    bce_free_tx_chain(struct bce_softc *);
 
 static int     bce_encap(struct bce_softc *, struct mbuf **);
+static int     bce_tso_setup(struct bce_softc *, struct mbuf **,
+                   uint16_t *, uint16_t *);
 static void    bce_start(struct ifnet *);
 static int     bce_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
 static void    bce_watchdog(struct ifnet *);
@@ -952,7 +957,7 @@ bce_attach(device_t dev)
        ifp->if_poll = bce_poll;
 #endif
        ifp->if_mtu = ETHERMTU;
-       ifp->if_hwassist = BCE_IF_HWASSIST;
+       ifp->if_hwassist = BCE_CSUM_FEATURES | CSUM_TSO;
        ifp->if_capabilities = BCE_IF_CAPABILITIES;
        ifp->if_capenable = ifp->if_capabilities;
        ifq_set_maxlen(&ifp->if_snd, USABLE_TX_BD(sc));
@@ -2436,8 +2441,8 @@ bce_dma_alloc(struct bce_softc *sc)
        rc = bus_dma_tag_create(sc->parent_tag, 1, 0,
                                BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
                                NULL, NULL,
-                               /* BCE_MAX_JUMBO_ETHER_MTU_VLAN */MCLBYTES,
-                               BCE_MAX_SEGMENTS, MCLBYTES,
+                               IP_MAXPACKET + sizeof(struct ether_vlan_header),
+                               BCE_MAX_SEGMENTS, PAGE_SIZE,
                                BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
                                BUS_DMA_ONEBPAGE,
                                &sc->tx_mbuf_tag);
@@ -4845,13 +4850,18 @@ bce_encap(struct bce_softc *sc, struct mbuf **m_head)
        bus_dmamap_t map, tmp_map;
        struct mbuf *m0 = *m_head;
        struct tx_bd *txbd = NULL;
-       uint16_t vlan_tag = 0, flags = 0;
+       uint16_t vlan_tag = 0, flags = 0, mss = 0;
        uint16_t chain_prod, chain_prod_start, prod;
        uint32_t prod_bseq;
        int i, error, maxsegs, nsegs;
 
        /* Transfer any checksum offload flags to the bd. */
-       if (m0->m_pkthdr.csum_flags) {
+       if (m0->m_pkthdr.csum_flags & CSUM_TSO) {
+               error = bce_tso_setup(sc, m_head, &flags, &mss);
+               if (error)
+                       return ENOBUFS;
+               m0 = *m_head;
+       } else if (m0->m_pkthdr.csum_flags & BCE_CSUM_FEATURES) {
                if (m0->m_pkthdr.csum_flags & CSUM_IP)
                        flags |= TX_BD_FLAGS_IP_CKSUM;
                if (m0->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
@@ -4901,9 +4911,11 @@ bce_encap(struct bce_softc *sc, struct mbuf **m_head)
 
                txbd->tx_bd_haddr_lo = htole32(BCE_ADDR_LO(segs[i].ds_addr));
                txbd->tx_bd_haddr_hi = htole32(BCE_ADDR_HI(segs[i].ds_addr));
-               txbd->tx_bd_mss_nbytes = htole16(segs[i].ds_len);
+               txbd->tx_bd_mss_nbytes = htole32(mss << 16) |
+                   htole16(segs[i].ds_len);
                txbd->tx_bd_vlan_tag = htole16(vlan_tag);
                txbd->tx_bd_flags = htole16(flags);
+
                prod_bseq += segs[i].ds_len;
                if (i == 0)
                        txbd->tx_bd_flags |= htole16(TX_BD_FLAGS_START);
@@ -5106,10 +5118,17 @@ bce_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
 
                if (mask & IFCAP_HWCSUM) {
                        ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
-                       if (IFCAP_HWCSUM & ifp->if_capenable)
-                               ifp->if_hwassist = BCE_IF_HWASSIST;
+                       if (ifp->if_capenable & IFCAP_TXCSUM)
+                               ifp->if_hwassist |= BCE_CSUM_FEATURES;
+                       else
+                               ifp->if_hwassist &= ~BCE_CSUM_FEATURES;
+               }
+               if (mask & IFCAP_TSO) {
+                       ifp->if_capenable ^= IFCAP_TSO;
+                       if (ifp->if_capenable & IFCAP_TSO)
+                               ifp->if_hwassist |= CSUM_TSO;
                        else
-                               ifp->if_hwassist = 0;
+                               ifp->if_hwassist &= ~CSUM_TSO;
                }
                break;
 
@@ -7913,3 +7932,47 @@ bce_coal_change(struct bce_softc *sc)
 
        sc->bce_coalchg_mask = 0;
 }
+
+static int
+bce_tso_setup(struct bce_softc *sc, struct mbuf **mp,
+    uint16_t *flags0, uint16_t *mss0)
+{
+       struct mbuf *m;
+       uint16_t flags;
+       int thoff, iphlen, hoff;
+
+       m = *mp;
+       KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
+
+       hoff = m->m_pkthdr.csum_lhlen;
+       iphlen = m->m_pkthdr.csum_iphlen;
+       thoff = m->m_pkthdr.csum_thlen;
+
+       KASSERT(hoff >= sizeof(struct ether_header),
+           ("invalid ether header len %d", hoff));
+       KASSERT(iphlen >= sizeof(struct ip),
+           ("invalid ip header len %d", iphlen));
+       KASSERT(thoff >= sizeof(struct tcphdr),
+           ("invalid tcp header len %d", thoff));
+
+       if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
+               m = m_pullup(m, hoff + iphlen + thoff);
+               if (m == NULL) {
+                       *mp = NULL;
+                       return ENOBUFS;
+               }
+               *mp = m;
+       }
+
+       /* Set the LSO flag in the TX BD */
+       flags = TX_BD_FLAGS_SW_LSO;
+
+       /* Set the length of IP + TCP options (in 32 bit words) */
+       flags |= (((iphlen + thoff -
+           sizeof(struct ip) - sizeof(struct tcphdr)) >> 2) << 8);
+
+       *mss0 = htole16(m->m_pkthdr.tso_segsz);
+       *flags0 = flags;
+
+       return 0;
+}

diff --git a/sys/dev/netif/bce/if_bcereg.h b/sys/dev/netif/bce/if_bcereg.h
index 88d9c6b..cd06cf1 100644 (file)
--- a/sys/dev/netif/bce/if_bcereg.h
+++ b/sys/dev/netif/bce/if_bcereg.h
@@ -5669,7 +5669,7 @@ struct l2_fhdr {
 #define TOTAL_TX_BD(sc)                (TOTAL_TX_BD_PER_PAGE * (sc)->tx_pages)
 #define USABLE_TX_BD(sc)       (USABLE_TX_BD_PER_PAGE * (sc)->tx_pages)
 #define MAX_TX_BD(sc)          (TOTAL_TX_BD((sc)) - 1)
-#define BCE_TX_SPARE_SPACE     5
+#define BCE_TX_SPARE_SPACE     33
 
 #define RX_PAGES_DEFAULT       2
 #define RX_PAGES_MAX           256
@@ -5800,7 +5800,7 @@ struct fw_info {
 
 #define BCE_TX_TIMEOUT         5
 
-#define BCE_MAX_SEGMENTS       32
+#define BCE_MAX_SEGMENTS       40
 #define BCE_DMA_ALIGN          8
 #define BCE_DMA_RX_ALIGN       16
 #define BCE_DMA_BOUNDARY       0
@@ -5819,16 +5819,16 @@ struct fw_info {
  * solution later.
  */
 #ifdef BCE_IP_CSUM
-#define BCE_IF_HWASSIST                (CSUM_IP | CSUM_TCP | CSUM_UDP)
+#define BCE_CSUM_FEATURES      (CSUM_IP | CSUM_TCP | CSUM_UDP)
 #else
-#define BCE_IF_HWASSIST                (CSUM_TCP | CSUM_UDP)
+#define BCE_CSUM_FEATURES      (CSUM_TCP | CSUM_UDP)
 #endif
 
 /* NOTE: This hardware also can do VLAN csum offload */
 #define BCE_IF_CAPABILITIES    (IFCAP_VLAN_MTU |       \
                                 IFCAP_VLAN_HWTAGGING | \
                                 IFCAP_HWCSUM |         \
-                                IFCAP_JUMBO_MTU)
+                                IFCAP_TSO)
 
 #define BCE_MIN_MTU                    60
 #define BCE_MIN_ETHER_MTU              64