Index: sys/dev/e1000/if_em.h =================================================================== --- sys/dev/e1000/if_em.h +++ sys/dev/e1000/if_em.h @@ -266,7 +266,7 @@ #define HW_DEBUGOUT1(S, A) if (DEBUG_HW) printf(S "\n", A) #define HW_DEBUGOUT2(S, A, B) if (DEBUG_HW) printf(S "\n", A, B) -#define EM_MAX_SCATTER 32 +#define EM_MAX_SCATTER 64 #define EM_VFTA_SIZE 128 #define EM_TSO_SIZE (65535 + sizeof(struct ether_vlan_header)) #define EM_TSO_SEG_SIZE 4096 /* Max dma segment size */ Index: sys/dev/e1000/if_em.c =================================================================== --- sys/dev/e1000/if_em.c +++ sys/dev/e1000/if_em.c @@ -1875,10 +1875,11 @@ u32 txd_upper = 0, txd_lower = 0, txd_used = 0; int ip_off, poff; int nsegs, i, j, first, last = 0; - int error, do_tso, tso_desc = 0, remap = 1; + int error, tso_desc = 0, remap = 1; + bool do_tso; m_head = *m_headp; - do_tso = ((m_head->m_pkthdr.csum_flags & CSUM_TSO) != 0); + do_tso = (m_head->m_pkthdr.csum_flags & CSUM_TSO); ip_off = poff = 0; /* @@ -1914,74 +1915,77 @@ * for IPv6 yet. */ ip_off = sizeof(struct ether_header); - m_head = m_pullup(m_head, ip_off); - if (m_head == NULL) { - *m_headp = NULL; - return (ENOBUFS); + if (m_head->m_len < ip_off) { + m_head = m_pullup(m_head, ip_off); + if (m_head == NULL) { + *m_headp = NULL; + return (ENOBUFS); + } } eh = mtod(m_head, struct ether_header *); if (eh->ether_type == htons(ETHERTYPE_VLAN)) { ip_off = sizeof(struct ether_vlan_header); - m_head = m_pullup(m_head, ip_off); + if (m_head->m_len < ip_off) { + m_head = m_pullup(m_head, ip_off); + if (m_head == NULL) { + *m_headp = NULL; + return (ENOBUFS); + } + } + } + if (m_head->m_len < ip_off + sizeof(struct ip)) { + m_head = m_pullup(m_head, ip_off + sizeof(struct ip)); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } } - m_head = m_pullup(m_head, ip_off + sizeof(struct ip)); - if (m_head == NULL) { - *m_headp = NULL; - return (ENOBUFS); - } ip = (struct ip *)(mtod(m_head, char *) + ip_off); poff = ip_off + (ip->ip_hl << 2); - if (do_tso) { - m_head = m_pullup(m_head, poff + sizeof(struct tcphdr)); - if (m_head == NULL) { - *m_headp = NULL; - return (ENOBUFS); + + if (do_tso || (m_head->m_pkthdr.csum_flags & CSUM_TCP)) { + if (m_head->m_len < poff + sizeof(struct tcphdr)) { + m_head = m_pullup(m_head, poff + + sizeof(struct tcphdr)); + if (m_head == NULL) { + *m_headp = NULL; + return (ENOBUFS); + } } tp = (struct tcphdr *)(mtod(m_head, char *) + poff); - /* - * TSO workaround: - * pull 4 more bytes of data into it. - */ - m_head = m_pullup(m_head, poff + (tp->th_off << 2) + 4); - if (m_head == NULL) { - *m_headp = NULL; - return (ENOBUFS); + if (m_head->m_len < poff + (tp->th_off << 2)) { + m_head = m_pullup(m_head, poff + + (tp->th_off << 2)); + if (m_head == NULL) { + *m_headp = NULL; + return (ENOBUFS); + } } ip = (struct ip *)(mtod(m_head, char *) + ip_off); - ip->ip_len = 0; - ip->ip_sum = 0; - /* - * The pseudo TCP checksum does not include TCP payload - * length so driver should recompute the checksum here - * what hardware expect to see. This is adherence of - * Microsoft's Large Send specification. - */ - tp = (struct tcphdr *)(mtod(m_head, char *) + poff); - tp->th_sum = in_pseudo(ip->ip_src.s_addr, - ip->ip_dst.s_addr, htons(IPPROTO_TCP)); - } else if (m_head->m_pkthdr.csum_flags & CSUM_TCP) { - m_head = m_pullup(m_head, poff + sizeof(struct tcphdr)); - if (m_head == NULL) { - *m_headp = NULL; - return (ENOBUFS); - } tp = (struct tcphdr *)(mtod(m_head, char *) + poff); - m_head = m_pullup(m_head, poff + (tp->th_off << 2)); - if (m_head == NULL) { - *m_headp = NULL; - return (ENOBUFS); + if (do_tso) { + ip->ip_len = htons(m_head->m_pkthdr.tso_segsz + + (ip->ip_hl << 2) + + (tp->th_off << 2)); + ip->ip_sum = 0; + /* + * The pseudo TCP checksum does not include TCP + * payload length so driver should recompute + * the checksum here what hardware expect to + * see. This is adherence of Microsoft's Large + * Send specification. + */ + tp->th_sum = in_pseudo(ip->ip_src.s_addr, + ip->ip_dst.s_addr, htons(IPPROTO_TCP)); } - ip = (struct ip *)(mtod(m_head, char *) + ip_off); - tp = (struct tcphdr *)(mtod(m_head, char *) + poff); } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) { - m_head = m_pullup(m_head, poff + sizeof(struct udphdr)); - if (m_head == NULL) { - *m_headp = NULL; - return (ENOBUFS); + if (m_head->m_len < poff + sizeof(struct udphdr)) { + m_head = m_pullup(m_head, poff + + sizeof(struct udphdr)); + if (m_head == NULL) { + *m_headp = NULL; + return (ENOBUFS); + } } ip = (struct ip *)(mtod(m_head, char *) + ip_off); } @@ -2029,9 +2033,6 @@ /* Try it again, but only once */ remap = 0; goto retry; - } else if (error == ENOMEM) { - adapter->no_tx_dma_setup++; - return (error); } else if (error != 0) { adapter->no_tx_dma_setup++; m_freem(*m_headp); @@ -2045,13 +2046,13 @@ * it follows a TSO burst, then we need to add a * sentinel descriptor to prevent premature writeback. */ - if ((do_tso == 0) && (txr->tx_tso == TRUE)) { + if ((!do_tso) && (txr->tx_tso == TRUE)) { if (nsegs == 1) tso_desc = TRUE; txr->tx_tso = FALSE; } - if (nsegs > (txr->tx_avail - 2)) { + if (nsegs > (txr->tx_avail - EM_MAX_SCATTER)) { txr->no_desc_avail++; bus_dmamap_unload(txr->txtag, map); return (ENOBUFS); @@ -2091,34 +2092,31 @@ ** If this is the last descriptor, we want to ** split it so we have a small final sentinel */ - if (tso_desc && (j == (nsegs -1)) && (seg_len > 8)) { + if (tso_desc && (j == (nsegs - 1)) && (seg_len > 8)) { seg_len -= 4; ctxd->buffer_addr = htole64(seg_addr); ctxd->lower.data = htole32( - adapter->txd_cmd | txd_lower | seg_len); - ctxd->upper.data = - htole32(txd_upper); + adapter->txd_cmd | txd_lower | seg_len); + ctxd->upper.data = htole32(txd_upper); if (++i == adapter->num_tx_desc) i = 0; /* Now make the sentinel */ ++txd_used; /* using an extra txd */ ctxd = &txr->tx_base[i]; tx_buffer = &txr->tx_buffers[i]; - ctxd->buffer_addr = - htole64(seg_addr + seg_len); + + ctxd->buffer_addr = htole64(seg_addr + seg_len); ctxd->lower.data = htole32( - adapter->txd_cmd | txd_lower | 4); - ctxd->upper.data = - htole32(txd_upper); + adapter->txd_cmd | txd_lower | 4); + ctxd->upper.data = htole32(txd_upper); last = i; if (++i == adapter->num_tx_desc) i = 0; } else { ctxd->buffer_addr = htole64(seg_addr); ctxd->lower.data = htole32( - adapter->txd_cmd | txd_lower | seg_len); - ctxd->upper.data = - htole32(txd_upper); + adapter->txd_cmd | txd_lower | seg_len); + ctxd->upper.data = htole32(txd_upper); last = i; if (++i == adapter->num_tx_desc) i = 0; @@ -3033,6 +3031,11 @@ if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST); if_setioctlfn(ifp, em_ioctl); if_setgetcounterfn(ifp, em_get_counter); + /* TSO parameters */ + ifp->if_hw_tsomax = EM_TSO_SIZE; + ifp->if_hw_tsomaxsegcount = EM_MAX_SCATTER; + ifp->if_hw_tsomaxsegsize = EM_TSO_SEG_SIZE; + #ifdef EM_MULTIQUEUE /* Multiqueue stack interface */ if_settransmitfn(ifp, em_mq_start); @@ -4891,8 +4894,8 @@ u32 ims_mask = IMS_ENABLE_MASK; if (hw->mac.type == e1000_82574) { - E1000_WRITE_REG(hw, EM_EIAC, EM_MSIX_MASK); - ims_mask |= EM_MSIX_MASK; + E1000_WRITE_REG(hw, EM_EIAC, adapter->ims); + ims_mask |= adapter->ims; } E1000_WRITE_REG(hw, E1000_IMS, ims_mask); }