myri10ge updates
Subsystems (2)
cover myri10ge updates
Hi Jeff,
No big change in myri10ge these days, mainly just a workaround for boards that were shipped with a bug in their firmware (patch #1, for 2.6.21).
Since things are pretty calm here, I am also resending the Large Receive Offload patch for inclusion (patch #2, maybe only for 2.6.22?). I didn't hear of anybody working on a generic LRO for the core stack, while our driver performance suffered from missing LRO for several months already. The code has been intensively tested by multiple customers.
Thanks,
Brice
PATCH 1/2 myri10ge: workaround buggy adopted firmwares
Work around a bug which occurs when adopting firmware versions 1.4.4 though 1.4.11 where broadcasts are filtered as if they were multicasts.
--- drivers/net/myri10ge/myri10ge.c | 27 +++++++++++++++++++++++---- 1 file changed, 23 insertions(+), 4 deletions(-)
drivers/net/myri10ge/myri10ge.c +23 -4
@@ -195,6 +195,10 @@char *fw_name;char eeprom_strings[MYRI10GE_EEPROM_STRINGS_SIZE ];char fw_version[128 ]; +int fw_ver_major; +int fw_ver_minor; +int fw_ver_tiny; +int adopted_rx_filter_bug;u8 mac_addr[6 ];/* eeprom mac address */ unsigned long serial_number;int vendor_specific_offset;
@@ -447,7 +451,6 @@ struct mcp_gen_header *hdr) {struct device *dev = &mgp->pdev->dev; -int major, minor;/* check firmware type */ if (ntohl (hdr->mcp_type) !=MCP_TYPE_ETH ) {
@@ -458,9 +461,11 @@/* save firmware version for ethtool */ strncpy (mgp->fw_version, hdr->version,sizeof (mgp->fw_version)); -sscanf (mgp->fw_version,"%d.%d" , &major, &minor); +sscanf (mgp->fw_version,"%d.%d.%d" , &mgp->fw_ver_major, + &mgp->fw_ver_minor, &mgp->fw_ver_tiny); -if (!(major ==MXGEFW_VERSION_MAJOR && minor ==MXGEFW_VERSION_MINOR )) { +if (!(mgp->fw_ver_major ==MXGEFW_VERSION_MAJOR + && mgp->fw_ver_minor ==MXGEFW_VERSION_MINOR )) {dev_err (dev,"Found firmware version %s\n" , mgp->fw_version);dev_err (dev,"Driver needs %d.%d\n" ,MXGEFW_VERSION_MAJOR ,MXGEFW_VERSION_MINOR );
@@ -561,6 +566,18 @@memcpy_fromio (hdr, mgp->sram + hdr_offset, bytes); status =myri10ge_validate_firmware (mgp, hdr);kfree (hdr); + +/* check to see if adopted firmware has bug where adopting +* it will cause broadcasts to be filtered unless the NIC +* is kept in ALLMULTI mode */ +if (mgp->fw_ver_major ==1 && mgp->fw_ver_minor ==4 && + mgp->fw_ver_tiny >=4 && mgp->fw_ver_tiny <=11 ) { + mgp->adopted_rx_filter_bug =1 ; +dev_warn (dev,"Adopting fw %d.%d.%d: " +"working around rx filter bug\n" , + mgp->fw_ver_major, mgp->fw_ver_minor, + mgp->fw_ver_tiny); + }return status; }
@@ -794,6 +811,8 @@ status =myri10ge_update_mac_address (mgp, mgp->dev->dev_addr);myri10ge_change_promisc (mgp,0 ,0 );myri10ge_change_pause (mgp, mgp->pause); +if (mgp->adopted_rx_filter_bug) + (void )myri10ge_send_cmd (mgp,MXGEFW_ENABLE_ALLMULTI , &cmd,1 );return status; }
@@ -2239,7 +2258,7 @@myri10ge_change_promisc (mgp, dev->flags &IFF_PROMISC ,1 );/* This firmware is known to not support multicast */ -if (!mgp->fw_multicast_support) +if (!mgp->fw_multicast_support || mgp->adopted_rx_filter_bug)return ;/* Disable multicast filtering */
PATCH 2/2 myri10ge: large receive offload
In reply to Brice GoglinAdd Large Receive Offload implemented in software.
--- drivers/net/myri10ge/myri10ge.c | 422 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 422 insertions(+)
drivers/net/myri10ge/myri10ge.c +422 -0
@@ -61,6 +61,8 @@#include <linux/moduleparam.h> #include <linux/io.h> #include <net/checksum.h> +#include <net/ip.h> +#include <net/tcp.h> #include <asm/byteorder.h> #include <asm/io.h> #include <asm/processor.h>
@@ -145,11 +147,32 @@int pkt_done;/* packets completed */ }; +struct myri10ge_lro_packet { +struct hlist_node lro_node; +struct sk_buff *skb; +int timestamp; +__u32 tsval; +__u32 tsecr; +__u32 source_ip; +__u32 dest_ip; +__u32 next_seq; +__u32 ack_seq; +__wsum data_csum; +__u16 window; +__u16 source_port; +__u16 dest_port; +__u16 append_cnt; +__u16 mss; +__u16 vlan_tci; +}; +struct myri10ge_rx_done {struct mcp_slot *entry;dma_addr_t bus;int cnt;int idx; +struct hlist_head lro_active; +struct hlist_head lro_free; };struct myri10ge_priv {
@@ -161,6 +184,10 @@struct myri10ge_rx_done rx_done;int small_bytes;int big_bytes; +int lro_flushed; +int lro_queued; +int lro_too_many_streams; +int lro_bad_csum;struct net_device *dev;struct net_device_stats stats;u8 __iomem *sram;
@@ -274,6 +301,10 @@module_param (myri10ge_debug, int,0 );MODULE_PARM_DESC (myri10ge_debug,"Debug level (0=none,...,16=all)" ); +static int myri10ge_lro =8 ; +module_param (myri10ge_lro, int,S_IRUGO ); +MODULE_PARM_DESC (myri10ge_lro,"Enable large N receive offload queues\n" ); +static int myri10ge_fill_thresh =256 ;module_param (myri10ge_fill_thresh, int,S_IRUGO |S_IWUSR );MODULE_PARM_DESC (myri10ge_fill_thresh,"Number of empty rx slots allowed\n" );
@@ -808,6 +839,9 @@ mgp->rx_done.idx =0 ; mgp->rx_done.cnt =0 ; mgp->link_changes =0 ; + mgp->lro_queued =0 ; + mgp->lro_flushed =0 ; + mgp->lro_too_many_streams =0 ; status =myri10ge_update_mac_address (mgp, mgp->dev->dev_addr);myri10ge_change_promisc (mgp,0 ,0 );myri10ge_change_pause (mgp, mgp->pause);
@@ -876,6 +910,357 @@skb_pull (skb,MXGEFW_PAD ); } +/* debug aid to check for "bad" hardware */ + +static void +myri10ge_frag_trim (struct skb_frag_struct *rx_frags,int old_len,int trim) +{ +struct skb_frag_struct *frag; +int offset =0 ; +int new_len = old_len - trim; +int old_size; + +/* find the frag where the IP payload ends. This +* should almost always be the 1st fragment */ + frag = rx_frags; +while (offset + frag->size < new_len) { + offset += frag->size; + frag++; + } +/* adjust its length */ + old_size = frag->size; + frag->size = new_len - offset; + +/* release any excess pages */ + offset += old_size; +while (offset < old_len) { + frag++; + offset += frag->size; +put_page (frag->page); + } +} + +static inline int myri10ge_lro_csum (int tcplen,struct iphdr *iph,__wsum csum) +{ +if (unlikely (ip_fast_csum ((u8 *) iph, iph->ihl))) +return -1 ; + +if (unlikely (csum_tcpudp_magic (iph->saddr, iph->daddr, + tcplen,IPPROTO_TCP , csum))) +return -1 ; +return 0 ; +} + +static inline void +myri10ge_lro_flush (struct myri10ge_priv *mgp,struct myri10ge_lro_packet *lro) +{ +struct iphdr *iph; +struct tcphdr *th; +struct sk_buff *skb; +u32 *ts_ptr; +u32 tcplen; + + skb = lro->skb; + +if (lro->append_cnt) { +/* incorporate the new len into the ip header and +* re-calculate the checksum, Note that +* eth_type_trans() left skb->data at the start of +* the vlan header, so we need to skip past it to +* get to the IP header */ +if (lro->vlan_tci) { + iph = (struct iphdr *)(skb->data +VLAN_HLEN ); + iph->tot_len =htons (skb->len -VLAN_HLEN ); + }else { + iph = (struct iphdr *)skb->data; + iph->tot_len =htons (skb->len); + } + iph->check =0 ; + iph->check =ip_fast_csum ((unsigned char *)iph, iph->ihl); + +/* incorporate the latest ack into the tcp header */ + th = (struct tcphdr *)(iph +1 ); + th->ack_seq = lro->ack_seq; + th->window = lro->window; + +/* incorporate latest timestamp into the tcp header */ +if (lro->timestamp) { + ts_ptr = (u32 *) (th +1 ); + ts_ptr[1 ] =htonl (lro->tsval); + ts_ptr[2 ] = lro->tsecr; + } + +/* +* update checksum in tcp header by re-calculating the +* tcp pseudoheader checksum, and adding it to the checksum +* of the tcp payload data +*/ + th->check =0 ; + tcplen =ntohs (iph->tot_len) -sizeof (*iph); + th->check =tcp_v4_check (tcplen, iph->saddr, iph->daddr, +csum_partial ((char *)th, + th->doff <<2 , + lro->data_csum)); + + skb->truesize = skb->len +sizeof (struct sk_buff ); + } + +skb_shinfo (skb)->gso_size = lro->mss; +netif_receive_skb (skb); + mgp->dev->last_rx = jiffies; + mgp->lro_queued += lro->append_cnt +1 ; + mgp->lro_flushed++; + lro->skb =NULL ; + lro->timestamp =0 ; + lro->append_cnt =0 ; +hlist_add_head (&lro->lro_node, &mgp->rx_done.lro_free); +} + +static int +myri10ge_lro_rx (struct myri10ge_priv *mgp,u8 * va, +struct skb_frag_struct *rx_frags,int *len,__wsum csum) +{ +struct ethhdr *eh; +struct vlan_ethhdr *vh; +struct iphdr *iph; +struct tcphdr *th; +struct myri10ge_lro_packet *lro; +u32 *ts_ptr =NULL ;/* XXX -Wuninitialized */ +struct sk_buff *skb; +struct skb_frag_struct *skb_frags; +struct hlist_node *node; +int opt_bytes, tcp_data_len, tcp_hdr_len, hlen, trim, llhlen; +__u32 seq; +__u16 ip_len, vlan_tci; + +/* check to see that it is IP */ + eh = (struct ethhdr *)(va +MXGEFW_PAD ); +if (eh->h_proto ==ntohs (ETH_P_IP )) { + llhlen =ETH_HLEN ; + vlan_tci =0 ; + }else if (eh->h_proto ==ntohs (ETH_P_8021Q )) { + vh = (struct vlan_ethhdr *)(va +MXGEFW_PAD ); +if (vh->h_vlan_encapsulated_proto !=ntohs (ETH_P_IP )) +return -1 ; + llhlen =VLAN_ETH_HLEN ; + vlan_tci = vh->h_vlan_TCI; +/* HW checksum starts after the ethernet header, we +* must subtract off the VLAN header's checksum before +* csum can be used */ + csum =csum_sub (csum, +csum_partial (va +MXGEFW_PAD +ETH_HLEN , +VLAN_HLEN ,0 )); + }else { +return -1 ; + } + +/* now check to see if it is TCP */ + iph = (struct iphdr *)(va + llhlen +MXGEFW_PAD ); +if (iph->protocol !=IPPROTO_TCP ) +return -1 ; + +/* ensure there are no options */ +if ((iph->ihl <<2 ) !=sizeof (*iph)) +return -1 ; + +/* .. and the packet is not fragmented */ +if (iph->frag_off &htons (IP_MF |IP_OFFSET )) +return -1 ; + +/* find the TCP header */ + th = (struct tcphdr *)(iph +1 ); + +/* ensure no bits set besides ack or psh */ +if (th->fin || th->syn || th->rst || th->urg || th->ece + || th->cwr || !th->ack) +return -1 ; + +/* check for timestamps. Since the only option we handle are +* timestamps, we only have to handle the simple case of +* aligned timestamps */ + + opt_bytes = (th->doff <<2 ) -sizeof (*th); + tcp_hdr_len =sizeof (*th) + opt_bytes; +if (opt_bytes !=0 ) { + ts_ptr = (u32 *) (th +1 ); +if (unlikely (opt_bytes !=TCPOLEN_TSTAMP_ALIGNED ) || + (*ts_ptr !=ntohl ((TCPOPT_NOP <<24 ) | (TCPOPT_NOP <<16 ) + | (TCPOPT_TIMESTAMP <<8 ) + |TCPOLEN_TIMESTAMP ))) { +return -1 ; + } + } + + ip_len =ntohs (iph->tot_len); + tcp_data_len = ip_len - (th->doff <<2 ) -sizeof (*iph); + +/* +* If frame is padded beyond the end of the IP packet, +* then we must trim the extra bytes off the end. +*/ + trim = *len - (ip_len + llhlen +MXGEFW_PAD ); +if (trim !=0 ) { +/* ensure we received the full frame */ +if (unlikely (trim <0 )) +return -1 ; +/* trim off any padding */ +myri10ge_frag_trim (rx_frags, *len, trim); + *len -= trim; + } + + hlen = ip_len + llhlen - tcp_data_len; + + seq =ntohl (th->seq); + +if (unlikely (myri10ge_lro_csum (tcp_hdr_len + tcp_data_len, iph, csum))) { + mgp->lro_bad_csum++; +return -1 ; + } + +/* now we have a packet that might be eligible for LRO, +* so see if it matches anything we might expect */ + +hlist_for_each_entry (lro, node, &mgp->rx_done.lro_active, lro_node) { +if (lro->source_port == th->source && + lro->dest_port == th->dest && + lro->source_ip == iph->saddr && + lro->dest_ip == iph->daddr && lro->vlan_tci == vlan_tci) { +/* Try to append it */ + +if (unlikely (seq != lro->next_seq)) { +/* out of order packet */ +hlist_del (&lro->lro_node); +myri10ge_lro_flush (mgp, lro); +return -1 ; + } +if (lro->timestamp) { +__u32 tsval =ntohl (*(ts_ptr +1 )); +/* make sure timestamp values are increasing */ +if (unlikely (lro->tsval > tsval || + *(ts_ptr +2 ) ==0 )) { +return -1 ; + } + lro->tsval = tsval; + lro->tsecr = *(ts_ptr +2 ); + } + lro->next_seq += tcp_data_len; + lro->ack_seq = th->ack_seq; + skb = lro->skb; + +/* subtract off the checksum of the tcp header +* from the hardware checksum, and add it to the +* stored tcp data checksum. csum_block_add() +* is used, as the total length so far may be +* odd +*/ + lro->data_csum = +csum_block_add (lro->data_csum, +csum_sub (csum, +csum_partial ((u8 *) th, + tcp_hdr_len, +0 )), + skb->data_len); + lro->window = th->window; + skb->data_len += tcp_data_len; + skb->len += tcp_data_len; +if (tcp_data_len > lro->mss) + lro->mss = tcp_data_len; + +/* pull off the header and firmware pad +* before we copy the data */ + + hlen +=MXGEFW_PAD ; + rx_frags[0 ].page_offset += hlen; + rx_frags[0 ].size -= hlen; + *len -= hlen; + skb_frags = + &skb_shinfo (skb)->frags[skb_shinfo (skb)->nr_frags]; +/* if it was just header (like a TCP ack with +* no data), release the page */ +if (*len <=0 ) { +put_page (rx_frags[0 ].page); + }else { +while (*len >0 ) { +memcpy (skb_frags, rx_frags, +sizeof (*skb_frags)); + *len -= rx_frags->size; + rx_frags++; + skb_frags++; +skb_shinfo (skb)->nr_frags++; + } + } + + lro->append_cnt++; + +/* cheap, conservative test. We may waste +* some slots with a 1500 byte mtu */ +if (skb_shinfo (skb)->nr_frags + +MYRI10GE_MAX_FRAGS_PER_FRAME >MAX_SKB_FRAGS + || mgp->dev->mtu + skb->len >65535 ) { +hlist_del (&lro->lro_node); +myri10ge_lro_flush (mgp, lro); + } +return 0 ; + } + } + +/* start a new packet */ +if (!hlist_empty (&mgp->rx_done.lro_free)) { + lro =hlist_entry (mgp->rx_done.lro_free.first, + struct myri10ge_lro_packet, lro_node); +/* allocate an skb to attach the page(s) to */ + + skb =netdev_alloc_skb (mgp->dev, hlen +16 ); +if (unlikely (skb ==NULL )) +return -1 ; + +myri10ge_rx_skb_build (skb, va, rx_frags, *len, + hlen +MXGEFW_PAD ); + skb->protocol =eth_type_trans (skb, mgp->dev); + skb->dev = mgp->dev; + skb->ip_summed =CHECKSUM_UNNECESSARY ; + lro->skb = skb; + lro->source_ip = iph->saddr; + lro->dest_ip = iph->daddr; + lro->source_port = th->source; + lro->dest_port = th->dest; + lro->next_seq = seq + tcp_data_len; + lro->mss = tcp_data_len; + lro->ack_seq = th->ack_seq; + +/* save the checksum of just the TCP payload by +* subtracting off the checksum of the TCP header from +* the entire hardware checksum +*/ + lro->data_csum =csum_sub (csum, +csum_partial ((u8 *) th, + tcp_hdr_len,0 )); + lro->window = th->window; + lro->vlan_tci = vlan_tci; +/* record timestamp if it is present */ +if (opt_bytes) { + lro->timestamp =1 ; + lro->tsval =ntohl (*(ts_ptr +1 )); + lro->tsecr = *(ts_ptr +2 ); + } +/* remove first packet from freelist.. */ +hlist_del (&lro->lro_node); +/* .. and insert at the front of the active list */ +hlist_add_head (&lro->lro_node, &mgp->rx_done.lro_active); + +/* release the page if there was no data. We do it +* down here since the code above refers to the +* contents of the page */ +if (skb_shinfo (skb)->frags[0 ].size <=0 ) { +put_page (skb_shinfo (skb)->frags[0 ].page); +skb_shinfo (skb)->nr_frags =0 ; + } +return 0 ; + } + mgp->lro_too_many_streams++; +return -1 ; +} +static void myri10ge_alloc_rx_pages (struct myri10ge_priv *mgp,struct myri10ge_rx_buf *rx,int bytes,int watchdog)
@@ -983,9 +1368,14 @@ remainder -=MYRI10GE_ALLOC_SIZE ; } +if (mgp->csum_flag && myri10ge_lro && + (0 ==myri10ge_lro_rx (mgp, va, rx_frags, &len, csum))) +return 1 ; hlen =MYRI10GE_HLEN > len ? len :MYRI10GE_HLEN ;/* allocate an skb to attach the page(s) to. */ +/* This is done +* after trying LRO, so as to avoid skb allocation overheads */ skb =netdev_alloc_skb (dev,MYRI10GE_HLEN +16 );if (unlikely (skb ==NULL )) {
@@ -1073,6 +1463,8 @@static inline void myri10ge_clean_rx_done (struct myri10ge_priv *mgp,int *limit) {struct myri10ge_rx_done *rx_done = &mgp->rx_done; +struct hlist_node *node, *node2; +struct myri10ge_lro_packet *lro;unsigned long rx_bytes =0 ;unsigned long rx_packets =0 ;unsigned long rx_ok;
@@ -1105,6 +1497,11 @@ } rx_done->idx = idx; rx_done->cnt = cnt; +hlist_for_each_entry_safe (lro, node, node2, &mgp->rx_done.lro_active, + lro_node) { +hlist_del (&lro->lro_node); +myri10ge_lro_flush (mgp, lro); + } mgp->stats.rx_packets += rx_packets; mgp->stats.rx_bytes += rx_bytes;
@@ -1338,6 +1735,7 @@"read_dma_bw_MBs" ,"write_dma_bw_MBs" ,"read_write_dma_bw_MBs" ,"serial_number" ,"tx_pkt_start" ,"tx_pkt_done" ,"tx_req" ,"tx_done" ,"rx_small_cnt" ,"rx_big_cnt" , +"lro_queued" ,"lro_flushed" ,"lro_too_many_streams" ,"lro_bad_csum" ,"wake_queue" ,"stop_queue" ,"watchdog_resets" ,"tx_linearized" ,"link_changes" ,"link_up" ,"dropped_link_overflow" , "dropped_link_error_or_filtered", "dropped_multicast_filtered",
@@ -1388,6 +1786,10 @@ data[i++] = (unsigned int )mgp->tx.done; data[i++] = (unsigned int )mgp->rx_small.cnt; data[i++] = (unsigned int )mgp->rx_big.cnt; + data[i++] = (unsigned int )mgp->lro_queued; + data[i++] = (unsigned int )mgp->lro_flushed; + data[i++] = (unsigned int )mgp->lro_too_many_streams; + data[i++] = (unsigned int )mgp->lro_bad_csum; data[i++] = (unsigned int )mgp->wake_queue; data[i++] = (unsigned int )mgp->stop_queue; data[i++] = (unsigned int )mgp->watchdog_resets;
@@ -1527,6 +1929,18 @@ goto abort_with_rx_big_ring; } + bytes =sizeof (struct myri10ge_lro_packet ); +INIT_HLIST_HEAD (&mgp->rx_done.lro_free); +INIT_HLIST_HEAD (&mgp->rx_done.lro_active); +for (i =0 ; i < myri10ge_lro; i++) { +struct myri10ge_lro_packet *lro; + lro =kzalloc (bytes,GFP_KERNEL ); +if (lro !=NULL ) { +INIT_HLIST_NODE (&lro->lro_node); +hlist_add_head (&lro->lro_node, &mgp->rx_done.lro_free); + } + } +return 0 ; abort_with_rx_big_ring:
@@ -1573,10 +1987,18 @@struct myri10ge_priv *mgp;struct sk_buff *skb;struct myri10ge_tx_buf *tx; +struct hlist_node *node, *node2; +struct myri10ge_lro_packet *lro;int i, len, idx; mgp =netdev_priv (dev); +hlist_for_each_entry_safe (lro, node, node2, &mgp->rx_done.lro_active, + lro_node) { +hlist_del (&lro->lro_node); +kfree (lro); + } +for (i = mgp->rx_big.cnt; i < mgp->rx_big.fill_cnt; i++) { idx = i & mgp->rx_big.mask;if (i == mgp->rx_big.fill_cnt -1 )
[PATCH 3/2] myri10ge: fix copyright and license
In reply to Brice GoglinFix copyright and license ("regents" should not have ever been used).
--- drivers/net/myri10ge/myri10ge.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)
drivers/net/myri10ge/myri10ge.c +11 -11
@@ -1,7 +1,7 @@/************************************************************************* * myri10ge.c: Myricom Myri-10G Ethernet driver. * -* Copyright (C) 2005, 2006 Myricom, Inc. +* Copyright (C) 2005 - 2007 Myricom, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without
@@ -16,17 +16,17 @@* may be used to endorse or promote products derived from this software * without specific prior written permission. * -* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND -* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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 REGENTS 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. +* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER 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. * * * If the eeprom on your board is not recent enough, you will need to get a
Quoted 6 lines · Brice Goglin
Hi Jeff,
No big change in myri10ge these days, mainly just a workaround for boards that were shipped with a bug in their firmware (patch #1, for 2.6.21).
Quoted 9 lines
Since things are pretty calm here, I am also resending the Large Receive Offload patch for inclusion (patch #2, maybe only for 2.6.22?). I didn't hear of anybody working on a generic LRO for the core stack, while our driver performance suffered from missing LRO for several months already. The code has been intensively tested by multiple customers.
Thanks,
Brice
I just got a last-minute copyright/license fix. I sent it as patch #3. Please apply this one for 2.6.21.
thanks,
Brice
Re: [PATCH 2/2] myri10ge: large receive offload
In reply to Brice GoglinQuoted 6 lines · Brice Goglin
Add Large Receive Offload implemented in software.
Signed-off-by Brice Goglin <[email protected]>--- drivers/net/myri10ge/myri10ge.c | 422 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 422 insertions(+)
Another patch to do this for another driver just came up as well. I would rather see some helpers in the net core make this easier on the NIC author, rather than reimplementing the same concept over and over again.
Re: [PATCH 3/2] myri10ge: fix copyright and license
In reply to Brice GoglinQuoted 6 lines · Brice Goglin
Fix copyright and license ("regents" should not have ever been used).
Signed-off-by Brice Goglin <[email protected]>--- drivers/net/myri10ge/myri10ge.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)
ACK, but dropped since patch #2 was dropped
Jeff Garzik
AppliedRe: [PATCH 1/2] myri10ge: workaround buggy adopted firmwares
In reply to Brice GoglinQuoted 8 lines · Brice Goglin
Work around a bug which occurs when adopting firmware versions 1.4.4 though 1.4.11 where broadcasts are filtered as if they were multicasts.
Signed-off-by Brice Goglin <[email protected]>--- drivers/net/myri10ge/myri10ge.c | 27 +++++++++++++++++++++++---- 1 file changed, 23 insertions(+), 4 deletions(-)
applied