10d160211SJeremy Fitzhardinge /* 20d160211SJeremy Fitzhardinge * Virtual network driver for conversing with remote driver backends. 30d160211SJeremy Fitzhardinge * 40d160211SJeremy Fitzhardinge * Copyright (c) 2002-2005, K A Fraser 50d160211SJeremy Fitzhardinge * Copyright (c) 2005, XenSource Ltd 60d160211SJeremy Fitzhardinge * 70d160211SJeremy Fitzhardinge * This program is free software; you can redistribute it and/or 80d160211SJeremy Fitzhardinge * modify it under the terms of the GNU General Public License version 2 90d160211SJeremy Fitzhardinge * as published by the Free Software Foundation; or, when distributed 100d160211SJeremy Fitzhardinge * separately from the Linux kernel or incorporated into other 110d160211SJeremy Fitzhardinge * software packages, subject to the following license: 120d160211SJeremy Fitzhardinge * 130d160211SJeremy Fitzhardinge * Permission is hereby granted, free of charge, to any person obtaining a copy 140d160211SJeremy Fitzhardinge * of this source file (the "Software"), to deal in the Software without 150d160211SJeremy Fitzhardinge * restriction, including without limitation the rights to use, copy, modify, 160d160211SJeremy Fitzhardinge * merge, publish, distribute, sublicense, and/or sell copies of the Software, 170d160211SJeremy Fitzhardinge * and to permit persons to whom the Software is furnished to do so, subject to 180d160211SJeremy Fitzhardinge * the following conditions: 190d160211SJeremy Fitzhardinge * 200d160211SJeremy Fitzhardinge * The above copyright notice and this permission notice shall be included in 210d160211SJeremy Fitzhardinge * all copies or substantial portions of the Software. 220d160211SJeremy Fitzhardinge * 230d160211SJeremy Fitzhardinge * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 240d160211SJeremy Fitzhardinge * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 250d160211SJeremy Fitzhardinge * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 260d160211SJeremy Fitzhardinge * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 270d160211SJeremy Fitzhardinge * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 280d160211SJeremy Fitzhardinge * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 290d160211SJeremy Fitzhardinge * IN THE SOFTWARE. 300d160211SJeremy Fitzhardinge */ 310d160211SJeremy Fitzhardinge 320d160211SJeremy Fitzhardinge #include <linux/module.h> 330d160211SJeremy Fitzhardinge #include <linux/kernel.h> 340d160211SJeremy Fitzhardinge #include <linux/netdevice.h> 350d160211SJeremy Fitzhardinge #include <linux/etherdevice.h> 360d160211SJeremy Fitzhardinge #include <linux/skbuff.h> 370d160211SJeremy Fitzhardinge #include <linux/ethtool.h> 380d160211SJeremy Fitzhardinge #include <linux/if_ether.h> 399ecd1a75SWei Liu #include <net/tcp.h> 400d160211SJeremy Fitzhardinge #include <linux/udp.h> 410d160211SJeremy Fitzhardinge #include <linux/moduleparam.h> 420d160211SJeremy Fitzhardinge #include <linux/mm.h> 435a0e3ad6STejun Heo #include <linux/slab.h> 440d160211SJeremy Fitzhardinge #include <net/ip.h> 450d160211SJeremy Fitzhardinge 46ca981633SStefano Stabellini #include <asm/xen/page.h> 471ccbf534SJeremy Fitzhardinge #include <xen/xen.h> 480d160211SJeremy Fitzhardinge #include <xen/xenbus.h> 490d160211SJeremy Fitzhardinge #include <xen/events.h> 500d160211SJeremy Fitzhardinge #include <xen/page.h> 51b9136d20SIgor Mammedov #include <xen/platform_pci.h> 520d160211SJeremy Fitzhardinge #include <xen/grant_table.h> 530d160211SJeremy Fitzhardinge 540d160211SJeremy Fitzhardinge #include <xen/interface/io/netif.h> 550d160211SJeremy Fitzhardinge #include <xen/interface/memory.h> 560d160211SJeremy Fitzhardinge #include <xen/interface/grant_table.h> 570d160211SJeremy Fitzhardinge 580fc0b732SStephen Hemminger static const struct ethtool_ops xennet_ethtool_ops; 590d160211SJeremy Fitzhardinge 600d160211SJeremy Fitzhardinge struct netfront_cb { 613683243bSIan Campbell int pull_to; 620d160211SJeremy Fitzhardinge }; 630d160211SJeremy Fitzhardinge 640d160211SJeremy Fitzhardinge #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) 650d160211SJeremy Fitzhardinge 660d160211SJeremy Fitzhardinge #define RX_COPY_THRESHOLD 256 670d160211SJeremy Fitzhardinge 680d160211SJeremy Fitzhardinge #define GRANT_INVALID_REF 0 690d160211SJeremy Fitzhardinge 70667c78afSJeremy Fitzhardinge #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE) 71667c78afSJeremy Fitzhardinge #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE) 7240206dd9SWei Liu #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256) 730d160211SJeremy Fitzhardinge 74e00f85beSstephen hemminger struct netfront_stats { 75e00f85beSstephen hemminger u64 rx_packets; 76e00f85beSstephen hemminger u64 tx_packets; 77e00f85beSstephen hemminger u64 rx_bytes; 78e00f85beSstephen hemminger u64 tx_bytes; 79e00f85beSstephen hemminger struct u64_stats_sync syncp; 80e00f85beSstephen hemminger }; 81e00f85beSstephen hemminger 820d160211SJeremy Fitzhardinge struct netfront_info { 830d160211SJeremy Fitzhardinge struct list_head list; 840d160211SJeremy Fitzhardinge struct net_device *netdev; 850d160211SJeremy Fitzhardinge 86bea3348eSStephen Hemminger struct napi_struct napi; 870d160211SJeremy Fitzhardinge 8884284d3cSJeremy Fitzhardinge unsigned int evtchn; 8984284d3cSJeremy Fitzhardinge struct xenbus_device *xbdev; 900d160211SJeremy Fitzhardinge 910d160211SJeremy Fitzhardinge spinlock_t tx_lock; 9284284d3cSJeremy Fitzhardinge struct xen_netif_tx_front_ring tx; 9384284d3cSJeremy Fitzhardinge int tx_ring_ref; 940d160211SJeremy Fitzhardinge 950d160211SJeremy Fitzhardinge /* 960d160211SJeremy Fitzhardinge * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries 970d160211SJeremy Fitzhardinge * are linked from tx_skb_freelist through skb_entry.link. 980d160211SJeremy Fitzhardinge * 990d160211SJeremy Fitzhardinge * NB. Freelist index entries are always going to be less than 1000d160211SJeremy Fitzhardinge * PAGE_OFFSET, whereas pointers to skbs will always be equal or 1010d160211SJeremy Fitzhardinge * greater than PAGE_OFFSET: we use this property to distinguish 1020d160211SJeremy Fitzhardinge * them. 1030d160211SJeremy Fitzhardinge */ 1040d160211SJeremy Fitzhardinge union skb_entry { 1050d160211SJeremy Fitzhardinge struct sk_buff *skb; 1061ffb40b8SIsaku Yamahata unsigned long link; 1070d160211SJeremy Fitzhardinge } tx_skbs[NET_TX_RING_SIZE]; 1080d160211SJeremy Fitzhardinge grant_ref_t gref_tx_head; 1090d160211SJeremy Fitzhardinge grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; 1100d160211SJeremy Fitzhardinge unsigned tx_skb_freelist; 1110d160211SJeremy Fitzhardinge 11284284d3cSJeremy Fitzhardinge spinlock_t rx_lock ____cacheline_aligned_in_smp; 11384284d3cSJeremy Fitzhardinge struct xen_netif_rx_front_ring rx; 11484284d3cSJeremy Fitzhardinge int rx_ring_ref; 11584284d3cSJeremy Fitzhardinge 11684284d3cSJeremy Fitzhardinge /* Receive-ring batched refills. */ 11784284d3cSJeremy Fitzhardinge #define RX_MIN_TARGET 8 11884284d3cSJeremy Fitzhardinge #define RX_DFL_MIN_TARGET 64 11984284d3cSJeremy Fitzhardinge #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) 12084284d3cSJeremy Fitzhardinge unsigned rx_min_target, rx_max_target, rx_target; 12184284d3cSJeremy Fitzhardinge struct sk_buff_head rx_batch; 12284284d3cSJeremy Fitzhardinge 12384284d3cSJeremy Fitzhardinge struct timer_list rx_refill_timer; 12484284d3cSJeremy Fitzhardinge 1250d160211SJeremy Fitzhardinge struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; 1260d160211SJeremy Fitzhardinge grant_ref_t gref_rx_head; 1270d160211SJeremy Fitzhardinge grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; 1280d160211SJeremy Fitzhardinge 1290d160211SJeremy Fitzhardinge unsigned long rx_pfn_array[NET_RX_RING_SIZE]; 1300d160211SJeremy Fitzhardinge struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1]; 1310d160211SJeremy Fitzhardinge struct mmu_update rx_mmu[NET_RX_RING_SIZE]; 132e0ce4af9SIan Campbell 133e0ce4af9SIan Campbell /* Statistics */ 134e00f85beSstephen hemminger struct netfront_stats __percpu *stats; 135e00f85beSstephen hemminger 136e9a799eaSIan Campbell unsigned long rx_gso_checksum_fixup; 1370d160211SJeremy Fitzhardinge }; 1380d160211SJeremy Fitzhardinge 1390d160211SJeremy Fitzhardinge struct netfront_rx_info { 1400d160211SJeremy Fitzhardinge struct xen_netif_rx_response rx; 1410d160211SJeremy Fitzhardinge struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; 1420d160211SJeremy Fitzhardinge }; 1430d160211SJeremy Fitzhardinge 1441ffb40b8SIsaku Yamahata static void skb_entry_set_link(union skb_entry *list, unsigned short id) 1451ffb40b8SIsaku Yamahata { 1461ffb40b8SIsaku Yamahata list->link = id; 1471ffb40b8SIsaku Yamahata } 1481ffb40b8SIsaku Yamahata 1491ffb40b8SIsaku Yamahata static int skb_entry_is_link(const union skb_entry *list) 1501ffb40b8SIsaku Yamahata { 1511ffb40b8SIsaku Yamahata BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); 152807540baSEric Dumazet return (unsigned long)list->skb < PAGE_OFFSET; 1531ffb40b8SIsaku Yamahata } 1541ffb40b8SIsaku Yamahata 1550d160211SJeremy Fitzhardinge /* 1560d160211SJeremy Fitzhardinge * Access macros for acquiring freeing slots in tx_skbs[]. 1570d160211SJeremy Fitzhardinge */ 1580d160211SJeremy Fitzhardinge 1590d160211SJeremy Fitzhardinge static void add_id_to_freelist(unsigned *head, union skb_entry *list, 1600d160211SJeremy Fitzhardinge unsigned short id) 1610d160211SJeremy Fitzhardinge { 1621ffb40b8SIsaku Yamahata skb_entry_set_link(&list[id], *head); 1630d160211SJeremy Fitzhardinge *head = id; 1640d160211SJeremy Fitzhardinge } 1650d160211SJeremy Fitzhardinge 1660d160211SJeremy Fitzhardinge static unsigned short get_id_from_freelist(unsigned *head, 1670d160211SJeremy Fitzhardinge union skb_entry *list) 1680d160211SJeremy Fitzhardinge { 1690d160211SJeremy Fitzhardinge unsigned int id = *head; 1700d160211SJeremy Fitzhardinge *head = list[id].link; 1710d160211SJeremy Fitzhardinge return id; 1720d160211SJeremy Fitzhardinge } 1730d160211SJeremy Fitzhardinge 1740d160211SJeremy Fitzhardinge static int xennet_rxidx(RING_IDX idx) 1750d160211SJeremy Fitzhardinge { 1760d160211SJeremy Fitzhardinge return idx & (NET_RX_RING_SIZE - 1); 1770d160211SJeremy Fitzhardinge } 1780d160211SJeremy Fitzhardinge 1790d160211SJeremy Fitzhardinge static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np, 1800d160211SJeremy Fitzhardinge RING_IDX ri) 1810d160211SJeremy Fitzhardinge { 1820d160211SJeremy Fitzhardinge int i = xennet_rxidx(ri); 1830d160211SJeremy Fitzhardinge struct sk_buff *skb = np->rx_skbs[i]; 1840d160211SJeremy Fitzhardinge np->rx_skbs[i] = NULL; 1850d160211SJeremy Fitzhardinge return skb; 1860d160211SJeremy Fitzhardinge } 1870d160211SJeremy Fitzhardinge 1880d160211SJeremy Fitzhardinge static grant_ref_t xennet_get_rx_ref(struct netfront_info *np, 1890d160211SJeremy Fitzhardinge RING_IDX ri) 1900d160211SJeremy Fitzhardinge { 1910d160211SJeremy Fitzhardinge int i = xennet_rxidx(ri); 1920d160211SJeremy Fitzhardinge grant_ref_t ref = np->grant_rx_ref[i]; 1930d160211SJeremy Fitzhardinge np->grant_rx_ref[i] = GRANT_INVALID_REF; 1940d160211SJeremy Fitzhardinge return ref; 1950d160211SJeremy Fitzhardinge } 1960d160211SJeremy Fitzhardinge 1970d160211SJeremy Fitzhardinge #ifdef CONFIG_SYSFS 1980d160211SJeremy Fitzhardinge static int xennet_sysfs_addif(struct net_device *netdev); 1990d160211SJeremy Fitzhardinge static void xennet_sysfs_delif(struct net_device *netdev); 2000d160211SJeremy Fitzhardinge #else /* !CONFIG_SYSFS */ 2010d160211SJeremy Fitzhardinge #define xennet_sysfs_addif(dev) (0) 2020d160211SJeremy Fitzhardinge #define xennet_sysfs_delif(dev) do { } while (0) 2030d160211SJeremy Fitzhardinge #endif 2040d160211SJeremy Fitzhardinge 2053ad9b358SMichał Mirosław static bool xennet_can_sg(struct net_device *dev) 2060d160211SJeremy Fitzhardinge { 2073ad9b358SMichał Mirosław return dev->features & NETIF_F_SG; 2080d160211SJeremy Fitzhardinge } 2090d160211SJeremy Fitzhardinge 2100d160211SJeremy Fitzhardinge 2110d160211SJeremy Fitzhardinge static void rx_refill_timeout(unsigned long data) 2120d160211SJeremy Fitzhardinge { 2130d160211SJeremy Fitzhardinge struct net_device *dev = (struct net_device *)data; 214bea3348eSStephen Hemminger struct netfront_info *np = netdev_priv(dev); 215288379f0SBen Hutchings napi_schedule(&np->napi); 2160d160211SJeremy Fitzhardinge } 2170d160211SJeremy Fitzhardinge 2180d160211SJeremy Fitzhardinge static int netfront_tx_slot_available(struct netfront_info *np) 2190d160211SJeremy Fitzhardinge { 220807540baSEric Dumazet return (np->tx.req_prod_pvt - np->tx.rsp_cons) < 221807540baSEric Dumazet (TX_MAX_TARGET - MAX_SKB_FRAGS - 2); 2220d160211SJeremy Fitzhardinge } 2230d160211SJeremy Fitzhardinge 2240d160211SJeremy Fitzhardinge static void xennet_maybe_wake_tx(struct net_device *dev) 2250d160211SJeremy Fitzhardinge { 2260d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 2270d160211SJeremy Fitzhardinge 2280d160211SJeremy Fitzhardinge if (unlikely(netif_queue_stopped(dev)) && 2290d160211SJeremy Fitzhardinge netfront_tx_slot_available(np) && 2300d160211SJeremy Fitzhardinge likely(netif_running(dev))) 2310d160211SJeremy Fitzhardinge netif_wake_queue(dev); 2320d160211SJeremy Fitzhardinge } 2330d160211SJeremy Fitzhardinge 2340d160211SJeremy Fitzhardinge static void xennet_alloc_rx_buffers(struct net_device *dev) 2350d160211SJeremy Fitzhardinge { 2360d160211SJeremy Fitzhardinge unsigned short id; 2370d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 2380d160211SJeremy Fitzhardinge struct sk_buff *skb; 2390d160211SJeremy Fitzhardinge struct page *page; 2400d160211SJeremy Fitzhardinge int i, batch_target, notify; 2410d160211SJeremy Fitzhardinge RING_IDX req_prod = np->rx.req_prod_pvt; 2420d160211SJeremy Fitzhardinge grant_ref_t ref; 2430d160211SJeremy Fitzhardinge unsigned long pfn; 2440d160211SJeremy Fitzhardinge void *vaddr; 2450d160211SJeremy Fitzhardinge struct xen_netif_rx_request *req; 2460d160211SJeremy Fitzhardinge 2470d160211SJeremy Fitzhardinge if (unlikely(!netif_carrier_ok(dev))) 2480d160211SJeremy Fitzhardinge return; 2490d160211SJeremy Fitzhardinge 2500d160211SJeremy Fitzhardinge /* 2510d160211SJeremy Fitzhardinge * Allocate skbuffs greedily, even though we batch updates to the 2520d160211SJeremy Fitzhardinge * receive ring. This creates a less bursty demand on the memory 2530d160211SJeremy Fitzhardinge * allocator, so should reduce the chance of failed allocation requests 2540d160211SJeremy Fitzhardinge * both for ourself and for other kernel subsystems. 2550d160211SJeremy Fitzhardinge */ 2560d160211SJeremy Fitzhardinge batch_target = np->rx_target - (req_prod - np->rx.rsp_cons); 2570d160211SJeremy Fitzhardinge for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) { 258617a20bbSIsaku Yamahata skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN, 2590d160211SJeremy Fitzhardinge GFP_ATOMIC | __GFP_NOWARN); 2600d160211SJeremy Fitzhardinge if (unlikely(!skb)) 2610d160211SJeremy Fitzhardinge goto no_skb; 2620d160211SJeremy Fitzhardinge 263617a20bbSIsaku Yamahata /* Align ip header to a 16 bytes boundary */ 264617a20bbSIsaku Yamahata skb_reserve(skb, NET_IP_ALIGN); 265617a20bbSIsaku Yamahata 2660d160211SJeremy Fitzhardinge page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); 2670d160211SJeremy Fitzhardinge if (!page) { 2680d160211SJeremy Fitzhardinge kfree_skb(skb); 2690d160211SJeremy Fitzhardinge no_skb: 2700d160211SJeremy Fitzhardinge /* Any skbuffs queued for refill? Force them out. */ 2710d160211SJeremy Fitzhardinge if (i != 0) 2720d160211SJeremy Fitzhardinge goto refill; 2730d160211SJeremy Fitzhardinge /* Could not allocate any skbuffs. Try again later. */ 2740d160211SJeremy Fitzhardinge mod_timer(&np->rx_refill_timer, 2750d160211SJeremy Fitzhardinge jiffies + (HZ/10)); 2760d160211SJeremy Fitzhardinge break; 2770d160211SJeremy Fitzhardinge } 2780d160211SJeremy Fitzhardinge 27901c68026SIan Campbell __skb_fill_page_desc(skb, 0, page, 0, 0); 2800d160211SJeremy Fitzhardinge skb_shinfo(skb)->nr_frags = 1; 2810d160211SJeremy Fitzhardinge __skb_queue_tail(&np->rx_batch, skb); 2820d160211SJeremy Fitzhardinge } 2830d160211SJeremy Fitzhardinge 2840d160211SJeremy Fitzhardinge /* Is the batch large enough to be worthwhile? */ 2850d160211SJeremy Fitzhardinge if (i < (np->rx_target/2)) { 2860d160211SJeremy Fitzhardinge if (req_prod > np->rx.sring->req_prod) 2870d160211SJeremy Fitzhardinge goto push; 2880d160211SJeremy Fitzhardinge return; 2890d160211SJeremy Fitzhardinge } 2900d160211SJeremy Fitzhardinge 2910d160211SJeremy Fitzhardinge /* Adjust our fill target if we risked running out of buffers. */ 2920d160211SJeremy Fitzhardinge if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) && 2930d160211SJeremy Fitzhardinge ((np->rx_target *= 2) > np->rx_max_target)) 2940d160211SJeremy Fitzhardinge np->rx_target = np->rx_max_target; 2950d160211SJeremy Fitzhardinge 2960d160211SJeremy Fitzhardinge refill: 2975dcddfaeSJeremy Fitzhardinge for (i = 0; ; i++) { 2980d160211SJeremy Fitzhardinge skb = __skb_dequeue(&np->rx_batch); 2990d160211SJeremy Fitzhardinge if (skb == NULL) 3000d160211SJeremy Fitzhardinge break; 3010d160211SJeremy Fitzhardinge 3020d160211SJeremy Fitzhardinge skb->dev = dev; 3030d160211SJeremy Fitzhardinge 3040d160211SJeremy Fitzhardinge id = xennet_rxidx(req_prod + i); 3050d160211SJeremy Fitzhardinge 3060d160211SJeremy Fitzhardinge BUG_ON(np->rx_skbs[id]); 3070d160211SJeremy Fitzhardinge np->rx_skbs[id] = skb; 3080d160211SJeremy Fitzhardinge 3090d160211SJeremy Fitzhardinge ref = gnttab_claim_grant_reference(&np->gref_rx_head); 3100d160211SJeremy Fitzhardinge BUG_ON((signed short)ref < 0); 3110d160211SJeremy Fitzhardinge np->grant_rx_ref[id] = ref; 3120d160211SJeremy Fitzhardinge 31301c68026SIan Campbell pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0])); 31401c68026SIan Campbell vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0])); 3150d160211SJeremy Fitzhardinge 3160d160211SJeremy Fitzhardinge req = RING_GET_REQUEST(&np->rx, req_prod + i); 3170d160211SJeremy Fitzhardinge gnttab_grant_foreign_access_ref(ref, 3180d160211SJeremy Fitzhardinge np->xbdev->otherend_id, 3190d160211SJeremy Fitzhardinge pfn_to_mfn(pfn), 3200d160211SJeremy Fitzhardinge 0); 3210d160211SJeremy Fitzhardinge 3220d160211SJeremy Fitzhardinge req->id = id; 3230d160211SJeremy Fitzhardinge req->gref = ref; 3240d160211SJeremy Fitzhardinge } 3250d160211SJeremy Fitzhardinge 3260d160211SJeremy Fitzhardinge wmb(); /* barrier so backend seens requests */ 3270d160211SJeremy Fitzhardinge 3280d160211SJeremy Fitzhardinge /* Above is a suitable barrier to ensure backend will see requests. */ 3290d160211SJeremy Fitzhardinge np->rx.req_prod_pvt = req_prod + i; 3300d160211SJeremy Fitzhardinge push: 3310d160211SJeremy Fitzhardinge RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify); 3320d160211SJeremy Fitzhardinge if (notify) 3330d160211SJeremy Fitzhardinge notify_remote_via_irq(np->netdev->irq); 3340d160211SJeremy Fitzhardinge } 3350d160211SJeremy Fitzhardinge 3360d160211SJeremy Fitzhardinge static int xennet_open(struct net_device *dev) 3370d160211SJeremy Fitzhardinge { 3380d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 3390d160211SJeremy Fitzhardinge 340bea3348eSStephen Hemminger napi_enable(&np->napi); 341bea3348eSStephen Hemminger 3420d160211SJeremy Fitzhardinge spin_lock_bh(&np->rx_lock); 3430d160211SJeremy Fitzhardinge if (netif_carrier_ok(dev)) { 3440d160211SJeremy Fitzhardinge xennet_alloc_rx_buffers(dev); 3450d160211SJeremy Fitzhardinge np->rx.sring->rsp_event = np->rx.rsp_cons + 1; 3460d160211SJeremy Fitzhardinge if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) 347288379f0SBen Hutchings napi_schedule(&np->napi); 3480d160211SJeremy Fitzhardinge } 3490d160211SJeremy Fitzhardinge spin_unlock_bh(&np->rx_lock); 3500d160211SJeremy Fitzhardinge 3510b1ab1b8SEduardo Habkost netif_start_queue(dev); 3520d160211SJeremy Fitzhardinge 3530d160211SJeremy Fitzhardinge return 0; 3540d160211SJeremy Fitzhardinge } 3550d160211SJeremy Fitzhardinge 3560d160211SJeremy Fitzhardinge static void xennet_tx_buf_gc(struct net_device *dev) 3570d160211SJeremy Fitzhardinge { 3580d160211SJeremy Fitzhardinge RING_IDX cons, prod; 3590d160211SJeremy Fitzhardinge unsigned short id; 3600d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 3610d160211SJeremy Fitzhardinge struct sk_buff *skb; 3620d160211SJeremy Fitzhardinge 3630d160211SJeremy Fitzhardinge BUG_ON(!netif_carrier_ok(dev)); 3640d160211SJeremy Fitzhardinge 3650d160211SJeremy Fitzhardinge do { 3660d160211SJeremy Fitzhardinge prod = np->tx.sring->rsp_prod; 3670d160211SJeremy Fitzhardinge rmb(); /* Ensure we see responses up to 'rp'. */ 3680d160211SJeremy Fitzhardinge 3690d160211SJeremy Fitzhardinge for (cons = np->tx.rsp_cons; cons != prod; cons++) { 3700d160211SJeremy Fitzhardinge struct xen_netif_tx_response *txrsp; 3710d160211SJeremy Fitzhardinge 3720d160211SJeremy Fitzhardinge txrsp = RING_GET_RESPONSE(&np->tx, cons); 373f942dc25SIan Campbell if (txrsp->status == XEN_NETIF_RSP_NULL) 3740d160211SJeremy Fitzhardinge continue; 3750d160211SJeremy Fitzhardinge 3760d160211SJeremy Fitzhardinge id = txrsp->id; 3770d160211SJeremy Fitzhardinge skb = np->tx_skbs[id].skb; 3780d160211SJeremy Fitzhardinge if (unlikely(gnttab_query_foreign_access( 3790d160211SJeremy Fitzhardinge np->grant_tx_ref[id]) != 0)) { 3800d160211SJeremy Fitzhardinge printk(KERN_ALERT "xennet_tx_buf_gc: warning " 3810d160211SJeremy Fitzhardinge "-- grant still in use by backend " 3820d160211SJeremy Fitzhardinge "domain.\n"); 3830d160211SJeremy Fitzhardinge BUG(); 3840d160211SJeremy Fitzhardinge } 3850d160211SJeremy Fitzhardinge gnttab_end_foreign_access_ref( 3860d160211SJeremy Fitzhardinge np->grant_tx_ref[id], GNTMAP_readonly); 3870d160211SJeremy Fitzhardinge gnttab_release_grant_reference( 3880d160211SJeremy Fitzhardinge &np->gref_tx_head, np->grant_tx_ref[id]); 3890d160211SJeremy Fitzhardinge np->grant_tx_ref[id] = GRANT_INVALID_REF; 3900d160211SJeremy Fitzhardinge add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id); 3910d160211SJeremy Fitzhardinge dev_kfree_skb_irq(skb); 3920d160211SJeremy Fitzhardinge } 3930d160211SJeremy Fitzhardinge 3940d160211SJeremy Fitzhardinge np->tx.rsp_cons = prod; 3950d160211SJeremy Fitzhardinge 3960d160211SJeremy Fitzhardinge /* 3970d160211SJeremy Fitzhardinge * Set a new event, then check for race with update of tx_cons. 3980d160211SJeremy Fitzhardinge * Note that it is essential to schedule a callback, no matter 3990d160211SJeremy Fitzhardinge * how few buffers are pending. Even if there is space in the 4000d160211SJeremy Fitzhardinge * transmit ring, higher layers may be blocked because too much 4010d160211SJeremy Fitzhardinge * data is outstanding: in such cases notification from Xen is 4020d160211SJeremy Fitzhardinge * likely to be the only kick that we'll get. 4030d160211SJeremy Fitzhardinge */ 4040d160211SJeremy Fitzhardinge np->tx.sring->rsp_event = 4050d160211SJeremy Fitzhardinge prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; 4060d160211SJeremy Fitzhardinge mb(); /* update shared area */ 4070d160211SJeremy Fitzhardinge } while ((cons == prod) && (prod != np->tx.sring->rsp_prod)); 4080d160211SJeremy Fitzhardinge 4090d160211SJeremy Fitzhardinge xennet_maybe_wake_tx(dev); 4100d160211SJeremy Fitzhardinge } 4110d160211SJeremy Fitzhardinge 4120d160211SJeremy Fitzhardinge static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev, 4130d160211SJeremy Fitzhardinge struct xen_netif_tx_request *tx) 4140d160211SJeremy Fitzhardinge { 4150d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 4160d160211SJeremy Fitzhardinge char *data = skb->data; 4170d160211SJeremy Fitzhardinge unsigned long mfn; 4180d160211SJeremy Fitzhardinge RING_IDX prod = np->tx.req_prod_pvt; 4190d160211SJeremy Fitzhardinge int frags = skb_shinfo(skb)->nr_frags; 4200d160211SJeremy Fitzhardinge unsigned int offset = offset_in_page(data); 4210d160211SJeremy Fitzhardinge unsigned int len = skb_headlen(skb); 4220d160211SJeremy Fitzhardinge unsigned int id; 4230d160211SJeremy Fitzhardinge grant_ref_t ref; 4240d160211SJeremy Fitzhardinge int i; 4250d160211SJeremy Fitzhardinge 4260d160211SJeremy Fitzhardinge /* While the header overlaps a page boundary (including being 4270d160211SJeremy Fitzhardinge larger than a page), split it it into page-sized chunks. */ 4280d160211SJeremy Fitzhardinge while (len > PAGE_SIZE - offset) { 4290d160211SJeremy Fitzhardinge tx->size = PAGE_SIZE - offset; 430f942dc25SIan Campbell tx->flags |= XEN_NETTXF_more_data; 4310d160211SJeremy Fitzhardinge len -= tx->size; 4320d160211SJeremy Fitzhardinge data += tx->size; 4330d160211SJeremy Fitzhardinge offset = 0; 4340d160211SJeremy Fitzhardinge 4350d160211SJeremy Fitzhardinge id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); 4360d160211SJeremy Fitzhardinge np->tx_skbs[id].skb = skb_get(skb); 4370d160211SJeremy Fitzhardinge tx = RING_GET_REQUEST(&np->tx, prod++); 4380d160211SJeremy Fitzhardinge tx->id = id; 4390d160211SJeremy Fitzhardinge ref = gnttab_claim_grant_reference(&np->gref_tx_head); 4400d160211SJeremy Fitzhardinge BUG_ON((signed short)ref < 0); 4410d160211SJeremy Fitzhardinge 4420d160211SJeremy Fitzhardinge mfn = virt_to_mfn(data); 4430d160211SJeremy Fitzhardinge gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, 4440d160211SJeremy Fitzhardinge mfn, GNTMAP_readonly); 4450d160211SJeremy Fitzhardinge 4460d160211SJeremy Fitzhardinge tx->gref = np->grant_tx_ref[id] = ref; 4470d160211SJeremy Fitzhardinge tx->offset = offset; 4480d160211SJeremy Fitzhardinge tx->size = len; 4490d160211SJeremy Fitzhardinge tx->flags = 0; 4500d160211SJeremy Fitzhardinge } 4510d160211SJeremy Fitzhardinge 4520d160211SJeremy Fitzhardinge /* Grant backend access to each skb fragment page. */ 4530d160211SJeremy Fitzhardinge for (i = 0; i < frags; i++) { 4540d160211SJeremy Fitzhardinge skb_frag_t *frag = skb_shinfo(skb)->frags + i; 455f36c3747SIan Campbell struct page *page = skb_frag_page(frag); 456f36c3747SIan Campbell 457f36c3747SIan Campbell len = skb_frag_size(frag); 458f36c3747SIan Campbell offset = frag->page_offset; 459f36c3747SIan Campbell 460f36c3747SIan Campbell /* Data must not cross a page boundary. */ 461f36c3747SIan Campbell BUG_ON(len + offset > PAGE_SIZE<<compound_order(page)); 462f36c3747SIan Campbell 463f36c3747SIan Campbell /* Skip unused frames from start of page */ 464f36c3747SIan Campbell page += offset >> PAGE_SHIFT; 465f36c3747SIan Campbell offset &= ~PAGE_MASK; 466f36c3747SIan Campbell 467f36c3747SIan Campbell while (len > 0) { 468f36c3747SIan Campbell unsigned long bytes; 469f36c3747SIan Campbell 470f36c3747SIan Campbell BUG_ON(offset >= PAGE_SIZE); 471f36c3747SIan Campbell 472f36c3747SIan Campbell bytes = PAGE_SIZE - offset; 473f36c3747SIan Campbell if (bytes > len) 474f36c3747SIan Campbell bytes = len; 4750d160211SJeremy Fitzhardinge 476f942dc25SIan Campbell tx->flags |= XEN_NETTXF_more_data; 4770d160211SJeremy Fitzhardinge 478f36c3747SIan Campbell id = get_id_from_freelist(&np->tx_skb_freelist, 479f36c3747SIan Campbell np->tx_skbs); 4800d160211SJeremy Fitzhardinge np->tx_skbs[id].skb = skb_get(skb); 4810d160211SJeremy Fitzhardinge tx = RING_GET_REQUEST(&np->tx, prod++); 4820d160211SJeremy Fitzhardinge tx->id = id; 4830d160211SJeremy Fitzhardinge ref = gnttab_claim_grant_reference(&np->gref_tx_head); 4840d160211SJeremy Fitzhardinge BUG_ON((signed short)ref < 0); 4850d160211SJeremy Fitzhardinge 486f36c3747SIan Campbell mfn = pfn_to_mfn(page_to_pfn(page)); 487f36c3747SIan Campbell gnttab_grant_foreign_access_ref(ref, 488f36c3747SIan Campbell np->xbdev->otherend_id, 4890d160211SJeremy Fitzhardinge mfn, GNTMAP_readonly); 4900d160211SJeremy Fitzhardinge 4910d160211SJeremy Fitzhardinge tx->gref = np->grant_tx_ref[id] = ref; 492f36c3747SIan Campbell tx->offset = offset; 493f36c3747SIan Campbell tx->size = bytes; 4940d160211SJeremy Fitzhardinge tx->flags = 0; 495f36c3747SIan Campbell 496f36c3747SIan Campbell offset += bytes; 497f36c3747SIan Campbell len -= bytes; 498f36c3747SIan Campbell 499f36c3747SIan Campbell /* Next frame */ 500f36c3747SIan Campbell if (offset == PAGE_SIZE && len) { 501f36c3747SIan Campbell BUG_ON(!PageCompound(page)); 502f36c3747SIan Campbell page++; 503f36c3747SIan Campbell offset = 0; 504f36c3747SIan Campbell } 505f36c3747SIan Campbell } 5060d160211SJeremy Fitzhardinge } 5070d160211SJeremy Fitzhardinge 5080d160211SJeremy Fitzhardinge np->tx.req_prod_pvt = prod; 5090d160211SJeremy Fitzhardinge } 5100d160211SJeremy Fitzhardinge 511f36c3747SIan Campbell /* 512f36c3747SIan Campbell * Count how many ring slots are required to send the frags of this 513f36c3747SIan Campbell * skb. Each frag might be a compound page. 514f36c3747SIan Campbell */ 515f36c3747SIan Campbell static int xennet_count_skb_frag_slots(struct sk_buff *skb) 516f36c3747SIan Campbell { 517f36c3747SIan Campbell int i, frags = skb_shinfo(skb)->nr_frags; 518f36c3747SIan Campbell int pages = 0; 519f36c3747SIan Campbell 520f36c3747SIan Campbell for (i = 0; i < frags; i++) { 521f36c3747SIan Campbell skb_frag_t *frag = skb_shinfo(skb)->frags + i; 522f36c3747SIan Campbell unsigned long size = skb_frag_size(frag); 523f36c3747SIan Campbell unsigned long offset = frag->page_offset; 524f36c3747SIan Campbell 525f36c3747SIan Campbell /* Skip unused frames from start of page */ 526f36c3747SIan Campbell offset &= ~PAGE_MASK; 527f36c3747SIan Campbell 528f36c3747SIan Campbell pages += PFN_UP(offset + size); 529f36c3747SIan Campbell } 530f36c3747SIan Campbell 531f36c3747SIan Campbell return pages; 532f36c3747SIan Campbell } 533f36c3747SIan Campbell 5340d160211SJeremy Fitzhardinge static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) 5350d160211SJeremy Fitzhardinge { 5360d160211SJeremy Fitzhardinge unsigned short id; 5370d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 538e00f85beSstephen hemminger struct netfront_stats *stats = this_cpu_ptr(np->stats); 5390d160211SJeremy Fitzhardinge struct xen_netif_tx_request *tx; 5400d160211SJeremy Fitzhardinge char *data = skb->data; 5410d160211SJeremy Fitzhardinge RING_IDX i; 5420d160211SJeremy Fitzhardinge grant_ref_t ref; 5430d160211SJeremy Fitzhardinge unsigned long mfn; 5440d160211SJeremy Fitzhardinge int notify; 545f36c3747SIan Campbell int slots; 5460d160211SJeremy Fitzhardinge unsigned int offset = offset_in_page(data); 5470d160211SJeremy Fitzhardinge unsigned int len = skb_headlen(skb); 548cf66f9d4SKonrad Rzeszutek Wilk unsigned long flags; 5490d160211SJeremy Fitzhardinge 5509ecd1a75SWei Liu /* If skb->len is too big for wire format, drop skb and alert 5519ecd1a75SWei Liu * user about misconfiguration. 5529ecd1a75SWei Liu */ 5539ecd1a75SWei Liu if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { 5549ecd1a75SWei Liu net_alert_ratelimited( 5559ecd1a75SWei Liu "xennet: skb->len = %u, too big for wire format\n", 5569ecd1a75SWei Liu skb->len); 5579ecd1a75SWei Liu goto drop; 5589ecd1a75SWei Liu } 5599ecd1a75SWei Liu 560f36c3747SIan Campbell slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) + 561f36c3747SIan Campbell xennet_count_skb_frag_slots(skb); 562f36c3747SIan Campbell if (unlikely(slots > MAX_SKB_FRAGS + 1)) { 563f36c3747SIan Campbell net_alert_ratelimited( 564f36c3747SIan Campbell "xennet: skb rides the rocket: %d slots\n", slots); 5650d160211SJeremy Fitzhardinge goto drop; 5660d160211SJeremy Fitzhardinge } 5670d160211SJeremy Fitzhardinge 568cf66f9d4SKonrad Rzeszutek Wilk spin_lock_irqsave(&np->tx_lock, flags); 5690d160211SJeremy Fitzhardinge 5700d160211SJeremy Fitzhardinge if (unlikely(!netif_carrier_ok(dev) || 571f36c3747SIan Campbell (slots > 1 && !xennet_can_sg(dev)) || 572fc741216SJesse Gross netif_needs_gso(skb, netif_skb_features(skb)))) { 573cf66f9d4SKonrad Rzeszutek Wilk spin_unlock_irqrestore(&np->tx_lock, flags); 5740d160211SJeremy Fitzhardinge goto drop; 5750d160211SJeremy Fitzhardinge } 5760d160211SJeremy Fitzhardinge 5770d160211SJeremy Fitzhardinge i = np->tx.req_prod_pvt; 5780d160211SJeremy Fitzhardinge 5790d160211SJeremy Fitzhardinge id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); 5800d160211SJeremy Fitzhardinge np->tx_skbs[id].skb = skb; 5810d160211SJeremy Fitzhardinge 5820d160211SJeremy Fitzhardinge tx = RING_GET_REQUEST(&np->tx, i); 5830d160211SJeremy Fitzhardinge 5840d160211SJeremy Fitzhardinge tx->id = id; 5850d160211SJeremy Fitzhardinge ref = gnttab_claim_grant_reference(&np->gref_tx_head); 5860d160211SJeremy Fitzhardinge BUG_ON((signed short)ref < 0); 5870d160211SJeremy Fitzhardinge mfn = virt_to_mfn(data); 5880d160211SJeremy Fitzhardinge gnttab_grant_foreign_access_ref( 5890d160211SJeremy Fitzhardinge ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly); 5900d160211SJeremy Fitzhardinge tx->gref = np->grant_tx_ref[id] = ref; 5910d160211SJeremy Fitzhardinge tx->offset = offset; 5920d160211SJeremy Fitzhardinge tx->size = len; 5930d160211SJeremy Fitzhardinge 5940d160211SJeremy Fitzhardinge tx->flags = 0; 5950d160211SJeremy Fitzhardinge if (skb->ip_summed == CHECKSUM_PARTIAL) 5960d160211SJeremy Fitzhardinge /* local packet? */ 597f942dc25SIan Campbell tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; 5980d160211SJeremy Fitzhardinge else if (skb->ip_summed == CHECKSUM_UNNECESSARY) 5990d160211SJeremy Fitzhardinge /* remote but checksummed. */ 600f942dc25SIan Campbell tx->flags |= XEN_NETTXF_data_validated; 6010d160211SJeremy Fitzhardinge 6020d160211SJeremy Fitzhardinge if (skb_shinfo(skb)->gso_size) { 6030d160211SJeremy Fitzhardinge struct xen_netif_extra_info *gso; 6040d160211SJeremy Fitzhardinge 6050d160211SJeremy Fitzhardinge gso = (struct xen_netif_extra_info *) 6060d160211SJeremy Fitzhardinge RING_GET_REQUEST(&np->tx, ++i); 6070d160211SJeremy Fitzhardinge 608f942dc25SIan Campbell tx->flags |= XEN_NETTXF_extra_info; 6090d160211SJeremy Fitzhardinge 6100d160211SJeremy Fitzhardinge gso->u.gso.size = skb_shinfo(skb)->gso_size; 6110d160211SJeremy Fitzhardinge gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; 6120d160211SJeremy Fitzhardinge gso->u.gso.pad = 0; 6130d160211SJeremy Fitzhardinge gso->u.gso.features = 0; 6140d160211SJeremy Fitzhardinge 6150d160211SJeremy Fitzhardinge gso->type = XEN_NETIF_EXTRA_TYPE_GSO; 6160d160211SJeremy Fitzhardinge gso->flags = 0; 6170d160211SJeremy Fitzhardinge } 6180d160211SJeremy Fitzhardinge 6190d160211SJeremy Fitzhardinge np->tx.req_prod_pvt = i + 1; 6200d160211SJeremy Fitzhardinge 6210d160211SJeremy Fitzhardinge xennet_make_frags(skb, dev, tx); 6220d160211SJeremy Fitzhardinge tx->size = skb->len; 6230d160211SJeremy Fitzhardinge 6240d160211SJeremy Fitzhardinge RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify); 6250d160211SJeremy Fitzhardinge if (notify) 6260d160211SJeremy Fitzhardinge notify_remote_via_irq(np->netdev->irq); 6270d160211SJeremy Fitzhardinge 628e00f85beSstephen hemminger u64_stats_update_begin(&stats->syncp); 629e00f85beSstephen hemminger stats->tx_bytes += skb->len; 630e00f85beSstephen hemminger stats->tx_packets++; 631e00f85beSstephen hemminger u64_stats_update_end(&stats->syncp); 63210a273a6SJeremy Fitzhardinge 63310a273a6SJeremy Fitzhardinge /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ 6340d160211SJeremy Fitzhardinge xennet_tx_buf_gc(dev); 6350d160211SJeremy Fitzhardinge 6360d160211SJeremy Fitzhardinge if (!netfront_tx_slot_available(np)) 6370d160211SJeremy Fitzhardinge netif_stop_queue(dev); 6380d160211SJeremy Fitzhardinge 639cf66f9d4SKonrad Rzeszutek Wilk spin_unlock_irqrestore(&np->tx_lock, flags); 6400d160211SJeremy Fitzhardinge 6416ed10654SPatrick McHardy return NETDEV_TX_OK; 6420d160211SJeremy Fitzhardinge 6430d160211SJeremy Fitzhardinge drop: 64409f75cd7SJeff Garzik dev->stats.tx_dropped++; 6450d160211SJeremy Fitzhardinge dev_kfree_skb(skb); 6466ed10654SPatrick McHardy return NETDEV_TX_OK; 6470d160211SJeremy Fitzhardinge } 6480d160211SJeremy Fitzhardinge 6490d160211SJeremy Fitzhardinge static int xennet_close(struct net_device *dev) 6500d160211SJeremy Fitzhardinge { 6510d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 6520d160211SJeremy Fitzhardinge netif_stop_queue(np->netdev); 653bea3348eSStephen Hemminger napi_disable(&np->napi); 6540d160211SJeremy Fitzhardinge return 0; 6550d160211SJeremy Fitzhardinge } 6560d160211SJeremy Fitzhardinge 6570d160211SJeremy Fitzhardinge static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb, 6580d160211SJeremy Fitzhardinge grant_ref_t ref) 6590d160211SJeremy Fitzhardinge { 6600d160211SJeremy Fitzhardinge int new = xennet_rxidx(np->rx.req_prod_pvt); 6610d160211SJeremy Fitzhardinge 6620d160211SJeremy Fitzhardinge BUG_ON(np->rx_skbs[new]); 6630d160211SJeremy Fitzhardinge np->rx_skbs[new] = skb; 6640d160211SJeremy Fitzhardinge np->grant_rx_ref[new] = ref; 6650d160211SJeremy Fitzhardinge RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; 6660d160211SJeremy Fitzhardinge RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; 6670d160211SJeremy Fitzhardinge np->rx.req_prod_pvt++; 6680d160211SJeremy Fitzhardinge } 6690d160211SJeremy Fitzhardinge 6700d160211SJeremy Fitzhardinge static int xennet_get_extras(struct netfront_info *np, 6710d160211SJeremy Fitzhardinge struct xen_netif_extra_info *extras, 6720d160211SJeremy Fitzhardinge RING_IDX rp) 6730d160211SJeremy Fitzhardinge 6740d160211SJeremy Fitzhardinge { 6750d160211SJeremy Fitzhardinge struct xen_netif_extra_info *extra; 6760d160211SJeremy Fitzhardinge struct device *dev = &np->netdev->dev; 6770d160211SJeremy Fitzhardinge RING_IDX cons = np->rx.rsp_cons; 6780d160211SJeremy Fitzhardinge int err = 0; 6790d160211SJeremy Fitzhardinge 6800d160211SJeremy Fitzhardinge do { 6810d160211SJeremy Fitzhardinge struct sk_buff *skb; 6820d160211SJeremy Fitzhardinge grant_ref_t ref; 6830d160211SJeremy Fitzhardinge 6840d160211SJeremy Fitzhardinge if (unlikely(cons + 1 == rp)) { 6850d160211SJeremy Fitzhardinge if (net_ratelimit()) 6860d160211SJeremy Fitzhardinge dev_warn(dev, "Missing extra info\n"); 6870d160211SJeremy Fitzhardinge err = -EBADR; 6880d160211SJeremy Fitzhardinge break; 6890d160211SJeremy Fitzhardinge } 6900d160211SJeremy Fitzhardinge 6910d160211SJeremy Fitzhardinge extra = (struct xen_netif_extra_info *) 6920d160211SJeremy Fitzhardinge RING_GET_RESPONSE(&np->rx, ++cons); 6930d160211SJeremy Fitzhardinge 6940d160211SJeremy Fitzhardinge if (unlikely(!extra->type || 6950d160211SJeremy Fitzhardinge extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { 6960d160211SJeremy Fitzhardinge if (net_ratelimit()) 6970d160211SJeremy Fitzhardinge dev_warn(dev, "Invalid extra type: %d\n", 6980d160211SJeremy Fitzhardinge extra->type); 6990d160211SJeremy Fitzhardinge err = -EINVAL; 7000d160211SJeremy Fitzhardinge } else { 7010d160211SJeremy Fitzhardinge memcpy(&extras[extra->type - 1], extra, 7020d160211SJeremy Fitzhardinge sizeof(*extra)); 7030d160211SJeremy Fitzhardinge } 7040d160211SJeremy Fitzhardinge 7050d160211SJeremy Fitzhardinge skb = xennet_get_rx_skb(np, cons); 7060d160211SJeremy Fitzhardinge ref = xennet_get_rx_ref(np, cons); 7070d160211SJeremy Fitzhardinge xennet_move_rx_slot(np, skb, ref); 7080d160211SJeremy Fitzhardinge } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); 7090d160211SJeremy Fitzhardinge 7100d160211SJeremy Fitzhardinge np->rx.rsp_cons = cons; 7110d160211SJeremy Fitzhardinge return err; 7120d160211SJeremy Fitzhardinge } 7130d160211SJeremy Fitzhardinge 7140d160211SJeremy Fitzhardinge static int xennet_get_responses(struct netfront_info *np, 7150d160211SJeremy Fitzhardinge struct netfront_rx_info *rinfo, RING_IDX rp, 7160d160211SJeremy Fitzhardinge struct sk_buff_head *list) 7170d160211SJeremy Fitzhardinge { 7180d160211SJeremy Fitzhardinge struct xen_netif_rx_response *rx = &rinfo->rx; 7190d160211SJeremy Fitzhardinge struct xen_netif_extra_info *extras = rinfo->extras; 7200d160211SJeremy Fitzhardinge struct device *dev = &np->netdev->dev; 7210d160211SJeremy Fitzhardinge RING_IDX cons = np->rx.rsp_cons; 7220d160211SJeremy Fitzhardinge struct sk_buff *skb = xennet_get_rx_skb(np, cons); 7230d160211SJeremy Fitzhardinge grant_ref_t ref = xennet_get_rx_ref(np, cons); 7240d160211SJeremy Fitzhardinge int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); 7257158ff6dSWei Liu int slots = 1; 7260d160211SJeremy Fitzhardinge int err = 0; 7270d160211SJeremy Fitzhardinge unsigned long ret; 7280d160211SJeremy Fitzhardinge 729f942dc25SIan Campbell if (rx->flags & XEN_NETRXF_extra_info) { 7300d160211SJeremy Fitzhardinge err = xennet_get_extras(np, extras, rp); 7310d160211SJeremy Fitzhardinge cons = np->rx.rsp_cons; 7320d160211SJeremy Fitzhardinge } 7330d160211SJeremy Fitzhardinge 7340d160211SJeremy Fitzhardinge for (;;) { 7350d160211SJeremy Fitzhardinge if (unlikely(rx->status < 0 || 7360d160211SJeremy Fitzhardinge rx->offset + rx->status > PAGE_SIZE)) { 7370d160211SJeremy Fitzhardinge if (net_ratelimit()) 7380d160211SJeremy Fitzhardinge dev_warn(dev, "rx->offset: %x, size: %u\n", 7390d160211SJeremy Fitzhardinge rx->offset, rx->status); 7400d160211SJeremy Fitzhardinge xennet_move_rx_slot(np, skb, ref); 7410d160211SJeremy Fitzhardinge err = -EINVAL; 7420d160211SJeremy Fitzhardinge goto next; 7430d160211SJeremy Fitzhardinge } 7440d160211SJeremy Fitzhardinge 7450d160211SJeremy Fitzhardinge /* 7460d160211SJeremy Fitzhardinge * This definitely indicates a bug, either in this driver or in 7470d160211SJeremy Fitzhardinge * the backend driver. In future this should flag the bad 748697089dcSWei Liu * situation to the system controller to reboot the backend. 7490d160211SJeremy Fitzhardinge */ 7500d160211SJeremy Fitzhardinge if (ref == GRANT_INVALID_REF) { 7510d160211SJeremy Fitzhardinge if (net_ratelimit()) 7520d160211SJeremy Fitzhardinge dev_warn(dev, "Bad rx response id %d.\n", 7530d160211SJeremy Fitzhardinge rx->id); 7540d160211SJeremy Fitzhardinge err = -EINVAL; 7550d160211SJeremy Fitzhardinge goto next; 7560d160211SJeremy Fitzhardinge } 7570d160211SJeremy Fitzhardinge 7580d160211SJeremy Fitzhardinge ret = gnttab_end_foreign_access_ref(ref, 0); 7590d160211SJeremy Fitzhardinge BUG_ON(!ret); 7600d160211SJeremy Fitzhardinge 7610d160211SJeremy Fitzhardinge gnttab_release_grant_reference(&np->gref_rx_head, ref); 7620d160211SJeremy Fitzhardinge 7630d160211SJeremy Fitzhardinge __skb_queue_tail(list, skb); 7640d160211SJeremy Fitzhardinge 7650d160211SJeremy Fitzhardinge next: 766f942dc25SIan Campbell if (!(rx->flags & XEN_NETRXF_more_data)) 7670d160211SJeremy Fitzhardinge break; 7680d160211SJeremy Fitzhardinge 7697158ff6dSWei Liu if (cons + slots == rp) { 7700d160211SJeremy Fitzhardinge if (net_ratelimit()) 7717158ff6dSWei Liu dev_warn(dev, "Need more slots\n"); 7720d160211SJeremy Fitzhardinge err = -ENOENT; 7730d160211SJeremy Fitzhardinge break; 7740d160211SJeremy Fitzhardinge } 7750d160211SJeremy Fitzhardinge 7767158ff6dSWei Liu rx = RING_GET_RESPONSE(&np->rx, cons + slots); 7777158ff6dSWei Liu skb = xennet_get_rx_skb(np, cons + slots); 7787158ff6dSWei Liu ref = xennet_get_rx_ref(np, cons + slots); 7797158ff6dSWei Liu slots++; 7800d160211SJeremy Fitzhardinge } 7810d160211SJeremy Fitzhardinge 7827158ff6dSWei Liu if (unlikely(slots > max)) { 7830d160211SJeremy Fitzhardinge if (net_ratelimit()) 784697089dcSWei Liu dev_warn(dev, "Too many slots\n"); 7850d160211SJeremy Fitzhardinge err = -E2BIG; 7860d160211SJeremy Fitzhardinge } 7870d160211SJeremy Fitzhardinge 7880d160211SJeremy Fitzhardinge if (unlikely(err)) 7897158ff6dSWei Liu np->rx.rsp_cons = cons + slots; 7900d160211SJeremy Fitzhardinge 7910d160211SJeremy Fitzhardinge return err; 7920d160211SJeremy Fitzhardinge } 7930d160211SJeremy Fitzhardinge 7940d160211SJeremy Fitzhardinge static int xennet_set_skb_gso(struct sk_buff *skb, 7950d160211SJeremy Fitzhardinge struct xen_netif_extra_info *gso) 7960d160211SJeremy Fitzhardinge { 7970d160211SJeremy Fitzhardinge if (!gso->u.gso.size) { 7980d160211SJeremy Fitzhardinge if (net_ratelimit()) 7990d160211SJeremy Fitzhardinge printk(KERN_WARNING "GSO size must not be zero.\n"); 8000d160211SJeremy Fitzhardinge return -EINVAL; 8010d160211SJeremy Fitzhardinge } 8020d160211SJeremy Fitzhardinge 8030d160211SJeremy Fitzhardinge /* Currently only TCPv4 S.O. is supported. */ 8040d160211SJeremy Fitzhardinge if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) { 8050d160211SJeremy Fitzhardinge if (net_ratelimit()) 8060d160211SJeremy Fitzhardinge printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type); 8070d160211SJeremy Fitzhardinge return -EINVAL; 8080d160211SJeremy Fitzhardinge } 8090d160211SJeremy Fitzhardinge 8100d160211SJeremy Fitzhardinge skb_shinfo(skb)->gso_size = gso->u.gso.size; 8110d160211SJeremy Fitzhardinge skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; 8120d160211SJeremy Fitzhardinge 8130d160211SJeremy Fitzhardinge /* Header must be checked, and gso_segs computed. */ 8140d160211SJeremy Fitzhardinge skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 8150d160211SJeremy Fitzhardinge skb_shinfo(skb)->gso_segs = 0; 8160d160211SJeremy Fitzhardinge 8170d160211SJeremy Fitzhardinge return 0; 8180d160211SJeremy Fitzhardinge } 8190d160211SJeremy Fitzhardinge 8200d160211SJeremy Fitzhardinge static RING_IDX xennet_fill_frags(struct netfront_info *np, 8210d160211SJeremy Fitzhardinge struct sk_buff *skb, 8220d160211SJeremy Fitzhardinge struct sk_buff_head *list) 8230d160211SJeremy Fitzhardinge { 8240d160211SJeremy Fitzhardinge struct skb_shared_info *shinfo = skb_shinfo(skb); 8250d160211SJeremy Fitzhardinge int nr_frags = shinfo->nr_frags; 8260d160211SJeremy Fitzhardinge RING_IDX cons = np->rx.rsp_cons; 8270d160211SJeremy Fitzhardinge struct sk_buff *nskb; 8280d160211SJeremy Fitzhardinge 8290d160211SJeremy Fitzhardinge while ((nskb = __skb_dequeue(list))) { 8300d160211SJeremy Fitzhardinge struct xen_netif_rx_response *rx = 8310d160211SJeremy Fitzhardinge RING_GET_RESPONSE(&np->rx, ++cons); 83201c68026SIan Campbell skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; 8330d160211SJeremy Fitzhardinge 83401c68026SIan Campbell __skb_fill_page_desc(skb, nr_frags, 83501c68026SIan Campbell skb_frag_page(nfrag), 83601c68026SIan Campbell rx->offset, rx->status); 8370d160211SJeremy Fitzhardinge 8380d160211SJeremy Fitzhardinge skb->data_len += rx->status; 8390d160211SJeremy Fitzhardinge 8400d160211SJeremy Fitzhardinge skb_shinfo(nskb)->nr_frags = 0; 8410d160211SJeremy Fitzhardinge kfree_skb(nskb); 8420d160211SJeremy Fitzhardinge 8430d160211SJeremy Fitzhardinge nr_frags++; 8440d160211SJeremy Fitzhardinge } 8450d160211SJeremy Fitzhardinge 8460d160211SJeremy Fitzhardinge shinfo->nr_frags = nr_frags; 8470d160211SJeremy Fitzhardinge return cons; 8480d160211SJeremy Fitzhardinge } 8490d160211SJeremy Fitzhardinge 850e0ce4af9SIan Campbell static int checksum_setup(struct net_device *dev, struct sk_buff *skb) 8510d160211SJeremy Fitzhardinge { 8520d160211SJeremy Fitzhardinge struct iphdr *iph; 8530d160211SJeremy Fitzhardinge unsigned char *th; 8540d160211SJeremy Fitzhardinge int err = -EPROTO; 855e0ce4af9SIan Campbell int recalculate_partial_csum = 0; 856e0ce4af9SIan Campbell 857e0ce4af9SIan Campbell /* 858e0ce4af9SIan Campbell * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy 859e0ce4af9SIan Campbell * peers can fail to set NETRXF_csum_blank when sending a GSO 860e0ce4af9SIan Campbell * frame. In this case force the SKB to CHECKSUM_PARTIAL and 861e0ce4af9SIan Campbell * recalculate the partial checksum. 862e0ce4af9SIan Campbell */ 863e0ce4af9SIan Campbell if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { 864e0ce4af9SIan Campbell struct netfront_info *np = netdev_priv(dev); 865e0ce4af9SIan Campbell np->rx_gso_checksum_fixup++; 866e0ce4af9SIan Campbell skb->ip_summed = CHECKSUM_PARTIAL; 867e0ce4af9SIan Campbell recalculate_partial_csum = 1; 868e0ce4af9SIan Campbell } 869e0ce4af9SIan Campbell 870e0ce4af9SIan Campbell /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ 871e0ce4af9SIan Campbell if (skb->ip_summed != CHECKSUM_PARTIAL) 872e0ce4af9SIan Campbell return 0; 8730d160211SJeremy Fitzhardinge 8740d160211SJeremy Fitzhardinge if (skb->protocol != htons(ETH_P_IP)) 8750d160211SJeremy Fitzhardinge goto out; 8760d160211SJeremy Fitzhardinge 8770d160211SJeremy Fitzhardinge iph = (void *)skb->data; 8780d160211SJeremy Fitzhardinge th = skb->data + 4 * iph->ihl; 8790d160211SJeremy Fitzhardinge if (th >= skb_tail_pointer(skb)) 8800d160211SJeremy Fitzhardinge goto out; 8810d160211SJeremy Fitzhardinge 8820d160211SJeremy Fitzhardinge skb->csum_start = th - skb->head; 8830d160211SJeremy Fitzhardinge switch (iph->protocol) { 8840d160211SJeremy Fitzhardinge case IPPROTO_TCP: 8850d160211SJeremy Fitzhardinge skb->csum_offset = offsetof(struct tcphdr, check); 886e0ce4af9SIan Campbell 887e0ce4af9SIan Campbell if (recalculate_partial_csum) { 888e0ce4af9SIan Campbell struct tcphdr *tcph = (struct tcphdr *)th; 889e0ce4af9SIan Campbell tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 890e0ce4af9SIan Campbell skb->len - iph->ihl*4, 891e0ce4af9SIan Campbell IPPROTO_TCP, 0); 892e0ce4af9SIan Campbell } 8930d160211SJeremy Fitzhardinge break; 8940d160211SJeremy Fitzhardinge case IPPROTO_UDP: 8950d160211SJeremy Fitzhardinge skb->csum_offset = offsetof(struct udphdr, check); 896e0ce4af9SIan Campbell 897e0ce4af9SIan Campbell if (recalculate_partial_csum) { 898e0ce4af9SIan Campbell struct udphdr *udph = (struct udphdr *)th; 899e0ce4af9SIan Campbell udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 900e0ce4af9SIan Campbell skb->len - iph->ihl*4, 901e0ce4af9SIan Campbell IPPROTO_UDP, 0); 902e0ce4af9SIan Campbell } 9030d160211SJeremy Fitzhardinge break; 9040d160211SJeremy Fitzhardinge default: 9050d160211SJeremy Fitzhardinge if (net_ratelimit()) 9060d160211SJeremy Fitzhardinge printk(KERN_ERR "Attempting to checksum a non-" 9070d160211SJeremy Fitzhardinge "TCP/UDP packet, dropping a protocol" 9080d160211SJeremy Fitzhardinge " %d packet", iph->protocol); 9090d160211SJeremy Fitzhardinge goto out; 9100d160211SJeremy Fitzhardinge } 9110d160211SJeremy Fitzhardinge 9120d160211SJeremy Fitzhardinge if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb)) 9130d160211SJeremy Fitzhardinge goto out; 9140d160211SJeremy Fitzhardinge 9150d160211SJeremy Fitzhardinge err = 0; 9160d160211SJeremy Fitzhardinge 9170d160211SJeremy Fitzhardinge out: 9180d160211SJeremy Fitzhardinge return err; 9190d160211SJeremy Fitzhardinge } 9200d160211SJeremy Fitzhardinge 9210d160211SJeremy Fitzhardinge static int handle_incoming_queue(struct net_device *dev, 9220d160211SJeremy Fitzhardinge struct sk_buff_head *rxq) 9230d160211SJeremy Fitzhardinge { 924e00f85beSstephen hemminger struct netfront_info *np = netdev_priv(dev); 925e00f85beSstephen hemminger struct netfront_stats *stats = this_cpu_ptr(np->stats); 9260d160211SJeremy Fitzhardinge int packets_dropped = 0; 9270d160211SJeremy Fitzhardinge struct sk_buff *skb; 9280d160211SJeremy Fitzhardinge 9290d160211SJeremy Fitzhardinge while ((skb = __skb_dequeue(rxq)) != NULL) { 9303683243bSIan Campbell int pull_to = NETFRONT_SKB_CB(skb)->pull_to; 9310d160211SJeremy Fitzhardinge 9323683243bSIan Campbell __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); 9330d160211SJeremy Fitzhardinge 9340d160211SJeremy Fitzhardinge /* Ethernet work: Delayed to here as it peeks the header. */ 9350d160211SJeremy Fitzhardinge skb->protocol = eth_type_trans(skb, dev); 9360d160211SJeremy Fitzhardinge 937e0ce4af9SIan Campbell if (checksum_setup(dev, skb)) { 9380d160211SJeremy Fitzhardinge kfree_skb(skb); 9390d160211SJeremy Fitzhardinge packets_dropped++; 94009f75cd7SJeff Garzik dev->stats.rx_errors++; 9410d160211SJeremy Fitzhardinge continue; 9420d160211SJeremy Fitzhardinge } 9430d160211SJeremy Fitzhardinge 944e00f85beSstephen hemminger u64_stats_update_begin(&stats->syncp); 945e00f85beSstephen hemminger stats->rx_packets++; 946e00f85beSstephen hemminger stats->rx_bytes += skb->len; 947e00f85beSstephen hemminger u64_stats_update_end(&stats->syncp); 9480d160211SJeremy Fitzhardinge 9490d160211SJeremy Fitzhardinge /* Pass it up. */ 9500d160211SJeremy Fitzhardinge netif_receive_skb(skb); 9510d160211SJeremy Fitzhardinge } 9520d160211SJeremy Fitzhardinge 9530d160211SJeremy Fitzhardinge return packets_dropped; 9540d160211SJeremy Fitzhardinge } 9550d160211SJeremy Fitzhardinge 956bea3348eSStephen Hemminger static int xennet_poll(struct napi_struct *napi, int budget) 9570d160211SJeremy Fitzhardinge { 958bea3348eSStephen Hemminger struct netfront_info *np = container_of(napi, struct netfront_info, napi); 959bea3348eSStephen Hemminger struct net_device *dev = np->netdev; 9600d160211SJeremy Fitzhardinge struct sk_buff *skb; 9610d160211SJeremy Fitzhardinge struct netfront_rx_info rinfo; 9620d160211SJeremy Fitzhardinge struct xen_netif_rx_response *rx = &rinfo.rx; 9630d160211SJeremy Fitzhardinge struct xen_netif_extra_info *extras = rinfo.extras; 9640d160211SJeremy Fitzhardinge RING_IDX i, rp; 965bea3348eSStephen Hemminger int work_done; 9660d160211SJeremy Fitzhardinge struct sk_buff_head rxq; 9670d160211SJeremy Fitzhardinge struct sk_buff_head errq; 9680d160211SJeremy Fitzhardinge struct sk_buff_head tmpq; 9690d160211SJeremy Fitzhardinge unsigned long flags; 9700d160211SJeremy Fitzhardinge int err; 9710d160211SJeremy Fitzhardinge 9720d160211SJeremy Fitzhardinge spin_lock(&np->rx_lock); 9730d160211SJeremy Fitzhardinge 9740d160211SJeremy Fitzhardinge skb_queue_head_init(&rxq); 9750d160211SJeremy Fitzhardinge skb_queue_head_init(&errq); 9760d160211SJeremy Fitzhardinge skb_queue_head_init(&tmpq); 9770d160211SJeremy Fitzhardinge 9780d160211SJeremy Fitzhardinge rp = np->rx.sring->rsp_prod; 9790d160211SJeremy Fitzhardinge rmb(); /* Ensure we see queued responses up to 'rp'. */ 9800d160211SJeremy Fitzhardinge 9810d160211SJeremy Fitzhardinge i = np->rx.rsp_cons; 9820d160211SJeremy Fitzhardinge work_done = 0; 9830d160211SJeremy Fitzhardinge while ((i != rp) && (work_done < budget)) { 9840d160211SJeremy Fitzhardinge memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); 9850d160211SJeremy Fitzhardinge memset(extras, 0, sizeof(rinfo.extras)); 9860d160211SJeremy Fitzhardinge 9870d160211SJeremy Fitzhardinge err = xennet_get_responses(np, &rinfo, rp, &tmpq); 9880d160211SJeremy Fitzhardinge 9890d160211SJeremy Fitzhardinge if (unlikely(err)) { 9900d160211SJeremy Fitzhardinge err: 9910d160211SJeremy Fitzhardinge while ((skb = __skb_dequeue(&tmpq))) 9920d160211SJeremy Fitzhardinge __skb_queue_tail(&errq, skb); 99309f75cd7SJeff Garzik dev->stats.rx_errors++; 9940d160211SJeremy Fitzhardinge i = np->rx.rsp_cons; 9950d160211SJeremy Fitzhardinge continue; 9960d160211SJeremy Fitzhardinge } 9970d160211SJeremy Fitzhardinge 9980d160211SJeremy Fitzhardinge skb = __skb_dequeue(&tmpq); 9990d160211SJeremy Fitzhardinge 10000d160211SJeremy Fitzhardinge if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { 10010d160211SJeremy Fitzhardinge struct xen_netif_extra_info *gso; 10020d160211SJeremy Fitzhardinge gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; 10030d160211SJeremy Fitzhardinge 10040d160211SJeremy Fitzhardinge if (unlikely(xennet_set_skb_gso(skb, gso))) { 10050d160211SJeremy Fitzhardinge __skb_queue_head(&tmpq, skb); 10060d160211SJeremy Fitzhardinge np->rx.rsp_cons += skb_queue_len(&tmpq); 10070d160211SJeremy Fitzhardinge goto err; 10080d160211SJeremy Fitzhardinge } 10090d160211SJeremy Fitzhardinge } 10100d160211SJeremy Fitzhardinge 10113683243bSIan Campbell NETFRONT_SKB_CB(skb)->pull_to = rx->status; 10123683243bSIan Campbell if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) 10133683243bSIan Campbell NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; 10140d160211SJeremy Fitzhardinge 10153683243bSIan Campbell skb_shinfo(skb)->frags[0].page_offset = rx->offset; 10163683243bSIan Campbell skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); 10173683243bSIan Campbell skb->data_len = rx->status; 10180d160211SJeremy Fitzhardinge 10190d160211SJeremy Fitzhardinge i = xennet_fill_frags(np, skb, &tmpq); 10200d160211SJeremy Fitzhardinge 10210d160211SJeremy Fitzhardinge /* 1022d9a58a78SIan Campbell * Truesize is the actual allocation size, even if the 1023d9a58a78SIan Campbell * allocation is only partially used. 10240d160211SJeremy Fitzhardinge */ 1025d9a58a78SIan Campbell skb->truesize += PAGE_SIZE * skb_shinfo(skb)->nr_frags; 10260d160211SJeremy Fitzhardinge skb->len += skb->data_len; 10270d160211SJeremy Fitzhardinge 1028f942dc25SIan Campbell if (rx->flags & XEN_NETRXF_csum_blank) 10290d160211SJeremy Fitzhardinge skb->ip_summed = CHECKSUM_PARTIAL; 1030f942dc25SIan Campbell else if (rx->flags & XEN_NETRXF_data_validated) 10310d160211SJeremy Fitzhardinge skb->ip_summed = CHECKSUM_UNNECESSARY; 10320d160211SJeremy Fitzhardinge 10330d160211SJeremy Fitzhardinge __skb_queue_tail(&rxq, skb); 10340d160211SJeremy Fitzhardinge 10350d160211SJeremy Fitzhardinge np->rx.rsp_cons = ++i; 10360d160211SJeremy Fitzhardinge work_done++; 10370d160211SJeremy Fitzhardinge } 10380d160211SJeremy Fitzhardinge 103956cfe5d0SWang Chen __skb_queue_purge(&errq); 10400d160211SJeremy Fitzhardinge 10410d160211SJeremy Fitzhardinge work_done -= handle_incoming_queue(dev, &rxq); 10420d160211SJeremy Fitzhardinge 10430d160211SJeremy Fitzhardinge /* If we get a callback with very few responses, reduce fill target. */ 10440d160211SJeremy Fitzhardinge /* NB. Note exponential increase, linear decrease. */ 10450d160211SJeremy Fitzhardinge if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > 10460d160211SJeremy Fitzhardinge ((3*np->rx_target) / 4)) && 10470d160211SJeremy Fitzhardinge (--np->rx_target < np->rx_min_target)) 10480d160211SJeremy Fitzhardinge np->rx_target = np->rx_min_target; 10490d160211SJeremy Fitzhardinge 10500d160211SJeremy Fitzhardinge xennet_alloc_rx_buffers(dev); 10510d160211SJeremy Fitzhardinge 10520d160211SJeremy Fitzhardinge if (work_done < budget) { 1053bea3348eSStephen Hemminger int more_to_do = 0; 1054bea3348eSStephen Hemminger 10550d160211SJeremy Fitzhardinge local_irq_save(flags); 10560d160211SJeremy Fitzhardinge 10570d160211SJeremy Fitzhardinge RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do); 10580d160211SJeremy Fitzhardinge if (!more_to_do) 1059288379f0SBen Hutchings __napi_complete(napi); 10600d160211SJeremy Fitzhardinge 10610d160211SJeremy Fitzhardinge local_irq_restore(flags); 10620d160211SJeremy Fitzhardinge } 10630d160211SJeremy Fitzhardinge 10640d160211SJeremy Fitzhardinge spin_unlock(&np->rx_lock); 10650d160211SJeremy Fitzhardinge 1066bea3348eSStephen Hemminger return work_done; 10670d160211SJeremy Fitzhardinge } 10680d160211SJeremy Fitzhardinge 10690d160211SJeremy Fitzhardinge static int xennet_change_mtu(struct net_device *dev, int mtu) 10700d160211SJeremy Fitzhardinge { 10719ecd1a75SWei Liu int max = xennet_can_sg(dev) ? 10729ecd1a75SWei Liu XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN; 10730d160211SJeremy Fitzhardinge 10740d160211SJeremy Fitzhardinge if (mtu > max) 10750d160211SJeremy Fitzhardinge return -EINVAL; 10760d160211SJeremy Fitzhardinge dev->mtu = mtu; 10770d160211SJeremy Fitzhardinge return 0; 10780d160211SJeremy Fitzhardinge } 10790d160211SJeremy Fitzhardinge 1080e00f85beSstephen hemminger static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev, 1081e00f85beSstephen hemminger struct rtnl_link_stats64 *tot) 1082e00f85beSstephen hemminger { 1083e00f85beSstephen hemminger struct netfront_info *np = netdev_priv(dev); 1084e00f85beSstephen hemminger int cpu; 1085e00f85beSstephen hemminger 1086e00f85beSstephen hemminger for_each_possible_cpu(cpu) { 1087e00f85beSstephen hemminger struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu); 1088e00f85beSstephen hemminger u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 1089e00f85beSstephen hemminger unsigned int start; 1090e00f85beSstephen hemminger 1091e00f85beSstephen hemminger do { 1092e00f85beSstephen hemminger start = u64_stats_fetch_begin_bh(&stats->syncp); 1093e00f85beSstephen hemminger 1094e00f85beSstephen hemminger rx_packets = stats->rx_packets; 1095e00f85beSstephen hemminger tx_packets = stats->tx_packets; 1096e00f85beSstephen hemminger rx_bytes = stats->rx_bytes; 1097e00f85beSstephen hemminger tx_bytes = stats->tx_bytes; 1098e00f85beSstephen hemminger } while (u64_stats_fetch_retry_bh(&stats->syncp, start)); 1099e00f85beSstephen hemminger 1100e00f85beSstephen hemminger tot->rx_packets += rx_packets; 1101e00f85beSstephen hemminger tot->tx_packets += tx_packets; 1102e00f85beSstephen hemminger tot->rx_bytes += rx_bytes; 1103e00f85beSstephen hemminger tot->tx_bytes += tx_bytes; 1104e00f85beSstephen hemminger } 1105e00f85beSstephen hemminger 1106e00f85beSstephen hemminger tot->rx_errors = dev->stats.rx_errors; 1107e00f85beSstephen hemminger tot->tx_dropped = dev->stats.tx_dropped; 1108e00f85beSstephen hemminger 1109e00f85beSstephen hemminger return tot; 1110e00f85beSstephen hemminger } 1111e00f85beSstephen hemminger 11120d160211SJeremy Fitzhardinge static void xennet_release_tx_bufs(struct netfront_info *np) 11130d160211SJeremy Fitzhardinge { 11140d160211SJeremy Fitzhardinge struct sk_buff *skb; 11150d160211SJeremy Fitzhardinge int i; 11160d160211SJeremy Fitzhardinge 11170d160211SJeremy Fitzhardinge for (i = 0; i < NET_TX_RING_SIZE; i++) { 11180d160211SJeremy Fitzhardinge /* Skip over entries which are actually freelist references */ 11191ffb40b8SIsaku Yamahata if (skb_entry_is_link(&np->tx_skbs[i])) 11200d160211SJeremy Fitzhardinge continue; 11210d160211SJeremy Fitzhardinge 11220d160211SJeremy Fitzhardinge skb = np->tx_skbs[i].skb; 11230d160211SJeremy Fitzhardinge gnttab_end_foreign_access_ref(np->grant_tx_ref[i], 11240d160211SJeremy Fitzhardinge GNTMAP_readonly); 11250d160211SJeremy Fitzhardinge gnttab_release_grant_reference(&np->gref_tx_head, 11260d160211SJeremy Fitzhardinge np->grant_tx_ref[i]); 11270d160211SJeremy Fitzhardinge np->grant_tx_ref[i] = GRANT_INVALID_REF; 11280d160211SJeremy Fitzhardinge add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i); 11290d160211SJeremy Fitzhardinge dev_kfree_skb_irq(skb); 11300d160211SJeremy Fitzhardinge } 11310d160211SJeremy Fitzhardinge } 11320d160211SJeremy Fitzhardinge 11330d160211SJeremy Fitzhardinge static void xennet_release_rx_bufs(struct netfront_info *np) 11340d160211SJeremy Fitzhardinge { 11350d160211SJeremy Fitzhardinge struct mmu_update *mmu = np->rx_mmu; 11360d160211SJeremy Fitzhardinge struct multicall_entry *mcl = np->rx_mcl; 11370d160211SJeremy Fitzhardinge struct sk_buff_head free_list; 11380d160211SJeremy Fitzhardinge struct sk_buff *skb; 11390d160211SJeremy Fitzhardinge unsigned long mfn; 11400d160211SJeremy Fitzhardinge int xfer = 0, noxfer = 0, unused = 0; 11410d160211SJeremy Fitzhardinge int id, ref; 11420d160211SJeremy Fitzhardinge 11430d160211SJeremy Fitzhardinge dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n", 11440d160211SJeremy Fitzhardinge __func__); 11450d160211SJeremy Fitzhardinge return; 11460d160211SJeremy Fitzhardinge 11470d160211SJeremy Fitzhardinge skb_queue_head_init(&free_list); 11480d160211SJeremy Fitzhardinge 11490d160211SJeremy Fitzhardinge spin_lock_bh(&np->rx_lock); 11500d160211SJeremy Fitzhardinge 11510d160211SJeremy Fitzhardinge for (id = 0; id < NET_RX_RING_SIZE; id++) { 11520d160211SJeremy Fitzhardinge ref = np->grant_rx_ref[id]; 11530d160211SJeremy Fitzhardinge if (ref == GRANT_INVALID_REF) { 11540d160211SJeremy Fitzhardinge unused++; 11550d160211SJeremy Fitzhardinge continue; 11560d160211SJeremy Fitzhardinge } 11570d160211SJeremy Fitzhardinge 11580d160211SJeremy Fitzhardinge skb = np->rx_skbs[id]; 11590d160211SJeremy Fitzhardinge mfn = gnttab_end_foreign_transfer_ref(ref); 11600d160211SJeremy Fitzhardinge gnttab_release_grant_reference(&np->gref_rx_head, ref); 11610d160211SJeremy Fitzhardinge np->grant_rx_ref[id] = GRANT_INVALID_REF; 11620d160211SJeremy Fitzhardinge 11630d160211SJeremy Fitzhardinge if (0 == mfn) { 11640d160211SJeremy Fitzhardinge skb_shinfo(skb)->nr_frags = 0; 11650d160211SJeremy Fitzhardinge dev_kfree_skb(skb); 11660d160211SJeremy Fitzhardinge noxfer++; 11670d160211SJeremy Fitzhardinge continue; 11680d160211SJeremy Fitzhardinge } 11690d160211SJeremy Fitzhardinge 11700d160211SJeremy Fitzhardinge if (!xen_feature(XENFEAT_auto_translated_physmap)) { 11710d160211SJeremy Fitzhardinge /* Remap the page. */ 117201c68026SIan Campbell const struct page *page = 117301c68026SIan Campbell skb_frag_page(&skb_shinfo(skb)->frags[0]); 11740d160211SJeremy Fitzhardinge unsigned long pfn = page_to_pfn(page); 11750d160211SJeremy Fitzhardinge void *vaddr = page_address(page); 11760d160211SJeremy Fitzhardinge 11770d160211SJeremy Fitzhardinge MULTI_update_va_mapping(mcl, (unsigned long)vaddr, 11780d160211SJeremy Fitzhardinge mfn_pte(mfn, PAGE_KERNEL), 11790d160211SJeremy Fitzhardinge 0); 11800d160211SJeremy Fitzhardinge mcl++; 11810d160211SJeremy Fitzhardinge mmu->ptr = ((u64)mfn << PAGE_SHIFT) 11820d160211SJeremy Fitzhardinge | MMU_MACHPHYS_UPDATE; 11830d160211SJeremy Fitzhardinge mmu->val = pfn; 11840d160211SJeremy Fitzhardinge mmu++; 11850d160211SJeremy Fitzhardinge 11860d160211SJeremy Fitzhardinge set_phys_to_machine(pfn, mfn); 11870d160211SJeremy Fitzhardinge } 11880d160211SJeremy Fitzhardinge __skb_queue_tail(&free_list, skb); 11890d160211SJeremy Fitzhardinge xfer++; 11900d160211SJeremy Fitzhardinge } 11910d160211SJeremy Fitzhardinge 11920d160211SJeremy Fitzhardinge dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n", 11930d160211SJeremy Fitzhardinge __func__, xfer, noxfer, unused); 11940d160211SJeremy Fitzhardinge 11950d160211SJeremy Fitzhardinge if (xfer) { 11960d160211SJeremy Fitzhardinge if (!xen_feature(XENFEAT_auto_translated_physmap)) { 11970d160211SJeremy Fitzhardinge /* Do all the remapping work and M2P updates. */ 11980d160211SJeremy Fitzhardinge MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu, 119979ea13ceSAl Viro NULL, DOMID_SELF); 12000d160211SJeremy Fitzhardinge mcl++; 12010d160211SJeremy Fitzhardinge HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl); 12020d160211SJeremy Fitzhardinge } 12030d160211SJeremy Fitzhardinge } 12040d160211SJeremy Fitzhardinge 120556cfe5d0SWang Chen __skb_queue_purge(&free_list); 12060d160211SJeremy Fitzhardinge 12070d160211SJeremy Fitzhardinge spin_unlock_bh(&np->rx_lock); 12080d160211SJeremy Fitzhardinge } 12090d160211SJeremy Fitzhardinge 12100d160211SJeremy Fitzhardinge static void xennet_uninit(struct net_device *dev) 12110d160211SJeremy Fitzhardinge { 12120d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 12130d160211SJeremy Fitzhardinge xennet_release_tx_bufs(np); 12140d160211SJeremy Fitzhardinge xennet_release_rx_bufs(np); 12150d160211SJeremy Fitzhardinge gnttab_free_grant_references(np->gref_tx_head); 12160d160211SJeremy Fitzhardinge gnttab_free_grant_references(np->gref_rx_head); 12170d160211SJeremy Fitzhardinge } 12180d160211SJeremy Fitzhardinge 1219c8f44affSMichał Mirosław static netdev_features_t xennet_fix_features(struct net_device *dev, 1220c8f44affSMichał Mirosław netdev_features_t features) 12218f7b01a1SEric Dumazet { 12228f7b01a1SEric Dumazet struct netfront_info *np = netdev_priv(dev); 12238f7b01a1SEric Dumazet int val; 12248f7b01a1SEric Dumazet 12258f7b01a1SEric Dumazet if (features & NETIF_F_SG) { 12268f7b01a1SEric Dumazet if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg", 12278f7b01a1SEric Dumazet "%d", &val) < 0) 12288f7b01a1SEric Dumazet val = 0; 12298f7b01a1SEric Dumazet 12308f7b01a1SEric Dumazet if (!val) 12318f7b01a1SEric Dumazet features &= ~NETIF_F_SG; 12328f7b01a1SEric Dumazet } 12338f7b01a1SEric Dumazet 12348f7b01a1SEric Dumazet if (features & NETIF_F_TSO) { 12358f7b01a1SEric Dumazet if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, 12368f7b01a1SEric Dumazet "feature-gso-tcpv4", "%d", &val) < 0) 12378f7b01a1SEric Dumazet val = 0; 12388f7b01a1SEric Dumazet 12398f7b01a1SEric Dumazet if (!val) 12408f7b01a1SEric Dumazet features &= ~NETIF_F_TSO; 12418f7b01a1SEric Dumazet } 12428f7b01a1SEric Dumazet 12438f7b01a1SEric Dumazet return features; 12448f7b01a1SEric Dumazet } 12458f7b01a1SEric Dumazet 1246c8f44affSMichał Mirosław static int xennet_set_features(struct net_device *dev, 1247c8f44affSMichał Mirosław netdev_features_t features) 12488f7b01a1SEric Dumazet { 12498f7b01a1SEric Dumazet if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { 12508f7b01a1SEric Dumazet netdev_info(dev, "Reducing MTU because no SG offload"); 12518f7b01a1SEric Dumazet dev->mtu = ETH_DATA_LEN; 12528f7b01a1SEric Dumazet } 12538f7b01a1SEric Dumazet 12548f7b01a1SEric Dumazet return 0; 12558f7b01a1SEric Dumazet } 12568f7b01a1SEric Dumazet 1257cf66f9d4SKonrad Rzeszutek Wilk static irqreturn_t xennet_interrupt(int irq, void *dev_id) 1258cf66f9d4SKonrad Rzeszutek Wilk { 1259cf66f9d4SKonrad Rzeszutek Wilk struct net_device *dev = dev_id; 1260cf66f9d4SKonrad Rzeszutek Wilk struct netfront_info *np = netdev_priv(dev); 1261cf66f9d4SKonrad Rzeszutek Wilk unsigned long flags; 1262cf66f9d4SKonrad Rzeszutek Wilk 1263cf66f9d4SKonrad Rzeszutek Wilk spin_lock_irqsave(&np->tx_lock, flags); 1264cf66f9d4SKonrad Rzeszutek Wilk 1265cf66f9d4SKonrad Rzeszutek Wilk if (likely(netif_carrier_ok(dev))) { 1266cf66f9d4SKonrad Rzeszutek Wilk xennet_tx_buf_gc(dev); 1267cf66f9d4SKonrad Rzeszutek Wilk /* Under tx_lock: protects access to rx shared-ring indexes. */ 1268cf66f9d4SKonrad Rzeszutek Wilk if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) 1269cf66f9d4SKonrad Rzeszutek Wilk napi_schedule(&np->napi); 1270cf66f9d4SKonrad Rzeszutek Wilk } 1271cf66f9d4SKonrad Rzeszutek Wilk 1272cf66f9d4SKonrad Rzeszutek Wilk spin_unlock_irqrestore(&np->tx_lock, flags); 1273cf66f9d4SKonrad Rzeszutek Wilk 1274cf66f9d4SKonrad Rzeszutek Wilk return IRQ_HANDLED; 1275cf66f9d4SKonrad Rzeszutek Wilk } 1276cf66f9d4SKonrad Rzeszutek Wilk 1277cf66f9d4SKonrad Rzeszutek Wilk #ifdef CONFIG_NET_POLL_CONTROLLER 1278cf66f9d4SKonrad Rzeszutek Wilk static void xennet_poll_controller(struct net_device *dev) 1279cf66f9d4SKonrad Rzeszutek Wilk { 1280cf66f9d4SKonrad Rzeszutek Wilk xennet_interrupt(0, dev); 1281cf66f9d4SKonrad Rzeszutek Wilk } 1282cf66f9d4SKonrad Rzeszutek Wilk #endif 1283cf66f9d4SKonrad Rzeszutek Wilk 12840a0b9d2eSStephen Hemminger static const struct net_device_ops xennet_netdev_ops = { 12850a0b9d2eSStephen Hemminger .ndo_open = xennet_open, 12860a0b9d2eSStephen Hemminger .ndo_uninit = xennet_uninit, 12870a0b9d2eSStephen Hemminger .ndo_stop = xennet_close, 12880a0b9d2eSStephen Hemminger .ndo_start_xmit = xennet_start_xmit, 12890a0b9d2eSStephen Hemminger .ndo_change_mtu = xennet_change_mtu, 1290e00f85beSstephen hemminger .ndo_get_stats64 = xennet_get_stats64, 12910a0b9d2eSStephen Hemminger .ndo_set_mac_address = eth_mac_addr, 12920a0b9d2eSStephen Hemminger .ndo_validate_addr = eth_validate_addr, 1293fb507934SMichał Mirosław .ndo_fix_features = xennet_fix_features, 1294fb507934SMichał Mirosław .ndo_set_features = xennet_set_features, 1295cf66f9d4SKonrad Rzeszutek Wilk #ifdef CONFIG_NET_POLL_CONTROLLER 1296cf66f9d4SKonrad Rzeszutek Wilk .ndo_poll_controller = xennet_poll_controller, 1297cf66f9d4SKonrad Rzeszutek Wilk #endif 12980a0b9d2eSStephen Hemminger }; 12990a0b9d2eSStephen Hemminger 13008e0e46bbSBill Pemberton static struct net_device *xennet_create_dev(struct xenbus_device *dev) 13010d160211SJeremy Fitzhardinge { 13020d160211SJeremy Fitzhardinge int i, err; 13030d160211SJeremy Fitzhardinge struct net_device *netdev; 13040d160211SJeremy Fitzhardinge struct netfront_info *np; 13050d160211SJeremy Fitzhardinge 13060d160211SJeremy Fitzhardinge netdev = alloc_etherdev(sizeof(struct netfront_info)); 130741de8d4cSJoe Perches if (!netdev) 13080d160211SJeremy Fitzhardinge return ERR_PTR(-ENOMEM); 13090d160211SJeremy Fitzhardinge 13100d160211SJeremy Fitzhardinge np = netdev_priv(netdev); 13110d160211SJeremy Fitzhardinge np->xbdev = dev; 13120d160211SJeremy Fitzhardinge 13130d160211SJeremy Fitzhardinge spin_lock_init(&np->tx_lock); 13140d160211SJeremy Fitzhardinge spin_lock_init(&np->rx_lock); 13150d160211SJeremy Fitzhardinge 13160d160211SJeremy Fitzhardinge skb_queue_head_init(&np->rx_batch); 13170d160211SJeremy Fitzhardinge np->rx_target = RX_DFL_MIN_TARGET; 13180d160211SJeremy Fitzhardinge np->rx_min_target = RX_DFL_MIN_TARGET; 13190d160211SJeremy Fitzhardinge np->rx_max_target = RX_MAX_TARGET; 13200d160211SJeremy Fitzhardinge 13210d160211SJeremy Fitzhardinge init_timer(&np->rx_refill_timer); 13220d160211SJeremy Fitzhardinge np->rx_refill_timer.data = (unsigned long)netdev; 13230d160211SJeremy Fitzhardinge np->rx_refill_timer.function = rx_refill_timeout; 13240d160211SJeremy Fitzhardinge 1325e00f85beSstephen hemminger err = -ENOMEM; 1326e00f85beSstephen hemminger np->stats = alloc_percpu(struct netfront_stats); 1327e00f85beSstephen hemminger if (np->stats == NULL) 1328e00f85beSstephen hemminger goto exit; 1329e00f85beSstephen hemminger 13300d160211SJeremy Fitzhardinge /* Initialise tx_skbs as a free chain containing every entry. */ 13310d160211SJeremy Fitzhardinge np->tx_skb_freelist = 0; 13320d160211SJeremy Fitzhardinge for (i = 0; i < NET_TX_RING_SIZE; i++) { 13331ffb40b8SIsaku Yamahata skb_entry_set_link(&np->tx_skbs[i], i+1); 13340d160211SJeremy Fitzhardinge np->grant_tx_ref[i] = GRANT_INVALID_REF; 13350d160211SJeremy Fitzhardinge } 13360d160211SJeremy Fitzhardinge 13370d160211SJeremy Fitzhardinge /* Clear out rx_skbs */ 13380d160211SJeremy Fitzhardinge for (i = 0; i < NET_RX_RING_SIZE; i++) { 13390d160211SJeremy Fitzhardinge np->rx_skbs[i] = NULL; 13400d160211SJeremy Fitzhardinge np->grant_rx_ref[i] = GRANT_INVALID_REF; 13410d160211SJeremy Fitzhardinge } 13420d160211SJeremy Fitzhardinge 13430d160211SJeremy Fitzhardinge /* A grant for every tx ring slot */ 13440d160211SJeremy Fitzhardinge if (gnttab_alloc_grant_references(TX_MAX_TARGET, 13450d160211SJeremy Fitzhardinge &np->gref_tx_head) < 0) { 13460d160211SJeremy Fitzhardinge printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n"); 13470d160211SJeremy Fitzhardinge err = -ENOMEM; 1348e00f85beSstephen hemminger goto exit_free_stats; 13490d160211SJeremy Fitzhardinge } 13500d160211SJeremy Fitzhardinge /* A grant for every rx ring slot */ 13510d160211SJeremy Fitzhardinge if (gnttab_alloc_grant_references(RX_MAX_TARGET, 13520d160211SJeremy Fitzhardinge &np->gref_rx_head) < 0) { 13530d160211SJeremy Fitzhardinge printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n"); 13540d160211SJeremy Fitzhardinge err = -ENOMEM; 13550d160211SJeremy Fitzhardinge goto exit_free_tx; 13560d160211SJeremy Fitzhardinge } 13570d160211SJeremy Fitzhardinge 13580a0b9d2eSStephen Hemminger netdev->netdev_ops = &xennet_netdev_ops; 13590a0b9d2eSStephen Hemminger 1360bea3348eSStephen Hemminger netif_napi_add(netdev, &np->napi, xennet_poll, 64); 1361fb507934SMichał Mirosław netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | 1362fb507934SMichał Mirosław NETIF_F_GSO_ROBUST; 1363fb507934SMichał Mirosław netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO; 13640d160211SJeremy Fitzhardinge 1365fc3e5941SIan Campbell /* 1366fc3e5941SIan Campbell * Assume that all hw features are available for now. This set 1367fc3e5941SIan Campbell * will be adjusted by the call to netdev_update_features() in 1368fc3e5941SIan Campbell * xennet_connect() which is the earliest point where we can 1369fc3e5941SIan Campbell * negotiate with the backend regarding supported features. 1370fc3e5941SIan Campbell */ 1371fc3e5941SIan Campbell netdev->features |= netdev->hw_features; 1372fc3e5941SIan Campbell 13730d160211SJeremy Fitzhardinge SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops); 13740d160211SJeremy Fitzhardinge SET_NETDEV_DEV(netdev, &dev->dev); 13750d160211SJeremy Fitzhardinge 13769ecd1a75SWei Liu netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER); 13779ecd1a75SWei Liu 13780d160211SJeremy Fitzhardinge np->netdev = netdev; 13790d160211SJeremy Fitzhardinge 13800d160211SJeremy Fitzhardinge netif_carrier_off(netdev); 13810d160211SJeremy Fitzhardinge 13820d160211SJeremy Fitzhardinge return netdev; 13830d160211SJeremy Fitzhardinge 13840d160211SJeremy Fitzhardinge exit_free_tx: 13850d160211SJeremy Fitzhardinge gnttab_free_grant_references(np->gref_tx_head); 1386e00f85beSstephen hemminger exit_free_stats: 1387e00f85beSstephen hemminger free_percpu(np->stats); 13880d160211SJeremy Fitzhardinge exit: 13890d160211SJeremy Fitzhardinge free_netdev(netdev); 13900d160211SJeremy Fitzhardinge return ERR_PTR(err); 13910d160211SJeremy Fitzhardinge } 13920d160211SJeremy Fitzhardinge 13930d160211SJeremy Fitzhardinge /** 13940d160211SJeremy Fitzhardinge * Entry point to this code when a new device is created. Allocate the basic 13950d160211SJeremy Fitzhardinge * structures and the ring buffers for communication with the backend, and 13960d160211SJeremy Fitzhardinge * inform the backend of the appropriate details for those. 13970d160211SJeremy Fitzhardinge */ 13988e0e46bbSBill Pemberton static int netfront_probe(struct xenbus_device *dev, 13990d160211SJeremy Fitzhardinge const struct xenbus_device_id *id) 14000d160211SJeremy Fitzhardinge { 14010d160211SJeremy Fitzhardinge int err; 14020d160211SJeremy Fitzhardinge struct net_device *netdev; 14030d160211SJeremy Fitzhardinge struct netfront_info *info; 14040d160211SJeremy Fitzhardinge 14050d160211SJeremy Fitzhardinge netdev = xennet_create_dev(dev); 14060d160211SJeremy Fitzhardinge if (IS_ERR(netdev)) { 14070d160211SJeremy Fitzhardinge err = PTR_ERR(netdev); 14080d160211SJeremy Fitzhardinge xenbus_dev_fatal(dev, err, "creating netdev"); 14090d160211SJeremy Fitzhardinge return err; 14100d160211SJeremy Fitzhardinge } 14110d160211SJeremy Fitzhardinge 14120d160211SJeremy Fitzhardinge info = netdev_priv(netdev); 14131b713e00SGreg Kroah-Hartman dev_set_drvdata(&dev->dev, info); 14140d160211SJeremy Fitzhardinge 14150d160211SJeremy Fitzhardinge err = register_netdev(info->netdev); 14160d160211SJeremy Fitzhardinge if (err) { 14170d160211SJeremy Fitzhardinge printk(KERN_WARNING "%s: register_netdev err=%d\n", 14180d160211SJeremy Fitzhardinge __func__, err); 14190d160211SJeremy Fitzhardinge goto fail; 14200d160211SJeremy Fitzhardinge } 14210d160211SJeremy Fitzhardinge 14220d160211SJeremy Fitzhardinge err = xennet_sysfs_addif(info->netdev); 14230d160211SJeremy Fitzhardinge if (err) { 14240d160211SJeremy Fitzhardinge unregister_netdev(info->netdev); 14250d160211SJeremy Fitzhardinge printk(KERN_WARNING "%s: add sysfs failed err=%d\n", 14260d160211SJeremy Fitzhardinge __func__, err); 14270d160211SJeremy Fitzhardinge goto fail; 14280d160211SJeremy Fitzhardinge } 14290d160211SJeremy Fitzhardinge 14300d160211SJeremy Fitzhardinge return 0; 14310d160211SJeremy Fitzhardinge 14320d160211SJeremy Fitzhardinge fail: 14330d160211SJeremy Fitzhardinge free_netdev(netdev); 14341b713e00SGreg Kroah-Hartman dev_set_drvdata(&dev->dev, NULL); 14350d160211SJeremy Fitzhardinge return err; 14360d160211SJeremy Fitzhardinge } 14370d160211SJeremy Fitzhardinge 14380d160211SJeremy Fitzhardinge static void xennet_end_access(int ref, void *page) 14390d160211SJeremy Fitzhardinge { 14400d160211SJeremy Fitzhardinge /* This frees the page as a side-effect */ 14410d160211SJeremy Fitzhardinge if (ref != GRANT_INVALID_REF) 14420d160211SJeremy Fitzhardinge gnttab_end_foreign_access(ref, 0, (unsigned long)page); 14430d160211SJeremy Fitzhardinge } 14440d160211SJeremy Fitzhardinge 14450d160211SJeremy Fitzhardinge static void xennet_disconnect_backend(struct netfront_info *info) 14460d160211SJeremy Fitzhardinge { 14470d160211SJeremy Fitzhardinge /* Stop old i/f to prevent errors whilst we rebuild the state. */ 14480d160211SJeremy Fitzhardinge spin_lock_bh(&info->rx_lock); 14490d160211SJeremy Fitzhardinge spin_lock_irq(&info->tx_lock); 14500d160211SJeremy Fitzhardinge netif_carrier_off(info->netdev); 14510d160211SJeremy Fitzhardinge spin_unlock_irq(&info->tx_lock); 14520d160211SJeremy Fitzhardinge spin_unlock_bh(&info->rx_lock); 14530d160211SJeremy Fitzhardinge 14540d160211SJeremy Fitzhardinge if (info->netdev->irq) 14550d160211SJeremy Fitzhardinge unbind_from_irqhandler(info->netdev->irq, info->netdev); 14560d160211SJeremy Fitzhardinge info->evtchn = info->netdev->irq = 0; 14570d160211SJeremy Fitzhardinge 14580d160211SJeremy Fitzhardinge /* End access and free the pages */ 14590d160211SJeremy Fitzhardinge xennet_end_access(info->tx_ring_ref, info->tx.sring); 14600d160211SJeremy Fitzhardinge xennet_end_access(info->rx_ring_ref, info->rx.sring); 14610d160211SJeremy Fitzhardinge 14620d160211SJeremy Fitzhardinge info->tx_ring_ref = GRANT_INVALID_REF; 14630d160211SJeremy Fitzhardinge info->rx_ring_ref = GRANT_INVALID_REF; 14640d160211SJeremy Fitzhardinge info->tx.sring = NULL; 14650d160211SJeremy Fitzhardinge info->rx.sring = NULL; 14660d160211SJeremy Fitzhardinge } 14670d160211SJeremy Fitzhardinge 14680d160211SJeremy Fitzhardinge /** 14690d160211SJeremy Fitzhardinge * We are reconnecting to the backend, due to a suspend/resume, or a backend 14700d160211SJeremy Fitzhardinge * driver restart. We tear down our netif structure and recreate it, but 14710d160211SJeremy Fitzhardinge * leave the device-layer structures intact so that this is transparent to the 14720d160211SJeremy Fitzhardinge * rest of the kernel. 14730d160211SJeremy Fitzhardinge */ 14740d160211SJeremy Fitzhardinge static int netfront_resume(struct xenbus_device *dev) 14750d160211SJeremy Fitzhardinge { 14761b713e00SGreg Kroah-Hartman struct netfront_info *info = dev_get_drvdata(&dev->dev); 14770d160211SJeremy Fitzhardinge 14780d160211SJeremy Fitzhardinge dev_dbg(&dev->dev, "%s\n", dev->nodename); 14790d160211SJeremy Fitzhardinge 14800d160211SJeremy Fitzhardinge xennet_disconnect_backend(info); 14810d160211SJeremy Fitzhardinge return 0; 14820d160211SJeremy Fitzhardinge } 14830d160211SJeremy Fitzhardinge 14840d160211SJeremy Fitzhardinge static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) 14850d160211SJeremy Fitzhardinge { 14860d160211SJeremy Fitzhardinge char *s, *e, *macstr; 14870d160211SJeremy Fitzhardinge int i; 14880d160211SJeremy Fitzhardinge 14890d160211SJeremy Fitzhardinge macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); 14900d160211SJeremy Fitzhardinge if (IS_ERR(macstr)) 14910d160211SJeremy Fitzhardinge return PTR_ERR(macstr); 14920d160211SJeremy Fitzhardinge 14930d160211SJeremy Fitzhardinge for (i = 0; i < ETH_ALEN; i++) { 14940d160211SJeremy Fitzhardinge mac[i] = simple_strtoul(s, &e, 16); 14950d160211SJeremy Fitzhardinge if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { 14960d160211SJeremy Fitzhardinge kfree(macstr); 14970d160211SJeremy Fitzhardinge return -ENOENT; 14980d160211SJeremy Fitzhardinge } 14990d160211SJeremy Fitzhardinge s = e+1; 15000d160211SJeremy Fitzhardinge } 15010d160211SJeremy Fitzhardinge 15020d160211SJeremy Fitzhardinge kfree(macstr); 15030d160211SJeremy Fitzhardinge return 0; 15040d160211SJeremy Fitzhardinge } 15050d160211SJeremy Fitzhardinge 15060d160211SJeremy Fitzhardinge static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info) 15070d160211SJeremy Fitzhardinge { 15080d160211SJeremy Fitzhardinge struct xen_netif_tx_sring *txs; 15090d160211SJeremy Fitzhardinge struct xen_netif_rx_sring *rxs; 15100d160211SJeremy Fitzhardinge int err; 15110d160211SJeremy Fitzhardinge struct net_device *netdev = info->netdev; 15120d160211SJeremy Fitzhardinge 15130d160211SJeremy Fitzhardinge info->tx_ring_ref = GRANT_INVALID_REF; 15140d160211SJeremy Fitzhardinge info->rx_ring_ref = GRANT_INVALID_REF; 15150d160211SJeremy Fitzhardinge info->rx.sring = NULL; 15160d160211SJeremy Fitzhardinge info->tx.sring = NULL; 15170d160211SJeremy Fitzhardinge netdev->irq = 0; 15180d160211SJeremy Fitzhardinge 15190d160211SJeremy Fitzhardinge err = xen_net_read_mac(dev, netdev->dev_addr); 15200d160211SJeremy Fitzhardinge if (err) { 15210d160211SJeremy Fitzhardinge xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); 15220d160211SJeremy Fitzhardinge goto fail; 15230d160211SJeremy Fitzhardinge } 15240d160211SJeremy Fitzhardinge 1525a144ff09SIan Campbell txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 15260d160211SJeremy Fitzhardinge if (!txs) { 15270d160211SJeremy Fitzhardinge err = -ENOMEM; 15280d160211SJeremy Fitzhardinge xenbus_dev_fatal(dev, err, "allocating tx ring page"); 15290d160211SJeremy Fitzhardinge goto fail; 15300d160211SJeremy Fitzhardinge } 15310d160211SJeremy Fitzhardinge SHARED_RING_INIT(txs); 15320d160211SJeremy Fitzhardinge FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); 15330d160211SJeremy Fitzhardinge 15340d160211SJeremy Fitzhardinge err = xenbus_grant_ring(dev, virt_to_mfn(txs)); 15351ca2983aSWei Liu if (err < 0) 15361ca2983aSWei Liu goto grant_tx_ring_fail; 15370d160211SJeremy Fitzhardinge 15380d160211SJeremy Fitzhardinge info->tx_ring_ref = err; 1539a144ff09SIan Campbell rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 15400d160211SJeremy Fitzhardinge if (!rxs) { 15410d160211SJeremy Fitzhardinge err = -ENOMEM; 15420d160211SJeremy Fitzhardinge xenbus_dev_fatal(dev, err, "allocating rx ring page"); 15431ca2983aSWei Liu goto alloc_rx_ring_fail; 15440d160211SJeremy Fitzhardinge } 15450d160211SJeremy Fitzhardinge SHARED_RING_INIT(rxs); 15460d160211SJeremy Fitzhardinge FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); 15470d160211SJeremy Fitzhardinge 15480d160211SJeremy Fitzhardinge err = xenbus_grant_ring(dev, virt_to_mfn(rxs)); 15491ca2983aSWei Liu if (err < 0) 15501ca2983aSWei Liu goto grant_rx_ring_fail; 15510d160211SJeremy Fitzhardinge info->rx_ring_ref = err; 15520d160211SJeremy Fitzhardinge 15530d160211SJeremy Fitzhardinge err = xenbus_alloc_evtchn(dev, &info->evtchn); 15540d160211SJeremy Fitzhardinge if (err) 15551ca2983aSWei Liu goto alloc_evtchn_fail; 15560d160211SJeremy Fitzhardinge 15570d160211SJeremy Fitzhardinge err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt, 1558ab392d2dSJavier Martinez Canillas 0, netdev->name, netdev); 15590d160211SJeremy Fitzhardinge if (err < 0) 15601ca2983aSWei Liu goto bind_fail; 15610d160211SJeremy Fitzhardinge netdev->irq = err; 15620d160211SJeremy Fitzhardinge return 0; 15630d160211SJeremy Fitzhardinge 15641ca2983aSWei Liu /* If we fail to setup netfront, it is safe to just revoke access to 15651ca2983aSWei Liu * granted pages because backend is not accessing it at this point. 15661ca2983aSWei Liu */ 15671ca2983aSWei Liu bind_fail: 15681ca2983aSWei Liu xenbus_free_evtchn(dev, info->evtchn); 15691ca2983aSWei Liu alloc_evtchn_fail: 15701ca2983aSWei Liu gnttab_end_foreign_access_ref(info->rx_ring_ref, 0); 15711ca2983aSWei Liu grant_rx_ring_fail: 15721ca2983aSWei Liu free_page((unsigned long)rxs); 15731ca2983aSWei Liu alloc_rx_ring_fail: 15741ca2983aSWei Liu gnttab_end_foreign_access_ref(info->tx_ring_ref, 0); 15751ca2983aSWei Liu grant_tx_ring_fail: 15761ca2983aSWei Liu free_page((unsigned long)txs); 15770d160211SJeremy Fitzhardinge fail: 15780d160211SJeremy Fitzhardinge return err; 15790d160211SJeremy Fitzhardinge } 15800d160211SJeremy Fitzhardinge 15810d160211SJeremy Fitzhardinge /* Common code used when first setting up, and when resuming. */ 1582f502bf2bSIan Campbell static int talk_to_netback(struct xenbus_device *dev, 15830d160211SJeremy Fitzhardinge struct netfront_info *info) 15840d160211SJeremy Fitzhardinge { 15850d160211SJeremy Fitzhardinge const char *message; 15860d160211SJeremy Fitzhardinge struct xenbus_transaction xbt; 15870d160211SJeremy Fitzhardinge int err; 15880d160211SJeremy Fitzhardinge 15890d160211SJeremy Fitzhardinge /* Create shared ring, alloc event channel. */ 15900d160211SJeremy Fitzhardinge err = setup_netfront(dev, info); 15910d160211SJeremy Fitzhardinge if (err) 15920d160211SJeremy Fitzhardinge goto out; 15930d160211SJeremy Fitzhardinge 15940d160211SJeremy Fitzhardinge again: 15950d160211SJeremy Fitzhardinge err = xenbus_transaction_start(&xbt); 15960d160211SJeremy Fitzhardinge if (err) { 15970d160211SJeremy Fitzhardinge xenbus_dev_fatal(dev, err, "starting transaction"); 15980d160211SJeremy Fitzhardinge goto destroy_ring; 15990d160211SJeremy Fitzhardinge } 16000d160211SJeremy Fitzhardinge 16010d160211SJeremy Fitzhardinge err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u", 16020d160211SJeremy Fitzhardinge info->tx_ring_ref); 16030d160211SJeremy Fitzhardinge if (err) { 16040d160211SJeremy Fitzhardinge message = "writing tx ring-ref"; 16050d160211SJeremy Fitzhardinge goto abort_transaction; 16060d160211SJeremy Fitzhardinge } 16070d160211SJeremy Fitzhardinge err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u", 16080d160211SJeremy Fitzhardinge info->rx_ring_ref); 16090d160211SJeremy Fitzhardinge if (err) { 16100d160211SJeremy Fitzhardinge message = "writing rx ring-ref"; 16110d160211SJeremy Fitzhardinge goto abort_transaction; 16120d160211SJeremy Fitzhardinge } 16130d160211SJeremy Fitzhardinge err = xenbus_printf(xbt, dev->nodename, 16140d160211SJeremy Fitzhardinge "event-channel", "%u", info->evtchn); 16150d160211SJeremy Fitzhardinge if (err) { 16160d160211SJeremy Fitzhardinge message = "writing event-channel"; 16170d160211SJeremy Fitzhardinge goto abort_transaction; 16180d160211SJeremy Fitzhardinge } 16190d160211SJeremy Fitzhardinge 16200d160211SJeremy Fitzhardinge err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", 16210d160211SJeremy Fitzhardinge 1); 16220d160211SJeremy Fitzhardinge if (err) { 16230d160211SJeremy Fitzhardinge message = "writing request-rx-copy"; 16240d160211SJeremy Fitzhardinge goto abort_transaction; 16250d160211SJeremy Fitzhardinge } 16260d160211SJeremy Fitzhardinge 16270d160211SJeremy Fitzhardinge err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); 16280d160211SJeremy Fitzhardinge if (err) { 16290d160211SJeremy Fitzhardinge message = "writing feature-rx-notify"; 16300d160211SJeremy Fitzhardinge goto abort_transaction; 16310d160211SJeremy Fitzhardinge } 16320d160211SJeremy Fitzhardinge 16330d160211SJeremy Fitzhardinge err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); 16340d160211SJeremy Fitzhardinge if (err) { 16350d160211SJeremy Fitzhardinge message = "writing feature-sg"; 16360d160211SJeremy Fitzhardinge goto abort_transaction; 16370d160211SJeremy Fitzhardinge } 16380d160211SJeremy Fitzhardinge 16390d160211SJeremy Fitzhardinge err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); 16400d160211SJeremy Fitzhardinge if (err) { 16410d160211SJeremy Fitzhardinge message = "writing feature-gso-tcpv4"; 16420d160211SJeremy Fitzhardinge goto abort_transaction; 16430d160211SJeremy Fitzhardinge } 16440d160211SJeremy Fitzhardinge 16450d160211SJeremy Fitzhardinge err = xenbus_transaction_end(xbt, 0); 16460d160211SJeremy Fitzhardinge if (err) { 16470d160211SJeremy Fitzhardinge if (err == -EAGAIN) 16480d160211SJeremy Fitzhardinge goto again; 16490d160211SJeremy Fitzhardinge xenbus_dev_fatal(dev, err, "completing transaction"); 16500d160211SJeremy Fitzhardinge goto destroy_ring; 16510d160211SJeremy Fitzhardinge } 16520d160211SJeremy Fitzhardinge 16530d160211SJeremy Fitzhardinge return 0; 16540d160211SJeremy Fitzhardinge 16550d160211SJeremy Fitzhardinge abort_transaction: 16560d160211SJeremy Fitzhardinge xenbus_transaction_end(xbt, 1); 16570d160211SJeremy Fitzhardinge xenbus_dev_fatal(dev, err, "%s", message); 16580d160211SJeremy Fitzhardinge destroy_ring: 16590d160211SJeremy Fitzhardinge xennet_disconnect_backend(info); 16600d160211SJeremy Fitzhardinge out: 16610d160211SJeremy Fitzhardinge return err; 16620d160211SJeremy Fitzhardinge } 16630d160211SJeremy Fitzhardinge 16640d160211SJeremy Fitzhardinge static int xennet_connect(struct net_device *dev) 16650d160211SJeremy Fitzhardinge { 16660d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(dev); 16670d160211SJeremy Fitzhardinge int i, requeue_idx, err; 16680d160211SJeremy Fitzhardinge struct sk_buff *skb; 16690d160211SJeremy Fitzhardinge grant_ref_t ref; 16700d160211SJeremy Fitzhardinge struct xen_netif_rx_request *req; 16710d160211SJeremy Fitzhardinge unsigned int feature_rx_copy; 16720d160211SJeremy Fitzhardinge 16730d160211SJeremy Fitzhardinge err = xenbus_scanf(XBT_NIL, np->xbdev->otherend, 16740d160211SJeremy Fitzhardinge "feature-rx-copy", "%u", &feature_rx_copy); 16750d160211SJeremy Fitzhardinge if (err != 1) 16760d160211SJeremy Fitzhardinge feature_rx_copy = 0; 16770d160211SJeremy Fitzhardinge 16780d160211SJeremy Fitzhardinge if (!feature_rx_copy) { 16790d160211SJeremy Fitzhardinge dev_info(&dev->dev, 1680898eb71cSJoe Perches "backend does not support copying receive path\n"); 16810d160211SJeremy Fitzhardinge return -ENODEV; 16820d160211SJeremy Fitzhardinge } 16830d160211SJeremy Fitzhardinge 1684f502bf2bSIan Campbell err = talk_to_netback(np->xbdev, np); 16850d160211SJeremy Fitzhardinge if (err) 16860d160211SJeremy Fitzhardinge return err; 16870d160211SJeremy Fitzhardinge 16881ba37c51SIan Campbell rtnl_lock(); 1689fb507934SMichał Mirosław netdev_update_features(dev); 16901ba37c51SIan Campbell rtnl_unlock(); 16910d160211SJeremy Fitzhardinge 16920d160211SJeremy Fitzhardinge spin_lock_bh(&np->rx_lock); 16930d160211SJeremy Fitzhardinge spin_lock_irq(&np->tx_lock); 16940d160211SJeremy Fitzhardinge 16950d160211SJeremy Fitzhardinge /* Step 1: Discard all pending TX packet fragments. */ 16960d160211SJeremy Fitzhardinge xennet_release_tx_bufs(np); 16970d160211SJeremy Fitzhardinge 16980d160211SJeremy Fitzhardinge /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ 16990d160211SJeremy Fitzhardinge for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { 170001c68026SIan Campbell skb_frag_t *frag; 170101c68026SIan Campbell const struct page *page; 17020d160211SJeremy Fitzhardinge if (!np->rx_skbs[i]) 17030d160211SJeremy Fitzhardinge continue; 17040d160211SJeremy Fitzhardinge 17050d160211SJeremy Fitzhardinge skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i); 17060d160211SJeremy Fitzhardinge ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); 17070d160211SJeremy Fitzhardinge req = RING_GET_REQUEST(&np->rx, requeue_idx); 17080d160211SJeremy Fitzhardinge 170901c68026SIan Campbell frag = &skb_shinfo(skb)->frags[0]; 171001c68026SIan Campbell page = skb_frag_page(frag); 17110d160211SJeremy Fitzhardinge gnttab_grant_foreign_access_ref( 17120d160211SJeremy Fitzhardinge ref, np->xbdev->otherend_id, 171301c68026SIan Campbell pfn_to_mfn(page_to_pfn(page)), 17140d160211SJeremy Fitzhardinge 0); 17150d160211SJeremy Fitzhardinge req->gref = ref; 17160d160211SJeremy Fitzhardinge req->id = requeue_idx; 17170d160211SJeremy Fitzhardinge 17180d160211SJeremy Fitzhardinge requeue_idx++; 17190d160211SJeremy Fitzhardinge } 17200d160211SJeremy Fitzhardinge 17210d160211SJeremy Fitzhardinge np->rx.req_prod_pvt = requeue_idx; 17220d160211SJeremy Fitzhardinge 17230d160211SJeremy Fitzhardinge /* 17240d160211SJeremy Fitzhardinge * Step 3: All public and private state should now be sane. Get 17250d160211SJeremy Fitzhardinge * ready to start sending and receiving packets and give the driver 17260d160211SJeremy Fitzhardinge * domain a kick because we've probably just requeued some 17270d160211SJeremy Fitzhardinge * packets. 17280d160211SJeremy Fitzhardinge */ 17290d160211SJeremy Fitzhardinge netif_carrier_on(np->netdev); 17300d160211SJeremy Fitzhardinge notify_remote_via_irq(np->netdev->irq); 17310d160211SJeremy Fitzhardinge xennet_tx_buf_gc(dev); 17320d160211SJeremy Fitzhardinge xennet_alloc_rx_buffers(dev); 17330d160211SJeremy Fitzhardinge 17340d160211SJeremy Fitzhardinge spin_unlock_irq(&np->tx_lock); 17350d160211SJeremy Fitzhardinge spin_unlock_bh(&np->rx_lock); 17360d160211SJeremy Fitzhardinge 17370d160211SJeremy Fitzhardinge return 0; 17380d160211SJeremy Fitzhardinge } 17390d160211SJeremy Fitzhardinge 17400d160211SJeremy Fitzhardinge /** 17410d160211SJeremy Fitzhardinge * Callback received when the backend's state changes. 17420d160211SJeremy Fitzhardinge */ 1743f502bf2bSIan Campbell static void netback_changed(struct xenbus_device *dev, 17440d160211SJeremy Fitzhardinge enum xenbus_state backend_state) 17450d160211SJeremy Fitzhardinge { 17461b713e00SGreg Kroah-Hartman struct netfront_info *np = dev_get_drvdata(&dev->dev); 17470d160211SJeremy Fitzhardinge struct net_device *netdev = np->netdev; 17480d160211SJeremy Fitzhardinge 17490d160211SJeremy Fitzhardinge dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); 17500d160211SJeremy Fitzhardinge 17510d160211SJeremy Fitzhardinge switch (backend_state) { 17520d160211SJeremy Fitzhardinge case XenbusStateInitialising: 17530d160211SJeremy Fitzhardinge case XenbusStateInitialised: 1754b78c9512SNoboru Iwamatsu case XenbusStateReconfiguring: 1755b78c9512SNoboru Iwamatsu case XenbusStateReconfigured: 17560d160211SJeremy Fitzhardinge case XenbusStateUnknown: 17570d160211SJeremy Fitzhardinge case XenbusStateClosed: 17580d160211SJeremy Fitzhardinge break; 17590d160211SJeremy Fitzhardinge 17600d160211SJeremy Fitzhardinge case XenbusStateInitWait: 17610d160211SJeremy Fitzhardinge if (dev->state != XenbusStateInitialising) 17620d160211SJeremy Fitzhardinge break; 17630d160211SJeremy Fitzhardinge if (xennet_connect(netdev) != 0) 17640d160211SJeremy Fitzhardinge break; 17650d160211SJeremy Fitzhardinge xenbus_switch_state(dev, XenbusStateConnected); 176608e34eb1SLaszlo Ersek break; 176708e34eb1SLaszlo Ersek 176808e34eb1SLaszlo Ersek case XenbusStateConnected: 1769ee89bab1SAmerigo Wang netdev_notify_peers(netdev); 17700d160211SJeremy Fitzhardinge break; 17710d160211SJeremy Fitzhardinge 17720d160211SJeremy Fitzhardinge case XenbusStateClosing: 17730d160211SJeremy Fitzhardinge xenbus_frontend_closed(dev); 17740d160211SJeremy Fitzhardinge break; 17750d160211SJeremy Fitzhardinge } 17760d160211SJeremy Fitzhardinge } 17770d160211SJeremy Fitzhardinge 1778e0ce4af9SIan Campbell static const struct xennet_stat { 1779e0ce4af9SIan Campbell char name[ETH_GSTRING_LEN]; 1780e0ce4af9SIan Campbell u16 offset; 1781e0ce4af9SIan Campbell } xennet_stats[] = { 1782e0ce4af9SIan Campbell { 1783e0ce4af9SIan Campbell "rx_gso_checksum_fixup", 1784e0ce4af9SIan Campbell offsetof(struct netfront_info, rx_gso_checksum_fixup) 1785e0ce4af9SIan Campbell }, 1786e0ce4af9SIan Campbell }; 1787e0ce4af9SIan Campbell 1788e0ce4af9SIan Campbell static int xennet_get_sset_count(struct net_device *dev, int string_set) 1789e0ce4af9SIan Campbell { 1790e0ce4af9SIan Campbell switch (string_set) { 1791e0ce4af9SIan Campbell case ETH_SS_STATS: 1792e0ce4af9SIan Campbell return ARRAY_SIZE(xennet_stats); 1793e0ce4af9SIan Campbell default: 1794e0ce4af9SIan Campbell return -EINVAL; 1795e0ce4af9SIan Campbell } 1796e0ce4af9SIan Campbell } 1797e0ce4af9SIan Campbell 1798e0ce4af9SIan Campbell static void xennet_get_ethtool_stats(struct net_device *dev, 1799e0ce4af9SIan Campbell struct ethtool_stats *stats, u64 * data) 1800e0ce4af9SIan Campbell { 1801e0ce4af9SIan Campbell void *np = netdev_priv(dev); 1802e0ce4af9SIan Campbell int i; 1803e0ce4af9SIan Campbell 1804e0ce4af9SIan Campbell for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 180548f26d51SEric Dumazet data[i] = *(unsigned long *)(np + xennet_stats[i].offset); 1806e0ce4af9SIan Campbell } 1807e0ce4af9SIan Campbell 1808e0ce4af9SIan Campbell static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) 1809e0ce4af9SIan Campbell { 1810e0ce4af9SIan Campbell int i; 1811e0ce4af9SIan Campbell 1812e0ce4af9SIan Campbell switch (stringset) { 1813e0ce4af9SIan Campbell case ETH_SS_STATS: 1814e0ce4af9SIan Campbell for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 1815e0ce4af9SIan Campbell memcpy(data + i * ETH_GSTRING_LEN, 1816e0ce4af9SIan Campbell xennet_stats[i].name, ETH_GSTRING_LEN); 1817e0ce4af9SIan Campbell break; 1818e0ce4af9SIan Campbell } 1819e0ce4af9SIan Campbell } 1820e0ce4af9SIan Campbell 18210fc0b732SStephen Hemminger static const struct ethtool_ops xennet_ethtool_ops = 18220d160211SJeremy Fitzhardinge { 18230d160211SJeremy Fitzhardinge .get_link = ethtool_op_get_link, 1824e0ce4af9SIan Campbell 1825e0ce4af9SIan Campbell .get_sset_count = xennet_get_sset_count, 1826e0ce4af9SIan Campbell .get_ethtool_stats = xennet_get_ethtool_stats, 1827e0ce4af9SIan Campbell .get_strings = xennet_get_strings, 18280d160211SJeremy Fitzhardinge }; 18290d160211SJeremy Fitzhardinge 18300d160211SJeremy Fitzhardinge #ifdef CONFIG_SYSFS 18310d160211SJeremy Fitzhardinge static ssize_t show_rxbuf_min(struct device *dev, 18320d160211SJeremy Fitzhardinge struct device_attribute *attr, char *buf) 18330d160211SJeremy Fitzhardinge { 18340d160211SJeremy Fitzhardinge struct net_device *netdev = to_net_dev(dev); 18350d160211SJeremy Fitzhardinge struct netfront_info *info = netdev_priv(netdev); 18360d160211SJeremy Fitzhardinge 18370d160211SJeremy Fitzhardinge return sprintf(buf, "%u\n", info->rx_min_target); 18380d160211SJeremy Fitzhardinge } 18390d160211SJeremy Fitzhardinge 18400d160211SJeremy Fitzhardinge static ssize_t store_rxbuf_min(struct device *dev, 18410d160211SJeremy Fitzhardinge struct device_attribute *attr, 18420d160211SJeremy Fitzhardinge const char *buf, size_t len) 18430d160211SJeremy Fitzhardinge { 18440d160211SJeremy Fitzhardinge struct net_device *netdev = to_net_dev(dev); 18450d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(netdev); 18460d160211SJeremy Fitzhardinge char *endp; 18470d160211SJeremy Fitzhardinge unsigned long target; 18480d160211SJeremy Fitzhardinge 18490d160211SJeremy Fitzhardinge if (!capable(CAP_NET_ADMIN)) 18500d160211SJeremy Fitzhardinge return -EPERM; 18510d160211SJeremy Fitzhardinge 18520d160211SJeremy Fitzhardinge target = simple_strtoul(buf, &endp, 0); 18530d160211SJeremy Fitzhardinge if (endp == buf) 18540d160211SJeremy Fitzhardinge return -EBADMSG; 18550d160211SJeremy Fitzhardinge 18560d160211SJeremy Fitzhardinge if (target < RX_MIN_TARGET) 18570d160211SJeremy Fitzhardinge target = RX_MIN_TARGET; 18580d160211SJeremy Fitzhardinge if (target > RX_MAX_TARGET) 18590d160211SJeremy Fitzhardinge target = RX_MAX_TARGET; 18600d160211SJeremy Fitzhardinge 18610d160211SJeremy Fitzhardinge spin_lock_bh(&np->rx_lock); 18620d160211SJeremy Fitzhardinge if (target > np->rx_max_target) 18630d160211SJeremy Fitzhardinge np->rx_max_target = target; 18640d160211SJeremy Fitzhardinge np->rx_min_target = target; 18650d160211SJeremy Fitzhardinge if (target > np->rx_target) 18660d160211SJeremy Fitzhardinge np->rx_target = target; 18670d160211SJeremy Fitzhardinge 18680d160211SJeremy Fitzhardinge xennet_alloc_rx_buffers(netdev); 18690d160211SJeremy Fitzhardinge 18700d160211SJeremy Fitzhardinge spin_unlock_bh(&np->rx_lock); 18710d160211SJeremy Fitzhardinge return len; 18720d160211SJeremy Fitzhardinge } 18730d160211SJeremy Fitzhardinge 18740d160211SJeremy Fitzhardinge static ssize_t show_rxbuf_max(struct device *dev, 18750d160211SJeremy Fitzhardinge struct device_attribute *attr, char *buf) 18760d160211SJeremy Fitzhardinge { 18770d160211SJeremy Fitzhardinge struct net_device *netdev = to_net_dev(dev); 18780d160211SJeremy Fitzhardinge struct netfront_info *info = netdev_priv(netdev); 18790d160211SJeremy Fitzhardinge 18800d160211SJeremy Fitzhardinge return sprintf(buf, "%u\n", info->rx_max_target); 18810d160211SJeremy Fitzhardinge } 18820d160211SJeremy Fitzhardinge 18830d160211SJeremy Fitzhardinge static ssize_t store_rxbuf_max(struct device *dev, 18840d160211SJeremy Fitzhardinge struct device_attribute *attr, 18850d160211SJeremy Fitzhardinge const char *buf, size_t len) 18860d160211SJeremy Fitzhardinge { 18870d160211SJeremy Fitzhardinge struct net_device *netdev = to_net_dev(dev); 18880d160211SJeremy Fitzhardinge struct netfront_info *np = netdev_priv(netdev); 18890d160211SJeremy Fitzhardinge char *endp; 18900d160211SJeremy Fitzhardinge unsigned long target; 18910d160211SJeremy Fitzhardinge 18920d160211SJeremy Fitzhardinge if (!capable(CAP_NET_ADMIN)) 18930d160211SJeremy Fitzhardinge return -EPERM; 18940d160211SJeremy Fitzhardinge 18950d160211SJeremy Fitzhardinge target = simple_strtoul(buf, &endp, 0); 18960d160211SJeremy Fitzhardinge if (endp == buf) 18970d160211SJeremy Fitzhardinge return -EBADMSG; 18980d160211SJeremy Fitzhardinge 18990d160211SJeremy Fitzhardinge if (target < RX_MIN_TARGET) 19000d160211SJeremy Fitzhardinge target = RX_MIN_TARGET; 19010d160211SJeremy Fitzhardinge if (target > RX_MAX_TARGET) 19020d160211SJeremy Fitzhardinge target = RX_MAX_TARGET; 19030d160211SJeremy Fitzhardinge 19040d160211SJeremy Fitzhardinge spin_lock_bh(&np->rx_lock); 19050d160211SJeremy Fitzhardinge if (target < np->rx_min_target) 19060d160211SJeremy Fitzhardinge np->rx_min_target = target; 19070d160211SJeremy Fitzhardinge np->rx_max_target = target; 19080d160211SJeremy Fitzhardinge if (target < np->rx_target) 19090d160211SJeremy Fitzhardinge np->rx_target = target; 19100d160211SJeremy Fitzhardinge 19110d160211SJeremy Fitzhardinge xennet_alloc_rx_buffers(netdev); 19120d160211SJeremy Fitzhardinge 19130d160211SJeremy Fitzhardinge spin_unlock_bh(&np->rx_lock); 19140d160211SJeremy Fitzhardinge return len; 19150d160211SJeremy Fitzhardinge } 19160d160211SJeremy Fitzhardinge 19170d160211SJeremy Fitzhardinge static ssize_t show_rxbuf_cur(struct device *dev, 19180d160211SJeremy Fitzhardinge struct device_attribute *attr, char *buf) 19190d160211SJeremy Fitzhardinge { 19200d160211SJeremy Fitzhardinge struct net_device *netdev = to_net_dev(dev); 19210d160211SJeremy Fitzhardinge struct netfront_info *info = netdev_priv(netdev); 19220d160211SJeremy Fitzhardinge 19230d160211SJeremy Fitzhardinge return sprintf(buf, "%u\n", info->rx_target); 19240d160211SJeremy Fitzhardinge } 19250d160211SJeremy Fitzhardinge 19260d160211SJeremy Fitzhardinge static struct device_attribute xennet_attrs[] = { 19270d160211SJeremy Fitzhardinge __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min), 19280d160211SJeremy Fitzhardinge __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max), 19290d160211SJeremy Fitzhardinge __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL), 19300d160211SJeremy Fitzhardinge }; 19310d160211SJeremy Fitzhardinge 19320d160211SJeremy Fitzhardinge static int xennet_sysfs_addif(struct net_device *netdev) 19330d160211SJeremy Fitzhardinge { 19340d160211SJeremy Fitzhardinge int i; 19350d160211SJeremy Fitzhardinge int err; 19360d160211SJeremy Fitzhardinge 19370d160211SJeremy Fitzhardinge for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) { 19380d160211SJeremy Fitzhardinge err = device_create_file(&netdev->dev, 19390d160211SJeremy Fitzhardinge &xennet_attrs[i]); 19400d160211SJeremy Fitzhardinge if (err) 19410d160211SJeremy Fitzhardinge goto fail; 19420d160211SJeremy Fitzhardinge } 19430d160211SJeremy Fitzhardinge return 0; 19440d160211SJeremy Fitzhardinge 19450d160211SJeremy Fitzhardinge fail: 19460d160211SJeremy Fitzhardinge while (--i >= 0) 19470d160211SJeremy Fitzhardinge device_remove_file(&netdev->dev, &xennet_attrs[i]); 19480d160211SJeremy Fitzhardinge return err; 19490d160211SJeremy Fitzhardinge } 19500d160211SJeremy Fitzhardinge 19510d160211SJeremy Fitzhardinge static void xennet_sysfs_delif(struct net_device *netdev) 19520d160211SJeremy Fitzhardinge { 19530d160211SJeremy Fitzhardinge int i; 19540d160211SJeremy Fitzhardinge 19550d160211SJeremy Fitzhardinge for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) 19560d160211SJeremy Fitzhardinge device_remove_file(&netdev->dev, &xennet_attrs[i]); 19570d160211SJeremy Fitzhardinge } 19580d160211SJeremy Fitzhardinge 19590d160211SJeremy Fitzhardinge #endif /* CONFIG_SYSFS */ 19600d160211SJeremy Fitzhardinge 196173db144bSJan Beulich static const struct xenbus_device_id netfront_ids[] = { 19620d160211SJeremy Fitzhardinge { "vif" }, 19630d160211SJeremy Fitzhardinge { "" } 19640d160211SJeremy Fitzhardinge }; 19650d160211SJeremy Fitzhardinge 19660d160211SJeremy Fitzhardinge 19678e0e46bbSBill Pemberton static int xennet_remove(struct xenbus_device *dev) 19680d160211SJeremy Fitzhardinge { 19691b713e00SGreg Kroah-Hartman struct netfront_info *info = dev_get_drvdata(&dev->dev); 19700d160211SJeremy Fitzhardinge 19710d160211SJeremy Fitzhardinge dev_dbg(&dev->dev, "%s\n", dev->nodename); 19720d160211SJeremy Fitzhardinge 19730d160211SJeremy Fitzhardinge xennet_disconnect_backend(info); 19740d160211SJeremy Fitzhardinge 19750d160211SJeremy Fitzhardinge xennet_sysfs_delif(info->netdev); 19760d160211SJeremy Fitzhardinge 19776bc96d04SIan Campbell unregister_netdev(info->netdev); 19786bc96d04SIan Campbell 19796bc96d04SIan Campbell del_timer_sync(&info->rx_refill_timer); 19806bc96d04SIan Campbell 1981e00f85beSstephen hemminger free_percpu(info->stats); 1982e00f85beSstephen hemminger 19830d160211SJeremy Fitzhardinge free_netdev(info->netdev); 19840d160211SJeremy Fitzhardinge 19850d160211SJeremy Fitzhardinge return 0; 19860d160211SJeremy Fitzhardinge } 19870d160211SJeremy Fitzhardinge 198873db144bSJan Beulich static DEFINE_XENBUS_DRIVER(netfront, , 19890d160211SJeremy Fitzhardinge .probe = netfront_probe, 19908e0e46bbSBill Pemberton .remove = xennet_remove, 19910d160211SJeremy Fitzhardinge .resume = netfront_resume, 1992f502bf2bSIan Campbell .otherend_changed = netback_changed, 199373db144bSJan Beulich ); 19940d160211SJeremy Fitzhardinge 19950d160211SJeremy Fitzhardinge static int __init netif_init(void) 19960d160211SJeremy Fitzhardinge { 19976e833587SJeremy Fitzhardinge if (!xen_domain()) 19980d160211SJeremy Fitzhardinge return -ENODEV; 19990d160211SJeremy Fitzhardinge 2000e95ae5a4SIgor Mammedov if (xen_hvm_domain() && !xen_platform_pci_unplug) 2001b9136d20SIgor Mammedov return -ENODEV; 2002b9136d20SIgor Mammedov 20030d160211SJeremy Fitzhardinge printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n"); 20040d160211SJeremy Fitzhardinge 2005ffb78a26SAl Viro return xenbus_register_frontend(&netfront_driver); 20060d160211SJeremy Fitzhardinge } 20070d160211SJeremy Fitzhardinge module_init(netif_init); 20080d160211SJeremy Fitzhardinge 20090d160211SJeremy Fitzhardinge 20100d160211SJeremy Fitzhardinge static void __exit netif_exit(void) 20110d160211SJeremy Fitzhardinge { 2012ffb78a26SAl Viro xenbus_unregister_driver(&netfront_driver); 20130d160211SJeremy Fitzhardinge } 20140d160211SJeremy Fitzhardinge module_exit(netif_exit); 20150d160211SJeremy Fitzhardinge 20160d160211SJeremy Fitzhardinge MODULE_DESCRIPTION("Xen virtual network device frontend"); 20170d160211SJeremy Fitzhardinge MODULE_LICENSE("GPL"); 2018d2f0c52bSMark McLoughlin MODULE_ALIAS("xen:vif"); 20194f93f09bSMark McLoughlin MODULE_ALIAS("xennet"); 2020