drivers/net/xen-netback/common.h

   1 /*
   2  * This program is free software; you can redistribute it and/or
   3  * modify it under the terms of the GNU General Public License version 2
   4  * as published by the Free Software Foundation; or, when distributed
   5  * separately from the Linux kernel or incorporated into other
   6  * software packages, subject to the following license:
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a copy
   9  * of this source file (the "Software"), to deal in the Software without
  10  * restriction, including without limitation the rights to use, copy, modify,
  11  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  12  * and to permit persons to whom the Software is furnished to do so, subject to
  13  * the following conditions:
  14  *
  15  * The above copyright notice and this permission notice shall be included in
  16  * all copies or substantial portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  21  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  23  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  24  * IN THE SOFTWARE.
  25  */
  26
  27 #ifndef __XEN_NETBACK__COMMON_H__
  28 #define __XEN_NETBACK__COMMON_H__
  29
  30 #define pr_fmt(fmt) KBUILD_MODNAME ":%s: " fmt, __func__
  31
  32 #include <linux/module.h>
  33 #include <linux/interrupt.h>
  34 #include <linux/slab.h>
  35 #include <linux/ip.h>
  36 #include <linux/in.h>
  37 #include <linux/io.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/etherdevice.h>
  40 #include <linux/wait.h>
  41 #include <linux/sched.h>
  42
  43 #include <xen/interface/io/netif.h>
  44 #include <xen/interface/grant_table.h>
  45 #include <xen/grant_table.h>
  46 #include <xen/xenbus.h>
  47 #include <xen/page.h>
  48 #include <linux/debugfs.h>
  49
  50 typedef unsigned int pending_ring_idx_t;
  51 #define INVALID_PENDING_RING_IDX (~0U)
  52
  53 struct pending_tx_info {
  54         struct xen_netif_tx_request req; /* tx request */
  55         unsigned int extra_count;
  56         /* Callback data for released SKBs. The callback is always
  57          * xenvif_zerocopy_callback, desc contains the pending_idx, which is
  58          * also an index in pending_tx_info array. It is initialized in
  59          * xenvif_alloc and it never changes.
  60          * skb_shinfo(skb)->destructor_arg points to the first mapped slot's
  61          * callback_struct in this array of struct pending_tx_info's, then ctx
  62          * to the next, or NULL if there is no more slot for this skb.
  63          * ubuf_to_vif is a helper which finds the struct xenvif from a pointer
  64          * to this field.
  65          */
  66         struct ubuf_info callback_struct;
  67 };
  68
  69 #define XEN_NETIF_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
  70 #define XEN_NETIF_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
  71
  72 struct xenvif_rx_meta {
  73         int id;
  74         int size;
  75         int gso_type;
  76         int gso_size;
  77 };
  78
  79 #define GSO_BIT(type) \
  80         (1 << XEN_NETIF_GSO_TYPE_ ## type)
  81
  82 /* Discriminate from any valid pending_idx value. */
  83 #define INVALID_PENDING_IDX 0xFFFF
  84
  85 #define MAX_BUFFER_OFFSET XEN_PAGE_SIZE
  86
  87 #define MAX_PENDING_REQS XEN_NETIF_TX_RING_SIZE
  88
  89 /* The maximum number of frags is derived from the size of a grant (same
  90  * as a Xen page size for now).
  91  */
  92 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
  93
  94 /* It's possible for an skb to have a maximal number of frags
  95  * but still be less than MAX_BUFFER_OFFSET in size. Thus the
  96  * worst-case number of copy operations is MAX_XEN_SKB_FRAGS per
  97  * ring slot.
  98  */
  99 #define MAX_GRANT_COPY_OPS (MAX_XEN_SKB_FRAGS * XEN_NETIF_RX_RING_SIZE)
 100
 101 #define NETBACK_INVALID_HANDLE -1
 102
 103 /* To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating
 104  * the maximum slots a valid packet can use. Now this value is defined
 105  * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by
 106  * all backend.
 107  */
 108 #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN
 109
 110 /* Queue name is interface name with "-qNNN" appended */
 111 #define QUEUE_NAME_SIZE (IFNAMSIZ + 5)
 112
 113 /* IRQ name is queue name with "-tx" or "-rx" appended */
 114 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
 115
 116 struct xenvif;
 117
 118 struct xenvif_stats {
 119         /* Stats fields to be updated per-queue.
 120          * A subset of struct net_device_stats that contains only the
 121          * fields that are updated in netback.c for each queue.
 122          */
 123         unsigned int rx_bytes;
 124         unsigned int rx_packets;
 125         unsigned int tx_bytes;
 126         unsigned int tx_packets;
 127
 128         /* Additional stats used by xenvif */
 129         unsigned long rx_gso_checksum_fixup;
 130         unsigned long tx_zerocopy_sent;
 131         unsigned long tx_zerocopy_success;
 132         unsigned long tx_zerocopy_fail;
 133         unsigned long tx_frag_overflow;
 134 };
 135
 136 struct xenvif_queue { /* Per-queue data for xenvif */
 137         unsigned int id; /* Queue ID, 0-based */
 138         char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
 139         struct xenvif *vif; /* Parent VIF */
 140
 141         /* Use NAPI for guest TX */
 142         struct napi_struct napi;
 143         /* When feature-split-event-channels = 0, tx_irq = rx_irq. */
 144         unsigned int tx_irq;
 145         /* Only used when feature-split-event-channels = 1 */
 146         char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
 147         struct xen_netif_tx_back_ring tx;
 148         struct sk_buff_head tx_queue;
 149         struct page *mmap_pages[MAX_PENDING_REQS];
 150         pending_ring_idx_t pending_prod;
 151         pending_ring_idx_t pending_cons;
 152         u16 pending_ring[MAX_PENDING_REQS];
 153         struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
 154         grant_handle_t grant_tx_handle[MAX_PENDING_REQS];
 155
 156         struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
 157         struct gnttab_map_grant_ref tx_map_ops[MAX_PENDING_REQS];
 158         struct gnttab_unmap_grant_ref tx_unmap_ops[MAX_PENDING_REQS];
 159         /* passed to gnttab_[un]map_refs with pages under (un)mapping */
 160         struct page *pages_to_map[MAX_PENDING_REQS];
 161         struct page *pages_to_unmap[MAX_PENDING_REQS];
 162
 163         /* This prevents zerocopy callbacks  to race over dealloc_ring */
 164         spinlock_t callback_lock;
 165         /* This prevents dealloc thread and NAPI instance to race over response
 166          * creation and pending_ring in xenvif_idx_release. In xenvif_tx_err
 167          * it only protect response creation
 168          */
 169         spinlock_t response_lock;
 170         pending_ring_idx_t dealloc_prod;
 171         pending_ring_idx_t dealloc_cons;
 172         u16 dealloc_ring[MAX_PENDING_REQS];
 173         struct task_struct *dealloc_task;
 174         wait_queue_head_t dealloc_wq;
 175         atomic_t inflight_packets;
 176
 177         /* Use kthread for guest RX */
 178         struct task_struct *task;
 179         wait_queue_head_t wq;
 180         /* When feature-split-event-channels = 0, tx_irq = rx_irq. */
 181         unsigned int rx_irq;
 182         /* Only used when feature-split-event-channels = 1 */
 183         char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
 184         struct xen_netif_rx_back_ring rx;
 185         struct sk_buff_head rx_queue;
 186
 187         unsigned int rx_queue_max;
 188         unsigned int rx_queue_len;
 189         unsigned long last_rx_time;
 190         bool stalled;
 191
 192         struct gnttab_copy grant_copy_op[MAX_GRANT_COPY_OPS];
 193
 194         /* We create one meta structure per ring request we consume, so
 195          * the maximum number is the same as the ring size.
 196          */
 197         struct xenvif_rx_meta meta[XEN_NETIF_RX_RING_SIZE];
 198
 199         /* Transmit shaping: allow 'credit_bytes' every 'credit_usec'. */
 200         unsigned long   credit_bytes;
 201         unsigned long   credit_usec;
 202         unsigned long   remaining_credit;
 203         struct timer_list credit_timeout;
 204         u64 credit_window_start;
 205
 206         /* Statistics */
 207         struct xenvif_stats stats;
 208 };
 209
 210 enum state_bit_shift {
 211         /* This bit marks that the vif is connected */
 212         VIF_STATUS_CONNECTED,
 213 };
 214
 215 struct xenvif_mcast_addr {
 216         struct list_head entry;
 217         struct rcu_head rcu;
 218         u8 addr[6];
 219 };
 220
 221 #define XEN_NETBK_MCAST_MAX 64
 222
 223 struct xenvif {
 224         /* Unique identifier for this interface. */
 225         domid_t          domid;
 226         unsigned int     handle;
 227
 228         u8               fe_dev_addr[6];
 229         struct list_head fe_mcast_addr;
 230         unsigned int     fe_mcast_count;
 231
 232         /* Frontend feature information. */
 233         int gso_mask;
 234         int gso_prefix_mask;
 235
 236         u8 can_sg:1;
 237         u8 ip_csum:1;
 238         u8 ipv6_csum:1;
 239         u8 multicast_control:1;
 240
 241         /* Is this interface disabled? True when backend discovers
 242          * frontend is rogue.
 243          */
 244         bool disabled;
 245         unsigned long status;
 246         unsigned long drain_timeout;
 247         unsigned long stall_timeout;
 248
 249         /* Queues */
 250         struct xenvif_queue *queues;
 251         unsigned int num_queues; /* active queues, resource allocated */
 252         unsigned int stalled_queues;
 253
 254         struct xenbus_watch credit_watch;
 255         struct xenbus_watch mcast_ctrl_watch;
 256
 257         spinlock_t lock;
 258
 259 #ifdef CONFIG_DEBUG_FS
 260         struct dentry *xenvif_dbg_root;
 261 #endif
 262
 263         /* Miscellaneous private stuff. */
 264         struct net_device *dev;
 265 };
 266
 267 struct xenvif_rx_cb {
 268         unsigned long expires;
 269         int meta_slots_used;
 270 };
 271
 272 #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb)
 273
 274 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif)
 275 {
 276         return to_xenbus_device(vif->dev->dev.parent);
 277 }
 278
 279 void xenvif_tx_credit_callback(unsigned long data);
 280
 281 struct xenvif *xenvif_alloc(struct device *parent,
 282                             domid_t domid,
 283                             unsigned int handle);
 284
 285 int xenvif_init_queue(struct xenvif_queue *queue);
 286 void xenvif_deinit_queue(struct xenvif_queue *queue);
 287
 288 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
 289                    unsigned long rx_ring_ref, unsigned int tx_evtchn,
 290                    unsigned int rx_evtchn);
 291 void xenvif_disconnect(struct xenvif *vif);
 292 void xenvif_free(struct xenvif *vif);
 293
 294 int xenvif_xenbus_init(void);
 295 void xenvif_xenbus_fini(void);
 296
 297 int xenvif_schedulable(struct xenvif *vif);
 298
 299 int xenvif_queue_stopped(struct xenvif_queue *queue);
 300 void xenvif_wake_queue(struct xenvif_queue *queue);
 301
 302 /* (Un)Map communication rings. */
 303 void xenvif_unmap_frontend_rings(struct xenvif_queue *queue);
 304 int xenvif_map_frontend_rings(struct xenvif_queue *queue,
 305                               grant_ref_t tx_ring_ref,
 306                               grant_ref_t rx_ring_ref);
 307
 308 /* Check for SKBs from frontend and schedule backend processing */
 309 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue);
 310
 311 /* Prevent the device from generating any further traffic. */
 312 void xenvif_carrier_off(struct xenvif *vif);
 313
 314 int xenvif_tx_action(struct xenvif_queue *queue, int budget);
 315
 316 int xenvif_kthread_guest_rx(void *data);
 317 void xenvif_kick_thread(struct xenvif_queue *queue);
 318
 319 int xenvif_dealloc_kthread(void *data);
 320
 321 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb);
 322
 323 void xenvif_carrier_on(struct xenvif *vif);
 324
 325 /* Callback from stack when TX packet can be released */
 326 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success);
 327
 328 /* Unmap a pending page and release it back to the guest */
 329 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
 330
 331 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue)
 332 {
 333         return MAX_PENDING_REQS -
 334                 queue->pending_prod + queue->pending_cons;
 335 }
 336
 337 irqreturn_t xenvif_interrupt(int irq, void *dev_id);
 338
 339 extern bool separate_tx_rx_irq;
 340
 341 extern unsigned int rx_drain_timeout_msecs;
 342 extern unsigned int rx_stall_timeout_msecs;
 343 extern unsigned int xenvif_max_queues;
 344
 345 #ifdef CONFIG_DEBUG_FS
 346 extern struct dentry *xen_netback_dbg_root;
 347 #endif
 348
 349 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
 350                                  struct sk_buff *skb);
 351 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue);
 352
 353 /* Multicast control */
 354 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr);
 355 void xenvif_mcast_addr_list_free(struct xenvif *vif);
 356
 357 #endif /* __XEN_NETBACK__COMMON_H__ */