Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 1 | #ifndef _VHOST_H |
| 2 | #define _VHOST_H |
| 3 | |
| 4 | #include <linux/eventfd.h> |
| 5 | #include <linux/vhost.h> |
| 6 | #include <linux/mm.h> |
| 7 | #include <linux/mutex.h> |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 8 | #include <linux/poll.h> |
| 9 | #include <linux/file.h> |
| 10 | #include <linux/skbuff.h> |
| 11 | #include <linux/uio.h> |
| 12 | #include <linux/virtio_config.h> |
| 13 | #include <linux/virtio_ring.h> |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 14 | #include <asm/atomic.h> |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 15 | |
| 16 | struct vhost_device; |
| 17 | |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 18 | struct vhost_work; |
| 19 | typedef void (*vhost_work_fn_t)(struct vhost_work *work); |
| 20 | |
| 21 | struct vhost_work { |
| 22 | struct list_head node; |
| 23 | vhost_work_fn_t fn; |
| 24 | wait_queue_head_t done; |
| 25 | int flushing; |
| 26 | unsigned queue_seq; |
| 27 | unsigned done_seq; |
| 28 | }; |
| 29 | |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 30 | /* Poll a file (eventfd or socket) */ |
| 31 | /* Note: there's nothing vhost specific about this structure. */ |
| 32 | struct vhost_poll { |
| 33 | poll_table table; |
| 34 | wait_queue_head_t *wqh; |
| 35 | wait_queue_t wait; |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 36 | struct vhost_work work; |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 37 | unsigned long mask; |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 38 | struct vhost_dev *dev; |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 39 | }; |
| 40 | |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 41 | void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn, |
| 42 | unsigned long mask, struct vhost_dev *dev); |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 43 | void vhost_poll_start(struct vhost_poll *poll, struct file *file); |
| 44 | void vhost_poll_stop(struct vhost_poll *poll); |
| 45 | void vhost_poll_flush(struct vhost_poll *poll); |
| 46 | void vhost_poll_queue(struct vhost_poll *poll); |
| 47 | |
| 48 | struct vhost_log { |
| 49 | u64 addr; |
| 50 | u64 len; |
| 51 | }; |
| 52 | |
| 53 | /* The virtqueue structure describes a queue attached to a device. */ |
| 54 | struct vhost_virtqueue { |
| 55 | struct vhost_dev *dev; |
| 56 | |
| 57 | /* The actual ring of buffers. */ |
| 58 | struct mutex mutex; |
| 59 | unsigned int num; |
| 60 | struct vring_desc __user *desc; |
| 61 | struct vring_avail __user *avail; |
| 62 | struct vring_used __user *used; |
| 63 | struct file *kick; |
| 64 | struct file *call; |
| 65 | struct file *error; |
| 66 | struct eventfd_ctx *call_ctx; |
| 67 | struct eventfd_ctx *error_ctx; |
| 68 | struct eventfd_ctx *log_ctx; |
| 69 | |
| 70 | struct vhost_poll poll; |
| 71 | |
| 72 | /* The routine to call when the Guest pings us, or timeout. */ |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 73 | vhost_work_fn_t handle_kick; |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 74 | |
| 75 | /* Last available index we saw. */ |
| 76 | u16 last_avail_idx; |
| 77 | |
| 78 | /* Caches available index value from user. */ |
| 79 | u16 avail_idx; |
| 80 | |
| 81 | /* Last index we used. */ |
| 82 | u16 last_used_idx; |
| 83 | |
| 84 | /* Used flags */ |
| 85 | u16 used_flags; |
| 86 | |
| 87 | /* Log writes to used structure. */ |
| 88 | bool log_used; |
| 89 | u64 log_addr; |
| 90 | |
Jason Wang | e0e9b40 | 2010-09-14 23:53:05 +0800 | [diff] [blame] | 91 | struct iovec iov[UIO_MAXIOV]; |
| 92 | /* hdr is used to store the virtio header. |
| 93 | * Since each iovec has >= 1 byte length, we never need more than |
| 94 | * header length entries to store the header. */ |
| 95 | struct iovec hdr[sizeof(struct virtio_net_hdr_mrg_rxbuf)]; |
| 96 | struct iovec *indirect; |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 97 | size_t vhost_hlen; |
| 98 | size_t sock_hlen; |
Jason Wang | e0e9b40 | 2010-09-14 23:53:05 +0800 | [diff] [blame] | 99 | struct vring_used_elem *heads; |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 100 | /* We use a kind of RCU to access private pointer. |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 101 | * All readers access it from worker, which makes it possible to |
| 102 | * flush the vhost_work instead of synchronize_rcu. Therefore readers do |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 103 | * not need to call rcu_read_lock/rcu_read_unlock: the beginning of |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 104 | * vhost_work execution acts instead of rcu_read_lock() and the end of |
Jason Wang | a290aec | 2010-11-29 13:48:40 +0800 | [diff] [blame] | 105 | * vhost_work execution acts instead of rcu_read_unlock(). |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 106 | * Writers use virtqueue mutex. */ |
Arnd Bergmann | 28457ee | 2010-03-09 19:24:45 +0100 | [diff] [blame] | 107 | void __rcu *private_data; |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 108 | /* Log write descriptors */ |
| 109 | void __user *log_base; |
Jason Wang | e0e9b40 | 2010-09-14 23:53:05 +0800 | [diff] [blame] | 110 | struct vhost_log *log; |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 111 | }; |
| 112 | |
| 113 | struct vhost_dev { |
| 114 | /* Readers use RCU to access memory table pointer |
| 115 | * log base pointer and features. |
| 116 | * Writers use mutex below.*/ |
Arnd Bergmann | 28457ee | 2010-03-09 19:24:45 +0100 | [diff] [blame] | 117 | struct vhost_memory __rcu *memory; |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 118 | struct mm_struct *mm; |
| 119 | struct mutex mutex; |
| 120 | unsigned acked_features; |
| 121 | struct vhost_virtqueue *vqs; |
| 122 | int nvqs; |
| 123 | struct file *log_file; |
| 124 | struct eventfd_ctx *log_ctx; |
Tejun Heo | c23f344 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 125 | spinlock_t work_lock; |
| 126 | struct list_head work_list; |
| 127 | struct task_struct *worker; |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 128 | }; |
| 129 | |
| 130 | long vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue *vqs, int nvqs); |
| 131 | long vhost_dev_check_owner(struct vhost_dev *); |
| 132 | long vhost_dev_reset_owner(struct vhost_dev *); |
| 133 | void vhost_dev_cleanup(struct vhost_dev *); |
| 134 | long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, unsigned long arg); |
| 135 | int vhost_vq_access_ok(struct vhost_virtqueue *vq); |
| 136 | int vhost_log_access_ok(struct vhost_dev *); |
| 137 | |
Michael S. Tsirkin | d5675bd | 2010-06-24 16:59:59 +0300 | [diff] [blame] | 138 | int vhost_get_vq_desc(struct vhost_dev *, struct vhost_virtqueue *, |
| 139 | struct iovec iov[], unsigned int iov_count, |
| 140 | unsigned int *out_num, unsigned int *in_num, |
| 141 | struct vhost_log *log, unsigned int *log_num); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 142 | void vhost_discard_vq_desc(struct vhost_virtqueue *, int n); |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 143 | |
| 144 | int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 145 | int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads, |
| 146 | unsigned count); |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 147 | void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *, |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 148 | unsigned int id, int len); |
| 149 | void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *, |
| 150 | struct vring_used_elem *heads, unsigned count); |
| 151 | void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *); |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 152 | void vhost_disable_notify(struct vhost_virtqueue *); |
| 153 | bool vhost_enable_notify(struct vhost_virtqueue *); |
| 154 | |
| 155 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, |
| 156 | unsigned int log_num, u64 len); |
| 157 | |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 158 | #define vq_err(vq, fmt, ...) do { \ |
| 159 | pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \ |
| 160 | if ((vq)->error_ctx) \ |
| 161 | eventfd_signal((vq)->error_ctx, 1);\ |
| 162 | } while (0) |
| 163 | |
| 164 | enum { |
| 165 | VHOST_FEATURES = (1 << VIRTIO_F_NOTIFY_ON_EMPTY) | |
| 166 | (1 << VIRTIO_RING_F_INDIRECT_DESC) | |
| 167 | (1 << VHOST_F_LOG_ALL) | |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 168 | (1 << VHOST_NET_F_VIRTIO_NET_HDR) | |
| 169 | (1 << VIRTIO_NET_F_MRG_RXBUF), |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 170 | }; |
| 171 | |
| 172 | static inline int vhost_has_feature(struct vhost_dev *dev, int bit) |
| 173 | { |
Arnd Bergmann | 28457ee | 2010-03-09 19:24:45 +0100 | [diff] [blame] | 174 | unsigned acked_features; |
| 175 | |
| 176 | acked_features = |
| 177 | rcu_dereference_index_check(dev->acked_features, |
| 178 | lockdep_is_held(&dev->mutex)); |
Michael S. Tsirkin | 3a4d5c9 | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 179 | return acked_features & (1 << bit); |
| 180 | } |
| 181 | |
| 182 | #endif |