X-Git-Url: https://err.no/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=net%2F9p%2Ftrans_virtio.c;h=42adc052b149251a46e3ac8a1bf039c91f00417f;hb=9a5467fd600669cda488771dac3e951034fe2b08;hp=42eea5fe262895587522ac689b1f06d2cf7ad7a1;hpb=1f7d4f8395093021ed2262296179cfe71bd5e2ec;p=linux-2.6 diff --git a/net/9p/trans_virtio.c b/net/9p/trans_virtio.c index 42eea5fe26..42adc052b1 100644 --- a/net/9p/trans_virtio.c +++ b/net/9p/trans_virtio.c @@ -1,17 +1,8 @@ /* * The Guest 9p transport driver * - * This is a trivial pipe-based transport driver based on the lguest console - * code: we use lguest's DMA mechanism to send bytes out, and register a - * DMA buffer to receive bytes in. It is assumed to be present and available - * from the very beginning of boot. - * - * This may be have been done by just instaniating another HVC console, - * but HVC's blocksize of 16 bytes is annoying and painful to performance. - * - * A more efficient transport could be built based on the virtio block driver - * but it requires some changes in the 9p transport model (which are in - * progress) + * This is a block based transport driver based on the lguest block driver + * code. * */ /* @@ -55,229 +46,410 @@ #include #include +#define VIRTQUEUE_NUM 128 + /* a single mutex to manage channel initialization and attachment */ -static DECLARE_MUTEX(virtio_9p_lock); +static DEFINE_MUTEX(virtio_9p_lock); /* global which tracks highest initialized channel */ static int chan_index; -/* We keep all per-channel information in a structure. +#define P9_INIT_MAXTAG 16 + + +/** + * enum p9_req_status_t - virtio request status + * @REQ_STATUS_IDLE: request slot unused + * @REQ_STATUS_SENT: request sent to server + * @REQ_STATUS_RCVD: response received from server + * @REQ_STATUS_FLSH: request has been flushed + * + * The @REQ_STATUS_IDLE state is used to mark a request slot as unused + * but use is actually tracked by the idpool structure which handles tag + * id allocation. + * + */ + +enum p9_req_status_t { + REQ_STATUS_IDLE, + REQ_STATUS_SENT, + REQ_STATUS_RCVD, + REQ_STATUS_FLSH, +}; + +/** + * struct p9_req_t - virtio request slots + * @status: status of this request slot + * @wq: wait_queue for the client to block on for this request + * + * The virtio transport uses an array to track outstanding requests + * instead of a list. While this may incurr overhead during initial + * allocation or expansion, it makes request lookup much easier as the + * tag id is a index into an array. (We use tag+1 so that we can accomodate + * the -1 tag for the T_VERSION request). + * This also has the nice effect of only having to allocate wait_queues + * once, instead of constantly allocating and freeing them. Its possible + * other resources could benefit from this scheme as well. + * + */ + +struct p9_req_t { + int status; + wait_queue_head_t *wq; +}; + +/** + * struct virtio_chan - per-instance transport information + * @initialized: whether the channel is initialized + * @inuse: whether the channel is in use + * @lock: protects multiple elements within this structure + * @vdev: virtio dev associated with this channel + * @vq: virtio queue associated with this channel + * @tagpool: accounting for tag ids (and request slots) + * @reqs: array of request slots + * @max_tag: current number of request_slots allocated + * @sg: scatter gather list which is used to pack a request (protected?) + * + * We keep all per-channel information in a structure. * This structure is allocated within the devices dev->mem space. * A pointer to the structure will get put in the transport private. + * */ + static struct virtio_chan { - bool initialized; /* channel is initialized */ - bool inuse; /* channel is in use */ + bool initialized; + bool inuse; + + spinlock_t lock; - struct virtqueue *in_vq, *out_vq; struct virtio_device *vdev; + struct virtqueue *vq; - /* This is our input buffer, and how much data is left in it. */ - unsigned int in_len; - char *in, *inbuf; + struct p9_idpool *tagpool; + struct p9_req_t *reqs; + int max_tag; - wait_queue_head_t wq; /* waitq for buffer */ + /* Scatterlist: can be too big for stack. */ + struct scatterlist sg[VIRTQUEUE_NUM]; } channels[MAX_9P_CHAN]; +/** + * p9_lookup_tag - Lookup requests by tag + * @c: virtio channel to lookup tag within + * @tag: numeric id for transaction + * + * this is a simple array lookup, but will grow the + * request_slots as necessary to accomodate transaction + * ids which did not previously have a slot. + * + * Bugs: there is currently no upper limit on request slots set + * here, but that should be constrained by the id accounting. + */ + +static struct p9_req_t *p9_lookup_tag(struct virtio_chan *c, u16 tag) +{ + /* This looks up the original request by tag so we know which + * buffer to read the data into */ + tag++; + + while (tag >= c->max_tag) { + int old_max = c->max_tag; + int count; + + if (c->max_tag) + c->max_tag *= 2; + else + c->max_tag = P9_INIT_MAXTAG; + + c->reqs = krealloc(c->reqs, sizeof(struct p9_req_t)*c->max_tag, + GFP_ATOMIC); + if (!c->reqs) { + printk(KERN_ERR "Couldn't grow tag array\n"); + BUG(); + } + for (count = old_max; count < c->max_tag; count++) { + c->reqs[count].status = REQ_STATUS_IDLE; + c->reqs[count].wq = kmalloc(sizeof(wait_queue_head_t), + GFP_ATOMIC); + if (!c->reqs[count].wq) { + printk(KERN_ERR "Couldn't grow tag array\n"); + BUG(); + } + init_waitqueue_head(c->reqs[count].wq); + } + } + + return &c->reqs[tag]; +} + + /* How many bytes left in this page. */ static unsigned int rest_of_page(void *data) { return PAGE_SIZE - ((unsigned long)data % PAGE_SIZE); } -static int p9_virtio_write(struct p9_trans *trans, void *buf, int count) -{ - struct virtio_chan *chan = (struct virtio_chan *) trans->priv; - struct virtqueue *out_vq = chan->out_vq; - struct scatterlist sg[1]; - unsigned int len; +/** + * p9_virtio_close - reclaim resources of a channel + * @trans: transport state + * + * This reclaims a channel by freeing its resources and + * reseting its inuse flag. + * + */ - P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio write (%d)\n", count); +static void p9_virtio_close(struct p9_trans *trans) +{ + struct virtio_chan *chan = trans->priv; + int count; + unsigned long flags; - /* keep it simple - make sure we don't overflow a page */ - if (rest_of_page(buf) < count) - count = rest_of_page(buf); + spin_lock_irqsave(&chan->lock, flags); + p9_idpool_destroy(chan->tagpool); + for (count = 0; count < chan->max_tag; count++) + kfree(chan->reqs[count].wq); + kfree(chan->reqs); + chan->max_tag = 0; + spin_unlock_irqrestore(&chan->lock, flags); - sg_init_one(sg, buf, count); + mutex_lock(&virtio_9p_lock); + chan->inuse = false; + mutex_unlock(&virtio_9p_lock); - /* add_buf wants a token to identify this buffer: we hand it any - * non-NULL pointer, since there's only ever one buffer. */ - if (out_vq->vq_ops->add_buf(out_vq, sg, 1, 0, (void *)1) == 0) { - /* Tell Host to go! */ - out_vq->vq_ops->kick(out_vq); - /* Chill out until it's done with the buffer. */ - while (!out_vq->vq_ops->get_buf(out_vq, &len)) - cpu_relax(); - } + kfree(trans); +} - P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio wrote (%d)\n", count); +/** + * req_done - callback which signals activity from the server + * @vq: virtio queue activity was received on + * + * This notifies us that the server has triggered some activity + * on the virtio channel - most likely a response to request we + * sent. Figure out which requests now have responses and wake up + * those threads. + * + * Bugs: could do with some additional sanity checking, but appears to work. + * + */ - /* We're expected to return the amount of data we wrote: all of it. */ - return count; +static void req_done(struct virtqueue *vq) +{ + struct virtio_chan *chan = vq->vdev->priv; + struct p9_fcall *rc; + unsigned int len; + unsigned long flags; + struct p9_req_t *req; + + spin_lock_irqsave(&chan->lock, flags); + while ((rc = chan->vq->vq_ops->get_buf(chan->vq, &len)) != NULL) { + req = p9_lookup_tag(chan, rc->tag); + req->status = REQ_STATUS_RCVD; + wake_up(req->wq); + } + /* In case queue is stopped waiting for more buffers. */ + spin_unlock_irqrestore(&chan->lock, flags); } -/* Create a scatter-gather list representing our input buffer and put it in the - * queue. */ -static void add_inbuf(struct virtio_chan *chan) -{ - struct scatterlist sg[1]; +/** + * pack_sg_list - pack a scatter gather list from a linear buffer + * @sg: scatter/gather list to pack into + * @start: which segment of the sg_list to start at + * @limit: maximum segment to pack data to + * @data: data to pack into scatter/gather list + * @count: amount of data to pack into the scatter/gather list + * + * sg_lists have multiple segments of various sizes. This will pack + * arbitrary data into an existing scatter gather list, segmenting the + * data as necessary within constraints. + * + */ - sg_init_one(sg, chan->inbuf, PAGE_SIZE); +static int +pack_sg_list(struct scatterlist *sg, int start, int limit, char *data, + int count) +{ + int s; + int index = start; + + while (count) { + s = rest_of_page(data); + if (s > count) + s = count; + sg_set_buf(&sg[index++], data, s); + count -= s; + data += s; + BUG_ON(index > limit); + } - /* We should always be able to add one buffer to an empty queue. */ - if (chan->in_vq->vq_ops->add_buf(chan->in_vq, sg, 0, 1, chan->inbuf)) - BUG(); - chan->in_vq->vq_ops->kick(chan->in_vq); + return index-start; } -static int p9_virtio_read(struct p9_trans *trans, void *buf, int count) -{ - struct virtio_chan *chan = (struct virtio_chan *) trans->priv; - struct virtqueue *in_vq = chan->in_vq; +/** + * p9_virtio_rpc - issue a request and wait for a response + * @t: transport state + * @tc: &p9_fcall request to transmit + * @rc: &p9_fcall to put reponse into + * + */ - P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio read (%d)\n", count); +static int +p9_virtio_rpc(struct p9_trans *t, struct p9_fcall *tc, struct p9_fcall **rc) +{ + int in, out; + int n, err, size; + struct virtio_chan *chan = t->priv; + char *rdata; + struct p9_req_t *req; + unsigned long flags; + + if (*rc == NULL) { + *rc = kmalloc(sizeof(struct p9_fcall) + t->msize, GFP_KERNEL); + if (!*rc) + return -ENOMEM; + } - /* If we don't have an input queue yet, we can't get input. */ - BUG_ON(!in_vq); + rdata = (char *)*rc+sizeof(struct p9_fcall); - /* No buffer? Try to get one. */ - if (!chan->in_len) { - chan->in = in_vq->vq_ops->get_buf(in_vq, &chan->in_len); - if (!chan->in) - return 0; + n = P9_NOTAG; + if (tc->id != P9_TVERSION) { + n = p9_idpool_get(chan->tagpool); + if (n < 0) + return -ENOMEM; } - /* You want more than we have to give? Well, try wanting less! */ - if (chan->in_len < count) - count = chan->in_len; + spin_lock_irqsave(&chan->lock, flags); + req = p9_lookup_tag(chan, n); + spin_unlock_irqrestore(&chan->lock, flags); - /* Copy across to their buffer and increment offset. */ - memcpy(buf, chan->in, count); - chan->in += count; - chan->in_len -= count; + p9_set_tag(tc, n); - /* Finished? Re-register buffer so Host will use it again. */ - if (chan->in_len == 0) - add_inbuf(chan); + P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio rpc tag %d\n", n); - P9_DPRINTK(P9_DEBUG_TRANS, "9p debug: virtio finished read (%d)\n", - count); + out = pack_sg_list(chan->sg, 0, VIRTQUEUE_NUM, tc->sdata, tc->size); + in = pack_sg_list(chan->sg, out, VIRTQUEUE_NUM-out, rdata, t->msize); - return count; -} + req->status = REQ_STATUS_SENT; -/* The poll function is used by 9p transports to determine if there - * is there is activity available on a particular channel. In our case - * we use it to wait for a callback from the input routines. - */ -static unsigned int -p9_virtio_poll(struct p9_trans *trans, struct poll_table_struct *pt) -{ - struct virtio_chan *chan = (struct virtio_chan *)trans->priv; - struct virtqueue *in_vq = chan->in_vq; - int ret = POLLOUT; /* we can always handle more output */ + if (chan->vq->vq_ops->add_buf(chan->vq, chan->sg, out, in, tc)) { + P9_DPRINTK(P9_DEBUG_TRANS, + "9p debug: virtio rpc add_buf returned failure"); + return -EIO; + } - poll_wait(NULL, &chan->wq, pt); + chan->vq->vq_ops->kick(chan->vq); - /* No buffer? Try to get one. */ - if (!chan->in_len) - chan->in = in_vq->vq_ops->get_buf(in_vq, &chan->in_len); + wait_event(*req->wq, req->status == REQ_STATUS_RCVD); - if (chan->in_len) - ret |= POLLIN; + size = le32_to_cpu(*(__le32 *) rdata); - return ret; -} + err = p9_deserialize_fcall(rdata, size, *rc, t->extended); + if (err < 0) { + P9_DPRINTK(P9_DEBUG_TRANS, + "9p debug: virtio rpc deserialize returned %d\n", err); + return err; + } -static void p9_virtio_close(struct p9_trans *trans) -{ - struct virtio_chan *chan = trans->priv; +#ifdef CONFIG_NET_9P_DEBUG + if ((p9_debug_level&P9_DEBUG_FCALL) == P9_DEBUG_FCALL) { + char buf[150]; - down(&virtio_9p_lock); - chan->inuse = false; - up(&virtio_9p_lock); + p9_printfcall(buf, sizeof(buf), *rc, t->extended); + printk(KERN_NOTICE ">>> %p %s\n", t, buf); + } +#endif - kfree(trans); -} + if (n != P9_NOTAG && p9_idpool_check(n, chan->tagpool)) + p9_idpool_put(n, chan->tagpool); -static void p9_virtio_intr(struct virtqueue *q) -{ - struct virtio_chan *chan = q->vdev->priv; + req->status = REQ_STATUS_IDLE; - P9_DPRINTK(P9_DEBUG_TRANS, "9p poll_wakeup: %p\n", &chan->wq); - wake_up_interruptible(&chan->wq); + return 0; } -static int p9_virtio_probe(struct virtio_device *dev) +/** + * p9_virtio_probe - probe for existence of 9P virtio channels + * @vdev: virtio device to probe + * + * This probes for existing virtio channels. At present only + * a single channel is in use, so in the future more work may need + * to be done here. + * + */ + +static int p9_virtio_probe(struct virtio_device *vdev) { int err; struct virtio_chan *chan; int index; - down(&virtio_9p_lock); + mutex_lock(&virtio_9p_lock); index = chan_index++; chan = &channels[index]; - up(&virtio_9p_lock); + mutex_unlock(&virtio_9p_lock); if (chan_index > MAX_9P_CHAN) { printk(KERN_ERR "9p: virtio: Maximum channels exceeded\n"); BUG(); - } - - chan->vdev = dev; - - /* This is the scratch page we use to receive console input */ - chan->inbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); - if (!chan->inbuf) { err = -ENOMEM; goto fail; } - /* Find the input queue. */ - dev->priv = chan; - chan->in_vq = dev->config->find_vq(dev, 0, p9_virtio_intr); - if (IS_ERR(chan->in_vq)) { - err = PTR_ERR(chan->in_vq); - goto free; - } + chan->vdev = vdev; - chan->out_vq = dev->config->find_vq(dev, 1, NULL); - if (IS_ERR(chan->out_vq)) { - err = PTR_ERR(chan->out_vq); - goto free_in_vq; + /* We expect one virtqueue, for requests. */ + chan->vq = vdev->config->find_vq(vdev, 0, req_done); + if (IS_ERR(chan->vq)) { + err = PTR_ERR(chan->vq); + goto out_free_vq; } + chan->vq->vdev->priv = chan; + spin_lock_init(&chan->lock); - init_waitqueue_head(&chan->wq); + sg_init_table(chan->sg, VIRTQUEUE_NUM); - /* Register the input buffer the first time. */ - add_inbuf(chan); chan->inuse = false; chan->initialized = true; - return 0; -free_in_vq: - dev->config->del_vq(chan->in_vq); -free: - kfree(chan->inbuf); +out_free_vq: + vdev->config->del_vq(chan->vq); fail: - down(&virtio_9p_lock); + mutex_lock(&virtio_9p_lock); chan_index--; - up(&virtio_9p_lock); + mutex_unlock(&virtio_9p_lock); return err; } -/* This sets up a transport channel for 9p communication. Right now + +/** + * p9_virtio_create - allocate a new virtio channel + * @devname: string identifying the channel to connect to (unused) + * @args: args passed from sys_mount() for per-transport options (unused) + * @msize: requested maximum packet size + * @extended: 9p2000.u enabled flag + * + * This sets up a transport channel for 9p communication. Right now * we only match the first available channel, but eventually we couldlook up * alternate channels by matching devname versus a virtio_config entry. * We use a simple reference count mechanism to ensure that only a single - * mount has a channel open at a time. */ -static struct p9_trans *p9_virtio_create(const char *devname, char *args) + * mount has a channel open at a time. + * + * Bugs: doesn't allow identification of a specific channel + * to allocate, channels are allocated sequentially. This was + * a pragmatic decision to get things rolling, but ideally some + * way of identifying the channel to attach to would be nice + * if we are going to support multiple channels. + * + */ + +static struct p9_trans * +p9_virtio_create(const char *devname, char *args, int msize, + unsigned char extended) { struct p9_trans *trans; - int index = 0; struct virtio_chan *chan = channels; + int index = 0; - down(&virtio_9p_lock); + mutex_lock(&virtio_9p_lock); while (index < MAX_9P_CHAN) { if (chan->initialized && !chan->inuse) { chan->inuse = true; @@ -287,28 +459,54 @@ static struct p9_trans *p9_virtio_create(const char *devname, char *args) chan = &channels[index]; } } - up(&virtio_9p_lock); + mutex_unlock(&virtio_9p_lock); if (index >= MAX_9P_CHAN) { - printk(KERN_ERR "9p: virtio: couldn't find a free channel\n"); - return NULL; + printk(KERN_ERR "9p: no channels available\n"); + return ERR_PTR(-ENODEV); + } + + chan->tagpool = p9_idpool_create(); + if (IS_ERR(chan->tagpool)) { + printk(KERN_ERR "9p: couldn't allocate tagpool\n"); + return ERR_PTR(-ENOMEM); } + p9_idpool_get(chan->tagpool); /* reserve tag 0 */ + chan->max_tag = 0; + chan->reqs = NULL; trans = kmalloc(sizeof(struct p9_trans), GFP_KERNEL); if (!trans) { printk(KERN_ERR "9p: couldn't allocate transport\n"); return ERR_PTR(-ENOMEM); } - - trans->write = p9_virtio_write; - trans->read = p9_virtio_read; + trans->extended = extended; + trans->msize = msize; trans->close = p9_virtio_close; - trans->poll = p9_virtio_poll; + trans->rpc = p9_virtio_rpc; trans->priv = chan; return trans; } +/** + * p9_virtio_remove - clean up resources associated with a virtio device + * @vdev: virtio device to remove + * + */ + +static void p9_virtio_remove(struct virtio_device *vdev) +{ + struct virtio_chan *chan = vdev->priv; + + BUG_ON(chan->inuse); + + if (chan->initialized) { + vdev->config->del_vq(chan->vq); + chan->initialized = false; + } +} + #define VIRTIO_ID_9P 9 static struct virtio_device_id id_table[] = { @@ -322,12 +520,13 @@ static struct virtio_driver p9_virtio_drv = { .driver.owner = THIS_MODULE, .id_table = id_table, .probe = p9_virtio_probe, + .remove = p9_virtio_remove, }; static struct p9_trans_module p9_virtio_trans = { .name = "virtio", .create = p9_virtio_create, - .maxsize = PAGE_SIZE, + .maxsize = PAGE_SIZE*16, .def = 0, }; @@ -343,7 +542,13 @@ static int __init p9_virtio_init(void) return register_virtio_driver(&p9_virtio_drv); } +static void __exit p9_virtio_cleanup(void) +{ + unregister_virtio_driver(&p9_virtio_drv); +} + module_init(p9_virtio_init); +module_exit(p9_virtio_cleanup); MODULE_DEVICE_TABLE(virtio, id_table); MODULE_AUTHOR("Eric Van Hensbergen ");