// TODO: right now there is one channel (remote), in the future, the caller
// may specify local which will cause it to give up the core to do the work.
// creation of additional remote channel also allows the caller to prioritize
// work, because the default policy for the kernel is to roundrobin between them.
int async_syscall(arsc_channel_t* chan, syscall_req_t* req, syscall_desc_t** desc_ptr2)
{
// Note that this assumes one global frontring (TODO)
// abort if there is no room for our request. ring size is currently 64.
// we could spin til it's free, but that could deadlock if this same thread
// is supposed to consume the requests it is waiting on later.
syscall_desc_t* desc = malloc(sizeof (syscall_desc_t));
desc->channel = chan;
syscall_front_ring_t *fr = &(desc->channel->sysfr);
//TODO: can do it locklessly using CAS, but could change with local async calls
struct mcs_lock_qnode local_qn = {0};
mcs_lock_lock(&(chan->aclock), &local_qn);
if (RING_FULL(fr)) {
errno = EBUSY;
return -1;
}
// req_prod_pvt comes in as the previously produced item. need to
// increment to the next available spot, which is the one we'll work on.
// at some point, we need to listen for the responses.
desc->idx = ++(fr->req_prod_pvt);
syscall_req_t* r = RING_GET_REQUEST(fr, desc->idx);
// CAS on the req->status perhaps
req->status = REQ_alloc;
memcpy(r, req, sizeof(syscall_req_t));
r->status = REQ_ready;
// push our updates to syscallfrontring.req_prod_pvt
// note: it is ok to push without protection since it is atomic and kernel
// won't process any requests until they are marked REQ_ready (also atomic)
RING_PUSH_REQUESTS(fr);
//cprintf("DEBUG: sring->req_prod: %d, sring->rsp_prod: %d\n",
mcs_lock_unlock(&desc->channel->aclock, &local_qn);
*desc_ptr2 = desc;
return 0;
}
struct blktap_request*
blktap_ring_make_request(struct blktap *tap)
{
struct blktap_ring *ring = &tap->ring;
struct blktap_request *request;
int usr_idx;
if (RING_FULL(&ring->ring))
return ERR_PTR(-ENOSPC);
request = blktap_request_alloc(tap);
if (!request)
return ERR_PTR(-ENOMEM);
for (usr_idx = 0; usr_idx < BLKTAP_RING_SIZE; usr_idx++)
if (!ring->pending[usr_idx])
break;
BUG_ON(usr_idx >= BLKTAP_RING_SIZE);
request->tap = tap;
request->usr_idx = usr_idx;
ring->pending[usr_idx] = request;
ring->n_pending++;
return request;
}
static void kick_pending_request_queues(struct blkfront_info *info)
{
if (!RING_FULL(&info->ring)) {
/* Re-enable calldowns. */
blk_start_queue(info->rq);
/* Kick things off immediately. */
do_blkif_request(info->rq);
}
}
static void blkfront_wait_slot(struct blkfront_dev *dev)
{
/* Wait for a slot */
if (RING_FULL(&dev->ring)) {
unsigned long flags;
DEFINE_WAIT(w);
local_irq_save(flags);
while (1) {
blkfront_aio_poll(dev);
if (!RING_FULL(&dev->ring))
break;
/* Really no slot, go to sleep. */
add_waiter(w, blkfront_queue);
local_irq_restore(flags);
schedule();
local_irq_save(flags);
}
remove_waiter(w, blkfront_queue);
local_irq_restore(flags);
}
}
static int write_block(FILE *p, blkif_sector_t sector, size_t amt)
{
static uint64_t next_reqid = 1;
blkif_response_t *rsp;
blkif_request_t *req;
int notify, work_to_do;
uint64_t reqid;
RING_IDX i;
/* wait until we can write something */
while(RING_FULL(&p->ring)) runtime_block(1);
/* write out the request */
i = p->ring.req_prod_pvt++;
req = RING_GET_REQUEST(&p->ring, i);
memset(req, 0, sizeof(blkif_request_t));
req->operation = BLKIF_OP_WRITE;
req->nr_segments = 1;
req->handle = p->disk_handle;
req->id = reqid = next_reqid++;
req->sector_number = sector;
req->seg[0].gref = p->block_grant;
req->seg[0].first_sect = 0;
req->seg[0].last_sect = (amt - 1) / 512;
wmb();
RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&p->ring, notify);
if(notify) channel_send(p->chan);
/* wait for it to be satisfied */
do {
while(!RING_HAS_UNCONSUMED_RESPONSES(&p->ring))
runtime_block(1);
i = p->ring.rsp_cons++;
rsp = RING_GET_RESPONSE(&p->ring, i);
} while(rsp->id != reqid);
/* was it successful? */
if(rsp->status != BLKIF_RSP_OKAY) {
printf("PROFILING: Block write failed!\n");
return 0;
}
/* we do writes one at a time, synchronously, so work_to_do should always
be false */
RING_FINAL_CHECK_FOR_RESPONSES(&p->ring, work_to_do);
assert(!work_to_do);
return 1;
}
static int scsifront_do_request(struct vscsifrnt_info *info,
struct vscsifrnt_shadow *shadow)
{
struct vscsiif_front_ring *ring = &(info->ring);
struct vscsiif_request *ring_req;
struct scsi_cmnd *sc = shadow->sc;
uint32_t id;
int i, notify;
if (RING_FULL(&info->ring))
return -EBUSY;
id = scsifront_get_rqid(info); /* use id in response */
if (id >= VSCSIIF_MAX_REQS)
return -EBUSY;
info->shadow[id] = shadow;
shadow->rqid = id;
ring_req = RING_GET_REQUEST(&(info->ring), ring->req_prod_pvt);
ring->req_prod_pvt++;
ring_req->rqid = id;
ring_req->act = shadow->act;
ring_req->ref_rqid = shadow->ref_rqid;
ring_req->nr_segments = shadow->nr_segments;
ring_req->id = sc->device->id;
ring_req->lun = sc->device->lun;
ring_req->channel = sc->device->channel;
ring_req->cmd_len = sc->cmd_len;
BUG_ON(sc->cmd_len > VSCSIIF_MAX_COMMAND_SIZE);
memcpy(ring_req->cmnd, sc->cmnd, sc->cmd_len);
ring_req->sc_data_direction = (uint8_t)sc->sc_data_direction;
ring_req->timeout_per_command = sc->request->timeout / HZ;
for (i = 0; i < (shadow->nr_segments & ~VSCSIIF_SG_GRANT); i++)
ring_req->seg[i] = shadow->seg[i];
RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(ring, notify);
if (notify)
notify_remote_via_irq(info->irq);
return 0;
}
/*
* do_blkif_request
* read a block; request is in a request queue
*/
static void do_blkif_request(struct request_queue *rq)
{
struct blkfront_info *info = NULL;
struct request *req;
int queued;
pr_debug("Entered do_blkif_request\n");
queued = 0;
while ((req = blk_peek_request(rq)) != NULL) {
info = req->rq_disk->private_data;
if (RING_FULL(&info->ring))
goto wait;
blk_start_request(req);
if (!blk_fs_request(req)) {
__blk_end_request_all(req, -EIO);
continue;
}
pr_debug("do_blk_req %p: cmd %p, sec %lx, "
"(%u/%u) buffer:%p [%s]\n",
req, req->cmd, (unsigned long)blk_rq_pos(req),
blk_rq_cur_sectors(req), blk_rq_sectors(req),
req->buffer, rq_data_dir(req) ? "write" : "read");
if (blkif_queue_request(req)) {
blk_requeue_request(rq, req);
wait:
/* Avoid pointless unplugs. */
blk_stop_queue(rq);
break;
}
queued++;
}
if (queued != 0)
flush_requests(info);
}
/* walk dirty map and enqueue read requests.
* returns: 0 when entire bitmap has been enqueued,
* 1 when the ring is full
* -1 on error
*/
static int writelog_enqueue_requests(struct writelog* wl)
{
struct disk_range* range = wl->shm;
log_request_t* req;
for (range = wl->cur; (void*)range < bmend(wl->shm); range++) {
if (!range->count)
break;
if (RING_FULL(&wl->fring))
break;
/* insert range into request stream */
/* 1. get next request slot from ring */
/* 2. ensure enough shm space is available */
BDPRINTF("enqueueing dirty extent: %"PRIu64":%u (ring space: %d/%d)",
range->sector, range->count, RING_FREE_REQUESTS(&wl->fring),
RING_SIZE(&wl->fring));
req = RING_GET_REQUEST(&wl->fring, wl->fring.req_prod_pvt);
req->sector = range->sector;
req->count = range->count;
/* ... */
req->offset = 0;
wl->fring.req_prod_pvt++;
wl->inflight++;
}
wl->cur = range;
if (range->count)
return 1;
return 0;
}
/*
* blkif_queue_request
*
* request block io
*
* id: for guest use only.
* operation: BLKIF_OP_{READ,WRITE,PROBE}
* buffer: buffer to read/write into. this should be a
* virtual address in the guest os.
*/
int ixp_queue_request(struct app_request *app_req, void *metadata)
{
struct ixpfront_info *info = (struct ixpfront_info *) metadata;
unsigned long buffer_mfn;
struct ixp_request *ring_req;
char *req_page = 0, *curr_pos;
unsigned long id;
int ref, err;
grant_ref_t gref_head;
if (unlikely(info->connected != IXP_STATE_CONNECTED))
return 1;
if (RING_FULL(&info->ring)) {
printk(KERN_ERR "%s:Ring full - returning backpressure\n", __FUNCTION__);
return 1;
}
if (gnttab_alloc_grant_references(
IXPIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
/*gnttab_request_free_callback(
&info->callback,
ixp_restart_queue_callback,
info,
IXP_MAX_SEGMENTS_PER_REQUEST);*/
return 1;
}
/* Fill out a communications ring structure. */
ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
id = get_id_from_freelist(info);
ring_req->id = id;
ring_req->handle = info->handle;
ring_req->operation = IXP_OP_3DES_ENCRYPT;
ring_req->nr_segments = 1;
BUG_ON(ring_req->nr_segments > IXPIF_MAX_SEGMENTS_PER_REQUEST);
req_page = (char *)__get_free_page(GFP_NOIO | __GFP_HIGH);
if(req_page == 0) {
printk(KERN_ERR "ixp_queue_request:Error allocating memory");
return 1;
}
((struct des_request *)req_page)->key_size = app_req->key_size;
((struct des_request *)req_page)->iv_size = app_req->iv_size;
((struct des_request *)req_page)->msg_size = app_req->msg_size;
curr_pos = req_page + sizeof(struct des_request);
memcpy(curr_pos, app_req->key, app_req->key_size);
curr_pos += app_req->key_size;
memcpy(curr_pos, app_req->iv, app_req->iv_size);
curr_pos += app_req->iv_size;
memcpy(curr_pos, app_req->msg, app_req->msg_size);
curr_pos += app_req->msg_size;
buffer_mfn = virt_to_mfn(req_page);
/* install a grant reference. */
ref = gnttab_claim_grant_reference(&gref_head);
BUG_ON(ref == -ENOSPC);
gnttab_grant_foreign_access_ref(
ref,
info->xbdev->otherend_id,
buffer_mfn,
0);
info->shadow[id].r_params.presp = app_req->presp;
info->shadow[id].r_params.callbk_tag = app_req->callbk_tag;
info->shadow[id].frame[0] = mfn_to_pfn(buffer_mfn);
info->shadow[id].req_page = req_page;
ring_req->seg[0] =
(struct ixp_request_segment) {
.gref = ref
};
info->ring.req_prod_pvt++;
/* Keep a private copy so we can reissue requests when recovering. */
info->shadow[id].req = *ring_req;
flush_requests(info);
//gnttab_free_grant_references(gref_head);
return 0;
}