static void end_cmd(struct nullb_cmd *cmd)
{
struct request_queue *q = NULL;
switch (queue_mode) {
case NULL_Q_MQ:
blk_mq_end_request(cmd->rq, 0);
return;
case NULL_Q_RQ:
INIT_LIST_HEAD(&cmd->rq->queuelist);
blk_end_request_all(cmd->rq, 0);
break;
case NULL_Q_BIO:
bio_endio(cmd->bio);
goto free_cmd;
}
if (cmd->rq)
q = cmd->rq->q;
/* Restart queue if needed, as we are freeing a tag */
if (q && !q->mq_ops && blk_queue_stopped(q)) {
unsigned long flags;
spin_lock_irqsave(q->queue_lock, flags);
if (blk_queue_stopped(q))
blk_start_queue(q);
spin_unlock_irqrestore(q->queue_lock, flags);
}
free_cmd:
free_cmd(cmd);
}
static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
{
struct completion_queue *cq;
struct llist_node *entry;
struct nullb_cmd *cmd;
cq = &per_cpu(completion_queues, smp_processor_id());
while ((entry = llist_del_all(&cq->list)) != NULL) {
entry = llist_reverse_order(entry);
do {
struct request_queue *q = NULL;
cmd = container_of(entry, struct nullb_cmd, ll_list);
entry = entry->next;
if (cmd->rq)
q = cmd->rq->q;
end_cmd(cmd);
if (q && !q->mq_ops && blk_queue_stopped(q)) {
spin_lock(q->queue_lock);
if (blk_queue_stopped(q))
blk_start_queue(q);
spin_unlock(q->queue_lock);
}
} while (entry);
}
return HRTIMER_NORESTART;
}
void __blk_run_queue(struct request_queue *q)
{
if (unlikely(blk_queue_stopped(q)))
return;
__blk_run_queue_uncond(q);
}
static void dm_old_stop_queue(struct request_queue *q)
{
unsigned long flags;
spin_lock_irqsave(q->queue_lock, flags);
if (!blk_queue_stopped(q))
blk_stop_queue(q);
spin_unlock_irqrestore(q->queue_lock, flags);
}
/*
* q->request_fn for old request-based dm.
* Called with the queue lock held.
*/
static void dm_old_request_fn(struct request_queue *q)
{
struct mapped_device *md = q->queuedata;
struct dm_target *ti = md->immutable_target;
struct request *rq;
struct dm_rq_target_io *tio;
sector_t pos = 0;
if (unlikely(!ti)) {
int srcu_idx;
struct dm_table *map = dm_get_live_table(md, &srcu_idx);
if (unlikely(!map)) {
dm_put_live_table(md, srcu_idx);
return;
}
ti = dm_table_find_target(map, pos);
dm_put_live_table(md, srcu_idx);
}
/*
* For suspend, check blk_queue_stopped() and increment
* ->pending within a single queue_lock not to increment the
* number of in-flight I/Os after the queue is stopped in
* dm_suspend().
*/
while (!blk_queue_stopped(q)) {
rq = blk_peek_request(q);
if (!rq)
return;
/* always use block 0 to find the target for flushes for now */
pos = 0;
if (req_op(rq) != REQ_OP_FLUSH)
pos = blk_rq_pos(rq);
if ((dm_old_request_peeked_before_merge_deadline(md) &&
md_in_flight(md) && rq->bio && !bio_multiple_segments(rq->bio) &&
md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) ||
(ti->type->busy && ti->type->busy(ti))) {
blk_delay_queue(q, 10);
return;
}
dm_start_request(md, rq);
tio = tio_from_request(rq);
init_tio(tio, rq, md);
/* Establish tio->ti before queuing work (map_tio_request) */
tio->ti = ti;
kthread_queue_work(&md->kworker, &tio->work);
BUG_ON(!irqs_disabled());
}
}
static void dm_mq_requeue_request(struct request *rq)
{
struct request_queue *q = rq->q;
unsigned long flags;
blk_mq_requeue_request(rq);
spin_lock_irqsave(q->queue_lock, flags);
if (!blk_queue_stopped(q))
blk_mq_kick_requeue_list(q);
spin_unlock_irqrestore(q->queue_lock, flags);
}
/*
* Input/Output thread.
*/
static int sd_io_thread(void *param)
{
struct sd_host *host = param;
struct request *req;
unsigned long flags;
int nr_sectors;
int error;
#if 0
/*
* We are going to perfom badly due to the read problem explained
* above. At least, be nice with other processes trying to use the
* cpu.
*/
set_user_nice(current, 0);
#endif
current->flags |= PF_NOFREEZE|PF_MEMALLOC;
mutex_lock(&host->io_mutex);
for (;;) {
req = NULL;
set_current_state(TASK_INTERRUPTIBLE);
spin_lock_irqsave(&host->queue_lock, flags);
if (!blk_queue_stopped(host->queue))
req = blk_fetch_request(host->queue);
spin_unlock_irqrestore(&host->queue_lock, flags);
if (!req) {
if (kthread_should_stop()) {
set_current_state(TASK_RUNNING);
break;
}
mutex_unlock(&host->io_mutex);
schedule();
mutex_lock(&host->io_mutex);
continue;
}
set_current_state(TASK_INTERRUPTIBLE);
nr_sectors = sd_do_request(host, req);
error = (nr_sectors < 0) ? nr_sectors : 0;
spin_lock_irqsave(&host->queue_lock, flags);
__blk_end_request(req, error, nr_sectors << 9);
spin_unlock_irqrestore(&host->queue_lock, flags);
}
mutex_unlock(&host->io_mutex);
return 0;
}
void blk_run_queue_async(struct request_queue *q)
{
if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
}
static void skd_request_fn(struct request_queue *q)
{
struct skd_device *skdev = q->queuedata;
struct skd_fitmsg_context *skmsg = NULL;
struct fit_msg_hdr *fmh = NULL;
struct skd_request_context *skreq;
struct request *req = NULL;
struct skd_scsi_request *scsi_req;
struct page *page;
unsigned long io_flags;
int error;
u32 lba;
u32 count;
int data_dir;
u32 be_lba;
u32 be_count;
u64 be_dmaa;
u64 cmdctxt;
u32 timo_slot;
void *cmd_ptr;
int flush, fua;
if (skdev->state != SKD_DRVR_STATE_ONLINE) {
skd_request_fn_not_online(q);
return;
}
if (blk_queue_stopped(skdev->queue)) {
if (skdev->skmsg_free_list == NULL ||
skdev->skreq_free_list == NULL ||
skdev->in_flight >= skdev->queue_low_water_mark)
/* There is still some kind of shortage */
return;
queue_flag_clear(QUEUE_FLAG_STOPPED, skdev->queue);
}
/*
* Stop conditions:
* - There are no more native requests
* - There are already the maximum number of requests in progress
* - There are no more skd_request_context entries
* - There are no more FIT msg buffers
*/
for (;; ) {
flush = fua = 0;
req = blk_peek_request(q);
/* Are there any native requests to start? */
if (req == NULL)
break;
lba = (u32)blk_rq_pos(req);
count = blk_rq_sectors(req);
data_dir = rq_data_dir(req);
io_flags = req->cmd_flags;
if (io_flags & REQ_FLUSH)
flush++;
if (io_flags & REQ_FUA)
fua++;
pr_debug("%s:%s:%d new req=%p lba=%u(0x%x) "
"count=%u(0x%x) dir=%d\n",
skdev->name, __func__, __LINE__,
req, lba, lba, count, count, data_dir);
/* At this point we know there is a request */
/* Are too many requets already in progress? */
if (skdev->in_flight >= skdev->cur_max_queue_depth) {
pr_debug("%s:%s:%d qdepth %d, limit %d\n",
skdev->name, __func__, __LINE__,
skdev->in_flight, skdev->cur_max_queue_depth);
break;
}
/* Is a skd_request_context available? */
skreq = skdev->skreq_free_list;
if (skreq == NULL) {
pr_debug("%s:%s:%d Out of req=%p\n",
skdev->name, __func__, __LINE__, q);
break;
}
SKD_ASSERT(skreq->state == SKD_REQ_STATE_IDLE);
SKD_ASSERT((skreq->id & SKD_ID_INCR) == 0);
/* Now we check to see if we can get a fit msg */
if (skmsg == NULL) {
if (skdev->skmsg_free_list == NULL) {
pr_debug("%s:%s:%d Out of msg\n",
skdev->name, __func__, __LINE__);
break;
}
}
skreq->flush_cmd = 0;
skreq->n_sg = 0;
skreq->sg_byte_count = 0;
skreq->discard_page = 0;
/*
* OK to now dequeue request from q.
*
* At this point we are comitted to either start or reject
* the native request. Note that skd_request_context is
* available but is still at the head of the free list.
*/
blk_start_request(req);
skreq->req = req;
skreq->fitmsg_id = 0;
/* Either a FIT msg is in progress or we have to start one. */
if (skmsg == NULL) {
/* Are there any FIT msg buffers available? */
skmsg = skdev->skmsg_free_list;
if (skmsg == NULL) {
pr_debug("%s:%s:%d Out of msg skdev=%p\n",
skdev->name, __func__, __LINE__,
skdev);
break;
}
SKD_ASSERT(skmsg->state == SKD_MSG_STATE_IDLE);
SKD_ASSERT((skmsg->id & SKD_ID_INCR) == 0);
skdev->skmsg_free_list = skmsg->next;
skmsg->state = SKD_MSG_STATE_BUSY;
skmsg->id += SKD_ID_INCR;
/* Initialize the FIT msg header */
fmh = (struct fit_msg_hdr *)skmsg->msg_buf;
memset(fmh, 0, sizeof(*fmh));
fmh->protocol_id = FIT_PROTOCOL_ID_SOFIT;
skmsg->length = sizeof(*fmh);
}
skreq->fitmsg_id = skmsg->id;
/*
* Note that a FIT msg may have just been started
* but contains no SoFIT requests yet.
*/
/*
* Transcode the request, checking as we go. The outcome of
* the transcoding is represented by the error variable.
*/
cmd_ptr = &skmsg->msg_buf[skmsg->length];
memset(cmd_ptr, 0, 32);
be_lba = cpu_to_be32(lba);
be_count = cpu_to_be32(count);
be_dmaa = cpu_to_be64((u64)skreq->sksg_dma_address);
cmdctxt = skreq->id + SKD_ID_INCR;
scsi_req = cmd_ptr;
scsi_req->hdr.tag = cmdctxt;
scsi_req->hdr.sg_list_dma_address = be_dmaa;
if (data_dir == READ)
skreq->sg_data_dir = SKD_DATA_DIR_CARD_TO_HOST;
else
skreq->sg_data_dir = SKD_DATA_DIR_HOST_TO_CARD;
if (io_flags & REQ_DISCARD) {
page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
if (!page) {
pr_err("request_fn:Page allocation failed.\n");
skd_end_request(skdev, skreq, -ENOMEM);
break;
}
skreq->discard_page = 1;
req->completion_data = page;
skd_prep_discard_cdb(scsi_req, skreq, page, lba, count);
} else if (flush == SKD_FLUSH_ZERO_SIZE_FIRST) {
skd_prep_zerosize_flush_cdb(scsi_req, skreq);
SKD_ASSERT(skreq->flush_cmd == 1);
} else {
skd_prep_rw_cdb(scsi_req, data_dir, lba, count);
}
if (fua)
scsi_req->cdb[1] |= SKD_FUA_NV;
if (!req->bio)
goto skip_sg;
error = skd_preop_sg_list(skdev, skreq);
if (error != 0) {
/*
* Complete the native request with error.
* Note that the request context is still at the
* head of the free list, and that the SoFIT request
* was encoded into the FIT msg buffer but the FIT
* msg length has not been updated. In short, the
* only resource that has been allocated but might
* not be used is that the FIT msg could be empty.
*/
pr_debug("%s:%s:%d error Out\n",
skdev->name, __func__, __LINE__);
skd_end_request(skdev, skreq, error);
continue;
}
skip_sg:
scsi_req->hdr.sg_list_len_bytes =
cpu_to_be32(skreq->sg_byte_count);
/* Complete resource allocations. */
skdev->skreq_free_list = skreq->next;
skreq->state = SKD_REQ_STATE_BUSY;
skreq->id += SKD_ID_INCR;
skmsg->length += sizeof(struct skd_scsi_request);
fmh->num_protocol_cmds_coalesced++;
/*
* Update the active request counts.
* Capture the timeout timestamp.
*/
skreq->timeout_stamp = skdev->timeout_stamp;
timo_slot = skreq->timeout_stamp & SKD_TIMEOUT_SLOT_MASK;
skdev->timeout_slot[timo_slot]++;
skdev->in_flight++;
pr_debug("%s:%s:%d req=0x%x busy=%d\n",
skdev->name, __func__, __LINE__,
skreq->id, skdev->in_flight);
/*
* If the FIT msg buffer is full send it.
*/
if (skmsg->length >= SKD_N_FITMSG_BYTES ||
fmh->num_protocol_cmds_coalesced >= skd_max_req_per_msg) {
skd_send_fitmsg(skdev, skmsg);
skmsg = NULL;
fmh = NULL;
}
}
/*
* Is a FIT msg in progress? If it is empty put the buffer back
* on the free list. If it is non-empty send what we got.
* This minimizes latency when there are fewer requests than
* what fits in a FIT msg.
*/
if (skmsg != NULL) {
/* Bigger than just a FIT msg header? */
if (skmsg->length > sizeof(struct fit_msg_hdr)) {
pr_debug("%s:%s:%d sending msg=%p, len %d\n",
skdev->name, __func__, __LINE__,
skmsg, skmsg->length);
skd_send_fitmsg(skdev, skmsg);
} else {
/*
* The FIT msg is empty. It means we got started
* on the msg, but the requests were rejected.
*/
skmsg->state = SKD_MSG_STATE_IDLE;
skmsg->id += SKD_ID_INCR;
skmsg->next = skdev->skmsg_free_list;
skdev->skmsg_free_list = skmsg;
}
skmsg = NULL;
fmh = NULL;
}
/*
* If req is non-NULL it means there is something to do but
* we are out of a resource.
*/
if (req)
blk_stop_queue(skdev->queue);
}
