diff options
author | Vasu Dev <vasu.dev@intel.com> | 2009-02-27 10:56:27 -0800 |
---|---|---|
committer | James Bottomley <James.Bottomley@HansenPartnership.com> | 2009-03-10 09:09:21 -0500 |
commit | c826a3145736e3baabebccfd0aecfbb6dae059f2 (patch) | |
tree | 4c156c85a8eec039973b369cff052206da128cbe /drivers/scsi/fcoe/libfcoe.c | |
parent | e904158159e9812d06646767b7c81846dc3b05e6 (diff) |
[SCSI] fcoe: Out of order tx frames was causing several check condition SCSI status
frames followed by these errors in log.
[sdp] Result: hostbyte=DID_OK driverbyte=DRIVER_SENSE,SUGGEST_OK
[sdp] Sense Key : Aborted Command [current]
[sdp] Add. Sense: Data phase error
This was causing some test apps to exit due to write failure under heavy
load.
This was due to a race around adding and removing tx frame skb in
fcoe_pending_queue, Chris Leech helped me to find that brief unlocking
period when pulling skb from fcoe_pending_queue in various contexts
(fcoe_watchdog and fcoe_xmit) and then adding skb back into fcoe_pending_queue
up on a failed fcoe_start_io could change skb/tx frame order in
fcoe_pending_queue. Thanks Chris.
This patch allows only single context to pull skb from fcoe_pending_queue
at any time to prevent above described ordering issue/race by use of
fcoe_pending_queue_active flag.
This patch simplified fcoe_watchdog with modified fcoe_check_wait_queue by
use of FCOE_LOW_QUEUE_DEPTH instead previously used several conditionals
to clear and set lp->qfull.
I think FCOE_MAX_QUEUE_DEPTH with FCOE_LOW_QUEUE_DEPTH will work better
in re/setting lp->qfull and these could be fine tuned for performance.
Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Diffstat (limited to 'drivers/scsi/fcoe/libfcoe.c')
-rw-r--r-- | drivers/scsi/fcoe/libfcoe.c | 45 |
1 files changed, 21 insertions, 24 deletions
diff --git a/drivers/scsi/fcoe/libfcoe.c b/drivers/scsi/fcoe/libfcoe.c index 7265e093799..14dd8a0402b 100644 --- a/drivers/scsi/fcoe/libfcoe.c +++ b/drivers/scsi/fcoe/libfcoe.c @@ -49,6 +49,7 @@ static int debug_fcoe; #define FCOE_MAX_QUEUE_DEPTH 256 +#define FCOE_LOW_QUEUE_DEPTH 32 /* destination address mode */ #define FCOE_GW_ADDR_MODE 0x00 @@ -723,21 +724,12 @@ static void fcoe_recv_flogi(struct fcoe_softc *fc, struct fc_frame *fp, u8 *sa) */ void fcoe_watchdog(ulong vp) { - struct fc_lport *lp; struct fcoe_softc *fc; - int qfilled = 0; read_lock(&fcoe_hostlist_lock); list_for_each_entry(fc, &fcoe_hostlist, list) { - lp = fc->lp; - if (lp) { - if (fc->fcoe_pending_queue.qlen > FCOE_MAX_QUEUE_DEPTH) - qfilled = 1; - if (fcoe_check_wait_queue(lp) < FCOE_MAX_QUEUE_DEPTH) { - if (qfilled) - lp->qfull = 0; - } - } + if (fc->lp) + fcoe_check_wait_queue(fc->lp); } read_unlock(&fcoe_hostlist_lock); @@ -753,8 +745,8 @@ void fcoe_watchdog(ulong vp) * * This empties the wait_queue, dequeue the head of the wait_queue queue * and calls fcoe_start_io() for each packet, if all skb have been - * transmitted, return 0 if a error occurs, then restore wait_queue and - * try again later. + * transmitted, return qlen or -1 if a error occurs, then restore + * wait_queue and try again later. * * The wait_queue is used when the skb transmit fails. skb will go * in the wait_queue which will be emptied by the time function OR @@ -764,33 +756,38 @@ void fcoe_watchdog(ulong vp) */ static int fcoe_check_wait_queue(struct fc_lport *lp) { - int rc; struct sk_buff *skb; struct fcoe_softc *fc; + int rc = -1; fc = lport_priv(lp); spin_lock_bh(&fc->fcoe_pending_queue.lock); - /* - * if interface pending queue full then set qfull in lport. - */ - if (fc->fcoe_pending_queue.qlen > FCOE_MAX_QUEUE_DEPTH) - lp->qfull = 1; + if (fc->fcoe_pending_queue_active) + goto out; + fc->fcoe_pending_queue_active = 1; if (fc->fcoe_pending_queue.qlen) { while ((skb = __skb_dequeue(&fc->fcoe_pending_queue)) != NULL) { spin_unlock_bh(&fc->fcoe_pending_queue.lock); rc = fcoe_start_io(skb); - if (rc) { + if (rc) fcoe_insert_wait_queue_head(lp, skb); - return rc; - } spin_lock_bh(&fc->fcoe_pending_queue.lock); + if (rc) + break; } - if (fc->fcoe_pending_queue.qlen < FCOE_MAX_QUEUE_DEPTH) + /* + * if interface pending queue is below FCOE_LOW_QUEUE_DEPTH + * then clear qfull flag. + */ + if (fc->fcoe_pending_queue.qlen < FCOE_LOW_QUEUE_DEPTH) lp->qfull = 0; } + fc->fcoe_pending_queue_active = 0; + rc = fc->fcoe_pending_queue.qlen; +out: spin_unlock_bh(&fc->fcoe_pending_queue.lock); - return fc->fcoe_pending_queue.qlen; + return rc; } /** |