[REPOST #3][PATCH 1/6] lpfc 8.1.7 : Adding statistics reset callback for FC transport

[REPOST #3][PATCH 1/6] lpfc 8.1.7 : Adding statistics reset callback for FC transport

[James Smart]

Sorry folks - a last code review identified one last reference
to seconds_since_last_reset that didn't get caught.

--

Patch updated for comments from Rolf Eike Beer
 - better handling of seconds_since_last_reset
 - remove some spaces around use of "sizeof()"

---

Adding statistics reset callback for FC transport

Signed-off-by: James Smart <James.Smart@emulex.com>




Adding statistics reset callback for FC transport

Signed-off-by: James Smart <James.Smart@emulex.com>


diff -upNr a/drivers/scsi/lpfc/lpfc_attr.c b/drivers/scsi/lpfc/lpfc_attr.c
--- a/drivers/scsi/lpfc/lpfc_attr.c	2006-07-31 10:41:15.000000000 -0400
+++ b/drivers/scsi/lpfc/lpfc_attr.c	2006-08-02 15:13:42.000000000 -0400
@@ -1210,8 +1210,10 @@ lpfc_get_stats(struct Scsi_Host *shost)
 	struct lpfc_hba *phba = (struct lpfc_hba *)shost->hostdata;
 	struct lpfc_sli *psli = &phba->sli;
 	struct fc_host_statistics *hs = &phba->link_stats;
+	struct lpfc_lnk_stat * lso = &psli->lnk_stat_offsets;
 	LPFC_MBOXQ_t *pmboxq;
 	MAILBOX_t *pmb;
+	unsigned long seconds;
 	int rc = 0;
 
 	pmboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
@@ -1272,22 +1274,103 @@ lpfc_get_stats(struct Scsi_Host *shost)
 	hs->invalid_crc_count = pmb->un.varRdLnk.crcCnt;
 	hs->error_frames = pmb->un.varRdLnk.crcCnt;
 
+	hs->link_failure_count -= lso->link_failure_count;
+	hs->loss_of_sync_count -= lso->loss_of_sync_count;
+	hs->loss_of_signal_count -= lso->loss_of_signal_count;
+	hs->prim_seq_protocol_err_count -= lso->prim_seq_protocol_err_count;
+	hs->invalid_tx_word_count -= lso->invalid_tx_word_count;
+	hs->invalid_crc_count -= lso->invalid_crc_count;
+	hs->error_frames -= lso->error_frames;
+
 	if (phba->fc_topology == TOPOLOGY_LOOP) {
 		hs->lip_count = (phba->fc_eventTag >> 1);
+		hs->lip_count -= lso->link_events;
 		hs->nos_count = -1;
 	} else {
 		hs->lip_count = -1;
 		hs->nos_count = (phba->fc_eventTag >> 1);
+		hs->nos_count -= lso->link_events;
 	}
 
 	hs->dumped_frames = -1;
 
-/* FIX ME */
-	/*hs->SecondsSinceLastReset = (jiffies - lpfc_loadtime) / HZ;*/
+	seconds = get_seconds();
+	if (seconds < psli->stats_start)
+		hs->seconds_since_last_reset = seconds +
+				((unsigned long)-1 - psli->stats_start);
+	else
+		hs->seconds_since_last_reset = seconds - psli->stats_start;
 
 	return hs;
 }
 
+static void
+lpfc_reset_stats(struct Scsi_Host *shost)
+{
+	struct lpfc_hba *phba = (struct lpfc_hba *)shost->hostdata;
+	struct lpfc_sli *psli = &phba->sli;
+	struct lpfc_lnk_stat * lso = &psli->lnk_stat_offsets;
+	LPFC_MBOXQ_t *pmboxq;
+	MAILBOX_t *pmb;
+	int rc = 0;
+
+	pmboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
+	if (!pmboxq)
+		return;
+	memset(pmboxq, 0, sizeof(LPFC_MBOXQ_t));
+
+	pmb = &pmboxq->mb;
+	pmb->mbxCommand = MBX_READ_STATUS;
+	pmb->mbxOwner = OWN_HOST;
+	pmb->un.varWords[0] = 0x1; /* reset request */
+	pmboxq->context1 = NULL;
+
+	if ((phba->fc_flag & FC_OFFLINE_MODE) ||
+		(!(psli->sli_flag & LPFC_SLI2_ACTIVE)))
+		rc = lpfc_sli_issue_mbox(phba, pmboxq, MBX_POLL);
+	else
+		rc = lpfc_sli_issue_mbox_wait(phba, pmboxq, phba->fc_ratov * 2);
+
+	if (rc != MBX_SUCCESS) {
+		if (rc == MBX_TIMEOUT)
+			pmboxq->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
+		else
+			mempool_free(pmboxq, phba->mbox_mem_pool);
+		return;
+	}
+
+	memset(pmboxq, 0, sizeof(LPFC_MBOXQ_t));
+	pmb->mbxCommand = MBX_READ_LNK_STAT;
+	pmb->mbxOwner = OWN_HOST;
+	pmboxq->context1 = NULL;
+
+	if ((phba->fc_flag & FC_OFFLINE_MODE) ||
+	    (!(psli->sli_flag & LPFC_SLI2_ACTIVE)))
+		rc = lpfc_sli_issue_mbox(phba, pmboxq, MBX_POLL);
+	else
+		rc = lpfc_sli_issue_mbox_wait(phba, pmboxq, phba->fc_ratov * 2);
+
+	if (rc != MBX_SUCCESS) {
+		if (rc == MBX_TIMEOUT)
+			pmboxq->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
+		else
+			mempool_free( pmboxq, phba->mbox_mem_pool);
+		return;
+	}
+
+	lso->link_failure_count = pmb->un.varRdLnk.linkFailureCnt;
+	lso->loss_of_sync_count = pmb->un.varRdLnk.lossSyncCnt;
+	lso->loss_of_signal_count = pmb->un.varRdLnk.lossSignalCnt;
+	lso->prim_seq_protocol_err_count = pmb->un.varRdLnk.primSeqErrCnt;
+	lso->invalid_tx_word_count = pmb->un.varRdLnk.invalidXmitWord;
+	lso->invalid_crc_count = pmb->un.varRdLnk.crcCnt;
+	lso->error_frames = pmb->un.varRdLnk.crcCnt;
+	lso->link_events = (phba->fc_eventTag >> 1);
+
+	psli->stats_start = get_seconds();
+
+	return;
+}
 
 /*
  * The LPFC driver treats linkdown handling as target loss events so there
@@ -1431,8 +1514,7 @@ struct fc_function_template lpfc_transpo
 	 */
 
 	.get_fc_host_stats = lpfc_get_stats,
-
-	/* the LPFC driver doesn't support resetting stats yet */
+	.reset_fc_host_stats = lpfc_reset_stats,
 
 	.dd_fcrport_size = sizeof(struct lpfc_rport_data),
 	.show_rport_maxframe_size = 1,
diff -upNr a/drivers/scsi/lpfc/lpfc_sli.c b/drivers/scsi/lpfc/lpfc_sli.c
--- a/drivers/scsi/lpfc/lpfc_sli.c	2006-07-31 10:41:15.000000000 -0400
+++ b/drivers/scsi/lpfc/lpfc_sli.c	2006-08-02 15:14:11.000000000 -0400
@@ -1792,6 +1792,9 @@ lpfc_sli_brdrestart(struct lpfc_hba * ph
 
 	spin_unlock_irq(phba->host->host_lock);
 
+	memset(&psli->lnk_stat_offsets, 0, sizeof(psli->lnk_stat_offsets));
+	psli->stats_start = get_seconds();
+
 	if (skip_post)
 		mdelay(100);
 	else
diff -upNr a/drivers/scsi/lpfc/lpfc_sli.h b/drivers/scsi/lpfc/lpfc_sli.h
--- a/drivers/scsi/lpfc/lpfc_sli.h	2006-07-31 10:41:15.000000000 -0400
+++ b/drivers/scsi/lpfc/lpfc_sli.h	2006-08-02 15:13:42.000000000 -0400
@@ -172,6 +172,18 @@ struct lpfc_sli_stat {
 	uint32_t mbox_busy;	 /* Mailbox cmd busy */
 };
 
+/* Structure to store link status values when port stats are reset */
+struct lpfc_lnk_stat {
+	uint32_t link_failure_count;
+	uint32_t loss_of_sync_count;
+	uint32_t loss_of_signal_count;
+	uint32_t prim_seq_protocol_err_count;
+	uint32_t invalid_tx_word_count;
+	uint32_t invalid_crc_count;
+	uint32_t error_frames;
+	uint32_t link_events;
+};
+
 /* Structure used to hold SLI information */
 struct lpfc_sli {
 	uint32_t num_rings;
@@ -201,6 +213,8 @@ struct lpfc_sli {
 	struct lpfc_iocbq ** iocbq_lookup; /* array to lookup IOCB by IOTAG */
 	size_t iocbq_lookup_len;           /* current lengs of the array */
 	uint16_t  last_iotag;              /* last allocated IOTAG */
+	unsigned long  stats_start;        /* in seconds */
+	struct lpfc_lnk_stat lnk_stat_offsets;
 };
 
 /* Given a pointer to the start of the ring, and the slot number of

Re: [REPOST #3][PATCH 1/6] lpfc 8.1.7 : Adding statistics reset callback for FC transport

[Andreas Herrmann]

On 02.08.2006 21:24 James Smart <James.Smart@Emulex.Com> wrote:
> Sorry folks - a last code review identified one last reference
> to seconds_since_last_reset that didn't get caught.

Sorry -- for speaking up so lately.

> Adding statistics reset callback for FC transport

Now lpfc is the second FC driver that stores old stats data in
order to reset FC host stats. (zfcp is the other one.)

Shouldn't this stuff be moved into scsi_transport_fc?
So the default for scsi_transport_fc would be to store
a stats-offset (as you call it) together with seconds since
last reset. So every FC lldd gets the reset-stats functionality
for free.
Of course an lldd is able to register its own reset-callback
(maybe because the reset is implemented in hardware)
to replace the default handler.

This just came into my mind when reviewing your patch.


Regards,

Andreas

--
Linux for eServer Development
Tel :  +49-7031-16-4640
Notes mail :  Andreas Herrmann/GERMANY/IBM@IBMDE
email :  aherrman@de.ibm.com

Re: [REPOST #3][PATCH 1/6] lpfc 8.1.7 : Adding statistics reset callback for FC transport

[James Smart]

Andreas Herrmann wrote:
>> Adding statistics reset callback for FC transport
> 
> Now lpfc is the second FC driver that stores old stats data in
> order to reset FC host stats. (zfcp is the other one.)
> 
> Shouldn't this stuff be moved into scsi_transport_fc?
> So the default for scsi_transport_fc would be to store
> a stats-offset (as you call it) together with seconds since
> last reset. So every FC lldd gets the reset-stats functionality
> for free.
> Of course an lldd is able to register its own reset-callback
> (maybe because the reset is implemented in hardware)
> to replace the default handler.
> 
> This just came into my mind when reviewing your patch.

In looking at what could be different:

- the handling of the "seconds from last reset" is common and could/should
   go into the transport.

- I don't know that i'd change the stats data though...
   I assume you are wanting to move the stats data from a LLD struct to a
   transport struct: I shied away from this initially, as I allowed for
   the data structure to be "hot" - e.g. the LLD is updating values in the
   fast path. There are better cacheline attributes if you let the driver
   manage it's own structure. Thus the call from the transport to fill in
   a reported structure. Granted, for lpfc (at least) much of the LLD
   data is passive as the hardware does the "hot" work.

   Another reason - I shy away from the driver reaching directly into
   transport structures to change values, in order to avoid race conditions.
   I've tried to only do this on static data, so the access is only at
   creation.

I'll look into posting a patch for the "seconds from last reset".
However, I would approach it that for now, the lpfc patches go in as is.
Then, the patch would add the transport support and update the drivers
accordingly (lpfc & zfcp included).

-- james s