When running Cisco-MDS diagnostics which perform driver-level frame loop
back, the switch is reporting errors. Diagnostic has a limit on latency
that is not being met by the driver.
The requirement of Latency frames is that they should be responded back
by the host with a maximum delay of few hundreds of microseconds. If the
switch doesn't get response frames within this time frame, it fails the
test.
Test is failing as the lpfc-wq workqueue was overwhelmed by the packet
rate and in some cases, the work element yielded to other kernel elements.
To resolve, reduce the outstanding load allowed by the adapter. This
ensures the driver spends a reasonable amount of time doing loopback and
can do so such that latency values can be met. Load is managed by reducing
the number of receive buffers posted such that the link can be
backpressured to reduce load.
Signed-off-by: Dick Kennedy <dick.kennedy@xxxxxxxxxxxx>
Signed-off-by: James Smart <jsmart2021@xxxxxxxxx>
---
drivers/scsi/lpfc/lpfc_sli.c | 13 +++++++++++--
1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/scsi/lpfc/lpfc_sli.c b/drivers/scsi/lpfc/lpfc_sli.c
index 12319f63cb45..e660ee98ad8b 100644
--- a/drivers/scsi/lpfc/lpfc_sli.c
+++ b/drivers/scsi/lpfc/lpfc_sli.c
@@ -4918,8 +4918,17 @@ static int
lpfc_sli4_rb_setup(struct lpfc_hba *phba)
{
phba->hbq_in_use = 1;
- phba->hbqs[LPFC_ELS_HBQ].entry_count =
- lpfc_hbq_defs[LPFC_ELS_HBQ]->entry_count;
+ /**
+ * Specific case when the MDS diagnostics is enabled and supported.
+ * The receive buffer count is truncated to manage the incoming
+ * traffic.
+ **/
+ if (phba->cfg_enable_mds_diags && phba->mds_diags_support)
+ phba->hbqs[LPFC_ELS_HBQ].entry_count =
+ lpfc_hbq_defs[LPFC_ELS_HBQ]->entry_count >> 1;
+ else
+ phba->hbqs[LPFC_ELS_HBQ].entry_count =
+ lpfc_hbq_defs[LPFC_ELS_HBQ]->entry_count;
phba->hbq_count = 1;
lpfc_sli_hbqbuf_init_hbqs(phba, LPFC_ELS_HBQ);
/* Initially populate or replenish the HBQs */
--
2.13.7
