Interrupts are being dropped under heavy IO on -rt .

From: jvrao
Date: Fri Apr 10 2009 - 12:20:50 EST


Qlogic driver uses MSI-EDGE interrupts by default.

On -rt kernel under heavy IO load we observed that the driver missed interrupts occasionally.

Missing interrupts caused spurious mailbox timeouts and path failovers.

Suspecting MSI interrupts, we disabled MSI and forcing the driver to use APIC interrupts (Patch attached).

With this change, our IO tests ran for extended period of time without any issues.

It would be interesting to see if the MSI infrastructure itself has issues on realtime, or it is the combination of qla2xxx/MSI.

Right now we don't have any setup to test out MSI with other adapters. Hence, wondering if anyone faced similar issues.

Thanks,
JV

Index: linux-2.6.29.1/drivers/scsi/qla2xxx/qla_isr.c
===================================================================
--- linux-2.6.29.1.orig/drivers/scsi/qla2xxx/qla_isr.c
+++ linux-2.6.29.1/drivers/scsi/qla2xxx/qla_isr.c
@@ -1983,7 +1983,13 @@ qla2x00_request_irqs(struct qla_hw_data
/* If possible, enable MSI-X. */
if (!IS_QLA2432(ha) && !IS_QLA2532(ha) &&
!IS_QLA8432(ha) && !IS_QLA8001(ha))
+#ifdef CONFIG_PREEMPT_RT
+ /* Observed IO timeouts with MSI iterrupts on -rt kernel. */
+ /* This code change forces the driver to use APIC interrupts. */
+ goto skip_msi;
+#else
goto skip_msix;
+#endif /* CONFIG_PREEMPT_RT */

if (IS_QLA2432(ha) && (ha->pdev->revision < QLA_MSIX_CHIP_REV_24XX ||
!QLA_MSIX_FW_MODE_1(ha->fw_attributes))) {
Index: linux-2.6.29.1/drivers/scsi/qla2xxx/qla_version.h
===================================================================
--- linux-2.6.29.1.orig/drivers/scsi/qla2xxx/qla_version.h
+++ linux-2.6.29.1/drivers/scsi/qla2xxx/qla_version.h
@@ -7,7 +7,7 @@
/*
* Driver version
*/
-#define QLA2XXX_VERSION "8.03.00-k4"
+#define QLA2XXX_VERSION "8.03.00-k4-rt1"

#define QLA_DRIVER_MAJOR_VER 8
#define QLA_DRIVER_MINOR_VER 3