Attachment 230321 Details for Bug 250266 – Patch to allow selecting IRQF_NODELAY over IRQF_SHARED for interrupt

Patch to allow selecting IRQF_NODELAY over IRQF_SHARED for interrupt

megasas_IRQF_NODELAY.patch (text/x-patch), 1.45 KB, created by Clark Williams on 2007-10-17 19:34:40 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Clark Williams

Created: 2007-10-17 19:34:40 UTC

Size: 1.45 KB

patch

obsolete

>--- vanilla/drivers/scsi/megaraid/megaraid_sas.c	2007-04-25 22:08:32.000000000 -0500
>+++ linux-2.6.21.x86_64/drivers/scsi/megaraid/megaraid_sas.c	2007-09-19 13:59:30.000000000 -0500
>@@ -51,6 +51,11 @@
> MODULE_AUTHOR("megaraidlinux@lsi.com");
> MODULE_DESCRIPTION("LSI Logic MegaRAID SAS Driver");
> 
>+static int megasas_intr_nodelay = 0;
>+module_param_named(nodelay, megasas_intr_nodelay, int, 0);
>+MODULE_PARM_DESC(nodelay,
>+	"Set to 1 to change interrupt type to NODELAY (default=SHARED)");
>+
> /*
>  * PCI ID table for all supported controllers
>  */
>@@ -2408,7 +2413,20 @@
> 	/*
> 	 * Register IRQ
> 	 */
>-	if (request_irq(pdev->irq, megasas_isr, IRQF_SHARED, "megasas", instance)) {
>+	/*
>+	 * Use of IRQF_NODELAY in CONFIG_PREEMPT_RT is know to fix a timeout
>+	 * problem that ultimately makes the driver consider the adapter dead,
>+	 * setting instance->hw_crit_error.  It is safe because megasas_isr
>+	 * doesn't take any locks, just using tasklet_schedule to defer the
>+	 * interrupt handling to softirq time. This is indeed a big hammer
>+	 * "solution", but at least systems with this hardware will work as
>+	 * expected till we find the real fix.
>+	 */
>+	printk("megasas: requesting irq %d as %s interrupt\n",
>+		pdev->irq, megasas_intr_nodelay ? "NODELAY" : "SHARED");
>+	if (request_irq(pdev->irq, megasas_isr,
>+			megasas_intr_nodelay ? 	IRQF_NODELAY : 	IRQF_SHARED,
>+			"megasas", instance)) {
> 		printk(KERN_DEBUG "megasas: Failed to register IRQ\n");
> 		goto fail_irq;
> 	}

Actions: View

Attachments on bug 250266: 160330 | 161776 | 204301 | 230321