Skip to contentSkip to conversation

powerpc/mpic: improve interrupt handling performance

PATCH In review Kumar Gala 2 participants 2 messages May 8, 2009 – May 11, 2009 PowerPC

lore
Share & tools
lore

Kumar Gala

author
#0 lore raw

PATCH 1/1 powerpc/mpic: improve interrupt handling performance

Before when we were setting up the irq host map for mpic we passed in just isu_size for the size of the linear map. However, for a number of mpic implementations we have no isu (thus pass in 0) and will end up with a no linear map (size = 0). This causes us to always call irq_find_mapping() from mpic_get_irq().

By moving the allocation of the host map to after we've determined the number of sources we can actually benefit from having a linear map for the non-isu users that covers all the interrupt sources.

---
Ben,

I leave it up to you if you want to put this in for 2.6.30. We see about a 1us reduction in time spent in mpic_get_irq() based on ftrace.

- k

 arch/powerpc/sysdev/mpic.c |   16 +++++++++-------
 1 files changed, 9 insertions(+), 7 deletions(-)
arch/powerpc/sysdev/mpic.c +9 -7
index 21b9567..0efc12d 100644
@@ -1057,13 +1057,6 @@ struct mpic * __init mpic_alloc(struct device_node *node,
 	memset(mpic, 0, sizeof(struct mpic));
 	mpic->name = name;
 
-	mpic->irqhost = irq_alloc_host(node, IRQ_HOST_MAP_LINEAR,
-				       isu_size, &mpic_host_ops,
-				       flags & MPIC_LARGE_VECTORS ? 2048 : 256);
-	if (mpic->irqhost == NULL)
-		return NULL;
-
-	mpic->irqhost->host_data = mpic;
 	mpic->hc_irq = mpic_irq_chip;
 	mpic->hc_irq.typename = name;
 	if (flags & MPIC_PRIMARY)
@@ -1213,6 +1206,15 @@ struct mpic * __init mpic_alloc(struct device_node *node,
 	mpic->isu_shift = 1 + __ilog2(mpic->isu_size - 1);
 	mpic->isu_mask = (1 << mpic->isu_shift) - 1;
 
+	mpic->irqhost = irq_alloc_host(node, IRQ_HOST_MAP_LINEAR,
+				       isu_size ? isu_size : mpic->num_sources,
+				       &mpic_host_ops,
+				       flags & MPIC_LARGE_VECTORS ? 2048 : 256);
+	if (mpic->irqhost == NULL)
+		return NULL;
+
+	mpic->irqhost->host_data = mpic;
+
 	/* Display version */
 	switch (greg_feature & MPIC_GREG_FEATURE_VERSION_MASK) {
 	case 1:
Signature · 1 line
1.6.0.6
#1 lore raw
Quoted 9 lines · Kumar Gala

Before when we were setting up the irq host map for mpic we passed in just isu_size for the size of the linear map. However, for a number of mpic implementations we have no isu (thus pass in 0) and will end up with a no linear map (size = 0). This causes us to always call irq_find_mapping() from mpic_get_irq().

By moving the allocation of the host map to after we've determined the number of sources we can actually benefit from having a linear map for the non-isu users that covers all the interrupt sources.

That's a nasty bug, so I think we should do:

arch/powerpc/kernel/irq.c +1 -0
index 7d46e5d..2b958d6 100644
@@ -537,6 +537,7 @@ struct irq_host *irq_alloc_host(struct device_node *of_node,
                }
                break;
        case IRQ_HOST_MAP_LINEAR:
+               WARN_ON(revmap_arg == 0);
                rmap = (unsigned int *)(host + 1);
                for (i = 0; i < revmap_arg; i++)
                        rmap[i] = NO_IRQ;

To save anyone else the embarrassment :)

cheers