On 10/31/2014 4:40 AM, Thomas Gleixner wrote:
On Fri, 31 Oct 2014, suravee.suthikulpanit@xxxxxxx wrote:
+/*
+ * alloc_msi_irq - Allocate MSIs from available MSI bitmap.
+ * @data: Pointer to v2m_data
+ * @nvec: Number of interrupts to allocate
+ * @irq: Pointer to the allocated irq
+ *
+ * Allocates interrupts only if the contiguous range of MSIs
+ * with specified nvec are available. Otherwise return the number
+ * of available interrupts. If none are available, then returns -ENOENT.
And the exact purpose of returning the number of available interrupts
is?
Initially, I intended to use this function to allocate irqs for both MSI
and multi-MSI case. But I'll simplify this and revisit it again when
adding the multi-MSI.
+ virq = __irq_domain_alloc_irqs(v2m->domain, hwirq,
+ 1, NUMA_NO_NODE, v2m, true);
And surely the ability of alloc_msi_irq() to allocate a contiguous
vector space is required to satisfy an hardcoded allocation of ONE
interrupt.
What is guaranteeing that the caller only requests a single interrupt?
Since this is calling from gicv2m_setup_msi_irq(), it should only setup
1 MSI interrupt.
>[...]
+err_out:
Single error exit which undoes the stuff in the same order it got
initialized is just plain wrong. Ever looked at proper error exits in
other kernel files?
I'll clean this up in V10. Thanks for pointing this out.
Suravee
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
