Found this comment in a Broadcom source code that has an interesting
comment; in addition, danitool, testing different kernel command lines
found that forcing all the interrupts to CPUx the network throughput
was increased more than a 15% (as far i remember).
Any comment will be welcome.
Pepe
.depth = 1,
.lock = __SPIN_LOCK_UNLOCKED(irq_desc->lock),
#ifdef CONFIG_SMP
+#if defined(CONFIG_MIPS_BRCM)
+ /* We don't have an interrupt controller to distribute
interrupts, so
+ initially assign all interrupts to one cpu, then make explicit
adjustments if needed */
+ .affinity = CPU_MASK_CPU0
+#else
.affinity = CPU_MASK_ALL
#endif
+#endif
}
};
_______________________________________________
openwrt-devel mailing list
[email protected]
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel
