Re: [PATCH] net/cpsw: don't disable_irqs() after an interrupt has been received.

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 04/16/2013 08:21 PM, Sebastian Andrzej Siewior wrote:
> In my testing I didn't notice anything unusual except one thing: I start
> a wget of a 126MiB file (which is sttored on MMC) from the beagle bone.
> On the first invocation I receive almost steady 5MiB/sec. In the
> following invocations of wget I see the performance floating between
> 1MiB/sec and 4MiB/sec. It usually ends with overall around 2MiB/sec.
> I couldn't notice nothing wrong. The only difference compared to the
> first invocation is (most likely) that the file is now served from
> memory and not read from MMC which should perform better but not worse.
> After executing "while ((1)); do /bin/true; done" on the beagle bone
> while the file was downloaded, I noticed that the speed rose to 9MiB/sec
> - 10/MiB/sec. Now this looks like power/idle optimization on the beagle
> bone side.

Yeah, disabling NO_HZ gives 11.2 MiB/s.

Sebastian