Re: [RFC 1/2] mmc: davinci: Eliminate spurious interrupts

[Ido Yariv <ido@wizery.com>]

Hello Sudhakar,

On Fri, Jan 27, 2012 at 08:11:55AM +0000, Rajashekhara, Sudhakar wrote:
> I tested both these patches and they work fine on the OMAP-L138 EVM. I
> observed that with these patches the number of interrupts during a transfer
> are less compared to earlier. For a 100 MB transfer, I could see around
> 700-800 interrupts less. Did you see any performance improvement with these
> patches?

Thanks for testing these.

The difference in the number of interrupts, or rather the ratio, depends
on the scenario you're working on.
The spurious patch has an effect on non-dma transfers, while the polling
optimization affects small (i.e. up to 128 bytes) transactions.
I expect these to have a greater impact on SDIO scenarios than on SD scenarios.

Regarding the performance improvement, these patches were tested using a
WLAN SDIO adapter, which requires quite a few small transactions.
The polling optimization had an impact of around 10-15% in throughput.
The spurious patch did not have any noticeable effect on throughput, but did
lower the number of interrupts by up to 25% in some cases.

Thanks,
Ido.