Re: Any way to do time keeping during kernel uncompression[ARM architecture]

[sandeep kumar <coolsandyforyou@xxxxxxxxx>]

Dear Abu

On our board we have a dedicated register to measure the ticks, right from the "POWER key press".

This register is incremented 32567 times every second.

We directly read the register with physical address(MMU is not turned on bootloader, during uncompress_kernel MMU& is on but still one on one mapping) values to know the current timing from boot up.

I profiled at the places where zImage copying happens & decompress_kernel functions(misc.c) & "restart:" label in head.S. and got the above results.

Regards

Sandeep.

On Thu, Feb 7, 2013 at 1:52 AM, Abu Rasheda <rcpilot2010@xxxxxxxxx> wrote:

On Wed, Feb 6, 2013 at 3:53 AM, sandeep kumar <coolsandyforyou@xxxxxxxxx> wrote:

Dear All

FYI:

I got the below data in my board...

For Zimage:

Copy from eMMC to RAM(bootloader) & other stuff in bootloader -- 200msec

Kernel relocation --- 1msec

Kernel uncompression -- 700msec

Total time taken --- 900msec approx

For Image:

Copy from eMMC to RAM(bootloader) & other stuff in bootloader -- 1400msec

So in case of Uncompressed kernel image.. even though it doesnt need kernel uncompression, it is taking 500msec more in bootloader. 

Thanks

Sandeep

How did you measure these times ? 

On Thu, Jan 31, 2013 at 12:46 PM, Dave Hylands <dhylands@xxxxxxxxx> wrote:

Hi Sandeep

On Wed, Jan 30, 2013 at 6:35 PM, sandeep kumar <coolsandyforyou@xxxxxxxxx> wrote:
>
> Hi Dave.
> Ok
> Can you tell me which thing runs faster ?
> 1)Copy zImage(compressed kernel) from flash to Ram, jump to start, Relocate compress kernel, uncompress it...
> 2)Copy Image(uncompressed one) from flash to RAM and run it right away...
>
> My gut feeling says second one is better, thats what i m trying to prove here... :)

It depends on numerous factors:

1 - Your flash speed

2 - Your RAM speed

3 - Your CPU speed

4 - Whether you have code and/or data caches enabled

If you enable the caches, use DMA from NAND (so make your NAND as fast as possible), then option 1 usually blows away option 2, but your mileage may vary.

Dave Hylands

--
With regards,
Sandeep Kumar Anantapalli,

_______________________________________________
Kernelnewbies mailing list
Kernelnewbies@xxxxxxxxxxxxxxxxx
http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies

--
With regards,
Sandeep Kumar Anantapalli,

_______________________________________________
Kernelnewbies mailing list
Kernelnewbies@xxxxxxxxxxxxxxxxx
http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies