[RFC PATCH 0/3] Add accurate boot timing to a Linux system

Hi Bjorn,

On Sun, Sep 25, 2011 at 5:54 AM, Bjorn Helgaas [off-list ref] wrote:

On Fri, Sep 23, 2011 at 5:03 PM, Simon Glass [off-list ref] wrote:

quoted

This experimental patch set adds boot timing to a Linux system. The
timing starts with the boot loader and extends through the kernel into
user space to the completion of the boot process. The timing starts when
the system leaves reset, not later when the kernel starts.

The concept is:
- Boot loader records a timestamp for key events during its operation
- These timestamps are passed to Linux, which adds more as it boots
- These timestamps are made available to user space, where more
timestamps are added as init does its job
- Finally the whole record is collected by a user-space script run at
the end of init. This is fed back through some mechanism to monitor
boot time in the field.

I think this is a cool idea. ?It's quite difficult to extract this
sort of information today, and making it easily and consistently
available should help focus attention and improve things.

There are difficult issues about which clock to use, how to correlate
bootloader & kernel timestamps, how to make sure the timestamps stay
sensible even when we use hwclock, ntp, etc., but I think it's worth
pushing on this to see how far you can go.

Yes it is not easy to get all of this right, especially trying to
cross two completely different projects. I think the timestamps should
be measured from reset. The kernel timestamps start not at zero, but
from the time the kernel starts. This guarantees a consistent time
base across the data. Also if there is a pre-boot program (such as a
NAND loader) then its time will be counted also.

A good timer would be an accurate monotonic wall clock, not
necessarily related to time of day. An ideal timer would be a
microsecond/nanosecond timer maintained by the hardware / SOC.

Regards,
Simon

Bjorn