What's the real story about KeQueryPerformanceCounter?

xxxxx@rahul.net wrote:
[…]

But nobody cares if it is zero absolute, or just close enough to zero to
be not significant.
[…]
So all that is required is to change the meaning of a previously reserved bit.

Agreed, in this context the cost is “zero”.
(With your definition that’s zero even considering account
documentation, application notes, version tracking, training, etc.)

The boot-time statement refers to this:

With any OS that has to run on *zillion configurations (like Windows
does), as soon as you want to use a feature that is not universally
present in each and every single one of these configurations, you have
to detect and specifically enable (or disable) support for it.

And with more and more features available, and also depending on the
complexity of the individual feature detection mechanism, this may add
considerably to the OS boot time. (As IMHO we can observe.)

The alternative is (of course) to use a kernel and application set
specifically linked for a specific hardware configuration.
Which removes any run-time cost, but adds configuration and probably
also support and administrative cost in a non-homogeneous environment.