Re: 2.6.19-rc <-> ThinkPads, summary

[Hugh Dickins <hugh@veritas.com>]

On Sat, 4 Nov 2006, Adrian Bunk wrote:
> 
> Hugh Dickins:
> - ThinkPad T43p
> - booting with "noapic nolapic" didn't make any difference
> - reverting commit cf4c6a2f27f5db810b69dcb1da7f194489e8ff88
>   ("i386: Factor out common io apic routing entry access") didn't help
> - question: Was your bisecting successful?

Not quite what I'd call successful, but I think I've just about arrived
at some kind of conclusion.

Short summary: forget my complaint, assume it's fixed in current -git.

Boring version:

I think I have two slightly different, perhaps not unrelated, issues.
One manifested in habitual usage, to and from work, suspending for quiet
at home, etc, etc.  As 2.6.19-rc progressed, I more and more often found
it impossible to re-suspend after the first time: suspend key ignored.
rc3 seemed worst, rc4 at first seemed okay, then not, perhaps because...

In order to bisect on this, I had to speed up the testing from a day
or two to a few minutes; and I'm now thinking that this may have
focussed on a different problem.  After several reset bisections
converging on absurd patches (e.g. sparse annotations or unbuilt
sources), I grew even more suspicious of my "good" cases, and
yesterday found even 2.6.18 and 2.6.17 (didn't try earlier) behave
like this: occasionally the suspend key gets ignored for about one
minute (in the few cases I timed).

So whatever I was bisecting on, it's not a regression in 2.6.19.
It may be a software bug, it would be worth fixing if I can work
it out (though the pleasure of bisection was not having to think,
I've grown addicted); but it's not anything to hold up 2.6.19.

And as far as habitual usage goes, experience so far with -gits
post rc4 suggests that that problem has gone away: it's a little
too early to tell for sure, but I've not had to go back to using
2.6.18 to avoid it yet.

Hugh