Re: linux-3.4 broken on chardonnay and huxelrebe (Re: [linux-3.4 test] 61301: regressions - FAIL)

[Ian Campbell <ian.campbell@citrix.com>]

On Thu, 2015-09-10 at 10:32 +0100, Ian Campbell wrote:
> The Chardonnay case suggests that either something has been backported into
> 3.4.x which has broken things (current real flights, which reliably fail,
> are running on 3.4.108) or that it is simply unreliable (or both). I think
> I need to repeat things a few times to confirm.

It turns out it was unreliable and these results were misleading. I setup
an adhoc job which simply installed Xen and rebooted 5 times and 3.4.x (for
x in increments of 10) failed reliably. In fact the fix wasn't until v3.7
-rc1 and some adhoc runs have fingered 65fe1f0f66a5 "ahci: implement
aggressive SATA device sleep support"[0]. I'm running a few more tests to
confirm but this looks reasonably certain.

That commit is a new feature, so it really shouldn't have the affect of
fixing bugs! I suspect this is something like a dodgy BIOS enabling the h/w
extension, which breaks until the kernel became aware of it and either
disables or explicitly copes with it being there.

Once the confirmation tests have run I will lock the machine and have a
poke around and see what I can see.

I could then take it to the Linux AHCI maintainer but I suspect that a
backport is only a slim possibility, as is someone taking the time to
determine which bit of this feature happened to fix these systems.

IOW I'm thinking that we should apply a minimum kernel version to
chardonnay as well as huxelrebe (once that feature exists).

Ian.

[0] git.kernel.org/torvalds/c/65fe1f0f66a57380229a4ced844188103135f37b