On Fri, 2007-03-09 at 00:15 +0100, Pavel Machek wrote:

That's a no-no. ATOMIC alocations can fail, and no, WARN_ON is not
enough. It is not a bug, they just fail.

But like I said in my post, there's no way we can disable suspend to
disk when they do, right now anyway. Also, this can't be called any
later than a late initcall or such since it's __init, and thus there
shouldn't be memory pressure yet that would cause this to fail.

In any case, I'd be much happier with having a "disable suspend"
variable so we could print a big warning and set that flag.

johannes