On Sun, Sep 05, 2010 at 07:03:24PM -0600, Bob Beck wrote: > > > > > > Basically by saying you will prevent swap is that you will make a > > situation > > > where you ask for a page of memory and don't get it because you are > > > suspending. > > > > Nothing the kernel needs to touch is swappable. by this point the > > suspend process only needs kernel stuff, userland has done running (but > > it still schedulnig until we disable interrupts). We don't swap device > > softcs etc, for obvious reasons. > > > > Ok, so there is one kernel thing that is swappable: pipe buffers, but > > they are not dependant for suspend (see above). > > > > As long as by the time we're fiddling with this we're ok for suspending > there is nothing > wrong with this. We have to be certain though. > > Failing that being a problem I see nothing fundamentally wrong with this, > but it will > need testing with swapkill, etc. I'd hate for it to introduce a bug which is > only found months later > when someone beats it up to try to work on something else.
I and a few others have beaten on this with swapkill. even when suspending a machine, i.e. run it deep into swap and do a supsend/resume this works fine and was totally expected as per what oga@ has already pointed out. So, I'm hunting for OKs now :)
