A development blog of what Con Kolivas is doing with code at the moment with the emphasis on linux kernel, MuQSS, BFS and -ck.
Tuesday, 26 November 2013
Experimental hibernation patch for BFS 443
In response to the numerous reports of problems with hibernate AKA suspend to disk, here is a purely experimental patch to apply to bfs443/3.12-ck1 to attempt to address the problem.
I was able to boot :-) Hibernate & resuming works "well", in the same way as without this patch (as written in the previous blog-thread): Serial mouse missing after hibernation and is recoverable after a suspend-to-ram + a following `setserial -a /dev/ttyS0`.
Sorry Manuel the serial mouse problem is a mainline one out of my control. Thanks for testing though. I'm particularly interested in those who can't successfully hibernate and resume.
Yes, I know, that you'd be right with your point.I only wanted to report back that your patch at least didn't cause problems/regressions on here. Manuel
No, both are not working. I just found the mentioned workaround for suspend to ram. But this doesn't work for resuming from hibernation because after loading the image from disk the cpu offline stuff is called.
Thanks, I'll try and put something up for your specific issue (which is odd). I still need someone who has suspend to ram working but not suspend to disk working to try this patch and see if it achieves anything at all.
I get this on all three of my machines with bfs. 1 sandybridge laptop (2 core + 2 ht) 2(4 core + 4ht) ivy bridge desktops. Maybe the problem is Intel related but i'm just guessing...
is it possible for you to post some prefered or must have config settings for the kernel in conjunction with BFS? As I remember, there were always some problems with CONFIG_HZ not set to 1000. And the BOOTPARAM_HOTPLUG_CPU0 should be NO. No dynticks with HIGH_RES_TIMERS too. And you mentioned, that NUMA would be better off. Maybe the problems could be better identified than.
Hotplugcpu0 should not be enable-able but the others should only be recommendations. They shouldn't lead to failures. Nowadays I use 1000Hz with dynticks. Numa is only relevant if you're on numa hardware, and then you should enable it.
Unfortunately, this patch doesn't fix hibernation issue for me. The symptom is the same — the system hibernates OK, but it fails to bring everything up after reading image while resuming.
Will I be able to boot?
ReplyDeleteI was able to boot :-) Hibernate & resuming works "well", in the same way as without this patch (as written in the previous blog-thread): Serial mouse missing after hibernation and is recoverable after a suspend-to-ram + a following `setserial -a /dev/ttyS0`.
ReplyDeleteBest regards, Manuel Krause
Sorry Manuel the serial mouse problem is a mainline one out of my control. Thanks for testing though. I'm particularly interested in those who can't successfully hibernate and resume.
DeleteYes, I know, that you'd be right with your point.I only wanted to report back that your patch at least didn't cause problems/regressions on here.
DeleteManuel
Meanwhile I managed to post this serial mouse issue to the kernel list. Let's see what's made of it. :-?
DeleteThank you for YOUR work,
Manuel
No luck for me but suspend to ram is also not working without my workaround (ps -eo pid | xargs -I'{}' taskset -pc 0 {} before suspending).
ReplyDeletebacktraces: http://imgur.com/bgJBn7l,tt521gl#0
Oh? I didn't know that suspend to ram wasn't working for you, I thought it was just to disk?
DeleteNo, both are not working. I just found the mentioned workaround for suspend to ram. But this doesn't work for resuming from hibernation because after loading the image from disk the cpu offline stuff is called.
DeleteThanks, I'll try and put something up for your specific issue (which is odd). I still need someone who has suspend to ram working but not suspend to disk working to try this patch and see if it achieves anything at all.
DeleteI get this on all three of my machines with bfs.
ReplyDelete1 sandybridge laptop (2 core + 2 ht) 2(4 core + 4ht) ivy bridge desktops.
Maybe the problem is Intel related but i'm just guessing...
Strange. Maybe a configuration issue cos my i7 with 12 threads doesn't have a problem.
DeleteWhat is your CONFIG_HZ ?
DeleteIf I use anything lower than 1000 it breaks all the time.
With 1000 it works mostly but breaks randomly on resume.
Hi con,
Deleteis it possible for you to post some prefered or must have config settings for the kernel in conjunction with BFS? As I remember, there were always some problems with CONFIG_HZ not set to 1000. And the BOOTPARAM_HOTPLUG_CPU0 should be NO. No dynticks with HIGH_RES_TIMERS too. And you mentioned, that NUMA would be better off. Maybe the problems could be better identified than.
cu sysitos
Hotplugcpu0 should not be enable-able but the others should only be recommendations. They shouldn't lead to failures. Nowadays I use 1000Hz with dynticks. Numa is only relevant if you're on numa hardware, and then you should enable it.
DeleteMike's question is good.
DeleteWould you also recommend to set CONFIG_HZ_PERIODIC to YES instead of NO_HZ_IDLE (Idle dynticks system for energy saving)?
Best regards, Manuel
Unfortunately, this patch doesn't fix hibernation issue for me. The symptom is the same — the system hibernates OK, but it fails to bring everything up after reading image while resuming.
ReplyDeleteJust in case, my CONFIG_HZ is 250.
DeleteThanks for testing. In which case this confirms it is not a fix.
Delete