Messages in this thread | | | Date | Thu, 29 Oct 2009 16:48:38 +0900 | From | Jin Dongming <> | Subject | Re: [PATCH] [RFC][Patch x86-tip] add notifier before kdump |
| |
Hi Eric
Thanks for your comment. I will consider the method you suggested. Thank you very much.
Best regards, Jin Dongming
Eric W. Biederman wrote: > Jin Dongming <jin.dongming@np.css.fujitsu.com> writes: > >> Vivek, Lon Hohberger >> >> Thanks for your comments. >> >> I also agree with your opinion, too. But I still have problems listed >> as following. >> 1. Nobody knows when the second kernel does not work well >> 2. Too much time cost to startup the second kernel >> >> So I hope that following work could be done. >> 1. Add some code before kdump to monitor the actions of second kernel >> Something we worried about is that nobody knows when the kdump will not >> work well. If the second kernel does not work well, nobody knows when >> the best time is to restart. So we need to add some code such as setting >> watchdog. If we want to monitor second kernel, some work need to be done >> before it starts to work. > > Any extra work done in the crashing kernel decreases the likely hood > of the second kernel working. If you want a watchdog I recommend you always > run it and have the interval set large enough for the second kernel to boot > before it starts petting it. That way no code needs to run after a kernel crash. > >> 2. Shorten the startup time of second kernel >> I think that the purpose of second kernel is used for backup information >> stored in memory to storage only. But the time cost is different >> according to the system architecture. And also I think that there are >> too many of device is not useful to get the memory information to >> storage. I think the startup time of second kernel should be shortened. >> I don't know much about second kernel, these are only my thought. > > The second kernel can be anything you like. Including a standalone executable > if needed for very precise requirements. Any improvements in boot time for a > normal linux image should apply to the dump kernel. > > Currently with a tight initrd and just dumping dmesg to disk I get away with > a 20M crashkernel reservation, and it typically runs and does it's work before > my watchdog fires. > > Eric > -- > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > Please read the FAQ at http://www.tux.org/lkml/ > >
| |