On Apr,Monday 18 2011, at 10:26 AM, Mindaugas Rasiukevicius wrote: > "Cherry G. Mathew" <cherry.g.mat...@gmail.com> wrote: >>>>> On Mon, Apr 18, 2011 at 03:04:32AM +0000, Mindaugas Rasiukevicius >>>>> wrote: >>>>>> Module Name: src >>>>>> Committed By: rmind >>>>>> Date: Mon Apr 18 03:04:31 UTC 2011 >>>>>> >>>>>> Modified Files: >>>>>> src/sys/arch/xen/xen: balloon.c >>>>>> >>>>>> Log Message: >>>>>> balloon_xenbus_attach: use KM_SLEEP for allocation. >>>>>> >>>>>> Note: please do not use KM_NOSLEEP. >>>>> >>>>> And, according to yamt@, KM_SLEEP can fail in the current design... >>>> >>>> IIRC yamt@ fixed it a year or few ago. >>>> >>> >>> And in the more specific immediate context: >>> http://mail-index.netbsd.org/port-xen/2011/04/07/msg006613.html >>> >>> >> >> Hi, >> >> PS: Can you please revert ? Unless it's breaking anything else, or you >> have a fix for the problem mentioned in the thread above, ie; > > Change is for balloon_xenbus_attach(), which is irrelevant to the issue > described in that email. > > Are you aware that KM_NOSLEEP might fail not only because there is no > memory (or KVA) available, though?
I had several problems with memory hungry zfs, where allocation failed just because some lock in uvm was hold when KM_NOSLEEP allocation was issued. We should avoid of using KM_NOSLEEP whenever is it possible. Regards Adam.