On Thu, Oct 19, 2017 at 10:18:28AM +0200, Manuel Bouyer wrote: > the key factor. From what I've seen it's virtual memory which is fragmented > too much, not physical memory. There is plenty of free space in the vmem, but > no chunk is big enough (from memory, hung processes are waiting for a 256kb > contigous chunk of VM space).
Yes, of course, virtual continous kernel VA is the issue, sorry, should have been more carefull with words. Are that xen guests 32bit? The exec args take that much and we have never implemented support for them being non-continous (IIRC, and not sure what exactly would be needed for that, I guess something in the pool code to add multiple mappings for a single allocation?). Martin
