On Fri, Jan 06, 2017 at 11:37:11AM +0000, Andy Smith wrote: > Andrew Cooper suggested trying two patches: > > https://github.com/xenserver/linux-3.x.pg/blob/master/master/series#L613-L614
[…] > Using a kernel built with those patches the problem has gone away for > me and has been stable for about a month of production load. Sarah Newman pointed me at another message from David Vrabel: https://lists.xenproject.org/archives/html/xen-devel/2016-01/msg03033.html > > As a workaround, you can give dom0 more than 4 GiB and then the > > driver will set a 64-bit DMA mask. I don't think the memory needs > > to populated so something like dom0_mem=2GiB,max:4GiB might work. I've booted the stock jessie kernel with hypervisor command line containing "dom0_mem=2GiB,max:4GiB" and have not yet been able to induce a crash, whereas without this I can do so within seconds by triggering a full-speed md array check. Normally I just have "dom0_mem=2GiB". This has been running less than 24hrs but is looking good so far. Thanks, Andy