On Wed, Sep 04, 2019 at 11:51:33PM -0700, Christoph Hellwig wrote: > On Thu, Sep 05, 2019 at 08:57:16AM +1000, Dave Chinner wrote: > > > And unfortunately generic/530 still hangs for me with this series. > > > > Where does it hang? > > > > > This is an x86-64 VM with 4G of RAM and virtio-blk, default mkfs.xfs > > > options from current xfsprogs, 20G test and scratch fs. > > > > That's pretty much what one of my test rigs is, except iscsi luns > > rather than virtio-blk. I haven't been able to reproduce the issues, > > so I'm kinda flying blind w.r.t. to testing them here. Can you > > get a trace of what is happening (xfs_trans*, xfs_log*, xfs_ail* > > tracepoints) so I can have a deeper look? > > console output below, traces attached. Thanks, I'll have a look in a minute. I'm pretty sure I know what it will show - I got a trace from Chandan earlier this afternoon and the problem is log recovery doesn't yeild the cpu until it runs out of transaction reservation space, so the push work doesn't run because workqueue default behaviour is strict "run work only on the CPU it is queued on".... I've got an additional patch in testing right now... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
