On Tue, Apr 15, 2014 at 10:59:25AM -0400, Brian Foster wrote: > Yeah, I noticed that the completion wouldn't fire as is since it appears > that doesn't occur if a b_iodone call is provided. FWIW, the thought > process was more to inherit the use of b_iowait in xlog_iodone(). That doesn't mean the b_iodone callback couldn't do the wakeup on b_iowait. I have to say it makes me a bit uneasy that we had to find this race the hard way for the superblock, and now for the logbufs again. Seems like all other uncached buffers are only read and written synchronously, so for this should be the last issue, but I fear about new ones showing up in the future and would prefer a more general solution. Reluctantly: Reviewed-by: Christoph Hellwig <hch@xxxxxx> _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
