During the analysis performed at

        LTP: madvise11 sometimes fails with madvise(..., MADV_SOFT_OFFLINE) returning -EBUSY

        https://issues.redhat.com/browse/RHEL-9292

Rafael Aquini and I noticed that the man-pages for madvise(2) do not mention the possibility of madvise() returning EBUSY as the result of a soft offlining (MADV_SOFT_OFFLINE) request. That is the case upstream, c9s and RHEL9.x and RHEL-8.x kernels that support MADV_SOFT_OFFLINE request for madvise().

Our man-pages package comes from http://www.kernel.org/doc/man-pages/. So the solution should first be posted and accepted there and then backported to the required RHEL streams.

The details for the cases where -EBUSY can be returned are described in RHEL-9292.

Also, there is an associated issue for the LTP test madvise11.c that fails because the test does not handle -EBUSY in the right way: https://issues.redhat.com/browse/RHEL-34106