Let's update the documentation that any signal is sufficient, and add a
comment that not only checking for fatal signals is historical baggage:
changing it now could break existing user space.  although unlikely.
For example, when an app provides a custom SIGALRM handler and triggers
memory offlining, the timeout cmd would no longer stop memory offlining,
because SIGALRM would no longer be considered a fatal signal.
Note that using signal_pending() instead of fatal_signal_pending() is
an anti-pattern, but slowly deprecating that behavior to eventually
change it in the far future is probably not worth the effort.  If this
ever becomes relevant for user-space, we might want to rethink.
Link: https://lkml.kernel.org/r/20230711174050.603820-1-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
 
 (-> BUG), memory offlining will keep retrying until it eventually succeeds.
 
 When offlining is triggered from user space, the offlining context can be
-terminated by sending a fatal signal. A timeout based offlining can easily be
+terminated by sending a signal. A timeout based offlining can easily be
 implemented via::
 
        % timeout $TIMEOUT offline_block | failure_handling
 
        do {
                pfn = start_pfn;
                do {
+                       /*
+                        * Historically we always checked for any signal and
+                        * can't limit it to fatal signals without eventually
+                        * breaking user space.
+                        */
                        if (signal_pending(current)) {
                                ret = -EINTR;
                                reason = "signal backoff";