OSDN Git Service

staging: lustre: simplify waiting in ldlm_completion_ast()
authorNeilBrown <neilb@suse.com>
Tue, 13 Feb 2018 20:17:30 +0000 (07:17 +1100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 16 Feb 2018 14:19:10 +0000 (15:19 +0100)
commite3382addbc0a5af8640db4f1f3b46ff22ad74400
treee6e64a49963eda3369a652d5cfd79aace53390cf
parentd7ce88fbb5dbc2263eb3a6a0c9888c17d65ffa97
staging: lustre: simplify waiting in ldlm_completion_ast()

If a signal-callback (lwi_on_signal) is set without lwi_allow_intr, as
is the case in ldlm_completion_ast(), the behavior depends on the
timeout set.

If a timeout is set, then signals are ignored.  If the timeout is
reached, the timeout handler is called.  If the timeout handler
return 0, which ldlm_expired_completion_wait() always does, the
l_wait_event() switches to exactly the behavior if no timeout was set.

If no timeout is set, then "fatal" signals are not ignored.  If one
arrives the callback is run, but as the callback is empty in this
case, that is not relevant.

This can be simplified to:
 if a timeout is wanted
     wait_event_idle_timeout()
     if that timed out, call the timeout handler
 l_wait_event_abortable()

i.e. the code always waits indefinitely.  Sometimes it performs a
non-abortable wait first.  Sometimes it doesn't.  But it only
aborts before the condition is true if it is signaled.
This doesn't quite agree with the comments and debug messages.

Now that we call the timeout handler (ldlm_expired_completion_wait())
wait directly, we can pass the two args directly rather then
using a special-purpose struct.

Reviewed-by: Patrick Farrell <paf@cray.com>
Reviewed-by: James Simmons <jsimmons@infradead.org>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lustre/ldlm/ldlm_request.c