aio: account I/O wait time properly
authorJeff Moyer <jmoyer@redhat.com>
Wed, 17 Oct 2007 06:27:20 +0000 (23:27 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Wed, 17 Oct 2007 15:42:53 +0000 (08:42 -0700)
Some months back I proposed changing the schedule() call in
read_events to an io_schedule():
http://osdir.com/ml/linux.kernel.aio.general/2006-10/msg00024.html
This was rejected as there are AIO operations that do not initiate
disk I/O.  I've had another look at the problem, and the only AIO
operation that will not initiate disk I/O is IOCB_CMD_NOOP.  However,
this command isn't even wired up!

Given that it doesn't work, and hasn't for *years*, I'm going to
suggest again that we do proper I/O accounting when using AIO.

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Acked-by: Zach Brown <zach.brown@oracle.com>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/aio.c

index ea2e198203818a4703f8dc56a6d8473cf6ada21a..d02f43b50a3d83e9fb6e20c8c64bb83df05eb1a3 100644 (file)
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -303,7 +303,7 @@ static void wait_for_all_aios(struct kioctx *ctx)
        set_task_state(tsk, TASK_UNINTERRUPTIBLE);
        while (ctx->reqs_active) {
                spin_unlock_irq(&ctx->ctx_lock);
-               schedule();
+               io_schedule();
                set_task_state(tsk, TASK_UNINTERRUPTIBLE);
                spin_lock_irq(&ctx->ctx_lock);
        }
@@ -323,7 +323,7 @@ ssize_t fastcall wait_on_sync_kiocb(struct kiocb *iocb)
                set_current_state(TASK_UNINTERRUPTIBLE);
                if (!iocb->ki_users)
                        break;
-               schedule();
+               io_schedule();
        }
        __set_current_state(TASK_RUNNING);
        return iocb->ki_user_data;
@@ -1170,7 +1170,7 @@ retry:
                        ret = 0;
                        if (to.timed_out)       /* Only check after read evt */
                                break;
-                       schedule();
+                       io_schedule();
                        if (signal_pending(tsk)) {
                                ret = -EINTR;
                                break;