In the ioctx destroy path, both exit_aio and io_destroy calls
wait_for_all_aios(). And in that function, it won't return until
there are no outstanding kiocb, tracked by ctx->reqs_active. So
the ref counting on kioctx for every individual kiocb is overly
excessive. We know we won't perform last put_ioctx when releasing
Kiocb. This should clear out the cache line conflict mentioned in
earlier post.
Signed-off-by: Ken Chen <[email protected]>
diff -Nurp linux-2.6.18/fs/aio.c linux-2.6.18.ken/fs/aio.c
--- linux-2.6.18/fs/aio.c 2006-09-19 20:42:06.000000000 -0700
+++ linux-2.6.18.ken/fs/aio.c 2006-10-12 13:33:09.000000000 -0700
@@ -423,7 +422,6 @@ static struct kiocb fastcall *__aio_get_
ring = kmap_atomic(ctx->ring_info.ring_pages[0], KM_USER0);
if (ctx->reqs_active < aio_ring_avail(&ctx->ring_info, ring)) {
list_add(&req->ki_list, &ctx->active_reqs);
- get_ioctx(ctx);
ctx->reqs_active++;
okay = 1;
}
@@ -534,8 +532,6 @@ int fastcall aio_put_req(struct kiocb *r
spin_lock_irq(&ctx->ctx_lock);
ret = __aio_put_req(ctx, req);
spin_unlock_irq(&ctx->ctx_lock);
- if (ret)
- put_ioctx(ctx);
return ret;
}
@@ -791,8 +787,7 @@ static int __aio_run_iocbs(struct kioctx
*/
iocb->ki_users++; /* grab extra reference */
aio_run_iocb(iocb);
- if (__aio_put_req(ctx, iocb)) /* drop extra ref */
- put_ioctx(ctx);
+ __aio_put_req(ctx, iocb);
}
if (!list_empty(&ctx->run_list))
return 1;
@@ -1015,9 +1010,6 @@ put_rq:
if (waitqueue_active(&ctx->wait))
wake_up(&ctx->wait);
- if (ret)
- put_ioctx(ctx);
-
return ret;
}
On Thu, Oct 12, 2006 at 04:54:46PM -0700, Chen, Kenneth W wrote:
> @@ -1015,9 +1010,6 @@ put_rq:
> if (waitqueue_active(&ctx->wait))
> wake_up(&ctx->wait);
>
> - if (ret)
> - put_ioctx(ctx);
> -
> return ret;
> }
This part makes me worry -- at this point we no longer have anything
pinning the ioctx, yet we touch ->wait after dropping the lock. The
only way around this is if rcu is introduced in the final free of an
ioctx to ensure the structure remains around for a sufficient grace
period.
-ben
--
"Time is of no importance, Mr. President, only life is important."
Don't Email: <[email protected]>.