Messages in this thread | | | Date | Wed, 26 Oct 2011 12:01:01 -0400 | From | Vivek Goyal <> | Subject | Re: [PATCH 04/13] block: make ioc get/put interface more conventional and fix race on alloction |
| |
On Tue, Oct 25, 2011 at 06:48:30PM -0700, Tejun Heo wrote:
[..] > @@ -1645,11 +1645,12 @@ static void blkiocg_attach_task(struct cgroup *cgrp, struct task_struct *tsk) > { > struct io_context *ioc; > > - task_lock(tsk); > - ioc = tsk->io_context; > - if (ioc) > + /* we don't lose anything even if ioc allocation fails */ > + ioc = get_task_io_context(tsk, GFP_ATOMIC, NUMA_NO_NODE); > + if (ioc) { > ioc->cgroup_changed = 1; > - task_unlock(tsk); > + put_io_context(ioc); > + }
We really don't need to create a new io context here if it is not present. cgroup_changed is set so that reference to old sync cfqq is dropped and a new one is assigned when new IO comes in. Given the fact that io context is not even present for the task being attached, there is no way there is associated cic and associated cfqq. So no need to set changed_cgroup=1.
Will it make sense to pass in additional parameter to get_task_io_context() so that it takes a reference if ioc exists otherwise returns NULL. Or may be another variant of get_task_io_context() function.
This is a minor issue though. Will just avoid creating iocontext memory for task if one is not using CFQ in the system.
[..] > > +extern char __blk_test_mode[2]; > +#define blk_test_mode (__blk_test_mode[0]) > +
Looks like leftover of your testing code.
Thanks Vivek
| |