Hi,
在 2023/01/09 9:38, Yu Kuai 写道:
So, it's not sysfs_lock but sysfs file deletion. When a kernfs, which
backs
sysfs, file is removed, it disables future operations and drains all
inflight ones before returning, so you remove the interface files before
cleaning up the object that it interacts with, you don't have to worry
about
racing against file operations as none can be in flight at that point.
I understand this know, kernfs_fop_write_iter() will grab
kernfs_node->active, and kobject_del() will wait for active to be
dropped in kernfs_drain().
Sorry that I've been asking you to go round and round but block device
add/remove paths have always been really tricky and we wanna avoid adding
more complications if at all possible. Can you see why the device is
being
destroyed before the queue attr is removed?
Sorry that I actually tested with patch 4 applied, and this is a bug
introduced by patch 4, my apologies. It set rqos to ERR_PTR() in
rq_qos_exit, and follow up rq_qos_issue() just check if rqos is NULL.
I'll wait for your patchset to be apllied, and then send a new version.
Just one thing to confirm, do you think it's better to use a global
mutex rather than a disk level mutex? I'm not sure because this will
cause different cgroup configurations from different disk can't
concurrent.
Thanks,
Kuai
