On Tue, Sep 17, 2024 at 1:53 PM Christoph Hellwig <hch@xxxxxx> wrote:
>
> On Tue, Sep 17, 2024 at 02:32:58PM +0900, Damien Le Moal wrote:
> > Commit 734e1a860312 ("block: Prevent deadlocks when switching
> > elevators") introduced the function elv_iosched_load_module() to allow
> > loading an elevator module outside of elv_iosched_store() with the
> > target device queue not frozen, to avoid deadlocks. However, the "none"
> > scheduler does not have a module and as a result,
> > elv_iosched_load_module() always returns an error when trying to switch
> > to this valid scheduler.
> >
> > Fix this by checking that the requested scheduler is "none" and doing
> > nothing in that case.
>
> The old code before this commit simply ignored the request_module,
> just as most callers of it do. I think that's the right approach
> here as well.
freeze queue is actually easy to cause deadlock, and old code is to not
do it everywhere.
Probably it may be more reliable to replace 'load_module' with 'no_freeze',
and not to freeze queue in case that 'no_freeze' is set in queue_attr_store().
queue_wb_lat_store() need no_freeze too since there is GFP_KERNEL
allocation involved.
Thanks,
