Re: [PATCH v2] mmc: core: Remove redundant rescan_disable flag

[Adrian Hunter <adrian.hunter@intel.com>]

On 10/05/16 12:01, Ulf Hansson wrote:
> On 29 April 2016 at 20:40, Adrian Hunter <adrian.hunter@intel.com> wrote:
>> On 28/04/2016 9:29 p.m., Ulf Hansson wrote:
>>>
>>> For the reasons explained below, it's safe to remove the rescan_disable
>>> flag.
>>>
>>> 1)
>>> cancel_delayed_work_sync() prevents an executing work from re-scheduling
>>> itself.
>>>
>>> 2)
>>> We are using the system_freezable_wq for the rescan works. As that queue
>>> becomes frozen when userspace becomes frozen during system PM, we don't
>>> need to disable the rescan works in the PM_SUSPEND_PREPARE phase.
>>
>>
>> What about the case when a SDIO card has to be removed?  Seems like a rescan
>> could theoretically get scheduled and run before the workqueue is frozen.
> 
> When the queue gets frozen, it means currently running works will be
> synced. Works that are scheduled (or becomes scheduled) will be put on
> hold and not allowed to run before the workqueue becomes un-frozen.

In the pm_notifier the workqueue is not frozen, so new work could be queued
and run, which would potentially race with the "host->bus_ops->remove(host)"
a few lines further on.