Re: [PATCH] driver core: Make sure device detached from driver before deleting it

[jeffy]

Hi Rafael,

Thanks for your reply.
On 10/18/2017 06:04 PM, Rafael J. Wysocki wrote:
But device_del() calls bus_remove_device() which in turn calls
device_release_driver(), so this looks like an ordering issue to me.

uh, right, didn't notice that, it indeed to be a ordering issue...

it turns out in device_del() we are calling device_links_purge() before 
bus_remove_device()...

What*exactly*  is not working?  Or rather, what symptoms do you see?

my board has these devices:
spi master device->spi child device->spi based pwm->pwm_bl

and i add a device link to the pwm and pwm_bl, and got a warning about 
the pwm not unbound:

static void device_links_purge(struct device *dev)
{
...
        list_for_each_entry_safe_reverse(link, ln, 
&dev->links.consumers, s_node) {
                WARN_ON(link->status != DL_STATE_DORMANT &&
                        link->status != DL_STATE_NONE); <-- warning here!
                __device_link_del(link);
        }


and i was confused by:
static void mac80211_hwsim_del_radio(struct mac80211_hwsim_data *data,
                                     const char *hwname,
                                     struct genl_info *info)
{
...
        device_release_driver(data->dev);
        device_unregister(data->dev);

and adding that device_release_driver hide that issue(oops...)


so we should move device_links_purge() after bus_remove_device() in 
device_del() right? maybe after device_remove_properties(), and that did 
solve the issue :)


Thanks,
Rafael