Re: [PATCH v2] driver core: bus: Remove an impossible error handling path in bus_add_driver()

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Mon, Oct 14, 2024 at 12:13:30AM +0800, Zijun Hu wrote:
> On 2024/10/14 00:00, Greg Kroah-Hartman wrote:
> > On Sun, Oct 13, 2024 at 11:46:46PM +0800, Zijun Hu wrote:
> >> On 2024/10/13 23:02, Greg Kroah-Hartman wrote:
> >>> On Tue, Sep 17, 2024 at 02:53:32PM +0800, Zijun Hu wrote:
> >>>> On 2024/9/17 14:49, Zijun Hu wrote:
> >>>>> From: Zijun Hu <quic_zijuhu@quicinc.com>
> >>>>>
> >>>>> For the following function call chain:
> >>>>> API driver_register() -> bus_add_driver() -> driver_attach()
> >>>>>
> >>>>> There is an error handling path for driver_attach() returning non-zero
> >>>>> or failure in bus_add_driver(), remove it with below reasons:
> >>>>>
> >>>>> - It is impossible for driver_attach() to have failure in bus_add_driver()
> >>>>>   For int driver_attach(const struct device_driver *drv), the only factor
> >>>>>   which makes it failed is that bus_to_subsys(@drv->bus) is NULL, but
> >>>>>   the factor has been excluded by bus_add_driver() before calling it.
> >>>>>
> >>>>> - driver_attach() is irrelevant with driver_register(), so the former's
> >>>>>   result should not also have an impact on the later.
> >>>>>
> >>>>> Signed-off-by: Zijun Hu <quic_zijuhu@quicinc.com>
> >>>>> ---
> >>>>> Changes in v2:
> >>>>> - Remove the error handling path instead of WARN_ON() it.
> >>>>> - Correct title and commit message
> >>>>> - Link to v1: https://lore.kernel.org/r/20240915-bus_add_driver_fix-v1-1-ce5cf1f66601@quicinc.com
> >>>>> ---
> >>>>>  drivers/base/bus.c | 4 ++--
> >>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
> >>>>>
> >>>>> diff --git a/drivers/base/bus.c b/drivers/base/bus.c
> >>>>> index 657c93c38b0d..54ff92aece92 100644
> >>>>> --- a/drivers/base/bus.c
> >>>>> +++ b/drivers/base/bus.c
> >>>>> @@ -674,7 +674,8 @@ int bus_add_driver(struct device_driver *drv)
> >>>>>  	if (sp->drivers_autoprobe) {
> >>>>>  		error = driver_attach(drv);
> >>>>>  		if (error)
> >>>>> -			goto out_del_list;
> >>>>> +			pr_warn("%s: failed to attach driver '%s' to bus '%s'\n",
> >>>>> +				__func__, drv->name, sp->bus->name);
> >>>>
> >>>> driver_attach() has __must_check attribute and this error may be
> >>>> inconsequential for driver_register(), so give pr_warn() here
> >>>
> >>> Yes, but you  now ignore the error, so someone will come back and add
> >>> that error handling in.  I'd just leave it as-is.
> >>>
> >>
> >> driver API driver_attach() may ONLY have below error -EINVAL.
> >> is it worthy of a __must_check attribute ?
> > 
> > Yes.
> > 
> >> i agree with you to leave it as-is if your answer is "YES".
> >> otherwise, i would like to also simply remove __must_check attribute.
> > 
> > Please don't.  If you do that, then callers will end up not checking the
> > results, and we do not want that.
> > 
> okay.
> 
> but as 2nd reason of commit message explained:
> driver_attach() failure should NOT cause driver_register() failure, so
> should ignore driver_attach() failure here.

How could driver_attach() fail and it still be ok to continue?

thanks,

greg k-h