Re: [Letux-kernel] BUG: drivers/pinctrl/core: races in pinctrl_groups and deferred probing

["H. Nikolaus Schaller" <hns@xxxxxxxxxxxxx>]

Hi Tony,

> Am 19.06.2018 um 08:11 schrieb Tony Lindgren <tony@xxxxxxxxxxx>:
> 
> * H. Nikolaus Schaller <hns@xxxxxxxxxxxxx> [180619 04:54]:
>>>> I had seen the call sequence
>>>> 
>>>> 
>>>> 
>>>> w/o any lock inside.
> 
> So the sequence above has mutex added around adding the pin
> controller specific functions and groups by the patch series
> I posted for both pcs_parse_bits_in_pinctrl_entry() and
> pcs_parse_one_pinctrl_entry(). So I think the above should
> be fixed now. But please confirm to make sure I'm not mistaken.

Ah, now I see.

My dump_stack() added to pinctrl_generic_add_group() reported

[    6.155487] Hardware name: Generic OMAP36xx (Flattened Device Tree)
[    6.162048] [<c01106ec>] (unwind_backtrace) from [<c010c058>] (show_stack+0x10/0x14)
[    6.170166] [<c010c058>] (show_stack) from [<c074bc28>] (dump_stack+0x7c/0x9c)
[    6.177734] [<c074bc28>] (dump_stack) from [<c042bbec>] (pinctrl_generic_add_group+0x48/0x90)
[    6.186614] [<c042bbec>] (pinctrl_generic_add_group) from [<c043203c>] (pcs_dt_node_to_map+0x4b0/0x81c)
[    6.196441] [<c043203c>] (pcs_dt_node_to_map) from [<c042ffd4>] (pinctrl_dt_to_map+0x1ec/0x2b8)
[    6.205535] [<c042ffd4>] (pinctrl_dt_to_map) from [<c042d028>] (create_pinctrl+0x58/0x2f8)
[    6.214141] [<c042d028>] (create_pinctrl) from [<c042d388>] (devm_pinctrl_get+0x2c/0x6c)
[    6.222625] [<c042d388>] (devm_pinctrl_get) from [<c04f2e9c>] (pinctrl_bind_pins+0x3c/0x138)
[    6.231445] [<c04f2e9c>] (pinctrl_bind_pins) from [<c04d3eb8>] (driver_probe_device+0xe8/0x318)
[    6.240509] [<c04d3eb8>] (driver_probe_device) from [<c04d4168>] (__driver_attach+0x80/0xa4)
[    6.249328] [<c04d4168>] (__driver_attach) from [<c04d264c>] (bus_for_each_dev+0x58/0x7c)

Apparently I didn't notice that pcs_parse_*_pinctrl_entry() are called
inside pcs_dt_node_to_map() and are part of the call sequence.

Hence your new mutex locks calls to pinctrl_generic_add_group() as
needed.

Obviously, the compiler has optimized away the nested calls to static
functions and I had no previous experience with how the whole pinctrl code
works (learning by debugging :).

So it looks sane and no need for further locks.

BR and thanks,
Nikolaus

--
To unsubscribe from this list: send the line "unsubscribe linux-gpio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html