Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Robert Jarzmik <robert.jarzmik@free.fr>
To:     Felipe Balbi <balbi@kernel.org>
Cc:     Jia-Ju Bai <baijiaju1990@gmail.com>, gregkh@linuxfoundation.org,
        nicolas.ferre@microchip.com, keescook@chromium.org,
        allen.lkml@gmail.com, linux-usb@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/2] usb: gadget: r8a66597: Fix two possible sleep-in-atomic-context bugs in init_controller()
References: <20180620035453.7721-1-baijiaju1990@gmail.com>
        <87efh0o6f9.fsf@belgarion.home> <87y3eydrs1.fsf@linux.intel.com>
        <87efgqkrzl.fsf@belgarion.home> <87k1qhevpd.fsf@linux.intel.com>
Date:   Fri, 29 Jun 2018 13:04:58 +0200
In-Reply-To: <87k1qhevpd.fsf@linux.intel.com> (Felipe Balbi's message of "Fri,
        29 Jun 2018 13:25:02 +0300")
Message-ID: <87a7rdlup1.fsf@belgarion.home>
User-Agent: Gnus/5.130008 (Ma Gnus v0.8) Emacs/24.5 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Felipe Balbi <balbi@kernel.org> writes:

> Hi,
>
> Robert Jarzmik <robert.jarzmik@free.fr> writes:
>>>>> The driver may sleep with holding a spinlock.
>>>>> The function call paths (from bottom to top) in Linux-4.16.7 are:
>>>>>
>>>>> [FUNC] msleep
>>>>> drivers/usb/gadget/udc/r8a66597-udc.c, 839: 
>>>>> 		msleep in init_controller
>>>>> drivers/usb/gadget/udc/r8a66597-udc.c, 96: 
>>>>> 		init_controller in r8a66597_usb_disconnect
>>>>> drivers/usb/gadget/udc/r8a66597-udc.c, 93: 
>>>>> 		spin_lock in r8a66597_usb_disconnect
>>>>
>>>> That should not happen...
>>>>
>>>> If think the issue you have is that your usb_connect() and usb_disconnect() are
>>>> called from interrupt context. I think the proper fix, as what is done in most
>>>> udc phys, is to schedule a workqueue, see drivers/usb/phy/phy-gpio-vbus-usb.c,
>>>> gpio_vbus_data.vbus.
>>>
>>> argh, no. No workqueues needed here. Sorry
>> Technically why ?
>
> well, strictly technically there's nothing wrong. But it opens a can of
> worms. We've seen time and time again drivers growing into
> unmaintainable mess because of workqueues being fired in several places.
I see.

>>
>> And as bonus question, why is it better to have mdelay() calls in the driver ?
>
> As a bugfix, it's the smallest fix possible, right? Ideally, we wouldn't
> need either of them. Perhaps there's a bit which can be polled instead?
Ideally yes. Do you remember if a "threaded interrupt" might use msleep() ? I
seem to remember that they can, so won't that be another alternative ?

Cheers.

-- 
Robert