Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Sender: linyyuan=codeaurora.org@mg.codeaurora.org
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII;
 format=flowed
Content-Transfer-Encoding: 7bit
Date:   Sun, 27 Jun 2021 10:48:56 +0800
From:   linyyuan@codeaurora.org
To:     Alan Stern <stern@rowland.harvard.edu>
Cc:     Felipe Balbi <balbi@kernel.org>,
        Thinh Nguyen <Thinh.Nguyen@synopsys.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        linux-usb@vger.kernel.org, linux-kernel@vger.kernel.org,
        Jack Pham <jackp@codeaurora.org>
Subject: Re: [PATCH] usb: dwc3: fix race of usb_gadget_driver operation
In-Reply-To: <20210626150304.GA601624@rowland.harvard.edu>
References: <20210625104415.8072-1-linyyuan@codeaurora.org>
 <20210625163707.GC574023@rowland.harvard.edu>
 <b24825113327c72c742d55e89ec2726e@codeaurora.org>
 <20210626150304.GA601624@rowland.harvard.edu>
Message-ID: <1d1f06763c7cdeb67264128537c6a8f4@codeaurora.org>
User-Agent: Roundcube Webmail/1.3.9
Precedence: bulk

On 2021-06-26 23:03, Alan Stern wrote:
> On Sat, Jun 26, 2021 at 09:16:25AM +0800, linyyuan@codeaurora.org 
> wrote:
>> On 2021-06-26 00:37, Alan Stern wrote:
>> > On Fri, Jun 25, 2021 at 06:44:15PM +0800, Linyu Yuan wrote:
> 
>> > > --- a/drivers/usb/dwc3/ep0.c
>> > > +++ b/drivers/usb/dwc3/ep0.c
>> > > @@ -597,10 +597,11 @@ static int dwc3_ep0_set_address(struct dwc3
>> > > *dwc, struct usb_ctrlrequest *ctrl)
>> > >
>> > >  static int dwc3_ep0_delegate_req(struct dwc3 *dwc, struct
>> > > usb_ctrlrequest *ctrl)
>> > >  {
>> > > -	int ret;
>> > > +	int ret = 0;
>> > >
>> > >  	spin_unlock(&dwc->lock);
>> > > -	ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
>> > > +	if (dwc->async_callbacks)
>> > > +		ret = dwc->gadget_driver->setup(dwc->gadget, ctrl);
>> > >  	spin_lock(&dwc->lock);
>> >
>> > Here and in the other places, you should test dwc->async_callbacks
>> > _before_ dropping the spinlock.  Otherwise there is a race (the flag
>> > could be written at about the same time it is checked).
>> thanks for your comments,
>> 
>> if you think there is race here, how to make sure gadget_driver 
>> pointer is
>> safe,
>> this is closest place where we can confirm it is non-NULL by checking
>> async_callbacks ?
> 
> I explained this twice already: We know that gadget_driver is not
> NULL because usb_gadget_remove_driver calls synchronize_irq before
> doing usb_gadget_udc_stop.
> 
> Look at this timing diagram:
> 
> 	CPU0				CPU1
> 	----				----
> 	IRQ happens for setup packet
> 	  Handler sees async_callbacks
> 	    is enabled
> 	  Handler unlocks dwc->lock
> 					usb_gadget_remove_driver runs
> 					  Disables async callbacks
> 					  Calls synchronize_irq
> 	  Handler calls dwc->		  . waits for IRQ handler to
> 	    gadget_driver->setup	  .   return
> 	  Handler locks dwc-lock	  .
> 	  ...				  .
> 	  Handler returns		  .
> 					  . synchronize_irq returns
> 					  Calls usb_gadget_udc_stop
> 					    dwc->gadget_driver is
> 					      set to NULL
> 
> As you can see, dwc->gadget_driver is non-NULL when CPU0 uses it,
> even though async_callbacks gets cleared during the time when the
> lock is released.
thanks for your patient explanation,
but from this part, seem it is synchronize_irq() help to avoid NULL 
pointer crash.

can you also explain how async_callbacks flag help here  ?
> 
> Alan Stern