linux-kernel - Re: [PATCH] usb: dwc3: Clear DWC3_EVENT

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <4ced9c3e-c7b5-e0a0-88ec-1ac383d893a2@quicinc.com>
Date:   Tue, 10 Jan 2023 11:05:40 +0800
From:   Linyu Yuan <quic_linyyuan@...cinc.com>
To:     Thinh Nguyen <Thinh.Nguyen@...opsys.com>
CC:     정재훈 <jh0801.jung@...sung.com>,
        'Felipe Balbi' <balbi@...nel.org>,
        'Greg Kroah-Hartman' <gregkh@...uxfoundation.org>,
        "'open list:USB XHCI DRIVER'" <linux-usb@...r.kernel.org>,
        'open list' <linux-kernel@...r.kernel.org>,
        'Seungchull Suh' <sc.suh@...sung.com>,
        'Daehwan Jung' <dh10.jung@...sung.com>
Subject: Re: [PATCH] usb: dwc3: Clear DWC3_EVENT_PENDING when count is 0


On 1/10/2023 10:53 AM, Thinh Nguyen wrote:
> On Tue, Jan 10, 2023, Linyu Yuan wrote:
>> On 1/10/2023 2:28 AM, Thinh Nguyen wrote:
>>> On Fri, Jan 06, 2023, Linyu Yuan wrote:
>>>> On 1/5/2023 5:54 PM, 정재훈 wrote:
>>>>>> -----Original Message-----
>>>>>> From: Linyu Yuan [mailto:quic_linyyuan@...cinc.com]
>>>>>> Sent: Thursday, January 5, 2023 12:35 PM
>>>>>> To: JaeHun Jung; Felipe Balbi; Greg Kroah-Hartman; Thinh Nguyen
>>>>>> Cc: open list:USB XHCI DRIVER; open list; Seungchull Suh; Daehwan Jung
>>>>>> Subject: Re: [PATCH] usb: dwc3: Clear DWC3_EVENT_PENDING when count is 0
>>>>>>
>>>>>>
>>>>>> On 1/5/2023 11:29 AM, Linyu Yuan wrote:
>>>>>>> On 1/2/2023 1:08 PM, JaeHun Jung wrote:
>>>>>>>> Sometimes very rarely, The count is 0 and the DWC3 flag is set has
>>>>>>>> status.
>>>>>>>> It must not have these status. Because, It can make happen interrupt
>>>>>>>> storming status.
>>>>>>> could you help explain without clear the flag, how interrupt storming
>>>>>>> happen ?
>>>>>>>
>>>>>>> as your change didn't touch any hardware register, i don't know how it
>>>>>>> fix storming.
>>>>>>>
>>>>> H/W interrupts are still occur on IP.
>>>> I guess we should fix it from IP layer.
>>>>
>>> How are you certain the problem is from IP layer?
>> I think all IRQ is from DWC3 controller IP. if it is not IP layer, could you
>> share how to prevent from SW layer ?
>>
>> seem IRQ can happen when event count is zero ,  why this can happen ? does
>> it mean event count register is not trust ?
> When the interrupt is unmasked, the controller will generate interrupts
> as events are received. Normally, the flag checking for pending event
> should be cleared before unmasking the interrupt, but we clear it after
> to account for possible false interrupt due to the nature of legacy pci
> interrupt. This exposes a gap where the interrupts can come but the flag
> isn't cleared. While it should be rare and shouldn't be too much of a
> problem, we can avoid this scenario with some additional checks.
>
>>>> but when checking DWC3_EVENT_PENDING flag, it will auto clear in dwc3 thread
>>>> irq handler.
>>>>
>>>> there is one possible root cause is it cleared only after irq enabled in
>>>> dwc3_process_event_buf(),
>>>>
>>>> we should move unmask irq operation at end of this function.
>>>>
>>> This interrupt storm can happen because we clear the evt->flags _after_
>>> we unmask the interrupt. This was done to prevent false interrupt from
>>> delay interrupt deassertion, which can be a problem for legacy pci
>>> interrupt.
>> thanks for explain, i didn't know that.
>>> see 7441b273388b ("usb: dwc3: gadget: Fix event pending check")
>>>
>>> The change JaeHun Jung did should be fine.
>> agree.
> The change may still need some additional check as suggested in my
> response:
> https://lore.kernel.org/linux-usb/20230109190914.3blihjfjdcszazdd@synopsys.com/T/#m7b907aa6da4023cb20fa00a57813d31fd84e081f
  do you think we need to read event count before checking 
DWC3_EVENT_PENDING  ?
>
> BR,
> Thinh