linux-kernel - Re: [PATCH 2/5] usb: gadget: Add function wakeup support

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: <1fac4c0f-4e8b-f333-7208-c50353a806f4@synopsys.com>
Date:   Sat, 13 Aug 2022 00:46:01 +0000
From:   Thinh Nguyen <Thinh.Nguyen@...opsys.com>
To:     Thinh Nguyen <Thinh.Nguyen@...opsys.com>,
        Elson Serrao <quic_eserrao@...cinc.com>,
        "balbi@...nel.org" <balbi@...nel.org>,
        "gregkh@...uxfoundation.org" <gregkh@...uxfoundation.org>
CC:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linux-usb@...r.kernel.org" <linux-usb@...r.kernel.org>,
        "quic_wcheng@...cinc.com" <quic_wcheng@...cinc.com>,
        "quic_jackp@...cinc.com" <quic_jackp@...cinc.com>,
        "quic_mrana@...cinc.com" <quic_mrana@...cinc.com>
Subject: Re: [PATCH 2/5] usb: gadget: Add function wakeup support

On 8/11/2022, Thinh Nguyen wrote:
> On 8/11/2022, Thinh Nguyen wrote:
>> On 8/11/2022, Elson Serrao wrote:
>>>
>>>
>>> On 8/9/2022 6:08 PM, Thinh Nguyen wrote:
>>
>> <snip>
>>
>>
>>>> To summarize the points:
>>>>
>>>> 1) The host only arms function remote wakeup if the device is capable of
>>>> remote wakeup (check USB_CONFIG_ATT_WAKEUP in bmAttributes and hardware
>>>> capability)
>>>>
>>>> 2) If the device is in suspend, the device can do remote wakeup (through
>>>> LFPS handshake) if the function is armed for remote wakeup (through
>>>> SET_FEATURE(FUNC_SUSPEND)).
>>>>
>>>> 3) If the link transitions to U0 after the device triggering a remote
>>>> wakeup, the device will also send device notification function wake for
>>>> all the interfaces armed with remote wakeup.
>>>>
>>>> 4) If the device is not in suspend, the device can send device
>>>> notification function wake if it's in U0.
>>>>
>>>>
>>>> Now, remote wakeup and function wake device notification are 2 separate
>>>> operations. We have the usb_gadget_ops->wakeup() for remote wakeup. I
>>>> suggested to maybe add usb_gadget_ops->send_wakeup_notification(gadget,
>>>> intf_id) for the device notification. What you did was combining both
>>>> operations in usb_gadget_ops->func_wakeup(). That may only work for
>>>> point 4) (assuming you fix the U0 check), but not point 3).
>>>
>>> Thank you for your feedback and summary. I will rename func_wakeup to
>>> send_wakeup_notification to better align with the approach. The reason I
>>> have combined remote_wakeup and function wake notification in
>>> usb_gadget_ops->func_wakeup() is because since the implementation is at
>>> function/composite level it has no knowledge on the link state. So I
>>> have delegated that task to controller driver to handle the notification
>>> accordingly. That is do a LFPS handshake first if the device is
>>> suspended and then send notification (explained below). But we can
>>> definitely separate this by adding an additional flag in the composite
>>> layer to set the link state based on the gadget suspend callback called
>>> when U3 SUSPEND interrupt is received. Let me know if you feel
>>> separating the two is a better approach.
>>>
>>
>> The reason I think we need to separate it is because of point 3. As I
>> note earlier, the spec states that "If remote wake event occurs in
>> multiple functions, each function shall send a Function Wake Notification."
>>
>> But if there's no remote wake event, and the host brought the device up
>> instead, then the function suspend state is retained.
>>
>> If we separate these 2 operations, the caller can check whether the
>> operation went through properly. For example, if the wakeup() is
>> initiated properly, but the function wake device notification didn't go
>> through. We would only need to resend the device notification rather
>> than initiate remote wakeup again.
> 
> If we don't have to send device notification for other interfaces, we
> can combine the operations here as you did.
> 

I still think it's better to split up the operations. The way you're
handling it now is not clear.

If the func_awake() returns -EAGAIN, I'd expect that the remote wake did
not go through and expect user to retry again. But here it does initiate
remote wake, but it just does not send device notification yet. This is
confusing.

Also, instead of all the function wake handling coming from the function
driver, now we depend on the controller driver to call function resume()
on state change to U0, which will trigger device notification. What
happen if it doesn't call resume(). There's too many dependencies and it
seems fragile.

I think all this can be handled in the function driver. You can fix the
dwc3 wakeup() and poll for U0/ON state rather than RECOVERY state, which
is what it's supposed to poll. On usb_gadget_wakeup() returns
successful, we'd expect the device is linked up and woken up. then you
can send device notification through a different api such as
usb_gadget_send_wake_notification().

Thanks,
Thinh