Hi Doug,

On (26/01/28 10:05), Doug Anderson wrote:

quoted

rtl8152 can trigger device reset during reset which
potentially can result in a deadlock:

 **** DPM device timeout after 10 seconds; 15 seconds until panic ****
 Call Trace:
 <TASK>
 schedule+0x483/0x1370
 schedule_preempt_disabled+0x15/0x30
 __mutex_lock_common+0x1fd/0x470
 __rtl8152_set_mac_address+0x80/0x1f0
 dev_set_mac_address+0x7f/0x150
 rtl8152_post_reset+0x72/0x150
 usb_reset_device+0x1d0/0x220
 rtl8152_resume+0x99/0xc0
 usb_resume_interface+0x3e/0xc0
 usb_resume_both+0x104/0x150
 usb_resume+0x22/0x110

The problem is that rtl8152 resume calls reset under
tp->control mutex while reset basically re-enters rtl8152
and attempts to acquire the same tp->control lock once
again.

Reset INACCESSIBLE device outside of tp->control mutex
scope to avoid recursive mutex_lock() deadlock.

Fixes: 4933b066fefb ("r8152: If inaccessible at resume time, issue a reset")
Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
---
 drivers/net/usb/r8152.c | 27 ++++++++++++++-------------
 1 file changed, 14 insertions(+), 13 deletions(-)

This is effectively v2 of:

https://lore.kernel.org/r/20241018141337.316807-1-danielgeorgem@chromium.org/ (local)

...and you've incorporated my feedback there. Thanks! :-)

Oh, nice :)

quoted

@@ -8674,6 +8662,19 @@ static int rtl8152_resume(struct usb_interface *intf)

        mutex_unlock(&tp->control);

+       /* If the device is RTL8152_INACCESSIBLE here then we should do a
+        * reset. This is important because the usb_lock_device_for_reset()
+        * that happens as a result of usb_queue_reset_device() will silently
+        * fail if the device was suspended or if too much time passed.
+        *
+        * NOTE: The device is locked here so we can directly do the reset.
+        * We don't need usb_lock_device_for_reset() because that's just a
+        * wrapper over device_lock() and device_resume() (which calls us)
+        * does that for us.
+        */
+       if (system_resume && test_bit(RTL8152_INACCESSIBLE, &tp->flags))
+               ret = usb_reset_device(tp->udev);
+
        return ret;

Question when looking at the above again: have you thought about the
consequences of clobbering `ret` above? I guess it's fine since
rtl8152_system_resume() always returns 0, but it looks a little
awkward. It's been long enough since I thought through all this code
that I'm not 100% sure what it _should_ do if rtl8152_system_resume()
was ever changed to return an error. Shouldn't it honor the existing
error instead of trying to reset the device and clearing the error?

Right... so that "ret" thing, I thought about it and at the end I
just decided that returning an actual device reset error from resume
is still better than "return 0 but device is inaccessible" ("mission
failed successfully" kind of a thing).  I'm not entirely sure what
would be the best way to handle this.  Like you said, for the time
being, rtl8152_system_resume() always returns 0.  Do we expect this
to change in the future?  Probably not.  On the other hand if
RTL8152_INACCESSIBLE bit is not cleared then user-space will
figure it out eventually (ioctl calls will fail, etc).  So maybe I
can just keep the existing code and ignore usb_reset_device() return
value.

Also: I guess you've added the `system_resume` variable here, which is
different than the earlier patch. It seems fine to me, though maybe
you want to consistently use the `system_resume` variable earlier in
the function too?

Sounds good!

In any case, both of the above are pretty nitty, so I'm OK with:

Reviewed-by: Douglas Anderson <dianders@chromium.org>

Thanks!