于 2012年12月16日 01:04, Alan Stern 写道:
> On Sat, 15 Dec 2012, Chen Gang wrote:
>
>> > Hello Alan Stern:
>> >
>> > in drivers/usb/host/ohci-q.c, function finish_urb:
>> > when we finish call urb_free_priv, we not set urb->hcpriv = NULL (line 46)
>> > urb_free_priv call kfree for urb_priv (which is urb->hcpriv, line 29)
>> > within finish_urb, we not set urb->hcpriv = NULL (line 43..81)
>> >
>> > in drivers/usb/host/ohci-hdc.c, function ohci_urb_dequeue:
>> > it judges urb->hcpriv == NULL to decide whether call finish_urb (line 317..322)
>> > within ohci_urb_dequeue, we not set urb->hcpriv = NULL (line 321..326)
>> > another functions (finish_unlinks, takeback_td...), can call finish_urb, too.
>> > they do not set urb->hcpriv = NULL, either.
>> > they do not judge urb->hcpriv == NULL before calling finish_urb.
>> >
>> > although I can not say it is surely a bug.
> It isn't. ohci_urb_dequeue calls usb_hcd_check_unlink_urb before
> looking at urb->hcpriv. If urb_free_priv has been called already then
> usb_hcd_check_unlink_urb will return a non-zero value. This is because
> finish_urb calls usb_hcd_unlink_urb_from_ep.
>
for ohci_urb_dequeue, it is surely as what you said above.
can you be sure that another (finish_unlinks, takeback_td...) also consider it ?
it seems we have already considered, but I am not quite sure
(since we already agree to sending a patch, it is not necessary to reply).
> Also, don't forget that the first think usb_hcd_giveback_urb does is
> set urb->hcpriv to NULL.
>
in finish_urb:
when call usb_hcd_giveback_urb, need unlock firstly (lock again, after finish).
kfree urb->hcpriv in lock protected, but set it NULL out of lock protected.
(at least, it seems not a good idea)
(since we already agree to sending a patch, it is not necessary to reply).
>> > I still suggest:
>> > in finish_urb, set urb->hcpriv = NULL, after finish calling urb_free_priv.
>> > in urb_free_priv, better to judge whether urb_priv == NULL, firstly.
> If you want to send in a patch to do this, go ahead.
>
ok, I will send (although it may be not a bug, better to do it)
excuse me, please wait for some days (I should construct environments for test it).
thanks.
> Alan Stern
>
>
>
--
Chen Gang
Asianux Corporation
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
