On Tuesday, August 24, 2021 2:09:10 PM CEST Pavel Skripkin wrote:
> On 8/24/21 3:01 PM, Fabio M. De Francesco wrote:
> > On Tuesday, August 24, 2021 1:07:46 PM CEST Pavel Skripkin wrote:
> >>
> >> Btw, not related to your patch, but I start think, that this check:
> >>
> >>
> >> if (!pIo_buf) {
> >> DBG_88E("[%s] pIo_buf == NULL\n", __func__);
> >> status = -ENOMEM;
> >> goto release_mutex;
> >> }
> >>
> >> Should be wrapped as
> >>
> >> if (WARN_ON(unlikely(!pIo_buf)) {
> >> ...
> >> }
> >>
> >> Since usb_vendor_req_buf is initialized in ->probe() and I can't see
> >> possible calltrace, which can cause zeroing this pointer.
> >
> > I see that usb_vendor_req_buf is initialized in rtw_init_intf_priv(). It depends on a
> > kzalloc() success on allocating memory. Obviously it could fail to allocate. If it fails,
> > rtw_init_intf_priv() returns _FAIL to its caller(s) (whichever they are - I didn't go too
> > deep in understanding the possible calls chains).
> >
>
> Call chain is the most interesting part here :)
>
> rtw_drv_init() <-- probe()
> usb_dvobj_init()
> rtw_init_intf_priv()
>
> If kzalloc fails, then whole ->probe() routine fails, i.e device will be
> disconnected.
I guess that if probe fails and then the device get disconnected it's not a
big problem, in the sense that nothing of very bad could happen.
> There is no read() calls before rtw_init_intf_priv(), so
> if kzalloc() call was successful, there is no way how usb_vendor_req_buf
> can be NULL, since read() can happen only in case of successfully
> connected device.
Yes, though I have very little knowledge of how drivers work, it make sense
to me too that read(s) can happen only in case of successful connection.
> Anyway, it can be NULL in case of out-of-bound write or smth else,
This is really something I don't know.
> but
> there is no explicit usb_alloc_vendor_req_buf = NULL in this driver.
> We should complain about completely wrong driver behavior, IMO :)
>
> Does it make sense?
I'm not sure, whether or not we now have an answer to your question
about the necessity to use WARN_ON... I think it's up to your judgement,
because I cannot help on this topic :(
Regards,
Fabio
> With regards,
> Pavel Skripkin
>
