On Wed, Jan 22, 2025, Andy Shevchenko wrote:
> On Wed, Jan 22, 2025 at 01:44:02AM +0000, Thinh Nguyen wrote:
> > On Fri, Jan 17, 2025, Andy Shevchenko wrote:
> > > On Thu, Jan 16, 2025 at 11:35:19PM +0000, Thinh Nguyen wrote:
> > > > On Thu, Jan 16, 2025, Andy Shevchenko wrote:
>
> ...
>
> > > > > for (epnum = 0; epnum < total; epnum++) {
> > > > > - int ret;
> > > > > + for (num = 0; num < count; num++) {
> > > > > + if (epnum == eps[num])
> > > > > + break;
> > > > > + }
> > > > > + if (num < count)
> > > > > + continue;
> > > >
> > > > You can probably rewrite this logic better.
> > >
> > > Any suggestions?
> > >
> > > Thanks for the review!
> >
> > From the first look, is the list sorted? If so, you don't need another
> > for-loop.
>
> Even if it's sorted it's not 1:1 mapped by indices. I do not understand how we
> can avoid the second loop. The only possibility is indeed to sort the list and
> sparse it in accordance to the endpoint numbers, but if we are going this way,
> it's much easier to switch to bitmap and the respective bitops.
If it's sorted, it can be something like this. Just a quick logic and not tested:
num = 0
for (epnum = 0; epnum < total; epnum++) {
if (num < count && epnum == eps[num]) {
num++;
continue;
}
...
}
>
> > Also, we loop over the number of endpoints throughout the driver, but
> > you only skip it here during init. Please double check for invalid
> > accessing of endpoints in other places.
> >
> > Perhaps set the dwc->eps[reserved_ep] to ERR_PTR(-EINVAL) or something
> > when you parse the reserved endpoints so you can skip them in your loop.
>
> Note, this is only for UDC, in USB host these are okay to be used.
> Does your suggestion imply that?
>
No. We track the total num_eps in dwc->num_eps. Then we do for-loop to
dwc->eps[i] and access the endpoint. Often we check if the endpoint is
NULL before accessing dwc->eps[i]. However, we don't do it everywhere.
So I ask for you to review to make sure that this change doesn't break
elsewhere where we may try to access dwc->eps[i] to an uninit endpoint
(Note I see at least 1 place e.g. dwc3_gadget_clear_tx_fifos that may
break)
BR,
Thinh
