Thanks Peter. v2 sent with your suggested comment added.
On Thu, Mar 19, 2026 at 6:34 PM Peter Maydell <[email protected]> wrote:
>
> On Wed, 18 Mar 2026 at 19:36, Jenny Guanni Qu <[email protected]> wrote:
> >
> > When a guest sets MaxPacketSize to 0 in an OHCI Endpoint Descriptor,
> > ohci_service_td() transfers 0 bytes per iteration. The Transfer
> > Descriptor never completes because CBP never advances toward BE,
> > causing ohci_service_ed_list() to loop indefinitely and hang QEMU.
> >
> > Add a check for MPS==0 after extracting the field from ED flags.
> > If MPS is zero, call ohci_die() to reset the controller and return
> > an error, preventing the infinite loop.
> >
> > Fixes: CVE-2026-3890
> > Reported-by: Jenny Guanni Qu <[email protected]>
> > Signed-off-by: Jenny Guanni Qu <[email protected]>
> > ---
> > hw/usb/hcd-ohci.c | 4 ++++
> > 1 file changed, 4 insertions(+)
> >
> > diff --git a/hw/usb/hcd-ohci.c b/hw/usb/hcd-ohci.c
> > index 1aeed9286f..b93a99593e 100644
> > --- a/hw/usb/hcd-ohci.c
> > +++ b/hw/usb/hcd-ohci.c
> > @@ -956,6 +956,10 @@ static int ohci_service_td(OHCIState *ohci, struct
> > ohci_ed *ed)
> > if (len && dir != OHCI_TD_DIR_IN) {
> > /* The endpoint may not allow us to transfer it all now */
> > pktlen = (ed->flags & OHCI_ED_MPS_MASK) >> OHCI_ED_MPS_SHIFT;
> > + if (pktlen == 0) {
> > + ohci_die(ohci);
> > + return 1;
> > + }
>
> The OHCI spec doesn't (as far as I can tell) say what the
> hardware should do for this situation, so I guess "flag it
> as an error" is a reasonable choice. I think it would be
> useful to note in a comment that this is our best-guess, though:
>
> /*
> * The OHCI spec doesn't say what to do if the guest hands us
> * an endpoint descriptor which specifies a MaximumPacketSize
> * of zero, which would mean we can never actually make forward
> * progress transferring data to it. We choose to treat it as
> * an error.
> */
>
> With something like that,
>
> Reviewed-by: Peter Maydell <[email protected]>
>
> (Or if I've missed something in the spec, let me know...)
>
> thanks
> -- PMM
