On Tue, Nov 21, 2017 at 01:24:53PM +0530, Kishon Vijay Abraham I wrote:
> Hi Lorenzo,
>
> On Saturday 18 November 2017 12:13 AM, Lorenzo Pieralisi wrote:
> > [+Kishon - please CC him next time]
> >
> > On Fri, Nov 17, 2017 at 04:00:40PM +0100, Niklas Cassel wrote:
> >> find_first_zero_bit()'s parameter 'size' is defined in bits,
> >> not in bytes.
> >>
> >> find_first_zero_bit() was called with size in bytes rather than bits,
> >> which thus defined a too low upper limit, causing
> >> dw_pcie_ep_inbound_atu() to assign iatu index #4 to both bar 4
> >> and bar 5, which made bar 5 overwrite the settings set by bar 4.
> >>
> >> Fix this by using replacing find_first_zero_bit() with ffz(),
> >> since ffz() only works on a single 'unsigned long' and therefore
> >> does not need a size argument.
> >>
> >> Fixes: f8aed6ec624f ("PCI: dwc: designware: Add EP mode support")
> >> Signed-off-by: Niklas Cassel <niklas.cassel@xxxxxxxx>
> >> ---
> >> drivers/pci/dwc/pcie-designware-ep.c | 6 ++----
> >> 1 file changed, 2 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/drivers/pci/dwc/pcie-designware-ep.c b/drivers/pci/dwc/pcie-designware-ep.c
> >> index d53d5f168363..ab9a9e160daf 100644
> >> --- a/drivers/pci/dwc/pcie-designware-ep.c
> >> +++ b/drivers/pci/dwc/pcie-designware-ep.c
> >> @@ -70,8 +70,7 @@ static int dw_pcie_ep_inbound_atu(struct dw_pcie_ep *ep, enum pci_barno bar,
> >> u32 free_win;
> >> struct dw_pcie *pci = to_dw_pcie_from_ep(ep);
> >>
> >> - free_win = find_first_zero_bit(&ep->ib_window_map,
> >> - sizeof(ep->ib_window_map));
> >> + free_win = ffz(ep->ib_window_map);
> >
> > You fix the bug at hand but code suffers from the same issue since we
> > should be checking the value against ~0UL first ie if size in bits <
> > num_ib_windows (that I think it is unlikely anyway but while at it let's
> > think if we can improve this).
> >
> > @Kishon: maybe we should add a static upper size and declare a
> > static bitmap accordingly (and still use find_first_zero_bit()) ?
> >
> > I think the check against ~0UL should be added anyway, let's get
> > Kishon's opinion before a v3 (if any).
>
> Yeah, for both ffz and find_first_zero_bit, the results are undefined if no
> zero exists.
>
> I'm thinking if we should just use
> find_first_zero_bit(&ep->ib_window_map, ep->num_ib_windows) after checking if
> there are some '0' bits left.
Well yes but first the &ep->ib_window_map must be sized accordingly.
Lorenzo
