Re: [PATCH v2] PCI: endpoint: Skip odd BAR when skipping 64bit BAR

From: Alan Mikhak
Date: Fri May 24 2019 - 14:53:57 EST


Hi Kishon,

Yes. This change is still applicable even when the platform specifies
that it only supports 64-bit BARs by setting the bar_fixed_64bit
member of epc_features.

The issue being fixed is this: If the 'continue' statement is executed
within the loop, the loop index 'bar' needs to advanced by two, not
one, when the BAR is 64-bit. Otherwise the next loop iteration will be
on an odd BAR which doesn't exist.

The PCI_BASE_ADDRESS_MEM_TYPE_64 flag in epf_bar->flag reflects the
value set by the platform in the bar_fixed_64bit member of
epc_features.

This patch moves the checking of PCI_BASE_ADDRESS_MEM_TYPE_64 in
epf_bar->flags to before the 'continue' statement to advance the 'bar'
loop index accordingly. The comment you see about 'pci_epc_set_bar()'
preceding the moved code is the original comment and was also moved
along with the code.

Regards,
Alan Mikhak

On Fri, May 24, 2019 at 1:51 AM Kishon Vijay Abraham I <kishon@xxxxxx> wrote:
>
> Hi,
>
> On 24/05/19 5:25 AM, Alan Mikhak wrote:
> > +Bjorn Helgaas, +Gustavo Pimentel, +Wen Yang, +Kangjie Lu
> >
> > On Thu, May 23, 2019 at 2:55 PM Alan Mikhak <alan.mikhak@xxxxxxxxxx> wrote:
> >>
> >> Always skip odd bar when skipping 64bit BARs in pci_epf_test_set_bar()
> >> and pci_epf_test_alloc_space().
> >>
> >> Otherwise, pci_epf_test_set_bar() will call pci_epc_set_bar() on odd loop
> >> index when skipping reserved 64bit BAR. Moreover, pci_epf_test_alloc_space()
> >> will call pci_epf_alloc_space() on bind for odd loop index when BAR is 64bit
> >> but leaks on subsequent unbind by not calling pci_epf_free_space().
> >>
> >> Signed-off-by: Alan Mikhak <alan.mikhak@xxxxxxxxxx>
> >> Reviewed-by: Paul Walmsley <paul.walmsley@xxxxxxxxxx>
> >> ---
> >> drivers/pci/endpoint/functions/pci-epf-test.c | 25 ++++++++++++-------------
> >> 1 file changed, 12 insertions(+), 13 deletions(-)
> >>
> >> diff --git a/drivers/pci/endpoint/functions/pci-epf-test.c b/drivers/pci/endpoint/functions/pci-epf-test.c
> >> index 27806987e93b..96156a537922 100644
> >> --- a/drivers/pci/endpoint/functions/pci-epf-test.c
> >> +++ b/drivers/pci/endpoint/functions/pci-epf-test.c
> >> @@ -389,7 +389,7 @@ static void pci_epf_test_unbind(struct pci_epf *epf)
> >>
> >> static int pci_epf_test_set_bar(struct pci_epf *epf)
> >> {
> >> - int bar;
> >> + int bar, add;
> >> int ret;
> >> struct pci_epf_bar *epf_bar;
> >> struct pci_epc *epc = epf->epc;
> >> @@ -400,8 +400,14 @@ static int pci_epf_test_set_bar(struct pci_epf *epf)
> >>
> >> epc_features = epf_test->epc_features;
> >>
> >> - for (bar = BAR_0; bar <= BAR_5; bar++) {
> >> + for (bar = BAR_0; bar <= BAR_5; bar += add) {
> >> epf_bar = &epf->bar[bar];
> >> + /*
> >> + * pci_epc_set_bar() sets PCI_BASE_ADDRESS_MEM_TYPE_64
> >> + * if the specific implementation required a 64-bit BAR,
> >> + * even if we only requested a 32-bit BAR.
> >> + */
>
> set_bar shouldn't set PCI_BASE_ADDRESS_MEM_TYPE_64. If a platform supports only
> 64-bit BAR, that should be specified in epc_features bar_fixed_64bit member.
>
> Thanks
> Kishon