Re: [PATCH] PCI: set BAR size bits correctly in Resize BAR control register

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Sun, Jul 21, 2019 at 8:17 PM Koenig, Christian
<Christian.Koenig@xxxxxxx> wrote:
>
> Am 19.07.19 um 19:56 schrieb Sumit Saxena:
> > +Christian Koenig
> >
> > On Tue, Jul 16, 2019 at 3:41 PM Sumit Saxena <sumit.saxena@xxxxxxxxxxxx> wrote:
> >> In Resize BAR control register, bits[8:12] represents size of BAR.
> >> As per PCIe specification, below is encoded values in register bits
> >> to actual BAR size table:
> >>
> >> Bits  BAR size
> >> 0     1 MB
> >> 1     2 MB
> >> 2     4 MB
> >> 3     8 MB
> >> --
> >>
> >> For 1 MB BAR size, BAR size bits should be set to 0 but incorrectly
> >> these bits are set to "1f".
> >> Latest megaraid_sas and mpt3sas adapters which support Resizable BAR
> >> with 1 MB BAR size fails to initialize during system resume from S3 sleep.
> >>
> >> Fix: Correctly set BAR size bits to "0" for 1MB BAR size.
> >>
> >> CC: stable@xxxxxxxxxxxxxxx # v4.16+
> >> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=203939
> >> Fixes: d3252ace0bc652a1a244455556b6a549f969bf99 ("PCI: Restore resized BAR state on resume")
> >> Signed-off-by: Sumit Saxena <sumit.saxena@xxxxxxxxxxxx>
> >> ---
> >>   drivers/pci/pci.c | 5 +++--
> >>   1 file changed, 3 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> >> index 8abc843..b651f32 100644
> >> --- a/drivers/pci/pci.c
> >> +++ b/drivers/pci/pci.c
> >> @@ -1417,12 +1417,13 @@ static void pci_restore_rebar_state(struct pci_dev *pdev)
> >>
> >>          for (i = 0; i < nbars; i++, pos += 8) {
> >>                  struct resource *res;
> >> -               int bar_idx, size;
> >> +               int bar_idx, size, order;
> >>
> >>                  pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
> >>                  bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
> >>                  res = pdev->resource + bar_idx;
> >> -               size = order_base_2((resource_size(res) >> 20) | 1) - 1;
> >> +               order = order_base_2((resource_size(res) >> 20) | 1);
> >> +               size = order ? order - 1 : 0;
>
> That actually doesn't looks like it is correct or at least it's
> unnecessary complex.
>
> The " >> 20) | 1" seems like a copy & paste error from the code in
> amdgpu where the BAR needs to larger than the VRAM size (which is not a
> power of two).
>
> So just using "size = order_base_2(resource_size(res) >> 20);" should be
> sufficient here.
Agreed, thanks for feedback. I will send simplified version of the
patch using just-
"size = order_base_2(resource_size(res) >> 20);".

Thanks,
Sumit
>
> Regards,
> Christian.
>
> >>                  ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
> >>                  ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
> >>                  pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
> >> --
> >> 1.8.3.1
> >>
>



[Index of Archives]     [DMA Engine]     [Linux Coverity]     [Linux USB]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Greybus]

  Powered by Linux