Re: [Qemu-devel] [PATCH v2] Align PCI capabilities in pci_find_space

[Alex Williamson <alex.williamson@redhat.com>]

On Tue, 2012-09-25 at 21:08 -0600, Alex Williamson wrote:
> On Tue, 2012-09-25 at 20:01 -0500, mjr@cs.wisc.edu wrote:
> > From: Matt Renzelmann <mjr@cs.wisc.edu>
> > 
> > The current implementation of pci_find_space does not correctly align
> > PCI capabilities in the PCI configuration space.  This patch fixes
> > this issue.
> > 
> > Signed-off-by: Matt Renzelmann <mjr@cs.wisc.edu>
> > ---
> > 
> > Alex Williamson <alex.williamson@redhat.com> wrote:
> > > I think you could just search every 4th byte.  In fact, this whole used
> > > byte-map could be turned into a single uint64_t bitmap for standard
> > > config space.  Thanks,
> > 
> > I've not tested this version of the patch, in contrast to the last, so
> > I'm a bit less confident of its correctness.  I did not reimplement it
> > as suggested as I'm not that familiar with this code, and instead just
> > applied the every 4th byte strategy.
> > 
> >  hw/pci.c |   12 ++++++++----
> >  1 files changed, 8 insertions(+), 4 deletions(-)
> > 
> > diff --git a/hw/pci.c b/hw/pci.c
> > index f855cf3..e99866a 100644
> > --- a/hw/pci.c
> > +++ b/hw/pci.c
> > @@ -1631,11 +1631,15 @@ static int pci_find_space(PCIDevice *pdev, uint8_t size)
> >      int config_size = pci_config_size(pdev);
> >      int offset = PCI_CONFIG_HEADER_SIZE;
> >      int i;
> > -    for (i = PCI_CONFIG_HEADER_SIZE; i < config_size; ++i)
> > -        if (pdev->used[i])
> > -            offset = i + 1;
> > -        else if (i - offset + 1 == size)
> > +
> > +    for (i = PCI_CONFIG_HEADER_SIZE; i < config_size; i += 4) {
> > +        if (pdev->used[i]) {
> > +            offset = i + 4;
> > +        } else if (i - offset + 1 == size) {
> 
> This test needs to change as well.  Looks like it should now be:
> 
>  (i - offset + 4 >= size)
> 
> Whereas we were previously calculating the difference from the offset to
> the current pointer plus the current unused byte, we're now assuming the
> current dword is empty because we're only handing out dword aligned
> offsets and it would be broken for something to not mark the first entry
> used.  Probably worthwhile to also add a comment noting the PCI spec
> requires dword alignment for capabilities.  Thanks,

BTW, rather than assume the rest of the dword is empty, we could just
check each dword instead of each byte, something like

uint32_t *dword_used = &pdev->used[PCI_CONFIG_HEADER_SIZE];

for (i = PCI_CONFIG_HEADER_SIZE; i < config_size; i +=4, dword_used++) {
    if (*dword_used) {
        offset = i + 4;
    } else if (i - offset + 4 >= size) {
        return offset;
    }
}

It also occurs to me that this function is broken for PCIe devices as we
should stop at PCI_CONFIG_SPACE_SIZE instead of config_size.  There
should be a separate allocator for extended config space, or a flag to
this function to indicate standard or extended.  Thanks,

Alex

> >              return offset;
> > +        }
> > +    }
> > +
> >      return 0;
> >  }
> >  
> 
>