On Tue, Mar 31, 2015 at 07:57:46PM -0700, Yinghai Lu wrote:
> Fix regression by commit d63e2e1f3df9 ("sparc/PCI: Clip bridge windows to
> fit in upstream windows").
>
> That cause bridge bar get clipped wrongly.
>
> The sparc64 dma_addr_t is 32-bit, we can not use it to check if we can
> use 64bit bar, introduce pci_bus_addr_t.
>
> sparc ofpci does not parse 64bit mem for root bus, add code to
> make sure that we get correct resource for root.
>
> Also there are device 64-bit res does not have pref bit flag, but bridges
> do have pref bit set, that cause extra "no compatible window".
> Set pref bit for them according to the errata.
Ping, any updates on this? AFAICT, we should add a new pci_bus_addr_t
distinct from dma_addr_t, but Ben raised an issue that needs to be
resolved first.
Bjorn
On Sat, May 16, 2015 at 10:25:37AM -0500, Bjorn Helgaas wrote:
> On Tue, Mar 31, 2015 at 07:57:46PM -0700, Yinghai Lu wrote:
> > Fix regression by commit d63e2e1f3df9 ("sparc/PCI: Clip bridge windows to
> > fit in upstream windows").
> >
> > That cause bridge bar get clipped wrongly.
> >
> > The sparc64 dma_addr_t is 32-bit, we can not use it to check if we can
> > use 64bit bar, introduce pci_bus_addr_t.
> >
> > sparc ofpci does not parse 64bit mem for root bus, add code to
> > make sure that we get correct resource for root.
> >
> > Also there are device 64-bit res does not have pref bit flag, but bridges
> > do have pref bit set, that cause extra "no compatible window".
> > Set pref bit for them according to the errata.
>
> Ping, any updates on this? AFAICT, we should add a new pci_bus_addr_t
> distinct from dma_addr_t, but Ben raised an issue that needs to be
> resolved first.
BTW, if it makes sense to add pci_bus_addr_t first, without the other
sparc64/prefetchability changes, that's fine with me. But I think there
were some doc and maybe Kconfig updates related to that, so please fold
those in and repost it.
Bjorn