On Tue, Jan 07, 2014 at 01:25:37PM +0800, Mark Zhang wrote:
> On 12/05/2013 08:25 PM, Hiroshi Doyu wrote:
[...]
> > @@ -526,6 +530,21 @@ static int smmu_setup_regs(struct smmu_device *smmu)
> > return 0;
> > }
> >
> > +static void flush_ptc_by_addr(struct smmu_device *smmu, unsigned long *pte,
> > + struct page *page)
> > +{
> > + u32 val;
> > +
> > + val = VA_PAGE_TO_PA_HI(pte, page);
> > + if (val)
> > + smmu_write(smmu, val, SMMU_PTC_FLUSH_1);
> > +
>
> This is not correct, according to my tests. We should write
> "SMMU_PTC_FLUSH_1" even when the "val" is zero.
>
> So I just copied Pavan's original work here, after applied this, the
> SMMU works correctly:
>
> - val = VA_PAGE_TO_PA_HI(pte, page);
> - if (val)
> + if (!pte) {
> + smmu_write(smmu, SMMU_PTC_FLUSH_TYPE_ALL, SMMU_PTC_FLUSH);
> + return;
> + }
> +
> + if (of_machine_is_compatible("nvidia,tegra124")) {
This check should be replaced by some flag so we don't have to compare
strings every time the PTC is flushed.
Thierry
Attachment:
pgp6AySeGGKpU.pgp
Description: PGP signature
