On Fri, Oct 24, 2014 at 10:45:46PM +0300, Konstantin Belousov wrote:
> On Fri, Oct 24, 2014 at 07:33:07PM +0000, Rui Paulo wrote:
> > On Oct 24, 2014, at 12:20 PM, Konstantin Belousov <kostik...@gmail.com>
> > wrote:
> >
> > > +static int
> > > +hpet_mmap(struct cdev *cdev, vm_ooffset_t offset, vm_paddr_t *paddr,
> > > + int nprot, vm_memattr_t *memattr)
> > > +{
> > > + struct hpet_softc *sc;
> > > +
> > > + sc = cdev->si_drv1;
> > > + if (offset > rman_get_size(sc->mem_res))
> > > + return (EINVAL);
> > > + if (!sc->mmap_allow_write && (nprot & PROT_WRITE))
> > > + return (EPERM);
> > > + *paddr = rman_get_start(sc->mem_res) + offset;
> > What is the memattr for the backing page ? Is it set to non-cached
> > mode somehow ? I was not able to find place where would this happen.
> >
> > I expect it to be set to non-cached since it's a device anyway, but I don't
> > know where it is. During my testing, I did not see any problems with
> > cached values, though.
> >
> I am not claiming that it is wrong, only that I do not see an easy reason
> why it is right. Just printing the *memattr would provide the confidence.
>
Ok, I did looked at the pte of the HPET page. I get the value
0x80000000fed00025
which coincides with the resource address 0xfed00000 reported
by devinfo -vr for hpet (to double-check my findings). The low
bits indicate that PAT0 pat entry is used for the page caching
mode. Corresponding PAT MSR 0x277 has the following value:
sandy% sudo cpucontrol -m 0x277 /dev/cpuctl0
MSR 0x277: 0x00010506 0x00070406
i.e. memory type is 6, which is write-back, according to SDM.
This is wrong, as I feared.
The patch below fixes the issue. The pte for HPET page is equal to
0x80000000fed0003d
after the patch is applied, PAT3 is used, and its value is 0 ==
UNCACHEABLE, as it must be.
Do you agree with the patch ?
diff --git a/sys/dev/acpica/acpi_hpet.c b/sys/dev/acpica/acpi_hpet.c
index 6b35f5c..0da8bae 100644
--- a/sys/dev/acpica/acpi_hpet.c
+++ b/sys/dev/acpica/acpi_hpet.c
@@ -356,6 +356,7 @@ hpet_mmap(struct cdev *cdev, vm_ooffset_t offset,
vm_paddr_t *paddr,
if (!sc->mmap_allow_write && (nprot & PROT_WRITE))
return (EPERM);
*paddr = rman_get_start(sc->mem_res) + offset;
+ *memattr = VM_MEMATTR_UNCACHEABLE;
return (0);
}
_______________________________________________
svn-src-head@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
