MacBookPro 5,1

[Jung-uk Kim]

On Wednesday 03 November 2010 12:47 pm, John Baldwin wrote:
> On Wednesday, November 03, 2010 12:25:37 pm Jung-uk Kim wrote:
> > On Wednesday 03 November 2010 08:28 am, John Baldwin wrote:
> > > On Tuesday, November 02, 2010 6:32:12 pm Jung-uk Kim wrote:
> > > > On Tuesday 02 November 2010 05:26 pm, John Baldwin wrote:
> > > > > On Tuesday, November 02, 2010 4:50:18 pm Jung-uk Kim wrote:
> > > > > > On Tuesday 02 November 2010 04:24 pm, John Baldwin wrote:
> > > > > > > On Tuesday, November 02, 2010 4:14:05 pm Jung-uk Kim 
wrote:
> > > > > > > > On Tuesday 02 November 2010 03:41 pm, John Baldwin 
wrote:
> > > > > > > > > On Tuesday, November 02, 2010 3:29:01 pm Jung-uk
> > > > > > > > > Kim
> >
> > wrote:
> > > > > > > > > > On Tuesday 02 November 2010 11:29 am, Andriy
> > > > > > > > > > Gapon
> >
> > wrote:
> > > > > > > > > > > on 29/10/2010 08:51 Andriy Gapon said the 
following:
> > > > > > > > > > > > I guess that a general problem here is that
> > > > > > > > > > > > it is incorrect to merely use memcpy/bcopy to
> > > > > > > > > > > > create a copy of a resource if the resource
> > > > > > > > > > > > has ACPI_RESOURCE_SOURCE field in it.
> > > > > > > > > > >
> > > > > > > > > > > Hans,
> > > > > > > > > > >
> > > > > > > > > > > could you please test the following patch?
> > > > > > > > > > >
> > > > > > > > > > > diff --git a/sys/dev/acpica/acpi_pci_link.c
> > > > > > > > > > > b/sys/dev/acpica/acpi_pci_link.c index
> > > > > > > > > > > dcf101d..e842635 100644 ---
> > > > > > > > > > > a/sys/dev/acpica/acpi_pci_link.c +++
> > > > > > > > > > > b/sys/dev/acpica/acpi_pci_link.c
> > > > > > > > > > > @@ -767,6 +767,8 @@ acpi_pci_link_srs_from_crs
> > > > > > > > > > >  				    link->l_irq;
> > > > > > > > > > >  			else
> > > > > > > > > > >  				resptr->Data.ExtendedIrq.Interrupts[0] =
> > > > > > > > > > > 0;
> > > > > > > > > > > +			memset(&resptr->Data.ExtendedIrq.ResourceSo
> > > > > > > > > > >urce , 0, +			    sizeof(ACPI_RESOURCE_SOURCE));
> > > > > > > > > > > link++;
> > > > > > > > > > >  			i++;
> > > > > > > > > > >  			break;
> > > > > > > > > >
> > > > > > > > > > Hmm...  Very interesting.  Can you please try
> > > > > > > > > > this, too?
> > > > > > > > >
> > > > > > > > > Linux doesn't set the resource source bits up at
> > > > > > > > > all when doing _SRS, so I'd rather just do that.  I
> > > > > > > > > think what I'd prefer is that we not use the
> > > > > > > > > prs_template, perhaps just save the type of the
> > > > > > > > > resource and build a new resource object from
> > > > > > > > > scratch where the resource is zero'd, the
> > > > > > > > > appropriate bits are set and then that resource is
> > > > > > > > > appended to the buffer being built.
> > > > > > > >
> > > > > > > > "Linux doesn't do it" is wrong if I am reading the
> > > > > > > > spec. correctly, i.e., _CRS, _PRS and _SRS must have
> > > > > > > > the same format and size.
> > > > > > >
> > > > > > > Umm, but we aren't setting up the raw bits for _SRS. 
> > > > > > > We are creating a list of ACPI_RESOURCE objects that
> > > > > > > ACPICA then encodes into a buffer to send to _SRS.
> > > > > >
> > > > > > Yes, I understand.  However, ACPICA is expecting the same
> > > > > > size of buffer *including* the optional parts if I am
> > > > > > reading the code right. Besides, I don't think there is
> > > > > > any harm in doing the right thing. ;-)
> > > > >
> > > > > To be clear, I am suggesting to take an ACPI_RESOURCE
> > > > > object, bzero it, then set the type and the IRQ and that's
> > > > > it.  Leave the ResourceSource bits as zero.  The size will
> > > > > still be set based on the actual type (or if needed we can
> > > > > use the cached size from the template copy we save from
> > > > > _PRS).  The point would be to start from a zero structure
> > > > > instead of from a copy of what we got from _PRS.
> > > >
> > > > It may work if we don't use l_prs_template.
> > >
> > > Well, we still need much of the info from the _PRS resource
> > > (the type, etc.), but I think we should not blindly use the
> > > template directly when building the buffer for _SRS.
> >
> > Actually, I think we should get the information directly from
> > _CRS as ACPI spec. is suggesting.
>
> I would be fine with that, but that does not work if _CRS doesn't
> work (the acpi_pci_link_srs_from_links() case).

For that case, we must use the template, of course.  In fact, my patch 
is more useful for this particular case. :-)

> Are we allowed to modify the buffer ACPICA gives us from _CRS and
> then pass that back to _SRS?

I believe so.  If we go with that route, we don't have to worry about 
ResourceSource.StringPtr or acpi_AppendBufferResource() copying stale 
pointers.

Jung-uk Kim