Message-ID: <20200821165440.GA1683523@bjorn-Precision-5520>
Date:   Fri, 21 Aug 2020 11:54:40 -0500
From:   Bjorn Helgaas <helgaas@...nel.org>
To:     Jonathan Cameron <Jonathan.Cameron@...wei.com>
Cc:     linux-mm@...ck.org, linux-acpi@...r.kernel.org,
        linux-arm-kernel@...ts.infradead.org, x86@...nel.org,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        Bjorn Helgaas <bhelgaas@...gle.com>, rafael@...nel.org,
        linux-kernel@...r.kernel.org, Ingo Molnar <mingo@...hat.com>,
        Thomas Gleixner <tglx@...utronix.de>, linuxarm@...wei.com,
        Dan Williams <dan.j.williams@...el.com>,
        Brice Goglin <Brice.Goglin@...ia.fr>,
        Sean V Kelley <sean.v.kelley@...ux.intel.com>,
        linux-api@...r.kernel.org, Keith Busch <kbusch@...nel.org>
Subject: Re: [PATCH v9 4/6] ACPI: HMAT: Fix handling of changes from ACPI 6.2
 to ACPI 6.3

On Fri, Aug 21, 2020 at 05:37:18PM +0100, Jonathan Cameron wrote:
> On Fri, 21 Aug 2020 08:46:22 -0500
> Bjorn Helgaas <helgaas@...nel.org> wrote:
> 
> > On Fri, Aug 21, 2020 at 01:59:01PM +0100, Jonathan Cameron wrote:
> > > On Fri, 21 Aug 2020 07:13:56 -0500
> > > Bjorn Helgaas <helgaas@...nel.org> wrote:
> > >   
> > > > [+cc Keith, author of 3accf7ae37a9 ("acpi/hmat: Parse and report
> > > > heterogeneous memory")]
> > > > 
> > > > On Fri, Aug 21, 2020 at 09:42:58AM +0100, Jonathan Cameron wrote:  
> > > > > On Thu, 20 Aug 2020 17:21:29 -0500
> > > > > Bjorn Helgaas <helgaas@...nel.org> wrote:
> > > > >     
> > > > > > On Wed, Aug 19, 2020 at 10:51:09PM +0800, Jonathan Cameron wrote:    
> > > > > > > In ACPI 6.3, the Memory Proximity Domain Attributes Structure
> > > > > > > changed substantially.  One of those changes was that the flag
> > > > > > > for "Memory Proximity Domain field is valid" was deprecated.
> > > > > > > 
> > > > > > > This was because the field "Proximity Domain for the Memory"
> > > > > > > became a required field and hence having a validity flag makes
> > > > > > > no sense.
> > > > > > > 
> > > > > > > So the correct logic is to always assume the field is there.
> > > > > > > Current code assumes it never is.
> > > > > > > 
> > > > > > > Signed-off-by: Jonathan Cameron <Jonathan.Cameron@...wei.com>
> > > > > > > ---
> > > > > > >  drivers/acpi/numa/hmat.c | 2 +-
> > > > > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > > > 
> > > > > > > diff --git a/drivers/acpi/numa/hmat.c b/drivers/acpi/numa/hmat.c
> > > > > > > index 2c32cfb72370..07cfe50136e0 100644
> > > > > > > --- a/drivers/acpi/numa/hmat.c
> > > > > > > +++ b/drivers/acpi/numa/hmat.c
> > > > > > > @@ -424,7 +424,7 @@ static int __init hmat_parse_proximity_domain(union acpi_subtable_headers *heade
> > > > > > >  		pr_info("HMAT: Memory Flags:%04x Processor Domain:%u Memory Domain:%u\n",
> > > > > > >  			p->flags, p->processor_PD, p->memory_PD);
> > > > > > >  
> > > > > > > -	if (p->flags & ACPI_HMAT_MEMORY_PD_VALID && hmat_revision == 1) {
> > > > > > > +	if ((p->flags & ACPI_HMAT_MEMORY_PD_VALID && hmat_revision == 1) || hmat_revision == 2) {      
> > > > > > 
> > > > > > I hope/assume the spec is written in such a way that p->memory_PD is
> > > > > > required for any revision > 1?  So maybe this should be:
> > > > > > 
> > > > > >   if ((p->flags & ACPI_HMAT_MEMORY_PD_VALID && hmat_revision == 1) ||
> > > > > >       hmat_revision > 1) {    
> > > > 
> > > > I should have said simply:
> > > > 
> > > >   if (hmat_revision == 1 && p->flags & ACPI_HMAT_MEMORY_PD_VALID)
> > > > 
> > > > We shouldn't even test p->flags for ACPI_HMAT_MEMORY_PD_VALID unless
> > > > we already know it's revision 1.
> > > > 
> > > > And unless there was a revision 0 of HMAT, there's no need to look for
> > > > hmat_revison > 1.  
> > > 
> > > It needs to stay as an or statement as you had the first time.
> > > The field is always valid for hmat_revision > 1, and valid for
> > > hmat_revision == 1 with the flag set.  You could express it as
> > > 
> > > if ((p->flags & ACPI_HMAT_MEMORY_PD_VALID) || (hmat_revision != 1))
> > > 
> > > but that seems more confusing to me.  
> > 
> > Oh, you're right, sorry!  There are two questions here:
> > 
> > 1) In what order should we test "p->flags & ACPI_HMAT_MEMORY_PD_VALID"
> >    and "hmat_revision == 1"?  ACPI_HMAT_MEMORY_PD_VALID is defined
> >    only when "hmat_revision == 1", so I think we should test the
> >    revision first.
> > 
> >    When "hmat_revision == 2", ACPI_HMAT_MEMORY_PD_VALID is reserved,
> >    so we shouldn't test it, even if we later check the revision and
> >    discard the result of the flag test.  This is a tiny thing,
> >    admittedly, but I think it follows the spec more clearly.
> 
> Agreed.
> 
> > 
> > 2) Do we need to test hmat_revision for anything other than 1?  Yes,
> >    you're right, see below.
> > 
> > > > > Good point.  We have existing protections elsewhere against
> > > > > hmat_revision being anything other than 1 or 2, so we should aim to
> > > > > keep that in only one place.    
> > > > 
> > > > I think the "Ignoring HMAT: Unknown revision" test in hmat_init(),
> > > > added by 3accf7ae37a9 ("acpi/hmat: Parse and report heterogeneous
> > > > memory"), is a mistake.
> > > > 
> > > > And I think hmat_normalize() has a similar mistake in that it tests
> > > > explicitly for hmat_revision == 2 when it should accept 2 AND anything
> > > > later.
> > > > 
> > > > We should assume that future spec revisions will be backwards
> > > > compatible.  Otherwise we're forced to make kernel changes when we
> > > > otherwise would not have to.  
> > > 
> > > I disagree with this. There is no rule in ACPI about maintaining
> > > backwards compatibility. The assumption is that the version number
> > > will always be checked.  The meaning of fields changed between
> > > version 1 and version 2 so it would be bold to assume that won't
> > > happen in the future!  
> > 
> > There *is* a rule about maintaining backwards compatibility.  ACPI
> > v6.3, sec 5.2.2, says:
> > 
> >   All versions of the ACPI tables must maintain backward
> >   compatibility. To accomplish this, modifications of the tables
> >   consist of redefinition of previously reserved fields and values
> >   plus appending data to the 1.0 tables. Modifications of the ACPI
> >   tables require that the version numbers of the modified tables be
> >   incremented.
> 
> Fair point.  Unfortunately it's not true here...  The field we
> are talking about here is probably fine, but the latency units
> changed between v1 and v2.  

Oops.  Sounds like this should have been done in a way that didn't
break old kernels reading new tables.  It's OK if old kernels can't
use new features, but not OK if things that used to work are broken
by new tables.

Bjorn

Powered by blists - more mailing lists