Date:   Mon, 28 Jun 2021 10:34:20 +0200
From:   Andrew Jones <drjones@...hat.com>
To:     Gavin Shan <gshan@...hat.com>
Cc:     devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        rdunlap@...radead.org, robh+dt@...nel.org, shan.gavin@...il.com
Subject: Re: [PATCH v4] Documentation, dt, numa: Add note to empty NUMA node

On Fri, Jun 25, 2021 at 09:14:49PM +1000, Gavin Shan wrote:
> On 6/25/21 8:18 PM, Andrew Jones wrote:
> > On Fri, Jun 25, 2021 at 06:36:48PM +1000, Gavin Shan wrote:
> > > On 6/25/21 5:06 PM, Andrew Jones wrote:
> > > > On Fri, Jun 25, 2021 at 09:02:17AM +0200, Andrew Jones wrote:
> > > > > On Fri, Jun 25, 2021 at 01:23:38PM +0800, Gavin Shan wrote:
> > > > > > The empty memory nodes, where no memory resides in, are allowed.
> > > > > > For these empty memory nodes, the 'len' of 'reg' property is zero.
> > > > > > The NUMA node IDs are still valid and parsed, but memory can be
> > > > > > added to them through hotplug afterwards. I finds difficulty to
> > > > > > get where it's properly documented.
> > > > > > 
> > > > > > So lets add note to empty memory nodes in the NUMA binding doc.
> > > > > > 
> > > > > > Signed-off-by: Gavin Shan <gshan@...hat.com>
> > > > > > ---
> > > > > >    Documentation/devicetree/bindings/numa.txt | 4 ++++
> > > > > >    1 file changed, 4 insertions(+)
> > > > > > 
> > > > > > diff --git a/Documentation/devicetree/bindings/numa.txt b/Documentation/devicetree/bindings/numa.txt
> > > > > > index 21b35053ca5a..0fc882e44270 100644
> > > > > > --- a/Documentation/devicetree/bindings/numa.txt
> > > > > > +++ b/Documentation/devicetree/bindings/numa.txt
> > > > > > @@ -109,6 +109,10 @@ Example:
> > > > > >    Dual socket system consists of 2 boards connected through ccn bus and
> > > > > >    each board having one socket/soc of 8 cpus, memory and pci bus.
> > > > > > +Note that empty memory nodes, which no memory resides in, are allowed.
> > > > > > +The NUMA node IDs in these empty memory nodes are still valid, but
> > > > > > +memory can be added into them through hotplug afterwards.
> > > > > 
> > > > > Please change the second sentence to:
> > > > > 
> > > > >     The NUMA node IDs in these empty memory nodes are still valid and
> > > > >     memory may be added into them through hotplug afterwards.
> > > > > 
> > > > > But, this doesn't look like the right place for this paragraph. You're
> > > > > adding the paragraph to the example section, but the example doesn't have
> > > > > any empty memory nodes.
> > > > > 
> > > > > I think the paragraph should be added to section "2 - numa-node-id" and an
> > > > 
> > > > Or maybe even create a new section for it.
> > > > 
> > > > > example empty memory node should be provided. Also, the commit message
> > > > > talks about the length of 'reg' being zero, which is an important
> > > > > distinction which should also be documented.
> > > > > 
> > > 
> > > Drew, thanks for your comments. Yeah, it sounds sensible to create
> > > a new section for it and an example would be more helpful. Please
> > > check if below changes are fine to you. I probably need Randy's review
> > > again.
> > > 
> > > I'm trying to avoid too many revisions for this sort of trivial patch,
> > > even though I already had. However, it's time frame for v5.14 and I'm
> > > pushing this to be merged during the cycle.
> > 
> > We must revise until we're satisfied... Also, I wouldn't call writing
> > specifications trivial.
> > 
> 
> [...]
> 
> > I'll just rewrite it:
> > 
> >   Empty memory nodes, which no memory resides in, are allowed. The 'length'
> >   field of the 'reg' property is zero, but the 'base-address' is a dummy
> >   address and invalid. However, the NUMA node IDs and distance maps are
> >   still valid and memory may be added into them through hotplug afterwards.
> > 
> 
> [...]
> 
> > 
> > You should write how you've selected the dummy addresses for the empty
> > memory nodes, perhaps here in the /* Empty memory node */ comments or
> > above in the paragraph or both.
> > 
> > > +	memory@...00000000 {
> > > +		device_type = "memory";
> > > +		reg = <0x1 0x0 0x0 0x0>;
> > > +		numa-node-id = <2>;
> > > +	};
> > > +
> > > +	/* Empty memory node */
> > > +	memory@...80000000 {
> > > +		device_type = "memory";
> > > +		reg = <0x1 0x80000000 0x0 0x0>;
> > > +		numa-node-id = <3>;
> > > +	};
> > > +
> 
> [...]
> 
> Thanks, Drew. As 'base-address' is dummy address and invalid. So we
> would note it as suggestion. With your suggestion integrated, how about
> something like this?
> 
> ==============================================================================
> 4 - Empty memory nodes
> ==============================================================================
> 
> Empty memory nodes, which no memory resides in, are allowed. The 'length'
> field of the 'reg' property is zero, but the 'base-address' is a dummy
> address and invalid. The 'base-address' could be the summation of highest
> memory address plus the NUMA node ID. However, the NUMA node IDs and
> distance maps are still valid and memory may be added into them through
> hotplug afterwards.
> 
>        :
> 
>        memory@0 {
>                 device_type = "memory";
>                 reg = <0x0 0x0 0x0 0x80000000>;
>                 numa-node-id = <0>;
>         };
> 
>         memory@...0000000 {
>                 device_type = "memory";
>                 reg = <0x0 0x80000000 0x0 0x80000000>;
>                 numa-node-id = <1>;
>         };
> 
>         /* Empty memory node */
>         memory@...00000002 {
>                 device_type = "memory";
>                 reg = <0x1 0x2 0x0 0x0>;
>                 numa-node-id = <2>;
>         };
> 
>         /* Empty memory node */
>         memory@...00000003 {
>                 device_type = "memory";
>                 reg = <0x1 0x3 0x0 0x0>;
>                 numa-node-id = <3>;
>         };
> 
>         :
> 
> Thanks,
> Gavin
> 
>

Looks good to me.

Thanks,
drew 

Powered by blists - more mailing lists