Re: [PATCH v2 6/6] ACPI: HMAT: Attach a device for each soft-reserved range

[Dan Williams]

On Wed, Mar 25, 2020 at 4:10 AM Will Deacon <will@xxxxxxxxxx> wrote:
>
> On Sun, Mar 22, 2020 at 09:12:58AM -0700, Dan Williams wrote:
> > The hmem enabling in commit 'cf8741ac57ed ("ACPI: NUMA: HMAT: Register
> > "soft reserved" memory as an "hmem" device")' only registered ranges to
> > the hmem driver for each soft-reservation that also appeared in the
> > HMAT. While this is meant to encourage platform firmware to "do the
> > right thing" and publish an HMAT, the corollary is that platforms that
> > fail to publish an accurate HMAT will strand memory from Linux usage.
> > Additionally, the "efi_fake_mem" kernel command line option enabling
> > will strand memory by default without an HMAT.
> >
> > Arrange for "soft reserved" memory that goes unclaimed by HMAT entries
> > to be published as raw resource ranges for the hmem driver to consume.
> >
> > Include a module parameter to disable either this fallback behavior, or
> > the hmat enabling from creating hmem devices. The module parameter
> > requires the hmem device enabling to have unique name in the module
> > namespace: "device_hmem".
> >
> > Rather than mark this x86-only, include an interim phys_to_target_node()
> > implementation for arm64.
> >
> > Cc: Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx>
> > Cc: Brice Goglin <Brice.Goglin@xxxxxxxx>
> > Cc: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
> > Cc: "Rafael J. Wysocki" <rjw@xxxxxxxxxxxxx>
> > Cc: Jeff Moyer <jmoyer@xxxxxxxxxx>
> > Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> > Cc: Will Deacon <will@xxxxxxxxxx>
> > Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > ---
> >  arch/arm64/mm/numa.c      |   13 +++++++++++++
> >  drivers/dax/Kconfig       |    1 +
> >  drivers/dax/hmem/Makefile |    3 ++-
> >  drivers/dax/hmem/device.c |   33 +++++++++++++++++++++++++++++++++
> >  4 files changed, 49 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/arm64/mm/numa.c b/arch/arm64/mm/numa.c
> > index 4decf1659700..00fba21eaec0 100644
> > --- a/arch/arm64/mm/numa.c
> > +++ b/arch/arm64/mm/numa.c
> > @@ -468,3 +468,16 @@ int memory_add_physaddr_to_nid(u64 addr)
> >       pr_warn("Unknown node for memory at 0x%llx, assuming node 0\n", addr);
> >       return 0;
> >  }
> > +
> > +/*
> > + * device-dax instance registrations want a valid target-node in case
> > + * they are ever onlined as memory (see hmem_register_device()).
> > + *
> > + * TODO: consult cached numa info
> > + */
> > +int phys_to_target_node(phys_addr_t addr)
> > +{
> > +     pr_warn_once("Unknown target node for memory at 0x%llx, assuming node 0\n",
> > +                     addr);
> > +     return 0;
> > +}
>
> Could you implement a generic version of this by iterating over the nodes
> with for_each_{,online_}node() and checking for intersection with
> node_{start,end}_pfn()?

Interesting. The gap is that node_{start,end}_pfn() requires
node_data[] which to date architectures have only setup for online
nodes. Recall a target node is an offline node that could come online
later. However, reworking offline data into node_data could be the
local solution for arm64, I'd just ask that each of the 6
memory-hotplug capable archs go make that opt-in decision themselves.