[PATCH v4 0/6] Memory Hierarchy: Enable target node lookups for reserved memory

[PATCH v4 0/6] Memory Hierarchy: Enable target node lookups for reserved memory

[Dan Williams]

Changes since v3 [1]:
- Cleanup numa_map_to_online_node() to remove redundant "if
  (!node_online(node))" (Aneesh)

[1]: http://lore.kernel.org/r/157954696789.2239526.17707265517154476652.stgit@dwillia2-desk3.amr.corp.intel.com

---

Merge notes:

x86 folks: This has an ack from Rafael for ACPI, and Michael for Power.
With an x86 ack I plan to take this through the libnvdimm tree provided
the x86 touches look ok to you.

---

Cover:

Arrange for platform numa info to be preserved for determining
'target_node' data. Where a 'target_node' is the node a reserved memory
range will become when it is onlined.

This new infrastructure is expected to be more valuable over time for
Memory Tiers / Hierarchy management as more platforms (via the ACPI HMAT
and EFI Specific Purpose Memory) publish reserved or "soft-reserved"
ranges to Linux. Linux system administrators will expect to be able to
interact with those ranges with a unique numa node number when/if that
memory is onlined via the dax_kmem driver [2].

One configuration that currently fails to properly convey the target
node for the resulting memory hotplug operation is persistent memory
defined by the memmap=nn!ss parameter. For example, today if node1 is a
memory only node, and all the memory from node1 is specified to
memmap=nn!ss and subsequently onlined, it will end up being onlined as
node0 memory. As it stands, memory_add_physaddr_to_nid() can only
identify online nodes and since node1 in this example has no online cpus
/ memory the target node is initialized node0.

The fix is to preserve rather than discard the numa_meminfo entries that
are relevant for reserved memory ranges, and to uplevel the node
distance helper for determining the "local" (closest) node relative to
an initiator node.

[2]: https://pmem.io/ndctl/daxctl-reconfigure-device.html

---

Dan Williams (6):
      ACPI: NUMA: Up-level "map to online node" functionality
      mm/numa: Skip NUMA_NO_NODE and online nodes in numa_map_to_online_node()
      powerpc/papr_scm: Switch to numa_map_to_online_node()
      x86/mm: Introduce CONFIG_KEEP_NUMA
      x86/numa: Provide a range-to-target_node lookup facility
      libnvdimm/e820: Retrieve and populate correct 'target_node' info


 arch/powerpc/platforms/pseries/papr_scm.c |   21 --------
 arch/x86/Kconfig                          |    1 
 arch/x86/mm/numa.c                        |   74 +++++++++++++++++++++++------
 drivers/acpi/numa/srat.c                  |   41 ----------------
 drivers/nvdimm/e820.c                     |   18 ++-----
 include/linux/acpi.h                      |   23 +++++++++
 include/linux/numa.h                      |   23 +++++++++
 mm/Kconfig                                |    5 ++
 mm/mempolicy.c                            |   31 ++++++++++++
 9 files changed, 145 insertions(+), 92 deletions(-)
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v4 1/6] ACPI: NUMA: Up-level "map to online node" functionality

[Dan Williams]

The acpi_map_pxm_to_online_node() helper is used to find the closest
online node to a given proximity domain. This is used to map devices in
a proximity domain with no online memory or cpus to the closest online
node and populate a device's 'numa_node' property. The numa_node
property allows applications to be migrated "close" to a resource.

In preparation for providing a generic facility to optionally map an
address range to its closest online node, or the node the range would
represent were it to be onlined (target_node), up-level the core of
acpi_map_pxm_to_online_node() to a generic mm/numa helper.

Cc: Michal Hocko <mhocko@suse.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/acpi/numa/srat.c |   41 -----------------------------------------
 include/linux/acpi.h     |   23 ++++++++++++++++++++++-
 include/linux/numa.h     |    9 +++++++++
 mm/mempolicy.c           |   30 ++++++++++++++++++++++++++++++
 4 files changed, 61 insertions(+), 42 deletions(-)

diff --git a/drivers/acpi/numa/srat.c b/drivers/acpi/numa/srat.c
index eadbf90e65d1..47b4969d9b93 100644
--- a/drivers/acpi/numa/srat.c
+++ b/drivers/acpi/numa/srat.c
@@ -72,47 +72,6 @@ int acpi_map_pxm_to_node(int pxm)
 }
 EXPORT_SYMBOL(acpi_map_pxm_to_node);
 
-/**
- * acpi_map_pxm_to_online_node - Map proximity ID to online node
- * @pxm: ACPI proximity ID
- *
- * This is similar to acpi_map_pxm_to_node(), but always returns an online
- * node.  When the mapped node from a given proximity ID is offline, it
- * looks up the node distance table and returns the nearest online node.
- *
- * ACPI device drivers, which are called after the NUMA initialization has
- * completed in the kernel, can call this interface to obtain their device
- * NUMA topology from ACPI tables.  Such drivers do not have to deal with
- * offline nodes.  A node may be offline when a device proximity ID is
- * unique, SRAT memory entry does not exist, or NUMA is disabled, ex.
- * "numa=off" on x86.
- */
-int acpi_map_pxm_to_online_node(int pxm)
-{
-	int node, min_node;
-
-	node = acpi_map_pxm_to_node(pxm);
-
-	if (node == NUMA_NO_NODE)
-		node = 0;
-
-	min_node = node;
-	if (!node_online(node)) {
-		int min_dist = INT_MAX, dist, n;
-
-		for_each_online_node(n) {
-			dist = node_distance(node, n);
-			if (dist < min_dist) {
-				min_dist = dist;
-				min_node = n;
-			}
-		}
-	}
-
-	return min_node;
-}
-EXPORT_SYMBOL(acpi_map_pxm_to_online_node);
-
 static void __init
 acpi_table_print_srat_entry(struct acpi_subtable_header *header)
 {
diff --git a/include/linux/acpi.h b/include/linux/acpi.h
index 0f37a7d5fa77..69b73ecfbee4 100644
--- a/include/linux/acpi.h
+++ b/include/linux/acpi.h
@@ -401,9 +401,30 @@ extern void acpi_osi_setup(char *str);
 extern bool acpi_osi_is_win8(void);
 
 #ifdef CONFIG_ACPI_NUMA
-int acpi_map_pxm_to_online_node(int pxm);
 int acpi_map_pxm_to_node(int pxm);
 int acpi_get_node(acpi_handle handle);
+
+/**
+ * acpi_map_pxm_to_online_node - Map proximity ID to online node
+ * @pxm: ACPI proximity ID
+ *
+ * This is similar to acpi_map_pxm_to_node(), but always returns an online
+ * node.  When the mapped node from a given proximity ID is offline, it
+ * looks up the node distance table and returns the nearest online node.
+ *
+ * ACPI device drivers, which are called after the NUMA initialization has
+ * completed in the kernel, can call this interface to obtain their device
+ * NUMA topology from ACPI tables.  Such drivers do not have to deal with
+ * offline nodes.  A node may be offline when a device proximity ID is
+ * unique, SRAT memory entry does not exist, or NUMA is disabled, ex.
+ * "numa=off" on x86.
+ */
+static inline int acpi_map_pxm_to_online_node(int pxm)
+{
+	int node = acpi_map_pxm_to_node(pxm);
+
+	return numa_map_to_online_node(node);
+}
 #else
 static inline int acpi_map_pxm_to_online_node(int pxm)
 {
diff --git a/include/linux/numa.h b/include/linux/numa.h
index 110b0e5d0fb0..20f4e44b186c 100644
--- a/include/linux/numa.h
+++ b/include/linux/numa.h
@@ -13,4 +13,13 @@
 
 #define	NUMA_NO_NODE	(-1)
 
+#ifdef CONFIG_NUMA
+int numa_map_to_online_node(int node);
+#else
+static inline int numa_map_to_online_node(int node)
+{
+	return NUMA_NO_NODE;
+}
+#endif
+
 #endif /* _LINUX_NUMA_H */
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 067cf7d3daf5..4cff069279f6 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -127,6 +127,36 @@ static struct mempolicy default_policy = {
 
 static struct mempolicy preferred_node_policy[MAX_NUMNODES];
 
+/**
+ * numa_map_to_online_node - Find closest online node
+ * @nid: Node id to start the search
+ *
+ * Lookup the next closest node by distance if @nid is not online.
+ */
+int numa_map_to_online_node(int node)
+{
+	int min_node;
+
+	if (node == NUMA_NO_NODE)
+		node = 0;
+
+	min_node = node;
+	if (!node_online(node)) {
+		int min_dist = INT_MAX, dist, n;
+
+		for_each_online_node(n) {
+			dist = node_distance(node, n);
+			if (dist < min_dist) {
+				min_dist = dist;
+				min_node = n;
+			}
+		}
+	}
+
+	return min_node;
+}
+EXPORT_SYMBOL_GPL(numa_map_to_online_node);
+
 struct mempolicy *get_task_policy(struct task_struct *p)
 {
 	struct mempolicy *pol = p->mempolicy;
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v4 2/6] mm/numa: Skip NUMA_NO_NODE and online nodes in numa_map_to_online_node()

[Dan Williams]

Update numa_map_to_online_node() to stop falling back to numa node 0
when the input is NUMA_NO_NODE. Also, skip the lookup if @node is
online. This makes the routine compatible with other arch node mapping
routines.

Reported-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Link: https://lore.kernel.org/r/157401275716.43284.13185549705765009174.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 mm/mempolicy.c |   20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 4cff069279f6..bcb012645809 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -135,21 +135,17 @@ static struct mempolicy preferred_node_policy[MAX_NUMNODES];
  */
 int numa_map_to_online_node(int node)
 {
-	int min_node;
+	int min_dist = INT_MAX, dist, n, min_node;
 
-	if (node == NUMA_NO_NODE)
-		node = 0;
+	if (node == NUMA_NO_NODE || node_online(node))
+		return node;
 
 	min_node = node;
-	if (!node_online(node)) {
-		int min_dist = INT_MAX, dist, n;
-
-		for_each_online_node(n) {
-			dist = node_distance(node, n);
-			if (dist < min_dist) {
-				min_dist = dist;
-				min_node = n;
-			}
+	for_each_online_node(n) {
+		dist = node_distance(node, n);
+		if (dist < min_dist) {
+			min_dist = dist;
+			min_node = n;
 		}
 	}
 
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v4 3/6] powerpc/papr_scm: Switch to numa_map_to_online_node()

[Dan Williams]

Now that the core exports numa_map_to_online_node() switch to that
instead of the locally coded duplicate.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: "Oliver O'Halloran" <oohall@gmail.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Reported-by: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Link: https://lore.kernel.org/r/157401276263.43284.12616818803654229788.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/powerpc/platforms/pseries/papr_scm.c |   21 +--------------------
 1 file changed, 1 insertion(+), 20 deletions(-)

diff --git a/arch/powerpc/platforms/pseries/papr_scm.c b/arch/powerpc/platforms/pseries/papr_scm.c
index c2ef320ba1bf..057ed703e882 100644
--- a/arch/powerpc/platforms/pseries/papr_scm.c
+++ b/arch/powerpc/platforms/pseries/papr_scm.c
@@ -284,25 +284,6 @@ int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 	return 0;
 }
 
-static inline int papr_scm_node(int node)
-{
-	int min_dist = INT_MAX, dist;
-	int nid, min_node;
-
-	if ((node == NUMA_NO_NODE) || node_online(node))
-		return node;
-
-	min_node = first_online_node;
-	for_each_online_node(nid) {
-		dist = node_distance(node, nid);
-		if (dist < min_dist) {
-			min_dist = dist;
-			min_node = nid;
-		}
-	}
-	return min_node;
-}
-
 static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
 {
 	struct device *dev = &p->pdev->dev;
@@ -347,7 +328,7 @@ static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
 
 	memset(&ndr_desc, 0, sizeof(ndr_desc));
 	target_nid = dev_to_node(&p->pdev->dev);
-	online_nid = papr_scm_node(target_nid);
+	online_nid = numa_map_to_online_node(target_nid);
 	ndr_desc.numa_node = online_nid;
 	ndr_desc.target_node = target_nid;
 	ndr_desc.res = &p->res;
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v4 4/6] x86/mm: Introduce CONFIG_KEEP_NUMA

[Dan Williams]

Currently x86 numa_meminfo is marked __initdata in the
CONFIG_MEMORY_HOTPLUG=n case. In support of a new facility to allow
drivers to map reserved memory to a 'target_node'
(phys_to_target_node()), add support for removing the __initdata
designation for those users. Both memory hotplug and
phys_to_target_node() users select CONFIG_KEEP_NUMA to tell the arch to
maintain its physical address to numa mapping infrastructure post init.

Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <x86@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/x86/mm/numa.c   |    6 +-----
 include/linux/numa.h |    6 ++++++
 mm/Kconfig           |    5 +++++
 3 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index 99f7a68738f0..5289d9d6799a 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c
@@ -25,11 +25,7 @@ nodemask_t numa_nodes_parsed __initdata;
 struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
 EXPORT_SYMBOL(node_data);
 
-static struct numa_meminfo numa_meminfo
-#ifndef CONFIG_MEMORY_HOTPLUG
-__initdata
-#endif
-;
+static struct numa_meminfo numa_meminfo __initdata_numa;
 
 static int numa_distance_cnt;
 static u8 *numa_distance;
diff --git a/include/linux/numa.h b/include/linux/numa.h
index 20f4e44b186c..c005ed6b807b 100644
--- a/include/linux/numa.h
+++ b/include/linux/numa.h
@@ -13,6 +13,12 @@
 
 #define	NUMA_NO_NODE	(-1)
 
+#ifdef CONFIG_KEEP_NUMA
+#define __initdata_numa
+#else
+#define __initdata_numa __initdata
+#endif
+
 #ifdef CONFIG_NUMA
 int numa_map_to_online_node(int node);
 #else
diff --git a/mm/Kconfig b/mm/Kconfig
index ab80933be65f..001f1185eadf 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -139,6 +139,10 @@ config HAVE_FAST_GUP
 config ARCH_KEEP_MEMBLOCK
 	bool
 
+# Keep arch numa mapping infrastructure post-init.
+config KEEP_NUMA
+	bool
+
 config MEMORY_ISOLATION
 	bool
 
@@ -154,6 +158,7 @@ config MEMORY_HOTPLUG
 	bool "Allow for memory hot-add"
 	depends on SPARSEMEM || X86_64_ACPI_NUMA
 	depends on ARCH_ENABLE_MEMORY_HOTPLUG
+	select KEEP_NUMA if NUMA
 
 config MEMORY_HOTPLUG_SPARSE
 	def_bool y
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 4/6] x86/mm: Introduce CONFIG_KEEP_NUMA

[Ingo Molnar]

* Dan Williams <dan.j.williams@intel.com> wrote:

> Currently x86 numa_meminfo is marked __initdata in the
> CONFIG_MEMORY_HOTPLUG=n case. In support of a new facility to allow
> drivers to map reserved memory to a 'target_node'
> (phys_to_target_node()), add support for removing the __initdata
> designation for those users. Both memory hotplug and
> phys_to_target_node() users select CONFIG_KEEP_NUMA to tell the arch to
> maintain its physical address to numa mapping infrastructure post init.
> 
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Cc: Andy Lutomirski <luto@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Borislav Petkov <bp@alien8.de>
> Cc: "H. Peter Anvin" <hpa@zytor.com>
> Cc: <x86@kernel.org>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: David Hildenbrand <david@redhat.com>
> Cc: Michal Hocko <mhocko@suse.com>
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>  arch/x86/mm/numa.c   |    6 +-----
>  include/linux/numa.h |    6 ++++++
>  mm/Kconfig           |    5 +++++
>  3 files changed, 12 insertions(+), 5 deletions(-)

The concept and the x86 portions look sane, just a few minor nits:

> 
> diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
> index 99f7a68738f0..5289d9d6799a 100644
> --- a/arch/x86/mm/numa.c
> +++ b/arch/x86/mm/numa.c
> @@ -25,11 +25,7 @@ nodemask_t numa_nodes_parsed __initdata;
>  struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
>  EXPORT_SYMBOL(node_data);
>  
> -static struct numa_meminfo numa_meminfo
> -#ifndef CONFIG_MEMORY_HOTPLUG
> -__initdata
> -#endif
> -;
> +static struct numa_meminfo numa_meminfo __initdata_numa;
>  
>  static int numa_distance_cnt;
>  static u8 *numa_distance;
> diff --git a/include/linux/numa.h b/include/linux/numa.h
> index 20f4e44b186c..c005ed6b807b 100644
> --- a/include/linux/numa.h
> +++ b/include/linux/numa.h
> @@ -13,6 +13,12 @@
>  
>  #define	NUMA_NO_NODE	(-1)
>  
> +#ifdef CONFIG_KEEP_NUMA
> +#define __initdata_numa
> +#else
> +#define __initdata_numa __initdata
> +#endif
> +
>  #ifdef CONFIG_NUMA
>  int numa_map_to_online_node(int node);
>  #else
> diff --git a/mm/Kconfig b/mm/Kconfig
> index ab80933be65f..001f1185eadf 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -139,6 +139,10 @@ config HAVE_FAST_GUP
>  config ARCH_KEEP_MEMBLOCK
>  	bool
>  
> +# Keep arch numa mapping infrastructure post-init.

s/numa/NUMA

Please also capitalize consistently in the rest of the series.

> +config KEEP_NUMA
> +	bool


So most of our recent new NUMA options followed the naming pattern of:

  CONFIG_NUMA_*

Such as CONFIG_NUMA_BALANCING or CONFIG_NUMA_EMU.

So I'd suggesting naming it to CONFIG_NUMA_KEEP, or, a bit more 
descriptively, such as CONFIG_NUMA_KEEP_MAPPING or such?

'Keeping NUMA' is kind of lame - of course we keep NUMA. ;-)

Thanks,

	Ingo
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 4/6] x86/mm: Introduce CONFIG_KEEP_NUMA

[Dan Williams]

On Thu, Feb 13, 2020 at 1:32 AM Ingo Molnar <mingo@kernel.org> wrote:
>
>
> * Dan Williams <dan.j.williams@intel.com> wrote:
>
> > Currently x86 numa_meminfo is marked __initdata in the
> > CONFIG_MEMORY_HOTPLUG=n case. In support of a new facility to allow
> > drivers to map reserved memory to a 'target_node'
> > (phys_to_target_node()), add support for removing the __initdata
> > designation for those users. Both memory hotplug and
> > phys_to_target_node() users select CONFIG_KEEP_NUMA to tell the arch to
> > maintain its physical address to numa mapping infrastructure post init.
> >
> > Cc: Dave Hansen <dave.hansen@linux.intel.com>
> > Cc: Andy Lutomirski <luto@kernel.org>
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Cc: Thomas Gleixner <tglx@linutronix.de>
> > Cc: Ingo Molnar <mingo@redhat.com>
> > Cc: Borislav Petkov <bp@alien8.de>
> > Cc: "H. Peter Anvin" <hpa@zytor.com>
> > Cc: <x86@kernel.org>
> > Cc: Andrew Morton <akpm@linux-foundation.org>
> > Cc: David Hildenbrand <david@redhat.com>
> > Cc: Michal Hocko <mhocko@suse.com>
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > ---
> >  arch/x86/mm/numa.c   |    6 +-----
> >  include/linux/numa.h |    6 ++++++
> >  mm/Kconfig           |    5 +++++
> >  3 files changed, 12 insertions(+), 5 deletions(-)
>
> The concept and the x86 portions look sane, just a few minor nits:
>
> >
> > diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
> > index 99f7a68738f0..5289d9d6799a 100644
> > --- a/arch/x86/mm/numa.c
> > +++ b/arch/x86/mm/numa.c
> > @@ -25,11 +25,7 @@ nodemask_t numa_nodes_parsed __initdata;
> >  struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
> >  EXPORT_SYMBOL(node_data);
> >
> > -static struct numa_meminfo numa_meminfo
> > -#ifndef CONFIG_MEMORY_HOTPLUG
> > -__initdata
> > -#endif
> > -;
> > +static struct numa_meminfo numa_meminfo __initdata_numa;
> >
> >  static int numa_distance_cnt;
> >  static u8 *numa_distance;
> > diff --git a/include/linux/numa.h b/include/linux/numa.h
> > index 20f4e44b186c..c005ed6b807b 100644
> > --- a/include/linux/numa.h
> > +++ b/include/linux/numa.h
> > @@ -13,6 +13,12 @@
> >
> >  #define      NUMA_NO_NODE    (-1)
> >
> > +#ifdef CONFIG_KEEP_NUMA
> > +#define __initdata_numa
> > +#else
> > +#define __initdata_numa __initdata
> > +#endif
> > +
> >  #ifdef CONFIG_NUMA
> >  int numa_map_to_online_node(int node);
> >  #else
> > diff --git a/mm/Kconfig b/mm/Kconfig
> > index ab80933be65f..001f1185eadf 100644
> > --- a/mm/Kconfig
> > +++ b/mm/Kconfig
> > @@ -139,6 +139,10 @@ config HAVE_FAST_GUP
> >  config ARCH_KEEP_MEMBLOCK
> >       bool
> >
> > +# Keep arch numa mapping infrastructure post-init.
>
> s/numa/NUMA
>
> Please also capitalize consistently in the rest of the series.
>
> > +config KEEP_NUMA
> > +     bool
>
>
> So most of our recent new NUMA options followed the naming pattern of:
>
>   CONFIG_NUMA_*
>
> Such as CONFIG_NUMA_BALANCING or CONFIG_NUMA_EMU.
>
> So I'd suggesting naming it to CONFIG_NUMA_KEEP, or, a bit more
> descriptively, such as CONFIG_NUMA_KEEP_MAPPING or such?
>
> 'Keeping NUMA' is kind of lame - of course we keep NUMA. ;-)

Ok, I settled on CONFIG_NUMA_KEEP_MEMINFO, and will fix up the
lowercase "numa" instances in the set.
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 4/6] x86/mm: Introduce CONFIG_KEEP_NUMA

[Thomas Gleixner]

Dan Williams <dan.j.williams@intel.com> writes:
> +#ifdef CONFIG_KEEP_NUMA
> +#define __initdata_numa
> +#else
> +#define __initdata_numa __initdata
> +#endif

TBH, I find this conditional annotation mightingly confusing.

__initdata_numa still suggest that this is __initdata, just a different
section and some extra rules or whatever.

Something like __initdata_or_keepnuma (sorry I could not come up with
something prettier, but you get the idea.

Thanks,

        tglx
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 4/6] x86/mm: Introduce CONFIG_KEEP_NUMA

[Dan Williams]

On Thu, Feb 13, 2020 at 3:22 AM Thomas Gleixner <tglx@linutronix.de> wrote:
>
> Dan Williams <dan.j.williams@intel.com> writes:
> > +#ifdef CONFIG_KEEP_NUMA
> > +#define __initdata_numa
> > +#else
> > +#define __initdata_numa __initdata
> > +#endif
>
> TBH, I find this conditional annotation mightingly confusing.
>
> __initdata_numa still suggest that this is __initdata, just a different
> section and some extra rules or whatever.
>
> Something like __initdata_or_keepnuma (sorry I could not come up with
> something prettier, but you get the idea.

Yes, and to dovetail with Ingo's feedback I think
__initdata_or_meminfo conveys it's optionally init vs runtime data.
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v4 5/6] x86/numa: Provide a range-to-target_node lookup facility

[Dan Williams]

The DEV_DAX_KMEM facility is a generic mechanism to allow device-dax
instances, fronting performance-differentiated-memory like pmem, to be
added to the System RAM pool. The numa node for that hot-added memory is
derived from the device-dax instance's 'target_node' attribute.

Recall that the 'target_node' is the ACPI-PXM-to-node translation for
memory when it comes online whereas the 'numa_node' attribute of the
device represents the closest online cpu node.

Presently useful target_node information from the ACPI SRAT is discarded
with the expectation that "Reserved" memory will never be onlined. Now,
DEV_DAX_KMEM violates that assumption, there is a need to retain the
translation. Move, rather than discard, numa_memblk data to a secondary
array that memory_add_physaddr_to_target_node() may consider at a later
point in time.

Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <x86@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/x86/mm/numa.c   |   68 +++++++++++++++++++++++++++++++++++++++++++-------
 include/linux/numa.h |    8 +++++-
 mm/mempolicy.c       |    5 ++++
 3 files changed, 70 insertions(+), 11 deletions(-)

diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index 5289d9d6799a..f2c8fca36f28 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c
@@ -26,6 +26,7 @@ struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
 EXPORT_SYMBOL(node_data);
 
 static struct numa_meminfo numa_meminfo __initdata_numa;
+static struct numa_meminfo numa_reserved_meminfo __initdata_numa;
 
 static int numa_distance_cnt;
 static u8 *numa_distance;
@@ -164,6 +165,26 @@ void __init numa_remove_memblk_from(int idx, struct numa_meminfo *mi)
 		(mi->nr_blks - idx) * sizeof(mi->blk[0]));
 }
 
+/**
+ * numa_move_memblk - Move one numa_memblk from one numa_meminfo to another
+ * @dst: numa_meminfo to move block to
+ * @idx: Index of memblk to remove
+ * @src: numa_meminfo to remove memblk from
+ *
+ * If @dst is non-NULL add it at the @dst->nr_blks index and increment
+ * @dst->nr_blks, then remove it from @src.
+ */
+static void __init numa_move_memblk(struct numa_meminfo *dst, int idx,
+		struct numa_meminfo *src)
+{
+	if (dst) {
+		memcpy(&dst->blk[dst->nr_blks], &src->blk[idx],
+				sizeof(struct numa_memblk));
+		dst->nr_blks++;
+	}
+	numa_remove_memblk_from(idx, src);
+}
+
 /**
  * numa_add_memblk - Add one numa_memblk to numa_meminfo
  * @nid: NUMA node ID of the new memblk
@@ -233,14 +254,19 @@ int __init numa_cleanup_meminfo(struct numa_meminfo *mi)
 	for (i = 0; i < mi->nr_blks; i++) {
 		struct numa_memblk *bi = &mi->blk[i];
 
-		/* make sure all blocks are inside the limits */
+		/* move / save reserved memory ranges */
+		if (!memblock_overlaps_region(&memblock.memory,
+					bi->start, bi->end - bi->start)) {
+			numa_move_memblk(&numa_reserved_meminfo, i--, mi);
+			continue;
+		}
+
+		/* make sure all non-reserved blocks are inside the limits */
 		bi->start = max(bi->start, low);
 		bi->end = min(bi->end, high);
 
-		/* and there's no empty or non-exist block */
-		if (bi->start >= bi->end ||
-		    !memblock_overlaps_region(&memblock.memory,
-			bi->start, bi->end - bi->start))
+		/* and there's no empty block */
+		if (bi->start >= bi->end)
 			numa_remove_memblk_from(i--, mi);
 	}
 
@@ -877,16 +903,38 @@ EXPORT_SYMBOL(cpumask_of_node);
 
 #endif	/* !CONFIG_DEBUG_PER_CPU_MAPS */
 
-#ifdef CONFIG_MEMORY_HOTPLUG
-int memory_add_physaddr_to_nid(u64 start)
+#ifdef CONFIG_KEEP_NUMA
+static int meminfo_to_nid(struct numa_meminfo *mi, u64 start)
 {
-	struct numa_meminfo *mi = &numa_meminfo;
-	int nid = mi->blk[0].nid;
 	int i;
 
 	for (i = 0; i < mi->nr_blks; i++)
 		if (mi->blk[i].start <= start && mi->blk[i].end > start)
-			nid = mi->blk[i].nid;
+			return mi->blk[i].nid;
+	return NUMA_NO_NODE;
+}
+
+int phys_to_target_node(phys_addr_t start)
+{
+	int nid = meminfo_to_nid(&numa_meminfo, start);
+
+	/*
+	 * Prefer online nodes, but if reserved memory might be
+	 * hot-added continue the search with reserved ranges.
+	 */
+	if (nid != NUMA_NO_NODE)
+		return nid;
+
+	return meminfo_to_nid(&numa_reserved_meminfo, start);
+}
+EXPORT_SYMBOL_GPL(phys_to_target_node);
+
+int memory_add_physaddr_to_nid(u64 start)
+{
+	int nid = meminfo_to_nid(&numa_meminfo, start);
+
+	if (nid == NUMA_NO_NODE)
+		nid = numa_meminfo.blk[0].nid;
 	return nid;
 }
 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
diff --git a/include/linux/numa.h b/include/linux/numa.h
index c005ed6b807b..cad0ab165619 100644
--- a/include/linux/numa.h
+++ b/include/linux/numa.h
@@ -1,7 +1,7 @@
 /* SPDX-License-Identifier: GPL-2.0 */
 #ifndef _LINUX_NUMA_H
 #define _LINUX_NUMA_H
-
+#include <linux/types.h>
 
 #ifdef CONFIG_NODES_SHIFT
 #define NODES_SHIFT     CONFIG_NODES_SHIFT
@@ -21,11 +21,17 @@
 
 #ifdef CONFIG_NUMA
 int numa_map_to_online_node(int node);
+int phys_to_target_node(phys_addr_t addr);
 #else
 static inline int numa_map_to_online_node(int node)
 {
 	return NUMA_NO_NODE;
 }
+
+static inline int phys_to_target_node(phys_addr_t addr)
+{
+	return NUMA_NO_NODE;
+}
 #endif
 
 #endif /* _LINUX_NUMA_H */
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index bcb012645809..ed376d57e527 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -3011,3 +3011,8 @@ void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
 		p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
 			       nodemask_pr_args(&nodes));
 }
+
+__weak int phys_to_target_node(phys_addr_t addr)
+{
+	return NUMA_NO_NODE;
+}
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 5/6] x86/numa: Provide a range-to-target_node lookup facility

[Ingo Molnar]

* Dan Williams <dan.j.williams@intel.com> wrote:

> The DEV_DAX_KMEM facility is a generic mechanism to allow device-dax
> instances, fronting performance-differentiated-memory like pmem, to be
> added to the System RAM pool. The numa node for that hot-added memory is
> derived from the device-dax instance's 'target_node' attribute.
> 
> Recall that the 'target_node' is the ACPI-PXM-to-node translation for
> memory when it comes online whereas the 'numa_node' attribute of the
> device represents the closest online cpu node.
> 
> Presently useful target_node information from the ACPI SRAT is discarded
> with the expectation that "Reserved" memory will never be onlined. Now,
> DEV_DAX_KMEM violates that assumption, there is a need to retain the
> translation. Move, rather than discard, numa_memblk data to a secondary
> array that memory_add_physaddr_to_target_node() may consider at a later
> point in time.
> 
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Cc: Andy Lutomirski <luto@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Borislav Petkov <bp@alien8.de>
> Cc: "H. Peter Anvin" <hpa@zytor.com>
> Cc: <x86@kernel.org>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: David Hildenbrand <david@redhat.com>
> Cc: Michal Hocko <mhocko@suse.com>
> Reported-by: kbuild test robot <lkp@intel.com>
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>  arch/x86/mm/numa.c   |   68 +++++++++++++++++++++++++++++++++++++++++++-------
>  include/linux/numa.h |    8 +++++-
>  mm/mempolicy.c       |    5 ++++
>  3 files changed, 70 insertions(+), 11 deletions(-)
> 
> diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
> index 5289d9d6799a..f2c8fca36f28 100644
> --- a/arch/x86/mm/numa.c
> +++ b/arch/x86/mm/numa.c
> @@ -26,6 +26,7 @@ struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
>  EXPORT_SYMBOL(node_data);
>  
>  static struct numa_meminfo numa_meminfo __initdata_numa;
> +static struct numa_meminfo numa_reserved_meminfo __initdata_numa;
>  
>  static int numa_distance_cnt;
>  static u8 *numa_distance;
> @@ -164,6 +165,26 @@ void __init numa_remove_memblk_from(int idx, struct numa_meminfo *mi)
>  		(mi->nr_blks - idx) * sizeof(mi->blk[0]));
>  }
>  
> +/**
> + * numa_move_memblk - Move one numa_memblk from one numa_meminfo to another
> + * @dst: numa_meminfo to move block to
> + * @idx: Index of memblk to remove
> + * @src: numa_meminfo to remove memblk from
> + *
> + * If @dst is non-NULL add it at the @dst->nr_blks index and increment
> + * @dst->nr_blks, then remove it from @src.
> + */
> +static void __init numa_move_memblk(struct numa_meminfo *dst, int idx,
> +		struct numa_meminfo *src)

Nit, this is obviously not how we format function definitions if 
checkpatch complains about the col80 limit.


> +{
> +	if (dst) {
> +		memcpy(&dst->blk[dst->nr_blks], &src->blk[idx],
> +				sizeof(struct numa_memblk));

This linebreak is actually unnecessary ...

Thanks,

	Ingo
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 5/6] x86/numa: Provide a range-to-target_node lookup facility

[Thomas Gleixner]

Dan Williams <dan.j.williams@intel.com> writes:
> +/**
> + * numa_move_memblk - Move one numa_memblk from one numa_meminfo to another
> + * @dst: numa_meminfo to move block to
> + * @idx: Index of memblk to remove
> + * @src: numa_meminfo to remove memblk from
> + *
> + * If @dst is non-NULL add it at the @dst->nr_blks index and increment
> + * @dst->nr_blks, then remove it from @src.

This is not correct. It's suggesting that these operations are only
happening when @dst is non-NULL. Remove is unconditional though.

Also this is called with &numa_reserved_meminfo as @dst argument, which is:

> +static struct numa_meminfo numa_reserved_meminfo __initdata_numa;

So how would @dst ever be NULL?
 
> + */
> +static void __init numa_move_memblk(struct numa_meminfo *dst, int idx,
> +		struct numa_meminfo *src)
> +{
> +	if (dst) {
> +		memcpy(&dst->blk[dst->nr_blks], &src->blk[idx],
> +				sizeof(struct numa_memblk));
> +		dst->nr_blks++;
> +	}
> +	numa_remove_memblk_from(idx, src);
> +}

...

> -		/* make sure all blocks are inside the limits */
> +		/* move / save reserved memory ranges */
> +		if (!memblock_overlaps_region(&memblock.memory,
> +					bi->start, bi->end - bi->start)) {
> +			numa_move_memblk(&numa_reserved_meminfo, i--, mi);

Thanks,

        tglx
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 5/6] x86/numa: Provide a range-to-target_node lookup facility

[Dan Williams]

On Thu, Feb 13, 2020 at 3:38 AM Thomas Gleixner <tglx@linutronix.de> wrote:
>
> Dan Williams <dan.j.williams@intel.com> writes:
> > +/**
> > + * numa_move_memblk - Move one numa_memblk from one numa_meminfo to another
> > + * @dst: numa_meminfo to move block to
> > + * @idx: Index of memblk to remove
> > + * @src: numa_meminfo to remove memblk from
> > + *
> > + * If @dst is non-NULL add it at the @dst->nr_blks index and increment
> > + * @dst->nr_blks, then remove it from @src.
>
> This is not correct. It's suggesting that these operations are only
> happening when @dst is non-NULL. Remove is unconditional though.
>
> Also this is called with &numa_reserved_meminfo as @dst argument, which is:
>
> > +static struct numa_meminfo numa_reserved_meminfo __initdata_numa;
>
> So how would @dst ever be NULL?

Ugh, something I should have caught. An earlier version of this patch
optionally defined numa_reserved_meminfo [1], but I later switched to
the current / cleaner __initdata_or_meminfo scheme. Will clean this
up.

[1]: https://lore.kernel.org/linux-mm/157309907296.1582359.7986676987778026949.stgit@dwillia2-desk3.amr.corp.intel.com/
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v4 6/6] libnvdimm/e820: Retrieve and populate correct 'target_node' info

[Dan Williams]

Use the new phys_to_target_node() and numa_map_to_online_node() helpers
to retrieve the correct id for the 'numa_node' ("local" / online
initiator node) and 'target_node' (offline target memory node) sysfs
attributes.

Below is an example from a 4 numa node system where all the memory on
node2 is pmem / reserved. It should be noted that with the arrival of
the ACPI HMAT table and EFI Specific Purpose Memory the kernel will
start to see more platforms with reserved / performance differentiated
memory in its own numa node. Hence all the stakeholders on the Cc for
what is ostensibly a libnvdimm local patch.

=== Before ===

/* Notice no online memory on node2 at start */

# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3708 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node   0   1   3
  0:  10  21  21
  1:  21  10  21
  3:  21  21  10

/*
 * Put the pmem namespace into devdax mode so it can be assigned to the
 * kmem driver
 */

# ndctl create-namespace -e namespace0.0 -m devdax -f
{
  "dev":"namespace0.0",
  "mode":"devdax",
  "map":"dev",
  "size":"3.94 GiB (4.23 GB)",
  "uuid":"1650af9b-9ba3-4704-acd6-10178399d9a3",
  [..]
}

/* Online Persistent Memory as System RAM */

# daxctl reconfigure-device --mode=system-ram dax0.0
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
[
  {
    "chardev":"dax0.0",
    "size":4225761280,
    "target_node":0,
    "mode":"system-ram"
  }
]
reconfigured 1 device

/* Note that the memory is onlined by default to the wrong node, node0 */

# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 7926 MB
node 0 free: 7655 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node   0   1   3
  0:  10  21  21
  1:  21  10  21
  3:  21  21  10


=== After ===

/* Notice that the "phys_index" error messages are gone */

# daxctl reconfigure-device --mode=system-ram dax0.0
[
  {
    "chardev":"dax0.0",
    "size":4225761280,
    "target_node":2,
    "mode":"system-ram"
  }
]
reconfigured 1 device

/* Notice that node2 is now correctly populated */

# numactl --hardware
available: 4 nodes (0-3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3793 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3851 MB
node 2 cpus:
node 2 size: 3968 MB
node 2 free: 3968 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3908 MB
node distances:
node   0   1   2   3
  0:  10  21  21  21
  1:  21  10  21  21
  2:  21  21  10  21
  3:  21  21  21  10

Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/x86/Kconfig      |    1 +
 drivers/nvdimm/e820.c |   18 ++++--------------
 2 files changed, 5 insertions(+), 14 deletions(-)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 5e8949953660..3a827fe7afd6 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1660,6 +1660,7 @@ config X86_PMEM_LEGACY
 	depends on PHYS_ADDR_T_64BIT
 	depends on BLK_DEV
 	select X86_PMEM_LEGACY_DEVICE
+	select KEEP_NUMA if NUMA
 	select LIBNVDIMM
 	help
 	  Treat memory marked using the non-standard e820 type of 12 as used
diff --git a/drivers/nvdimm/e820.c b/drivers/nvdimm/e820.c
index e02f60ad6c99..4cd18be9d0e9 100644
--- a/drivers/nvdimm/e820.c
+++ b/drivers/nvdimm/e820.c
@@ -7,6 +7,7 @@
 #include <linux/memory_hotplug.h>
 #include <linux/libnvdimm.h>
 #include <linux/module.h>
+#include <linux/numa.h>
 
 static int e820_pmem_remove(struct platform_device *pdev)
 {
@@ -16,27 +17,16 @@ static int e820_pmem_remove(struct platform_device *pdev)
 	return 0;
 }
 
-#ifdef CONFIG_MEMORY_HOTPLUG
-static int e820_range_to_nid(resource_size_t addr)
-{
-	return memory_add_physaddr_to_nid(addr);
-}
-#else
-static int e820_range_to_nid(resource_size_t addr)
-{
-	return NUMA_NO_NODE;
-}
-#endif
-
 static int e820_register_one(struct resource *res, void *data)
 {
 	struct nd_region_desc ndr_desc;
 	struct nvdimm_bus *nvdimm_bus = data;
+	int nid = phys_to_target_node(res->start);
 
 	memset(&ndr_desc, 0, sizeof(ndr_desc));
 	ndr_desc.res = res;
-	ndr_desc.numa_node = e820_range_to_nid(res->start);
-	ndr_desc.target_node = ndr_desc.numa_node;
+	ndr_desc.numa_node = numa_map_to_online_node(nid);
+	ndr_desc.target_node = nid;
 	set_bit(ND_REGION_PAGEMAP, &ndr_desc.flags);
 	if (!nvdimm_pmem_region_create(nvdimm_bus, &ndr_desc))
 		return -ENXIO;
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 0/6] Memory Hierarchy: Enable target node lookups for reserved memory

[Dan Williams]

On Tue, Jan 21, 2020 at 7:20 PM Dan Williams <dan.j.williams@intel.com> wrote:
>
> Changes since v3 [1]:
> - Cleanup numa_map_to_online_node() to remove redundant "if
>   (!node_online(node))" (Aneesh)
>
> [1]: http://lore.kernel.org/r/157954696789.2239526.17707265517154476652.stgit@dwillia2-desk3.amr.corp.intel.com
>
> ---
>
> Merge notes:
>
> x86 folks: This has an ack from Rafael for ACPI, and Michael for Power.
> With an x86 ack I plan to take this through the libnvdimm tree provided
> the x86 touches look ok to you.

Ping x86 folks. There's no additional changes identified for this
series. Can I request an ack to take it through libnvdimm.git? Do you
need a resend?

    x86/mm: Introduce CONFIG_KEEP_NUMA
    x86/numa: Provide a range-to-target_node lookup facility


>
> ---
>
> Cover:
>
> Arrange for platform numa info to be preserved for determining
> 'target_node' data. Where a 'target_node' is the node a reserved memory
> range will become when it is onlined.
>
> This new infrastructure is expected to be more valuable over time for
> Memory Tiers / Hierarchy management as more platforms (via the ACPI HMAT
> and EFI Specific Purpose Memory) publish reserved or "soft-reserved"
> ranges to Linux. Linux system administrators will expect to be able to
> interact with those ranges with a unique numa node number when/if that
> memory is onlined via the dax_kmem driver [2].
>
> One configuration that currently fails to properly convey the target
> node for the resulting memory hotplug operation is persistent memory
> defined by the memmap=nn!ss parameter. For example, today if node1 is a
> memory only node, and all the memory from node1 is specified to
> memmap=nn!ss and subsequently onlined, it will end up being onlined as
> node0 memory. As it stands, memory_add_physaddr_to_nid() can only
> identify online nodes and since node1 in this example has no online cpus
> / memory the target node is initialized node0.
>
> The fix is to preserve rather than discard the numa_meminfo entries that
> are relevant for reserved memory ranges, and to uplevel the node
> distance helper for determining the "local" (closest) node relative to
> an initiator node.
>
> [2]: https://pmem.io/ndctl/daxctl-reconfigure-device.html
>
> ---
>
> Dan Williams (6):
>       ACPI: NUMA: Up-level "map to online node" functionality
>       mm/numa: Skip NUMA_NO_NODE and online nodes in numa_map_to_online_node()
>       powerpc/papr_scm: Switch to numa_map_to_online_node()
>       x86/mm: Introduce CONFIG_KEEP_NUMA
>       x86/numa: Provide a range-to-target_node lookup facility
>       libnvdimm/e820: Retrieve and populate correct 'target_node' info
>
>
>  arch/powerpc/platforms/pseries/papr_scm.c |   21 --------
>  arch/x86/Kconfig                          |    1
>  arch/x86/mm/numa.c                        |   74 +++++++++++++++++++++++------
>  drivers/acpi/numa/srat.c                  |   41 ----------------
>  drivers/nvdimm/e820.c                     |   18 ++-----
>  include/linux/acpi.h                      |   23 +++++++++
>  include/linux/numa.h                      |   23 +++++++++
>  mm/Kconfig                                |    5 ++
>  mm/mempolicy.c                            |   31 ++++++++++++
>  9 files changed, 145 insertions(+), 92 deletions(-)
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v4 0/6] Memory Hierarchy: Enable target node lookups for reserved memory

[Ingo Molnar]

* Dan Williams <dan.j.williams@intel.com> wrote:

> On Tue, Jan 21, 2020 at 7:20 PM Dan Williams <dan.j.williams@intel.com> wrote:
> >
> > Changes since v3 [1]:
> > - Cleanup numa_map_to_online_node() to remove redundant "if
> >   (!node_online(node))" (Aneesh)
> >
> > [1]: http://lore.kernel.org/r/157954696789.2239526.17707265517154476652.stgit@dwillia2-desk3.amr.corp.intel.com
> >
> > ---
> >
> > Merge notes:
> >
> > x86 folks: This has an ack from Rafael for ACPI, and Michael for Power.
> > With an x86 ack I plan to take this through the libnvdimm tree provided
> > the x86 touches look ok to you.
> 
> Ping x86 folks. There's no additional changes identified for this
> series. Can I request an ack to take it through libnvdimm.git? Do you
> need a resend?
> 
>     x86/mm: Introduce CONFIG_KEEP_NUMA
>     x86/numa: Provide a range-to-target_node lookup facility

If the minor complaints I outlined are addressed:

Reviewed-by: Ingo Molnar <mingo@kernel.org>

Thanks,

	Ingo
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v5 0/6] Memory Hierarchy: Enable target node lookups for reserved memory

[Dan Williams]

Changes since v4 [1]:
- Rename CONFIG_KEEP_NUMA to CONFIG_NUMA_KEEP_MEMINFO (Ingo)
- Rename __initdata_numa to __initdata_or_meminfo (Thomas)
- Capitalize NUMA throughout (Ingo)
- Replace explicit memcpy with implicit structure copy to address an 80
  column violation, and fixup a function definition line-wrap (Ingo)
- Rename numa_move_memblk() to numa_move_tail_memblk(), and remove the
  stale kernel-doc that implied @dst was optional (Thomas)
- Comment that phys_to_target_node() is an optional arch implementation
  detail that consumers must gate with "depends on $ARCH"
- Apply Ingo's conditional reviewed-by

[1]: http://lore.kernel.org/r/157966227494.2508551.7206194169374588977.stgit@dwillia2-desk3.amr.corp.intel.com

---

Merge notes: I believe this addresses all outstanding comments, barring
additional feedback I will push to libnvdimm-for-next.

---

Cover:

Arrange for platform NUMA info to be preserved for determining
'target_node' data. Where a 'target_node' is the node a reserved memory
range will become when it is onlined.

This new infrastructure is expected to be more valuable over time for
Memory Tiers / Hierarchy management as more platforms (via the ACPI HMAT
and EFI Specific Purpose Memory) publish reserved or "soft-reserved"
ranges to Linux. Linux system administrators will expect to be able to
interact with those ranges with a unique NUMA node number when/if that
memory is onlined via the dax_kmem driver [2].

One configuration that currently fails to properly convey the target
node for the resulting memory hotplug operation is persistent memory
defined by the memmap=nn!ss parameter. For example, today if node1 is a
memory only node, and all the memory from node1 is specified to
memmap=nn!ss and subsequently onlined, it will end up being onlined as
node0 memory. As it stands, memory_add_physaddr_to_nid() can only
identify online nodes and since node1 in this example has no online cpus
/ memory the target node is initialized node0.

The fix is to preserve rather than discard the numa_meminfo entries that
are relevant for reserved memory ranges, and to uplevel the node
distance helper for determining the "local" (closest) node relative to
an initiator node.

[2]: https://pmem.io/ndctl/daxctl-reconfigure-device.html

---

Dan Williams (6):
      ACPI: NUMA: Up-level "map to online node" functionality
      mm/numa: Skip NUMA_NO_NODE and online nodes in numa_map_to_online_node()
      powerpc/papr_scm: Switch to numa_map_to_online_node()
      x86/mm: Introduce CONFIG_NUMA_KEEP_MEMINFO
      x86/NUMA: Provide a range-to-target_node lookup facility
      libnvdimm/e820: Retrieve and populate correct 'target_node' info


 arch/powerpc/platforms/pseries/papr_scm.c |   21 ---------
 arch/x86/Kconfig                          |    1 
 arch/x86/mm/numa.c                        |   67 +++++++++++++++++++++++------
 drivers/acpi/numa/srat.c                  |   41 ------------------
 drivers/nvdimm/e820.c                     |   18 ++------
 include/linux/acpi.h                      |   23 ++++++++++
 include/linux/numa.h                      |   30 +++++++++++++
 mm/Kconfig                                |    5 ++
 mm/mempolicy.c                            |   26 +++++++++++
 9 files changed, 140 insertions(+), 92 deletions(-)

base-commit: bb6d3fb354c5ee8d6bde2d576eb7220ea09862b9
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v5 1/6] ACPI: NUMA: Up-level "map to online node" functionality

[Dan Williams]

The acpi_map_pxm_to_online_node() helper is used to find the closest
online node to a given proximity domain. This is used to map devices in
a proximity domain with no online memory or cpus to the closest online
node and populate a device's 'numa_node' property. The numa_node
property allows applications to be migrated "close" to a resource.

In preparation for providing a generic facility to optionally map an
address range to its closest online node, or the node the range would
represent were it to be onlined (target_node), up-level the core of
acpi_map_pxm_to_online_node() to a generic mm/numa helper.

Cc: Michal Hocko <mhocko@suse.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/acpi/numa/srat.c |   41 -----------------------------------------
 include/linux/acpi.h     |   23 ++++++++++++++++++++++-
 include/linux/numa.h     |    9 +++++++++
 mm/mempolicy.c           |   30 ++++++++++++++++++++++++++++++
 4 files changed, 61 insertions(+), 42 deletions(-)

diff --git a/drivers/acpi/numa/srat.c b/drivers/acpi/numa/srat.c
index eadbf90e65d1..47b4969d9b93 100644
--- a/drivers/acpi/numa/srat.c
+++ b/drivers/acpi/numa/srat.c
@@ -72,47 +72,6 @@ int acpi_map_pxm_to_node(int pxm)
 }
 EXPORT_SYMBOL(acpi_map_pxm_to_node);
 
-/**
- * acpi_map_pxm_to_online_node - Map proximity ID to online node
- * @pxm: ACPI proximity ID
- *
- * This is similar to acpi_map_pxm_to_node(), but always returns an online
- * node.  When the mapped node from a given proximity ID is offline, it
- * looks up the node distance table and returns the nearest online node.
- *
- * ACPI device drivers, which are called after the NUMA initialization has
- * completed in the kernel, can call this interface to obtain their device
- * NUMA topology from ACPI tables.  Such drivers do not have to deal with
- * offline nodes.  A node may be offline when a device proximity ID is
- * unique, SRAT memory entry does not exist, or NUMA is disabled, ex.
- * "numa=off" on x86.
- */
-int acpi_map_pxm_to_online_node(int pxm)
-{
-	int node, min_node;
-
-	node = acpi_map_pxm_to_node(pxm);
-
-	if (node == NUMA_NO_NODE)
-		node = 0;
-
-	min_node = node;
-	if (!node_online(node)) {
-		int min_dist = INT_MAX, dist, n;
-
-		for_each_online_node(n) {
-			dist = node_distance(node, n);
-			if (dist < min_dist) {
-				min_dist = dist;
-				min_node = n;
-			}
-		}
-	}
-
-	return min_node;
-}
-EXPORT_SYMBOL(acpi_map_pxm_to_online_node);
-
 static void __init
 acpi_table_print_srat_entry(struct acpi_subtable_header *header)
 {
diff --git a/include/linux/acpi.h b/include/linux/acpi.h
index 0f24d701fbdc..3839363081f3 100644
--- a/include/linux/acpi.h
+++ b/include/linux/acpi.h
@@ -416,9 +416,30 @@ extern void acpi_osi_setup(char *str);
 extern bool acpi_osi_is_win8(void);
 
 #ifdef CONFIG_ACPI_NUMA
-int acpi_map_pxm_to_online_node(int pxm);
 int acpi_map_pxm_to_node(int pxm);
 int acpi_get_node(acpi_handle handle);
+
+/**
+ * acpi_map_pxm_to_online_node - Map proximity ID to online node
+ * @pxm: ACPI proximity ID
+ *
+ * This is similar to acpi_map_pxm_to_node(), but always returns an online
+ * node.  When the mapped node from a given proximity ID is offline, it
+ * looks up the node distance table and returns the nearest online node.
+ *
+ * ACPI device drivers, which are called after the NUMA initialization has
+ * completed in the kernel, can call this interface to obtain their device
+ * NUMA topology from ACPI tables.  Such drivers do not have to deal with
+ * offline nodes.  A node may be offline when a device proximity ID is
+ * unique, SRAT memory entry does not exist, or NUMA is disabled, ex.
+ * "numa=off" on x86.
+ */
+static inline int acpi_map_pxm_to_online_node(int pxm)
+{
+	int node = acpi_map_pxm_to_node(pxm);
+
+	return numa_map_to_online_node(node);
+}
 #else
 static inline int acpi_map_pxm_to_online_node(int pxm)
 {
diff --git a/include/linux/numa.h b/include/linux/numa.h
index 110b0e5d0fb0..20f4e44b186c 100644
--- a/include/linux/numa.h
+++ b/include/linux/numa.h
@@ -13,4 +13,13 @@
 
 #define	NUMA_NO_NODE	(-1)
 
+#ifdef CONFIG_NUMA
+int numa_map_to_online_node(int node);
+#else
+static inline int numa_map_to_online_node(int node)
+{
+	return NUMA_NO_NODE;
+}
+#endif
+
 #endif /* _LINUX_NUMA_H */
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 977c641f78cf..756d6e5bb59f 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -127,6 +127,36 @@ static struct mempolicy default_policy = {
 
 static struct mempolicy preferred_node_policy[MAX_NUMNODES];
 
+/**
+ * numa_map_to_online_node - Find closest online node
+ * @nid: Node id to start the search
+ *
+ * Lookup the next closest node by distance if @nid is not online.
+ */
+int numa_map_to_online_node(int node)
+{
+	int min_node;
+
+	if (node == NUMA_NO_NODE)
+		node = 0;
+
+	min_node = node;
+	if (!node_online(node)) {
+		int min_dist = INT_MAX, dist, n;
+
+		for_each_online_node(n) {
+			dist = node_distance(node, n);
+			if (dist < min_dist) {
+				min_dist = dist;
+				min_node = n;
+			}
+		}
+	}
+
+	return min_node;
+}
+EXPORT_SYMBOL_GPL(numa_map_to_online_node);
+
 struct mempolicy *get_task_policy(struct task_struct *p)
 {
 	struct mempolicy *pol = p->mempolicy;
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v5 2/6] mm/numa: Skip NUMA_NO_NODE and online nodes in numa_map_to_online_node()

[Dan Williams]

Update numa_map_to_online_node() to stop falling back to numa node 0
when the input is NUMA_NO_NODE. Also, skip the lookup if @node is
online. This makes the routine compatible with other arch node mapping
routines.

Reported-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Link: https://lore.kernel.org/r/157401275716.43284.13185549705765009174.stgit@dwillia2-desk3.amr.corp.intel.com
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 mm/mempolicy.c |   20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 756d6e5bb59f..19f7e71945a7 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -135,21 +135,17 @@ static struct mempolicy preferred_node_policy[MAX_NUMNODES];
  */
 int numa_map_to_online_node(int node)
 {
-	int min_node;
+	int min_dist = INT_MAX, dist, n, min_node;
 
-	if (node == NUMA_NO_NODE)
-		node = 0;
+	if (node == NUMA_NO_NODE || node_online(node))
+		return node;
 
 	min_node = node;
-	if (!node_online(node)) {
-		int min_dist = INT_MAX, dist, n;
-
-		for_each_online_node(n) {
-			dist = node_distance(node, n);
-			if (dist < min_dist) {
-				min_dist = dist;
-				min_node = n;
-			}
+	for_each_online_node(n) {
+		dist = node_distance(node, n);
+		if (dist < min_dist) {
+			min_dist = dist;
+			min_node = n;
 		}
 	}
 
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v5 3/6] powerpc/papr_scm: Switch to numa_map_to_online_node()

[Dan Williams]

Now that the core exports numa_map_to_online_node() switch to that
instead of the locally coded duplicate.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: "Oliver O'Halloran" <oohall@gmail.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Reported-by: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Link: https://lore.kernel.org/r/157401276263.43284.12616818803654229788.stgit@dwillia2-desk3.amr.corp.intel.com
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/powerpc/platforms/pseries/papr_scm.c |   21 +--------------------
 1 file changed, 1 insertion(+), 20 deletions(-)

diff --git a/arch/powerpc/platforms/pseries/papr_scm.c b/arch/powerpc/platforms/pseries/papr_scm.c
index 0b4467e378e5..3cc66224ec1f 100644
--- a/arch/powerpc/platforms/pseries/papr_scm.c
+++ b/arch/powerpc/platforms/pseries/papr_scm.c
@@ -285,25 +285,6 @@ int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 	return 0;
 }
 
-static inline int papr_scm_node(int node)
-{
-	int min_dist = INT_MAX, dist;
-	int nid, min_node;
-
-	if ((node == NUMA_NO_NODE) || node_online(node))
-		return node;
-
-	min_node = first_online_node;
-	for_each_online_node(nid) {
-		dist = node_distance(node, nid);
-		if (dist < min_dist) {
-			min_dist = dist;
-			min_node = nid;
-		}
-	}
-	return min_node;
-}
-
 static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
 {
 	struct device *dev = &p->pdev->dev;
@@ -349,7 +330,7 @@ static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
 
 	memset(&ndr_desc, 0, sizeof(ndr_desc));
 	target_nid = dev_to_node(&p->pdev->dev);
-	online_nid = papr_scm_node(target_nid);
+	online_nid = numa_map_to_online_node(target_nid);
 	ndr_desc.numa_node = online_nid;
 	ndr_desc.target_node = target_nid;
 	ndr_desc.res = &p->res;
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v5 4/6] x86/mm: Introduce CONFIG_NUMA_KEEP_MEMINFO

[Dan Williams]

Currently x86 numa_meminfo is marked __initdata in the
CONFIG_MEMORY_HOTPLUG=n case. In support of a new facility to allow
drivers to map reserved memory to a 'target_node'
(phys_to_target_node()), add support for removing the __initdata
designation for those users. Both memory hotplug and
phys_to_target_node() users select CONFIG_NUMA_KEEP_MEMINFO to tell the
arch to maintain its physical address to NUMA mapping infrastructure
post init.

Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <x86@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/x86/mm/numa.c   |    6 +-----
 include/linux/numa.h |    7 +++++++
 mm/Kconfig           |    5 +++++
 3 files changed, 13 insertions(+), 5 deletions(-)

diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index 99f7a68738f0..2450b21cc28a 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c
@@ -25,11 +25,7 @@ nodemask_t numa_nodes_parsed __initdata;
 struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
 EXPORT_SYMBOL(node_data);
 
-static struct numa_meminfo numa_meminfo
-#ifndef CONFIG_MEMORY_HOTPLUG
-__initdata
-#endif
-;
+static struct numa_meminfo numa_meminfo __initdata_or_meminfo;
 
 static int numa_distance_cnt;
 static u8 *numa_distance;
diff --git a/include/linux/numa.h b/include/linux/numa.h
index 20f4e44b186c..5773cd2613fc 100644
--- a/include/linux/numa.h
+++ b/include/linux/numa.h
@@ -13,6 +13,13 @@
 
 #define	NUMA_NO_NODE	(-1)
 
+/* optionally keep NUMA memory info available post init */
+#ifdef CONFIG_NUMA_KEEP_MEMINFO
+#define __initdata_or_meminfo
+#else
+#define __initdata_or_meminfo __initdata
+#endif
+
 #ifdef CONFIG_NUMA
 int numa_map_to_online_node(int node);
 #else
diff --git a/mm/Kconfig b/mm/Kconfig
index ab80933be65f..328268473fec 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -139,6 +139,10 @@ config HAVE_FAST_GUP
 config ARCH_KEEP_MEMBLOCK
 	bool
 
+# Keep arch NUMA mapping infrastructure post-init.
+config NUMA_KEEP_MEMINFO
+	bool
+
 config MEMORY_ISOLATION
 	bool
 
@@ -154,6 +158,7 @@ config MEMORY_HOTPLUG
 	bool "Allow for memory hot-add"
 	depends on SPARSEMEM || X86_64_ACPI_NUMA
 	depends on ARCH_ENABLE_MEMORY_HOTPLUG
+	select NUMA_KEEP_MEMINFO if NUMA
 
 config MEMORY_HOTPLUG_SPARSE
 	def_bool y
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v5 4/6] x86/mm: Introduce CONFIG_NUMA_KEEP_MEMINFO

[Thomas Gleixner]

eviewed-by: Thomas Gleixner <tglx@linutronix.de>
Dan Williams <dan.j.williams@intel.com> writes:

> Currently x86 numa_meminfo is marked __initdata in the
> CONFIG_MEMORY_HOTPLUG=n case. In support of a new facility to allow
> drivers to map reserved memory to a 'target_node'
> (phys_to_target_node()), add support for removing the __initdata
> designation for those users. Both memory hotplug and
> phys_to_target_node() users select CONFIG_NUMA_KEEP_MEMINFO to tell the
> arch to maintain its physical address to NUMA mapping infrastructure
> post init.
>
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Cc: Andy Lutomirski <luto@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Borislav Petkov <bp@alien8.de>
> Cc: "H. Peter Anvin" <hpa@zytor.com>
> Cc: <x86@kernel.org>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: David Hildenbrand <david@redhat.com>
> Cc: Michal Hocko <mhocko@suse.com>
> Reviewed-by: Ingo Molnar <mingo@kernel.org>
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>

Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v5 5/6] x86/NUMA: Provide a range-to-target_node lookup facility

[Dan Williams]

The DEV_DAX_KMEM facility is a generic mechanism to allow device-dax
instances, fronting performance-differentiated-memory like pmem, to be
added to the System RAM pool. The NUMA node for that hot-added memory is
derived from the device-dax instance's 'target_node' attribute.

Recall that the 'target_node' is the ACPI-PXM-to-node translation for
memory when it comes online whereas the 'numa_node' attribute of the
device represents the closest online cpu node.

Presently useful target_node information from the ACPI SRAT is discarded
with the expectation that "Reserved" memory will never be onlined. Now,
DEV_DAX_KMEM violates that assumption, there is a need to retain the
translation. Move, rather than discard, numa_memblk data to a secondary
array that memory_add_physaddr_to_target_node() may consider at a later
point in time.

Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <x86@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Reported-by: kbuild test robot <lkp@intel.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/x86/mm/numa.c   |   61 ++++++++++++++++++++++++++++++++++++++++++--------
 include/linux/numa.h |   14 +++++++++++
 2 files changed, 64 insertions(+), 11 deletions(-)

diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index 2450b21cc28a..59ba008504dc 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c
@@ -26,6 +26,7 @@ struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
 EXPORT_SYMBOL(node_data);
 
 static struct numa_meminfo numa_meminfo __initdata_or_meminfo;
+static struct numa_meminfo numa_reserved_meminfo __initdata_or_meminfo;
 
 static int numa_distance_cnt;
 static u8 *numa_distance;
@@ -164,6 +165,19 @@ void __init numa_remove_memblk_from(int idx, struct numa_meminfo *mi)
 		(mi->nr_blks - idx) * sizeof(mi->blk[0]));
 }
 
+/**
+ * numa_move_tail_memblk - Move a numa_memblk from one numa_meminfo to another
+ * @dst: numa_meminfo to append block to
+ * @idx: Index of memblk to remove
+ * @src: numa_meminfo to remove memblk from
+ */
+static void __init numa_move_tail_memblk(struct numa_meminfo *dst, int idx,
+					 struct numa_meminfo *src)
+{
+	dst->blk[dst->nr_blks++] = src->blk[idx];
+	numa_remove_memblk_from(idx, src);
+}
+
 /**
  * numa_add_memblk - Add one numa_memblk to numa_meminfo
  * @nid: NUMA node ID of the new memblk
@@ -233,14 +247,19 @@ int __init numa_cleanup_meminfo(struct numa_meminfo *mi)
 	for (i = 0; i < mi->nr_blks; i++) {
 		struct numa_memblk *bi = &mi->blk[i];
 
-		/* make sure all blocks are inside the limits */
+		/* move / save reserved memory ranges */
+		if (!memblock_overlaps_region(&memblock.memory,
+					bi->start, bi->end - bi->start)) {
+			numa_move_tail_memblk(&numa_reserved_meminfo, i--, mi);
+			continue;
+		}
+
+		/* make sure all non-reserved blocks are inside the limits */
 		bi->start = max(bi->start, low);
 		bi->end = min(bi->end, high);
 
-		/* and there's no empty or non-exist block */
-		if (bi->start >= bi->end ||
-		    !memblock_overlaps_region(&memblock.memory,
-			bi->start, bi->end - bi->start))
+		/* and there's no empty block */
+		if (bi->start >= bi->end)
 			numa_remove_memblk_from(i--, mi);
 	}
 
@@ -877,16 +896,38 @@ EXPORT_SYMBOL(cpumask_of_node);
 
 #endif	/* !CONFIG_DEBUG_PER_CPU_MAPS */
 
-#ifdef CONFIG_MEMORY_HOTPLUG
-int memory_add_physaddr_to_nid(u64 start)
+#ifdef CONFIG_NUMA_KEEP_MEMINFO
+static int meminfo_to_nid(struct numa_meminfo *mi, u64 start)
 {
-	struct numa_meminfo *mi = &numa_meminfo;
-	int nid = mi->blk[0].nid;
 	int i;
 
 	for (i = 0; i < mi->nr_blks; i++)
 		if (mi->blk[i].start <= start && mi->blk[i].end > start)
-			nid = mi->blk[i].nid;
+			return mi->blk[i].nid;
+	return NUMA_NO_NODE;
+}
+
+int phys_to_target_node(phys_addr_t start)
+{
+	int nid = meminfo_to_nid(&numa_meminfo, start);
+
+	/*
+	 * Prefer online nodes, but if reserved memory might be
+	 * hot-added continue the search with reserved ranges.
+	 */
+	if (nid != NUMA_NO_NODE)
+		return nid;
+
+	return meminfo_to_nid(&numa_reserved_meminfo, start);
+}
+EXPORT_SYMBOL_GPL(phys_to_target_node);
+
+int memory_add_physaddr_to_nid(u64 start)
+{
+	int nid = meminfo_to_nid(&numa_meminfo, start);
+
+	if (nid == NUMA_NO_NODE)
+		nid = numa_meminfo.blk[0].nid;
 	return nid;
 }
 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
diff --git a/include/linux/numa.h b/include/linux/numa.h
index 5773cd2613fc..a42df804679e 100644
--- a/include/linux/numa.h
+++ b/include/linux/numa.h
@@ -1,7 +1,7 @@
 /* SPDX-License-Identifier: GPL-2.0 */
 #ifndef _LINUX_NUMA_H
 #define _LINUX_NUMA_H
-
+#include <linux/types.h>
 
 #ifdef CONFIG_NODES_SHIFT
 #define NODES_SHIFT     CONFIG_NODES_SHIFT
@@ -21,12 +21,24 @@
 #endif
 
 #ifdef CONFIG_NUMA
+/* Generic implementation available */
 int numa_map_to_online_node(int node);
+
+/*
+ * Optional architecture specific implementation, users need a "depends
+ * on $ARCH"
+ */
+int phys_to_target_node(phys_addr_t addr);
 #else
 static inline int numa_map_to_online_node(int node)
 {
 	return NUMA_NO_NODE;
 }
+
+static inline int phys_to_target_node(phys_addr_t addr)
+{
+	return NUMA_NO_NODE;
+}
 #endif
 
 #endif /* _LINUX_NUMA_H */
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

Re: [PATCH v5 5/6] x86/NUMA: Provide a range-to-target_node lookup facility

[Thomas Gleixner]

Dan Williams <dan.j.williams@intel.com> writes:

> The DEV_DAX_KMEM facility is a generic mechanism to allow device-dax
> instances, fronting performance-differentiated-memory like pmem, to be
> added to the System RAM pool. The NUMA node for that hot-added memory is
> derived from the device-dax instance's 'target_node' attribute.
>
> Recall that the 'target_node' is the ACPI-PXM-to-node translation for
> memory when it comes online whereas the 'numa_node' attribute of the
> device represents the closest online cpu node.
>
> Presently useful target_node information from the ACPI SRAT is discarded
> with the expectation that "Reserved" memory will never be onlined. Now,
> DEV_DAX_KMEM violates that assumption, there is a need to retain the
> translation. Move, rather than discard, numa_memblk data to a secondary
> array that memory_add_physaddr_to_target_node() may consider at a later
> point in time.
>
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Cc: Andy Lutomirski <luto@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Borislav Petkov <bp@alien8.de>
> Cc: "H. Peter Anvin" <hpa@zytor.com>
> Cc: <x86@kernel.org>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: David Hildenbrand <david@redhat.com>
> Cc: Michal Hocko <mhocko@suse.com>
> Reported-by: kbuild test robot <lkp@intel.com>
> Reviewed-by: Ingo Molnar <mingo@kernel.org>
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>

Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

[PATCH v5 6/6] libnvdimm/e820: Retrieve and populate correct 'target_node' info

[Dan Williams]

Use the new phys_to_target_node() and numa_map_to_online_node() helpers
to retrieve the correct id for the 'numa_node' ("local" / online
initiator node) and 'target_node' (offline target memory node) sysfs
attributes.

Below is an example from a 4 NUMA node system where all the memory on
node2 is pmem / reserved. It should be noted that with the arrival of
the ACPI HMAT table and EFI Specific Purpose Memory the kernel will
start to see more platforms with reserved / performance differentiated
memory in its own NUMA node. Hence all the stakeholders on the Cc for
what is ostensibly a libnvdimm local patch.

=== Before ===

/* Notice no online memory on node2 at start */

# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3708 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node   0   1   3
  0:  10  21  21
  1:  21  10  21
  3:  21  21  10

/*
 * Put the pmem namespace into devdax mode so it can be assigned to the
 * kmem driver
 */

# ndctl create-namespace -e namespace0.0 -m devdax -f
{
  "dev":"namespace0.0",
  "mode":"devdax",
  "map":"dev",
  "size":"3.94 GiB (4.23 GB)",
  "uuid":"1650af9b-9ba3-4704-acd6-10178399d9a3",
  [..]
}

/* Online Persistent Memory as System RAM */

# daxctl reconfigure-device --mode=system-ram dax0.0
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
[
  {
    "chardev":"dax0.0",
    "size":4225761280,
    "target_node":0,
    "mode":"system-ram"
  }
]
reconfigured 1 device

/* Note that the memory is onlined by default to the wrong node, node0 */

# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 7926 MB
node 0 free: 7655 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node   0   1   3
  0:  10  21  21
  1:  21  10  21
  3:  21  21  10


=== After ===

/* Notice that the "phys_index" error messages are gone */

# daxctl reconfigure-device --mode=system-ram dax0.0
[
  {
    "chardev":"dax0.0",
    "size":4225761280,
    "target_node":2,
    "mode":"system-ram"
  }
]
reconfigured 1 device

/* Notice that node2 is now correctly populated */

# numactl --hardware
available: 4 nodes (0-3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3793 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3851 MB
node 2 cpus:
node 2 size: 3968 MB
node 2 free: 3968 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3908 MB
node distances:
node   0   1   2   3
  0:  10  21  21  21
  1:  21  10  21  21
  2:  21  21  10  21
  3:  21  21  21  10

Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 arch/x86/Kconfig      |    1 +
 drivers/nvdimm/e820.c |   18 ++++--------------
 2 files changed, 5 insertions(+), 14 deletions(-)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index beea77046f9b..d4e446daf457 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1664,6 +1664,7 @@ config X86_PMEM_LEGACY
 	depends on PHYS_ADDR_T_64BIT
 	depends on BLK_DEV
 	select X86_PMEM_LEGACY_DEVICE
+	select NUMA_KEEP_MEMINFO if NUMA
 	select LIBNVDIMM
 	help
 	  Treat memory marked using the non-standard e820 type of 12 as used
diff --git a/drivers/nvdimm/e820.c b/drivers/nvdimm/e820.c
index e02f60ad6c99..4cd18be9d0e9 100644
--- a/drivers/nvdimm/e820.c
+++ b/drivers/nvdimm/e820.c
@@ -7,6 +7,7 @@
 #include <linux/memory_hotplug.h>
 #include <linux/libnvdimm.h>
 #include <linux/module.h>
+#include <linux/numa.h>
 
 static int e820_pmem_remove(struct platform_device *pdev)
 {
@@ -16,27 +17,16 @@ static int e820_pmem_remove(struct platform_device *pdev)
 	return 0;
 }
 
-#ifdef CONFIG_MEMORY_HOTPLUG
-static int e820_range_to_nid(resource_size_t addr)
-{
-	return memory_add_physaddr_to_nid(addr);
-}
-#else
-static int e820_range_to_nid(resource_size_t addr)
-{
-	return NUMA_NO_NODE;
-}
-#endif
-
 static int e820_register_one(struct resource *res, void *data)
 {
 	struct nd_region_desc ndr_desc;
 	struct nvdimm_bus *nvdimm_bus = data;
+	int nid = phys_to_target_node(res->start);
 
 	memset(&ndr_desc, 0, sizeof(ndr_desc));
 	ndr_desc.res = res;
-	ndr_desc.numa_node = e820_range_to_nid(res->start);
-	ndr_desc.target_node = ndr_desc.numa_node;
+	ndr_desc.numa_node = numa_map_to_online_node(nid);
+	ndr_desc.target_node = nid;
 	set_bit(ND_REGION_PAGEMAP, &ndr_desc.flags);
 	if (!nvdimm_pmem_region_create(nvdimm_bus, &ndr_desc))
 		return -ENXIO;
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org