Skip to contentSkip to conversation

Device tree based NUMA support for Arm - Part#3

PATCH In review Wei Chen 3 participants 34 messages Jan 10, 2023 – Jan 16, 2023

lore
Share & tools
lore

Wei Chen

author
#0 lore raw

v2 cover Device tree based NUMA support for Arm - Part#3

(The Arm device tree based NUMA support patch set contains 35 patches. In order to make stuff easier for reviewers, I split them into 3 parts:
1. Preparation. I have re-sorted the patch series. And moved independent patches to the head of the series - merged in [1]
2. Move generically usable code from x86 to common - merged in [2]
3. Add new code to support Arm - this series.

This series only contains the third part patches. As the whole NUMA series has been reviewed for 1 round in [3], so this series would be v2)

Xen memory allocation and scheduler modules are NUMA aware. But actually, on x86 has implemented the architecture APIs to support NUMA. Arm was providing a set of fake architecture APIs to make it compatible with NUMA awared memory allocation and scheduler.

Arm system was working well as a single node NUMA system with these fake APIs, because we didn't have multiple nodes NUMA system on Arm. But in recent years, more and more Arm devices support multiple nodes NUMA system.

So now we have a new problem. When Xen is running on these Arm devices, Xen still treat them as single node SMP systems. The NUMA affinity capability of Xen memory allocation and scheduler becomes meaningless. Because they rely on input data that does not reflect real NUMA layout.

Xen still think the access time for all of the memory is the same for all CPUs. However, Xen may allocate memory to a VM from different NUMA nodes with different access speeds. This difference can be amplified in workloads inside VM, causing performance instability and timeouts.

So in this patch series, we implement a set of NUMA API to use device tree to describe the NUMA layout. We reuse most of the code of x86 NUMA to create and maintain the mapping between memory and CPU, create the matrix between any two NUMA nodes. Except ACPI and some x86 specified code, we have moved other code to common. In next stage, when we implement ACPI based NUMA for Arm64, we may move the ACPI NUMA code to common too, but in current stage, we keep it as x86 only.

This patch serires has been tested and booted well on one Arm64 NUMA machine and one HPE x86 NUMA machine.

[1] https://lists.xenproject.org/archives/html/xen-devel/2022-06/msg00499.html
[2] https://lists.xenproject.org/archives/html/xen-devel/2022-11/msg01043.html
[3] https://lists.xenproject.org/archives/html/xen-devel/2021-09/msg01903.html

Henry Wang (1):
xen/arm: Set correct per-cpu cpu_core_mask

Wei Chen (16):
xen/arm: use NR_MEM_BANKS to override default NR_NODE_MEMBLKS xen/arm: implement helpers to get and update NUMA status xen/arm: implement node distance helpers for Arm xen/arm: use arch_get_ram_range to memory ranges from bootinfo xen/arm: build NUMA cpu_to_node map in dt_smp_init_cpus xen/arm: Add boot and secondary CPU to NUMA system xen/arm: introduce a helper to parse device tree processor node xen/arm: introduce a helper to parse device tree memory node xen/arm: introduce a helper to parse device tree NUMA distance map xen/arm: unified entry to parse all NUMA data from device tree xen/arm: keep guest still be NUMA unware
xen/arm: enable device tree based NUMA in system init xen/arm: implement numa_node_to_arch_nid for device tree NUMA xen/arm: use CONFIG_NUMA to gate node_online_map in smpboot xen/arm: Provide Kconfig options for Arm to enable NUMA
docs: update numa command line to support Arm

 SUPPORT.md                        |   1 +
 docs/misc/xen-command-line.pandoc |   2 +-
 xen/arch/arm/Kconfig              |  11 ++
 xen/arch/arm/Makefile             |   2 +
 xen/arch/arm/domain_build.c       |   6 +
 xen/arch/arm/include/asm/numa.h   |  87 ++++++++-
 xen/arch/arm/numa.c               | 183 +++++++++++++++++++
 xen/arch/arm/numa_device_tree.c   | 290 ++++++++++++++++++++++++++++++
 xen/arch/arm/setup.c              |  17 ++
 xen/arch/arm/smpboot.c            |  38 ++++
 xen/arch/x86/include/asm/numa.h   |   2 -
 xen/arch/x86/srat.c               |   2 +-
 xen/include/xen/numa.h            |  11 ++
 13 files changed, 647 insertions(+), 5 deletions(-)
 create mode 100644 xen/arch/arm/numa.c
 create mode 100644 xen/arch/arm/numa_device_tree.c
Signature · 1 line
2.25.1
author
#1 lore raw

PATCH v2 1/17 xen/arm: use NR_MEM_BANKS to override default NR_NODE_MEMBLKS

As a memory range described in device tree cannot be split across multiple nodes. And it is very likely than if you have more than 64 nodes, you may need a lot more than 2 regions per node. So the default NR_NODE_MEMBLKS value (MAX_NUMNODES * 2) makes no sense on Arm.

So, for Arm, we would just define NR_NODE_MEMBLKS as an alias to NR_MEM_BANKS. And in the future NR_MEM_BANKS will be user-configurable via kconfig, but for now leave NR_MEM_BANKS as 128 on Arm. This avoid to have different way to define the value based NUMA vs non-NUMA.

Further discussions can be found here[1].

[1] https://lists.xenproject.org/archives/html/xen-devel/2021-09/msg02322.html

---
v1 -> v2:
1. Add code comments to explain using NR_MEM_BANKS for Arm
2. Refine commit messages.

---
 xen/arch/arm/include/asm/numa.h | 19 ++++++++++++++++++-
 xen/include/xen/numa.h          |  9 +++++++++
 2 files changed, 27 insertions(+), 1 deletion(-)
xen/arch/arm/include/asm/numa.h +18 -1
index e2bee2bd82..7d6ae36a19 100644
@@ -3,9 +3,26 @@
 
 #include <xen/mm.h>
 
+#include <asm/setup.h>
+
 typedef u8 nodeid_t;
 
-#ifndef CONFIG_NUMA
+#ifdef CONFIG_NUMA
+
+/*
+ * It is very likely that if you have more than 64 nodes, you may
+ * need a lot more than 2 regions per node. So, for Arm, we would
+ * just define NR_NODE_MEMBLKS as an alias to NR_MEM_BANKS.
+ * And in the future NR_MEM_BANKS will be bumped for new platforms,
+ * but for now leave NR_MEM_BANKS as it is on Arm. This avoid to
+ * have different way to define the value based NUMA vs non-NUMA.
+ *
+ * Further discussions can be found here:
+ * https://lists.xenproject.org/archives/html/xen-devel/2021-09/msg02322.html
+ */
+#define NR_NODE_MEMBLKS NR_MEM_BANKS
+
+#else
 
 /* Fake one node for now. See also node_online_map. */
 #define cpu_to_node(cpu) 0
xen/include/xen/numa.h +9 -0
index 29b8c2df89..b86d0851fc 100644
@@ -13,7 +13,16 @@
 #define MAX_NUMNODES 1
 #endif
 
+/*
+ * Some architectures may have different considerations for
+ * number of node memory blocks. They can define their
+ * NR_NODE_MEMBLKS in asm/numa.h to reflect their architectural
+ * implementation. If the arch does not have specific implementation,
+ * the following default NR_NODE_MEMBLKS will be used.
+ */
+#ifndef NR_NODE_MEMBLKS
 #define NR_NODE_MEMBLKS (MAX_NUMNODES * 2)
+#endif
 
 #define vcpu_to_node(v) (cpu_to_node((v)->processor))
 
Signature · 1 line
2.25.1
#2 lore raw

Re: [PATCH v2 01/17] xen/arm: use NR_MEM_BANKS to override default NR_NODE_MEMBLKS

Quoted 15 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +6 -1
@@ -3,9 +3,26 @@
 
 #include <xen/mm.h>
 
+#include <asm/setup.h>
+
 typedef u8 nodeid_t;
 
-#ifndef CONFIG_NUMA
+#ifdef CONFIG_NUMA
+
+/*
+ * It is very likely that if you have more than 64 nodes, you may

xen/arch/arm/include/asm/numa.h line 23

Quoted 10 lines
xen/arch/arm/include/asm/numa.h +10 -0
+ * need a lot more than 2 regions per node. So, for Arm, we would
+ * just define NR_NODE_MEMBLKS as an alias to NR_MEM_BANKS.
+ * And in the future NR_MEM_BANKS will be bumped for new platforms,
+ * but for now leave NR_MEM_BANKS as it is on Arm. This avoid to
+ * have different way to define the value based NUMA vs non-NUMA.
+ *
+ * Further discussions can be found here:
+ * https://lists.xenproject.org/archives/html/xen-devel/2021-09/msg02322.html
+ */
+#define NR_NODE_MEMBLKS NR_MEM_BANKS

But isn't NR_MEM_BANKS a system-wide setting, which doesn't really make sense to use as a per-node one? IOW aren't you now allowing NR_MEM_BANKS regions on each node, which taken together will be much more than NR_MEM_BANKS that you can deal with on the whole system?

Jan

author
#3 lore raw

RE: [PATCH v2 01/17] xen/arm: use NR_MEM_BANKS to override default NR_NODE_MEMBLKS

Hi Jan,

Quoted 37 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月10日 18:00
To: Wei Chen <[email protected]>
Cc: nd <[email protected]>; Stefano Stabellini <[email protected]>; Julien Grall <[email protected]>; Bertrand Marquis <[email protected]>; Volodymyr Babchuk <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Wei Liu <[email protected]>; [email protected]
Subject: Re: [PATCH v2 01/17] xen/arm: use NR_MEM_BANKS to override default NR_NODE_MEMBLKS

Quoted 13 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +4 -1
@@ -3,9 +3,26 @@
 
 #include <xen/mm.h>
 
+#include <asm/setup.h>
+
 typedef u8 nodeid_t;
 
-#ifndef CONFIG_NUMA
+#ifdef CONFIG_NUMA
+

xen/arch/arm/include/asm/numa.h line 21

Quoted 10 lines
xen/arch/arm/include/asm/numa.h +10 -0
+/*
+ * It is very likely that if you have more than 64 nodes, you may
+ * need a lot more than 2 regions per node. So, for Arm, we would
+ * just define NR_NODE_MEMBLKS as an alias to NR_MEM_BANKS.
+ * And in the future NR_MEM_BANKS will be bumped for new platforms,
+ * but for now leave NR_MEM_BANKS as it is on Arm. This avoid to
+ * have different way to define the value based NUMA vs non-NUMA.
+ *
+ * Further discussions can be found here:
+ * https://lists.xenproject.org/archives/html/xen-devel/2021-

09/msg02322.html

Quoted 9 lines

xen/arch/arm/include/asm/numa.h line 23

Quoted 2 lines
xen/arch/arm/include/asm/numa.h +2 -0
+ */
+#define NR_NODE_MEMBLKS NR_MEM_BANKS

But isn't NR_MEM_BANKS a system-wide setting, which doesn't really make sense to use as a per-node one? IOW aren't you now allowing NR_MEM_BANKS regions on each node, which taken together will be much more than NR_MEM_BANKS that you can deal with on the whole system?

Thanks of pointing out this. Yes NR_MEM_BANKS a system-wide setting, but we just use it to define the MAX memory banks for single node, this does not mean that there are really so many banks on these nodes. When a system only contains one node NR_MEM_BANKS equals to NR_NODE_MEMBLKS. Our idea is that, if the total memory banks of all nodes exceed the NR_MEM_BANKS, we will bump NR_MEM_BANKS.

But I am open to this question, if there are more suggestions from maintainers.

Cheers,
Wei Chen

Quoted 1 line

Jan

author
#4 lore raw

PATCH v2 2/17 xen/arm: implement helpers to get and update NUMA status

In reply to Wei Chen

NUMA has one global and one implementation specific switches. For ACPI NUMA implementation, Xen has acpi_numa, so we introduce device_tree_numa for device tree NUMA implementation. And use enumerations to indicate init, off and on status.

arch_numa_disabled will get device_tree_numa status, but for arch_numa_setup we have not provided boot arguments to setup device_tree_numa. So we just return -EINVAL in this patch.

---
v1 -> v2:
1. Use arch_numa_disabled to replace numa_enable_with_firmware.
2. Introduce enumerations for device tree numa status.
3. Use common numa_disabled, drop Arm version numa_disabled.
4. Introduce arch_numa_setup for Arm.
5. Rename bad_srat to numa_bad.
6. Add numa_enable_with_firmware helper.
7. Add numa_disabled helper.
8. Refine commit message.

---
 xen/arch/arm/include/asm/numa.h | 17 +++++++++++++
 xen/arch/arm/numa.c             | 44 +++++++++++++++++++++++++++++++++
 xen/arch/x86/include/asm/numa.h |  1 -
 xen/include/xen/numa.h          |  1 +
 4 files changed, 62 insertions(+), 1 deletion(-)
 create mode 100644 xen/arch/arm/numa.c
xen/arch/arm/include/asm/numa.h +17 -0
index 7d6ae36a19..52ca414e47 100644
@@ -22,6 +22,12 @@ typedef u8 nodeid_t;
  */
 #define NR_NODE_MEMBLKS NR_MEM_BANKS
 
+enum dt_numa_status {
+    DT_NUMA_INIT,
+    DT_NUMA_ON,
+    DT_NUMA_OFF,
+};
+
 #else
 
 /* Fake one node for now. See also node_online_map. */
@@ -39,6 +45,17 @@ extern mfn_t first_valid_mfn;
 #define node_start_pfn(nid) (mfn_x(first_valid_mfn))
 #define __node_distance(a, b) (20)
 
+#define numa_disabled() (true)
+static inline bool arch_numa_unavailable(void)
+{
+    return true;
+}
+
+static inline bool arch_numa_broken(void)
+{
+    return true;
+}
+
 #endif
 
 #define arch_want_default_dmazone() (false)
xen/arch/arm/numa.c new file +44 -0

44 lines not on this page · raw message

xen/arch/x86/include/asm/numa.h +0 -1
index 7866afa408..61efe60a95 100644
@@ -12,7 +12,6 @@ extern unsigned int numa_node_to_arch_nid(nodeid_t n);
 
 #define ZONE_ALIGN (1UL << (MAX_ORDER+PAGE_SHIFT))
 
-extern bool numa_disabled(void);
 extern nodeid_t setup_node(unsigned int pxm);
 extern void srat_detect_node(int cpu);
 
xen/include/xen/numa.h +1 -0
index b86d0851fc..7d7aeb3a3c 100644
@@ -55,6 +55,7 @@ extern void numa_init_array(void);
 extern void numa_set_node(unsigned int cpu, nodeid_t node);
 extern void numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn);
 extern void numa_fw_bad(void);
+extern bool numa_disabled(void);
 
 extern int arch_numa_setup(const char *opt);
 extern bool arch_numa_unavailable(void);
Signature · 1 line
2.25.1
#5 lore raw

Re: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 26

Quoted 8 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +2 -0
@@ -22,6 +22,12 @@ typedef u8 nodeid_t;
  */
 #define NR_NODE_MEMBLKS NR_MEM_BANKS
 
+enum dt_numa_status {
+    DT_NUMA_INIT,

I don't see any use of this. I also think the name isn't good, as INIT can be taken for "initializer" as well as "initialized". Suggesting an alternative would require knowing what the future plans with this are; right now ...

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 29

Quoted 3 lines
xen/arch/arm/include/asm/numa.h +3 -0
+    DT_NUMA_ON,
+    DT_NUMA_OFF,
+};

... the other two are also used only in a single file, at which point their placing in a header is also questionable.

Quoted 16 lines
xen/arch/arm/numa.c new file +13 -0
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Arm Architecture support layer for NUMA.
+ *
+ * Copyright (C) 2021 Arm Ltd
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

xen/arch/arm/numa.c line 23

Quoted 10 lines
xen/arch/arm/numa.c new file +10 -0
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+#include <xen/init.h>
+#include <xen/numa.h>
+
+static enum dt_numa_status __read_mostly device_tree_numa;

__ro_after_init?

Quoted 13 lines
xen/arch/x86/include/asm/numa.h +0 -1
@@ -12,7 +12,6 @@ extern unsigned int numa_node_to_arch_nid(nodeid_t n);
 
 #define ZONE_ALIGN (1UL << (MAX_ORDER+PAGE_SHIFT))
 
-extern bool numa_disabled(void);
 extern nodeid_t setup_node(unsigned int pxm);
 extern void srat_detect_node(int cpu);
 

xen/include/xen/numa.h extern void numa_init_array(void); · line 61

Quoted 7 lines
xen/include/xen/numa.h +1 -0
@@ -55,6 +55,7 @@ extern void numa_init_array(void);
 extern void numa_set_node(unsigned int cpu, nodeid_t node);
 extern void numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn);
 extern void numa_fw_bad(void);
+extern bool numa_disabled(void);
 
 extern int arch_numa_setup(const char *opt);
 extern bool arch_numa_unavailable(void);

How is this movement of a declaration related to the subject of the patch?

Jan

author
#6 lore raw

RE: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

Hi Jan,

Quoted 20 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:38
To: Wei Chen <[email protected]>
Cc: nd <[email protected]>; Stefano Stabellini <[email protected]>; Julien Grall <[email protected]>; Bertrand Marquis <[email protected]>; Volodymyr Babchuk <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; xen- [email protected]
Subject: Re: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -22,6 +22,12 @@ typedef u8 nodeid_t;
  */
 #define NR_NODE_MEMBLKS NR_MEM_BANKS
 
Quoted 8 lines

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 26

Quoted 2 lines
xen/arch/arm/include/asm/numa.h +2 -0
+enum dt_numa_status {
+    DT_NUMA_INIT,

I don't see any use of this. I also think the name isn't good, as INIT can be taken for "initializer" as well as "initialized". Suggesting an alternative would require knowing what the future plans with this are; right now ...

static enum dt_numa_status __read_mostly device_tree_numa; We use DT_NUMA_INIT to indicate device_tree_numa is in a default value (System's initial value, hasn't done initialization). Maybe rename it To DT_NUMA_UNINIT be better?

Quoted 7 lines

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 29

Quoted 3 lines
xen/arch/arm/include/asm/numa.h +3 -0
+    DT_NUMA_ON,
+    DT_NUMA_OFF,
+};

... the other two are also used only in a single file, at which point their placing in a header is also questionable.

This is a good point, I will move them to that file.

Quoted 20 lines
Quoted 20 lines
xen/arch/arm/numa.c new file +17 -0
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Arm Architecture support layer for NUMA.
+ *
+ * Copyright (C) 2021 Arm Ltd
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
Quoted 9 lines

xen/arch/arm/numa.c line 23

Quoted 6 lines
xen/arch/arm/numa.c new file +6 -0
+ *
+ */
+#include <xen/init.h>
+#include <xen/numa.h>
+
+static enum dt_numa_status __read_mostly device_tree_numa;

__ro_after_init?

OK.

Quoted 16 lines
Quoted 13 lines
xen/arch/x86/include/asm/numa.h +0 -1
@@ -12,7 +12,6 @@ extern unsigned int numa_node_to_arch_nid(nodeid_t n);
 
 #define ZONE_ALIGN (1UL << (MAX_ORDER+PAGE_SHIFT))
 
-extern bool numa_disabled(void);
 extern nodeid_t setup_node(unsigned int pxm);
 extern void srat_detect_node(int cpu);
 

xen/include/xen/numa.h extern void numa_init_array(void); · line 56

Quoted 2 lines
xen/include/xen/numa.h +0 -0
@@ -55,6 +55,7 @@ extern void numa_init_array(void);
 extern void numa_set_node(unsigned int cpu, nodeid_t node);
 extern void numa_initmem_init(unsigned long start_pfn, unsigned long

end_pfn);

Quoted 8 lines

xen/include/xen/numa.h extern void numa_init_array(void); · line 61

Quoted 5 lines
xen/include/xen/numa.h +1 -0
 extern void numa_fw_bad(void);
+extern bool numa_disabled(void);
 
 extern int arch_numa_setup(const char *opt);
 extern bool arch_numa_unavailable(void);

How is this movement of a declaration related to the subject of the patch?

Can I add some messages in commit log to say something like "As we have Implemented numa_disabled for Arm, so we move numa_disabled to common header"?

Cheers,
Wei Chen

Quoted 1 line

Jan

#7 lore raw

Re: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

Quoted 20 lines · Wei Chen
Quoted 11 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:38

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -22,6 +22,12 @@ typedef u8 nodeid_t;
  */
 #define NR_NODE_MEMBLKS NR_MEM_BANKS
 
Quoted 8 lines

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 26

Quoted 2 lines
xen/arch/arm/include/asm/numa.h +2 -0
+enum dt_numa_status {
+    DT_NUMA_INIT,

I don't see any use of this. I also think the name isn't good, as INIT can be taken for "initializer" as well as "initialized". Suggesting an alternative would require knowing what the future plans with this are; right now ...

Quoted 1 line

static enum dt_numa_status __read_mostly device_tree_numa;

There's no DT_NUMA_INIT here. You _imply_ it having a value of zero.

Quoted 3 lines

We use DT_NUMA_INIT to indicate device_tree_numa is in a default value (System's initial value, hasn't done initialization). Maybe rename it To DT_NUMA_UNINIT be better?

Perhaps, yes.

Quoted 19 lines
Quoted 19 lines
Quoted 13 lines
xen/arch/x86/include/asm/numa.h +0 -1
@@ -12,7 +12,6 @@ extern unsigned int numa_node_to_arch_nid(nodeid_t n);
 
 #define ZONE_ALIGN (1UL << (MAX_ORDER+PAGE_SHIFT))
 
-extern bool numa_disabled(void);
 extern nodeid_t setup_node(unsigned int pxm);
 extern void srat_detect_node(int cpu);
 

xen/include/xen/numa.h extern void numa_init_array(void); · line 56

Quoted 2 lines
xen/include/xen/numa.h +0 -0
@@ -55,6 +55,7 @@ extern void numa_init_array(void);
 extern void numa_set_node(unsigned int cpu, nodeid_t node);
 extern void numa_initmem_init(unsigned long start_pfn, unsigned long

end_pfn);

Quoted 3 lines
xen/include/xen/numa.h +1 -0
 extern void numa_fw_bad(void);
+extern bool numa_disabled(void);
 
Quoted 8 lines
Quoted 5 lines

xen/include/xen/numa.h extern void numa_init_array(void); · line 61

Quoted 2 lines
xen/include/xen/numa.h +0 -0
 extern int arch_numa_setup(const char *opt);
 extern bool arch_numa_unavailable(void);

How is this movement of a declaration related to the subject of the patch?

Can I add some messages in commit log to say something like "As we have Implemented numa_disabled for Arm, so we move numa_disabled to common header"?

See your own patch 3, where you have a similar statement (albeit you mean "declaration" there, not "definition"). However, right now numa_disabled() is a #define on Arm, so the declaration becoming common isn't really warranted. In fact it'll get in the way of converting function-like macros to inline functions for Misra.

Jan

author
#8 lore raw

Re: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

Hi Jan,

Quoted 21 lines · Jan Beulich
Quoted 20 lines · Wei Chen
Quoted 19 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:38

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 26

Quoted 8 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +2 -0
@@ -22,6 +22,12 @@ typedef u8 nodeid_t;
   */
  #define NR_NODE_MEMBLKS NR_MEM_BANKS
 
+enum dt_numa_status {
+    DT_NUMA_INIT,

I don't see any use of this. I also think the name isn't good, as INIT can be taken for "initializer" as well as "initialized". Suggesting an alternative would require knowing what the future plans with this are; right now ...

Quoted 4 lines
Quoted 1 line

static enum dt_numa_status __read_mostly device_tree_numa;

There's no DT_NUMA_INIT here. You _imply_ it having a value of zero.

How about I assign device_tree_numa explicitly like:
... __read_mostly device_tree_numa = DT_NUMA_UNINIT;

Quoted 31 lines
Quoted 3 lines

We use DT_NUMA_INIT to indicate device_tree_numa is in a default value (System's initial value, hasn't done initialization). Maybe rename it To DT_NUMA_UNINIT be better?

Perhaps, yes.

Quoted 25 lines
Quoted 24 lines
Quoted 13 lines
xen/arch/x86/include/asm/numa.h +0 -1
@@ -12,7 +12,6 @@ extern unsigned int numa_node_to_arch_nid(nodeid_t n);
 
  #define ZONE_ALIGN (1UL << (MAX_ORDER+PAGE_SHIFT))
 
-extern bool numa_disabled(void);
  extern nodeid_t setup_node(unsigned int pxm);
  extern void srat_detect_node(int cpu);
 

xen/include/xen/numa.h extern void numa_init_array(void); · line 56

Quoted 2 lines
xen/include/xen/numa.h +0 -0
@@ -55,6 +55,7 @@ extern void numa_init_array(void);
  extern void numa_set_node(unsigned int cpu, nodeid_t node);
  extern void numa_initmem_init(unsigned long start_pfn, unsigned long

end_pfn);

xen/include/xen/numa.h extern void numa_init_array(void); · line 61

Quoted 5 lines
xen/include/xen/numa.h +1 -0
  extern void numa_fw_bad(void);
+extern bool numa_disabled(void);
 
  extern int arch_numa_setup(const char *opt);
  extern bool arch_numa_unavailable(void);

How is this movement of a declaration related to the subject of the patch?

Quoted 9 lines
Quoted 2 lines

Can I add some messages in commit log to say something like "As we have Implemented numa_disabled for Arm, so we move numa_disabled to common header"?

See your own patch 3, where you have a similar statement (albeit you mean "declaration" there, not "definition"). However, right now numa_disabled() is a #define on Arm, so the declaration becoming common isn't really warranted. In fact it'll get in the way of converting function-like macros to inline functions for Misra.

Yes, I think you're right. This may also seem a little strange,when we disable NUMA, there are two headers have numa_disabled statement. I will revert this change. And I also will covert the macro to a static inline function.

Cheers,
Wei Chen

Quoted 1 line

Jan

#9 lore raw

Re: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

Quoted 24 lines · Wei Chen

Hi Jan,

Quoted 21 lines · Jan Beulich
Quoted 20 lines · Wei Chen
Quoted 19 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:38

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 26

Quoted 8 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +2 -0
@@ -22,6 +22,12 @@ typedef u8 nodeid_t;
   */
  #define NR_NODE_MEMBLKS NR_MEM_BANKS
 
+enum dt_numa_status {
+    DT_NUMA_INIT,

I don't see any use of this. I also think the name isn't good, as INIT can be taken for "initializer" as well as "initialized". Suggesting an alternative would require knowing what the future plans with this are; right now ...

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

static enum dt_numa_status __read_mostly device_tree_numa;

There's no DT_NUMA_INIT here. You _imply_ it having a value of zero.

How about I assign device_tree_numa explicitly like:
... __read_mostly device_tree_numa = DT_NUMA_UNINIT;

Well, yes, this is what I was asking for when mentioning the lack of use of the enumerator. Irrespective of that I remain unhappy with the name, though.

Jan

author
#10 lore raw

RE: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

Hi Jan,

Quoted 45 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月16日 19:15
To: Wei Chen <[email protected]>
Cc: nd <[email protected]>; Stefano Stabellini <[email protected]>; Julien Grall <[email protected]>; Bertrand Marquis <[email protected]>; Volodymyr Babchuk <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; xen- [email protected]
Subject: Re: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

Quoted 19 lines · Wei Chen

Hi Jan,

Quoted 16 lines · Jan Beulich
Quoted 15 lines · Wei Chen
Quoted 15 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:38

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 26

Quoted 8 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +2 -0
@@ -22,6 +22,12 @@ typedef u8 nodeid_t;
   */
  #define NR_NODE_MEMBLKS NR_MEM_BANKS
 
+enum dt_numa_status {
+    DT_NUMA_INIT,

I don't see any use of this. I also think the name isn't good, as

INIT

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

can be taken for "initializer" as well as "initialized". Suggesting

an

Quoted 9 lines
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

alternative would require knowing what the future plans with this are; right now ...

static enum dt_numa_status __read_mostly device_tree_numa;

There's no DT_NUMA_INIT here. You _imply_ it having a value of zero.

Quoted 7 lines
Quoted 2 lines

How about I assign device_tree_numa explicitly like:
... __read_mostly device_tree_numa = DT_NUMA_UNINIT;

Well, yes, this is what I was asking for when mentioning the lack of use of the enumerator. Irrespective of that I remain unhappy with the name, though.

How about DT_NUMA_DEF or do you have some suggestions for the name?

Cheers,
Wei Chen

Quoted 1 line

Jan

#11 lore raw

Re: [PATCH v2 02/17] xen/arm: implement helpers to get and update NUMA status

Quoted 37 lines · Wei Chen
Quoted 37 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月16日 19:15

Quoted 13 lines · Wei Chen
Quoted 12 lines · Jan Beulich
Quoted 11 lines · Wei Chen
Quoted 11 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:38

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -22,6 +22,12 @@ typedef u8 nodeid_t;
   */
  #define NR_NODE_MEMBLKS NR_MEM_BANKS
 
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

xen/arch/arm/include/asm/numa.h typedef u8 nodeid_t; · line 26

Quoted 2 lines
xen/arch/arm/include/asm/numa.h +2 -0
+enum dt_numa_status {
+    DT_NUMA_INIT,

I don't see any use of this. I also think the name isn't good, as

INIT

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

can be taken for "initializer" as well as "initialized". Suggesting

an

Quoted 11 lines
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

alternative would require knowing what the future plans with this are; right now ...

static enum dt_numa_status __read_mostly device_tree_numa;

There's no DT_NUMA_INIT here. You _imply_ it having a value of zero.

How about I assign device_tree_numa explicitly like:
... __read_mostly device_tree_numa = DT_NUMA_UNINIT;

Quoted 6 lines
Quoted 4 lines

Well, yes, this is what I was asking for when mentioning the lack of use of the enumerator. Irrespective of that I remain unhappy with the name, though.

How about DT_NUMA_DEF or do you have some suggestions for the name?

Yeah, "DEFAULT" is probably the least bad one.

Jan

author
#12 lore raw

PATCH v2 3/17 xen/arm: implement node distance helpers for Arm

In reply to Wei Chen

We will parse NUMA nodes distances from device tree. So we need a matrix to record the distances between any two nodes we parsed. Accordingly, we provide this node_set_distance API for device tree NUMA to set the distance for any two nodes in this patch. When NUMA initialization failed, __node_distance will return NUMA_REMOTE_DISTANCE, this will help us avoid doing rollback for distance maxtrix when NUMA initialization failed.

As both x86 and Arm have implemented __node_distance, so we move its definition from asm/numa.h to xen/numa.h. At same time, the outdated u8 return value of x86 has been changed to unsigned char.

---
v1 -> v2:
1. Use unsigned int/char instead of uint32_t/u8.
2. Re-org the commit message.

---
 xen/arch/arm/Makefile           |  1 +
 xen/arch/arm/include/asm/numa.h | 14 +++++++++
 xen/arch/arm/numa.c             | 52 ++++++++++++++++++++++++++++++++-
 xen/arch/x86/include/asm/numa.h |  1 -
 xen/arch/x86/srat.c             |  2 +-
 xen/include/xen/numa.h          |  1 +
 6 files changed, 68 insertions(+), 3 deletions(-)
xen/arch/arm/Makefile +1 -0
index 4d076b278b..9073398d6e 100644
@@ -38,6 +38,7 @@ obj-$(CONFIG_LIVEPATCH) += livepatch.o
 obj-y += mem_access.o
 obj-y += mm.o
 obj-y += monitor.o
+obj-$(CONFIG_NUMA) += numa.o
 obj-y += p2m.o
 obj-y += percpu.o
 obj-y += platform.o
xen/arch/arm/include/asm/numa.h +14 -0
index 52ca414e47..dbdb632711 100644
@@ -28,6 +28,20 @@ enum dt_numa_status {
     DT_NUMA_OFF,
 };
 
+/*
+ * In ACPI spec, 0-9 are the reserved values for node distance,
+ * 10 indicates local node distance, 20 indicates remote node
+ * distance. Set node distance map in device tree will follow
+ * the ACPI's definition.
+ */
+#define NUMA_DISTANCE_UDF_MIN   0
+#define NUMA_DISTANCE_UDF_MAX   9
+#define NUMA_LOCAL_DISTANCE     10
+#define NUMA_REMOTE_DISTANCE    20
+
+extern void numa_set_distance(nodeid_t from, nodeid_t to,
+                              unsigned int distance);
+
 #else
 
 /* Fake one node for now. See also node_online_map. */
xen/arch/arm/numa.c +51 -1

67 lines not on this page · raw message

xen/arch/x86/include/asm/numa.h +0 -1
index 61efe60a95..18b71ddfef 100644
@@ -21,7 +21,6 @@ extern void init_cpu_to_node(void);
 #define arch_want_default_dmazone() (num_online_nodes() > 1)
 
 void srat_parse_regions(paddr_t addr);
-extern u8 __node_distance(nodeid_t a, nodeid_t b);
 unsigned int arch_get_dma_bitsize(void);
 
 #endif
xen/arch/x86/srat.c +1 -1
index 56749ddca5..50faf5d352 100644
@@ -328,7 +328,7 @@ unsigned int numa_node_to_arch_nid(nodeid_t n)
 	return 0;
 }
 
-u8 __node_distance(nodeid_t a, nodeid_t b)
+unsigned char __node_distance(nodeid_t a, nodeid_t b)
 {
 	unsigned index;
 	u8 slit_val;
xen/include/xen/numa.h +1 -0
index 7d7aeb3a3c..cff4fb8ccc 100644
@@ -115,6 +115,7 @@ extern bool numa_memblks_available(void);
 extern bool numa_update_node_memblks(nodeid_t node, unsigned int arch_nid,
                                      paddr_t start, paddr_t size, bool hotplug);
 extern void numa_set_processor_nodes_parsed(nodeid_t node);
+extern unsigned char __node_distance(nodeid_t a, nodeid_t b);
 
 #else
 
Signature · 1 line
2.25.1
#13 lore raw

Re: [PATCH v2 03/17] xen/arm: implement node distance helpers for Arm

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -28,6 +28,20 @@ enum dt_numa_status {
     DT_NUMA_OFF,
 };
 

xen/arch/arm/include/asm/numa.h enum dt_numa_status { · line 40

Quoted 10 lines
xen/arch/arm/include/asm/numa.h +10 -0
+/*
+ * In ACPI spec, 0-9 are the reserved values for node distance,
+ * 10 indicates local node distance, 20 indicates remote node
+ * distance. Set node distance map in device tree will follow
+ * the ACPI's definition.
+ */
+#define NUMA_DISTANCE_UDF_MIN   0
+#define NUMA_DISTANCE_UDF_MAX   9
+#define NUMA_LOCAL_DISTANCE     10
+#define NUMA_REMOTE_DISTANCE    20

In the absence of a caller of numa_set_distance() it is entirely unclear whether this tying to ACPI used values is actually appropriate.

xen/arch/arm/numa.c line 5

Quoted 8 lines
xen/arch/arm/numa.c +1 -1
@@ -2,7 +2,7 @@
 /*
  * Arm Architecture support layer for NUMA.
  *
- * Copyright (C) 2021 Arm Ltd
+ * Copyright (C) 2022 Arm Ltd

I don't think it makes sense to change such after the fact. And certainly not in an unrelated patch.

xen/arch/arm/numa.c line 28

Quoted 8 lines
xen/arch/arm/numa.c +4 -0
@@ -22,6 +22,11 @@
 
 static enum dt_numa_status __read_mostly device_tree_numa;
 
+static unsigned char __read_mostly
+node_distance_map[MAX_NUMNODES][MAX_NUMNODES] = {
+    { 0 }
+};

__ro_after_init?

Quoted 9 lines
xen/arch/arm/numa.c +5 -0
@@ -42,3 +47,48 @@ int __init arch_numa_setup(const char *opt)
 {
     return -EINVAL;
 }
+
+void __init numa_set_distance(nodeid_t from, nodeid_t to,
+                              unsigned int distance)
+{
+    if ( from >= MAX_NUMNODES || to >= MAX_NUMNODES )

xen/arch/arm/numa.c int __init arch_numa_setup(const char *opt) · line 64

Quoted 10 lines
xen/arch/arm/numa.c +10 -0
+    {
+        printk(KERN_WARNING
+               "NUMA: invalid nodes: from=%"PRIu8" to=%"PRIu8" MAX=%"PRIu8"\n",
+               from, to, MAX_NUMNODES);
+        return;
+    }
+
+    /* NUMA defines 0xff as an unreachable node and 0-9 are undefined */
+    if ( distance >= NUMA_NO_DISTANCE ||
+        (distance >= NUMA_DISTANCE_UDF_MIN &&

Nit: Indentation.

Quoted 14 lines
xen/arch/arm/numa.c +14 -0
+         distance <= NUMA_DISTANCE_UDF_MAX) ||
+        (from == to && distance != NUMA_LOCAL_DISTANCE) )
+    {
+        printk(KERN_WARNING
+               "NUMA: invalid distance: from=%"PRIu8" to=%"PRIu8" distance=%"PRIu32"\n",
+               from, to, distance);
+        return;
+    }
+
+    node_distance_map[from][to] = distance;
+}
+
+unsigned char __node_distance(nodeid_t from, nodeid_t to)
+{

xen/arch/arm/numa.c int __init arch_numa_setup(const char *opt) · line 88

Quoted 10 lines
xen/arch/arm/numa.c +10 -0
+    /* When NUMA is off, any distance will be treated as remote. */
+    if ( numa_disabled() )
+        return NUMA_REMOTE_DISTANCE;
+
+    /*
+     * Check whether the nodes are in the matrix range.
+     * When any node is out of range, except from and to nodes are the
+     * same, we treat them as unreachable (return 0xFF)
+     */
+    if ( from >= MAX_NUMNODES || to >= MAX_NUMNODES )

I guess using ARRAY_SIZE() here would be more future-proof.

Jan

author
#14 lore raw

RE: [PATCH v2 03/17] xen/arm: implement node distance helpers for Arm

Hi Jan,

Quoted 20 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:47
To: Wei Chen <[email protected]>
Cc: nd <[email protected]>; Stefano Stabellini <[email protected]>; Julien Grall <[email protected]>; Bertrand Marquis <[email protected]>; Volodymyr Babchuk <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; xen- [email protected]
Subject: Re: [PATCH v2 03/17] xen/arm: implement node distance helpers for Arm

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -28,6 +28,20 @@ enum dt_numa_status {
     DT_NUMA_OFF,
 };
 
Quoted 14 lines

xen/arch/arm/include/asm/numa.h enum dt_numa_status { · line 40

Quoted 10 lines
xen/arch/arm/include/asm/numa.h +10 -0
+/*
+ * In ACPI spec, 0-9 are the reserved values for node distance,
+ * 10 indicates local node distance, 20 indicates remote node
+ * distance. Set node distance map in device tree will follow
+ * the ACPI's definition.
+ */
+#define NUMA_DISTANCE_UDF_MIN   0
+#define NUMA_DISTANCE_UDF_MAX   9
+#define NUMA_LOCAL_DISTANCE     10
+#define NUMA_REMOTE_DISTANCE    20

In the absence of a caller of numa_set_distance() it is entirely unclear whether this tying to ACPI used values is actually appropriate.

From Kernel's NUMA device tree binding, it seems DT NUMA are reusing ACPI used values for distances [1].

Quoted 12 lines

xen/arch/arm/numa.c line 5

Quoted 8 lines
xen/arch/arm/numa.c +1 -1
@@ -2,7 +2,7 @@
 /*
  * Arm Architecture support layer for NUMA.
  *
- * Copyright (C) 2021 Arm Ltd
+ * Copyright (C) 2022 Arm Ltd

I don't think it makes sense to change such after the fact. And certainly not in an unrelated patch.

I will retore it, and add a SPDX header.

Quoted 11 lines

xen/arch/arm/numa.c line 28

Quoted 8 lines
xen/arch/arm/numa.c +4 -0
@@ -22,6 +22,11 @@
 
 static enum dt_numa_status __read_mostly device_tree_numa;
 
+static unsigned char __read_mostly
+node_distance_map[MAX_NUMNODES][MAX_NUMNODES] = {
+    { 0 }
+};

__ro_after_init?

Yes.

Quoted 18 lines

xen/arch/arm/numa.c int __init arch_numa_setup(const char *opt) · line 57

Quoted 12 lines
xen/arch/arm/numa.c +8 -0
@@ -42,3 +47,48 @@ int __init arch_numa_setup(const char *opt)
 {
     return -EINVAL;
 }
+
+void __init numa_set_distance(nodeid_t from, nodeid_t to,
+                              unsigned int distance)
+{
+    if ( from >= MAX_NUMNODES || to >= MAX_NUMNODES )
+    {
+        printk(KERN_WARNING
+               "NUMA: invalid nodes: from=%"PRIu8" to=%"PRIu8"

MAX=%"PRIu8"\n",

xen/arch/arm/numa.c int __init arch_numa_setup(const char *opt)

Quoted 5 lines
xen/arch/arm/numa.c +5 -0
+               from, to, MAX_NUMNODES);
+        return;
+    }
+
+    /* NUMA defines 0xff as an unreachable node and 0-9 are undefined
Quoted 6 lines

*/

xen/arch/arm/numa.c int __init arch_numa_setup(const char *opt) · line 64

Quoted 2 lines
xen/arch/arm/numa.c +2 -0
+    if ( distance >= NUMA_NO_DISTANCE ||
+        (distance >= NUMA_DISTANCE_UDF_MIN &&

Nit: Indentation.

Ok.

Quoted 19 lines

xen/arch/arm/numa.c int __init arch_numa_setup(const char *opt)

Quoted 5 lines
xen/arch/arm/numa.c +5 -0
+         distance <= NUMA_DISTANCE_UDF_MAX) ||
+        (from == to && distance != NUMA_LOCAL_DISTANCE) )
+    {
+        printk(KERN_WARNING
+               "NUMA: invalid distance: from=%"PRIu8" to=%"PRIu8"

distance=%"PRIu32"\n",

Quoted 13 lines
xen/arch/arm/numa.c +13 -0
+               from, to, distance);
+        return;
+    }
+
+    node_distance_map[from][to] = distance;
+}
+
+unsigned char __node_distance(nodeid_t from, nodeid_t to)
+{
+    /* When NUMA is off, any distance will be treated as remote. */
+    if ( numa_disabled() )
+        return NUMA_REMOTE_DISTANCE;
+
Quoted 9 lines

xen/arch/arm/numa.c int __init arch_numa_setup(const char *opt) · line 88

Quoted 6 lines
xen/arch/arm/numa.c +6 -0
+    /*
+     * Check whether the nodes are in the matrix range.
+     * When any node is out of range, except from and to nodes are the
+     * same, we treat them as unreachable (return 0xFF)
+     */
+    if ( from >= MAX_NUMNODES || to >= MAX_NUMNODES )

I guess using ARRAY_SIZE() here would be more future-proof.

I will use it in next version.

[1]https://www.kernel.org/doc/Documentation/devicetree/bindings/numa.txt

Thanks,
Wei Chen

Quoted 1 line

Jan

#15 lore raw

Re: [PATCH v2 03/17] xen/arm: implement node distance helpers for Arm

Quoted 26 lines · Wei Chen
Quoted 11 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:47

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -28,6 +28,20 @@ enum dt_numa_status {
     DT_NUMA_OFF,
 };
 
Quoted 14 lines

xen/arch/arm/include/asm/numa.h enum dt_numa_status { · line 40

Quoted 10 lines
xen/arch/arm/include/asm/numa.h +10 -0
+/*
+ * In ACPI spec, 0-9 are the reserved values for node distance,
+ * 10 indicates local node distance, 20 indicates remote node
+ * distance. Set node distance map in device tree will follow
+ * the ACPI's definition.
+ */
+#define NUMA_DISTANCE_UDF_MIN   0
+#define NUMA_DISTANCE_UDF_MAX   9
+#define NUMA_LOCAL_DISTANCE     10
+#define NUMA_REMOTE_DISTANCE    20

In the absence of a caller of numa_set_distance() it is entirely unclear whether this tying to ACPI used values is actually appropriate.

Quoted 2 lines

From Kernel's NUMA device tree binding, it seems DT NUMA are reusing ACPI used values for distances [1].

I can't find any mention of ACPI in that doc, so the example values used there matching ACPI's may also be coincidental. In no event can a Linux kernel doc serve as DT specification. If values are to match ACPI's, I expect a DT spec to actually say so.

Jan

author
#16 lore raw

RE: [PATCH v2 03/17] xen/arm: implement node distance helpers for Arm

Hi Jan,

Quoted 41 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月12日 16:11
To: Wei Chen <[email protected]>
Cc: nd <[email protected]>; Stefano Stabellini <[email protected]>; Julien Grall <[email protected]>; Bertrand Marquis <[email protected]>; Volodymyr Babchuk <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; xen- [email protected]
Subject: Re: [PATCH v2 03/17] xen/arm: implement node distance helpers for Arm

Quoted 11 lines · Wei Chen
Quoted 11 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:47

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -28,6 +28,20 @@ enum dt_numa_status {
     DT_NUMA_OFF,
 };
 
Quoted 12 lines
Quoted 4 lines
Quoted 4 lines
xen/arch/arm/include/asm/numa.h +4 -0
+/*
+ * In ACPI spec, 0-9 are the reserved values for node distance,
+ * 10 indicates local node distance, 20 indicates remote node
+ * distance. Set node distance map in device tree will follow
Quoted 8 lines

xen/arch/arm/include/asm/numa.h enum dt_numa_status { · line 40

Quoted 6 lines
xen/arch/arm/include/asm/numa.h +6 -0
+ * the ACPI's definition.
+ */
+#define NUMA_DISTANCE_UDF_MIN   0
+#define NUMA_DISTANCE_UDF_MAX   9
+#define NUMA_LOCAL_DISTANCE     10
+#define NUMA_REMOTE_DISTANCE    20

In the absence of a caller of numa_set_distance() it is entirely

unclear

Quoted 3 lines
Quoted 2 lines

whether this tying to ACPI used values is actually appropriate.

Quoted 8 lines
Quoted 2 lines

From Kernel's NUMA device tree binding, it seems DT NUMA are reusing ACPI used values for distances [1].

I can't find any mention of ACPI in that doc, so the example values used there matching ACPI's may also be coincidental. In no event can a Linux kernel doc serve as DT specification. If values are to match ACPI's, I expect a DT spec to actually say so.

Unfortunately, the latest device tree spec doesn't have any NUMA description [1]. But if we define different values for DT NUMA in Xen, we may have an incompatible with Linux DT.

[1] https://github.com/devicetree-org/devicetree-specification/releases/download/v0.4-rc1/devicetree-specification-v0.4-rc1.pdf

Cheers,
Wei Chen

Quoted 1 line

Jan

#17 lore raw

Re: [PATCH v2 03/17] xen/arm: implement node distance helpers for Arm

In reply to Jan Beulich

Hi,

Quoted 12 lines · Jan Beulich
Quoted 11 lines · Wei Chen
Quoted 11 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:47

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -28,6 +28,20 @@ enum dt_numa_status {
      DT_NUMA_OFF,
  };
 
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

xen/arch/arm/include/asm/numa.h enum dt_numa_status { · line 35

Quoted 5 lines
xen/arch/arm/include/asm/numa.h +5 -0
+/*
+ * In ACPI spec, 0-9 are the reserved values for node distance,
+ * 10 indicates local node distance, 20 indicates remote node
+ * distance. Set node distance map in device tree will follow
+ * the ACPI's definition.

Looking at the ACPI spec, I agree that the local node distance is defined to 10. But I couldn't find any mention of the value 20.

How is NUMA_REMOTE_DISTANCE is meant to be used? Is this a default value? If so, maybe we should add "DEFAULT" in the name.

Quoted 10 lines
Quoted 10 lines
Quoted 9 lines

xen/arch/arm/include/asm/numa.h enum dt_numa_status { · line 40

Quoted 5 lines
xen/arch/arm/include/asm/numa.h +5 -0
+ */
+#define NUMA_DISTANCE_UDF_MIN   0
+#define NUMA_DISTANCE_UDF_MAX   9
+#define NUMA_LOCAL_DISTANCE     10
+#define NUMA_REMOTE_DISTANCE    20

In the absence of a caller of numa_set_distance() it is entirely unclear whether this tying to ACPI used values is actually appropriate.

Quoted 6 lines
Quoted 2 lines

From Kernel's NUMA device tree binding, it seems DT NUMA are reusing
ACPI used values for distances [1].

I can't find any mention of ACPI in that doc, so the example values used there matching ACPI's may also be coincidental. In no event can a Linux kernel doc serve as DT specification.

The directory Documentation/devicetree is the de-facto place where all the bindings are described. This is used by most (to not say all) users.

I vaguely remember there were plan in the past to move the bindings out of the kernel. But it looks like this has not been done. Yet, they tend to be reviewed independently from the kernel.

So, as Wei pointed, if we don't follow them then we will not be able to re-use Device-Tree shipped.

Quoted 2 lines

If values are to match ACPI's, I
expect a DT spec to actually say so.

I don't think it is necessary to say that. Bindings are not meant to change and a developer can rely on the local distance value to not change with the current bindings.

So IMHO, it is OK to assume that the local distance value is the same between ACPI and DT. That would definitely simplify the parsing and code.

Cheers,

Signature · 1 line
Julien Grall
author
#18 lore raw

Re: [PATCH v2 03/17] xen/arm: implement node distance helpers for Arm

Hi Julien,

Quoted 23 lines · Julien Grall

Hi,

Quoted 12 lines · Jan Beulich
Quoted 11 lines · Wei Chen
Quoted 11 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 2023年1月11日 0:47

Quoted 6 lines · Wei Chen
xen/arch/arm/include/asm/numa.h +0 -0
@@ -28,6 +28,20 @@ enum dt_numa_status {
       DT_NUMA_OFF,
   };
 
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

xen/arch/arm/include/asm/numa.h enum dt_numa_status { · line 35

Quoted 5 lines
xen/arch/arm/include/asm/numa.h +5 -0
+/*
+ * In ACPI spec, 0-9 are the reserved values for node distance,
+ * 10 indicates local node distance, 20 indicates remote node
+ * distance. Set node distance map in device tree will follow
+ * the ACPI's definition.

Looking at the ACPI spec, I agree that the local node distance is defined to 10. But I couldn't find any mention of the value 20.

Quoted 2 lines

How is NUMA_REMOTE_DISTANCE is meant to be used? Is this a default value? If so, maybe we should add "DEFAULT" in the name.

I think you're right, maybe we can rename NUMA_REMOTE_DISTANCE
to NUMA_DEFAULT_DISTANCE.Different pairs of nodes may have different remote distance values, I can not define one value for them.

And why use 20 as default value, I have followed the ACPI's logic. In ACPI NUMA, when ACPI doesn't provide SLIT table, 20 will be used for all nodes default distance.

Quoted 21 lines
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines

xen/arch/arm/include/asm/numa.h enum dt_numa_status { · line 40

Quoted 5 lines
xen/arch/arm/include/asm/numa.h +5 -0
+ */
+#define NUMA_DISTANCE_UDF_MIN   0
+#define NUMA_DISTANCE_UDF_MAX   9
+#define NUMA_LOCAL_DISTANCE     10
+#define NUMA_REMOTE_DISTANCE    20

In the absence of a caller of numa_set_distance() it is entirely unclear whether this tying to ACPI used values is actually appropriate.

Quoted 6 lines
Quoted 2 lines

 From Kernel's NUMA device tree binding, it seems DT NUMA are reusing ACPI used values for distances [1].

I can't find any mention of ACPI in that doc, so the example values used there matching ACPI's may also be coincidental. In no event can a Linux kernel doc serve as DT specification.

The directory Documentation/devicetree is the de-facto place where all the bindings are described. This is used by most (to not say all) users.

I vaguely remember there were plan in the past to move the bindings out of the kernel. But it looks like this has not been done. Yet, they tend to be reviewed independently from the kernel.

Quoted 8 lines

So, as Wei pointed, if we don't follow them then we will not be able to re-use Device-Tree shipped.

Quoted 2 lines

If values are to match ACPI's, I
expect a DT spec to actually say so.

I don't think it is necessary to say that. Bindings are not meant to change and a developer can rely on the local distance value to not change with the current bindings.

So IMHO, it is OK to assume that the local distance value is the same between ACPI and DT. That would definitely simplify the parsing and code.

Thanks for clarifying this.

Cheers,
Wei Chen

Quoted 2 lines

Cheers,

author
#19 lore raw

PATCH v2 4/17 xen/arm: use arch_get_ram_range to memory ranges from bootinfo

In reply to Wei Chen

Implement the same helper "arch_get_ram_range" as x86 for NUMA code to get memory bank from Arm bootinfo.

---
v2 -> v3:
1. Use arch_get_ram_range instead of arch_get_memory_map. v1 -> v2:
1. Use arch_get_memory_map to replace arch_get_memory_bank_range and arch_get_memory_bank_number.

---
 xen/arch/arm/numa.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
xen/arch/arm/numa.c +11 -0
index 34851ceacf..dcfcd85fcf 100644
@@ -92,3 +92,14 @@ unsigned char __node_distance(nodeid_t from, nodeid_t to)
 }
 
 EXPORT_SYMBOL(__node_distance);
+
+int __init arch_get_ram_range(unsigned int idx, paddr_t *start, paddr_t *end)
+{
+    if ( idx >= bootinfo.mem.nr_banks )
+        return -ENOENT;
+
+    *start = bootinfo.mem.bank[idx].start;
+    *end = *start + bootinfo.mem.bank[idx].size;
+
+    return 0;
+}
Signature · 1 line
2.25.1
author
#20 lore raw

PATCH v2 5/17 xen/arm: build NUMA cpu_to_node map in dt_smp_init_cpus

In reply to Wei Chen

NUMA implementation has a cpu_to_node array to store CPU to NODE map. Xen is using CPU logical ID in runtime components, so we use CPU logical ID as CPU index in cpu_to_node.

In device tree case, cpu_logical_map is created in dt_smp_init_cpus. So, when NUMA is enabled, dt_smp_init_cpus will fetch CPU NUMA id at the same time for cpu_to_node.

---
v1 -> v2:
1. Use static inline to replace macros to perform function paramerters type check.
2. Add numa_disabled to gate the numa-node-id check for CONFIG_NUMA on but numa disabled user case.
3. Use macro instead of static inline function to stub numa_set_node.

---
 xen/arch/arm/include/asm/numa.h |  4 ++++
 xen/arch/arm/smpboot.c          | 36 +++++++++++++++++++++++++++++++++
 2 files changed, 40 insertions(+)
xen/arch/arm/include/asm/numa.h +4 -0
index dbdb632711..3bc28416b4 100644
@@ -70,6 +70,10 @@ static inline bool arch_numa_broken(void)
     return true;
 }
 
+static inline void numa_set_node(unsigned int cpu, nodeid_t node)
+{
+}
+
 #endif
 
 #define arch_want_default_dmazone() (false)
xen/arch/arm/smpboot.c +36 -0
index 412ae22869..5ee6ab11e9 100644
@@ -118,7 +118,12 @@ static void __init dt_smp_init_cpus(void)
     {
         [0 ... NR_CPUS - 1] = MPIDR_INVALID
     };
+    static nodeid_t node_map[NR_CPUS] __initdata =
+    {
+        [0 ... NR_CPUS - 1] = NUMA_NO_NODE
+    };
     bool bootcpu_valid = false;
+    unsigned int nid = 0;
     int rc;
 
     mpidr = system_cpuinfo.mpidr.bits & MPIDR_HWID_MASK;
@@ -169,6 +174,28 @@ static void __init dt_smp_init_cpus(void)
             continue;
         }
 
+        if ( IS_ENABLED(CONFIG_NUMA) )
+        {
+            /*
+             * When CONFIG_NUMA is set, try to fetch numa infomation
+             * from CPU dts node, otherwise the nid is always 0.
+             */
+            if ( !dt_property_read_u32(cpu, "numa-node-id", &nid) )
+            {
+                printk(XENLOG_WARNING
+                       "cpu[%d] dts path: %s: doesn't have numa information!\n",
+                       cpuidx, dt_node_full_name(cpu));
+                /*
+                 * During the early stage of NUMA initialization, when Xen
+                 * found any CPU dts node doesn't have numa-node-id info, the
+                 * NUMA will be treated as off, all CPU will be set to a FAKE
+                 * node 0. So if we get numa-node-id failed here, we should
+                 * set nid to 0.
+                 */
+                nid = 0;
+            }
+        }
+
         /*
          * 8 MSBs must be set to 0 in the DT since the reg property
          * defines the MPIDR[23:0]
@@ -228,9 +255,13 @@ static void __init dt_smp_init_cpus(void)
         {
             printk("cpu%d init failed (hwid %"PRIregister"): %d\n", i, hwid, rc);
             tmp_map[i] = MPIDR_INVALID;
+            node_map[i] = NUMA_NO_NODE;
         }
         else
+        {
             tmp_map[i] = hwid;
+            node_map[i] = nid;
+        }
     }
 
     if ( !bootcpu_valid )
@@ -246,6 +277,11 @@ static void __init dt_smp_init_cpus(void)
             continue;
         cpumask_set_cpu(i, &cpu_possible_map);
         cpu_logical_map(i) = tmp_map[i];
+
+        nid = node_map[i];
+        if ( nid >= MAX_NUMNODES )
+            nid = 0;
+        numa_set_node(i, nid);
     }
 }
 
Signature · 1 line
2.25.1
author
#21 lore raw

PATCH v2 6/17 xen/arm: Add boot and secondary CPU to NUMA system

In reply to Wei Chen

In this patch, we make NUMA node online and add cpu to its NUMA node. This will make NUMA-aware components have NUMA affinity data to support their work.

To keep the mostly the same behavior of x86, we use numa_detect_cpu_node to online node. The difference is that, we have prepared cpu_to_node in dt_smp_init_cpus, so we don't need to setup cpu_to_node in numa_detect_cpu_node.

---
v2 -> v3:
1. Use unsigned int instead of int for cpu id.
2. Use static inline for stub to do type check. v1 -> v2:
1. Use numa_detect_cpu_node to online node.
2. Use macros instead of static inline functions to stub numa_detect_cpu_node.

---
 xen/arch/arm/include/asm/numa.h |  9 +++++++++
 xen/arch/arm/numa.c             | 10 ++++++++++
 xen/arch/arm/setup.c            |  5 +++++
 3 files changed, 24 insertions(+)
xen/arch/arm/include/asm/numa.h +9 -0
index 3bc28416b4..e0c909cbb7 100644
@@ -41,6 +41,7 @@ enum dt_numa_status {
 
 extern void numa_set_distance(nodeid_t from, nodeid_t to,
                               unsigned int distance);
+extern void numa_detect_cpu_node(unsigned int cpu);
 
 #else
 
@@ -74,6 +75,14 @@ static inline void numa_set_node(unsigned int cpu, nodeid_t node)
 {
 }
 
+static inline void numa_add_cpu(unsigned int cpu)
+{
+}
+
+static inline void numa_detect_cpu_node(unsigned int cpu)
+{
+}
+
 #endif
 
 #define arch_want_default_dmazone() (false)
xen/arch/arm/numa.c +10 -0
index dcfcd85fcf..4dd7cf10ba 100644
@@ -74,6 +74,16 @@ void __init numa_set_distance(nodeid_t from, nodeid_t to,
     node_distance_map[from][to] = distance;
 }
 
+void numa_detect_cpu_node(unsigned int cpu)
+{
+    nodeid_t node = cpu_to_node[cpu];
+
+    if ( node == NUMA_NO_NODE )
+        node = 0;
+
+    node_set_online(node);
+}
+
 unsigned char __node_distance(nodeid_t from, nodeid_t to)
 {
     /* When NUMA is off, any distance will be treated as remote. */
xen/arch/arm/setup.c +5 -0
index 1f26f67b90..8c02cf6cd4 100644
@@ -1115,6 +1115,11 @@ void __init start_xen(unsigned long boot_phys_offset,
 
     for_each_present_cpu ( i )
     {
+        /* Detect and online node based on cpu_to_node[]. */
+        numa_detect_cpu_node(i);
+        /* Set up node_to_cpumask based on cpu_to_node[]. */
+        numa_add_cpu(i);
+
         if ( (num_online_cpus() < nr_cpu_ids) && !cpu_online(i) )
         {
             int ret = cpu_up(i);
Signature · 1 line
2.25.1
author
#22 lore raw

PATCH v2 7/17 xen/arm: introduce a helper to parse device tree processor node

In reply to Wei Chen

Processor NUMA ID information is stored in device tree's processor node as "numa-node-id". We need a new helper to parse this ID from processor node. If we get this ID from processor node, this ID's validity still need to be checked. Once we got a invalid NUMA ID from any processor node, the device tree will be marked as NUMA information invalid.

---
v1 -> v2:
1. Move numa_disabled from fdt_numa_processor_affinity_init to fdt_parse_numa_cpu_node.
2. Move invalid NUMA id check to fdt_parse_numa_cpu_node.
3. Return ENODATA for normal dtb without NUMA info.
4. Use NUMA status helpers instead of SRAT functions.

---
 xen/arch/arm/Makefile           |  1 +
 xen/arch/arm/include/asm/numa.h |  2 ++
 xen/arch/arm/numa.c             |  2 +-
 xen/arch/arm/numa_device_tree.c | 64 +++++++++++++++++++++++++++++++++
 4 files changed, 68 insertions(+), 1 deletion(-)
 create mode 100644 xen/arch/arm/numa_device_tree.c
xen/arch/arm/Makefile +1 -0
index 9073398d6e..bbc68e3735 100644
@@ -39,6 +39,7 @@ obj-y += mem_access.o
 obj-y += mm.o
 obj-y += monitor.o
 obj-$(CONFIG_NUMA) += numa.o
+obj-$(CONFIG_DEVICE_TREE_NUMA) += numa_device_tree.o
 obj-y += p2m.o
 obj-y += percpu.o
 obj-y += platform.o
xen/arch/arm/include/asm/numa.h +2 -0
index e0c909cbb7..923ffbfd42 100644
@@ -28,6 +28,8 @@ enum dt_numa_status {
     DT_NUMA_OFF,
 };
 
+extern enum dt_numa_status device_tree_numa;
+
 /*
  * In ACPI spec, 0-9 are the reserved values for node distance,
  * 10 indicates local node distance, 20 indicates remote node
xen/arch/arm/numa.c +1 -1
index 4dd7cf10ba..3e02cec646 100644
@@ -20,7 +20,7 @@
 #include <xen/init.h>
 #include <xen/numa.h>
 
-static enum dt_numa_status __read_mostly device_tree_numa;
+enum dt_numa_status __read_mostly device_tree_numa;
 
 static unsigned char __read_mostly
 node_distance_map[MAX_NUMNODES][MAX_NUMNODES] = {
xen/arch/arm/numa_device_tree.c new file +64 -0

64 lines not on this page · raw message

Signature · 1 line
2.25.1
author
#23 lore raw

PATCH v2 8/17 xen/arm: introduce a helper to parse device tree memory node

In reply to Wei Chen

Memory blocks' NUMA ID information is stored in device tree's memory nodes as "numa-node-id". We need a new helper to parse and verify this ID from memory nodes.

---
v1->v2:
1. Move numa_disabled check to fdt_parse_numa_memory_node.
2. Use numa_bad to replace bad_srat.
3. Replace tabs by spaces.
4. Align parameters.
5. return ENODATA for a normal dtb without numa info.
6. Un-addressed comment:
"Why not parse numa-node-id and call fdt_numa_memory_affinity_init from xen/arch/arm/bootfdt.c:device_tree_get_meminfo. Is it because device_tree_get_meminfo is called too early?"
I checked the device_tree_get_meminfo code and I think the answer is similar as I reply in RFC. I prefer a unify numa initialization entry. Don't want to make numa parse code in different places.
7. Use node id as dummy PXM for numa_update_node_memblks.

---
 xen/arch/arm/numa_device_tree.c | 89 +++++++++++++++++++++++++++++++++
 1 file changed, 89 insertions(+)
xen/arch/arm/numa_device_tree.c +89 -0
index c031053d24..793a410fd4 100644
@@ -34,6 +34,26 @@ static int __init fdt_numa_processor_affinity_init(nodeid_t node)
     return 0;
 }
 
+/* Callback for parsing of the memory regions affinity */
+static int __init fdt_numa_memory_affinity_init(nodeid_t node,
+                                                paddr_t start, paddr_t size)
+{
+    if ( !numa_memblks_available() )
+    {
+        dprintk(XENLOG_WARNING,
+                "Too many numa entry, try bigger NR_NODE_MEMBLKS\n");
+        return -EINVAL;
+    }
+
+    numa_fw_nid_name = "numa-node-id";
+    if ( !numa_update_node_memblks(node, node, start, size, false) )
+        return -EINVAL;
+
+    device_tree_numa = DT_NUMA_ON;
+
+    return 0;
+}
+
 /* Parse CPU NUMA node info */
 static int __init fdt_parse_numa_cpu_node(const void *fdt, int node)
 {
@@ -62,3 +82,72 @@ static int __init fdt_parse_numa_cpu_node(const void *fdt, int node)
 
     return fdt_numa_processor_affinity_init(nid);
 }
+
+/* Parse memory node NUMA info */
+static int __init fdt_parse_numa_memory_node(const void *fdt, int node,
+                                             const char *name,
+                                             unsigned int addr_cells,
+                                             unsigned int size_cells)
+{
+    unsigned int nid;
+    int ret = 0, len;
+    paddr_t addr, size;
+    const struct fdt_property *prop;
+    unsigned int idx, ranges;
+    const __be32 *addresses;
+
+    if ( numa_disabled() )
+        return -EINVAL;
+
+    /*
+     * device_tree_get_u32 will return NUMA_NO_NODE when this memory
+     * DT node doesn't have numa-node-id. This can help us to
+     * distinguish a bad DTB and a normal DTB without NUMA info.
+     */
+    nid = device_tree_get_u32(fdt, node, "numa-node-id", NUMA_NO_NODE);
+    if ( node == NUMA_NO_NODE )
+    {
+        numa_fw_bad();
+        return -ENODATA;
+    }
+    else if ( nid >= MAX_NUMNODES )
+    {
+        printk(XENLOG_WARNING "Node id %u exceeds maximum value\n", nid);
+        goto invalid_data;
+    }
+
+    prop = fdt_get_property(fdt, node, "reg", &len);
+    if ( !prop )
+    {
+        printk(XENLOG_WARNING
+               "fdt: node `%s': missing `reg' property\n", name);
+        goto invalid_data;
+    }
+
+    addresses = (const __be32 *)prop->data;
+    ranges = len / (sizeof(__be32)* (addr_cells + size_cells));
+    for ( idx = 0; idx < ranges; idx++ )
+    {
+        device_tree_get_reg(&addresses, addr_cells, size_cells, &addr, &size);
+        /* Skip zero size ranges */
+        if ( !size )
+            continue;
+
+        ret = fdt_numa_memory_affinity_init(nid, addr, size);
+        if ( ret )
+            goto invalid_data;
+    }
+
+    if ( idx == 0 )
+    {
+        printk(XENLOG_ERR
+               "bad property in memory node, idx=%d ret=%d\n", idx, ret);
+        goto invalid_data;
+    }
+
+    return 0;
+
+invalid_data:
+    numa_fw_bad();
+    return -EINVAL;
+}
Signature · 1 line
2.25.1
author
#24 lore raw

PATCH v2 9/17 xen/arm: introduce a helper to parse device tree NUMA distance map

In reply to Wei Chen

A NUMA aware device tree will provide a "distance-map" node to describe distance between any two nodes. This patch introduce a new helper to parse this distance map.

---
v1 -> v2:
1. Get rid of useless braces.
2. Use new NUMA status helper.
3. Use PRIu32 to replace u in print messages.
4. Fix opposite = __node_distance(to, from).
5. disable dtb numa info table when we find an invalid data in dtb.

---
 xen/arch/arm/numa_device_tree.c | 107 ++++++++++++++++++++++++++++++++
 1 file changed, 107 insertions(+)
xen/arch/arm/numa_device_tree.c +107 -0
index 793a410fd4..88056e7ef8 100644
@@ -151,3 +151,110 @@ invalid_data:
     numa_fw_bad();
     return -EINVAL;
 }
+
+/* Parse NUMA distance map v1 */
+static int __init fdt_parse_numa_distance_map_v1(const void *fdt, int node)
+{
+    const struct fdt_property *prop;
+    const __be32 *matrix;
+    unsigned int i, entry_count;
+    int len;
+
+    printk(XENLOG_INFO "NUMA: parsing numa-distance-map\n");
+
+    prop = fdt_get_property(fdt, node, "distance-matrix", &len);
+    if ( !prop )
+    {
+        printk(XENLOG_WARNING
+               "NUMA: No distance-matrix property in distance-map\n");
+        goto invalid_data;
+    }
+
+    if ( len % sizeof(__be32) != 0 )
+    {
+        printk(XENLOG_WARNING
+               "distance-matrix in node is not a multiple of u32\n");
+        goto invalid_data;
+    }
+
+    entry_count = len / sizeof(__be32);
+    if ( entry_count == 0 )
+    {
+        printk(XENLOG_WARNING "NUMA: Invalid distance-matrix\n");
+        goto invalid_data;
+    }
+
+    matrix = (const __be32 *)prop->data;
+    for ( i = 0; i + 2 < entry_count; i += 3 )
+    {
+        unsigned int from, to, distance, opposite;
+
+        from = dt_next_cell(1, &matrix);
+        to = dt_next_cell(1, &matrix);
+        distance = dt_next_cell(1, &matrix);
+        if ( (from == to && distance != NUMA_LOCAL_DISTANCE) ||
+            (from != to && distance <= NUMA_LOCAL_DISTANCE) )
+        {
+            printk(XENLOG_WARNING
+                   "NUMA: Invalid distance: NODE#%"PRIu32"->NODE#%"PRIu32":%"PRIu32"\n",
+                   from, to, distance);
+            goto invalid_data;
+        }
+
+        printk(XENLOG_INFO "NUMA: distance: NODE#%"PRIu32"->NODE#%"PRIu32":%"PRIu32"\n",
+               from, to, distance);
+
+        /* Get opposite way distance */
+        opposite = __node_distance(to, from);
+        if ( opposite == 0 )
+        {
+            /* Bi-directions are not set, set both */
+            numa_set_distance(from, to, distance);
+            numa_set_distance(to, from, distance);
+        }
+        else
+        {
+            /*
+             * Opposite way distance has been set to a different value.
+             * It may be a firmware device tree bug?
+             */
+            if ( opposite != distance )
+            {
+                /*
+                 * In device tree NUMA distance-matrix binding:
+                 * https://www.kernel.org/doc/Documentation/devicetree/bindings/numa.txt
+                 * There is a notes mentions:
+                 * "Each entry represents distance from first node to
+                 *  second node. The distances are equal in either
+                 *  direction."
+                 *
+                 * That means device tree doesn't permit this case.
+                 * But in ACPI spec, it cares to specifically permit this
+                 * case:
+                 * "Except for the relative distance from a System Locality
+                 *  to itself, each relative distance is stored twice in the
+                 *  matrix. This provides the capability to describe the
+                 *  scenario where the relative distances for the two
+                 *  directions between System Localities is different."
+                 *
+                 * That means a real machine allows such NUMA configuration.
+                 * So, place a WARNING here to notice system administrators,
+                 * is it the specail case that they hijack the device tree
+                 * to support their rare machines?
+                 */
+                printk(XENLOG_WARNING
+                       "Un-matched bi-direction! NODE#%"PRIu32"->NODE#%"PRIu32":%"PRIu32", NODE#%"PRIu32"->NODE#%"PRIu32":%"PRIu32"\n",
+                       from, to, distance, to, from, opposite);
+            }
+
+            /* Opposite way distance has been set, just set this way */
+            numa_set_distance(from, to, distance);
+        }
+    }
+
+    return 0;
+
+invalid_data:
+    numa_fw_bad();
+    return -EINVAL;
+}
Signature · 1 line
2.25.1
author
#25 lore raw

PATCH v2 10/17 xen/arm: unified entry to parse all NUMA data from device tree

In reply to Wei Chen

In this function, we scan the whole device tree to parse CPU node id, memory node id and distance-map. Though early_scan_node will invoke a handler to process memory nodes. If we want to parse memory node id in that handler, we have to embed NUMA parse code in that handler. But we still need to scan whole device tree to find CPU NUMA id and distance-map. In this case, we include memory NUMA id parse in this function too. Another benefit is that we have a unique entry for device tree NUMA data parse.

---
v1->v2:
1. Fix typos in commit message.
2. Fix code style and align parameters.
3. Use strncmp to replace memcmp.

---
 xen/arch/arm/include/asm/numa.h |  1 +
 xen/arch/arm/numa_device_tree.c | 30 ++++++++++++++++++++++++++++++
 2 files changed, 31 insertions(+)
xen/arch/arm/include/asm/numa.h +1 -0
index 923ffbfd42..1213d30ce0 100644
@@ -44,6 +44,7 @@ extern enum dt_numa_status device_tree_numa;
 extern void numa_set_distance(nodeid_t from, nodeid_t to,
                               unsigned int distance);
 extern void numa_detect_cpu_node(unsigned int cpu);
+extern int numa_device_tree_init(const void *fdt);
 
 #else
 
xen/arch/arm/numa_device_tree.c +30 -0
index 88056e7ef8..4009b9b6de 100644
@@ -258,3 +258,33 @@ invalid_data:
     numa_fw_bad();
     return -EINVAL;
 }
+
+static int __init fdt_scan_numa_nodes(const void *fdt, int node,
+                                      const char *uname, int depth,
+                                      unsigned int address_cells,
+                                      unsigned int size_cells, void *data)
+{
+    int len, ret = 0;
+    const void *prop;
+
+    prop = fdt_getprop(fdt, node, "device_type", &len);
+    if ( prop )
+    {
+        if ( strncmp(prop, "cpu", len) == 0 )
+            ret = fdt_parse_numa_cpu_node(fdt, node);
+        else if ( strncmp(prop, "memory", len) == 0 )
+            ret = fdt_parse_numa_memory_node(fdt, node, uname,
+                                address_cells, size_cells);
+    }
+    else if ( fdt_node_check_compatible(fdt, node,
+                                        "numa-distance-map-v1") == 0 )
+        ret = fdt_parse_numa_distance_map_v1(fdt, node);
+
+    return ret;
+}
+
+/* Initialize NUMA from device tree */
+int __init numa_device_tree_init(const void *fdt)
+{
+    return device_tree_for_each_node(fdt, 0, fdt_scan_numa_nodes, NULL);
+}
Signature · 1 line
2.25.1
author
#26 lore raw

PATCH v2 11/17 xen/arm: keep guest still be NUMA unware

In reply to Wei Chen

The NUMA information provided in the host Device-Tree are only for Xen. For dom0, we want to hide them as they may be different (for now, dom0 is still not aware of NUMA) The CPU and memory nodes are recreated from scratch for the domain. So we already skip the "numa-node-id" property for these two types of nodes.

However, some devices like PCIe may have "numa-node-id" property too. We have to skip them as well.

---
v1 -> v2:
1. Add Rb

---
 xen/arch/arm/domain_build.c | 6 ++++++
 1 file changed, 6 insertions(+)
xen/arch/arm/domain_build.c +6 -0
index 829cea8de8..8258048d0e 100644
@@ -1185,6 +1185,10 @@ static int __init write_properties(struct domain *d, struct kernel_info *kinfo,
                 continue;
         }
 
+        /* Dom0 is currently NUMA unaware */
+        if ( dt_property_name_is_equal(prop, "numa-node-id") )
+            continue;
+
         res = fdt_property(kinfo->fdt, prop->name, prop_data, prop_len);
 
         if ( res )
@@ -2540,6 +2544,8 @@ static int __init handle_node(struct domain *d, struct kernel_info *kinfo,
         DT_MATCH_TYPE("memory"),
         /* The memory mapped timer is not supported by Xen. */
         DT_MATCH_COMPATIBLE("arm,armv7-timer-mem"),
+        /* Numa info doesn't need to be exposed to Domain-0 */
+        DT_MATCH_COMPATIBLE("numa-distance-map-v1"),
         { /* sentinel */ },
     };
     static const struct dt_device_match timer_matches[] __initconst =
Signature · 1 line
2.25.1
author
#27 lore raw

PATCH v2 12/17 xen/arm: enable device tree based NUMA in system init

In reply to Wei Chen

In this patch, we can start to create NUMA system that is based on device tree.

---
v1->v2:
1. replace ~0 by INVALID_PADDR.
2. only print error messages for invalid dtb data.
3. remove unnecessary return.
4. remove the parameter of numa_init.

---
 xen/arch/arm/include/asm/numa.h |  5 +++
 xen/arch/arm/numa.c             | 57 +++++++++++++++++++++++++++++++++
 xen/arch/arm/setup.c            |  7 ++++
 3 files changed, 69 insertions(+)
xen/arch/arm/include/asm/numa.h +5 -0
index 1213d30ce0..34eec9378c 100644
@@ -45,6 +45,7 @@ extern void numa_set_distance(nodeid_t from, nodeid_t to,
                               unsigned int distance);
 extern void numa_detect_cpu_node(unsigned int cpu);
 extern int numa_device_tree_init(const void *fdt);
+extern void numa_init(void);
 
 #else
 
@@ -86,6 +87,10 @@ static inline void numa_detect_cpu_node(unsigned int cpu)
 {
 }
 
+static inline void numa_init(void)
+{
+}
+
 #endif
 
 #define arch_want_default_dmazone() (false)
xen/arch/arm/numa.c +57 -0

70 lines not on this page · raw message

xen/arch/arm/setup.c +7 -0
index 8c02cf6cd4..4cdc7e2edb 100644
@@ -1031,6 +1031,13 @@ void __init start_xen(unsigned long boot_phys_offset,
     /* Parse the ACPI tables for possible boot-time configuration */
     acpi_boot_table_init();
 
+    /*
+     * Try to initialize NUMA system, if failed, the system will
+     * fallback to uniform system which means system has only 1
+     * NUMA node.
+     */
+    numa_init();
+
     end_boot_allocator();
 
     /*
Signature · 1 line
2.25.1
author
#28 lore raw

PATCH v2 13/17 xen/arm: implement numa_node_to_arch_nid for device tree NUMA

In reply to Wei Chen

Device tree based NUMA doesn't have the proximity domain like ACPI. So we can return node id directly as arch nid.

---
v1 -> v2:
1. Use numa_node_to_arch_nid instead of dummy node_to_pxm.

---
 xen/arch/arm/include/asm/numa.h | 9 +++++++++
 1 file changed, 9 insertions(+)
xen/arch/arm/include/asm/numa.h +9 -0
index 34eec9378c..a0b8d7a11c 100644
@@ -47,6 +47,15 @@ extern void numa_detect_cpu_node(unsigned int cpu);
 extern int numa_device_tree_init(const void *fdt);
 extern void numa_init(void);
 
+/*
+ * Device tree NUMA doesn't have architecural node id.
+ * So we can just return node id as arch nid.
+ */
+static inline unsigned int numa_node_to_arch_nid(nodeid_t n)
+{
+    return n;
+}
+
 #else
 
 /* Fake one node for now. See also node_online_map. */
Signature · 1 line
2.25.1
author
#29 lore raw

PATCH v2 14/17 xen/arm: use CONFIG_NUMA to gate node_online_map in smpboot

In reply to Wei Chen

node_online_map in smpboot still need for Arm when NUMA is turn off by Kconfig.

---
v1 -> v2:
1. No change.

---
 xen/arch/arm/smpboot.c | 2 ++
 1 file changed, 2 insertions(+)
xen/arch/arm/smpboot.c +2 -0
index 5ee6ab11e9..3ae359bbeb 100644
@@ -41,8 +41,10 @@ integer_param("maxcpus", max_cpus);
 /* CPU logical map: map xen cpuid to an MPIDR */
 register_t __cpu_logical_map[NR_CPUS] = { [0 ... NR_CPUS-1] = MPIDR_INVALID };
 
+#ifndef CONFIG_NUMA
 /* Fake one node for now. See also asm/numa.h */
 nodemask_t __read_mostly node_online_map = { { [0] = 1UL } };
+#endif
 
 /* Xen stack for bringing up the first CPU. */
 static unsigned char __initdata cpu0_boot_stack[STACK_SIZE]
Signature · 1 line
2.25.1
author
#30 lore raw

PATCH v2 15/17 xen/arm: Set correct per-cpu cpu_core_mask

In reply to Wei Chen

From: Henry Wang <[email protected]>

In the common sysctl command XEN_SYSCTL_physinfo, the cores_per_socket is calculated based on the cpu_core_mask of CPU0. Currently on Arm this is a fixed value 1 (can be checked via xl info), which is not correct. This is because during the Arm cpu online process, set_cpu_sibling_map() only sets the per-cpu cpu_core_mask for itself.

cores_per_socket refers to the number of cores that belong to the same socket (NUMA node). Therefore, this commit introduces a helper function numa_set_cpu_core_mask(cpu), which sets the per-cpu cpu_core_mask to the cpus in the same NUMA node as cpu. Calling this function at the boot time can ensure the correct cpu_core_mask, leading to the correct cores_per_socket to be returned by XEN_SYSCTL_physinfo.

---
v1 -> v2:
1. New patch

---
 xen/arch/arm/include/asm/numa.h |  7 +++++++
 xen/arch/arm/numa.c             | 11 +++++++++++
 xen/arch/arm/setup.c            |  5 +++++
 3 files changed, 23 insertions(+)
xen/arch/arm/include/asm/numa.h +7 -0
index a0b8d7a11c..e66fb0a11f 100644
@@ -46,6 +46,7 @@ extern void numa_set_distance(nodeid_t from, nodeid_t to,
 extern void numa_detect_cpu_node(unsigned int cpu);
 extern int numa_device_tree_init(const void *fdt);
 extern void numa_init(void);
+extern void numa_set_cpu_core_mask(int cpu);
 
 /*
  * Device tree NUMA doesn't have architecural node id.
@@ -62,6 +63,12 @@ static inline unsigned int numa_node_to_arch_nid(nodeid_t n)
 #define cpu_to_node(cpu) 0
 #define node_to_cpumask(node)   (cpu_online_map)
 
+static inline void numa_set_cpu_core_mask(int cpu)
+{
+    cpumask_or(per_cpu(cpu_core_mask, cpu),
+               per_cpu(cpu_core_mask, cpu), &cpu_possible_map);
+}
+
 /*
  * TODO: make first_valid_mfn static when NUMA is supported on Arm, this
  * is required because the dummy helpers are using it.
xen/arch/arm/numa.c +11 -0
index e9081d45ce..ef245e39a8 100644
@@ -52,6 +52,17 @@ int __init arch_numa_setup(const char *opt)
     return -EINVAL;
 }
 
+void numa_set_cpu_core_mask(int cpu)
+{
+    nodeid_t node = cpu_to_node[cpu];
+
+    if ( node == NUMA_NO_NODE )
+        node = 0;
+
+    cpumask_or(per_cpu(cpu_core_mask, cpu),
+               per_cpu(cpu_core_mask, cpu), &node_to_cpumask(node));
+}
+
 void __init numa_set_distance(nodeid_t from, nodeid_t to,
                               unsigned int distance)
 {
xen/arch/arm/setup.c +5 -0
index 4cdc7e2edb..d45becedee 100644
@@ -1136,6 +1136,11 @@ void __init start_xen(unsigned long boot_phys_offset,
     }
 
     printk("Brought up %ld CPUs\n", (long)num_online_cpus());
+
+    /* Set per-cpu cpu_core_mask to cpus that belongs to the same NUMA node. */
+    for_each_online_cpu ( i )
+        numa_set_cpu_core_mask(i);
+
     /* TODO: smp_cpus_done(); */
 
     /* This should be done in a vpmu driver but we do not have one yet. */
Signature · 1 line
2.25.1
author
#31 lore raw

PATCH v2 16/17 xen/arm: Provide Kconfig options for Arm to enable NUMA

In reply to Wei Chen

Arm platforms support both ACPI and device tree. We don't want users to select device tree NUMA or ACPI NUMA manually. We hope users can just enable NUMA for Arm, and device tree NUMA and ACPI NUMA can be selected depends on device tree feature and ACPI feature status automatically. In this case, these two kinds of NUMA support code can be co-exist in one Xen binary. Xen can check feature flags to decide using device tree or ACPI as NUMA based firmware.

So in this patch, we introduce a generic option:
CONFIG_ARM_NUMA for users to enable NUMA for Arm. And one CONFIG_DEVICE_TREE_NUMA option for ARM_NUMA to select when HAS_DEVICE_TREE option is enabled. Once when ACPI NUMA for Arm is supported, ACPI_NUMA can be selected here too.

---
v1 -> v2:
1. Remove the condition of selecting DEVICE_TREE_NUMA.

---
 xen/arch/arm/Kconfig | 11 +++++++++++
 1 file changed, 11 insertions(+)
xen/arch/arm/Kconfig +11 -0
index 239d3aed3c..e751ad50d1 100644
@@ -39,6 +39,17 @@ config ACPI
 config ARM_EFI
 	bool
 
+config ARM_NUMA
+	bool "Arm NUMA (Non-Uniform Memory Access) Support (UNSUPPORTED)" if UNSUPPORTED
+	depends on HAS_DEVICE_TREE
+	select DEVICE_TREE_NUMA
+	help
+	  Enable Non-Uniform Memory Access (NUMA) for Arm architecutres
+
+config DEVICE_TREE_NUMA
+	bool
+	select NUMA
+
 config GICV3
 	bool "GICv3 driver"
 	depends on !NEW_VGIC
Signature · 1 line
2.25.1
author
#32 lore raw

PATCH v2 17/17 docs: update numa command line to support Arm

In reply to Wei Chen

Current numa command in documentation is x86 only. Remove x86 from numa command's arch limitation in this patch.

---
v1 -> v2:
1. Update Arm NUMA status in SUPPORT.md to "Tech Preview".

---
 SUPPORT.md                        | 1 +
 docs/misc/xen-command-line.pandoc | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)
SUPPORT.md +1 -0
index 295369998e..b73d04a028 100644
@@ -401,6 +401,7 @@ on embedded platforms and the x86 PV shim.
 Enables NUMA aware scheduling in Xen
 
     Status, x86: Supported
+    Status, Arm: Tech Preview
 
 ## Scalability
 
docs/misc/xen-command-line.pandoc +1 -1
index 923910f553..40da980b67 100644
@@ -1878,7 +1878,7 @@ i.e. a limit on the number of guests it is possible to start each having
 assigned a device sharing a common interrupt line.  Accepts values between
 1 and 255.
 
-### numa (x86)
+### numa
 > `= on | off | fake=<integer> | noacpi`
 
 > Default: `on`
Signature · 1 line
2.25.1
Acked-by on v2 17/17
#33 lore raw

Re: [PATCH v2 17/17] docs: update numa command line to support Arm

Quoted 4 lines · Wei Chen

Current numa command in documentation is x86 only. Remove x86 from numa command's arch limitation in this patch.

Of course not on its own, only on top of all the earlier patches.

Jan