[Qemu-ppc] [PATCH v3 1/1] numa: equally distribute memory on nodes

[Laurent Vivier]

When there are more nodes than available memory to put the minimum
allowed memory by node, all the memory is put on the last node.

This is because we put (ram_size / nb_numa_nodes) &
~((1 << mc->numa_mem_align_shift) - 1); on each node, and in this
case the value is 0. This is particularly true with pseries,
as the memory must be aligned to 256MB.

To avoid this problem, this patch uses an error diffusion algorithm [1]
to distribute equally the memory on nodes.

We introduce numa_auto_assign_ram() function in MachineClass
to keep compatibility between machine type versions.
The legacy function is used with pseries-2.9, pc-q35-2.9 and
pc-i440fx-2.9 (and previous), the new one with all others.

Example:

qemu-system-ppc64 -S -nographic  -nodefaults -monitor stdio -m 1G -smp 8 \
                  -numa node -numa node -numa node \
                  -numa node -numa node -numa node

Before:

(qemu) info numa
6 nodes
node 0 cpus: 0 6
node 0 size: 0 MB
node 1 cpus: 1 7
node 1 size: 0 MB
node 2 cpus: 2
node 2 size: 0 MB
node 3 cpus: 3
node 3 size: 0 MB
node 4 cpus: 4
node 4 size: 0 MB
node 5 cpus: 5
node 5 size: 1024 MB

After:
(qemu) info numa
6 nodes
node 0 cpus: 0 6
node 0 size: 0 MB
node 1 cpus: 1 7
node 1 size: 256 MB
node 2 cpus: 2
node 2 size: 0 MB
node 3 cpus: 3
node 3 size: 256 MB
node 4 cpus: 4
node 4 size: 256 MB
node 5 cpus: 5
node 5 size: 256 MB

[1] https://en.wikipedia.org/wiki/Error_diffusion

Signed-off-by: Laurent Vivier <address@hidden>
---
 hw/core/machine.c       |  2 ++
 hw/i386/pc_piix.c       |  1 +
 hw/i386/pc_q35.c        |  1 +
 hw/ppc/spapr.c          |  1 +
 include/hw/boards.h     |  2 ++
 include/qemu/typedefs.h |  1 +
 include/sysemu/numa.h   |  9 +++++++--
 numa.c                  | 49 ++++++++++++++++++++++++++++++++++++++-----------
 8 files changed, 53 insertions(+), 13 deletions(-)

diff --git a/hw/core/machine.c b/hw/core/machine.c
index ada9eea..2482c63 100644
--- a/hw/core/machine.c
+++ b/hw/core/machine.c
@@ -17,6 +17,7 @@
 #include "qapi/visitor.h"
 #include "hw/sysbus.h"
 #include "sysemu/sysemu.h"
+#include "sysemu/numa.h"
 #include "qemu/error-report.h"
 #include "qemu/cutils.h"
 
@@ -400,6 +401,7 @@ static void machine_class_init(ObjectClass *oc, void *data)
      * On Linux, each node's border has to be 8MB aligned
      */
     mc->numa_mem_align_shift = 23;
+    mc->numa_auto_assign_ram = numa_default_auto_assign_ram;
 
     object_class_property_add_str(oc, "accel",
         machine_get_accel, machine_set_accel, &error_abort);
diff --git a/hw/i386/pc_piix.c b/hw/i386/pc_piix.c
index 9f102aa..e1dbded 100644
--- a/hw/i386/pc_piix.c
+++ b/hw/i386/pc_piix.c
@@ -442,6 +442,7 @@ static void pc_i440fx_2_9_machine_options(MachineClass *m)
     pc_i440fx_machine_options(m);
     m->alias = "pc";
     m->is_default = 1;
+    m->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
 }
 
 DEFINE_I440FX_MACHINE(v2_9, "pc-i440fx-2.9", NULL,
diff --git a/hw/i386/pc_q35.c b/hw/i386/pc_q35.c
index dd792a8..b8a6fd7 100644
--- a/hw/i386/pc_q35.c
+++ b/hw/i386/pc_q35.c
@@ -305,6 +305,7 @@ static void pc_q35_2_9_machine_options(MachineClass *m)
 {
     pc_q35_machine_options(m);
     m->alias = "q35";
+    m->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
 }
 
 DEFINE_Q35_MACHINE(v2_9, "pc-q35-2.9", NULL,
diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index 80d12d0..bdc31ce 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -3242,6 +3242,7 @@ static void spapr_machine_2_9_class_options(MachineClass 
*mc)
 {
     spapr_machine_2_10_class_options(mc);
     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
+    mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
 }
 
 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
diff --git a/include/hw/boards.h b/include/hw/boards.h
index 31d9c72..99458eb 100644
--- a/include/hw/boards.h
+++ b/include/hw/boards.h
@@ -136,6 +136,8 @@ struct MachineClass {
     int minimum_page_bits;
     bool has_hotpluggable_cpus;
     int numa_mem_align_shift;
+    void (*numa_auto_assign_ram)(MachineClass *mc, NodeInfo *nodes,
+                                 int nb_nodes, ram_addr_t size);
 
     HotplugHandler *(*get_hotplug_handler)(MachineState *machine,
                                            DeviceState *dev);
diff --git a/include/qemu/typedefs.h b/include/qemu/typedefs.h
index f08d327..7d85057 100644
--- a/include/qemu/typedefs.h
+++ b/include/qemu/typedefs.h
@@ -97,5 +97,6 @@ typedef struct SSIBus SSIBus;
 typedef struct uWireSlave uWireSlave;
 typedef struct VirtIODevice VirtIODevice;
 typedef struct Visitor Visitor;
+typedef struct node_info NodeInfo;
 
 #endif /* QEMU_TYPEDEFS_H */
diff --git a/include/sysemu/numa.h b/include/sysemu/numa.h
index 8f09dcf..6270384 100644
--- a/include/sysemu/numa.h
+++ b/include/sysemu/numa.h
@@ -15,13 +15,13 @@ struct numa_addr_range {
     QLIST_ENTRY(numa_addr_range) entry;
 };
 
-typedef struct node_info {
+struct node_info {
     uint64_t node_mem;
     unsigned long *node_cpu;
     struct HostMemoryBackend *node_memdev;
     bool present;
     QLIST_HEAD(, numa_addr_range) addr; /* List to store address ranges */
-} NodeInfo;
+};
 
 extern NodeInfo numa_info[MAX_NODES];
 void parse_numa_opts(MachineClass *mc);
@@ -31,6 +31,11 @@ extern QemuOptsList qemu_numa_opts;
 void numa_set_mem_node_id(ram_addr_t addr, uint64_t size, uint32_t node);
 void numa_unset_mem_node_id(ram_addr_t addr, uint64_t size, uint32_t node);
 uint32_t numa_get_node(ram_addr_t addr, Error **errp);
+void numa_legacy_auto_assign_ram(MachineClass *mc, NodeInfo *nodes,
+                                 int nb_nodes, ram_addr_t size);
+void numa_default_auto_assign_ram(MachineClass *mc, NodeInfo *nodes,
+                                  int nb_nodes, ram_addr_t size);
+
 
 /* on success returns node index in numa_info,
  * on failure returns nb_numa_nodes */
diff --git a/numa.c b/numa.c
index 6fc2393..750fd95 100644
--- a/numa.c
+++ b/numa.c
@@ -294,6 +294,42 @@ static void validate_numa_cpus(void)
     g_free(seen_cpus);
 }
 
+void numa_legacy_auto_assign_ram(MachineClass *mc, NodeInfo *nodes,
+                                 int nb_nodes, ram_addr_t size)
+{
+    int i;
+    uint64_t usedmem = 0;
+
+    /* Align each node according to the alignment
+     * requirements of the machine class
+     */
+
+    for (i = 0; i < nb_nodes - 1; i++) {
+        nodes[i].node_mem = (size / nb_nodes) &
+                            ~((1 << mc->numa_mem_align_shift) - 1);
+        usedmem += nodes[i].node_mem;
+    }
+    nodes[i].node_mem = size - usedmem;
+}
+
+void numa_default_auto_assign_ram(MachineClass *mc, NodeInfo *nodes,
+                                  int nb_nodes, ram_addr_t size)
+{
+    int i;
+    uint64_t usedmem = 0, node_mem;
+    uint64_t granularity = size / nb_nodes;
+    uint64_t propagate = 0;
+
+    for (i = 0; i < nb_nodes - 1; i++) {
+        node_mem = (granularity + propagate) &
+                   ~((1 << mc->numa_mem_align_shift) - 1);
+        propagate = granularity + propagate - node_mem;
+        nodes[i].node_mem = node_mem;
+        usedmem += node_mem;
+    }
+    nodes[i].node_mem = ram_size - usedmem;
+}
+
 void parse_numa_opts(MachineClass *mc)
 {
     int i;
@@ -336,17 +372,8 @@ void parse_numa_opts(MachineClass *mc)
             }
         }
         if (i == nb_numa_nodes) {
-            uint64_t usedmem = 0;
-
-            /* Align each node according to the alignment
-             * requirements of the machine class
-             */
-            for (i = 0; i < nb_numa_nodes - 1; i++) {
-                numa_info[i].node_mem = (ram_size / nb_numa_nodes) &
-                                        ~((1 << mc->numa_mem_align_shift) - 1);
-                usedmem += numa_info[i].node_mem;
-            }
-            numa_info[i].node_mem = ram_size - usedmem;
+            assert(mc->numa_auto_assign_ram);
+            mc->numa_auto_assign_ram(mc, numa_info, nb_numa_nodes, ram_size);
         }
 
         numa_total = 0;
-- 
2.9.3