[Qemu-devel] [PATCH 1/4] added -numa cmdline parameter parser

[Andre Przywara]

adds a -numa command line parameter and sets a QEMU global array with
the memory sizes. The CPU-to-node assignemnt is written into the
CPUState. If no specific values for memory and CPUs are given,
all resources will be split equally across all nodes.
This code currently support only up to 64 virtual CPUs.

Signed-off-by: Andre Przywara <address@hidden>
---
 cpu-defs.h      |    1 +
 exec.c          |    1 +
 qemu-options.hx |    8 +++
 sysemu.h        |    6 ++-
 vl.c            |  133 +++++++++++++++++++++++++++++++++++++++++++++++++++++--
 5 files changed, 144 insertions(+), 5 deletions(-)

diff --git a/cpu-defs.h b/cpu-defs.h
index b462a9f..7cbf85d 100644
--- a/cpu-defs.h
+++ b/cpu-defs.h
@@ -205,6 +205,7 @@ typedef struct CPUWatchpoint {
                                                                         \
     CPUState *next_cpu; /* next CPU sharing TB cache */                 \
     int cpu_index; /* CPU index (informative) */                        \
+    int numa_node; /* NUMA node this cpu is belonging to  */            \
     int running; /* Nonzero if cpu is currently running(usermode).  */  \
     /* user data */                                                     \
     void *opaque;                                                       \
diff --git a/exec.c b/exec.c
index fc7e08c..8245ac0 100644
--- a/exec.c
+++ b/exec.c
@@ -554,6 +554,7 @@ void cpu_exec_init(CPUState *env)
         cpu_index++;
     }
     env->cpu_index = cpu_index;
+    env->numa_node = 0;
     TAILQ_INIT(&env->breakpoints);
     TAILQ_INIT(&env->watchpoints);
     *penv = env;
diff --git a/qemu-options.hx b/qemu-options.hx
index 1d783e5..1e39f25 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -47,6 +47,14 @@ CPUs are supported. On Sparc32 target, Linux limits the 
number of usable CPUs
 to 4.
 ETEXI
 
+DEF("numa", HAS_ARG, QEMU_OPTION_numa,
+    "-numa node[,mem=size][,cpus=cpu[-cpu]][,nodeid=node]\n")
+STEXI
address@hidden -numa @var{opts}
+Simulate a multi node NUMA system. If mem and cpus are omitted, resources
+are split equally.
+ETEXI
+
 DEF("fda", HAS_ARG, QEMU_OPTION_fda,
     "-fda/-fdb file  use 'file' as floppy disk 0/1 image\n")
 DEF("fdb", HAS_ARG, QEMU_OPTION_fdb, "")
diff --git a/sysemu.h b/sysemu.h
index 24b4bd1..cbfbb8e 100644
--- a/sysemu.h
+++ b/sysemu.h
@@ -108,6 +108,10 @@ extern int old_param;
 extern int kqemu_allowed;
 #endif
 
+#define MAX_NODES 64
+extern int nb_numa_nodes;
+extern uint64_t node_mem[MAX_NODES];
+
 #define MAX_OPTION_ROMS 16
 extern const char *option_rom[MAX_OPTION_ROMS];
 extern int nb_option_roms;
@@ -248,7 +252,7 @@ void do_usb_add(Monitor *mon, const char *devname);
 void do_usb_del(Monitor *mon, const char *devname);
 void usb_info(Monitor *mon);
 
-const char *get_opt_name(char *buf, int buf_size, const char *p);
+const char *get_opt_name(char *buf, int buf_size, const char *p, char delim);
 const char *get_opt_value(char *buf, int buf_size, const char *p);
 int get_param_value(char *buf, int buf_size,
                     const char *tag, const char *str);
diff --git a/vl.c b/vl.c
index 55a9bc5..a6d6801 100644
--- a/vl.c
+++ b/vl.c
@@ -265,6 +265,10 @@ const char *prom_envs[MAX_PROM_ENVS];
 int nb_drives_opt;
 struct drive_opt drives_opt[MAX_DRIVES];
 
+int nb_numa_nodes;
+uint64_t node_mem[MAX_NODES];
+uint64_t node_cpumask[MAX_NODES];
+
 static CPUState *cur_cpu;
 static CPUState *next_cpu;
 static int event_pending = 1;
@@ -1865,12 +1869,12 @@ static int socket_init(void)
 }
 #endif
 
-const char *get_opt_name(char *buf, int buf_size, const char *p)
+const char *get_opt_name(char *buf, int buf_size, const char *p, char delim)
 {
     char *q;
 
     q = buf;
-    while (*p != '\0' && *p != '=') {
+    while (*p != '\0' && *p != delim) {
         if (q && (q - buf) < buf_size - 1)
             *q++ = *p;
         p++;
@@ -1910,7 +1914,7 @@ int get_param_value(char *buf, int buf_size,
 
     p = str;
     for(;;) {
-        p = get_opt_name(option, sizeof(option), p);
+        p = get_opt_name(option, sizeof(option), p, '=');
         if (*p != '=')
             break;
         p++;
@@ -1935,7 +1939,7 @@ int check_params(char *buf, int buf_size,
 
     p = str;
     for(;;) {
-        p = get_opt_name(buf, buf_size, p);
+        p = get_opt_name(buf, buf_size, p, '=');
         if (*p != '=')
             return -1;
         p++;
@@ -2628,6 +2632,62 @@ int drive_init(struct drive_opt *arg, int snapshot, void 
*opaque)
     return drives_table_idx;
 }
 
+static void numa_add(const char *optarg)
+{
+    char option[128];
+    char *endptr;
+    unsigned long long value, endvalue;
+    int nodenr;
+
+    optarg = get_opt_name(option, 128, optarg, ',') + 1;
+    if (!strcmp(option, "node")) {
+        if (get_param_value(option, 128, "nodeid", optarg) == 0) {
+            nodenr = nb_numa_nodes;
+        } else {
+            nodenr = strtoull(option, NULL, 10);
+        }
+
+        if (get_param_value(option, 128, "mem", optarg) == 0) {
+            node_mem[nodenr] = 0;
+        } else {
+            value = strtoull(option, &endptr, 0);
+            switch (*endptr) {
+            case 0: case 'M': case 'm':
+                value <<= 20;
+                break;
+            case 'G': case 'g':
+                value <<= 30;
+                break;
+            }
+            node_mem[nodenr] = value;
+        }
+        if (get_param_value(option, 128, "cpus", optarg) == 0) {
+            node_cpumask[nodenr] = 0;
+        } else {
+            value = strtoull(option, &endptr, 10);
+            if (value >= 64) {
+                value = 63;
+                fprintf(stderr, "only 64 CPUs in NUMA mode supported.\n");
+            } else {
+                if (*endptr == '-') {
+                    endvalue = strtoull(endptr+1, &endptr, 10);
+                    if (endvalue >= 63) {
+                        endvalue = 62;
+                        fprintf(stderr,
+                            "only 63 CPUs in NUMA mode supported.\n");
+                    }
+                    value = (1 << (endvalue + 1)) - (1 << value);
+                } else {
+                    value = 1 << value;
+                }
+            }
+            node_cpumask[nodenr] = value;
+        }
+        nb_numa_nodes++;
+    }
+    return;
+}
+
 /***********************************************************/
 /* USB devices */
 
@@ -4290,6 +4350,7 @@ int main(int argc, char **argv, char **envp)
     const char *chroot_dir = NULL;
     const char *run_as = NULL;
 #endif
+    CPUState *env;
 
     qemu_cache_utils_init(envp);
 
@@ -4353,12 +4414,18 @@ int main(int argc, char **argv, char **envp)
         virtio_consoles[i] = NULL;
     virtio_console_index = 0;
 
+    for (i = 0; i < MAX_NODES; i++) {
+        node_mem[i] = 0;
+        node_cpumask[i] = 0;
+    }
+
     usb_devices_index = 0;
 
     nb_net_clients = 0;
     nb_bt_opts = 0;
     nb_drives = 0;
     nb_drives_opt = 0;
+    nb_numa_nodes = 0;
     hda_index = -1;
 
     nb_nics = 0;
@@ -4508,6 +4575,13 @@ int main(int argc, char **argv, char **envp)
                                     ",trans=none" : "");
                 }
                 break;
+            case QEMU_OPTION_numa:
+                if (nb_numa_nodes >= MAX_NODES) {
+                    fprintf(stderr, "qemu: too many NUMA nodes\n");
+                    exit(1);
+                }
+                numa_add(optarg);
+                break;
             case QEMU_OPTION_nographic:
                 nographic = 1;
                 break;
@@ -5211,6 +5285,48 @@ int main(int argc, char **argv, char **envp)
         }
     }
 
+    if (nb_numa_nodes > 0) {
+        int i;
+
+        if (nb_numa_nodes > smp_cpus) {
+            nb_numa_nodes = smp_cpus;
+        }
+
+        /* If no memory size if given for any node, assume the default case
+         * and distribute the available memory equally across all nodes
+         */
+        for (i = 0; i < nb_numa_nodes; i++) {
+            if (node_mem[i] != 0)
+                break;
+        }
+        if (i == nb_numa_nodes) {
+            uint64_t usedmem = 0;
+
+            /* On Linux, the each node's border has to be 8MB aligned,
+             * the final node gets the rest.
+             */
+            for (i = 0; i < nb_numa_nodes - 1; i++) {
+                node_mem[i] = (ram_size / nb_numa_nodes) & ~((1 << 23UL) - 1);
+                usedmem += node_mem[i];
+            }
+            node_mem[i] = ram_size - usedmem;
+        }
+
+        for (i = 0; i < nb_numa_nodes; i++) {
+            if (node_cpumask[i] != 0)
+                break;
+        }
+        /* assigning the VCPUs round-robin is easier to implement, guest OSes
+         * must cope with this anyway, because there are BIOSes out there in
+         * real machines which also use this scheme.
+         */
+        if (i == nb_numa_nodes) {
+            for (i = 0; i < smp_cpus; i++) {
+                node_cpumask[i % nb_numa_nodes] |= 1 << i;
+            }
+        }
+    }
+
     if (kvm_enabled()) {
         int ret;
 
@@ -5274,6 +5390,15 @@ int main(int argc, char **argv, char **envp)
     machine->init(ram_size, vga_ram_size, boot_devices,
                   kernel_filename, kernel_cmdline, initrd_filename, cpu_model);
 
+
+    for (env = first_cpu; env != NULL; env = env->next_cpu) {
+        for (i = 0; i < nb_numa_nodes; i++) {
+            if (node_cpumask[i] & (1 << env->cpu_index)) {
+                env->numa_node = i;
+            }
+        }
+    }
+
     current_machine = machine;
 
     /* Set KVM's vcpu state to qemu's initial CPUState. */
-- 
1.6.1.3