Skip to contentSkip to conversation

qemu, numa: non-contiguous cpusets

PATCH Borislav Petkov 3 participants 5 messages Sep 29, 2013 – Sep 30, 2013 KVM

lore
Share & tools
lore
#0 lore raw

Btw,

while I got your attention, on a not-really related topic: how do we feel about adding support for specifying a non-contiguous set of cpus for a numa node in qemu with the -numa option? I.e., like this, for example:

x86_64-softmmu/qemu-system-x86_64 -smp 8 -numa node,nodeid=0,cpus=0\;2\;4-5 -numa node,nodeid=1,cpus=1\;3\;6-7

The ';' needs to be escaped from the shell but I'm open for better suggestions.

Here's a diff:

---

vl.c +32 -13
index 4e709d5c1c20..82a6c8451fb0 100644
@@ -1261,9 +1261,27 @@ char *get_boot_devices_list(size_t *size)
     return list;
 }
 
+static int __numa_set_cpus(unsigned long *map, int start, int end)
+{
+    if (end >= MAX_CPUMASK_BITS) {
+        end = MAX_CPUMASK_BITS - 1;
+        fprintf(stderr,
+            "qemu: NUMA: A max of %d VCPUs are supported\n",
+             MAX_CPUMASK_BITS);
+        return -EINVAL;
+    }
+
+    if (end < start) {
+        return -EINVAL;
+    }
+
+    bitmap_set(map, start, end-start+1);
+    return 0;
+}
+
 static void numa_node_parse_cpus(int nodenr, const char *cpus)
 {
-    char *endptr;
+    char *endptr, *ptr = (char *)cpus;
     unsigned long long value, endvalue;
 
     /* Empty CPU range strings will be considered valid, they will simply
@@ -1273,7 +1291,8 @@ static void numa_node_parse_cpus(int nodenr, const char *cpus)
         return;
     }
 
-    if (parse_uint(cpus, &value, &endptr, 10) < 0) {
+fromthetop:
+    if (parse_uint(ptr, &value, &endptr, 10) < 0) {
         goto error;
     }
     if (*endptr == '-') {
@@ -1282,22 +1301,22 @@ static void numa_node_parse_cpus(int nodenr, const char *cpus)
         }
     } else if (*endptr == '\0') {
         endvalue = value;
-    } else {
-        goto error;
-    }
+    } else if (*endptr == ';') {
+	if (__numa_set_cpus(node_cpumask[nodenr], value, value) < 0) {
+		goto error;
+	}
+	endptr++;
+        if (*endptr == '\0')
+	    return;
 
-    if (endvalue >= MAX_CPUMASK_BITS) {
-        endvalue = MAX_CPUMASK_BITS - 1;
-        fprintf(stderr,
-            "qemu: NUMA: A max of %d VCPUs are supported\n",
-             MAX_CPUMASK_BITS);
-    }
+	ptr = endptr;
 
-    if (endvalue < value) {
+	goto fromthetop;
+    } else {
         goto error;
     }
 
-    bitmap_set(node_cpumask[nodenr], value, endvalue-value+1);
+    __numa_set_cpus(node_cpumask[nodenr], value, endvalue);
     return;
 
 error:

--

Thanks.

Signature · 5 lines
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--
#1 lore raw
Quoted 11 lines · Borislav Petkov

Btw,

while I got your attention, on a not-really related topic: how do we feel about adding support for specifying a non-contiguous set of cpus for a numa node in qemu with the -numa option? I.e., like this, for example:

x86_64-softmmu/qemu-system-x86_64 -smp 8 -numa node,nodeid=0,cpus=0\;2\;4-5 -numa node,nodeid=1,cpus=1\;3\;6-7

The ';' needs to be escaped from the shell but I'm open for better suggestions.

Use a ':' instead.

Daniel

Signature · 4 lines
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|
maintainer
#2 lore raw
In reply to Borislav Petkov

Il 29/09/2013 17:10, Borislav Petkov ha scritto:

Quoted 8 lines

Btw,

while I got your attention, on a not-really related topic: how do we feel about adding support for specifying a non-contiguous set of cpus for a numa node in qemu with the -numa option? I.e., like this, for example:

x86_64-softmmu/qemu-system-x86_64 -smp 8 -numa node,nodeid=0,cpus=0\;2\;4-5 -numa node,nodeid=1,cpus=1\;3\;6-7

I think there are already patches on the list to do that, as part of the NUMA memory binding series from Wanlong Gao.

Paolo

Quoted 87 lines

The ';' needs to be escaped from the shell but I'm open for better suggestions.

Here's a diff:

---

vl.c +32 -14
index 4e709d5c1c20..82a6c8451fb0 100644
@@ -1261,9 +1261,27 @@ char *get_boot_devices_list(size_t *size)
     return list;
 }
 
+static int __numa_set_cpus(unsigned long *map, int start, int end)
+{
+    if (end >= MAX_CPUMASK_BITS) {
+        end = MAX_CPUMASK_BITS - 1;
+        fprintf(stderr,
+            "qemu: NUMA: A max of %d VCPUs are supported\n",
+             MAX_CPUMASK_BITS);
+        return -EINVAL;
+    }
+
+    if (end < start) {
+        return -EINVAL;
+    }
+
+    bitmap_set(map, start, end-start+1);
+    return 0;
+}
+
 static void numa_node_parse_cpus(int nodenr, const char *cpus)
 {
-    char *endptr;
+    char *endptr, *ptr = (char *)cpus;
     unsigned long long value, endvalue;
 
     /* Empty CPU range strings will be considered valid, they will simply
@@ -1273,7 +1291,8 @@ static void numa_node_parse_cpus(int nodenr, const char *cpus)
         return;
     }
 
-    if (parse_uint(cpus, &value, &endptr, 10) < 0) {
+fromthetop:
+    if (parse_uint(ptr, &value, &endptr, 10) < 0) {
         goto error;
     }
     if (*endptr == '-') {
@@ -1282,22 +1301,22 @@ static void numa_node_parse_cpus(int nodenr, const char *cpus)
         }
     } else if (*endptr == '\0') {
         endvalue = value;
-    } else {
-        goto error;
-    }
+    } else if (*endptr == ';') {
+	if (__numa_set_cpus(node_cpumask[nodenr], value, value) < 0) {
+		goto error;
+	}
+	endptr++;
+        if (*endptr == '\0')
+	    return;
 
-    if (endvalue >= MAX_CPUMASK_BITS) {
-        endvalue = MAX_CPUMASK_BITS - 1;
-        fprintf(stderr,
-            "qemu: NUMA: A max of %d VCPUs are supported\n",
-             MAX_CPUMASK_BITS);
-    }
+	ptr = endptr;
 
-    if (endvalue < value) {
+	goto fromthetop;
+    } else {
         goto error;
     }
 
-    bitmap_set(node_cpumask[nodenr], value, endvalue-value+1);
+    __numa_set_cpus(node_cpumask[nodenr], value, endvalue);
     return;
 
 error:
--

Thanks.

#3 lore raw
Quoted 2 lines · Paolo Bonzini

I think there are already patches on the list to do that, as part of the NUMA memory binding series from Wanlong Gao.

Yeah: https://lists.gnu.org/archive/html/qemu-devel/2013-09/msg02833.html

Although I don't see from it how the syntax for -cpus will look like from that QAPI magic except that it is an

+ '*cpus': ['uint16'],

:-)

Thanks.

Signature · 5 lines
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--