[PATCH 5/5] [POST-4.0]: HVM NUMA guest: add xc_nodeload() function

[Andre Przywara <andre.przywara@amd.com>]

[-- Attachment #1: Type: text/plain, Size: 1022 bytes --]

This is actually work in progress, as it's just the base function which 
should be used libxc's setup_guest().
It mimics the current XendDomainInfo.py:find_relaxed_node() code, which 
can eventually go away.
The function iterates over all VCPUs in all domains to determine each 
node's load. It considers VCPUs pinned to single nodes vs. unpinned 
VCPUs which result in a lower load for each node (although the sum is 
equal). Also added is the amount of free memory, this can be used to 
sequentially determine the best host node for each guest node.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>

Regards,
Andre.

-- 
Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany
Tel: +49 351 488-3567-12
----to satisfy European Law for business letters:
Advanced Micro Devices GmbH
Karl-Hammerschmidt-Str. 34, 85609 Dornach b. Muenchen
Geschaeftsfuehrer: Andrew Bowd; Thomas M. McCoy; Giuliano Meroni
Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen
Registergericht Muenchen, HRB Nr. 43632

[-- Attachment #2: 05-hvm_numa_host_node_load.patch --]
[-- Type: text/plain, Size: 3734 bytes --]

commit 3199a0f02fdd8379bd8694ac57f307234075d0c2
Author: Andre Przywara <andre.przywara@amd.com>
Date:   Thu Feb 4 13:51:53 2010 +0100

    added xc_nodeload() function

diff --git a/tools/libxc/xc_misc.c b/tools/libxc/xc_misc.c
index 04da3d2..9ae61b6 100644
--- a/tools/libxc/xc_misc.c
+++ b/tools/libxc/xc_misc.c
@@ -424,6 +424,70 @@ xc_map_foreign_bulk(int xc_handle, uint32_t dom, int prot,
     return ret;
 }
 
+#define MAX_DOMS 8
+#define MAX_CPU_ID 255
+
+struct xc_nodeload* xc_getnodeload(int xc_handle,
+                                   int *ret_nrnodes)
+{
+       xc_dominfo_t dominfo[MAX_DOMS];
+       xc_vcpuinfo_t vcpuinfo;
+       int nextdom = 0;
+       int nrdoms, dom, vcpu, i;
+       xc_physinfo_t physinfo;
+       uint32_t nodemap[MAX_CPU_ID + 1];
+       int nrcpus, nrnodes, sum;
+       uint64_t cpumap;
+       int *curload;
+    struct xc_nodeload *nodeload;
+
+       set_xen_guest_handle(physinfo.cpu_to_node, nodemap);
+       physinfo.max_cpu_id = MAX_CPU_ID;
+
+    if (xc_physinfo(xc_handle, &physinfo) != 0)
+        return NULL;
+    nrnodes = physinfo.max_node_id + 1;
+    nrcpus = physinfo.max_cpu_id + 1;
+    curload = malloc(nrnodes * sizeof(int));
+    nodeload = malloc(nrnodes * sizeof(struct xc_nodeload));
+    for (i = 0; i < nrnodes; i++) {
+        nodeload[i].node = i;
+        nodeload[i].load = 0;
+        xc_availheap(xc_handle, 0, 0, i, &nodeload[i].freemem);
+        nodeload[i].freemem /= 1024;
+    }
+
+    while ((nrdoms = xc_domain_getinfo(xc_handle,
+                                       nextdom, MAX_DOMS, dominfo)) != -1) {
+        for (dom = 0; dom < nrdoms; dom++) {
+            for (vcpu = 0; vcpu <= dominfo[dom].max_vcpu_id; vcpu++) {
+                xc_vcpu_getinfo(xc_handle, dominfo[dom].domid, vcpu, &vcpuinfo);
+                if (!vcpuinfo.online)
+                    continue;
+                   xc_vcpu_getaffinity(xc_handle, dominfo[dom].domid, vcpu,
+                                       &cpumap);
+                memset(curload, 0, sizeof(int) * nrnodes);
+                for (i = sum = 0; i < nrcpus; i++) {
+                               if ((1ULL << i) & cpumap) {
+                                       if (curload[nodemap[i]] == 0)
+                                               sum++;
+                                       curload[nodemap[i]] = 1;
+                               }
+                }
+                for (i = 0; i < nrnodes; i++)
+                   nodeload[i].load += curload[i] * nrnodes / sum;
+            }
+        }
+        if (nrdoms < 2)
+            break;
+        nextdom = dominfo[nrdoms - 1].domid + 1;
+    }
+    free(curload);
+    if (ret_nrnodes != NULL)
+        *ret_nrnodes = nrnodes;
+    return nodeload;
+}
+
 /*
  * Local variables:
  * mode: C
diff --git a/tools/libxc/xenctrl.h b/tools/libxc/xenctrl.h
index 79c1fd4..0b6ef9a 100644
--- a/tools/libxc/xenctrl.h
+++ b/tools/libxc/xenctrl.h
@@ -613,6 +613,22 @@ typedef uint32_t xc_cpu_to_node_t;
 int xc_physinfo(int xc_handle,
                 xc_physinfo_t *info);
 
+struct xc_nodeload {
+    int node;
+    int load;
+    uint64_t freemem;
+};
+
+/* Iterate over all domains and VCPUs to get an estimate of the load
+ * distribution of the NUMA nodes. Returns the malloced array with
+ * an entry for each node. ret_nrnodes is optional (can be NULL) and
+ * contains the number of entries in the array.
+ * The node member contains the node number, useful if you sort the array.
+ * The load value is a relative metric, freemem is given in KB.
+ */
+struct xc_nodeload* xc_getnodeload(int xc_handle,
+                                   int *ret_nrnodes);
+
 int xc_sched_id(int xc_handle,
                 int *sched_id);
 

[-- Attachment #3: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel