virNumaGetPageInfo: Take huge pages into account

On the Linux kernel, if huge pages are allocated the size they cut off
from memory is accounted under the 'MemUsed' in the meminfo file.
However, we want the sum to be subtracted from 'MemTotal'. This patch
implements this feature. After this change, we can enable reporting
of the ordinary system pages in the capability XML:

<capabilities>

  <host>
    <uuid>01281cda-f352-cb11-a9db-e905fe22010c</uuid>
    <cpu>
      <arch>x86_64</arch>
      <model>Haswell</model>
      <vendor>Intel</vendor>
      <topology sockets='1' cores='1' threads='1'/>
      <feature/>
      <pages unit='KiB' size='4'/>
      <pages unit='KiB' size='2048'/>
      <pages unit='KiB' size='1048576'/>
    </cpu>
    <power_management/>
    <migration_features/>
    <topology>
      <cells num='4'>
        <cell id='0'>
          <memory unit='KiB'>4048248</memory>
          <pages unit='KiB' size='4'>748382</pages>
          <pages unit='KiB' size='2048'>3</pages>
          <pages unit='KiB' size='1048576'>1</pages>
          <distances/>
          <cpus num='1'>
            <cpu id='0' socket_id='0' core_id='0' siblings='0'/>
          </cpus>
        </cell>
        ...
      </cells>
    </topology>
  </host>
</capabilities>

You can see the beautiful thing about this: if you sum up all the
<pages/> you'll get <memory/>.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>

diff --git a/src/nodeinfo.c b/src/nodeinfo.c
index 8fe7adccaf380510ffc22c5c641d5d1e05f5e768..2c1c4379bbf34d32094645bbcff5b9e1bc2de47a 100644 (file)
--- a/src/nodeinfo.c
+++ b/src/nodeinfo.c
@@ -2038,7 +2038,7 @@ nodeGetFreePages(unsigned int npages,
             unsigned int page_size = pages[i];
             unsigned int page_free;
 
-            if (virNumaGetPageInfo(cell, page_size, NULL, &page_free) < 0)
+            if (virNumaGetPageInfo(cell, page_size, 0, NULL, &page_free) < 0)
                 goto cleanup;
 
             counts[ncounts++] = page_free;

diff --git a/src/util/virnuma.c b/src/util/virnuma.c
index 27b6ecb9a764857d08080328083b1ed563dd53c4..66d228181bf72575dc1bea3f0100aecfaef7bf2d 100644 (file)
--- a/src/util/virnuma.c
+++ b/src/util/virnuma.c
@@ -640,6 +640,8 @@ virNumaGetHugePageInfo(int node,
  * virNumaGetPageInfo:
  * @node: NUMA node id
  * @page_size: which huge page are we interested in (in KiB)
+ * @huge_page_sum: the sum of memory taken by huge pages (in
+ * bytes)
  * @page_avail: total number of huge pages in the pool
  * @page_free: the number of free huge pages in the pool
  *
@@ -647,6 +649,13 @@ virNumaGetHugePageInfo(int node,
  * total number of pages in the pool (both free and taken)
  * and count for free pages in the pool.
  *
+ * The @huge_page_sum parameter exists due to the Linux kernel
+ * limitation. The problem is, if there are some huge pages
+ * allocated, they are accounted under the 'MemUsed' field in the
+ * meminfo file instead of being subtracted from the 'MemTotal'.
+ * We must do the subtraction ourselves.
+ * If unsure, pass 0.
+ *
  * If you're interested in just one bit, pass NULL to the other one.
  *
  * As a special case, if @node == -1, overall info is fetched
@@ -657,6 +666,7 @@ virNumaGetHugePageInfo(int node,
 int
 virNumaGetPageInfo(int node,
                    unsigned int page_size,
+                   unsigned long long huge_page_sum,
                    unsigned int *page_avail,
                    unsigned int *page_free)
 {
@@ -666,7 +676,6 @@ virNumaGetPageInfo(int node,
     /* sysconf() returns page size in bytes,
      * the @page_size is however in kibibytes */
     if (page_size == system_page_size / 1024) {
-# if 0
         unsigned long long memsize, memfree;
 
         /* TODO: come up with better algorithm that takes huge pages into
@@ -679,16 +688,14 @@ virNumaGetPageInfo(int node,
                 goto cleanup;
         }
 
+        /* see description above */
+        memsize -= huge_page_sum;
+
         if (page_avail)
             *page_avail = memsize / system_page_size;
 
         if (page_free)
             *page_free = memfree / system_page_size;
-# else
-        virReportError(VIR_ERR_ARGUMENT_UNSUPPORTED, "%s",
-                       _("system page size are not supported yet"));
-        goto cleanup;
-# endif /* 0 */
     } else {
         if (virNumaGetHugePageInfo(node, page_size, page_avail, page_free) < 0)
             goto cleanup;
@@ -737,31 +744,18 @@ virNumaGetPages(int node,
     unsigned int ntmp = 0;
     size_t i;
     bool exchange;
-
-# if 0
-    /* This has to be disabled until the time the issue in
-     * virNumaGetPageInfo is resolved. Sorry. */
     long system_page_size;
+    unsigned long long huge_page_sum = 0;
 
     /* sysconf() returns page size in bytes,
      * but we are storing the page size in kibibytes. */
     system_page_size = sysconf(_SC_PAGESIZE) / 1024;
 
-    /* We know that ordinary system pages are supported
-     * if nothing else is. */
-    if (VIR_REALLOC_N(tmp_size, 1) < 0 ||
-        VIR_REALLOC_N(tmp_avail, 1) < 0 ||
-        VIR_REALLOC_N(tmp_free, 1) < 0)
-        goto cleanup;
-
-    if (virNumaGetPageInfo(node, system_page_size,
-                           &tmp_avail[ntmp], &tmp_free[ntmp]) < 0)
-        goto cleanup;
-    tmp_size[ntmp] = system_page_size;
-    ntmp++;
-# endif /* 0 */
-
-    /* Now that we got ordinary system pages, lets get info on huge pages */
+    /* Query huge pages at first.
+     * On Linux systems, the huge pages pool cuts off the available memory and
+     * is always shown as used memory. Here, however, we want to report
+     * slightly different information. So we take the total memory on a node
+     * and subtract memory taken by the huge pages. */
     if (virNumaGetHugePageInfoPath(&path, node, 0, NULL) < 0)
         goto cleanup;
 
@@ -792,9 +786,7 @@ virNumaGetPages(int node,
             goto cleanup;
         }
 
-        /* Querying more detailed info makes sense only sometimes */
-        if ((pages_avail || pages_free) &&
-            virNumaGetHugePageInfo(node, page_size,
+        if (virNumaGetHugePageInfo(node, page_size,
                                    &page_avail, &page_free) < 0)
             goto cleanup;
 
@@ -807,11 +799,27 @@ virNumaGetPages(int node,
         tmp_avail[ntmp] = page_avail;
         tmp_free[ntmp] = page_free;
         ntmp++;
+
+        /* page_size is in kibibytes while we want huge_page_sum
+         * in just bytes. */
+        huge_page_sum += 1024 * page_size * page_avail;
     }
 
     if (direrr < 0)
         goto cleanup;
 
+    /* Now append the ordinary system pages */
+    if (VIR_REALLOC_N(tmp_size, ntmp + 1) < 0 ||
+        VIR_REALLOC_N(tmp_avail, ntmp + 1) < 0 ||
+        VIR_REALLOC_N(tmp_free, ntmp + 1) < 0)
+        goto cleanup;
+
+    if (virNumaGetPageInfo(node, system_page_size, huge_page_sum,
+                           &tmp_avail[ntmp], &tmp_free[ntmp]) < 0)
+        goto cleanup;
+    tmp_size[ntmp] = system_page_size;
+    ntmp++;
+
     /* Just to produce nice output, sort the arrays by increasing page size */
     do {
         exchange = false;
@@ -854,6 +862,7 @@ virNumaGetPages(int node,
 int
 virNumaGetPageInfo(int node ATTRIBUTE_UNUSED,
                    unsigned int page_size ATTRIBUTE_UNUSED,
+                   unsigned long long huge_page_sum ATTRIBUTE_UNUSED,
                    unsigned int *page_avail ATTRIBUTE_UNUSED,
                    unsigned int *page_free ATTRIBUTE_UNUSED)
 {

diff --git a/src/util/virnuma.h b/src/util/virnuma.h
index a7435dc27ab03f323ee319eccc94dbfede543fa7..fbeb7a9559166512462ee4b731c526672004d01e 100644 (file)
--- a/src/util/virnuma.h
+++ b/src/util/virnuma.h
@@ -72,6 +72,7 @@ int virNumaGetNodeCPUs(int node, virBitmapPtr *cpus);
 
 int virNumaGetPageInfo(int node,
                        unsigned int page_size,
+                       unsigned long long huge_page_sum,
                        unsigned int *page_avail,
                        unsigned int *page_free);
 int virNumaGetPages(int node,