]> git.ipfire.org Git - thirdparty/qemu.git/commitdiff
migration: Fix possible division by zero on calc expected downtime
authorPeter Xu <peterx@redhat.com>
Mon, 11 May 2026 18:24:32 +0000 (14:24 -0400)
committerPeter Xu <peterx@redhat.com>
Tue, 19 May 2026 20:23:13 +0000 (16:23 -0400)
Commit dd4fe8844b changed the reporting of expected downtime behavior, so
that the value will be calculated on-demand.  One side effect on the change
is QEMU will allow the calculation to happen anytime even if there's no
transfer happening for a short while.

PeterM reported an ubsan report from clang when running migration-test with
aarch64 binary on x86_64 hosts.  I can also reproduce if I run the test
concurrently so some of the src QEMU may not get chance to push any data,
causing mbps to be 0:

../migration/migration.c:1051:12: runtime error: -nan is outside the range of representable values of type 'long'

Fix it by properly handle both Inf and Nan to return INT64_MAX.

Add a rich comment, per PeterM's suggestion.

Link: https://lore.kernel.org/r/CAFEAcA-MYH6C39xO0OLx4-M5pKurJpurwRsMqZe9q=W-NShAbw@mail.gmail.com
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Fixes: dd4fe8844b ("migration: Calculate expected downtime on demand")
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Link: https://lore.kernel.org/r/20260511182432.1333467-1-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
migration/migration.c

index b6f78eb3ac5c859c70ead48999ec7e068990ac5c..05f10e4576d9f289e85e64afd6da0d5aa5b54a79 100644 (file)
@@ -63,6 +63,7 @@
 #include "system/dirtylimit.h"
 #include "qemu/sockets.h"
 #include "system/kvm.h"
+#include "math.h"
 
 #define NOTIFIER_ELEM_INIT(array, elem)    \
     [elem] = NOTIFIER_WITH_RETURN_LIST_INITIALIZER((array)[elem])
@@ -1044,12 +1045,29 @@ static bool migrate_show_downtime(MigrationState *s)
 /* Return expected downtime (unit: milliseconds) */
 int64_t migration_downtime_calc_expected(MigrationState *s)
 {
+    double expected_ms;
+
     if (mig_stats.dirty_sync_count <= 1) {
         return migrate_downtime_limit();
     }
 
-    return mig_stats.dirty_bytes_last_sync /
+    expected_ms = mig_stats.dirty_bytes_last_sync /
         migration_get_switchover_bw(s) * 1000;
+
+    /*
+     * If we haven't been able to transfer any data, the result here could
+     * be NaN (for 0 / 0) or infinity (something else / 0).
+     *
+     * Return INT64_MAX as our best approximation to "this will take
+     * forever to complete". If the problem is transient (e.g. we just
+     * haven't started to transfer yet) we'll recalculate to a more
+     * accurate figure later.
+     */
+    if (isnan(expected_ms) || expected_ms >= (double)INT64_MAX) {
+        return INT64_MAX;
+    }
+
+    return (int64_t) expected_ms;
 }
 
 static void populate_time_info(MigrationInfo *info, MigrationState *s)