[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

[PATCH] qemu: Do not error out when setting affinity failed



At least in a particular scenario described in the code.  Basically when
libvirtd is running without CAP_SYS_NICE (e.g. in a container) and it is trying
to set QEMU affinity to all CPUs (because there is no setting requested in the
XML) it fails.  But if we ignore the failure in this particular case than you
can limit the CPUs used by controlling the affinity for libvirtd itself.

In any other case (anything requested in the XML, pinning a live domain, etc.)
the call is still considered fatal and the action errors out.

Resolves: https://bugzilla.redhat.com/1819801

Suggested-by: Daniel P. Berrangé <berrange redhat com>
Signed-off-by: Martin Kletzander <mkletzan redhat com>
---
 src/qemu/qemu_process.c | 41 ++++++++++++++++++++++++++++++++++++++---
 1 file changed, 38 insertions(+), 3 deletions(-)

diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c
index cfe09d632633..270bb37d3682 100644
--- a/src/qemu/qemu_process.c
+++ b/src/qemu/qemu_process.c
@@ -2528,6 +2528,7 @@ qemuProcessGetAllCpuAffinity(virBitmapPtr *cpumapRet)
 static int
 qemuProcessInitCpuAffinity(virDomainObjPtr vm)
 {
+    bool settingAll = false;
     g_autoptr(virBitmap) cpumapToSet = NULL;
     virDomainNumatuneMemMode mem_mode;
     qemuDomainObjPrivatePtr priv = vm->privateData;
@@ -2566,13 +2567,30 @@ qemuProcessInitCpuAffinity(virDomainObjPtr vm)
         if (!(cpumapToSet = virBitmapNewCopy(vm->def->cputune.emulatorpin)))
             return -1;
     } else {
+        settingAll = true;
         if (qemuProcessGetAllCpuAffinity(&cpumapToSet) < 0)
             return -1;
     }
 
     if (cpumapToSet &&
         virProcessSetAffinity(vm->pid, cpumapToSet) < 0) {
-        return -1;
+        /*
+         * We only want to error out if we failed to set the affinity to
+         * user-requested mapping.  If we are just trying to reset the affinity
+         * to all CPUs and this fails it can only be an issue if:
+         *  1) libvirtd does not have CAP_SYS_NICE
+         *  2) libvirtd does not run on all CPUs
+         *
+         * However since this scenario is very improbable, we rather skip
+         * reporting the error because it helps running libvirtd in a a scenario
+         * where pinning is handled by someone else.
+         *
+         * See also: https://bugzilla.redhat.com/1819801#c2
+         */
+        if (settingAll)
+            virResetLastError();
+        else
+            return -1;
     }
 
     return 0;
@@ -2726,8 +2744,25 @@ qemuProcessSetupPid(virDomainObjPtr vm,
         affinity_cpumask = use_cpumask;
 
     /* Setup legacy affinity. */
-    if (affinity_cpumask && virProcessSetAffinity(pid, affinity_cpumask) < 0)
-        goto cleanup;
+    if (affinity_cpumask && virProcessSetAffinity(pid, affinity_cpumask) < 0) {
+        /*
+         * We only want to error out if we failed to set the affinity to
+         * user-requested mapping.  If we are just trying to reset the affinity
+         * to all CPUs and this fails it can only be an issue if:
+         *  1) libvirtd does not have CAP_SYS_NICE
+         *  2) libvirtd does not run on all CPUs
+         *
+         * However since this scenario is very improbable, we rather skip
+         * reporting the error because it helps running libvirtd in a a scenario
+         * where pinning is handled by someone else.
+         *
+         * See also: https://bugzilla.redhat.com/1819801#c2
+         */
+        if (affinity_cpumask == hostcpumap)
+            virResetLastError();
+        else
+            goto cleanup;
+    }
 
     /* Set scheduler type and priority, but not for the main thread. */
     if (sched &&
-- 
2.28.0


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]