[libvirt] [PATCH v4 0/9] Implement Add/Del IOThreads

John Ferlan jferlan at redhat.com
Thu Apr 23 19:28:31 UTC 2015



So obviously I messed something up with the threading on the responses -
thought I had it right, but not quite I guess. Oh well - I think it's
obvious what I was trying to do ;-)

v5 updates:

Patch 1/10 (forgot to edit the 10)

 * Fixes the syntax check
 * Moves the PostParse autofill of iothreadids right after the possible
reading from XML

Patch 4/9

 * Add virDomainIOThreadIDArrayHasPin which will looking through the
niothreadids[] array for any iothreadid[] with a cpumask defined
 * Fail in virDomainIOThreadPinDefParseXML if the iothreadid cannot be found
 * Remove [n]iothreadspin references
 * Left the cpumask checking to default - I think all should be handled
in a later/separate patch
 * Fixed  qemuSetupCgroupForIOThreads cpumask ordering

Patch 5/9
 * Adjusted the test to use 'fifo' *and* priority='1' for second entry
(got a failure in a different test after my first pass)

Patch 7.5/9
 * NEW - Code that will remove the iothread_id bit from the ids mask. I
did test this and it worked like I expected...

Patch 8/9
 * Not quite sure what was meant by auto numa placement so I didn't add
that into the patch I sent, but if the following is what you meant, then
I can squash it in (and define cpumask appropriately)

+    if (vm->def->placement_mode == VIR_DOMAIN_CPU_PLACEMENT_MODE_AUTO)
+        cpumask = priv->autoCpuset;
+    else
+        cpumask = vm->def->cpumask;
+
+    if (cpumask) {
+        if (qemuDomainHotplugPinThread(cpumask, iothread_id,

 * Remove qemuProcessSetSchedParams

 * Remove niothreadspin

 * Remove the virNumaIsAvailable() hunk from qemuDomainChgIOThread

Tks,

John




More information about the libvir-list mailing list