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

Re: [libvirt] [PATCH 2/2] FreeBSD: stub out CPU affinity functions.



On 12/17/12 18:25, Daniel P. Berrange wrote:
On Mon, Dec 17, 2012 at 09:09:52PM +0400, Roman Bogorodskiy wrote:
   Daniel P. Berrange wrote:

On Sun, Dec 16, 2012 at 06:47:56PM +0400, Roman Bogorodskiy wrote:
---
  src/util/processinfo.c | 22 ++++++++++++++++++++++
  1 file changed, 22 insertions(+)

diff --git a/src/util/processinfo.c b/src/util/processinfo.c
index b1db049..4822bcc 100644
--- a/src/util/processinfo.c
+++ b/src/util/processinfo.c
@@ -168,6 +168,28 @@ realloc:
      return 0;
  }

+#elif defined(__FreeBSD__)
+
+int virProcessInfoSetAffinity(pid_t pid ATTRIBUTE_UNUSED,
+                              virBitmapPtr map ATTRIBUTE_UNUSED)
+{
+    return 0;
+}

Hmm, I'm somewhat loathe to be pretending that this works.

IMHO, we should look at 'map' and if it is all-1s then
we should accept it trivially. If it is not all-1s then
we should raise an error so the user knows their requested
config was not honoured

The reasoning behind this change is, as subject says, provide a stub
while there's no complete implementation available. I've been using
it for a couple of weeks and didn't notice any problems.

Would not it make sense to just throw a warning or not critical error to
user so he knows there's no affinity? I was planning to create a
full-feature implementation of that after I'm done with the networking
support.

Ok, since you're planning on implementing this properly, I'll accept
this patch as is for now

As I said in my review, I don't like this approach even if it's going to be improved in the future. Until it's approved we should go with Dan's idea about accepting all-1-map and erroring out on everything else.

I will add that to this patch and push it tomorrow.

Peter


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