On Aug 26, 2014, at 3:29 PM, Thomas Graf <tg...@noironetworks.com> wrote:
> On 08/26/14 at 12:02pm, Jarno Rajahalme wrote: >> Miniflows can also be dynamically allocated to have more inline space than >> the default amount. classifier.c makes use of that, but it never calls >> miniflow_move (even indirectly) so you are not getting an assert fail. The >> motivation for right-sized miniflows in the classifier is to reduce levels >> of indirection while not wasting memory. > > I even did a grep for additional alloc sites but missed the embedded > miniflow in cls_match. Neat but really non obvious ;-) > > The code looked so extremely wrong with dead code branch *plus* > out of bounds access of the array. Sorry for the noise. > No worries :-) It’s always nice to have another set of eyes on the code! >> /* Initializes 'dst' with the data in 'src', destroying 'src'. >> * The caller must eventually free 'dst' with miniflow_destroy(). >> * 'dst' must be regularly sized miniflow, but 'src' can have >> * larger than default inline values. */ > > Maybe also adjust this comment a bit. I didn't get that "larger > than default inline values" refers to the *size* of > inline_values. Done in a patch I just sent out. Jarno _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
