On Sat, Jul 19, 2014 at 10:07:16PM +0300, Jarno Rajahalme wrote:
>
> > On Jul 19, 2014, at 8:27 PM, Ben Pfaff wrote:
> >
> > We could make some operations lockless, like read and store,
>
> These would be enough for RCU cmap and classifier, right? Refcount
> needs the read-modify-write operatio
> On Jul 19, 2014, at 8:27 PM, Ben Pfaff wrote:
>
> We could make some operations lockless, like read and store,
These would be enough for RCU cmap and classifier, right? Refcount needs the
read-modify-write operations, though.
> but I don't
> know how to make read-modify-write operations loc
On Sat, Jul 19, 2014 at 11:33:36AM +0300, Jarno Rajahalme wrote:
>
>
> Sent from my iPhone
>
> > On Jul 19, 2014, at 10:03 AM, Ben Pfaff wrote:
> >
> >> On Fri, Jul 18, 2014 at 11:33:38PM -0700, Ben Pfaff wrote:
> >>> On Fri, Jul 18, 2014 at 09:05:49PM -0700, Jarno Rajahalme wrote:
> >>> Older
CMAP_FOR_EACH and CLS_FOR_EACH and their variants tried to use void ** as
a "pointer to any kind of pointer". That is a violation of the aliasing
rules in ISO C which technically yields undefined behavior. With GCC 4.1,
it causes both warnings and actual misbehavior. One option would to add
-fno
Signed-off-by: Ben Pfaff
---
lib/classifier.c | 49 +
1 file changed, 21 insertions(+), 28 deletions(-)
diff --git a/lib/classifier.c b/lib/classifier.c
index 2ff539d..af28070 100644
--- a/lib/classifier.c
+++ b/lib/classifier.c
@@ -1250,36 +1250,1
We could make some operations lockless, like read and store, but I don't
know how to make read-modify-write operations lockless generically.
What do you have in mind?
On Sat, Jul 19, 2014 at 11:37:42AM +0300, Jarno Rajahalme wrote:
> Would it be a bad idea to make the trivial types with sizes less
I didn't look at the classifier yet. Maybe it can't work the same way.
On Jul 19, 2014 1:33 AM, "Jarno Rajahalme" wrote:
>
>
> Sent from my iPhone
>
> > On Jul 19, 2014, at 10:03 AM, Ben Pfaff wrote:
> >
> >> On Fri, Jul 18, 2014 at 11:33:38PM -0700, Ben Pfaff wrote:
> >>> On Fri, Jul 18, 2014 a
Would it be a bad idea to make the trivial types with sizes less than void
pointer lockless in ovs-atomic-pthread just like in gcc4+, at least for x86/64?
Jarno
> On Jul 19, 2014, at 10:05 AM, Ben Pfaff wrote:
>
> I guess not, I see that xenddk-6.0.0-50762, at least, doesn't have a
> libato
Sent from my iPhone
> On Jul 19, 2014, at 10:03 AM, Ben Pfaff wrote:
>
>> On Fri, Jul 18, 2014 at 11:33:38PM -0700, Ben Pfaff wrote:
>>> On Fri, Jul 18, 2014 at 09:05:49PM -0700, Jarno Rajahalme wrote:
>>> Older GCC (e.g. 4.1.2) did not like the pointer casts used for
>>> initializing the iter
I guess not, I see that xenddk-6.0.0-50762, at least, doesn't have a
libatomic.
On Sat, Jul 19, 2014 at 12:04:25AM -0700, Ben Pfaff wrote:
> Does it make a difference if you add -latomic to the linker flags?
>
> On Fri, Jul 18, 2014 at 05:32:39PM -0700, Jarno Rajahalme wrote:
> > From config.log:
Does it make a difference if you add -latomic to the linker flags?
On Fri, Jul 18, 2014 at 05:32:39PM -0700, Jarno Rajahalme wrote:
> From config.log:
>
> configure:13921: checking whether gcc -std=gnu99 supports GCC 4.0+ atomic
> built-ins
> configure:13988: gcc -std=gnu99 -o conftest -g -O2
On Fri, Jul 18, 2014 at 11:33:38PM -0700, Ben Pfaff wrote:
> On Fri, Jul 18, 2014 at 09:05:49PM -0700, Jarno Rajahalme wrote:
> > Older GCC (e.g. 4.1.2) did not like the pointer casts used for
> > initializing the iteration cursors. This patch changes the code to
> > avoid the void casts for GCC,
12 matches
Mail list logo
