Re: [PATCH 1/7] eal: use different constructor priorities for initcalls

On Wednesday 15 February 2017 08:35 PM, Jan Blunck wrote:

On Wed, Feb 15, 2017 at 3:37 PM, Shreyansh Jain [off-list ref] wrote:

quoted

On Wednesday 15 February 2017 03:32 PM, Jan Blunck wrote:

quoted

--- a/lib/librte_eal/common/include/rte_eal.h
+++ b/lib/librte_eal/common/include/rte_eal.h

@@ -258,8 +258,16 @@ static inline int rte_gettid(void)
        return RTE_PER_LCORE(_thread_id);
 }

-#define RTE_INIT(func) \
-static void __attribute__((constructor, used)) func(void)
+#define RTE_EAL_INIT(func) \
+static void __attribute__((constructor(101), used)) func(void)
+
+#define RTE_POST_EAL_INIT(func) \
+static void __attribute__((constructor(102), used)) func(void)
+
+#define RTE_DEV_INIT(func) \
+static void __attribute__((constructor(103), used)) func(void)

Shouldn't we simply allow this priority to be default to allow for some
priority space between buses and default init?

The absolute numbers are not that important. We can always adjust
them. Important is the relative order. If you have a use-case for
something that needs to be initialized before the devices but can't
get initialized with the eal/post-eal then please speak up.

No use-case as of now.