[dpdk-dev] Calling rte_eal_init multiple times in the same process

[dpdk-dev] Calling rte_eal_init multiple times in the same process

[Tomas Vestelind]

Dear all,

I'm not sure who to address with this email so I'll just present my problem and my suggested solution.

The problem I'm facing is that I need to do rte_eal_init serveral times in the same process. The reason for this is mostly architectural as we want to divide certain functionality in to different, implementation independent, modules in which DPDK would be used for different purposes. Thus we would need to eventually call rte_eal_init several times because we can't say for sure if it has been called by another module or not.

While thinking about it I saw no good, existing solution to this (I might have missed it though). Would returning a "DPDK has already been initialized" error code in rte_eal_init be a good way to go about it? I haven't done this yet but I could probably provide a patch if it's interesting.

Thank you for your time and a very nice framework with good documentation!

BR,
Tomas

Re: [dpdk-dev] Calling rte_eal_init multiple times in the same process

[Thomas Monjalon]

Hello,

2014-04-17 09:09, Tomas Vestelind:
> The problem I'm facing is that I need to do rte_eal_init serveral times in
> the same process. The reason for this is mostly architectural as we want to
> divide certain functionality in to different, implementation independent,
> modules in which DPDK would be used for different purposes. Thus we would
> need to eventually call rte_eal_init several times because we can't say for
> sure if it has been called by another module or not.
> 
> While thinking about it I saw no good, existing solution to this (I might
> have missed it though). Would returning a "DPDK has already been
> initialized" error code in rte_eal_init be a good way to go about it? I
> haven't done this yet but I could probably provide a patch if it's
> interesting.

Don't hesitate to submit a patch. It's far easier to discuss on a patch basis.
If your patch fill your need and nobody complains about it, it will be 
accepted.

-- 
Thomas

Re: [dpdk-dev] Calling rte_eal_init multiple times in the same process

[Neil Horman]

On Thu, Apr 17, 2014 at 09:09:44AM +0000, Tomas Vestelind wrote:
> Dear all,
> 
> I'm not sure who to address with this email so I'll just present my problem and my suggested solution.
> 
> The problem I'm facing is that I need to do rte_eal_init serveral times in the same process. The reason for this is mostly architectural as we want to divide certain functionality in to different, implementation independent, modules in which DPDK would be used for different purposes. Thus we would need to eventually call rte_eal_init several times because we can't say for sure if it has been called by another module or not.
> 
> While thinking about it I saw no good, existing solution to this (I might have missed it though). Would returning a "DPDK has already been initialized" error code in rte_eal_init be a good way to go about it? I haven't done this yet but I could probably provide a patch if it's interesting.
> 
> Thank you for your time and a very nice framework with good documentation!
> 
> BR,
> Tomas
> 


You already (more or less) have that.  rte_eal_init panics or calls exit() if
something goes wrong during initialization.  If all goes well it returns 0.  The
only time that it returns with something else is if this code at the top of the
function fails:

if (!rte_atomic32_test_and_set(&run_once))
                return -1;

Thats there to specifically gate the init routine so it only gets called once.
All your code needs to do is check the return for -1.  Thats your indicator that
its already been initalized.  Anything >= 0 is success, and if your program
crashes, thats a problem :)

Neil