Re: [dpdk-dev] rte_eal_init() alternative?

[Panu Matilainen <pmatilai@redhat.com>]

On 10/09/2015 01:13 PM, Montorsi, Francesco wrote:
>>> It seems the patch missed the boat :)
>>
>> Correct, sorry. I'm attaching it now.
> Ok, for some reason the email client is removing the attachment... I'm copying and pasting it:
> (the points marked as TODO are functions that still contain rte_panic() calls...)

I actually did receive the attachment from the previous mail, but 
inlined patches are far better for commenting purposes.

> + */
> +struct internal_config;
> +int rte_eal_init_raw(const char* logid, struct internal_config *cfg);

Like the name indicates, struct internal_config is internal to 
librte_eal, you'd need to "export" the eal_internal_cfg.h header for 
this to be useful to users outside librte_eal itself. But I'd say 
there's a reason why its internal...

> -	if (rte_eal_pci_init() < 0)
> -		rte_panic("Cannot init PCI\n");
> +	if (rte_eal_pci_init() < 0) {
> +		RTE_LOG (ERR, EAL, "Cannot init PCI\n");
> +		return -1;
> +	}
>
>   #ifdef RTE_LIBRTE_IVSHMEM
> -	if (rte_eal_ivshmem_init() < 0)
> -		rte_panic("Cannot init IVSHMEM\n");
> +	if (rte_eal_ivshmem_init() < 0) {
> +		RTE_LOG (ERR, EAL, "Cannot init IVSHMEM\n");
> +		return -1;
> +	}
>   #endif
>
> -	if (rte_eal_memory_init() < 0)
> -		rte_panic("Cannot init memory\n");
> +	if (rte_eal_memory_init() < 0) {
> +		RTE_LOG (ERR, EAL, "Cannot init memory\n");
> +		return -1;
> +	}
[...]

Something like that, sure. The big question with this conversion is what 
to do with already allocated/initialized resources in case of failure, 
which I'd guess is the reason rte_panic() is there - to avoid having to 
deal with all that.

Getting to a point where all or even most inialization can be undone in 
case of failure is likely going to be a long road, I think many 
subsystems dont even have a shutdown function. To beging with, EAL 
itself doesn't have one :)

Anyway, one has to start someplace. But in order to make the cleanup 
eventually possible, I'd suggest using a common point of exit instead of 
a dozen returns, ie something in spirit of

{
	[...]

	if (rte_eal_pci_init() < 0) {
		RTE_LOG (ERR, EAL, "Cannot init PCI\n");
		goto err;
	}

	if (rte_eal_memory_init() < 0)
		RTE_LOG (ERR, EAL, "Cannot init memory\n");
		goto err;
	}

	[...]

	return 0;

err:
	/* TODO: undo all initialization work */
         return -1;
}

	- Panu -