[PATCH v2 2/5] random: defer seeding to EAL init

[Mattias Rönnblom]

On 2024-12-18 17:35, Stephen Hemminger wrote:
> On Tue, 17 Dec 2024 09:59:49 +0100
> David Marchand <david.marchand at redhat.com> wrote:
> 
>> The RNG is documented as being seeded as part of EAL init.
>>
>> Move the initialisation (seeding) helper out of a constructor and
>> call it explicitly from rte_eal_init() as it was done before commit
>> 3f002f069612 ("eal: replace libc-based random generation with LFSR").
>>
>> This also moves the unconditional lcore variable allocation out of a
>> constructor.
>>
>> While at it, mark local symbol rand_state as static.
>>
>> Fixes: 29c39cd3d54d ("random: keep PRNG state in lcore variable")
>> Cc: stable at dpdk.org
>>
>> Signed-off-by: David Marchand <david.marchand at redhat.com>
>> Reviewed-by: Mattias Rönnblom <mattias.ronnblom at ericsson.com>
>> Acked-by: Anatoly Burakov <anatoly.burakov at intel.com>
> 
> Probably need to add a check to rte_random() so it crashes
> if called before initialization, rather than returning an un-random
> number which could be a hidden long term bug.

If you do, do it either as a RTE_ASSERT() or an RTE_VERIFY() in the 
(lcore_id == LCORE_ID_ANY) path, since that is what will be taken.

Preferably, you should have as little as possible in rte_rand() fast 
path, because this function is used in packet processing.

That said, the "unrandom" number will always be 0, so it shouldn't go 
unnoticed for too long.