Re: [dpdk-dev] [PATCH] test/ipsec: fix test suite setup function

["Ananyev, Konstantin" <konstantin.ananyev@intel.com>]

Hi Bernard,

> > > > > Subject: [PATCH] test/ipsec: fix test suite setup function
> > > > >
> > > > > Check for valid crypto_null devices before continuing.
> > > > >
> > > > > Fixes: 05fe65eb66b2 ("test/ipsec: introduce functional test")
> > > > > Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com>
> > > > > ---
> > > > >  test/test/test_ipsec.c | 17 +++++++++++++++--
> > > > >  1 file changed, 15 insertions(+), 2 deletions(-)
> > > > >
> > > > > diff --git a/test/test/test_ipsec.c b/test/test/test_ipsec.c index
> > > > > ff1a1c4..4dfc55b 100644
> > > > > --- a/test/test/test_ipsec.c
> > > > > +++ b/test/test/test_ipsec.c
> > > > > @@ -46,6 +46,8 @@
> > > > >  #define BURST_SIZE		32
> > > > >  #define REORDER_PKTS	1
> > > > >
> > > > > +static int gbl_driver_id;
> > > > > +
> > > >
> > > > Why do you need that global here?
> > >
> > > test_ipsec.c is based on test_cryptodev.c.
> > > gbl_driver_id used to store the ID of the required driver.
> >
> > Sorry but referencing someone else code is not an answer.
> > Why do *you* need it *here*?
> 
> The global is not needed.
> I have renamed it to driver_id and added it as a local variable where it is used.
> 
> > > > >  struct user_params {
> > > > >  	enum rte_crypto_sym_xform_type auth;
> > > > >  	enum rte_crypto_sym_xform_type cipher; @@ -218,7 +220,7 @@
> > > > > testsuite_setup(void)  {
> > > > >  	struct ipsec_testsuite_params *ts_params = &testsuite_params;
> > > > >  	struct rte_cryptodev_info info;
> > > > > -	uint32_t nb_devs, dev_id;
> > > > > +	uint32_t i, nb_devs, dev_id;
> > > > >  	size_t sess_sz;
> > > > >
> > > > >  	memset(ts_params, 0, sizeof(*ts_params)); @@ -251,7 +253,18 @@
> > > > > testsuite_setup(void)
> > > > >  		return TEST_FAILED;
> > > > >  	}
> > > > >
> > > > > -	ts_params->valid_devs[ts_params->valid_dev_count++] = 0;
> > > > > +	gbl_driver_id = rte_cryptodev_driver_id_get(
> > > > > +				RTE_STR(CRYPTODEV_NAME_NULL_PMD));
> > >
> > > These tests only work with the crypto_null  PMD's,  gbl_driver_id is set to the
> > crypto_null PMD id here.
> > >
> > > > > +
> > > > > +	/* Create list of valid crypto devs */
> > > > > +	for (i = 0; i < nb_devs; i++) {
> > > > > +		rte_cryptodev_info_get(i, &info);
> > > > > +		if (info.driver_id == gbl_driver_id)
> > > > > +			ts_params->valid_devs[ts_params->valid_dev_count++]
> > > > = i;
> > > > > +	}
> > > >
> > > > I think you need to check driver capabilities, instead of relying on driver
> > name.
> > >
> > > I don't think it is necessary to check the driver capabilities.
> >
> > I still think that the valid way to check supported algorithms is to check device
> > capabilities, not the driver name.
> 
> In the testsuite_setup() function the parameters for the check_cryptodev_capability() are not setup. They are setup in the test functions of
> the testsuite.

Ok, so what prevents us to setup them earlier?

> 
> > > This is how it is done in test_cryptodev.c.
> > >  I think it makes sense to mirror the test_cryptodev.c implementation.
> > >
> > > > > +
> > > > > +	if (ts_params->valid_dev_count < 1)
> > > > > +		return TEST_FAILED;
> > > > >
> > > > >  	/* Set up all the qps on the first of the valid devices found */
> > > > >  	dev_id = ts_params->valid_devs[0];
> > > >
> > > > If we always use just valid_dev[0] to determine private session
> > > > size, why do you keep going though all devs in the loop above?
> > >
> > > There may be several crypto devs present for example, crypto_aesni_mb0,
> > crypto_aseni_mb1, crypto_null0 and  crypto_null1.
> >
> > Yes.
> >
> > > The valid_dev[] array will contain all devs of the requested type, in this case
> > crypto_null0 and crypto_null1.
> >
> > But we need/use only one.
> 
> I will change the code to replace the valid_devs[] with one valid_dev.
> 
> > > > Another thing, as I mentioned off-line - later you still use all
> > > > vald_devs[] to init
> > > > session:
> > > > s = rte_cryptodev_sym_session_create(qp->mp_session);
> > > >         if (s == NULL)
> > > >                 return -ENOMEM;
> > > >
> > > >         /* initiliaze SA crypto session for all supported devices */
> > > >         for (i = 0; i != devnum; i++) {
> > > >                 rc = rte_cryptodev_sym_session_init(devid[i], s,
> > > >                         ut->crypto_xforms, qp->mp_session_private);
> > > >                 if (rc != 0)
> > > >                         break;
> > > >         }
> > > >
> > > > I think we need either to determine max private session size based
> > > > on *all* valid_devs[], or just use one device that can do NULL algorithm.
> > >
> > > The valid_devs[] array only contains crypto_null PMD's The code is
> > > using the crypto_null PMD only.
> >
> > In fact there is no reason to be crypto_null only.
> > I think it could be any crypto-dev that does support NULL auth/cipher.
> 
> As discussed offline it should be sufficient to test with the crypto_dev NULL PMD.

As we discussed  offline - yes, I don't think it's too excessive to verify ipsec_autotest
with each existing driver that supports _NULL algs, but I don't see the reason why
it shouldn't support anything except crypto_null.
Konstantin 

> 
> > > > As we always enqueue/dequeuer into valid_devs[0] - I think there is
> > > > no point to have an arrays here, just single valid_dev should be sufficient.
> > >
> > > The test program may be started with several crypto_dev PMD's for example:
> > >
> > > test -c f -n 4 --vdev crypto_aesni_mb0 --vdev crypto_null0 --vdev
> > > crypto_aesni_mb1 --vdev crypto_dev_null1
> > >
> > > In this case the valid_devs[] array will contain crypto_dev_null0 and
> > crypto_dev_null1.
> 
> I will replace the valid_devs[] with valid_dev which contains the first crypto_null device found.
> 
> I will send a v2 patch
> 
> Regards,
> 
> Bernard.