Re: [dpdk-dev] [PATCH 1/4] lib/crypto: add support for ECDSA

["Kusztal, ArkadiuszX" <arkadiuszx.kusztal@intel.com>]

Hi guys,

> >
> > > >
> > > > > > +
> > > > > > +	rte_crypto_param k;
> > > > > > +	/**< The ECDSA per-message secret number, which is an
> > integer
> > > > > > +	 * in the interval (1, n-1)
> > > > > > +	 */
> > > > > [Arek] - If pmd can generate 'k' internally we could do something like:
> > > > > 'if k.data == NULL => PMD will generate 'k' internally, k.data
> > > > > remains untouched.'
> > > >
> > > > [Anoob] So that will be exposed as a new capability, right? Do you
> > > > suggest any changes here to accommodate that?
> > > [Arek] Or maybe feature flag, it would apply to DSA as well.
> >
> > [Anoob] I meant feature flag only! Capability is for net devs etc.
> >
> > > >
> > > > > Another option is to provide user with some callback function to
> > > > > generate CSRN which could be useful for RSA PSS, OAEP as well
> > > > > (we already discussed that internally in Intel, I will elaborate
> > > > > on this bit more in
> > > > different thread).
> > > >
> > > > [Anoob] Do you intend to keep the generated CSRN hidden in the PMD?
> > > [Arek] Openssl PMD does that with DSA but iam not a fan of that. But
> > > if some hw can do DSA/ECDSA by generating 'k' by itself it would
> > > have to be
> > added anyway.
> > > Other option is to add aforementioned callbacks (for HW that not
> > > generate
> > 'k'
> > > by itself)
> >
> > [Anoob] I think we can support either case. Our h/w also can generate
> > CSRN and so are open to ideas. Can you explain the proposed callback
> function?
> > Also, we can defer that discussion to a separate thread, if we have an
> > agreement on the approach here.
> >
> [Shally] Just a thought here. It can also be supported by exposing an API to
> application, like pre-compute / pre-setup which can generate 'k' or 'PRF' for
> other algo usage purpose. If PMD support it, return will be success with
> relevant o/p.. if not, return will be failure. Or no offload at all. In any case,
> generation of prf or 'k' is heavy on data path... app can call these pre-setup
> APIs to save cycles on actual sign op.
[AK] - We can start another thread on that. For now this implementation is good too, we can extend API by PMD CSRN option later.
> 
> Shally
> > > >
> > > > >
> > > > > > +
> > > > > > +	rte_crypto_param r;
> > > > > > +	/**< r component of elliptic curve signature
> > > > > > +	 *     output : for signature generation
> > > > > > +	 *     input  : for signature verification
> > > > > > +	 */
> > > > > > +	rte_crypto_param s;
> > > > > > +	/**< s component of elliptic curve signature
> > > > > > +	 *     output : for signature generation
> > > > > > +	 *     input  : for signature verification
> > > > > > +	 */
> > > > > [Arek] - Do we want to add any constraints like 'this field
> > > > > should be big enough to hold...'
> > > >
> > > > [Anoob] For every case where rte_crypto_param is used for
> > > > 'output', application should make sure the buffers are large
> > > > enough. Do you think we could document it somewhere common
> instead
> > > > of adding per
> > > operation?
> > > [Arek]
> > > Both options look good to me.
> >
> > [Anoob] How about updating the description of 'rte_crypto_param' to
> > reflect the same? I can update it in v2, if you can confirm.
[AK] It is ok to me.
> >
> > > >
> > > > > > +};
> > > > > > +
> > > > > > +/**
> > > > > >   * Asymmetric Cryptographic Operation.
> > > > > >   *
> > > > > >   * Structure describing asymmetric crypto operation params.
> > > > > > @@ -537,6 +619,7 @@ struct rte_crypto_asym_op {
> > > > > >  		struct rte_crypto_mod_op_param modinv;
> > > > > >  		struct rte_crypto_dh_op_param dh;
> > > > > >  		struct rte_crypto_dsa_op_param dsa;
> > > > > > +		struct rte_crypto_ecdsa_op_param ecdsa;
> > > > > >  	};
> > > > > >  };
> > > > > >
> > > > > > diff --git a/lib/librte_cryptodev/rte_cryptodev.c
> > > > > > b/lib/librte_cryptodev/rte_cryptodev.c
> > > > > > index 89aa2ed..0d6babb 100644
> > > > > > --- a/lib/librte_cryptodev/rte_cryptodev.c
> > > > > > +++ b/lib/librte_cryptodev/rte_cryptodev.c
> > > > > > @@ -173,6 +173,7 @@ const char
> > > > > > *rte_crypto_asym_xform_strings[]
> > = {
> > > > > >  	[RTE_CRYPTO_ASYM_XFORM_MODINV]	= "modinv",
> > > > > >  	[RTE_CRYPTO_ASYM_XFORM_DH]	= "dh",
> > > > > >  	[RTE_CRYPTO_ASYM_XFORM_DSA]	= "dsa",
> > > > > > +	[RTE_CRYPTO_ASYM_XFORM_ECDSA]	= "ecdsa",
> > > > > >  };
> > > > > >
> > > > > >  /**
> > > > > > --
> > > > > > 2.7.4
> > > > >
> > > > > Regards,
> > > > > Arek