Re: Re: [PATCH v10 4/5] crypto: Add Allwinner Security System crypto accelerator

[LABBE Corentin <clabbe.montjoie-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]

On Tue, Jul 07, 2015 at 09:49:36PM +0800, Herbert Xu wrote:
> On Mon, Jul 06, 2015 at 09:10:47PM +0200, LABBE Corentin wrote:
> >
> > +int sun4i_hash_init(struct ahash_request *areq)
> > +{
> > +	const char *hash_type;
> > +	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
> > +	struct crypto_ahash *tfm = crypto_ahash_reqtfm(areq);
> > +	struct ahash_alg *alg = __crypto_ahash_alg(tfm->base.__crt_alg);
> > +	struct sun4i_ss_alg_template *algt;
> > +	struct sun4i_ss_ctx *ss;
> > +
> > +	memset(op, 0, sizeof(struct sun4i_req_ctx));
> > +
> > +	algt = container_of(alg, struct sun4i_ss_alg_template, alg.hash);
> > +	ss = algt->ss;
> > +	op->ss = algt->ss;
> > +
> > +	hash_type = crypto_tfm_alg_name(areq->base.tfm);
> > +
> > +	if (strcmp(hash_type, "sha1") == 0)
> > +		op->mode = SS_OP_SHA1;
> > +	else if (strcmp(hash_type, "md5") == 0)
> > +		op->mode = SS_OP_MD5;
> 
> Please store these modes in algt and then you wouldn't need to do
> the strcmp.
> 

thanks, that's better

> > +int sun4i_hash_import_md5(struct ahash_request *areq, const void *in)
> > +{
> > +	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
> > +	const struct md5_state *ictx = in;
> > +	int i;
> 
> You need to initialise op->ss here.  Ditto for SHA.
> 

Ok

> > +	/* get the partial hash only if something was written */
> > +	if (op->mode == SS_OP_SHA1) {
> > +		for (i = 0; i < 5; i++)
> > +			op->hash[i] = readl(ss->base + SS_MD0 + i * 4);
> > +	} else {
> > +		for (i = 0; i < 4; i++)
> > +			op->hash[i] = readl(ss->base + SS_MD0 + i * 4);
> > +	}
> 
> You can avoid the op->mode check by storing the hash size in algt.
> 

It seems that I can use crypto_ahash_digestsize() that do the same thing.

> > +int sun4i_hash_finup(struct ahash_request *areq)
> > +{
> > +	int err;
> > +
> > +	err = sun4i_hash_update(areq);
> > +	if (err != 0)
> > +		return err;
> > +
> > +	return sun4i_hash_final(areq);
> 
> You can drop finup since the API provides a default finup that's
> exactly the same as this.  You only need to provide finup if it
> can optimise on this.
> 
> Ditto for digest.

If I remove digest(), loading the driver crash the kernel with:

Unable to handle kernel NULL pointer dereference at virtual address 00000000
pgd = c0004000
[00000000] *pgd=00000000
Internal error: Oops: 80000007 [#1] PREEMPT SMP ARM
CPU: 0 PID: 10120 Comm: cryptomgr_test Tainted: G            E   4.2.0-rc1+ #78
LR is at crypto_ahash_op+0x30/0x68

For using the default one does I need to do something particular ?

Thanks

LABBE Corentin