On Sat, Jan 04, 2025 at 08:45:10AM +0800, Herbert Xu wrote:
> On Thu, Dec 26, 2024 at 07:08:03PM +0100, Lukas Wunner wrote:
> >
> > diff --git a/crypto/ecdsa-p1363.c b/crypto/ecdsa-p1363.c
> > index eaae7214d69b..c4f458df18ed 100644
> > --- a/crypto/ecdsa-p1363.c
> > +++ b/crypto/ecdsa-p1363.c
> > @@ -21,7 +21,7 @@ static int ecdsa_p1363_verify(struct crypto_sig *tfm,
> > const void *digest, unsigned int dlen)
> > {
> > struct ecdsa_p1363_ctx *ctx = crypto_sig_ctx(tfm);
> > - unsigned int keylen = crypto_sig_keysize(ctx->child);
> > + unsigned int keylen = DIV_ROUND_UP(crypto_sig_keysize(ctx->child), 8);
>
> This may overflow unnecessarily, please rewrite these as:
>
> X / 8 + !!(X & 7)
Interesting. Wouldn't it make sense to have a DIV_ROUND_UP_SAFE() macro
for cases like this?
I'd expect this version to actually be faster than DIV_ROUND_UP():
There's the extra logical AND, as well as the negation.
But the "n / d" and "!!(n & (d - 1))" can be computed in parallel.
Thanks,
Lukas
