Re: [PATCH 0/2] wireguard-linux-compat: grsecurity compat patches

From: Mathias Krause <minipli@grsecurity.net>
To: "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc: WireGuard mailing list <wireguard@lists.zx2c4.com>
Subject: Re: [PATCH 0/2] wireguard-linux-compat: grsecurity compat patches
Date: Mon, 6 Dec 2021 22:00:16 +0100	[thread overview]
Message-ID: <3ed9270f-00da-5c88-cadd-59b4419b9a98@grsecurity.net> (raw)
In-Reply-To: <CAHmME9odv1V9wUrLsX16J68wVD3rq+op0VvaNxtsLGG8Rrz4tQ@mail.gmail.com>

Am 06.12.21 um 19:55 schrieb Jason A. Donenfeld:
> Nice detective work! I just loaded this up on the CI, so we'll see if
> this does work across the board.

Thanks.

> It sounds like the original code also had this bug -- the "r"(out)
> should be in the output constraints, not in the input constraints,
> right? Sounds like something I should report to the EverCrypt authors
> and also fix upstream too then.

Yes, probably, but you're mixing up the two. "r"(out) should be an input
operand as it's only read in the inline asm. The other two ('tmp' and
'f') are read and written, so need to be output operands (as they
already are!) but also marked as earlyclobber ("&") as they're modified
before all input operands are read ('out', in this case). It's just a
hint to the compiler to not make the register allocation overlap with
any (other) input operand register.

Thanks,
Mathias