Re: [GIT pull] locking/urgent for v5.12-rc3

From: Peter Zijlstra <peterz@infradead.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	the arch/x86 maintainers <x86@kernel.org>,
	Steven Rostedt <rostedt@goodmis.org>
Subject: Re: [GIT pull] locking/urgent for v5.12-rc3
Date: Mon, 15 Mar 2021 23:09:15 +0100	[thread overview]
Message-ID: <YE/bCyJM/xGsqsd3@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <CAHk-=wjE1Z2L87B_ahMMEd9E+jumMwsw=HDCgDkV4gmS=F9u6A@mail.gmail.com>

On Mon, Mar 15, 2021 at 11:59:12AM -0700, Linus Torvalds wrote:
> Is it only the static_call_sites entry itself that needs the
> alignment? Or do we end up depending on the static call function being
> at least 4-byte aligned too? The way it plays games with the key makes
> me worry.

The only thing that absolutely needs to be aligned is the
struct static_call_key address. We use the 2 LSB there.

The code address has no alignment requirements, due to x86 instruction
coding the actual CALL (or JMP for tail-calls) can be anywhere.

Now, static_call_site is PC32 encoded, that is:

	struct static_call_key *key =
		(void *)((unsigned long)&site->key + site->key);

And assuming &site->key is aligned, then site->key & 3 == key & 3.

Per the missing alignment for modules, the above went side-ways. The
patch in question fixed this by not relying on that and always computing
the absolute address first, then transfer the LSBs and then re-encoding
it.

Anyway, still good to also fix the alignment.