Date:   Mon, 10 Jan 2022 11:28:01 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Ingo Molnar <mingo@kernel.org>
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org,
        Andrew Morton <akpm@linux-foundation.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        "David S. Miller" <davem@davemloft.net>,
        Ard Biesheuvel <ardb@kernel.org>,
        Josh Poimboeuf <jpoimboe@redhat.com>,
        Jonathan Corbet <corbet@lwn.net>,
        Al Viro <viro@zeniv.linux.org.uk>
Subject: Re: [PATCH 0000/2297] [ANNOUNCE, RFC] "Fast Kernel Headers" Tree
 -v1: Eliminate the Linux kernel's "Dependency Hell"
Message-ID: <YdwKMfNkB7P1tm/m@hirez.programming.kicks-ass.net>
References: <YdIfz+LMewetSaEB@gmail.com>
 <YdLL0kaFhm6rp9NS@kroah.com>
 <YdLaMvaM9vq4W6f1@gmail.com>
 <YdL+IwQGTLFQyVz2@kroah.com>
 <YdMkTjGSQFLEV5VB@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YdMkTjGSQFLEV5VB@gmail.com>
Precedence: bulk

On Mon, Jan 03, 2022 at 05:29:02PM +0100, Ingo Molnar wrote:
> Yeah, so I *did* find this somewhat suboptimal too, and developed an 
> earlier version that used linker section tricks to gain the field offsets 
> more automatically.
> 
> It was an unmitigated disaster: was fragile on x86 already (which has a zoo 
> of linking quirks with no precedent of doing this before bounds.c 
> processing), but on ARM64 and probably on most of the other RISC-ish 
> architectures there was also a real runtime code generation cost of using 
> linker tricks: 2-3 extra instructions per per_task() use - clearly 
> unacceptable.
> 
> Found this out the hard way after making it boot & work on ARM64 and 
> looking at the assembly output, trying to figure out why the generated code 
> size increased. :-/

Right, I suggested you do the per-cpu thing. And then Mark reported that
code-gen issue on arm64.

I'm still thinking the toolchains ought to look at fixing that. It'll be
too late to use for per-task, but at least the current per-cpu usages
will (eventually) get better code-gen.