Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Daniel Axtens <dja@axtens.net>
To:     keescook@chromium.org
Cc:     catalin.marinas@arm.com, clang-built-linux@googlegroups.com,
        hca@linux.ibm.com, jarmo.tiitto@gmail.com,
        linux-kernel@vger.kernel.org, lukas.bulwahn@gmail.com,
        mark.rutland@arm.com, masahiroy@kernel.org, maskray@google.com,
        morbo@google.com, nathan@kernel.org, ndesaulniers@google.com,
        oberpar@linux.ibm.com, ojeda@kernel.org, peterz@infradead.org,
        samitolvanen@google.com, torvalds@linux-foundation.org,
        wcw@google.com, will@kernel.org, dja@axtens.net
Subject: ARCH_WANTS_NO_INSTR (Re: [GIT PULL] Clang feature updates for v5.14-rc1)
Date:   Wed,  6 Oct 2021 00:10:15 +1100
Message-Id: <20211005131015.3153458-1-dja@axtens.net>
In-Reply-To: <202106281231.E99B92BB13@keescook>
References: <202106281231.E99B92BB13@keescook>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Hi,

Apologies, I can't find the original email for this:

>      Kconfig: Introduce ARCH_WANTS_NO_INSTR and CC_HAS_NO_PROFILE_FN_ATTR

which is now commit 51c2ee6d121c ("Kconfig: Introduce ARCH_WANTS_NO_INSTR and
CC_HAS_NO_PROFILE_FN_ATTR"). It doesn't seem to show up on Google, this was the
best I could find.

Anyway, the commit message reads:

    Kconfig: Introduce ARCH_WANTS_NO_INSTR and CC_HAS_NO_PROFILE_FN_ATTR
    
    We don't want compiler instrumentation to touch noinstr functions,
    which are annotated with the no_profile_instrument_function function
    attribute. Add a Kconfig test for this and make GCOV depend on it, and
    in the future, PGO.
    
    If an architecture is using noinstr, it should denote that via this
    Kconfig value. That makes Kconfigs that depend on noinstr able to express
    dependencies in an architecturally agnostic way.

However, things in generic code (such as rcu_nmi_enter) are tagged with
`noinstr`, so I'm worried that this commit subtly breaks things like KASAN on
platforms that haven't opted in yet. (I stumbled across this while developing
KASAN on ppc64, but at least riscv and ppc32 have KASAN but not
ARCH_WANTS_NO_INSTR already.)

As I said, I haven't been able to find the original thread - is there any reason
this shouldn't be always on? Why would an arch not opt in? What's the motivation
for ignoring the noinstr markings?

Should generic KASAN/KCSAN/anything else marked in noinstr also have markings
requring ARCH_WANTS_NO_INSTR? AFAICT they should, right?

Kind regards,
Daniel