Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Miquel Raynal <miquel.raynal@bootlin.com>
To:     Arnd Bergmann <arnd@kernel.org>,
        Tudor Ambarus <tudor.ambarus@linaro.org>,
        Pratyush Yadav <pratyush@kernel.org>,
        Miquel Raynal <miquel.raynal@bootlin.com>,
        Richard Weinberger <richard@nod.at>,
        Vignesh Raghavendra <vigneshr@ti.com>
Cc:     Arnd Bergmann <arnd@arndb.de>, Peter Foley <pefoley2@pefoley.com>,
        Pedro Falcato <pedro.falcato@gmail.com>,
        Michael Walle <michael@walle.cc>,
        Mark Brown <broonie@kernel.org>,
        Takahiro Kuwano <Takahiro.Kuwano@infineon.com>,
        Dhruva Gole <d-gole@ti.com>, linux-mtd@lists.infradead.org,
        linux-kernel@vger.kernel.org, linux-spi@vger.kernel.org
Subject: Re: [PATCH] mtd: spi-nor: avoid holes in struct spi_mem_op
Date:   Thu, 27 Jul 2023 17:17:38 +0200
Message-Id: <20230727151738.586338-1-miquel.raynal@bootlin.com>
In-Reply-To: <20230719190045.4007391-1-arnd@kernel.org>
References: 
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

On Wed, 2023-07-19 at 19:00:25 UTC, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> gcc gets confused when -ftrivial-auto-var-init=pattern is used on sparse
> bit fields such as 'struct spi_mem_op', which caused the previous false
> positive warning about an uninitialized variable:
> 
> drivers/mtd/spi-nor/spansion.c: error: 'op' is used uninitialized [-Werror=uninitialized]
> 
> In fact, the variable is fully initialized and gcc does not see it being
> used, so the warning is entirely bogus. The problem appears to be
> a misoptimization in the initialization of single bit fields when the
> rest of the bytes are not initialized.
> 
> A previous workaround added another initialization, which ended up
> shutting up the warning in spansion.c, though it apparently still happens
> in other files as reported by Peter Foley in the gcc bugzilla. The
> workaround of adding a fake initialization seems particularly bad
> because it would set values that can never be correct but prevent the
> compiler from warning about actually missing initializations.
> 
> Revert the broken workaround and instead pad the structure to only
> have bitfields that add up to full bytes, which should avoid this
> behavior in all drivers.
> 
> I also filed a new bug against gcc with what I found, so this can
> hopefully be addressed in future gcc releases. At the moment, only
> gcc-12 and gcc-13 are affected.
> 
> Cc: Peter Foley <pefoley2@pefoley.com>
> Cc: Pedro Falcato <pedro.falcato@gmail.com>
> Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110743
> Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108402
> Link: https://godbolt.org/z/efMMsG1Kx
> Fixes: 420c4495b5e56 ("mtd: spi-nor: spansion: make sure local struct does not contain garbage")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> Acked-by: Mark Brown <broonie@kernel.org>
> Acked-by: Tudor Ambarus <tudor.ambarus@linaro.org>

Applied to https://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux.git mtd/fixes, thanks.

Miquel