Re: [PATCH net-next] net: mvpp2: Initialize link in mvpp2_isr_handle_{xlg,gmac_internal}

From: Nathan Chancellor <natechancellor@gmail.com>
To: Jakub Kicinski <kuba@kernel.org>
Cc: Russell King - ARM Linux admin <linux@armlinux.org.uk>,
	David Miller <davem@davemloft.net>,
	andrew@lunn.ch, netdev@vger.kernel.org,
	linux-kernel@vger.kernel.org, clang-built-linux@googlegroups.com
Subject: Re: [PATCH net-next] net: mvpp2: Initialize link in mvpp2_isr_handle_{xlg,gmac_internal}
Date: Fri, 11 Sep 2020 09:01:01 -0700	[thread overview]
Message-ID: <20200911160101.GA4061896@ubuntu-n2-xlarge-x86> (raw)
In-Reply-To: <20200911082236.7dfb7937@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>

On Fri, Sep 11, 2020 at 08:22:36AM -0700, Jakub Kicinski wrote:
> On Fri, 11 Sep 2020 12:11:58 +0100 Russell King - ARM Linux admin wrote:
> > On Thu, Sep 10, 2020 at 05:31:42PM -0700, Nathan Chancellor wrote:
> > > Ah great, that is indeed cleaner, thank you for letting me know!  
> > 
> > Hmm, I'm not sure why gcc didn't find that. Strangely, the 0-day bot
> > seems to have only picked up on it with clang, not gcc.
> 
> May be similar to: https://lkml.org/lkml/2019/2/25/1092
> 
> Recent GCC is so bad at catching uninitialized vars I was considering
> build testing with GCC8 :/

It is even simpler than that, the warning was straight up disabled in
commit 78a5255ffb6a ("Stop the ad-hoc games with -Wno-maybe-initialized").

clang's -Wuninitialized and -Wsometimes-uninitialized are generally more
accurate but can have some false positives because the semantic analysis
phase happens before inlining and dead code elimination.

Cheers,
Nathan