Re: [PATCH 1/2] riscv: vdso: Do not use LTO for the vDSO

From: Nathan Chancellor <nathan@kernel.org>
To: "Thomas Weißschuh" <thomas.weissschuh@linutronix.de>
Cc: Paul Walmsley <pjw@kernel.org>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	Albert Ou <aou@eecs.berkeley.edu>,
	Alexandre Ghiti <alex@ghiti.fr>,
	Conor Dooley <conor.dooley@microchip.com>,
	Wende Tan <twd2.me@gmail.com>,
	Palmer Dabbelt <palmer@rivosinc.com>,
	linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
	Nam Cao <namcao@linutronix.de>, kernel test robot <lkp@intel.com>,
	stable@vger.kernel.org
Subject: Re: [PATCH 1/2] riscv: vdso: Do not use LTO for the vDSO
Date: Tue, 7 Jul 2026 10:44:06 -0700	[thread overview]
Message-ID: <20260707174406.GA991448@ax162> (raw)
In-Reply-To: <20260707080753-4e88aca1-b88d-4f6c-b37a-f7f3064bda5e@linutronix.de>

On Tue, Jul 07, 2026 at 08:10:19AM +0200, Thomas Weißschuh wrote:
> On Mon, Jul 06, 2026 at 02:01:58PM -0700, Nathan Chancellor wrote:
> > On Wed, Jul 01, 2026 at 11:21:22AM +0200, Thomas Weißschuh wrote:
> > > With LTO enabled the compiler assumes that the vDSO functions are not
> > > used and optimizes them away completely. Currently this happens to
> > > __vdso_clock_getres(), __vdso_clock_gettime(), __vdso_getrandom(),
> > > __vdso_gettimeofday() and __vdso_riscv_hwprobe().
> > > 
> > > Disable LTO for the vDSO, as these functions are hand-optimized anyways.
> > > 
> > > Reported-by: kernel test robot <lkp@intel.com>
> > > Closes: https://lore.kernel.org/oe-kbuild-all/202606301855.WvkSC4kD-lkp@intel.com/
> > 
> > While this change seems correct, is this really the fix for that report?
> > It seems like that error happens in clang but I would expect this sort
> > of issue to only appear once LTO has run through ld.lld?
> 
> At this point the vDSO userspace library has already run through ld.lld.
> That has optimized away the futex symbols, which means their offsets are not
> defined when building the regular vDSO *kernel* code.

Ahhh, thanks for the additional clarification! That makes much more
sense, I had missed that we were in kernel code, not userspace.

-- 
Cheers,
Nathan