Re: [PATCH RFC] riscv: Allow to build only with LLVM >= 17.0.0

From: Conor Dooley <conor@kernel.org>
To: Alexandre Ghiti <alex@ghiti.fr>
Cc: "Alexandre Ghiti" <alexghiti@rivosinc.com>,
	"Paul Walmsley" <paul.walmsley@sifive.com>,
	"Palmer Dabbelt" <palmer@dabbelt.com>,
	"Albert Ou" <aou@eecs.berkeley.edu>,
	"Björn Töpel" <bjorn@rivosinc.com>,
	"Conor Dooley" <conor.dooley@microchip.com>,
	"Nathan Chancellor" <nathan@kernel.org>,
	linux-kernel@vger.kernel.org, linux-riscv@lists.infradead.org,
	"kernel test robot" <lkp@intel.com>
Subject: Re: [PATCH RFC] riscv: Allow to build only with LLVM >= 17.0.0
Date: Wed, 17 Jul 2024 14:06:42 +0100	[thread overview]
Message-ID: <20240717-commuting-channel-4ad202950e4c@spud> (raw)
In-Reply-To: <20240717-theft-resample-19c248bb2a26@spud>

[-- Attachment #1.1: Type: text/plain, Size: 1917 bytes --]

On Wed, Jul 17, 2024 at 01:06:39PM +0100, Conor Dooley wrote:
> On Wed, Jul 17, 2024 at 01:41:23PM +0200, Alexandre Ghiti wrote:
> > Hi Conor,
> > 
> > On 17/07/2024 13:32, Conor Dooley wrote:
> > > On Wed, Jul 17, 2024 at 01:17:16PM +0200, Alexandre Ghiti wrote:
> > > > The following build failure happens when using LLVM < 17.0.0:
> > > > 
> > > > kernel/sched/core.c:11873:7: error: cannot jump from this asm goto statement to one of its possible targets
> > > > 
> > > > This is a known issue [1] so let's upgrade the minimal requirement for
> > > > LLVM to the version 17.0.0, which is the first version to contain the
> > > > fix.
> > > I think doing this unilaterally is kinda insane, LLVM 17 isn't even a
> > > year old. Debian testing doesn't have anything later than 16.
> > 
> > 
> > Debian will very likely select the qspinlocks when available anyway, so
> > they'll need llvm >= 17. And Debian won't ship a kernel >= 6.11 until some
> > time right? So they'll probably update their infra to llvm >= 17 (and
> > they'll probably do to take advantages of the new extensions).
> 
> What I mean is that you are going to prevent people building the kernel
> with llvm on machines running anything but very recent rolling-release
> distros. Your patch would stop most developers, including those who don't
> care about your qspinlock stuff, even build testing with the version of
> LLVM that their distro provides. I'm not talking about distros building
> kernels in their build infrastructure.
> 
> > 
> > 
> > > Why does
> > > it need to be done unilaterally rather than just when the qspinlock
> > > stuff is built?
> > 
> > 
> > We can do that indeed, it may happen again and we can keep requiring llvm 17
> > on a per-config basis.

Nathan pointed out to me that I misunderstood the build failure, and
that it happens whether or not the option is enabled. /sigh.

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

[-- Attachment #2: Type: text/plain, Size: 161 bytes --]

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv