Re: [PATCH v5 00/13] riscv: improve boot time isa extensions handling

From: Guenter Roeck <linux@roeck-us.net>
To: Conor Dooley <conor@kernel.org>
Cc: Jisheng Zhang <jszhang@kernel.org>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	Paul Walmsley <paul.walmsley@sifive.com>,
	Albert Ou <aou@eecs.berkeley.edu>,
	Anup Patel <anup@brainfault.org>,
	Atish Patra <atishp@atishpatra.org>,
	Heiko Stuebner <heiko@sntech.de>,
	Andrew Jones <ajones@ventanamicro.com>,
	Conor Dooley <conor.dooley@microchip.com>,
	linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
	kvm@vger.kernel.org, kvm-riscv@lists.infradead.org
Subject: Re: [PATCH v5 00/13] riscv: improve boot time isa extensions handling
Date: Sun, 12 Feb 2023 10:38:26 -0800	[thread overview]
Message-ID: <48d2ac96-abdd-23bc-b167-08dc2c1a1dbe@roeck-us.net> (raw)
In-Reply-To: <Y+kt04c1iRlzUNLA@spud>

On 2/12/23 10:20, Conor Dooley wrote:
> On Sun, Feb 12, 2023 at 10:14:13AM -0800, Guenter Roeck wrote:
>> On 2/12/23 10:06, Conor Dooley wrote:
>>> On Sun, Feb 12, 2023 at 05:06:09PM +0000, Conor Dooley wrote:
>>>> On Sun, Feb 12, 2023 at 04:33:58PM +0000, Conor Dooley wrote:
>>>>> On Sun, Feb 12, 2023 at 03:59:59PM +0000, Conor Dooley wrote:
>>>>
>>>> So as not to lead anyone up the garden path, let me correct myself:
>>>>
>>>>> Hmm, so this appears to be us attempting to patch in alternatives where
>>>>> none actually exists - seemingly F & D.
>>>>
>>>> And of course that's not true, riscv_has_extension_likely() now uses
>>>> alternatives as of:
>>>> bdda5d554e43 ("riscv: introduce riscv_has_extension_[un]likely()")
>>>>
>>>>   From a quick look, it just happens that the only users are F & D.
>>>>
>>>
>>> Samuel pointed out that this is a lockdep splat on irc.
>>> There's a patch on the list that removes the lockdep annotation
>>> entirely:
>>> https://patchwork.kernel.org/project/linux-riscv/patch/20230202114116.3695793-1-changbin.du@huawei.com/
>>>
>>> So ye, no surprises that it was config based!
>>>
>>> Palmer posted a "better" fix for that lockdep warning a while ago:
>>> https://lore.kernel.org/all/20220322022331.32136-1-palmer@rivosinc.com/
>>>
>>> So we'd have to duplicate/reuse that for cpufeature/errata patching.
>>>
>>>
>>
>> This does not (only) happen in stop_machine().
> 
> Yah, sorry I meant that it's the same lockdep splat as is being
> addressed there.
> The first patch deletes the lockdep stuff entirely, so removes the
> splat. I was thinking that we'd need to take Palmer's (IMO better)
> patch and do the same thing for patching alternatives, but I figure we
> can just take the text_mutex itself for alternatives & not have to
> dance around the lock.
> 
> I'll go do that I suppose!

Thanks a lot for the clarification. That sounds like the backtrace
can be largely ignored. However, I still see that the patch series
results in boot hangs with the sifive_u qemu emulation, where
the log ends with "Oops - illegal instruction". Is that problem
being addressed as well ?

Thanks,
Guenter