[PATCH v2] ARM: initialize jump labels before setup_machine_fdt()

[Phil Elwell]

On 07/06/2022 16:44, Jason A. Donenfeld wrote:
> Hi Phil,
> 
> On Tue, Jun 07, 2022 at 04:35:32PM +0100, Phil Elwell wrote:
>> Jason,
>>
>> On 07/06/2022 16:14, Jason A. Donenfeld wrote:
>>> Hey again,
>>>
>>> On Tue, Jun 07, 2022 at 10:15:27AM +0100, Phil Elwell wrote:
>>>> On 07/06/2022 09:43, Jason A. Donenfeld wrote:
>>>>> Hi Phil,
>>>>>
>>>>> On Tue, Jun 7, 2022 at 10:29 AM Phil Elwell <phil at raspberrypi.com> wrote:
>>>>>>
>>>>>> This patch is fatal for me in the downstream Raspberry Pi kernel - it locks up
>>>>>> on boot even before the earlycon output is available. Hacking jump_label_init to
>>>>>> skip the jump_entry for "crng_is_ready" allows it to boot, but is likely to have
>>>>>> consequences further down the line.
>>>>>
>>>>> Also, reading this a few times, I'm not 100% sure I understand what
>>>>> you did to hack around this and why that works. Think you could paste
>>>>> your hackpatch just out of interest to the discussion (but obviously
>>>>> not to be applied)?
>>>>
>>>> This is the minimal version of my workaround patch that at least allows the
>>>> board to boot. Bear in mind that it was written with no previous knowledge of
>>>> jump labels and was arrived at by iteratively bisecting the list of jump_labels
>>>> until the first dangerous one was found, then later working out that there was
>>>> only one.
>>>
>>> Looks like this patch fails due to CONFIG_STRICT_KERNEL_RWX.
>>> Investigating deeper now, but that for starters seems to be the
>>> differentiating factor between my prior test rig and one that reproduces
>>> the error. I assume your raspi also sets CONFIG_STRICT_KERNEL_RWX.
>>
>> Yes, it does, as does multi_v7_defconfig.
> 
> Oh good. Adjusting my CI now to have that.
> 
> Having tickled arch/arm/ a little bit now, this is looking sort of
> complicated. So I think I might be leaning toward giving up and just
> rolling with <https://git.zx2c4.com/linux-rng/commit/?id=78f79dda>.
> 
> Unless of course somebody has some ARM chops and can think of a quick
> easy fix.

I've not looked at the performance implications (if any), but in terms of when the RNG initilialization completes and the lack of a WARN I'm happy with that patch, a.k.a. [1].

Phil

[1] https://lore.kernel.org/lkml/20220607100210.683136-1-Jason@zx2c4.com/