5.18.1 has thousands of similar lines in dmesg with CONFIG_WARN_ALL_UNSEEDED_RANDOM=y

5.18.1 has thousands of similar lines in dmesg with CONFIG_WARN_ALL_UNSEEDED_RANDOM=y

[Toralf Förster]

This deletes the early boot messages in dmesg:

/var/log # head dmesg
[    3.114274] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114286] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114291] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114297] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114304] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114315] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114323] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114329] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114333] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0
[    3.114345] random: get_random_u32 called from
allocate_slab+0x144/0x480 with crng_init=0

5.18.0 was fine

--
Toralf

Re: 5.18.1 has thousands of similar lines in dmesg with CONFIG_WARN_ALL_UNSEEDED_RANDOM=y

[Toralf Förster]

On 5/30/22 11:24, Toralf Förster wrote:
> This deletes the early boot messages in dmesg:
>

Well,

changing CONFIG_LOG_BUF_SHIFT=18 to CONFIG_LOG_BUF_SHIFT=20 helps, but
nevertheless:

  $ grep get_random_u dmesg-5.18.1 | wc
    5541   38787  440908

For 5.15.44 there's no similar effect in dmesg after upgrading the kernel.

--
Toralf

Re: 5.18.1 has thousands of similar lines in dmesg with CONFIG_WARN_ALL_UNSEEDED_RANDOM=y

[Jason A. Donenfeld]

Hi Toralf,

On Mon, May 30, 2022 at 11:24:55AM +0200, Toralf Förster wrote:
> This deletes the early boot messages in dmesg:
> 
> /var/log # head dmesg
> [    3.114274] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114286] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114291] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114297] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114304] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114315] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114323] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114329] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114333] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0
> [    3.114345] random: get_random_u32 called from
> allocate_slab+0x144/0x480 with crng_init=0

This was simplified here:

    https://lore.kernel.org/lkml/20220509172426.612649-1-Jason@zx2c4.com/
    https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=cc1e127bfa95b5fb2f9307e7168bf8b2b45b4c5e

I'm happy to consider other heuristics than the one there if you find
this one problematic. But I think as a debug option, the current
behavior encapsulates well what somebody might want to know when
debugging a kernel. Either way, please let me know if this poses a
problem for you.

Jason

Re: 5.18.1 has thousands of similar lines in dmesg with CONFIG_WARN_ALL_UNSEEDED_RANDOM=y

[John Ogness]

On 2022-05-30, Toralf Förster <toralf.foerster@gmx.de> wrote:
> On 5/30/22 11:24, Toralf Förster wrote:
>> This deletes the early boot messages in dmesg:
>
> changing CONFIG_LOG_BUF_SHIFT=18 to CONFIG_LOG_BUF_SHIFT=20 helps, but
> nevertheless:

FWIW, you can also use the boot argument "log_buf_len" to dynamically
allocate the dmesg buffer at boot.

For example: log_buf_len=16M

This is usually more convenient than reconfiguring/rebuilding your
kernel.

John Ogness

Re: 5.18.1 has thousands of similar lines in dmesg with CONFIG_WARN_ALL_UNSEEDED_RANDOM=y

[Toralf Förster]

On 5/30/22 21:52, John Ogness wrote:
> FWIW, you can also use the boot argument "log_buf_len" to dynamically
> allocate the dmesg buffer at boot.
indeed.
Thx.

--
Toralf

Re: 5.18.1 has thousands of similar lines in dmesg with CONFIG_WARN_ALL_UNSEEDED_RANDOM=y

[Toralf Förster]

On 5/30/22 21:47, Jason A. Donenfeld wrote:
> This was simplified here:
>
>      https://lore.kernel.org/lkml/20220509172426.612649-1-Jason@zx2c4.com/
>      https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=cc1e127bfa95b5fb2f9307e7168bf8b2b45b4c5e

Ah ok, understood the rationale behind that.
Thx.

--
Toralf