Received-SPF: pass (google.com: domain of linux-crypto+bounces-4210-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Message-ID: <eb3a2581-efc6-40c3-a7ea-551865017d40@foss.st.com>
Date: Thu, 16 May 2024 16:35:28 +0200
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH] [RFC] clk: stm32mp1: Keep RNG1 clock always running
To: Marek Vasut <marex@denx.de>, <linux-crypto@vger.kernel.org>
CC: =?UTF-8?Q?Uwe_Kleine-K=C3=B6nig?= <u.kleine-koenig@pengutronix.de>,
        Alexandre Torgue <alexandre.torgue@foss.st.com>,
        Gabriel Fernandez
	<gabriel.fernandez@foss.st.com>,
        Herbert Xu <herbert@gondor.apana.org.au>,
        Maxime Coquelin <mcoquelin.stm32@gmail.com>,
        Michael Turquette
	<mturquette@baylibre.com>,
        Olivia Mackall <olivia@selenic.com>, Rob Herring
	<robh@kernel.org>,
        Stephen Boyd <sboyd@kernel.org>,
        Yang Yingliang
	<yangyingliang@huawei.com>,
        <linux-arm-kernel@lists.infradead.org>, <linux-clk@vger.kernel.org>,
        <linux-stm32@st-md-mailman.stormreply.com>
References: <20240513220349.183568-1-marex@denx.de>
 <b2d0dfcb-37d6-4375-a4ad-ca96a5339840@foss.st.com>
 <cc6f98eb-f6b2-4a34-a8ed-c0f759fa4c79@denx.de>
 <51951dd4-8e8c-4e67-89f6-6a710022e34f@foss.st.com>
 <3257e8f8-5bb0-4c75-a3a3-e5685b65de2a@denx.de>
 <5b39b5b6-7008-4362-a578-3faab87cd23b@foss.st.com>
 <2eb2b80e-8650-46cf-9d8f-6dd6a884558a@denx.de>
Content-Language: en-US
From: Gatien CHEVALLIER <gatien.chevallier@foss.st.com>
In-Reply-To: <2eb2b80e-8650-46cf-9d8f-6dd6a884558a@denx.de>
Content-Type: text/plain; charset="UTF-8"; format=flowed
Content-Transfer-Encoding: 8bit


On 5/16/24 12:43, Marek Vasut wrote:
> On 5/16/24 9:42 AM, Gatien CHEVALLIER wrote:
> 
> Hi,
> 
>>>>>> What if you add a trace in a random generation function in random.c?
>>>>>
>>>>> Do you have a function name or line number for me ?
>>>>
>>>> I put a trace in _get_random_bytes() in drivers/char/random.c. I'm not
>>>> 100% sure but this should be the entry point when getting a random 
>>>> number.
>>>
>>> You're right, there is a read attempt right before the hang, and 
>>> __clk_is_enabled() returns 0 in stm32_read_rng() . In fact, it is the 
>>> pm_runtime_get_sync() which is returning -EACCES instead of zero, and 
>>> this is currently not checked so the failure is not detected before 
>>> register access takes place, to register file with clock disabled, 
>>> which triggers a hard hang.
>>>
>>> I'll be sending a patch shortly, thanks for this hint !
>>>
>>
>> Great news, indeed the return code isn't checked. Let's use
>> pm_runtime_resume_and_get().
> 
> Yes please.
> 
> I will wonder why we get EACCES though, that basically means we are 
> suspending already. Is it safe to return -errno from rng read function 
> in that case ?

The framework expects a function that can return an error code so I
don't see why not. Else the framework would have an issue.

I still haven't figured out what is happening.

Could it be that the kernel is getting entropy with hwrng_fillfn()
like it does periodically to feed the entropy pool and it happens at the
same time as your pm test sequence?

FYI, I have been running your script with (echo devices > 
/sys/power/pm_test) for 5 hours now and haven't been able to reproduce 
the issue.

> 
>>>>>> After this, I'll try to reproduce the issue.
>>>>>
>>>>> If you have a minute to test it on some ST MP15 board, that would 
>>>>> be real nice. Thanks !
>>>>
>>>> I tried to reproduce the issue you're facing on a STM32MP157C-DK2 no
>>>> SCMI on the 6.9-rc7 kernel tag. I uses OP-TEE and TF-A in the bootchain
>>>> but this should not have an impact here.
>>>>
>>>> How did you manage to test using "echo core > /sys/power/pm_test"?
>>>> In kernel/power/suspend.c, enter_state(). If the pm_test_level is core,
>>>> then an error is fired with the following trace:
>>>> "Unsupported test mode for suspend to idle, please choose 
>>>> none/freezer/devices/platform."
>>>
>>> Could this be firmware related ?
>>>
>>>> I've tried using "echo devices > /sys/power/pm_test" so that I can 
>>>> at least test that the driver is put to sleep then wakes up. I do not
>>>> reproduce your issue.
>>>
>>> Can you try 'processors' ?
>>>
>>
>> Given this:
>> #ifdef CONFIG_PM_DEBUG
>>          if (pm_test_level != TEST_NONE && pm_test_level <= TEST_CPUS) {
>>              pr_warn("Unsupported test mode for suspend to idle
> 
> You're supposed to be suspending to 'mem' , not 'idle' . Could that be it ?

Yes you're right, I've been missing that. I do not have "deep" available
in /sys/power/mem_sleep... not upstreamed yet maybe... Have you coded a
PSCI service for this in U-Boot?

I'm either missing something or I can't reproduce your setup.

Thanks,
Gatien