linux-kernel - Re: [PATCH] spi: Fix error code checking in spi_mem_exec

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date: Wed, 13 Mar 2024 12:34:01 -0700
From: Florian Fainelli <florian.fainelli@...adcom.com>
To: Pratyush Yadav <pratyush@...nel.org>
Cc: Michael Walle <mwalle@...nel.org>, linux-spi@...r.kernel.org,
 Mark Brown <broonie@...nel.org>, Miquel Raynal <miquel.raynal@...tlin.com>,
 Mika Westerberg <mika.westerberg@...ux.intel.com>,
 "Chia-Lin Kao (AceLan)" <acelan.kao@...onical.com>,
 open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] spi: Fix error code checking in spi_mem_exec_op()

On 3/13/24 12:29, Pratyush Yadav wrote:
> On Wed, Mar 13 2024, Florian Fainelli wrote:
> 
>> On 3/13/24 11:28, Pratyush Yadav wrote:
>>> On Wed, Mar 13 2024, Michael Walle wrote:
>>>
>>>> On Wed Mar 13, 2024 at 6:10 PM CET, Florian Fainelli wrote:
>>>>> After commit cff49d58f57e ("spi: Unify error codes by replacing -ENOTSUPP with
>>>>> -EOPNOTSUPP"), our SPI NOR flashes would stop probing with the following
>>>>> visible in the kernel log:
>>>>>
>>>>> [    2.196300] brcmstb_qspi f0440920.qspi: using bspi-mspi mode
>>>>> [    2.210295] spi-nor: probe of spi1.0 failed with error -95
>>>>>
>>>>> It turns out that the check in spi_mem_exec_op() was changed to check
>>>>> for -ENOTSUPP (old error code) or -EOPNOTSUPP (new error code), but this
>>>>> means that for drivers that were converted, the second condition is now
>>>>> true, and we stop falling through like we used to. Fix the error to
>>>>> check for neither error being neither -ENOTSUPP *nor* -EOPNOTSUPP.
>>>>>
>>>>> Fixes: cff49d58f57e ("spi: Unify error codes by replacing -ENOTSUPP with -EOPNOTSUPP")
>>>>> Signed-off-by: Florian Fainelli <florian.fainelli@...adcom.com>
>>>>> Change-Id: I4159811f6c582c4de2143382473d2000b8755872
>>>>
>>>> Ha, thank you!
>>>>
>>>> Reviewed-by: Michael Walle <mwalle@...nel.org>
>>>>
>>>> FWIW in next, there is commit
>>>> e63aef9c9121e ("spi: spi-mem: add statistics support to ->exec_op() calls")
>>>> that probably will conflict with this one.
>>>>
>>>> Also, - not for this patch - but with that logic, spi_mem_exec_op()
>>>> might return EOPNOTSUPP *or* ENOTSUPP, even for drivers which might
>>>> still return ENOTSUPP, because there is one condition in
>>>> spi_mem_exec_op() which will always return EOPNOTSUPP. That is
>>>> somewhat confusing, no?
>>> I agree. I suppose it would be better to do:
>>>       if (!ret)
>>>          return 0;
>>>       if (ret == -ENOTSUPP || ret == -EOPNOTSUPP)
>>>          return -EOPNOTSUPP;
>>>
>>
>> But with e63aef9c9121e ("spi: spi-mem: add statistics support to ->exec_op()
>> calls") applied, would not that mean duplicating the statistics gathering, or
>> were the statistics gathering only intended for when ret == 0?
> 
> Hmm, I didn't properly understand this. Ignore my suggestion. Your patch
> does the right thing.

What I meant is that e63aef9c9121e will increment statistics not just 
when we return 0 from ctlr->mem_ops->exec_op, but also if we return 
-ENOTSUPP or -EOPNOTSUPP, and I am  not sure if this is exactly what is 
intended. But this is somewhat orthogonal.

> 
> In this case we should return ret when:
> 
>      ret is 0
>      OR
>      when ret is not -EOPNOTSUPP or -ENOTSUPP.
> 
> So if we get either of the two we _won't_ return and continue forward.
> 
>  From looking at just this, spi_mem_exec_op() only returns -EOPNOTSUPP so
> far since it has:
> 
> 	if (!spi_mem_internal_supports_op(mem, op))
> 		return -EOPNOTSUPP;
> 
> But then looking further, it has:
> 
> 	ret = spi_sync(mem->spi, &msg);
> 
> 	if (ret)
> 		return ret;
> 
> spi_sync() can return -ENOTSUPP if it goes via __spi_async(). I suppose
> we would need to fix that if we want consistent return codes. But that
> isn't a problem this patch should fix. So with the merge conflict fixed
> up,

Thanks, although as I replied to Mark in the other branch of the thread, 
since this is a regression affecting v6.8, would not we want it to be 
fast tracked, and not based upon for-next?

> 
> Reviewed-by: Pratyush Yadav <pratyush@...nel.org>
> 

-- 
Florian


Download attachment "smime.p7s" of type "application/pkcs7-signature" (4221 bytes)