IronPort-SDR: zZe3tle/M9ZaDVgjdKR/kvXMgCTY0oJzESKYT1NLqsb9md0D1A1uAAMgjSlPIKoEs0vR/Fv7sM
 5jfjim8vSNo86y/PHAGGnZLwEbq56U8vEvySXbwS8sjyPQCKSlzF2PoMC+j7O4q77k893QVoz0
 NGYsixDEt2wQyVel6Jd4Ol3tTEYECC6D4Li9g9/mLDsaHlcVgx6+DkmbpjmAryJwyR1OWT+pHY
 tum0Uf/oMpEa0nKEHALl4wnZWFnpw+6w3Gy2dFfpT/WXza9p/hOpUbls9LbZyuyl+Auli01YhE
 VSMtofKtLFR0Mx1oCAwIvjNI
IronPort-SDR: AafbcMYtH23Ll5KNtYwkH2QpWDl2uYubZBotjMyjnAaXckm9cZgB8iTC/VOZCPHx5W2LH6gfIm
 LCM3+ygOzOrzNbhm6CepOqSE83BpbFjrcuDGEXMOaUpy17Hk2Iu2Uc2qe2NIfBQ85fMXGNmqw7
 Wrkec+3xePJQuaDGAl4iYfr7XrSJkhF5OG3456s+WES0OhRz8yx6q8Q3Z/G+smNWyD0mZIDENL
 CZVaH6+bIgqgC/kylh5juXhwa+GntZS6GL6N0nqxatVmKc1yMzX3Kj6d5WITltREovPaYLKTkY
 S/U=
WDCIronportException: Internal
Message-ID: <0d967bb4-0b80-c293-b7d5-f49c9cc38718@opensource.wdc.com>
Date:   Tue, 14 Dec 2021 06:36:00 +0900
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:91.0)
 Gecko/20100101 Thunderbird/91.4.0
Subject: Re: [PATCH v1 1/2] ata: libahci_platform: Get rid of dup message when
 IRQ can't be retrieved
Content-Language: en-US
To:     Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc:     Sergey Shtylyov <s.shtylyov@omp.ru>, linux-ide@vger.kernel.org,
        linux-kernel@vger.kernel.org, Hans de Goede <hdegoede@redhat.com>,
        Jens Axboe <axboe@kernel.dk>
References: <20211209145937.77719-1-andriy.shevchenko@linux.intel.com>
 <d91cf14d-c7d8-1c61-9071-102f38e8c924@opensource.wdc.com>
 <febc7f73-929f-d8a6-ea01-5056b9101b46@omp.ru>
 <549c1825-56e6-de9e-e109-77f0d06cfd0f@opensource.wdc.com>
 <5322dafd-86ad-a293-6005-29384cb96cc8@omp.ru>
 <de3dc434-8b87-5d9d-7fe8-bd44ff2bcbfb@opensource.wdc.com>
 <Ybcz85/ZoXRCmbbD@smile.fi.intel.com>
From:   Damien Le Moal <damien.lemoal@opensource.wdc.com>
Organization: Western Digital
In-Reply-To: <Ybcz85/ZoXRCmbbD@smile.fi.intel.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2021/12/13 20:52, Andy Shevchenko wrote:
> On Mon, Dec 13, 2021 at 07:39:31AM +0900, Damien Le Moal wrote:
>> On 2021/12/11 19:25, Sergey Shtylyov wrote:
>>> On 11.12.2021 2:45, Damien Le Moal wrote:
> 
> ...
> 
>>>>>> So 0 will be returned as-is. That is rather weird. That should be fixed to
>>>>>> return -ENXIO:
>>>>>>
>>>>>> 	if (WARN(ret == 0, "0 is an invalid IRQ number\n"))
>>>>>> 		return -ENXIO;
>>>>>> 	return ret;
>>>>>
>>>>>     My unmerged patch (https://marc.info/?l=linux-kernel&m=163623041902285) does this
>>>>> but returns -EINVAL instead.
>>>>
>>>> Thinking more about this, shouldn't this change go into platform_get_irq()
>>>> instead of platform_get_irq_optional() ?
>>>
>>>     Why? platform_get_irq() currently just calls platform_get_irq_optional()...
>>>
>>>> The way I see it, I think that the intended behavior for
>>>> platform_get_irq_optional() is:
>>>> 1) If have IRQ, return it, always > 0
>>>> 2) If no IRQ, return 0
>>>
>>>     That does include the IRQ0 case, right?
>>
>> IRQ 0 being invalid, I think that case should be dealt with internally within
>> platform_get_irq_optional() and warn/error return. IRQ 0 showing up would thus
>> be case (3), an error.
>>
>>>
>>>> 3) If error, return < 0
>>>> no ?
>>>
>>>    I completely agree, I (after thinking a bit) have no issues with that...
>>>
>>>> And for platform_get_irq(), case (2) becomes an error.
>>>> Is this the intended semantic ?
>>>
>>>     I don't see how it's different from the current behavior. But we can do 
>>> that as well, I just don't see whether it's really better...
>>
>> The problem I see is that the current behavior is unclear: what does
>> platform_get_irq_optional() returning 0 mean ? IRQ == 0 ? or "no IRQ" ? I think
>> it should be the latter rather than the former. Note that the function could
>> return ENOENT (or similar) for the "no IRQ" case. With that, case (2) goes away,
>> but then I do not see any difference between platform_get_irq_optional() and
>> platform_get_irq().
>>
>> If the preferred API semantic is to allow returning IRQ 0 with a warning, then
>> the kdoc comments of platform_get_irq_optional() and platform_get_irq() are
>> totally broken, and the code for many drivers is probably wrong too.
> 
> Yeah, what we need to do is that (roughly a roadmap):
>  - revisit callers of platform_get_irq_optional() to be prepared for
>    new behaviour
>  - rewrite platform_get_irq() to return -ENOENT
>  - rewrite platform_get_irq_optional() to return 0 on -ENOENT
> 
> This is how other similar (i.e. _optional) APIs do.

Sounds like a good plan to me. In the mean time though, your patch 1/2 should
keep the "if (!irq)" test and return an error for that case. No ?


-- 
Damien Le Moal
Western Digital Research