linux-kernel - Re: [PATCH] platform: finally disallow IRQ0 in platform_get

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]

Message-ID: <8a415980-990b-abae-6f60-dedd0c199583@omp.ru>
Date:   Tue, 4 Jan 2022 22:27:58 +0300
From:   Sergey Shtylyov <s.shtylyov@....ru>
To:     Andy Shevchenko <andriy.shevchenko@...el.com>
CC:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        "Rafael J. Wysocki" <rafael@...nel.org>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] platform: finally disallow IRQ0 in platform_get_irq() and
 its ilk

On 12/10/21 2:17 PM, Andy Shevchenko wrote:

[...]
>>>> The commit a85a6c86c25b ("driver core: platform: Clarify that IRQ 0 is
>>>> invalid") only calls WARN() when IRQ0 is about to be returned, however
>>>> using IRQ0 is considered invalid (according to Linus) outside the arch/
>>>> code where it's used by the i8253 drivers. Many driver subsystems treat
>>>> 0 specially (e.g. as an indication of the polling mode by libata), so
>>>> the users of platform_get_irq[_byname]() in them would have to filter
>>>> out IRQ0 explicitly and this (quite obviously) doesn't scale...
>>>> Let's finally get this straight and return -EINVAL instead of IRQ0!
>>>
>>> You are changing the return value of platform_get_irq_optional().
>>> The problem here is the proposed change doesn't bring any value in such
>>> case. platform_get_irq_optional() should be able (at the end of the day)
>>> to return 3 types of values (as other APIs do):
>>> 	 > 0: success
>>> 	== 0: IRQ not found
>>> 	 < 0: an error that must be consumed by the caller
>>
>>    I remember that was in your patch that got reverted right after being merged. ;-)
>> IMHO returning both error code and 0 on failure is a sign of a misdesigned API, it
>> makes the failure check unnecessarily complex and error prone.
> 
> I dunno what you are talking about when you mentioned "0 on failure" because 0
> is not the failure, that's what I'm trying to tell.

   OK.
 
>>> 0 is unexpected result for non-optional APIs and there you may try to play
>>> tricks (like replacing it by error code).
>>>
>>> There was a discussion around the topic:
>>> https://lore.kernel.org/lkml/20210331144526.19439-1-andriy.shevchenko@linux.intel.com/T/#u
>>
>>    I don't see much of the discussion there...
> 
> Indeed, it was split between two threads. Another one is this:
> https://lore.kernel.org/linux-serial/20210407101713.8694-1-andriy.shevchenko@linux.intel.com/T/#u

   OK.

>>> Wanna help?
>>
>>    No, I'm afraid you're on your own here... 

   Tell me please, how far you've got with this by now?
   (I've already started to add the fixups to your patch -- unfortunately, this change has to be
done atomically, not piecemeal.)

>>>> Fixes: a85a6c86c25b ("driver core: platform: Clarify that IRQ 0 is invalid")
>>>
>>> Not sure.
>>
>>    Why? It fixes gthe IRQ0 problem, so that you don't have to check for IRQ0 in many callers
>> (for the subsytems that treat 0 as s/th special, like polling mode)... If you have something
>> to improve, you can do that atop of this patch...
> 
> Because first we need to fix all users of platform_get_irq_optional().

   I still don't understand why your issue should be fixed 1st -- but I don't really care about
the order...

MBR, Sergey