[PATCH] Remove deprecated IRQF_DISABLED flag entirely

[Valentin Rothberg]

On Fri, Mar 6, 2015 at 8:41 PM, Dmitry Torokhov
<dmitry.torokhov at gmail.com> wrote:
> On Thu, Mar 5, 2015 at 5:11 AM, Hannes Reinecke <hare at suse.de> wrote:
>> On 03/05/2015 01:59 PM, Valentin Rothberg wrote:
>>> The IRQF_DISABLED is a NOOP and has been scheduled for removal since
>>> Linux v2.6.36 by commit 6932bf37bed4 ("genirq: Remove IRQF_DISABLED from
>>> core code").
>>>
>>> According to commit e58aa3d2d0cc ("genirq: Run irq handlers with
>>> interrupts disabled") running IRQ handlers with interrupts enabled can
>>> cause stack overflows when the interrupt line of the issuing device is
>>> still active.
>>>
>>> This patch ends the grace period for IRQF_DISABLED (i.e., SA_INTERRUPT
>>> in older versions of Linux) and removes the definition and all remaining
>>> usages of this flag.
>>>
>>> Signed-off-by: Valentin Rothberg <valentinrothberg at gmail.com>
>>> ---
>>> The bigger hunk in Documentation/scsi/ncr53c8xx.txt is removed entirely
>>> as IRQF_DISABLED is gone now; the usage in older kernel versions
>>> (including the old SA_INTERRUPT flag) should be discouraged.  The
>>> trouble of using IRQF_SHARED is a general problem and not specific to
>>> any driver.
>>>
>>> I left the reference in Documentation/PCI/MSI-HOWTO.txt untouched since
>>> it has already been removed in linux-next by commit b0e1ee8e1405
>>> ("MSI-HOWTO.txt: remove reference on IRQF_DISABLED").
>>>
>>> All remaining references are changelogs that I suggest to keep.
>>
>> While you're at it: having '0x0' as a value for the irq flags looks
>> a bit silly, and makes you wonder what the parameter is for.
>>
>> I would rather like to have
>>
>> #define IRQF_NONE 0x0
>>
>> and use it for these cases.
>> That way the scope of that parameter is clear.
>
> No, that would imply that IRQ never triggers whereas passing 0 means
> we keep triggers that have been set by the platform.

Are you against introducing a new flag or just don't like 'IRQF_NONE'?
 I think that passing 0 could mean anything when one does not know the
semantics.  Combining yours and Hannes' proposal could look like this:

#define IRQF_PLAT 0x0 - keep triggers that have been set by the platform

I wrote a Coccinelle script to check for such 0-flags and find 758
cases in current Linus' mainline.  The script only checks function
calls to {devm_}request_{threaded_}IRQ() but does not find flags
passed to wrapper functions or flags that are stored in a struct etc.

Kind regards,
 Valentin