Re: pull request: wireless 2011-11-22

[David Miller]

From: Larry Finger <Larry.Finger@xxxxxxxxxxxx>
Date: Tue, 22 Nov 2011 15:30:33 -0600

> On 11/22/2011 03:13 PM, David Miller wrote:
>> From: David Miller<davem@xxxxxxxxxxxxx>
>> Date: Tue, 22 Nov 2011 16:05:11 -0500 (EST)
>>
>>> From: "John W. Linville"<linville@xxxxxxxxxxxxx>
>>> Date: Tue, 22 Nov 2011 15:56:55 -0500
>>>
>>>> On Tue, Nov 22, 2011 at 03:14:29PM -0500, David Miller wrote:
>>>>> From: "John W. Linville"<linville@xxxxxxxxxxxxx>
>>>>> Date: Tue, 22 Nov 2011 14:35:05 -0500
>>>>>
>>>>>> Here is the latest batch of fixes intended for 3.2.  This includes a
>>>>>> correction for a user-visible error in mac80211's debugfs info, a fix
>>>>>> for a potential memory corrupter in prism54, an endian fix for rt2x00,
>>>>>> an endian fix for mac80211, a fix for a NULL derefernce in cfg80211, a
>>>>>> locking fix and a deadlock fix for p54spi, and a pair of rt2x00 fixes
>>>>>> for handling some spurious interrupts that hardware can generate.
>>>>>>
>>>>>> Please let me know if there are problems!
>>>>>
>>>>> The rt2800pci change doesn't look correct.
>>>>>
>>>>> If the IRQ line is shared with another device, this change will make
>>>>> it
>>>>> never see interrupts.  Once you say "IRQ_HANDLED" the IRQ dispatch
>>>>> stops processing the interrupt handler list.
>>>>
>>>> I thought this at first as well.  But looking at the code in
>>>> kernel/irq/handle.c doesn't support that conclusion.  In fact, every
>>>> handler gets invoked no matter what they all return.  All of the irq
>>>> handler return values are ORed together and passed to note_interrupt.
>>>> Only if every irq handler returns IRQ_NONE does the code in
>>>> kernel/irq/spurious.c start getting involved.
>>>>
>>>> Anyway, this seems to be safe even for shared interrupts.  That said,
>>>> this is a bit ugly.  But it makes a serious difference in performance
>>>> for those afflicted with this issue.
>>>
>>> It just means that we won't notice spurious interrupts if the device
>>> sharing the line with rt2800pci generates one.
>>>
>>> This change is wrong.
>>
>> BTW, look at it this way, if what you say is true John then what's the
>> point
>> in returning any specific value at all?
>>
>> Everyone can just return IRQ_HANDLED and everything would just work.
>>
>> But you know that's not the case, and that it's important that this
>> value
>> is returned accurately.
> 
> I was trying to find the thread that reported the improvement in
> performance with this change, but failed. Is it possible that their
> change just papered over an interrupt storm from some other device
> that shared that interrupt?

It doesn't fix a performance problem, it fixes a problem wherein the
IRQ line is disabled by the generic IRQ code because all handlers
return IRQ_NONE.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/