[BISECTED] 3.10-rc1 OMAP1 GPIO IRQ regression

[Jon Hunter]

On 22/05/13 22:20, Aaro Koskinen wrote:
> On Tue, May 21, 2013 at 08:37:16PM +0100, Jon Hunter wrote:
>> On 21/05/13 18:39, Aaro Koskinen wrote:
>>> On Mon, May 20, 2013 at 10:46:21AM -0700, Tony Lindgren wrote:
>>>> * Tony Lindgren <tony at atomide.com> [130516 14:50]:
>>>>> * Aaro Koskinen <aaro.koskinen at iki.fi> [130516 14:05]:
>>>>>> On Thu, May 16, 2013 at 11:09:34AM -0700, Tony Lindgren wrote:
>>>>>>> * Aaro Koskinen <aaro.koskinen at iki.fi> [130513 13:58]:
>>>>>>>> I tested 3.10-rc1 on OMAP1 / Nokia 770, and Retu MFD probe is broken:
>>>>>>>>
>>>>>>>> [    2.264221] retu-mfd 2-0001: Retu v3.2 found
>>>>>>>> [    2.281951] retu-mfd 2-0001: Failed to allocate IRQs: -12
>>>>>>>> [    2.300140] retu-mfd: probe of 2-0001 failed with error -12
>>>>>>>>
>>>>>>>> The error is coming from regmap code. According to git bisect, it is
>>>>>>>> caused by:
>>>>>>>>
>>>>>>>> 	commit ede4d7a5b9835510fd1f724367f68d2fa4128453
>>>>>>>> 	Author: Jon Hunter <jon-hunter at ti.com>
>>>>>>>> 	Date:   Fri Mar 1 11:22:47 2013 -0600
>>>>>>>>
>>>>>>>> 	    gpio/omap: convert gpio irq domain to linear mapping
>>>>>>>>
>>>>>>>> The commit does not anymore revert cleanly, and I haven't yet tried
>>>>>>>> crafting a manual revert, so any fix proposals/ideas are welcome...
>>>>>>>
>>>>>>> Hmm this might be a bit trickier to fix. Obviously the real solution
>>>>>>> is to convert omap1 to SPARSE_IRQ like we did for omap2+.
>>>>>>>
>>>>>>> For the -rc cycle, it might be possible to fix this by adding a
>>>>>>> different irq_to_gpio() and gpio_to_irq() functions for omap1.
>>>>>>
>>>>>> The commit reverts cleanly if we also revert
>>>>>> 3513cdeccc647d41c4a9ff923af17deaaac04a66 (gpio/omap: optimise interrupt
>>>>>> service routine), which seems to be just some minor optimization. The
>>>>>> result is below, and with it 770 works again.
>>>>>
>>>>> Hmm in this case it seems that we should just fix it rather than go back
>>>>> to the old code, so let's take a look at that first.
>>>>
>>>> Does the following fix it for you or do we need to fix something else
>>>> there too?
>>>
>>> Thanks, that fixes Retu probe on 770.
>>
>> Sorry for not responding sooner, but I have been moving. 
>>
>> From looking at the code it appears that the regmap_add_irq_chip()
>> is failing in the retu driver. I am not sure if this will work, 
>> but have you tried making the following change to the retu driver 
>> so that it also uses irq_domain_add_linear() as well? Just a thought ...
> 
> This will trigger WARNING: at drivers/base/regmap/regmap-irq.c:514.

Sorry, there is another change needed. Can you try ...

diff --git a/drivers/mfd/retu-mfd.c b/drivers/mfd/retu-mfd.c
index a183098..9080cd6 100644
--- a/drivers/mfd/retu-mfd.c
+++ b/drivers/mfd/retu-mfd.c
@@ -267,14 +267,13 @@ static int retu_probe(struct i2c_client *i2c, const struct i2c_device_id *id)
        if (ret < 0)
                return ret;
 
-       ret = regmap_add_irq_chip(rdev->regmap, i2c->irq, IRQF_ONESHOT, -1,
+       ret = regmap_add_irq_chip(rdev->regmap, i2c->irq, IRQF_ONESHOT, 0,
                                  rdat->irq_chip, &rdev->irq_data);
        if (ret < 0)
                return ret;
 
        ret = mfd_add_devices(rdev->dev, -1, rdat->children, rdat->nchildren,
-                             NULL, regmap_irq_chip_get_base(rdev->irq_data),
-                             NULL);
+                             NULL, 0, regmap_irq_get_domain(rdev->irq_data));
        if (ret < 0) {
                regmap_del_irq_chip(i2c->irq, rdev->irq_data);
                return ret;
 
> Also other drivers report GPIO IRQ issues later in the boot, e.g.
> 
> [    3.907928] genirq: Flags mismatch irq 0. 00000001 (serial wakeup) vs. 00002000 (RETU)
> [    3.923706] No interrupt for UART wake GPIO: 37

Looks like the gpio is just conflicting with the RETU driver.
 
> With Tony's patch (without any Retu modifications), the boot is clean. So
> I guess gpio-omap fix is needed.

Ideally it would be good if the RETU can use linear domains and then may be this
can be avoided.

Cheers
Jon