IRQ storm from PMU chip

[graeme@…]

Approx every 5 boots of a GTA02 I get an irq storm which slows the machine to a
crawl and makes it next to useless. This is what turns up in dmesg.

pcf50633_irq: entering(irq=53, pcf=c7d06c00): scheduling work
pcf50633_work: INT1=0x00 INT2=0x00 INT3=0x40 INT4=0x00 INT5=0x00
ADCRDY
pcf50633_irq: entering(irq=53, pcf=c7d06c00): scheduling work
pcf50633_work: INT1=0x80 INT2=0x00 INT3=0x40 INT4=0x00 INT5=0x00
SECOND ADCRDY
pcf50633_irq: entering(irq=53, pcf=c7d06c00): scheduling work
pcf50633_work: INT1=0x00 INT2=0x00 INT3=0x40 INT4=0x00 INT5=0x00
ADCRDY
pcf50633_irq: entering(irq=53, pcf=c7d06c00): scheduling work
pcf50633_work: INT1=0x80 INT2=0x00 INT3=0x40 INT4=0x00 INT5=0x00
SECOND ADCRDY
pcf50633_irq: entering(irq=53, pcf=c7d06c00): scheduling work
pcf50633_work: INT1=0x80 INT2=0x00 INT3=0x40 INT4=0x00 INT5=0x00
SECOND ADCRDY
pcf50633_irq: entering(irq=53, pcf=c7d06c00): scheduling work
pcf50633_work: INT1=0x80 INT2=0x00 INT3=0x40 INT4=0x00 INT5=0x00
SECOND ADCRDY
pcf50633_irq: entering(irq=53, pcf=c7d06c00): scheduling work
pcf50633_work: INT1=0x00 INT2=0x00 INT3=0x40 INT4=0x00 INT5=0x00
ADCRDY

[thomas@…]

I also get this (possibly on every boot), but it doesn't seem to be impacting on
performance.

[andy@…]

It is quite possible something is slowing the thing to a crawl, which would be
important to understand, but it shouldn't be this. The SECOND thing is an
interrupt from the PMU that happens once a second.

You can confirm this with cat /proc/interrupts.

[tony@…]

Need to track what cause this PMU interrupt strom during various situtioon

[andy@…]

I think this can be the JACK_INSERT issue... Miles had an A5 which would spew
false interrupts from this after boot for a while.

The JACK_INSERT circuit is not great, we should perhaps disable this interrupt
not just for wake but as an interrupt.

Maybe Graeme could eyeball /proc/interrupts.

[graeme@…]

I shall try and do this, and good news is I can probably make it happen when I
am in Taipei next week so people can see for themselves.

[zecke@…]

Any news? How can this be related to the head set? Could it be that this is the
periodic tick of the PMU?

[zecke@…]

I'm tempted to close this as WORKSFORSOME. The jack insert interrupt has been
fixed, the many messages on the console are from someone starting the ADC that
often (e.g. hald polling /proc/apm). So the crawl down/interrupt storm came from
something else. We don't know what though.

[andy]

I'm giving in to Holger's temptation and closing as WORKSFORME. A real interrupt storm would leave big numbers in /proc/interrupts against PMU. If it's something else we can hear about new symptoms in a new bug.