On Wed, Mar 20, 2024 at 03:37:34PM +0200, Jarkko Nikula wrote: > On 3/20/24 2:27 PM, Heiner Kallweit wrote: > > On 19.03.2024 22:11, Heiner Kallweit wrote: > > > On a N100-based mini pc I see the following. I found older reports with the same symptom, > > > but root cause seems to be different. > > > > > > - Interrupt 27 is not shared in my case > > > - I checked register values on entering the ISR. Interrupt mask and DW_IC_RAW_INTR_STAT > > > are both 0. > > > - After an interrupt storm of 100,000 interrupts the interrupt gets disabled > > > - The issue affects channel 0 only > > > > > > If not the I2C IP, then who else can touch the interrupt line? > > > > I noticed that after including INTEL_IDMA64 in my config the problem no longer occurs. > > So there seems to be a dependency. Should it be reflected in Kconfig, e.g. > > make MFD_INTEL_LPSS dependent on INTEL_IDMA64, or let it imply INTEL_IDMA64? > > > Hmm.. interesting. I'd say BIOS perhaps has left the IDMA active and is > generating interrupts until the idma64 driver acknowledges it. > > There should not be generic dependency since the i2c_designware is not using > the DMA and a quick test on one platform where idma64 and i2c_designware are > sharing the same interrupt without CONFIG_INTEL_IDMA64 not set didn't show > similar behavior. > > Andy: Do you have any additional ideas or debug hints to this? Can you share `cat /proc/interrupts` in non-working and working cases? Just to confirm: loading idma64 driver fixes the issue, correct? -- With Best Regards, Andy Shevchenko
