Spurious irq 95: 0xffffffdf, please flush posted write for irq 37 and Linux kernel thread "softirq" is consuming very high percentage of CPU - 95%

There have been at-least 3 major PSP releases since 02.01.03.11.

Do you have any specific reason to stay at 2.6.29 kernel?

Thank you so much for fast response Sanjeev

>> Do you have any specific reason to stay at 2.6.29 kernel?

We are in v2.6.29 (PSP  02.01.03.11) , because we have about 40+ kernel patches for our custom hardware based on OMAP 3503,  and it will take quite some time before migrate to latest release from TI.

Can you please let me know if this problem is resolved in the new release of kernel ? if yes which release ?

Is this a bug in the softirq kernel thread ?  which is fixed in the newer release ?

thanks

Pads

Resmi said:

Can you please let me know if this problem is resolved in the new release of kernel ? if yes which release ?

Resmi said:

Is this a bug in the softirq kernel thread ?  which is fixed in the newer release ?

Resmi said:

... because we have about 40+ kernel patches for our custom hardware based on OMAP 3503,

Dear Sanjeev

  Thank very much once again for the prompt response,

>>Can't point to a specific release, but here is the patch that should be fixing it

>>https://patchwork.kernel.org/patch/18244/

Even with this patch, I still see the problem , spurious interrupt message is displayed and the ksoftirqd kernel thread's CPU spikes and stays at 95% cpu usage.

below is the additional details

------------- dmesg output after issue has occurred ------------------------

.....

Beginning of smsc911x_open method
Beginning of smsc911x_soft_reset method
End of smsc911x_soft_reset method
net eth0: SMSC911x/921x identified at 0xa080c000, IRQ: 172
End of smsc911x_open method
ADDRCONF(NETDEV_UP): eth0: link is not ready
ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready
eth0: duplicate address detected!
Beginning of smsc911x_stop method
Spurious irq 95: 0xffffffdf, please flush posted write for irq 37                 
Spurious irq 95: 0xffffffdf, please flush posted write for irq 37                    <-------------- This msg is printed when execution is happening within smsc911x_stop() method of SMSC911x device driver, it is stuck at napi_disable() method , this issue occurs almost always in the napi_disable() which is called by smsc911x_stop() method, when eth0 interface is brought down to restart the eth0 interface.

.......

----------------------------  end -------------------------------------------------------

and the below is the output of  /proc/interrupt before and after this issue has occurred

------------------------ Before issue :  /proc/interrupts ----------------------------------------

           CPU0
 11:          0        INTC  prcm
 12:       4958        INTC  DMA
 24:          0        INTC  Omap 3 Camera ISP
 25:          0        INTC  OMAP DSS
 37:      49895        INTC  gp timer                     <------------------------this is low here
 56:        586        INTC  i2c_omap
 57:        175        INTC  i2c_omap
 61:         17        INTC  i2c_omap
 65:          0        INTC  omap_mcspi_isr
 72:          3        INTC  serial idle
 73:        448        INTC  serial idle
 74:       1649        INTC  serial idle, serial
 83:      15607        INTC  mmc0
 92:        170        INTC  musb_hdrc
 93:         63        INTC  musb_hdrc
172:        882        GPIO  eth0
174:          0        GPIO  maintenance_reset
176:          0        GPIO  cpld-power
369:          0     twl4030  twl4030_keypad
378:          0     twl4030  twl4030_usb
384:          0     twl4030  mmc0
Err:          0
-----------------------------------------end --------------------------------------

and

----------------- After issue: /proc/interrupts -------------------------------

           CPU0
 11:          0        INTC  prcm
 12:       8574        INTC  DMA
 24:          0        INTC  Omap 3 Camera ISP
 25:          0        INTC  OMAP DSS
 37:     183364        INTC  gp timer                    <------------------------this is very high here
 56:       1084        INTC  i2c_omap
 57:        178        INTC  i2c_omap
 61:         17        INTC  i2c_omap
 65:          0        INTC  omap_mcspi_isr
 72:         11        INTC  serial idle
 73:        448        INTC  serial idle
 74:       3607        INTC  serial idle, serial
 83:      22277        INTC  mmc0
 92:        635        INTC  musb_hdrc
 93:        528        INTC  musb_hdrc
172:       2305        GPIO  eth0
174:          0        GPIO  maintenance_reset
176:          0        GPIO  cpld-power
369:          0     twl4030  twl4030_keypad
378:          0     twl4030  twl4030_usb
384:          0     twl4030  mmc0
Err:          0
----------------------------------------------------------------------------

My observation is that whenever the eth0 interface is restarted (i.e. stopped and started), during stop sequence, in the ethernet device driver (in drivers/net/smsc911x.c )  in the smsc911x_stop() method, after  successfully invoking netif_stop_queue(dev); the   napi_disable(&pdata->napi) is called, and this is when the spurious irq message is printed and following kernel threads sudden start consuming high % of cpu ,  mmcqd and ksoftirqd , in top I see  98.9 % usage by softirq thread.

Any pointers/suggestions to resolve is greatly appreciated.

>>Personal opinion: you should consider the effort in porting against time spend in hitting the problems and debugging and finding that it has already been fixed; and

>>back-porting it. Remember this kernel version is really old.

I completely agree with you,  as of last evening, in parallel to finding a fix for the above issue,  I have initiated the process of porting our patches to latest release,

btw do you recommend moving to latest release or the one before the latest release  ?  as I am concerned about stability, need your expert inputs to decide on this.

Thank you ,

Pads,

Any updates ?