Fwd: [Prism54-devel] [patch 0/6] [prism54 svn trunk] various fixes for prism54 svn

[Jean-Baptiste Note]

Hello Bob, dear list,

>> >  the BUG in a trace!  Here it is, I hope it is useful.  This is the XG-601 with
>> >  no shared IRQ.
>> >
>> > ------------[ cut here ]------------
>> > kernel BUG at /home/bbeers/download/prism54/prism54-svn=20050516/ksrc/isl_38xx.c
>> > :248!
>> >  <7>CB drv Qs: [32032][31753][0][0][166][124]
>> > CB dev Qs: [62][32028][31753][0][0][162]
>> > Kernel panic - not syncing: Fatal exception in interrupt
>> >  <7>CB drv Qs: [32032][31753][0][0][166][124]
>> > CB dev Qs: [62][32029][31753][0][0][162]
>> > CB drv Qs: [32032][31753][0][0][166][124]
>> > CB dev Qs: [62][32030][31753][0][0][162]
>> > CB drv Qs: [32032][31753][0][0][166][124]
>> > CB dev Qs: [62][32031][31753][0][0][162]
>> > CB drv Qs: [32032][31753][0][0][166][124]
>> > CB dev Qs: [62][32032][31753][0][0][162]

It seems that either :
1/ we have serious stack corruption kernel-wise
2/ there's a bug in the kernel regarding coherent mappings

3/ the device lost it completely wrt the control block offsets. As you
can see, it goes on updating the rx queue in the control block. This can
be due to 
  a/ a firmware bug

  b/ a hardware bug of the dma engine

  c/ us breaking some implicit communication protocol rules with the
  hardware. Maybe doing the update in-tasklet was not a good idea after
  all ? -- the XH8196 driver does this though, i'll have a look at the
  original code by intersil. Additionally, i'll someday alleviate the
  update work in such a way that it could reasonbly be reintegrated into
  the interrupt handler.

I guess we should detect this error condition (queues indexes out to
lunch device-wise), and upon this 

1/ display the control block (debugging measure)

2/ fail and trigger a reset of the card, instead of BUGing on it, as we
currenlty do. 

Does this seem better and acceptable to you ?

> 1.0.4.3

I don't know how "reproducible" this bug is, but if you can trigger it
fairly easily, could you try with another firmware version to see how it
fares ?

> As you request, this reply is also sent to the list for all curious eyes.  I was
>  replying only to the list earlier in this thread, but after a couple of times
>  getting your replies off list, I thought I was perhaps breaking protocol, so
>  I began replying only to you.  

My mistake, hitting the wrong key to reply. I should learn and use
correctly my mailer.

> I am happy to continue the thread either way, I just want to help
> improve the driver.  I have a few different SBCs and 3 prism54 devices
> to test with.

I'm happy you can help !

JB

-- 
Jean-Baptiste Note
+33 (0)6 83 03 42 38
jean-baptiste.note at wanadoo.fr