Re: port-sparc64/54211: esiop stopped detecting boot disk

["Michael L. Hitch" <mhitch%montana.edu@localhost>]

On Thu, 14 Nov 2019, Michael L. Hitch wrote:

>  I noticed that sparc64 switched from gcc6 to gcc7 back in Februrary, so my 
> first wild guess is something in gcc7 may be causing the problem. There were 
> no noticable changes in the mpt code between NetBSD-8 and NetBSD-9, so I 
> don't know if the problem is in the mpt code, or something else it uses.
>
>  I will probably try this again with an older tree that still has gcc6 and I 
> can try with both versions).  I was doing that a year ago when I was having 
> trouble with the Ultra 45 crashing while configuring the audio driver.

  I had time over the Thanksgiving weekend to track this down.  I did 
determine that the switch to gcc7 did not cause this, and the commit that 
did was a few weeks later.  The culprit is revision 1.69 of sys/i2c/i2c.c 
(certainly not very obvious).

  That change resulted in the device probe trying to configure a 
non-existant i2c device on the Ultra 45.  Inspecting the dmesg output in 
this ticket shows the same thing.  A working kernel shows:

[   1.0000000] gpio at iic0 addr 0x18 not configured

  While a failing kernel shows:

[   1.0000000] admtemp0 at iic0 addr 0x18:  temperature sensor: id. 0xf0, 
rev. 0xf0

  Reverting this change results in a working kernel. both for -current and 
netbsd-9.

Mike


---
Michael L. Hitch                        mhitch%montana.edu@localhost
Operations Consulting,  University Information Technology
Montana State University, Bozeman, MT     USA