On Tue, 1 Nov 2016, Paul Burton wrote:

BTW, do you have a feel for whether there's a good r2k/r3k platform (ideal 
would be some software emulator if any are good enough) that we could hook up 
to our continuous integration system? That would help us to catch any 
regressions like this in future before they hit mainline.

 I know about no such platform I'm afraid.

 QEMU does not have the R2k/R3k exception/MMU/cache model and implementing 
that would be a considerable effort I see no volunteers for.  I haven't 
heard of any other simulator which might be closer to implementing that 
model.

 As to using real hardware -- I might be the closest myself to be capable 
of doing some automated testing as I have an R3k machine in my home lab 
wired for remote control.  It could track Ralf's `mips-for-linux-next' 
branch and watch out for regressions, by trying to build and boot kernels 
automatically on a regular basis; maybe doing some further validation 
even, such as running GCC or glibc regression testing.  But while the 
target is ready I'm still missing the host-side setup, which I haven't 
completed.  I don't think there's any other hardware readily available 
which could be hooked somewhere.

 So for the time being I think we need to continue relying on people 
spotting issues by hand.  I think we've been doing pretty good overall.

 Thanks for your review.

  Maciej