On Tue, 18 Apr 2006, Ludovic Court�s wrote:
In fact, I was using 2.6.14 (with Bob Breuer's SMP patch) which does
_not_ have this lock sequence change.
Now, I tried to apply this lock sequence change to my 2.6.14 tree to see
what happens and, well, the kernel does not hang as early as before but
I'm still getting "DMA error" messages and it panics soon enough as
well.
Where does it panic? Where you able to get a backtrace?
So far the reports on kernels with locking change reverted look like this:
Me, HyperSPARC processors, UP kernel - DMA problems gone.
Bertrand, HyperSPARC CPUs, SMP kernel - DMA problems still present.
Bertrand, SuperSPARC CPUs, SMP kernel - DMA problems gone.
You, HyperSparc CPUs, SMP kernel - DMA problems still present.
From this data set it appears that DMA problems is some combination of
the locking patch and SMP patch on HyperSPARCs only. It would be really
useful if you could try running HyperSPARC with UP kernel and locking
change reverted, that would at least make it clear that locking change
is responsible for at least a part of the problem.