Mark Morgan Lloyd a écrit :
BERTRAND Joël wrote:2.4 kernels (after 2.4.21) are not stable on sparc32-smp. On all sun4m-smp I have tested, they randomly hang with "watchdog reset".2.4.27 reliable here on an SS20 with 2x80MHz processors, but not with Ross modules. Irrespective of hardware, it looks as though if the kernel gets as far as running init then it's likely to be fairly solid. However one has only to look back through the ML to see how many people have problems before this point.
I don't agree. I have some SS20/SMP (with several configurations dual SM71, dual RT626, quad RT626, CG14, leo, CG6...) and I have seen the same bug with all kernel after the 2.4.21 on all configurations : system can crash with a "Watchdog reset" when X tries to initilize the framebuffer. Without X, system seems to be stable, but I suspect a serious bug.
The last kernel usable with ROSS modules is the 2.2. Today, I don't know any OS that is stable with more than one ROSS module. NetBSD randomly hangs with HyperSPARC/SMP (async register or NMI trap), there is a bug in memory management with linux 2.6 and HyperSPARC (even with only one module). With SM71/SMP and linux 2.6.19.5, I have seen three bugs : ESP driver is broken (device disconnected...), SunLANCE driver too, and pipes randomly hangs (tar xvfj linux-2.6.20.1.tar.bz2 hangs with Oops and syscall_too_hard()). Only Solaris 9 can be used but with dual RT626 (with three or quad RT626, system hans with "Watchdog reset"). If I have some time, I shall try to install a Solaris 2.7with quad-RT626...
I have tested all hardware with SUNvts and all motherboards, memory and CPU modules work fine (with Solaris of course ;-) ).
Regards, JKB