Blars Blarson a écrit :
In article <[🔎] 446508B5.6070003@systella.fr> mt1@systella.fr writes:I never see any SMP bug on sparc64 after 2.6.14 release... While debian/Sparc uses a 32 bits userland, I don't understand why we have to cut the sparc port in sparc32 and sparc64 ports. Today, the main problem comes from the worst support of SMP in sparc32 kernel, from the ESPFAS (and its DMA trouble) and from the HyperSPARC cache.My ultra-2 crashes randomly under high load with kernels as late as 2.6.17rc2. "RED state exception" is what it gets, in the 2.6.17rc it gets stuck in an infinite loop of them rather than rebooting. Have you tried putting a high load on your system? Compiling openoffice and the kernel at the same time is the type of load being discussed.
I have a lot of sparc64 systems with high load (the load average of one of them, an Ultra60/SMP is 14 for the last day) and I have never seen this trouble.
kant:[~] > uptime 08:31:17 up 103 days, 13:32, 31 user, load average: 16.54, 12.07, 14.39 kant:[~] > cat /proc/cpuinfo cpu : TI UltraSparc II (BlackBird) fpu : UltraSparc II integrated FPU promlib : Version 3 Revision 23 prom : 3.23.1 type : sun4u ncpus probed : 2 ncpus active : 2 D$ parity tl1 : 0 I$ parity tl1 : 0 Cpu0Bogo : 897.02 Cpu0ClkTck : 000000001ad28548 Cpu2Bogo : 897.02 Cpu2ClkTck : 000000001ad28548 MMU Type : Spitfire State: CPU0: online CPU2: online kant:[~] > uname -aLinux kant 2.6.15.1-netfilter-route-patch #3 Fri Jan 27 13:21:49 CET 2006 sparc64 GNU/Linux
Regards, JKB