From: Jan Engelhardt <jengelh@xxxxxxxxxx> Date: Thu, 3 Dec 2009 20:53:35 +0100 (CET) > > On Wednesday 2009-12-02 23:57, David Miller wrote: > >>From: Jan Engelhardt <jengelh@xxxxxxxxxx> >>Date: Wed, 2 Dec 2009 19:41:34 +0100 (CET) >> >>> the T1 is said to be multi-core, however, I am not seeing any MC sched >>> levels in dmesg as I do on x86. Why is that? >> >>The T1 cores are repesented as SMT units. >> >>On multi-chip T2 systems such as 5140 et al., we use MC to repesent >>the physical CPU chips. > > Well I wonder, because htop gives me only something like the following > on `nice -n20 make -j24` in a Linux kernel build, and it surprises > me that it does not use all threads. I can generated some userspace > compiling load that distributes equally over all 24 threads, though. > Could it be something with the scheduler, or this very fact that > they are represented as SIBLING rather than MC? I've never used htop, I'll have to check it out. When I have checked in the past when I do "make -jN" builds on T1 the scheduler fills processes the first of every 4 cpus, then the second on every 4 cpus, etc. etc. loading the cores evenly as the number of processes increases. And that's the desired behavior. On T2 we have 8 threads per core, but two integer units (one for every 4 cpu threads) so the scheduler fills (within a core) cpu 1 on all cores, then cpu 4 on all cores, then cpu 2 on all cores, then cpu 5 on all cores, etc. etc. If this isn't happening, please investigate further :-) -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
