On 02/03/2012 10:56 AM, Lars Heidieker wrote: > On 02/02/2012 12:28 AM, Lars Heidieker wrote: >> Hi, >> >> I've just posted a patch ( http://www.netbsd.org/~para/fix.patch ) >> - It moves uareas and buffer cache back to the kernel_map restoring >> the previous behavior. Sizing the kmem_arena is changed accordingly >> (Something I stepped on while checking evbmips on gxemul). >> - Code to drain pools if the kmem_arena runs out of space. >> > > As the kmem_arena size is sized by nkmempages, this sizing needs review > for certain archs. > As pool_allocations that used the pool_alocator_nointr are allocating > from the arena. > > amd64: we are fine with the direct mapped segment > i386: should increase NKMEMPAGES_MAX_DEFAULT for about 256mb to 320mb > sparc64: the limit needs to be raised > sparc: the limit needs to be raised > mips: should be fine with direct mapped segment > vax: a raise is likely to be required > > For machines with limited memory and therefor small kmem_arena we might > want to reduce the amount of quantum-caching that is done in the > kmem_va_arena > > Lars >
The code for proper kmem_arena sizing: http://www.netbsd.org/~para/kmemsizing.diff params done for i386/amd64/sparc64/arm32 Lars