On Jul 16,  3:49pm, Richard Earnshaw wrote:
} On Wed, 23 Feb 2005 01:51:27 +0100, Wolfgang Solfrank wrote:
} > 
} > >Running -current from about a week ago on an evbarm board (IQ31244) a 
} > >build of libgcj.so (from gcc 4 sources) is dying with the error:
} > >
} > >	sbrk: grow 16384 failed, error = 12
} > 
} > Hmm, since I just looked into this on the amd64 platform, it's almost
} > certainly a problem with the kernel running out of virtual memory.
} > Our VM system uses quite some memory for housekeeping, at least
} > 16 bytes per page on 32-bit machines plus the machine dependent stuff.
} > While I thought that this would mostly be a problem on 64-bit machines
} > (with its much larger address space, but until now similar kernel virtual
} > memory limits), your mail seems to indicate that at least some 32-bit
} > boxes have similar problems.
} 
} I've bumped NKMEMPAGES to give me 16M of kvm space (was previously 7M), 
} and it seems to have solved the problem, but I need to do some more 
} testing.

     I've seen similar problems with 2.0/sparc.  Is there any
conclusion to this.  How does one calculate an appropriate value for
NKMEMPAGES?  Has this been fixed in 2.0.1/2.1/3.0/-current?

}-- End of excerpt from Richard Earnshaw