dwight@jove.UUCP (01/23/84)
Here's the problem: We have an 11/780 with 14Meg of memory. However when we gen the kernel with the necessary changes to locore.s and cmap.h to allow us to use it all, the system gets very flakey (large (6Meg) thrashers have their address spaces corrupted, things randomly dump core and so on). The system finds all 14M and allocates buffer space accordingly and everything goes fine until any heavy virtual memory activity starts up. I'm sure it's not a minor problem with kernel loading or compilation (I removed all the .o's before recompiling etc.). The changes I made were those mentioned in the back of the config manual, index fields in cmap.h (c_next & c_prev) expanded to 14 bits and a constant in locore.s changed to 16384. We know the memory is ok. We are running ok on 8Meg now. Has anybody experienced this problem? ps. I found some hardwired constants in Fastreclaim() dealing with the cmap and tried fixing them but I managed to make things worse. Dwight Melcher {ucbvax!hplabs,allegra!nbires,decvax!kpno,harpo!seismo,ihnp4!kpno}!hao!boulder!dwight