From: Jeff D. <jd...@ka...> - 2000-03-17 17:29:54
|
I fixed the swapping bug. That turned out to be a long-standing bug and I'm surprised I didn't see it before. The problem was that the kernel vm area wasn't mapped in to new processes, and if they needed to swap immediately, they segfaulted when trying to read the swap map, which is vmalloced. I also moved thread cleanup, including killing the underlying process, up a bit. The killing of the process now happens right after the new zombie calls schedule. If this doesn't fix the out-of-processes problem that Lars saw, I don't know what will. In other news, I tried getting a major number for the block device, which I renamed vbd (== Virtual Block Device). In the immortal words of hpa, "Both these names are extremely poor choices" (he was also referring to /dev/disk/*). Discussions are continuing... Jeff |