CompuServe Thread

Virtual Memory Wishlist

3 messages in this thread
#117274From: Chris ScheersAug 19, 1990 9:31 PM
Actually, it makes some difference whether you are using the 68451 MMU or the 68851 MMU. But actually, where most of my concern lies is the area of sharing areas of memory at other than the page level. In particular, chip memory on the Amiga is a problem. Chip memory is a resource that you would want to protect, but the Amiga design is such that you go in and directly fiddle it. Even if you don't, a defective Copper list can trash chip memory independently of any MMU (assuming that you can still modify a Copper list with a protected OS). To modify AmigaDOS to be completely secure, i.e., one task can not adversely affect another, would require a lot of effort in this area and would be very likely to not be fully compatible with current programs. Chris
#117849From: Gary DunlapAug 25, 1990 4:27 PM
For chip memory, I would expect to allocate a page and use it repeatedly until the page is exausted (that page would not be availible to other task for allocation). There would be upto one page of waste per task (20 pages total) of chip memory, and the same for stack, heap, data, and code pages. If the page discriptor is duplicated into a seperate tree for the user task, then the other task trees would not be able to access it, only the OS, user task, and custom chips would see these pages. For message ports and other task to task communications, do something simmular. Allocate pages of shared memeory, and use them as needed. I would not expect the Amiga to become totaly crash proof with the first use of VM, but at least it would be much harder to crash the OS, and other task than without VM. >> Gary <<
#118105From: Chris ScheersAug 28, 1990 2:07 AM
I have no doubt that an Amiga OS using VM could be made very crash proof. What I doubt is that the current ADOS can be made to even come close to crash proof, even with VM. Chris