[bregma]

Well, that's a problem on many of the devices I've seen: zero is a valid memory address and dereferencing it does not cause any kind of crash. In fact some hardware requires reading or maybe even writing to that address. In an age of virtual memory there's no reason why zero should cause a crash and it wastes an entire page of memory for every application to make that happen, if it does.

[PhilipRoman]

I haven't had the chance to work on any MMU-less devices, but I don't quite follow your remark about wasting a page. Crashing is just the default behaviour in the absence of a mapping and consumes no resources.

[bregma]

The granularity of virtual address mapping is usually a page. On many systems, that's 4 kilobytes of address space. In order to trigger a fault when the address 0x0000000000000000 is dereferenced, it's necessary to map the entire address range from 0x0000000000000000 to 0x0000000000000fff to the same faulting behaviour.

That's a waste of a page.

[PhilipRoman]

Yeah but there is no "page" there. Assuming a process starts with an empty page table (meaning every single address will segfault), you don't have to do anything else to get the crashing behaviour for null.

Unless you're talking about some kind of device which supports virtual memory, but also by default sets up mappings, including at 0 which seems weird to me.

[bregma]

So if I have an empty page table 0x00000001 can be referenced as a valid address?

Memory mapping on all hardware with which I am familiar (which is not a small integer) is done by pages. If you have to fault on one address in a page (eg. zero page, stack canaries, heap guards) you have to fault on every single address in that page. But "memory is cheap". Until it's not.

[chuckadams]

Any system where losing one page would be too dear is likely not using a MMU in the first place.

Besides, memory access is twice as fast on the zero page! Wait, you're telling me the 6502 isn't the standard anymore? ;p