[duaneb]

> Second, I wonder if it would make sense to jump straight to using mmap() instead of the classic brk()/sbrk().

Yes, it will. brk/sbrk are terrible ways to allocate anything but more stack. Use mmap instead, both for more control over the layout in memory, and for more portability.

[nkurz]

Perhaps you know: on Linux, how do mmap() and brk() differ as far as page initialization? Does brk() conceptually include the effects of MAP_POPULATE? Do both of them cause virtual to physical memory mapping at the time of the call?

[duaneb]

I'm not sure as to the specifics, because brk is a strict subset of mmap; however, I would be surprised if it doesn't use the full VM substructure to allocate the mapping immediately.

I would amend my statement before to say that, if you're looking into memory allocation, conceptually, mmap is where everything happens these days. brk is kept for backwards compatibility.

[lgeek]

On the systems I've seen, brk makes a lazy allocation, i.e. it doesn't allocate physical memory until you write to a page.