Re: mmap and MAP_FIXED

[Daniel Jour <daniel.oertwig%gmail.com@localhost>]

Regarding the page / map size: At first I tried this with larger sizes
(around 98 kb), but no difference.


Johnny wrote:
> What is the pagesize of the machine?

pagesize yields 4096 on the system in question, so it shouldn't be of
relevance, I guess.


Michael wrote:
> But the start address in the test program is aligned to something much
> larger, so the first test should succeed if that was the problem. I
> wouldn't bet on the address range used being available though, but
> that's just a hunch.

Why shouldn't the range be available, though? /proc/PID/maps does not
show any mapping in this range. Is there some kernel related mapping
that doesn't show up in /proc/PID/maps?


Michael wrote:
> Also, why MMAP_FIXED in the first place?

Actually I don't know. A lot of CLISP's code is pretty old, so I'd guess
it's just a remnant or maybe due to portability to operating  systems that
cannot search a free address range in virtual memory.


Michael wrote:
> Finally, mmap()ing 1TB will probably run into all sorts of other
> problems before hitting mmap() itself. If the underlying filesystem is
> ffsv1 it probably doesn't even support files that big.

Hm, the example program is a bit misleading here. I don't intend to map
such a big region, the programs purpose was to find a (0x1000, thus a single
page) address range where mmap with MAP_FIXED works. Every single call to
mmap in that example program fails.


Martin wrote:
> # define TRY_EMPTY_VM_SPACE    0xb0000000

Using this address actually works :) So I guess for some reason I cannot
mmap addresses as high up as 0x4000000000000 ... the questions is:
Why, when it seems to be unmapped?