On Thu, 19 Mar 2020 at 15:13, Linus Walleij [off-list ref] wrote:

On Tue, Mar 17, 2020 at 12:58 PM Peter Maydell [off-list ref] wrote:

quoted

What in particular does this personality setting affect?
My copy of the personality(2) manpage just says:

       PER_LINUX32 (since Linux 2.2)
              [To be documented.]

which isn't very informative.

It's not a POSIX thing (not part of the Single Unix Specification)
so as with most Linux things it has some fuzzy semantics
defined by the community...

I usually just go to the source.

If we're going to decide that this is the way to say
"give me 32-bit semantics" we need to actually document
that and define in at least broad terms what we mean
by it, so that when new things are added that might or
might not check against the setting there is a reference
defining whether they should or not, and so that
userspace knows what it's opting into by setting the flag.
The kernel loves undocumented APIs but userspace
consumers of them are not so enamoured :-)

As a concrete example, should "give me 32-bit semantics
via PER_LINUX32" mean "mmap should always return addresses
within 4GB" ? That would seem like it would make sense --
but on the other hand it would make it absolutely unusable
for QEMU's purposes, because we want to be able to
do full 64-bit mmap() for our own internal allocations.
(This is a specific example of the general case that
I'm dubious about having this be a process-wide switch,
because QEMU is a single process which sometimes
makes syscalls on its own behalf and sometimes makes
syscalls on behalf of the guest program it is emulating.
We want 32-bit semantics for the latter but if we
also get them for the former then there might be
unintended side effects.)

I would not be surprised if running say i586 on x86_64
has the same problem, just that noone has reported
it yet. But what do I know. If they have the same problem
they can use the same solution. Hm QEMU supports
emulating i586 or even i386 ... maybe you could test?

Native i586 code on x86-64 should be fine, because it
will be using the compat syscalls, which ext4 already
ensures get the 32-bit sized hash (see hash2pos() and
friends).

thanks
-- PMM