Re: [PATCH 1/9] linux-user: Diagnose incorrect -R size

[Peter Maydell]

On Tue, 7 Mar 2023 at 10:12, Peter Maydell <peter.maydell@linaro.org> wrote:
>
> On Mon, 6 Mar 2023 at 21:24, Richard Henderson
> <richard.henderson@linaro.org> wrote:
> >
> > On 3/6/23 04:56, Peter Maydell wrote:
> > > On Mon, 6 Mar 2023 at 02:14, Richard Henderson
> > > <richard.henderson@linaro.org> wrote:
> > >>
> > >> Zero is the value for 'off', and should not be used with -R.
> > >> We have been enforcing host page alignment for the non-R
> > >> fallback of MAX_RESERVED_VA, but failing to enforce for -R.
> > >
> > > I'm pretty sure we have users who specifically use "-R 0" to
> > > ask for "definitely turn off any reserved VA".
> > > Here's a random example from an old gcc bug report:
> > > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60681
> > > and somebody using it via the environment variable:
> > > https://www.openembedded.org/pipermail/openembedded-core/2015-January/100572.html
> >
> > Odd.
> >
> > Well, it won't actually have the effect of "definitely turn off", it will 
> > merely leave
> > things as per the default, which *will* enable reserved va for 32-bit 
> > guests on 64-bit hosts.
> >
> > The only remaining question is whether we diagnose this oddness or silently 
> > accept it.  It
> > feels like someone playing with options they don't actually understand and 
> > an error is
> > warranted.
>
> I'm pretty sure I've issued the advice "turn off the reserved
> area stuff with -R 0" in the past, for working around various
> QEMU bugs where it wasn't able to allocate the whole reserved
> area it wanted to but the guest program didn't actually care.

It looks like we (inadvertently) broke "-R 0 means turn off"
in 2019 with commit dc18baaef36d95e5; prior to that the
64-on-32 default was set by the initial value of the global
variable and could be overridden on the command line. After
that we ended up doing the default-value stuff after the
command line was parsed instead.

thanks
-- PMM