[Bug 47883] New: Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

Bug ID: 47883
           Summary: Enable explicit large address aware override
           Product: Wine
           Version: 4.17
          Hardware: x86-64
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: -unknown
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

Some 32-bit games like Dragon Age: Origins start without sound in recent Wine
versions (see for example bug 43685), due to running out of 2 GB memory
limitations. One workaround is to use a patcher on the binary to enable large
address aware bit.

Such as: https://github.com/randomstuff/pe-set-laa

This is cumbersome however.

Please enable some explicit workaround (like some Wine environment variable for
example), to avoid using patching on the binaries.

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #1 from Austin English  ---
How does it behave on Windows?

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

--- Comment #2 from Shmerl  ---
(In reply to Austin English from comment #1)
> How does it behave on Windows?

Good question, I haven't test it personally. But it's quite possible that due
to different underlying libraries and different allocations, on Windows the
same program won't run out of virtual memory limit for 32-bit.

Another common case is for example using d9vk instead of wined3d. The former
might run out of RAM faster due to doing different allocations for Vulkan use
case. Since authors of such games didn't anticipate all these cases, they
didn't flip that LAA flag on the binary. Doing it with a patcher is a fix, but
the point is to have a a simpler option to force such behavior regardless of
the flag. It would provide an easy to access workaround.

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

Zebediah Figura  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]
           Severity|normal                      |enhancement

--- Comment #3 from Zebediah Figura  ---
(In reply to Austin English from comment #1)
> How does it behave on Windows?

The request is to add a generic configurable workaround for a bug that cannot
be properly fixed in a generic sense. The workaround is generic but not
universal, either because applications truncate the high bit(s) [which is why
the LAA flag exists] or because it's simply not enough to alleviate address
space pressure. There isn't really anything comparable on Windows; it "works"
because there just isn't that much pressure (except when there is).

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

--- Comment #4 from Rafał Mużyło  ---
So, as I've mentioned on irc, recently gstreamer began to hang some of wine
apps 
 (that were running fine before) by running out of resources.

A hint by zf suggested those resources is actually address space.

That led me to an experiment: I've set large address aware bit on the
executable in question.

Given that the hang was randomly triggered and took quite a bit of time, I
can't be sure it *definitely* solved the problem, but for the time being it
does seem so.

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

soredake  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

--- Comment #5 from Rafał Mużyło  ---
Finally got the app running long enough to get a valid result.

Now, again, I can't say I'm certain the crash is gone, but it didn't happen in
quite awhile.

Yet, winegstreamer is leaking something somewhere. After a long enough run,
'too many allocated handles, not caching' from wine server becomes quite common
in the output.

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

mirh  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #6 from mirh  ---
This is how proton does it. 
https://github.com/ValveSoftware/wine/commit/9de03fa9a529298d698b4be4c380739...

I guess it becomes kind of a very handy shortcut when you have programs (well,
games) starved for virtual memory left and right.. but if bug 44375 was to be
fixed then the use case would be just as "niche" as it already is on Windows.

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

--- Comment #7 from Shmerl  ---
(In reply to mirh from comment #6)
> This is how proton does it. 
> https://github.com/ValveSoftware/wine/commit/
> 9de03fa9a529298d698b4be4c380739c68958d00
>
> I guess it becomes kind of a very handy shortcut when you have programs
> (well, games) starved for virtual memory left and right.. but if bug 44375
> was to be fixed then the use case would be just as "niche" as it already is
> on Windows.

It would be good to upstream that variable override to Wine from Proton. And I
doubt such thing can be fixed in a generic fashion as above that will help all
use cases. Wine doesn't handle something like overrides with dxvk /
vkd3d-proton and such which can have their own RAM needs that differ from how
Windows does things. So simply having a switch gives some flexibility for those
who need it and it's easier than patching binaries.

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

--- Comment #8 from mirh  ---
As explained in bug 33858 (not really sure what that is eventually about tbh),
there's nothing to "fix" to begin with.

I *guess* that if some of the apis wine uses was proven to be "inherently and
inevitably" bigger, this evaluation may be subject to change then.. but for the
time being the mood seems to be that you are more likely to find bugs against
individual components that need improvement.

Even the infamous dxvk memory penalty was allegedly fixed last week, just with
some brain put on the allocation strategy.
(as for vkd3d, it seems a pretty bad example to make given that I can hardly
think to 32-bit dx12 applications)

[Bug 47883] Enable explicit large address aware override

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47883

soredake  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|broaden_acid002@simplelogin |
                   |.com                        |