Re: Increase Default vm_max_map_count to Improve Compatibility with Modern Games

[Pierre-Loup A. Griffais]

On 4/12/24 10:18 AM, Liam R. Howlett wrote:
> Adding Pierre-Loup to cc list.
>
>
> * David Hildenbrand <david@xxxxxxxxxx> [691231 23:00]:
> > On 02.04.24 09:34, Oleksandr Natalenko wrote:
> > > Hello.
> > >
> > > On středa 20. března 2024 22:05:34, CEST vincentdelor@xxxxxxx wrote:
> > > > Hello,
> > > >
> > > > I am writing to highlight an issue impacting many Linux users, especially those who enjoy modern gaming. The current default setting of `vm_max_map_count` at 65530 has been linked to crashes or launch failures in several contemporary games.
> > > >
> > > > To address this, I have opened a detailed bug report (218616 – Increase Default vm_max_map_count to Improve Gaming Experience on Linux) available at: 218616 – Increase Default vm_max_map_count to Improve Gaming Experience on Linux (kernel.org) .
>
> This is change is getting more traction as distributions switch to a
> higher number of VMAs.  I feel we need to educate them as to what this
> really means and why it is unnecessary and wrong.
>
> > > > We have identified that several modern games such as Hogwarts Legacy, Star Citizen, and others experience crashes or fail to start on Linux due to the default `vm_max_map_count` being set to 65530. These issues can be mitigated by increasing the `vm_max_map_count` value to over 1048576, which has been confirmed to resolve the crashes without introducing additional bugs related to map handling.
> > > >
> > > > This issue affects a wide range of users and has been noted in distributions like Fedora and Pop!_OS, which have already adjusted this value to accommodate modern gaming requirements.
> > > >
> > > > For reference, here is the change for Fedora:
> > > > https://fedoraproject.org/wiki/Changes/IncreaseVmMaxMapCount
> > > >
> > > > Here is a list of games affected by this low value in vm_max_map_count as reported to Valve:
> > > >
> > > > THE FINALS
> > > > https://github.com/ValveSoftware/Proton/issues/7317#issuecomment-1974837850
> > > >
> > > > Hogwarts Legacy
> > > > https://github.com/ValveSoftware/Proton/issues/6510#issuecomment-1422781100
> > > >
> > > > DayZ
> > > > https://github.com/ValveSoftware/Proton/issues/3899#issuecomment-1304397069
> > > >
> > > > Counter-Strike 2
> > > > https://github.com/ValveSoftware/Proton/issues/2704#issuecomment-1705199788
>
> Most of these do not have the vma information available anymore, if it
> was there (expired pastebin links, etc).
>
> ...
>
> > > > **Affected Games:**
> > > > - Hogwarts Legacy
> > > > - Star Citizen
> > > > - THE FINALS
> > > > - DayZ
> > > > - Counter-Strike 2
> > > > - Payday 2
> > > > - (and potentially others)
> > > >
> > > > **References:**
> > > > - Fedora's change documentation: https://fedoraproject.org/wiki/Changes/IncreaseVmMaxMapCount
> > > > - Various user reports and confirmations on gaming performance improvement with increased `vm_max_map_count`.
>
> Absolutely not.  This will do nothing for performance.  The game may run
> vs not run, but it won't get faster.

It's possible / likely the user was confused about the performance 
impact, here. The reason we make that change is to enable affected games 
to run at all, that would otherwise crash when they reach the limit.

> ...
>
> > Using a high VMA count usually implies that the application is doing
> > something suboptimal. Having many VMAs can degrade performance of MM
> > operations and result in memory waste.
> >
> > Running into VMA limits usually implies that the application is not
> > optimized and should handle things differently. Likely, the memory allocator
> > is doing something "bad" (e.g., mmap()+munmap() instead of MADV_DONTNEED)
> > and should be optimized.
>
> To be clear, what you are doing here is akin to adding more memory to
> your system when there is a memory leak.  This is not the solution you
> should be pushing.  Ironically, this is using more memory and performing
> worse than it should.  At best, the limit increase is a workaround for
> buggy programs.
>
> At worst, you are enabling bad things to keep happening and normalising
> poor programming choices.  Please put pressure on the applications that
> clearly have issues.

We don't get to prescribe what those applications do. The fact of the 
matter is that there are several high-performance memory allocators in 
wide use by game applications that make heavy internal use of mmap(), 
and that using hundreds of thousands of different memory mappings is 
well supported on the platform those applications were written for. (or 
mapping regions with different permissions, which results in different 
regions after platform translation to Linux happens within Wine)

Pointing out that there exists one game that doesn't happen to do that 
is not terribly useful for the purpose of this discussion.

The problem statement seems pretty simple - distributions that want to 
support those usecases out of the box can make that change, like we've 
done for years on SteamOS. On those that don't, users of those 
applications will have to discover and learn to apply the change by hand 
after having a likely sub-par experience trying to get their game up and 
running.

I've yet to hear a specific downside of making the change other than a 
real concern about DoS of kernel memory in another discussion - it seems 
to me like there is much lower hanging fruit for DoSing a Linux system 
you have shell access to, at the moment.

Thanks,
 - Pierre-Loup

> To put this into perspective, normal applications on Linux use in the
> 100s.  Insane applications (chromium) use 1000 to 2000. The heavier user
> is Android and that uses in the 1000s regularly (usually topping out at
> around 3000). You are allowing one process to use over 65,530 vmas.
>
> Again, this is per process.
>
> Currently, if you run Wolfenstein II: The New Colossus (proton game),
> there are 4 proton processes with 115, 104, 99, and 24 vmas.  I wanted a
> newer example, but steam (or nvidia) is having A Day on my gaming PC and
> won't start.
>
> You are enabling (and normalising across multiple popular distributions)
> a change from less than 120 to 1048576.  That is an increas of ~870x
> of virtual memory areas - not tiny chunks of memory, areas that can span
> large amounts of memory.  Assuming the *best* scenario in the buggy
> programs, you'd use 65531 vmas - that is still a multiple of 546x the
> number used by wolfenstein's largest thread.
>
> > I don't think we should be raising the limit for everybody out there.
>
> If there is an underlying technical reason for needing this number of
> vmas, then it isn't provided.  I'm going to guess it's DRM/anti-cheat
> that needs fixing.
>
> I'd like the problem to please be fixed and not hide it.  You are at a
> performance disadvantage with the current approach - and enabling others
> to do the same.
>
> Thank you,
> Liam