Message-ID: <fdea59fe-f570-489f-bf88-1ffd47119cac@nvidia.com>
Date: Wed, 12 Mar 2025 09:10:19 +1100
From: Balbir Singh <balbirs@...dia.com>
To: Bert Karwatzki <spasswolf@....de>
Cc: Ingo Molnar <mingo@...nel.org>, Kees Cook <kees@...nel.org>,
 Bjorn Helgaas <bhelgaas@...gle.com>,
 Linus Torvalds <torvalds@...ux-foundation.org>,
 Peter Zijlstra <peterz@...radead.org>, Andy Lutomirski <luto@...nel.org>,
 linux-kernel@...r.kernel.org
Subject: Re: commit 7ffb791423c7 breaks steam game

On 3/12/25 05:24, Bert Karwatzki wrote:
> Am Dienstag, dem 11.03.2025 um 12:15 +0100 schrieb Bert Karwatzki:
>> Am Dienstag, dem 11.03.2025 um 18:19 +1100 schrieb Balbir Singh:
>>> On 3/11/25 08:48, Balbir Singh wrote:
>>>> On 3/10/25 22:22, Bert Karwatzki wrote:
>>>>> Using linux next-20250307 to play the game stellaris via steam I noticed that
>>>>> loading the game gets sluggish with the progress bar getting stuck at 100%.
>>>>> In this situation mouse and keyboard inputs don't work properly anymore.
>>>>> Switching to a VT and killing stellaris somewhat fixes the situation though in
>>>>> one instance the touchpad did not work after that. I bisected this between
>>>>> v6.14-rc5 and next-20250307 and got this as the first bad commit:
>>>>>
>>>>> 7ffb791423c7c518269a9aad35039ef824a40adb is the first bad commit
>>>>> commit 7ffb791423c7c518269a9aad35039ef824a40adb (HEAD)
>>>>> Author: Balbir Singh <balbirs@...dia.com>
>>>>> Date:   Fri Feb 7 10:42:34 2025 +1100
>>>>>
>>>>>     x86/kaslr: Reduce KASLR entropy on most x86 systems
>>>>>
>>>>> Reverting commit 7ffb791423c7 in next-20250307 fixes the issue for me.
>>>>>
>>>>
>>>> Thanks for the report! Could you also share the dmesg? Do you have any proprietary
>>>> modules loaded? Could you also share the output of /proc/iomem (captured as sudo)?
>>>> The lspci output is useful, lspci -vvv (captured with sudo) would help see where
>>>> the BAR regions are, specifically before and after the patch. Could you also share
>>>> the kernel config?
>>>>
>>>> I assume your config has CONFIG_PCI_P2PDMA enabled. Did the system ever work with
>>>> nokaslr for you?
>>>>
>>>> I am a little surprised that reducing the entropy causes these issues and having
>>>> a larger direct map causes these issues.
>>>
>>> Hi, Bert
>>>
>>> I tried to reproduce the issue at my end, but could not. I am new to Steam and my
>>> hardware/games are also different. From what I could gather, Steam is a 32 bit
>>> application that uses SDL. For my testing I used Xorg, not wayland
>>>
>>> In addition to the above?
>>>
>>> 1. Could you confirm if the mouse/keyboard work outside of Steam?
>>> 2. Does evtest /dev/input/<mouse device> show anything?
>>> 3. We could run the kernel with nokaslr and see if the issue reproduces?
>>> 4. As a test, we could try disabling CONFIG_PCI_P2PDMA in the kernel config,
>>> build boot and test, this has the same effect has not having the patch
>>>
>>> Balbir
>>>
>>
>> 1. The problem is not with steam but specifically with the game stellaris (which
>> can be started without steam). As long as stellaris is running there is a ~2s
>> delay in keyboard and mouse input (for everything), killing stellaris restores
>> input to normal.
>> 2 evtest /dev/input/event{1,3,4} (keyboard, touchpad, mouse) seems to show the
>> same output whether stellaris is running or not, only if it's running there's
>> the mentioned delay.

Thanks, so the issue is specific to the game and running it?

>> 3. For some weird reason my kernel does not recongnize the nokaslr cmdline
>> parameter, so I build a kernel without CONFIG_RANDOMIZE_BASE and this does NOT
>> fix the issue.

Can you clarify if you're booting with the compressed image bzImage/vmlinuz or
with vmlinux?

>> 4. Most surprisingly removing CONFIG_PCI_P2PDMA also does NOT fix the issue.
>>


> 
> I've done more experimenting regarding 4.:
> next-20250307 with "CONFIG_RANDOMIZE_BASE=y" AND "CONFIG_PCI_P2PDMA is not set"
> works as expected (i.e. no input lag when stellaris is running)
> 
> next-20250307 with "CONFIG_RANDOMIZE_BASE is not set" AND "CONFIG_PCI_P2PDMA is
> not set" also shows the buggy behaviour (i.e. input lag when stellaris is
> running) (this was the configuration I tested before)

This is an interesting experiment, I am beginning to wonder if the system relies
on a reduced direct map for the game to work correctly. Can you also check in this
scenario if CONFIG_RANDOMIZE_MEMORY is disabled?

Can you please share the dmesg
1. output before and after the changes?
2. Do you see any warnings/errors in journalctl or game specific log files?
3. lspci -vvv output before and after the changes?


> 
> As a sidenote, I've tested several kernel with nokaslr as command line parameter
> (6.1.128, 6.8.12, 6.12.17 (the debian sid distributional kernel)) and nokaslr is
> not recognized as a command line parameter in any of them
> 

Please see my comment above about booting. How did you check if nokaslr is being
recognized, is it via looking up dmesg?

Thanks!
Balbir Singh

Powered by blists - more mailing lists