Re: [REGRESSION] PCI BAR allocation/D3cold failure: NVIDIA MX250 inaccessible since Kernel 6.x on Acer Swift 3 (SF314-56G)

On 5/16/26 12:16, Matthias Dittrich wrote:
> 
> I completed the bisect test between kernel 5.15 und 5.16 and found the
> commit and now have reproducible revert-test results. I used the
> bisect.sh (attached ) for automation. Testing criteria was status rev ff
> (bad) or rev a1 (good).  As it was the first indicator for me on Ubuntu.
> I also used olddefconfig as I know this config worked before.
>
> The identified commit (see final-bisect-log.txt) is:
> # first bad commit: [a1224f34d72a103829d6953935d6c6621f135b83] ACPI: PM:
> Check states of power resources during initialization

Thx for the bisection. Not my area of expertise, so the following might
send everyone in the wrong direction, nevertheless FWIW:

That commit caused trouble on two other machines and DMI quirks were
added to make them work again; see 9b04d99788cf47 ("ACPI: PM: Do not
turn of unused power resources on the Toshiba Click Mini") [v6.4-rc1]
and cd7ef20ba8c6e9 ("ACPI: PM: Add unused power resource quirk for
THUNDEROBOT ZERO") [v7.0-rc1] for details. Wonder if the problem on your
machine is similar.

Ciao, Thorsten

> Additionally testing (see Abschlusstest.txt):
> 
>   * Verification with Parent commit test: git checkout
>     a1224f34d72a103829d6953935d6c6621f135b83~1
>       o lspci reports `rev a1` and  `nvidia-smi` works correctly
>   * Revert test on 5.16 with git revert --no-commit
>     a1224f34d72a103829d6953935d6c6621f135b83
>       o     lspci reports `rev a1` and  `nvidia-smi` works correctly
>   * Revert test on 6.1 with git revert --no-commit
>     a1224f34d72a103829d6953935d6c6621f135b83
>       o lspci reports `rev a1` and  `nvidia-smi` works correctly
> 
> With kernels like 6.2 and newer "git revert" result in an error and
> stops (transleted to english):
> 
> git revert --no-commit a1224f34d72a103829d6953935d6c6621f135b83 Auto-
> merging drivers/acpi/power.c CONFLICT (content): Merge conflict in
> drivers/acpi/power.c error: could not revert a1224f34d72a... (ACPI: PM:
> Check states of power resources during initialization) Hint: After
> resolving the conflicts, mark the corrected paths with: Hint: 'git add
> <paths>' or 'git rm <paths>'
> 
> All tests have been done with Ubuntu 22.04. 
> 
> Let me know if I can test anything further.
> 
> 
> Test summary Ubuntu:
> 
> Test Case
> 
> 	
> 
> Kernel / Commit
> 
> 	
> 
> Modification
> 
> 	
> 
> PCI State
> 
> 	
> 
> |Nvidia-smi/
> NVIDIA Driver|
> 
> 	
> 
> Result
> 
> Baseline good
> Ubuntu 22.04
> 
> 	
> 
> Linux 5.15
> 
> 	
> 
> none
> 
> 	
> 
> |rev a1|
> 
> 	
> 
> works
> 
> 	
> 
> GOOD
> 
> Baseline bad
> 
> 	
> 
> Linux 5.16
> 
> 	
> 
> none
> 
> 	
> 
> |rev ff|
> 
> 	
> 
> fails
> 
> 	
> 
> BAD
> 
> Bisect parent
> 
> 	
> 
> |a1224f34...~1|
> 
> 	
> 
> none
> 
> 	
> 
> |rev a1|
> 
> 	
> 
> works
> 
> 	
> 
> GOOD
> 
> First bad commit
> 
> 	
> 
> |a1224f34...|
> 
> 	
> 
> none
> 
> 	
> 
> |rev ff|
> 
> 	
> 
> fails
> 
> 	
> 
> BAD
> 
> Revert verification
> 
> 	
> 
> |a1224f34...|
> 
> 	
> 
> reverted
> 
> 	
> 
> |rev a1|
> 
> 	
> 
> works
> 
> 	
> 
> GOOD
> 
> Forward revert test
> 
> 	
> 
> Linux 6.1
> 
> 	
> 
> reverted
> 
> 	
> 
> |rev a1|
> 
> 	
> 
> works
> 
> 	
> 
> GOOD
> 
> Ubuntu 22.04 (standard)
> 
> 	
> 
> Linux 6.8
> 
> 	
> 
> none
> 
> 	
> 
> |rev ff|
> 
> 	
> 
> fails
> 
> 	
> 
> BAD
> 
> Modern revert attempt
> 
> 	
> 
> Linux 6.2
> 
> 	
> 
> revert attempted
> 
> 	
> 
> conflict in |drivers/acpi/power.c|
> 
> 	
> 
> n/a
> 
> 	
> 
> not testable
> 
> *Linux 5.15 is the standard in Ubuntu 20.04 but also available as
> downgrade in Ubuntu 22.04
> 
> Test summary CachyOS:
> 
> Test Case
> 
> 	
> 
> Kernel / Commit
> 
> 	
> 
> Modification
> 
> 	
> 
> PCI State
> 
> 	
> 
> |Nvidia-smi/
> NVIDIA Driver|
> 
> 	
> 
> Result
> 
> Standard
> 
> 	
> 
> Linux 7.0.2
> 
> 	
> 
> none
> 
> 	
> 
> |rev a1|
> 
> 	
> 
> fails
> 
> 	
> 
> BAD
> 
> LTS
> 
> 	
> 
> Linux 6.18
> 
> 	
> 
> none
> 
> 	
> 
> |rev a1|
> 
> 	
> 
> fails
> 
> 	
> 
> BAD
> 
> Start parameter
> 
> 	
> 
> |Linux ||7.0.2|
> 
> 	
> 
> start parameter
> 
> 	
> 
> |-|
> 
> 	
> 
> fails
> 
> 	
> 
> BAD
> 
> git
> 
> 	
> 
> |Linux ||5.15|
> 
> 	
> 
> none
> 
> 	
> 
> 
> 	
> 
> 
> 	
> 
> not testable system freeze
> 
> git
> 
> 	
> 
> |Linux ||7.0.2|
> 
> 	
> 
> revert attempted
> 
> 	
> 
> 
> 	
> 
> 
> 	
> 
> not testable
> 
> 
> Kind regards,
> Matthias
> 
> Am 11.05.2026 um 11:34 schrieb Matthias Dittrich:
>>
>> Dear Bjorn,
>>
>> Thank you for the quick response. For transparency, I used the AI tool
>> Gemini during the bug-fixing and testing process. I am not an IT
>> specialist and therefore cannot fully verify all of the AI’s
>> conclusions in my last mail.
>>
>> All kernels were tested manually by installing the OS and kernel. The
>> recommended boot parameters were also tested manually. The logs sent
>> previously and in this mail are raw terminal outputs generated from
>> the commands listed in the "Skript" file.
>>
>>
>> The dmesg and |lspci|output from the newest CachyOS kernel is
>> attached. Additionally, the following commands were also run for the
>> Ubuntu 5.15 kernel for comparison:
>>
>> dmesg
>> sudo lspci -vv > lspci-vv-ubuntu2205-K515-good.txt
>> sudo lspci -vvv > lspci-vvv-ubuntu2205-K515-good.txt
>> sudo lspci -vvxxxx > lspci-vvxxxx-ubuntu2205-K515-good.txt
>> sudo lspci -s 02:00.0 -vvxxxx > lspci_gpu_$(uname -r).txt
>> sudo lspci -t > lspci_tree_$(uname -r).txt
>>
>>
>> Gemini recommended performing a git bisect with optimized parameters
>> to reduce compilation time.
>>
>>   * Before starting the bisect tests, I tested different kernel
>>     revisions. The "git" kernel 5.15 (with localmodconfig) appears to
>>     work just like the newer Ubuntu 5.15 kernel, as the graphics card
>>     status is shown as |(rev a1)|. The earliest kernel revision where
>>     the status changes to |(rev ff)|starts with kernel 5.16 (tested
>>     with localmodconfig). However, the NVIDIA driver itself could not
>>     be tested.
>>   * The bisect between 5.15 and 5.16 result in|
>>      [342a67f0884293639bd17ea44df754ead799e669]|| NFS: Label the
>>     dentry with a verifier in nfs_link(), nfs_symlink()|, which seems
>>     incorrect, as the verification using |git checkout
>>     342a67f0884293639bd17ea44df754ead799e669~1|also failed.
>>   * An automated bisect was attempted without localmodconfig, but
>>     there were technical issues and I have not yet been able to
>>     complete the testing.
>>
>> I also noticed that the |lspci|results differ between CachyOS and Ubuntu:
>>
>> 02:00.0 3D controller: NVIDIA Corporation GP108M [GeForce MX250] (rev a1)
>>
>>   * |(rev a1)|— Ubuntu 22.04 Kernel 5.15 (works), CachyOS Kernel 6.1
>>     LTS and Kernel 7, but NVIDIA MX150 does not work
>>   * |(rev ff)|— Ubuntu 22.04 Kernel 5.16 and newer (localmodconfig)
>>     and Kernel 6.8 and NVIDIA MX150 does not work
>>
>> Kind regards,
>> Matthias
>>
>>
>>
>> Am 01.05.2026 um 18:55 schrieb Bjorn Helgaas:
>>> [+cc Ilpo]
>>>
>>> On Fri, May 01, 2026 at 10:13:41AM +0200, Matthias Dittrich wrote:
>>>> Dear Kernel Developers,
>>>>
>>>> I found a regression and did my best to identify the issue and find a
>>>> workaround, but I am not a developer. I also published the regression in the
>>>> NVIDIA Linux Developer Forum, but it appears to be a kernel regression and
>>>> not a driver issue, as the latest drivers work on older kernels (e.g., 5.15)
>>>> but fail on anything 6.x+. I did not perform a git bisect as compiling on
>>>> this i5-8265U - because of its low performance I can only compile one kernel
>>>> over night.
>>>>
>>>> **Description:**
>>>> The NVIDIA MX250 GPU on Acer Swift 3 (SF314-56G / BIOS 1.14) is inaccessible
>>>> on any Linux distribution using Kernel 6.x or 7.x. The device fails to
>>>> initialize during the driver probe routine.
>>> Thanks for the report.  With the newest kernel available to you, can
>>> you please capture the complete dmesg and output of "sudo lspci -vv"?
>>> Don't filter out things that don't seem relevant; we can easily do
>>> that.
>>>
>>> Bjorn