5.18 up for testing 🧪

[anthraxx]

Hey folks,

I've just pushed the 5.18 testing branch. Please feel free to give it a try and come back with some feedback.

[hardfalcon]

5.18.x crashes on boot on two of my machines if I don't add either snd_hda_intel.snoop=1 or modprobe.blacklist=snd_hda_intel in the bootloader. The crash occurs also with a vanilla kernel 5.18 if I build the kernel using a hardened config. I'm currently bisecting this, and so far, I've been able to narrow it down to between commit 169e777 ("good", does not crash) and 3123109 ("bad", does crash). So while the bug is not present in linux 5.17, it was introduced somewhere before 5.18-rc1.

Interestingly, both affected machines use dedicated AMD GPUs, and for one of those machines, I also have an otherwise identical machine with a dedicated NVidia GPU that is not affected by the crashes. Still, booting with modprobe.backlist=amdgpu does not prevent the crashes. Also, one of the affected machines has an Intel Haswell-generation CPU with an integrated GPU (though it's not used and i915 does not get loaded AFAICT), whilst the other uses a Broadwell-generation Xeon without integrated GPU.

I'll keep you posted when I've found out more.

[hardfalcon]

Commit 9bf3fc5 is good, too.

[hardfalcon]

Narrowed it down to between dfdc1de (good/no crashes) and 74164d2 (bad/crashes).

[hardfalcon]

Narrowed it down to between 0a815d0 (good/no crashes) and 9ae2a14 (bad/crashes).

[hardfalcon]

Narrowed it down to 1930a6e (good/no crashes) and 9ae2a14 (bad(crashes).

However, I'm afraid I can't bisect further since all commits within the remaining range that I try seem to throw compiler errors… :/

[hardfalcon]

Narrowed it down to 37fcacb (good/no crashes) and 9ae2a14 (bad/crashes), any further bisecting results in compiler errors.

This is the config I've used for compiling the kernel (without the linux-hardened patchset applied!).

The relevant difference appear to be the following dmesg excerpts:
5.17.0-1-hardened-12883-g37fcacb50be7 (booted without snd_hda_intel.snoop parameter):

snd_hda_intel 0000:00:1b.0: enabling device (0000 -> 0002)
snd_hda_intel 0000:01:00.1: Force to non-snoop mode

5.17.0-1-hardened-12897-g9ae2a143081f (booted with snd_hda_intel.snoop=1 to prevent crash):

snd_hda_intel 0000:00:1b.0: enabling device (0000 -> 0002)
snd_hda_intel 0000:00:1b.0: Force to snoop mode by module option
snd_hda_intel 0000:01:00.1: Force to snoop mode by module option

Relevant lines from lspci -nn:

00:1b.0 Audio device [0403]: Intel Corporation 8 Series/C220 Series Chipset High Definition Audio Controller [8086:8c20] (rev 04)
01:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] Lexa PRO [Radeon 540/540X/550/550X / RX 540X/550/550X] [1002:699f] (rev c7)
01:00.1 Audio device [0403]: Advanced Micro Devices, Inc. [AMD/ATI] Baffin HDMI/DP Audio [Radeon RX 550 640SP / RX 560/560X] [1002:aae0]

[anthraxx]

Maybe it would be worth reporting this upstream, so far the scope we have reduced here should be relatively good for the subsystem maintainers to narrow down the actual issue.

As far as I see, this is the last blocker to officially release 5.18, however this isn't really strictly speaking a hardened blocker but a vanilla issue.

[hardfalcon]

Reported upstream:
https://bugzilla.kernel.org/show_bug.cgi?id=216112

[anthraxx]

Fixed by https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=9882d63bea14c8b3ed2c9360b9ab9f0e2f64ae2b