Am 04.10.26 um 13:24 schrieb Ben Hutchings:
Hi Ben,
On Sat, 2026-10-03 at 23:59 +0200, Bernhard Schmidt wrote:
[...]
I still don't get why my bisection attempt failed so badly,
6.12.111-stable (the one compiled from the kernel.org stable tree
following your bisection instructions) has been running for 36+ hours
before I voluntarily rebooted into your 6.12.112 kernel.
I believe this is because Debian's config sets a default mode for the
Intel IOMMU driver which is not present upstream.
When building from the upstream source, make sure you have
CONFIG_INTEL_IOMMU_DEFAULT_ON=y in the .config file, or add
"intel_iommu=on" to the kernel boot parameters.
Hrm, possible. I do have IOMMU messages on boot
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 12 under DRHD base
0xc5ffc000 IOMMU 6
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 11 under DRHD base
0xb87fc000 IOMMU 5
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 10 under DRHD base
0xaaffc000 IOMMU 4
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 18 under DRHD base
0xfbffc000 IOMMU 3
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 17 under DRHD base
0xee7fc000 IOMMU 2
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 16 under DRHD base
0xe0ffc000 IOMMU 1
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 15 under DRHD base
0xd37fc000 IOMMU 0
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 8 under DRHD base
0x9d7fc000 IOMMU 7
Oct 03 23:54:05 nm3 kernel: DMAR-IR: IOAPIC id 9 under DRHD base
0x9d7fc000 IOMMU 7
Oct 03 23:54:06 nm3 kernel: iommu: Default domain type: Translated
Oct 03 23:54:06 nm3 kernel: iommu: DMA domain TLB invalidation policy:
lazy mode
But the stock Debian kernel lists a bit more output when grepping for IOMMU
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 12 under DRHD base
0xc5ffc000 IOMMU 6
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 11 under DRHD base
0xb87fc000 IOMMU 5
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 10 under DRHD base
0xaaffc000 IOMMU 4
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 18 under DRHD base
0xfbffc000 IOMMU 3
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 17 under DRHD base
0xee7fc000 IOMMU 2
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 16 under DRHD base
0xe0ffc000 IOMMU 1
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 15 under DRHD base
0xd37fc000 IOMMU 0
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 8 under DRHD base
0x9d7fc000 IOMMU 7
Oct 03 22:43:51 nm3 kernel: DMAR-IR: IOAPIC id 9 under DRHD base
0x9d7fc000 IOMMU 7
Oct 03 22:43:51 nm3 kernel: iommu: Default domain type: Translated
Oct 03 22:43:51 nm3 kernel: iommu: DMA domain TLB invalidation policy:
lazy mode
Oct 03 22:43:51 nm3 kernel: pci 0000:17:00.0: Adding to iommu group 0
Oct 03 22:43:51 nm3 kernel: pci 0000:17:02.0: Adding to iommu group 1
Oct 03 22:43:51 nm3 kernel: pci 0000:18:00.0: Adding to iommu group 2
Oct 03 22:43:51 nm3 kernel: pci 0000:19:00.0: Adding to iommu group 3
Oct 03 22:43:51 nm3 kernel: pci 0000:19:00.1: Adding to iommu group 3
Oct 03 22:43:51 nm3 kernel: pci 0000:00:00.0: Adding to iommu group 4
[...]
Oct 03 22:43:51 nm3 kernel: pci 0000:d7:12.1: Adding to iommu group 125
Oct 03 22:43:51 nm3 kernel: pci 0000:d7:12.2: Adding to iommu group 125
Oct 03 22:43:51 nm3 kernel: pci 0000:d7:15.0: Adding to iommu group 126
Oct 03 22:43:51 nm3 kernel: pci 0000:d7:16.0: Adding to iommu group 127
Oct 03 22:43:51 nm3 kernel: pci 0000:d7:16.4: Adding to iommu group 127
What I don't get though, I followed Salvatore's instructions from a
stock Debian kernel which say
cp /boot/config-$(uname -r) .config
yes '' | make localmodconfig
make savedefconfig
mv defconfig arch/x86/configs/my_defconfig
so all settings should be the same. But as I wrote before, the kernel
sizes are wildly different, so possibly something went wrong here.
-rw-r--r-- 1 root root 283542 Oct 3 15:02 config-6.12+unreleased-amd64
-rw-r--r-- 1 root root 171116 Oct 2 01:00 config-6.12.107
-rw-r--r-- 1 root root 283337 Aug 29 06:19 config-6.12.107+deb13-amd64
-rw-r--r-- 1 root root 171310 Oct 2 01:05 config-6.12.111
-rw-r--r-- 1 root root 283542 Sep 28 14:18 config-6.12.111+deb13-amd64
root@nm3:/boot# grep CONFIG_INTEL_IOMMU_DEFAULT_ON config-6.12*
config-6.12+unreleased-amd64:# CONFIG_INTEL_IOMMU_DEFAULT_ON is not set
config-6.12+unreleased-amd64:CONFIG_INTEL_IOMMU_DEFAULT_ON_INTGPU_OFF=y
config-6.12.107:# CONFIG_INTEL_IOMMU_DEFAULT_ON is not set
config-6.12.107+deb13-amd64:# CONFIG_INTEL_IOMMU_DEFAULT_ON is not set
config-6.12.107+deb13-amd64:CONFIG_INTEL_IOMMU_DEFAULT_ON_INTGPU_OFF=y
config-6.12.111:# CONFIG_INTEL_IOMMU_DEFAULT_ON is not set
config-6.12.111+deb13-amd64:# CONFIG_INTEL_IOMMU_DEFAULT_ON is not set
config-6.12.111+deb13-amd64:CONFIG_INTEL_IOMMU_DEFAULT_ON_INTGPU_OFF=y
That setting is not set for both my kernels, the trixie kernels and the
one Salvatore sent out yesterday. So maybe it's enabled somewhere else
in Debian (a kernel patch)?
Anyway, I booted "my" .111 with intel_iommu=on now and it shows the
"Adding to iommu group" messages, so it should crash now.
Bernhard
