packaging/kernel: Add CONFIG_PCI_MMCONFIG to x86 guest kernel configuration #2311
Conversation
dgibson
added
wip
|
/test |
|
/test |
|
/test-vfio |
|
/retest-vfio |
dgibson
removed
the
wip
|
Rebased now that we're using ACPI hotplug, which should address the problems it had. |
|
/test |
|
Wait, I jumped the gun a little. This won't be ok until #1682 is actually merged. |
dgibson
added
the
wip
|
/retest |
dgibson
force-pushed
the
mmconfig
branch
2 times, most recently
from
4d80396
to
5acc24a
Compare
|
/test |
|
/retest-ubuntu-metrics |
|
/retest-fedora |
|
/retest-arm |
|
/retest-firecracker |
|
/retest-fc |
|
/cc @Jakob-Naucke (for s390x) and @jongwu (for ARM) |
|
They shall be fixed by #2736 |
|
/retest-arm /retest-s390x |
|
kata-containers/tests#4044 for tracking |
|
/test |
|
/retest-arm |
|
/test |
|
/retest-metrics |
|
/retest-ubuntu-metrics |
|
@marcel-apf now that you're back, could you have a look at this one. Now that the rescan removal is merged, this is no longer blocked. |
|
@Jakob-Naucke IIUC you were saying earlier that the ARM & s390x CI failures should be fixed by kata-containers/tests#4044. That's now closed, but the failures remain. Any ideas? |
|
/test |
|
/retest-centos |
|
/retest-ubuntu-metrics |
|
/retest-centos |
|
|
/retest-arm |
|
/retest-arm |
…ration The guest kernel configuration suggested for Kata, and which is used by the CI didn't include CONFIG_PCI_MMCONFIG. That's kind of weird, MMCONFIG is the modern normal way of handling configuration cycles. In addition, due to a complex set of interactions through the ACPI code, disabling MMCONFIG means that SHPC hotplug doesn't work: the driver is included in the guest kernel, but will fail to probe on PCI to PCI bridges, meaning it won't actually be activated. Enable MMCONFIG so that we suggest and testa more typical guest kernel configuration. fixes kata-containers#2288 Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
|
/test |
|
/retest-ubuntu-metrics |
|
Thanks @dgibson ! |
The guest kernel configuration suggested for Kata, and which is used by the
CI didn't include CONFIG_PCI_MMCONFIG. That's kind of weird, MMCONFIG is
the modern normal way of handling configuration cycles.
In addition, due to a complex set of interactions through the ACPI code,
disabling MMCONFIG means that SHPC hotplug doesn't work: the driver is
included in the guest kernel, but will fail to probe on PCI to PCI bridges,
meaning it won't actually be activated.
Enable MMCONFIG so that we suggest and testa more typical guest kernel
configuration.
fixes #2288
Signed-off-by: David Gibson david@gibson.dropbear.id.au