Public bug reported:
When a VM with device passthrough is launched by an unprivileged process
(e.g., Libvirt launches qemu process as an unprivileged user), IOMMUFD's
unified memory accounting causes ENOMEM errors for large device BARs
that exceed the process memlock limit. VFIO without IOMMUFD does not
check device memory against the limit and therefore does not show the
same behavior.
Device memory regions such as I/O device BARs do not consume host RAM
and should not count against the RLIMIT_MEMLOCK limit for how much
memory can be locked/pinned.
** Affects: linux-nvidia (Ubuntu)
Importance: Undecided
Status: New
** Affects: linux-nvidia-6.14 (Ubuntu)
Importance: Undecided
Status: New
** Affects: linux-nvidia-6.17 (Ubuntu)
Importance: Undecided
Status: New
** Also affects: linux-nvidia-6.14 (Ubuntu)
Importance: Undecided
Status: New
** Also affects: linux-nvidia (Ubuntu)
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2131970
Title:
VM BAR mapping ENOMEM from hitting RLIMIT_MEMLOCK limit when using
iommufd
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-nvidia/+bug/2131970/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
