Re: [PATCH v4 0/3] cgroup: New misc cgroup controller

[Xingyou Chen]

在 2021/3/30 12:42, Vipin Sharma 写道:
> Hello,
>
> This patch series is creating a new misc cgroup controller for limiting
> and tracking of resources which are not abstract like other cgroup
> controllers.
>
> This controller was initially proposed as encryption_id but after the
> feedbacks and use cases for other resources, it is now changed to misc
> cgroup.
> https://lore.kernel.org/lkml/20210108012846.4134815-2-vipinsh@xxxxxxxxxx/
>
> Most of the cloud infrastructure use cgroups for knowing the host state,
> track the resources usage, enforce limits on them, etc. They use this
> info to optimize work allocation in the fleet and make sure no rogue job
> consumes more than it needs and starves others.
>
> There are resources on a system which are not abstract enough like other
> cgroup controllers and are available in a limited quantity on a host.
>
> One of them is Secure Encrypted Virtualization (SEV) ASID on AMD CPU.
> SEV ASIDs are used for creating encrypted VMs. SEV is mostly be used by
> the cloud providers for providing confidential VMs. Since SEV ASIDs are
> limited, there is a need to schedule encrypted VMs in a cloud
> infrastructure based on SEV ASIDs availability and also to limit its
> usage.
>
> There are similar requirements for other resource types like TDX keys,
> IOASIDs and SEID.
>
> Adding these resources to a cgroup controller is a natural choice with
> least amount of friction. Cgroup itself says it is a mechanism to
> distribute system resources along the hierarchy in a controlled
> mechanism and configurable manner. Most of the resources in cgroups are
> abstracted enough but there are still some resources which are not
> abstract but have limited availability or have specific use cases.
>
> Misc controller is a generic controller which can be used by these
> kinds of resources.

Will we make this dynamic? Let resources be registered via something 
like misc_cg_res_{register,unregister}, at compile time or runtime, 
instead of hard coded into misc_res_name/misc_res_capacity etc.

There are needs as noted in drmcg session earlier this year. We may
make misc cgroup stable, and let device drivers to register their
own resources.

This may make misc cgroup controller more complex than expected, but
simpler than adding multiple similar controllers.

> One suggestion was to use BPF for this purpose, however, there are
> couple of things which might not be addressed with BPF:
> 1. Which controller to use in v1 case? These are not abstract resources
>     so in v1 where each controller have their own hierarchy it might not
>     be easy to identify the best controller to use for BPF.
>
> 2. Abstracting out a single BPF program which can help with all of the
>     resources types might not be possible, because resources we are
>     working with are not similar and abstract enough, for example network
>     packets, and there will be different places in the source code to use
>     these resources.
>
> A new cgroup controller tends to give much easier and well integrated
> solution when it comes to scheduling and limiting a resource with
> existing tools in a cloud infrastructure.
>
> Changes in RFC v4:
> 1. Misc controller patch is split into two patches. One for generic misc
>     controller and second for adding SEV and SEV-ES resource.
> 2. Using READ_ONCE and WRITE_ONCE for variable accesses.
> 3. Updated documentation.
> 4. Changed EXPORT_SYMBOL to EXPORT_SYMBOL_GPL.
> 5. Included cgroup header in misc_cgroup.h.
> 6. misc_cg_reduce_charge changed to misc_cg_cancel_charge.
> 7. misc_cg set to NULL after uncharge.
> 8. Added WARN_ON if misc_cg not NULL before charging in SEV/SEV-ES.
>
> Changes in RFC v3:
> 1. Changed implementation to support 64 bit counters.
> 2. Print kernel logs only once per resource per cgroup.
> 3. Capacity can be set less than the current usage.
>
> Changes in RFC v2:
> 1. Documentation fixes.
> 2. Added kernel log messages.
> 3. Changed charge API to treat misc_cg as input parameter.
> 4. Added helper APIs to get and release references on the cgroup.
>
> [1] https://lore.kernel.org/lkml/20210218195549.1696769-1-vipinsh@xxxxxxxxxx
> [2] https://lore.kernel.org/lkml/20210302081705.1990283-1-vipinsh@xxxxxxxxxx/
> [3] https://lore.kernel.org/lkml/20210304231946.2766648-1-vipinsh@xxxxxxxxxx/
>
> Vipin Sharma (3):
>    cgroup: Add misc cgroup controller
>    cgroup: Miscellaneous cgroup documentation.
>    svm/sev: Register SEV and SEV-ES ASIDs to the misc controller
>
>   Documentation/admin-guide/cgroup-v1/index.rst |   1 +
>   Documentation/admin-guide/cgroup-v1/misc.rst  |   4 +
>   Documentation/admin-guide/cgroup-v2.rst       |  73 +++-
>   arch/x86/kvm/svm/sev.c                        |  70 ++-
>   arch/x86/kvm/svm/svm.h                        |   1 +
>   include/linux/cgroup_subsys.h                 |   4 +
>   include/linux/misc_cgroup.h                   | 132 ++++++
>   init/Kconfig                                  |  14 +
>   kernel/cgroup/Makefile                        |   1 +
>   kernel/cgroup/misc.c                          | 407 ++++++++++++++++++
>   10 files changed, 695 insertions(+), 12 deletions(-)
>   create mode 100644 Documentation/admin-guide/cgroup-v1/misc.rst
>   create mode 100644 include/linux/misc_cgroup.h
>   create mode 100644 kernel/cgroup/misc.c