[PATCH RFC v3 0/2] mm: Refactor KVM guest_memfd to introduce guestmem library

Elliot Berman posted 2 patches 1 week, 2 days ago
There is a newer version of this series
MAINTAINERS              |   2 +
include/linux/guestmem.h |  33 +++++++
mm/Kconfig               |   3 +
mm/Makefile              |   1 +
mm/guestmem.c            | 232 +++++++++++++++++++++++++++++++++++++++++++++++
virt/kvm/Kconfig         |   1 +
virt/kvm/guest_memfd.c   | 107 +++++++---------------
7 files changed, 305 insertions(+), 74 deletions(-)
[PATCH RFC v3 0/2] mm: Refactor KVM guest_memfd to introduce guestmem library
Posted by Elliot Berman 1 week, 2 days ago
In preparation for adding more features to KVM's guest_memfd, refactor
and introduce a library which abtracts some of the core-mm decisions
about managing folios associated with guest memory. The goal of the
refactor serves two purposes:

1. Provide an easier way to reason about memory in guest_memfd. KVM
   needs to support multiple confidentiality models (TDX, SEV, pKVM, Arm
   CCA). These models support different semantics for when the host
   can(not) access guest memory. An abstraction for the allocator and
   managing the state of pages will make it eaiser to reason about the
   state of folios within the guest_memfd.

2. Provide a common implementation for other users such as Gunyah [1] and
   guestmemfs [2].

In this initial series, I'm seeking comments for the line I'm drawing
between library and user (KVM). I've not introduced new functionality in
this series; the first new feature will probably be Fuad's mappability
patches [3].

I've decided to only bring out the address_space from guest_memfd as it
seemed the simplest approach. In the current iteration, KVM "attaches"
the guestmem to the inode. I expect we'll want to provide some helpers
for inode, file, and vm operations when it's relevant to
mappability/accessiblity/faultability.

I'd appreciate any feedback, especially on how much we should pull into
the guestmem library.

[1]: https://lore.kernel.org/lkml/20240222-gunyah-v17-0-1e9da6763d38@quicinc.com/
[2]: https://lore.kernel.org/all/20240805093245.889357-1-jgowans@amazon.com/
[3]: https://lore.kernel.org/all/20241010085930.1546800-3-tabba@google.com/

Changes in v3:
 - Refactor/extract only the address_space
 - Link to v2: https://lore.kernel.org/all/20240829-guest-memfd-lib-v2-0-b9afc1ff3656@quicinc.com/

Changes in v2:
- Significantly reworked to introduce "accessible" and "safe" reference
  counters
- Link to v1: https://lore.kernel.org/r/20240805-guest-memfd-lib-v1-0-e5a29a4ff5d7@quicinc.com

Signed-off-by: Elliot Berman <quic_eberman@quicinc.com>
---
Elliot Berman (2):
      KVM: guest_memfd: Convert .free_folio() to .release_folio()
      mm: guestmem: Convert address_space operations to guestmem library

 MAINTAINERS              |   2 +
 include/linux/guestmem.h |  33 +++++++
 mm/Kconfig               |   3 +
 mm/Makefile              |   1 +
 mm/guestmem.c            | 232 +++++++++++++++++++++++++++++++++++++++++++++++
 virt/kvm/Kconfig         |   1 +
 virt/kvm/guest_memfd.c   | 107 +++++++---------------
 7 files changed, 305 insertions(+), 74 deletions(-)
---
base-commit: 5cb1659f412041e4780f2e8ee49b2e03728a2ba6
change-id: 20241112-guestmem-library-68363cb29186

Best regards,
-- 
Elliot Berman <quic_eberman@quicinc.com>
Re: [PATCH RFC v3 0/2] mm: Refactor KVM guest_memfd to introduce guestmem library
Posted by David Hildenbrand 1 week ago
On 13.11.24 23:34, Elliot Berman wrote:
> In preparation for adding more features to KVM's guest_memfd, refactor
> and introduce a library which abtracts some of the core-mm decisions
> about managing folios associated with guest memory. The goal of the
> refactor serves two purposes:
> 
> 1. Provide an easier way to reason about memory in guest_memfd. KVM
>     needs to support multiple confidentiality models (TDX, SEV, pKVM, Arm
>     CCA). These models support different semantics for when the host
>     can(not) access guest memory. An abstraction for the allocator and
>     managing the state of pages will make it eaiser to reason about the
>     state of folios within the guest_memfd.
> 
> 2. Provide a common implementation for other users such as Gunyah [1] and
>     guestmemfs [2].
> 
> In this initial series, I'm seeking comments for the line I'm drawing
> between library and user (KVM). I've not introduced new functionality in
> this series; the first new feature will probably be Fuad's mappability
> patches [3].

Right, or the dummy mmap + vma->set_policy patches for NUMA handling.

-- 
Cheers,

David / dhildenb