Re: [RFC PATCH v2 29/51] mm: guestmem_hugetlb: Wrap HugeTLB as an allocator for guest_memfd

[Sean Christopherson <seanjc@google.com>]

On Wed, May 14, 2025, Ackerley Tng wrote:
> guestmem_hugetlb is an allocator for guest_memfd. It wraps HugeTLB to
> provide huge folios for guest_memfd.
> 
> This patch also introduces guestmem_allocator_operations as a set of
> operations that allocators for guest_memfd can provide. In a later
> patch, guest_memfd will use these operations to manage pages from an
> allocator.
> 
> The allocator operations are memory-management specific and are placed
> in mm/ so key mm-specific functions do not have to be exposed
> unnecessarily.

This code doesn't have to be put in mm/, all of the #includes are to <linux/xxx.h>.
Unless I'm missing something, what you actually want to avoid is _exporting_ mm/
APIs, and for that all that is needed is ensure the code is built-in to the kernel
binary, not to kvm.ko.

diff --git a/virt/kvm/Makefile.kvm b/virt/kvm/Makefile.kvm
index d047d4cf58c9..c18c77e8a638 100644
--- a/virt/kvm/Makefile.kvm
+++ b/virt/kvm/Makefile.kvm
@@ -13,3 +13,5 @@ kvm-$(CONFIG_HAVE_KVM_IRQ_ROUTING) += $(KVM)/irqchip.o
 kvm-$(CONFIG_HAVE_KVM_DIRTY_RING) += $(KVM)/dirty_ring.o
 kvm-$(CONFIG_HAVE_KVM_PFNCACHE) += $(KVM)/pfncache.o
 kvm-$(CONFIG_KVM_GUEST_MEMFD) += $(KVM)/guest_memfd.o
+
+obj-$(subst m,y,$(CONFIG_KVM_GUEST_MEMFD)) += $(KVM)/guest_memfd_hugepages.o
\ No newline at end of file

People may want the code to live in mm/ for maintenance and ownership reasons
(or not, I haven't followed the discussions on hugepage support), but that's a
very different justification than what's described in the changelog.

And if the _only_ user is guest_memfd, putting this in mm/ feels quite weird.
And if we anticipate other users, the name guestmem_hugetlb is weird, because
AFAICT there's nothing in here that is in any way guest specific, it's just a
few APIs for allocating and accounting hugepages.

Personally, I don't see much point in trying to make this a "generic" library,
in quotes because the whole guestmem_xxx namespace makes it anything but generic.
I don't see anything in mm/guestmem_hugetlb.c that makes me go "ooh, that's nasty,
I'm glad this is handled by a library".  But if we want to go straight to a
library, it should be something that is really truly generic, i.e. not "guest"
specific in any way.

> Signed-off-by: Ackerley Tng <ackerleytng@google.com>
> 
> Change-Id: I3cafe111ea7b3c84755d7112ff8f8c541c11136d
> ---
>  include/linux/guestmem.h      |  20 +++++
>  include/uapi/linux/guestmem.h |  29 +++++++
>  mm/Kconfig                    |   5 +-
>  mm/guestmem_hugetlb.c         | 159 ++++++++++++++++++++++++++++++++++
>  4 files changed, 212 insertions(+), 1 deletion(-)
>  create mode 100644 include/linux/guestmem.h
>  create mode 100644 include/uapi/linux/guestmem.h


..

> diff --git a/include/uapi/linux/guestmem.h b/include/uapi/linux/guestmem.h
> new file mode 100644
> index 000000000000..2e518682edd5
> --- /dev/null
> +++ b/include/uapi/linux/guestmem.h

With my KVM hat on, NAK to defining uAPI in a library like this.  This subtly
defines uAPI for KVM, and effectively any other userspace-facing entity that
utilizes the library/allocator.  KVM's uAPI needs to be defined by KVM, period.

There's absolutely zero reason to have guestmem_hugetlb_setup() take in flags.
Explicitly pass the page size, or if preferred, the page_size_log, and let the
caller figure out how to communicate the size to the kernel.

IMO, the whole MAP_HUGE_xxx approach is a (clever) hack to squeeze the desired
size into mmap() flags.  I don't see any reason to carry that forward to guest_memfd.
For once, we had the foresight to reserve some space in KVM's uAPI structure, so
there's no need to squeeze things into flags.

E.g. we could do something like this:

diff --git include/uapi/linux/kvm.h include/uapi/linux/kvm.h
index 42053036d38d..b79914472d27 100644
--- include/uapi/linux/kvm.h
+++ include/uapi/linux/kvm.h
@@ -1605,11 +1605,16 @@ struct kvm_memory_attributes {
 #define KVM_CREATE_GUEST_MEMFD _IOWR(KVMIO,  0xd4, struct kvm_create_guest_memfd)
 #define GUEST_MEMFD_FLAG_MMAP          (1ULL << 0)
 #define GUEST_MEMFD_FLAG_INIT_SHARED   (1ULL << 1)
+#define GUEST_MEMFD_FLAG_HUGE_PAGES    (1ULL << 2)
 
 struct kvm_create_guest_memfd {
        __u64 size;
        __u64 flags;
-       __u64 reserved[6];
+       __u8 huge_page_size_log2;
+       __u8 reserve8;
+       __u16 reserve16;
+       __u32 reserve32;
+       __u64 reserved[5];
 };
 
 #define KVM_PRE_FAULT_MEMORY   _IOWR(KVMIO, 0xd5, struct kvm_pre_fault_memory)

And not have to burn 6 bits of flags to encode the size in a weird location.

But that's a detail for KVM to sort out, which is exactly my point; how this is
presented to userspace for guest_memfd is question for KVM.

> @@ -0,0 +1,29 @@
> +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
> +#ifndef _UAPI_LINUX_GUESTMEM_H
> +#define _UAPI_LINUX_GUESTMEM_H
> +
> +/*
> + * Huge page size must be explicitly defined when using the guestmem_hugetlb
> + * allocator for guest_memfd.  It is the responsibility of the application to
> + * know which sizes are supported on the running system.  See mmap(2) man page
> + * for details.
> + */
> +
> +#define GUESTMEM_HUGETLB_FLAG_SHIFT	58
> +#define GUESTMEM_HUGETLB_FLAG_MASK	0x3fUL
> +
> +#define GUESTMEM_HUGETLB_FLAG_16KB	(14UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_64KB	(16UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_512KB	(19UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_1MB	(20UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_2MB	(21UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_8MB	(23UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_16MB	(24UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_32MB	(25UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_256MB	(28UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_512MB	(29UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_1GB	(30UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_2GB	(31UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +#define GUESTMEM_HUGETLB_FLAG_16GB	(34UL << GUESTMEM_HUGETLB_FLAG_SHIFT)
> +
> +#endif /* _UAPI_LINUX_GUESTMEM_H */

...

> +const struct guestmem_allocator_operations guestmem_hugetlb_ops = {
> +	.inode_setup = guestmem_hugetlb_setup,
> +	.inode_teardown = guestmem_hugetlb_teardown,
> +	.alloc_folio = guestmem_hugetlb_alloc_folio,
> +	.nr_pages_in_folio = guestmem_hugetlb_nr_pages_in_folio,
> +};
> +EXPORT_SYMBOL_GPL(guestmem_hugetlb_ops);

Why are these bundled into a structure?  AFAICT, that adds layers of indirection
for absolutely no reason.  And especially on the KVM guest_memfd side, implementing
a pile of infrastructure to support "custom" allocators is very premature.  Without
a second "custom" allocator, it's impossible to determine if the indirection
provided is actually a good design.  I.e. all of the kvm_gmem_has_custom_allocator()
logic in guest_memfd.c is just HugeTLB logic buried behind a layer of unnecessary
indirection.