On 10.08.21 17:02, Kirill A. Shutemov wrote:
On Tue, Aug 10, 2021 at 09:48:04AM +0200, David Hildenbrand wrote:
On 10.08.21 08:26, Kirill A. Shutemov wrote:
UEFI Specification version 2.9 introduces concept of memory acceptance:
Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP,
requiring memory to be accepted before it can be used by the guest.
Accepting happens via a protocol specific for the Virtrual Machine
platform.
Accepting memory is costly and it makes VMM allocate memory for the
accepted guest physical address range. It's better to postpone memory
acceptation until memory is needed. It lowers boot time and reduces
memory overhead.
Support of such memory requires few changes in core-mm code:
- memblock has to accept memory on allocation;
- page allocator has to accept memory on the first allocation of the
page;
Memblock change is trivial.
Page allocator is modified to accept pages on the first allocation.
PageOffline() is used to indicate that the page requires acceptance.
The flag currently used by hotplug and balloon. Such pages are not
available to page allocator.
An architecture has to provide three helpers if it wants to support
unaccepted memory:
- accept_memory() makes a range of physical addresses accepted.
- maybe_set_page_offline() marks a page PageOffline() if it requires
acceptance. Used during boot to put pages on free lists.
- clear_page_offline() clears makes a page accepted and clears
PageOffline().
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
---
mm/internal.h | 14 ++++++++++++++
mm/memblock.c | 1 +
mm/page_alloc.c | 13 ++++++++++++-
3 files changed, 27 insertions(+), 1 deletion(-)
diff --git a/mm/internal.h b/mm/internal.h
index 31ff935b2547..d2fc8a17fbe0 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -662,4 +662,18 @@ void vunmap_range_noflush(unsigned long start, unsigned long end);
int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
unsigned long addr, int page_nid, int *flags);
+#ifndef CONFIG_UNACCEPTED_MEMORY
+static inline void maybe_set_page_offline(struct page *page, unsigned int order)
+{
+}
+
+static inline void clear_page_offline(struct page *page, unsigned int order)
+{
+}
+
+static inline void accept_memory(phys_addr_t start, phys_addr_t end)
+{
+}
Can we find better fitting names for the first two? The function names are
way too generic. For example:
accept_or_set_page_offline()
accept_and_clear_page_offline()
Sounds good.
I thought for a second if
PAGE_TYPE_OPS(Unaccepted, offline)
makes sense as well, not sure.
I find Offline fitting the situation. Don't see a reason to add more
terminology here.
Also, please update the description of PageOffline in page-flags.h to
include the additional usage with PageBuddy set at the same time.
Okay.
I assume you don't have to worry about page_offline_freeze/thaw ... as we
only set PageOffline initially, but not later at runtime when other
subsystems (/proc/kcore) might stumble over it.
I think so, but I would need to look at this code once again.
Another thing to look into would be teaching makedumpfile via vmcoreinfo
about these special buddy pages:
makedumpfile will naturally skip all PageOffline pages and skip
PageBuddy pages if requested to skip free pages. It detects these pages
via the mapcount value. You will want makedumpfile to treat them like
PageOffline pages: kernel/crash_core.c
#define PAGE_BUDDY_MAPCOUNT_VALUE (~PG_buddy)
VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
#define PAGE_OFFLINE_MAPCOUNT_VALUE (~PG_offline)
VMCOREINFO_NUMBER(PAGE_OFFLINE_MAPCOUNT_VALUE);
We could export PAGE_BUDDY_OFFLINE_MAPCOUNT_VALUE or just compute it
inside makedumpfile from the other two values.
--
Thanks,
David / dhildenb
