[PATCH makedumpfile] Add confidential VM unaccepted free pages support on Linux 6.16 and later
YAMAZAKI MASAMITSU(山崎 真光)
yamazaki-msmt at nec.com
Mon Jun 30 03:15:00 PDT 2025
>> UEFI Specification version 2.9 introduces the 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 Virtual
>> Machine platform[1].
>>
>> Before unaccepted memory is accepted by guest, any access from guest
>> will result in the guest failed, as well as the kexec'ed kernel. So,
>> the kexec'ed kernel will skip these pages and fill in zero data for the
>> reader of vmcore.
>>
>> However, it introduces a problem. When exclude a page filled with zero,
>> a pd_zero, which sizeof(page_desc_t) is 24 bytes, will be written into
>> cd_header and this part is not compressed by design. As the unaccepted
>> pages are exported as zero pages, which means ~1/170 of the capacity of
>> unaccepted memory will be added into vmcore additionally. In fact, they
>> should be considered as free pages and most of the time should be
>> excluded.
>>
>> Unaccepted memory is unusable free memory, so they are not managed by
>> buddy, instead, they are added to a new list zone.unaccepted_pages only
>> when the order is MAX_PAGE_ORDER each time conventionally. The new
>> page type, PGTY_unaccepted can be used to identify whether a page is
>> unaccepted[2]. Therefore, add following changes to exclude them like
>> free pages:
>>
>> 1. Add NUMBER(PAGE_UNACCEPTED_MAPCOUNT_VALUE) to identify a page is
>> unaccepted, a kernel patch[3] to export the value of page type
>> PAGE_UNACCEPTED_MAPCOUNT_VALUE since kernel 6.16.
>>
>> 2. Add a condition to exclude these unaccepted free pages.
>>
>> Dumping host kernel will not be impacted by the modification, because it
>> cannot enable CONFIG_UNACCEPTED_MEMORY, so the page type
>> PAGE_UNACCEPTED_MAPCOUNT_VALUE cannot be found in vmcoreinfo and skip
>> the step.
>>
>> Here is a vmcore size statistic of a freshly booted TD VM with different
>> memory sizes:
>>
>> VM.mem | Before After
>> -------+----------------
>> 512G | ~4.9G ~2.0G
>> 256G | ~2.0G ~1.1G
> Kindly ping.
I apologize for keeping you waiting.
This patch look likes right to me, so: Ack.
Thanks,Masa
More information about the kexec
mailing list