(commenting once more, to explain the actual reason:)
On 2/21/24 11:24, Gerd Hoffmann wrote:
> On Wed, Feb 21, 2024 at 03:48:25AM +0000, Ni, Ray wrote:
>>>
>>> + MaxCpusPerHob = (MAX_UINT16 - sizeof (EFI_HOB_GUID_TYPE) - sizeof
>>> (MP_HAND_OFF)) / sizeof (PROCESSOR_HAND_OFF);
>>
>> Above formula assumes the maximum HOB length could be 0xFFFF.
>
> Which is IMHO correct.
It is not correct:
>
>> But actually the maximum HOB length could be only 0xFFF8 because
>> PeiCore::PeiCreateHob() contains following logic:
>>
>> if (0x10000 - Length <= 0x7) {
>> return EFI_INVALID_PARAMETER;
>> }
>
> That Length is the *data* size, the HOB header is not included.
>
> The "- sizeof (EFI_HOB_GUID_TYPE)" in the formula above accounts the
> space needed for HOB header and GUID.
Yes, but the problem is not that we need the extra space for
EFI_HOB_GUID_TYPE (24 bytes) -- we need the extra space for padding /
alignment.
This is what the PI spec says (v1.8, section "III-4.5.2 HOB Construction
Rules"):
HOB construction must obey the following rules:
[...]
4. All HOBs must be multiples of 8 bytes in length. This requirement
meets the alignment restrictions of the Itanium® processor family.
[...]
And if your total *desired* HOB length exceeds 0xFFF8 (i.e., it falls in
[0xFFF9..0xFFFF]), then the required upwards rounding produces 0x1_0000.
But that rounded-up value cannot be expressed in
"EFI_HOB_GENERIC_HEADER.HobLength" -- a UINT16 field.
In short, you've got Itanium to thank for this. :)
Laszlo
-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#115768): https://edk2.groups.io/g/devel/message/115768
Mute This Topic: https://groups.io/mt/104472313/21656
Group Owner: devel+ow...@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
