ArmVirtPkg: ArmVirtQemu: expose only 64-bit entry point for v3.0+ SMBIOS

The legacy 32-bit SMBIOS entry point has little use on AARCH64 systems,
since many such systems have no 32-bit addressable physical RAM, and so
OSes that implement SMBIOS will have to be able to deal with the 64-bit
entry point anyway.

Given that the OS will map main memory in 1 GB chunks if it can, and that
punching a page sized hole (e.g., for SMBIOS data) into it will result in
the whole 1 GB chunk being mapped using 2 MB and 4 KB blocks instead, it
is important to group memory reservations from the OS as much as we can,
and allocating below 4 GB for no good reason interferes with that.

This is especially important under virtualization, considering that each
*level* of lookup at stage 1 (the guest virtual page table) will result in
a full page table walk at stage 2 (the guest PA to host PA mapping).

So expose only the 64-bit entry point when the SMBIOS tables adhere to
version 3.0 or later.

Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>

ArmVirtPkg/ArmVirtQemu.dsc

@@ -169,6 +169,14 @@
   # support anything bigger, even if the host hardware does
   gEmbeddedTokenSpaceGuid.PcdPrePiCpuMemorySize|40
 
+  # Clearing BIT0 in this PCD prevents installing a 32-bit SMBIOS entry point,
+  # if the entry point version is >= 3.0. AARCH64 OSes cannot assume the
+  # presence of the 32-bit entry point anyway (because many AARCH64 systems
+  # don't have 32-bit addressable physical RAM), and the additional allocations
+  # below 4 GB needlessly fragment the memory map. So expose the 64-bit entry
+  # point only, for entry point versions >= 3.0.
+  gEfiMdeModulePkgTokenSpaceGuid.PcdSmbiosEntryPointProvideMethod|0x2
+
 [PcdsDynamicDefault.common]
   ## If TRUE, OvmfPkg/AcpiPlatformDxe will not wait for PCI
   #  enumeration to complete before installing ACPI tables.