Re: [PATCH] ArmVirtPkg/ArmVirtQemu: use non-accelerated CopyMem for VariableRuntimeDxe

[Shannon Zhao <zhaoshenglong@huawei.com>]

On 2017/11/15 22:03, Ard Biesheuvel wrote:
> On 15 November 2017 at 13:51, Laszlo Ersek <lersek@redhat.com> wrote:
>> > On 11/14/17 11:22, Ard Biesheuvel wrote:
>>> >> The VariableRuntimeDxe driver may use CopyMem () on NOR flash regions,
>>> >> assuming such regions always have full memory semantics. Given that
>>> >> those regions cannot be mapped as ordinary memory on ARM (due to the
>>> >> fact that the NOR flash requires device semantics while in write mode)
>>> >> this prevents us from using BaseMemoryLibOptDxe in VariableRuntimeDxe,
>>> >> since it may use unaligned accesses and/or DC ZVA instructions, both
>>> >> of which are incompatible with mappings using device semantics.
>>> >>
>>> >> Note that there is no way we can work around this by changing the
>>> >> mapping type between 'memory' and 'device' when switching from read to
>>> >> write mode and back, because the runtime mapping is created by the OS,
>>> >> and cannot be changed at will.
>>> >>
>>> >> So let's just switch to the unaccelerated version of BaseMemoryLib which
>>> >> does not have the same problem.
>>> >>
>>> >> Contributed-under: TianoCore Contribution Agreement 1.1
>>> >> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
>>> >> ---
>>> >>  ArmVirtPkg/ArmVirtQemu.dsc       | 2 ++
>>> >>  ArmVirtPkg/ArmVirtQemuKernel.dsc | 2 ++
>>> >>  2 files changed, 4 insertions(+)
>>> >>
>>> >> diff --git a/ArmVirtPkg/ArmVirtQemu.dsc b/ArmVirtPkg/ArmVirtQemu.dsc
>>> >> index 8a60b61f2aa6..7b220d6e3c31 100644
>>> >> --- a/ArmVirtPkg/ArmVirtQemu.dsc
>>> >> +++ b/ArmVirtPkg/ArmVirtQemu.dsc
>>> >> @@ -261,6 +261,8 @@ [Components.common]
>>> >>    MdeModulePkg/Universal/Variable/RuntimeDxe/VariableRuntimeDxe.inf {
>>> >>      <LibraryClasses>
>>> >>        NULL|MdeModulePkg/Library/VarCheckUefiLib/VarCheckUefiLib.inf
>>> >> +      # don't use unaligned CopyMem () on the UEFI varstore NOR flash region
>>> >> +      BaseMemoryLib|MdePkg/Library/BaseMemoryLib/BaseMemoryLib.inf
>>> >>    }
>>> >>  !if $(SECURE_BOOT_ENABLE) == TRUE
>>> >>    MdeModulePkg/Universal/SecurityStubDxe/SecurityStubDxe.inf {
>>> >> diff --git a/ArmVirtPkg/ArmVirtQemuKernel.dsc b/ArmVirtPkg/ArmVirtQemuKernel.dsc
>>> >> index 9a31ec93ca06..7c032e1b07e0 100644
>>> >> --- a/ArmVirtPkg/ArmVirtQemuKernel.dsc
>>> >> +++ b/ArmVirtPkg/ArmVirtQemuKernel.dsc
>>> >> @@ -252,6 +252,8 @@ [Components.common]
>>> >>    MdeModulePkg/Universal/Variable/RuntimeDxe/VariableRuntimeDxe.inf {
>>> >>      <LibraryClasses>
>>> >>        NULL|MdeModulePkg/Library/VarCheckUefiLib/VarCheckUefiLib.inf
>>> >> +      # don't use unaligned CopyMem () on the UEFI varstore NOR flash region
>>> >> +      BaseMemoryLib|MdePkg/Library/BaseMemoryLib/BaseMemoryLib.inf
>>> >>    }
>>> >>  !if $(SECURE_BOOT_ENABLE) == TRUE
>>> >>    MdeModulePkg/Universal/SecurityStubDxe/SecurityStubDxe.inf {
>>> >>
>> >
>> > Reviewed-by: Laszlo Ersek <lersek@redhat.com>
>> >
>> > Given that we've never seen the symptom reported by Shannon, I must
>> > think Shannon is using some kind of new hardware. May I ask what
>> > hardware that is?
>> >
> I think this is equally likely to occur in any KVM hardware
> virtualization context. It is simply a regression after moving to the
> OptDxe BaseMemoryLib flavor.

Exactly. I'm using Huawei D05. It works well when I use a older edk2
while fails when updating to UDK2017 branch.

Tested-by: Shannon Zhao <zhaoshenglong@huawei.com>

Thanks,
-- 
Shannon