OvmfPkg/VirtioRngDxe: make VirtioRngGetRNG reentrant safe

[tpressure]

Description

This is a bugfix for the stuck boot issue that we experience from time to time. The reason was that VirtioRngGetRNG could be called in a nested way. The issue appears if a timer expires right when we are in VirtioFlush. If we get interrupted there, and the timer callback also calls VirtioRngGetRNG, we enter VirtioFlush a second time and all goes down the drain because the function does modify the virtqueue state.

See cloud-hypervisor/cloud-hypervisor#6779 for more details.

The solution is to raise the TPL to TPL_HIGH_LEVEL which is basically identical to interrupts being disabled. With this fix, we can no longer be interrupted when we fiddle around with the device and its virtqueues.

The callstack looks something like this

VirtioRngGetRNG (call 1)
└─ VirtioFlush
└─ gBS->Stall
└─ AcpiTimerLib::MicroSecondDelay
└─ Timer interrupt
└─ CoreSignalEvent
└─ CoreNotifyEvent
└─ [Notify callback]
└─ GetRandomNumberXX
└─ RngLib
└─ EFI_RNG_PROTOCOL
└─ VirtioRngGetRNG (call 2)

[phip1611]

Tremendous debugging, awesome outcome!