shared_pool_alloc: zero out allocated DMA buffers

[eric135]

DMA buffers are expected to be zeroed out upon allocation

MSFT#53036590

[eric135]

Rebasing on newest main

[smalis-msft]

Just sanity checking, there's nothing else that implements this trait today?

[yupavlen-ms]

Today it is LockedMemorySpawner and SharedPoolAllocator. Tomorrow also EmulatedDmaAllocator for unit tests.

[smalis-msft]

Do they all follow the new contract?

[yupavlen-ms]

This change benefits normal boot and all clients are compatible. For Restore functionality there is alternative path to restore buffers without zeroiniting them, at least how it's implemented in #210

After we're going to merge with #260 then I'll test end-to-end if it works too.

[smalis-msft]

That's not what I asked though. We've added a new requirement to this method, we need to make sure all types that implement this trait are meeting the new requirement.

[mattkur]

All prod types do (LockedMemorySpawner gets it by way of MAP_ANONYMOUS). It seems that the test code does not?

https://github.com/microsoft/openvmm/blob/04454925c9fdad1845b4450743b739e16d8a8b4d/vm/devices/user_driver/src/emulated.rs#L244C1-L250C2

calls:

openvmm/vm/devices/user_driver/src/emulated.rs

Line 168 in 0445492

pub fn alloc(&self, len: usize) -> Option<DmaBuffer> {

[mattkur]

Filed #299