feat: MaybeUnint sockets to save space by putting sockets in .bss

[sigvartmh]

Proposal

So this may be a bit more controversial but sockets currently don't get the 00 bit patteren meaning they can't be in .bss on embedded targets. Maybe there is a better way of not initializing them but this is the solution I found. Feel free to deny or come with feed back on it. This also adds unsafe to the code which may not be wanted.

Commit message:

Initialise a slice of uninitialised socket storage in place, returning the initialised slice.

This exists to keep socket storage out of .data on bare-metal targets.

SocketStorage is a niche-filled Option<Item>: None occupies zero bytes of payload and its bit pattern is chosen by the compiler, so EMPTY is not all-zero. A

static mut STORAGE: [SocketStorage; N] = [SocketStorage::EMPTY; N];

therefore cannot live in .bss; it goes in .data and the linker emits an initialiser image of N * size_of::<SocketStorage>() bytes into flash, almost all of which is zeros. Declaring the storage MaybeUninit instead places it in .bss, and this function fills it at startup:

static mut STORAGE: MaybeUninit<[SocketStorage; N]> = MaybeUninit::uninit();
let sockets = SocketSet::new(SocketStorage::init_slice(unsafe {
    &mut *(core::ptr::addr_of_mut!(STORAGE) as *mut [MaybeUninit<SocketStorage>; N])
}));

Measured on thumbv7em-none-eabihf with 7 sockets, opt-level="z" and fat LTO: 3616 bytes of .data removed, of which 3584 were zeros, at a cost of ~380 bytes of .rodata and a short init loop. RAM use is unchanged.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.61%. Comparing base (f676d39) to head (c1c9768).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1199      +/-   ##
==========================================
+ Coverage   82.57%   82.61%   +0.03%     
==========================================
  Files          83       83              
  Lines       26174    26194      +20     
==========================================
+ Hits        21613    21639      +26     
+ Misses       4561     4555       -6     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[whitequark]

I've designed to keep the core of smoltcp (excluding phy drivers) free of unsafe as an intentional tradeoff. I would like to keep that tradeoff intact. There are quite a few optimizations that can be done if you're OK writing unsafe code, but I think there is a place for a TCP/IP stack which chooses robustness over absolute highest performance (or code size, etc), so smoltcp is such a stack.

Regarding your particular problem: smoltcp has been originally designed to have its structures allocated on stack, perhaps in main. Would that solve your problem?

[sigvartmh]

I've designed to keep the core of smoltcp (excluding phy drivers) free of unsafe as an intentional tradeoff. I would like to keep that tradeoff intact. There are quite a few optimizations that can be done if you're OK writing unsafe code, but I think there is a place for a TCP/IP stack which chooses robustness over absolute highest performance (or code size, etc), so smoltcp is such a stack.

Regarding your particular problem: smoltcp has been originally designed to have its structures allocated on stack, perhaps in main. Would that solve your problem?

Sure if I self allocate them on the stack having thme being allocated/initialized from .bss that's fine thanks for your reply :)

I looked through the code and found a no places of unsafe which is why I assumed this maybe wouldn't go in anyways great work on the stack thanks for making it public 👍