Spi/SpiDma: move some fields into global state

[bugadani]

No description provided.

[bugadani]

/test-size

[github-actions]

Binary Size Report

esp32

sleep-timer Diff (PR vs. Base)

VM SIZE    
-------------- 
[ = ]       0    TOTAL
Filtering enabled (source_filter); omitted  201Ki of entries

embassy-dhcp Diff (PR vs. Base)

VM SIZE    
-------------- 
[ = ]       0    TOTAL
Filtering enabled (source_filter); omitted  286Ki of entries

[github-actions]

New commits in main has made this PR unmergable. Please resolve the conflicts.

[bugadani]

/test-size loopback_dma_psram

[github-actions]

Triggered binary size analysis for #5259. Results will be posted as a comment when ready.

[Copilot]

Pull request overview

This PR refactors the SPI master (Spi/SpiDma) drivers to move some per-instance fields (pin guards and DMA transfer-in-progress flags) into per-peripheral global state, reducing per-driver storage and centralizing cleanup.

Changes:

• Introduces SpiWrapper and moves PeripheralGuard ownership into it, with drop-time deinit of shared state.
• Moves SPI pin guards into the per-peripheral State (via RefCell<SpiPinGuard>), and updates with_* pin assignment APIs to mutate that shared state.
• Moves DMA “transfer in progress” flags into a per-peripheral global DMA State, and reworks SpiDma to own Spi instead of AnySpi + local flags.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
esp-hal/src/spi/master/mod.rs	Moves SPI pin guard tracking into the global per-peripheral State and adds SpiWrapper with drop-time deinit.
esp-hal/src/spi/master/dma.rs	Refactors SpiDma to wrap Spi, and moves DMA transfer flags into per-peripheral global DMA state.

[Copilot]

Storing SpiPinGuard inside a global State behind RefCell means pin connect/disconnect now relies on interior mutability without any explicit critical-section/mutex protection. Given this is global state for a peripheral, it would be safer to use a lock (e.g. NonReentrantMutex<SpiPinGuard>) rather than RefCell, so pin updates are explicitly serialized and don't depend on an unsafe Sync impl.

[Copilot]

unsafe impl Sync for State is not justified here: this DMA State is stored in a static and uses Cell<bool> fields. Cell is not thread-safe and on multi-core chips it can be accessed from multiple contexts if the driver is shared, leading to potential data races/UB. Consider using AtomicBool (or portable_atomic::AtomicBool) for these flags, or wrap the state in an appropriate mutex/critical-section type and avoid the unsafe Sync impl.

[Copilot]

The DMA helper types struct Info / struct State introduced here shadow super::Info / super::State imported via use super::*;, which makes the module harder to follow and increases the risk of accidental type confusion later. Consider renaming these to something more specific (e.g. DmaInfo/DmaState) or avoiding the wildcard import.

[Copilot]

State is made Sync via unsafe impl Sync for State, but it contains RefCell<SpiPinGuard> (and on esp32 also Cell). RefCell/Cell are not thread-safe; on multi-core targets or with interrupt preemption this Sync impl can permit data races/UB if the state is ever accessed from more than one execution context. Prefer a synchronization primitive for pins (e.g. esp_sync::NonReentrantMutex or critical_section::Mutex<RefCell<_>>), or otherwise restructure so State does not need an unsafe Sync impl.

Suggested change

unsafe impl Sync for State {}