Merge SpiDma and SpiDmaBus drivers

[bugadani]

We don't need two, when one can do the job. Further work needs to be done:

• avoid copying into the internal buffers when possible (so that they are only needed in some cases), and maybe not at all for ESP32-C5 and later...? - SpiDma read/write: copy as little as possible #5290
• introduce a config option to fall back to CPU-driven blocking operatons for very short transfers - it'll result in a small, but noticeable improvement with the lcd-async crate. SPI: allow forcing CPU-driven blocking transfers #5350
• a bunch of documentation work is necessary to explain the differences between the different styles in the API - SpiDma read/write: copy as little as possible #5290

This PR allows using the buffer-based API together with drivers that require embedded-hal(-async) implementations. It's implemented by adding copy buffers to the global state of the SpiDma driver. These buffers are used in place of EmptyBuf, which is mostly unnecessary at this point.

[Dominaezzz]

It's implemented by adding copy buffers to the global state of the SpiDma driver

Ah that works. My first thought is how big is the buffer gonna be, but I'll just read the PR.

I guess you're closing #3125?

[bugadani]

My first thought is how big is the buffer gonna be, but I'll just read the PR.

As much as you give it, with_buffers isn't going away just yet. But due to the 32000-some byte limit we also kinda just need 8 descriptors linked up if we don't need to copy data, so a good chunk of use cases can work without actually setting up buffers. How we'll end up explaining this is another matter...

I'm not ure about #3125 I'm not actually targeting that issue.

[github-actions]

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

[github-actions]

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

[github-actions]

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

[github-actions]

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

[Copilot]

Pull request overview

This PR consolidates the SPI DMA driver surface by removing the separate SpiDmaBus type and making SpiDma directly support both the slice-based API (copying via internal buffers) and the externally-managed buffer API (via DmaRxBuf/DmaTxBuf), including embedded-hal(-async) trait implementations.

Changes:

• Replace SpiDmaBus usage by moving copy-buffers into SpiDma global DMA state and exposing SpiDma::with_buffers(...) directly.
• Rename DMA buffer-oriented APIs to _buffer / _buffers suffixed methods (e.g. read_buffer, transfer_buffers, half_duplex_read_buffer).
• Update examples/tests to use the renamed buffer-based methods.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
qa-test/src/bin/qspi_flash.rs	Update QSPI flash QA test to use half_duplex_{read,write}_buffer APIs.
hil-test/src/bin/spi_half_duplex_write_psram.rs	Update HIL half-duplex PSRAM test to use half_duplex_write_buffer.
hil-test/src/bin/spi_half_duplex_slave_qspi.rs	Update HIL QSPI slave test to use half_duplex_{read,write}_buffer.
hil-test/src/bin/spi_full_duplex.rs	Update full-duplex DMA HIL tests to use read_buffer, write_buffer, transfer_buffers.
examples/peripheral/spi/loopback_dma_psram/src/main.rs	Update loopback DMA+PSRAM example to use transfer_buffers.
esp-hal/src/spi/master/dma.rs	Core refactor: merge SpiDmaBus into SpiDma, add embedded-hal(-async) trait impls, add async slice-based methods, and update buffer handling.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

[Copilot]

The SpiDma docs say the buffer-based API method names end with _buffer, but one of the renamed methods is transfer_buffers (plural). Consider tweaking the wording to mention both _buffer and _buffers (or list the method names) to avoid confusion.

Suggested change

	/// arguments as well as the number of bytes to transfer, and their names end with `_buffer`.
	/// arguments as well as the number of bytes to transfer, and their names end with `_buffer` or
	/// `_buffers`.

[github-actions]

New commits in main have made this PR unmergeable. Please resolve the conflicts.

[MabezDev]

Just some nits, overall LGTM!

[MabezDev]

Is this a limitation, or a design decision we've made for our ease of impl? IIRC eh discourages early flushing because we can be more efficient by queuing up transfers, but I will admit this nugget of knowledge is a bit fuzzy

[bugadani]

We can't not flush with DMA if we want any semblance of soundness - we can't release the borrow on the buffers before the transfers finish.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.