SpiDma read/write: copy as little as possible by bugadani · Pull Request #5290 · esp-rs/esp-hal

bugadani · 2026-03-31T13:44:45Z

Using code developed for AES DMA, this PR tries its best to avoid copying data before DMA-enabled SPI transfers. This PR enables us to send/receive data to/from DRAM and PSRAM without configuring copy buffers.

Non-copying transfers are disabled on ESP32 for unaligned data. ESP32 is the only device we have that requires a TX-side copy buffer due to alignment requirements, and the code we've had for AES-DMA doesn't implement that yet.

Closes #3125 by removing most of the copying that a display driver would do.

github-actions · 2026-03-31T13:44:57Z

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

bugadani · 2026-03-31T13:50:40Z

+                if copy_before > 0 {
+                    self.write_async_impl(MaybeCopyTxBuf::Copy(tx_buffer), &words[..copy_before])
+                        .await?;
+                }


This cries for a "block for short transfers" setting. Whether we want that even for CPU-controlled async transfers is a good question, I think really fast SPI busses justify it.

bugadani · 2026-04-01T13:12:20Z

/hil full

bugadani · 2026-04-01T13:12:56Z

/test-size embassy_spi esp32s3

github-actions · 2026-04-01T13:13:13Z

Binary Size Report

`esp32s3`

Diff (PR vs. Base)

VM SIZE    
-------------- 
+0.7%    +428    .bss
-1.5% -1.02Ki    .text
-0.4%    -612    TOTAL
Filtering enabled (source_filter); omitted  341Ki of entries

github-actions · 2026-04-01T14:37:17Z

Triggered full HIL run for #5290.

Run: https://github.com/esp-rs/esp-hal/actions/runs/23854189810

Status update: ❌ HIL (full) run failed (conclusion: failure).

bugadani · 2026-04-01T14:48:39Z

+            #[cfg(esp32)]
+            if !(buffer.as_ptr() as usize).is_multiple_of(4) {
+                return false;
+            }


I'll highlight this just to mention that I don't want to bother with this PR any more just to get it to work on ESP32. We still skip copying if the data is aligned, but resolving this here would take probably more changes than this PR needs to have. The solution would be to add a copy buffer to MaybeCopyTxBuffer::Direct, and pipe it through prepare_for_tx, but it can be done in the future if we really need to do it.

github-actions · 2026-04-02T12:10:53Z

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

github-actions · 2026-04-08T09:30:50Z

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

bugadani · 2026-04-14T10:00:01Z

+                #[cfg(esp32s2)]
+                if _direction == TransferDirection::In
+                    && !((buffer.as_ptr() as usize).is_multiple_of(16)
+                        && buffer.len().is_multiple_of(16))
+                {
+                    // When EDMA accesses external RAM, fields in linked list descriptors should be
+                    // aligned with block size. Specifically, size and buffer
+                    // address pointer in receive descriptors should be 16-byte, 32-byte or 64-byte
+                    // aligned. For data frame whose length is not a multiple of 16 bytes, 32 bytes,
+                    // or 64 bytes, EDMA adds padding bytes to the end.
+                    //
+                    // FIXME: the above should be handled by `prepare_for_rx`, but for now we just
+                    // skip the no-copy optimization.
+                    return false;
+                }


Makes me wonder why AES doesn't have this issue - are we lucky with the tests?

github-actions Bot added the merge-conflict Merge conflict detected. Automatically added/removed by CI. label Mar 31, 2026

bugadani commented Mar 31, 2026

View reviewed changes

Comment thread esp-hal/src/spi/master/dma.rs

bugadani commented Mar 31, 2026

View reviewed changes

Comment thread esp-hal/src/spi/master/dma.rs

bugadani commented Mar 31, 2026

View reviewed changes

This comment was marked as outdated.

Sign in to view

bugadani force-pushed the spi-dma-nocopy branch 2 times, most recently from 38fd120 to 85d8576 Compare April 1, 2026 09:33

github-actions Bot removed the merge-conflict Merge conflict detected. Automatically added/removed by CI. label Apr 1, 2026

bugadani force-pushed the spi-dma-nocopy branch 5 times, most recently from 5e99f7f to 38a1636 Compare April 1, 2026 12:36

bugadani mentioned this pull request Apr 1, 2026

Merge SpiDma and SpiDmaBus drivers #5272

Draft

This comment was marked as outdated.

Sign in to view

bugadani commented Apr 1, 2026

View reviewed changes

github-actions Bot added the merge-conflict Merge conflict detected. Automatically added/removed by CI. label Apr 2, 2026

bugadani force-pushed the spi-dma-nocopy branch from 62fc503 to 86de035 Compare April 7, 2026 08:04

github-actions Bot added merge-conflict Merge conflict detected. Automatically added/removed by CI. and removed merge-conflict Merge conflict detected. Automatically added/removed by CI. labels Apr 7, 2026

bugadani force-pushed the spi-dma-nocopy branch from be7262a to c3da565 Compare April 13, 2026 08:54

github-actions Bot removed the merge-conflict Merge conflict detected. Automatically added/removed by CI. label Apr 13, 2026

bugadani added 16 commits April 13, 2026 10:55

Rename methods

6a3a650

Merge DMA-driven drivers

44aa646

Reorganize SPI DMA driver

b6070ec

Clean up chunking code

47e138a

Extract constructor impl

47f9c8a

Fix example

79acbb3

Docs

1abb1b7

Skip empty operations

89851cb

Use a lifetimed buffer internally

4dfd6b7

Extract copying functions

c1d4add

Minimal copying async read/write

f5fe139

Update blocking functions

69dce67

Do not write back the owner bit

454070f

Force copying unaligned buffer on ESP32

7da6b15

Tweak documentation

8624b16

Simplify PCNT setup in SPI tests

f1a5a0d

bugadani force-pushed the spi-dma-nocopy branch 2 times, most recently from f66878d to 09ff5b1 Compare April 13, 2026 10:32

Leave cache writeback to the prepare impl

5f80969

bugadani force-pushed the spi-dma-nocopy branch from 09ff5b1 to 44dbde7 Compare April 13, 2026 15:17

Rewrite tests to use PSRAM if available, disable optimization on S2

049b4c1

bugadani force-pushed the spi-dma-nocopy branch from 44dbde7 to 049b4c1 Compare April 13, 2026 15:26

bugadani commented Apr 14, 2026

View reviewed changes

bugadani force-pushed the spi-dma-nocopy branch from 1e2fb66 to 9c1aa53 Compare April 14, 2026 14:31

Tweak ESP32 and S2

7fa46f6

bugadani force-pushed the spi-dma-nocopy branch from 9c1aa53 to 7fa46f6 Compare April 14, 2026 14:53

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

SpiDma read/write: copy as little as possible#5290

SpiDma read/write: copy as little as possible#5290
bugadani wants to merge 19 commits intoesp-rs:mainfrom
bugadani:spi-dma-nocopy

bugadani commented Mar 31, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Mar 31, 2026

Uh oh!

Uh oh!

Uh oh!

bugadani Mar 31, 2026

Uh oh!

This comment was marked as outdated.

bugadani commented Apr 1, 2026 •

edited

Loading

Uh oh!

bugadani commented Apr 1, 2026

Uh oh!

github-actions Bot commented Apr 1, 2026 •

edited

Loading

Uh oh!

This comment was marked as outdated.

This comment was marked as outdated.

github-actions Bot commented Apr 1, 2026 •

edited

Loading

Uh oh!

bugadani Apr 1, 2026

Uh oh!

github-actions Bot commented Apr 2, 2026

Uh oh!

github-actions Bot commented Apr 8, 2026

Uh oh!

bugadani Apr 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

bugadani commented Mar 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Mar 31, 2026

Uh oh!

Uh oh!

Uh oh!

bugadani Mar 31, 2026

Choose a reason for hiding this comment

Uh oh!

This comment was marked as outdated.

bugadani commented Apr 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

bugadani commented Apr 1, 2026

Uh oh!

github-actions Bot commented Apr 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Binary Size Report

esp32s3

Diff (PR vs. Base)

Uh oh!

This comment was marked as outdated.

This comment was marked as outdated.

github-actions Bot commented Apr 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

bugadani Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Apr 2, 2026

Uh oh!

github-actions Bot commented Apr 8, 2026

Uh oh!

bugadani Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

bugadani commented Mar 31, 2026 •

edited

Loading

bugadani commented Apr 1, 2026 •

edited

Loading

github-actions Bot commented Apr 1, 2026 •

edited

Loading

`esp32s3`

github-actions Bot commented Apr 1, 2026 •

edited

Loading