improved flashing speed on macos by sengmonkham · Pull Request #239 · beagleboard/bb-imager-rs

sengmonkham · 2026-02-04T10:32:05Z

This PR addresses the slow flashing speed issue observed on macOS by disabling OS-level file caching for the target device and optimizing buffer alignment.

#58

Changes
Enabled F_NOCACHE: Uses fcntl to set the F_NOCACHE flag on the file descriptor obtained via authopen. This forces direct I/O to the SD card, bypassing the unified buffer cache which was causing performance bottlenecks.

Increased Buffer Alignment: Updated DirectIoBuffer alignment from 512 bytes to 4096 bytes (page size) to ensure optimal compatibility with direct I/O operations and strict alignment requirements on macOS.

Flashing speed before changes

kawkoiharu@kawkois-MacBook-Air ~ % cargo run --manifest-path bb-flasher-sd/Cargo.toml --example flash_benchmark --features macos_authopen -- /dev/disk4
error: manifest path `bb-flasher-sd/Cargo.toml` does not exist
kawkoiharu@kawkois-MacBook-Air ~ % cd code/gsoc/bb-imager-rs
kawkoiharu@kawkois-MacBook-Air bb-imager-rs % clear
kawkoiharu@kawkois-MacBook-Air bb-imager-rs % cargo run --manifest-path bb-flasher-sd/Cargo.toml --example flash_benchmark --features macos_authopen -- /dev/disk4
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.39s
     Running `target/debug/examples/flash_benchmark /dev/disk4`
Creating dummy image (100MB)...
Starting flash to "/dev/disk4"...
Progress: 100.0%
Flash completed successfully.
Time taken: 33.98s
Average Speed: 2.94 MB/s

Flashing speed after the changes

Starting flash to "/dev/disk4"...
Progress: 100.0%
Flash completed successfully.
Time taken: 19.48s
Average Speed: 5.13 MB/s
kawkoiharu@kawkois-MacBook-Air bb-imager-rs %```

sengmonkham · 2026-02-04T10:52:11Z

@Ayush1325 please review

Ayush1325 · 2026-02-04T10:59:01Z

@sengmonkham What is the non-existent flash_benchmark example?
Also I can see you are using debug build. Please provide results with release build.

sengmonkham · 2026-02-04T13:22:09Z

@sengmonkham What is the non-existent flash_benchmark example? Also I can see you are using debug build. Please provide results with release build.

Sorry for the confusion. I originally used a local flash_benchmark script to bypass the CLI's strict SD card detection (since my test drive wasn't being picked up) and to measure the speeds.

Regarding the release build, I have verified the fix in release mode using the main CLI tool (patched to bypass detection for my setup).

kawkoiharu@kawkois-MacBook-Air bb-imager-rs % cargo run --release --manifest-path bb-imager-cli/Cargo.toml -- flash sd ~/micropython.img.xz /dev/disk4
Compiling bb-flasher v0.1.0 ...
Compiling bb-imager-cli v0.0.19 ...
Finished release profile [optimized] target(s) in 8.03s
Running target/release/bb-imager-cli flash sd /Users/kawkoiharu/micropython.img.xz /dev/disk4
[1] Preparing
[2] Flashing [██████████████████████...██████████████████████] [100 %]

Ayush1325 · 2026-02-04T15:52:11Z

@sengmonkham What is the non-existent flash_benchmark example? Also I can see you are using debug build. Please provide results with release build.

Sorry for the confusion. I originally used a local flash_benchmark script to bypass the CLI's strict SD card detection (since my test drive wasn't being picked up) and to measure the speeds.

Regarding the release build, I have verified the fix in release mode using the main CLI tool (patched to bypass detection for my setup).

kawkoiharu@kawkois-MacBook-Air bb-imager-rs % cargo run --release --manifest-path bb-imager-cli/Cargo.toml -- flash sd ~/micropython.img.xz /dev/disk4 Compiling bb-flasher v0.1.0 ... Compiling bb-imager-cli v0.0.19 ... Finished release profile [optimized] target(s) in 8.03s Running target/release/bb-imager-cli flash sd /Users/kawkoiharu/micropython.img.xz /dev/disk4 [1] Preparing [2] Flashing [██████████████████████...██████████████████████] [100 %]

Can you use time or something while running the CLI. I want the PR description to have info regarding how much real performance improvement this PR causes.

I will also need to check if the new alignment works properly on Windows and Linux myself.

giuliomoro · 2026-02-04T16:06:46Z

confirmed that the latest CI https://github.com/beagleboard/bb-imager-rs/releases/download/continuous-release/BeagleBoard.Imaging.Utility_alpha_aarch64.dmg flashes much faster than https://github.com/beagleboard/bb-imager-rs/releases/download/0.0.19/BeagleBoard.Imaging.Utility_0.0.19_aarch64.dmg

Ayush1325 · 2026-02-04T17:54:14Z

confirmed that the latest CI https://github.com/beagleboard/bb-imager-rs/releases/download/continuous-release/BeagleBoard.Imaging.Utility_alpha_aarch64.dmg flashes much faster than https://github.com/beagleboard/bb-imager-rs/releases/download/0.0.19/BeagleBoard.Imaging.Utility_0.0.19_aarch64.dmg

Not sure why that would be. That release does not contain this PR. And I haven't merged any macos flashing performance PR recently.

sengmonkham · 2026-02-04T18:46:36Z

2s difference
without the changes

Screen.Recording.2026-02-05.at.00.03.36.mov

with the changes
https://github.com/user-attachments/assets/d37879cf-0110-4e4f-9048-58eec76588a9

@sengmonkham What is the non-existent flash_benchmark example? Also I can see you are using debug build. Please provide results with release build.

Sorry for the confusion. I originally used a local flash_benchmark script to bypass the CLI's strict SD card detection (since my test drive wasn't being picked up) and to measure the speeds.
Regarding the release build, I have verified the fix in release mode using the main CLI tool (patched to bypass detection for my setup).
kawkoiharu@kawkois-MacBook-Air bb-imager-rs % cargo run --release --manifest-path bb-imager-cli/Cargo.toml -- flash sd ~/micropython.img.xz /dev/disk4 Compiling bb-flasher v0.1.0 ... Compiling bb-imager-cli v0.0.19 ... Finished release profile [optimized] target(s) in 8.03s Running target/release/bb-imager-cli flash sd /Users/kawkoiharu/micropython.img.xz /dev/disk4 [1] Preparing [2] Flashing [██████████████████████...██████████████████████] [100 %]

Can you use time or something while running the CLI. I want the PR description to have info regarding how much real performance improvement this PR causes.

I will also need to check if the new alignment works properly on Windows and Linux myself.

sengmonkham · 2026-02-05T09:32:26Z

@Ayush1325 as you can see above, my testing shows an improvement of 2s. Testing with a 130MB compressed image.
Without changes: 25s
With changes: 23s

Ayush1325 · 2026-02-05T09:34:52Z

@Ayush1325 as you can see above, my testing shows an improvement of 2s. Testing with a 130MB compressed image. Without changes: 25s With changes: 23s

Can you try with a bigger image. Maybe a BeagleY-AI image?

bb-flasher-sd/src/pal/macos.rs

bb-flasher-sd/src/flashing.rs

bb-imager-cli/Cargo.toml

sengmonkham · 2026-02-05T16:44:40Z

@Ayush1325 I have added an option for the cli users to switch to /dev/rdisk4 when trying to use /dev/disk4 for faster flashing experience. In my experience flashing the BeagleY-AI image it took 16mins using the rdisk4 and 23mins with disk4.

I also added a fix for the "Resource busy" error on macOS. The CLI now ensures the disk is unmounted before opening, which is required for raw device access. (This is essential for the rdisk workflow to work smoothly, so I included it here).

I could not manage to further improve the flashing speed, no mater what i did the speeds did not change.

bb-flasher-sd/src/pal/macos.rs

bb-imager-cli/Cargo.toml

bb-flasher-sd/src/pal/macos.rs

Ayush1325 · 2026-02-05T17:30:43Z

bb-flasher-sd/src/helpers.rs

 }

-#[repr(align(512))]
+#[repr(align(4096))]


Does this cause a speed increase?

tested it in isolation and got these results:
512: 16:51 mins
4096: 16:43 mins

Ayush1325 · 2026-02-05T20:31:33Z

bb-imager-cli/Cargo.toml

 "dist/.target/man" = "/usr/share/man/man1"
 "dist/.target/shell-comp/_bb-imager-cli" = "/usr/share/zsh/site-functions/_bb-imager-cli"
-"dist/.target/shell-comp/bb-imager-cli.bash" = "/usr/share/bash-completion/bb-imager-cli.bash"
+"dist/.target/shell-comp/bb-imager-cli.bash" = "/usr/share/bash-completion/bb-imager-cli.bash"


Not sure why this is showing up in diff. Probably some CRLF nonsense. Anyway remove this change.

Ayush1325 · 2026-02-05T20:33:53Z

bb-imager-cli/src/main.rs

+
+                    // Simple stdin read since we don't have dialoguer
+                    let mut input = String::new();
+                    if std::io::stdin().read_line(&mut input).is_ok() {


It's fine to use expect here.

Ayush1325 · 2026-02-05T20:35:00Z

bb-imager-cli/src/main.rs

            let wifi = wifi_ssid.map(|x| (x, wifi_password.unwrap()));

+            #[cfg(target_os = "macos")]
+            let dst = if dst.to_string_lossy().starts_with("/dev/disk")


Please extract this to it's own function. Maybe something like validate. You can define 2 versions, one for macos and other, a no-op for other OS.

Ayush1325 · 2026-02-05T20:35:35Z

bb-imager-cli/src/main.rs

+                        let input = input.trim().to_lowercase();
+                        if input.is_empty() || input == "y" || input == "yes" {
+                            let _ = term.write_line(&format!("Switching to {}\n", rdisk));
+                            PathBuf::from(rdisk)


Once you extract this to a separate function, you can just return here and remove all the else blocks with a single default return.

Ayush1325 · 2026-02-07T07:14:28Z

bb-imager-cli/Cargo.toml

 "dist/.target/man" = "/usr/share/man/man1"
-"dist/.target/shell-comp/_bb-imager-cli" = "/usr/share/zsh/site-functions/_bb-imager-cli"
-"dist/.target/shell-comp/bb-imager-cli.bash" = "/usr/share/bash-completion/bb-imager-cli.bash"
+"dist/.target/shell-comp/_bb-imager-cli" = "/usr/share/zsh/site-functions/_bb-imager-cli"


Why is this here?

I have absolutely no idea why it is showing as diff, all i did was remove the last change as you requested. should i re add those lines?

Ayush1325 · 2026-02-07T07:14:59Z

bb-flasher-sd/src/pal/macos.rs

-        .await
-        .unwrap()
-        .map_err(|source| Error::FailedToFormat { source })
+    tokio::task::spawn_blocking(move || {


Why change this?

- Increase buffer alignment to 4096 bytes for better Direct I/O performance. - Add interactive prompt in CLI to switch to raw device (/dev/rdisk) on macOS for significantly faster flashing. - Fix 'Resource busy' error by automatically unmounting the disk before flashing in the CLI. Signed-off-by: Sen <kawkoi@proton.mail>

Signed-off-by: sen <kawkoi@proton.me>

sengmonkham marked this pull request as draft February 4, 2026 10:42

sengmonkham marked this pull request as ready for review February 4, 2026 10:43

sengmonkham changed the title ~~improved flashing speed in macos~~ improved flashing speed on macos Feb 4, 2026

sengmonkham force-pushed the fix/macos-slow-flashing branch from 1db1888 to fd9b4a3 Compare February 5, 2026 13:04

Ayush1325 requested changes Feb 5, 2026

View reviewed changes

sengmonkham force-pushed the fix/macos-slow-flashing branch 5 times, most recently from bc9ecc6 to 8865ebc Compare February 5, 2026 16:06

Ayush1325 reviewed Feb 5, 2026

View reviewed changes

sengmonkham force-pushed the fix/macos-slow-flashing branch from 8865ebc to 6e7e01a Compare February 5, 2026 18:51

Ayush1325 requested changes Feb 5, 2026

View reviewed changes

Ayush1325 mentioned this pull request Feb 5, 2026

feat(macos): implement proper SD card ejection #243

Closed

Ayush1325 requested changes Feb 7, 2026

View reviewed changes

sengmonkham force-pushed the fix/macos-slow-flashing branch from e507953 to 3b45018 Compare February 7, 2026 09:36

Sen and others added 2 commits February 7, 2026 15:14

refactor: extract macOS device path check into separate function

1c778c8

Signed-off-by: sen <kawkoi@proton.me>

sengmonkham force-pushed the fix/macos-slow-flashing branch from 3b45018 to 1c778c8 Compare February 7, 2026 09:48

Ayush1325 approved these changes Feb 7, 2026

View reviewed changes

Ayush1325 merged commit 4cd7adf into beagleboard:main Feb 8, 2026
11 checks passed

Conversation

sengmonkham commented Feb 4, 2026

Uh oh!

sengmonkham commented Feb 4, 2026

Uh oh!

Ayush1325 commented Feb 4, 2026

Uh oh!

sengmonkham commented Feb 4, 2026

Uh oh!

Ayush1325 commented Feb 4, 2026

Uh oh!

giuliomoro commented Feb 4, 2026

Uh oh!

Ayush1325 commented Feb 4, 2026

Uh oh!

sengmonkham commented Feb 4, 2026

Uh oh!

sengmonkham commented Feb 5, 2026

Uh oh!

Ayush1325 commented Feb 5, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

sengmonkham commented Feb 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

sengmonkham commented Feb 5, 2026 •

edited

Loading