cpu/esp_common: erase block/sector-wise to leave interrupt time by crasbe · Pull Request #22945 · RIOT-OS/RIOT

crasbe · 2026-10-19T21:13:50Z

Contribution description

With the migration to ESP-IDF v5.4 in #21261, the erase function now erases a whole region instead of individual sectors/blocks. Depending on the size of the region, this can take a couple of seconds or more.
Since in that time, the interrupts have to be disabled, the interrupt that retriggers the watchdog is also disabled, causing a watchdog reset.

This PR changes the approach to erase the flash block- or sectorwise, leaving some space in between for interrupts.

Testing procedure

I extended the tests/drivers/mtd_raw test to write and erase a large section of the flash, which takes some seconds. This will trigger the watchdog reset reliably.

The rest ran on an esp32-wemos-d1-r32 board, Claude ran the tests for seeedstudio-xiao-esp32c3, seeedstudio-xiao-esp32s3, esp32h2-devkit and esp32c6-devkit as well cuz I'm lazy. Can prove traces if wished.

Behavior on `master`

Cherry-pick the commit with the test:

chris@ThinkPias:~/RIOTstuff/riot-upstream3/RIOT$ git log --oneline -n1
aa5db4178e (HEAD -> pr/esp32_flash_crash, origin/pr/esp32_flash_crash) tests/drivers/mtd_raw: add test for larger writes and erases
chris@ThinkPias:~/RIOTstuff/riot-upstream3/RIOT$ git checkout master
Zu Zweig »master« gewechselt
Ihr Branch ist auf demselben Stand wie 'origin/master'.
chris@ThinkPias:~/RIOTstuff/riot-upstream3/RIOT$ git cherry-pick aa5db4178e
[master b6b7537c2f] tests/drivers/mtd_raw: add test for larger writes and erases
 Date: Mon Oct 19 23:05:38 2026 +0200
 2 files changed, 44 insertions(+)

Welcome to pyterm!
Type '/exit' to exit.

> 
> mtd info
mtd info
mtd devices: 1
test 0
 -=[ MTD_0 ]=-
sectors: 763
pages per sector: 16
page size: 256
total: 3052 kiB
> test 0
[START]
[SUCCESS]
> test_erase_all 0
test_erase_all 0
[START]
ets Jul 29 2019 12:21:46

rst:0x7 (TG0WDT_SYS_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
configsip: 0, SPIWP:0xee
clk_drv:0x00,q_drv:0x00,d_drv:0x00,cs0_drv:0x00,hd_drv:0x00,wp_drv:0x00
mode:DOUT, clock div:2
load:0x3fff0030,len:1552
load:0x40078000,len:13736
load:0x40080400,len:4
ho 8 tail 4 room 4
load:0x40080404,len:2916
entry 0x40080414
Unicore app
Single core mode
detected chip: generic
flash io: dout

main(): This is RIOT! (Version: b6b7)
Manual MTD test
Timeout in expect script at "child.expect_exact("[SUCCESS]")" (tests/drivers/mtd_raw/tests/01-run.py:21)

Behavior with this PR

Welcome to pyterm!
Type '/exit' to exit.

> 
> mtd info
mtd info
mtd devices: 1
test 0
 -=[ MTD_0 ]=-
sectors: 763
pages per sector: 16
page size: 256
total: 3052 kiB
> test 0
[START]
[SUCCESS]
> test_erase_all 0
test_erase_all 0
[START]
[SUCCESS]
> 
make: Verzeichnis „/home/chris/RIOTstuff/riot-upstream3/RIOT/tests/drivers/mtd_raw“ wird verlassen

Issues/PRs references

Fixes the new issue raised in #14506.
Also fixes #16130 and reverts #16131.

Declaration of AI-Tools / LLMs usage:

AI-Tools / LLMs that were used are:

Claude Code Opus 5.5 found and fixed the issue when searching for the original issue in mdt_erase success, but vfs_format resets board (esp32-heltec-lora32-v2) #14506, which has since been fixed. I rewrote the comment and simplified the approach.

Erasing a big region can stall interrupts for too long which can trigger the watchdog and cause a reset. Therefore we erase sector- or blockwise, depending on the flash. Assisted-By: Claude Code Opus 5.5

Large writes and/or erases can lead to watchdog timeouts if the MTD subsystem does not leave enough interrupt time. Add a large write and erase to check for that. Assisted-By: Claude Code Opus 5.5

crasbe · 2026-10-19T21:30:15Z

Here comes the trust me bro Test Matrix:

Board	MTD	master	this PR
seeedstudio-xiao-esp32c3	3052 KiB	Watchdog-Reset	[SUCCESS]
seeedstudio-xiao-esp32s3	7148 KiB	Watchdog-Reset	[SUCCESS]
esp32c6-devkit	7148 KiB	Watchdog-Reset	[SUCCESS]
esp32h2-devkit	3052 KiB	[SUCCESS]	[SUCCESS]
esp32-wemos-d1-r32	3052 KiB	Watchdog-Reset	[SUCCESS]
esp32s2-devkit	3052 KiB	Watchdog-Reset	[SUCCESS]

The H2 is too fast for the watchdog to trigger.

gschorcht · 2026-10-19T22:25:00Z

I added the results for an ESP32-S2 board in comment #22945 (comment).

gschorcht · 2026-10-19T22:27:08Z

+    return 0;
+}
+
+SHELL_COMMAND(test_erase_all, "Write to every sector & erase them at once", cmd_test_erase_all);


Shouldn't the command simply be called test_erase, since you can specify the size to be deleted as the second argument?

This started as a command that erases everything, but for e.g. a multi-gigabyte SD card this would take forever, that's why it got changed.

Co-authored-by: Gunar Schorcht <gunar@schorcht.net> Co-authored-by: crasbe <crasbe@gmail.com>

…rases

benpicco

Just some minor comments from my side.

benpicco · 2026-10-19T21:42:30Z

+    uint32_t limit_mib = (argc > 2) ? strtoul(argv[2], NULL, 10) : 16;
+
+    if (dev == NULL || limit_mib == 0) {
+        printf("usage: %s <dev> [size in MiB, default 16, max flash size]\n", argv[0]);


Suggested change

printf("usage: %s <dev> [size in MiB, default 16, max flash size]\n", argv[0]);

printf("usage: %s <dev> [size in MiB, default 16]\n", argv[0]);

I think max flash size just adds confusion, at least it confused me.
You can't erase more than is there anyway.

benpicco · 2026-10-19T21:45:27Z

@@ -79,10 +80,27 @@ extern uint32_t spi_flash_get_id(void);

 static inline esp_err_t spi_flash_erase_range(size_t start_address, size_t size)


Suggested change

static inline esp_err_t spi_flash_erase_range(size_t start_address, size_t size)

static esp_err_t spi_flash_erase_range(size_t start_address, size_t size)

benpicco · 2026-10-19T22:50:23Z

+     * is erased block- or sectorwise. */
+    while ((res == ESP_OK) && size) {
+        size_t chunk = ((start_address % block_size) == 0) && (size >= block_size)
+                     ? block_size : sector_size;


I assume block_size >= sector_size?
Does it have an advantage to switch between the two? The esp8266 variant only uses sectors, but maybe that's all there is.

Just a short comment as to why this was done that way would be nice.

riot-ci · 2026-10-19T23:19:37Z

Murdock results

❌ FAILED

2ae5d12 fixup! fixup! tests/drivers/mtd_raw: add test for larger writes and erases

Success	Failures	Total	Runtime
2166	0	13104	03m:07s

Artifacts

crasbe added 2 commits October 19, 2026 22:26

cpu/esp_common: erase block/sector-wise to leave interrupt time

4f00a36

Erasing a big region can stall interrupts for too long which can trigger the watchdog and cause a reset. Therefore we erase sector- or blockwise, depending on the flash. Assisted-By: Claude Code Opus 5.5

tests/drivers/mtd_raw: add test for larger writes and erases

aa5db41

Large writes and/or erases can lead to watchdog timeouts if the MTD subsystem does not leave enough interrupt time. Add a large write and erase to check for that. Assisted-By: Claude Code Opus 5.5

crasbe requested a review from benpicco October 19, 2026 21:13

crasbe requested review from MichelRottleuthner, aabadie, gschorcht and leandrolanzieri as code owners October 19, 2026 21:13

github-actions Bot added Area: tests Area: tests and testing framework Platform: ESP Platform: This PR/issue effects ESP-based platforms Area: cpu Area: CPU/MCU ports labels Oct 19, 2026

crasbe linked an issue Oct 19, 2026 that may be closed by this pull request

mdt_erase success, but vfs_format resets board (esp32-heltec-lora32-v2) #14506

Open

tests/drivers/mtd_raw: re-enable for esp32-wroom-32

e905c4b

gschorcht reviewed Oct 19, 2026

View reviewed changes

Comment thread tests/drivers/mtd_raw/main.c Outdated

gschorcht reviewed Oct 19, 2026

View reviewed changes

Comment thread tests/drivers/mtd_raw/main.c Outdated

crasbe commented Oct 19, 2026

View reviewed changes

Comment thread tests/drivers/mtd_raw/tests/01-run.py Outdated

fixup! tests/drivers/mtd_raw: add test for larger writes and erases

a987d60

Co-authored-by: Gunar Schorcht <gunar@schorcht.net> Co-authored-by: crasbe <crasbe@gmail.com>

crasbe commented Oct 19, 2026

View reviewed changes

Comment thread tests/drivers/mtd_raw/main.c Outdated

fixup! fixup! tests/drivers/mtd_raw: add test for larger writes and e…

2ae5d12

…rases

benpicco reviewed Oct 19, 2026

View reviewed changes

Navigation Menu

Repository navigation

cpu/esp_common: erase block/sector-wise to leave interrupt time - #22945

cpu/esp_common: erase block/sector-wise to leave interrupt time#22945
crasbe wants to merge 5 commits into
RIOT-OS:masterfrom
crasbe:pr/esp32_flash_crash

crasbe commented Oct 19, 2026 •

edited

Loading

Uh oh!

crasbe commented Oct 19, 2026 •

edited by gschorcht

Loading

Uh oh!

gschorcht commented Oct 19, 2026

Uh oh!

gschorcht Oct 19, 2026

Uh oh!

crasbe Oct 19, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

benpicco left a comment

Uh oh!

benpicco Oct 19, 2026

Uh oh!

benpicco Oct 19, 2026

Uh oh!

benpicco Oct 19, 2026

Uh oh!

riot-ci commented Oct 19, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

	printf("usage: %s <dev> [size in MiB, default 16, max flash size]\n", argv[0]);
	printf("usage: %s <dev> [size in MiB, default 16]\n", argv[0]);

		@@ -79,10 +80,27 @@ extern uint32_t spi_flash_get_id(void);

		static inline esp_err_t spi_flash_erase_range(size_t start_address, size_t size)

	static inline esp_err_t spi_flash_erase_range(size_t start_address, size_t size)
	static esp_err_t spi_flash_erase_range(size_t start_address, size_t size)

Conversation

crasbe commented Oct 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Contribution description

Testing procedure

Behavior on master

Behavior with this PR

Issues/PRs references

Declaration of AI-Tools / LLMs usage:

Uh oh!

crasbe commented Oct 19, 2026 • edited by gschorcht Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gschorcht commented Oct 19, 2026

Uh oh!

gschorcht Oct 19, 2026

Choose a reason for hiding this comment

Uh oh!

crasbe Oct 19, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

benpicco left a comment

Choose a reason for hiding this comment

Uh oh!

benpicco Oct 19, 2026

Choose a reason for hiding this comment

Uh oh!

benpicco Oct 19, 2026

Choose a reason for hiding this comment

Uh oh!

benpicco Oct 19, 2026

Choose a reason for hiding this comment

Uh oh!

riot-ci commented Oct 19, 2026

Murdock results

Artifacts

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

crasbe commented Oct 19, 2026 •

edited

Loading

Behavior on `master`

crasbe commented Oct 19, 2026 •

edited by gschorcht

Loading