[embassy-sync] Using Watch, probe-rs and a --release build leads to an exception

[lure23]

I started using the new Watch functionality today. Unfortunately, haven't been able to see it work in release builds, yet. This happens:

3.559229 [INFO ] Init succeeded
3.905247 [ERROR] Exception 'Load access fault' mepc=0x42004a1e, mtval=0x4d421b00
3.905345 [ERROR] TrapFrame { ra: 1107327136, t0: 3905200, t1: 0, t2: 0, t3: 2181074082, t4: 0, t5: 0, t6: 0, a0: 1082556352, a1: 1, a2: 1, a3: 4, a4: 1107315144, a5: 1082171792, a6: 0, a7: 0, s0: 1082577216, s1: 2328, s2: 1082170664, s3: 4, s4: 4096, s5: 1082169368, s6: 0, s7: 1082579628, s8: 1296177920, s9: 1082169344, s10: 1082168999, s11: 1082169304, gp: 1082134720, tp: 0, sp: 1082577136, pc: 1107315230, mstatus: 6273, mcause: 5, mtval: 1296177920 }
3.905661 [ERROR] 0x4200789c
3.905693 [ERROR] 0x42009c40
3.905725 [ERROR] 0x42001264
3.905757 [ERROR] 0x420021b6
3.905789 [ERROR] 0x4200675c
3.905821 [ERROR] 0x42004894
3.905853 [ERROR] 0x42004650
3.905885 [ERROR] 0x4200c312
3.905917 [ERROR] 0x42000132

If I change to non---release (wanted to see the symbols in the stack trace), All is Fine.

I presume this is caused by inclusion of embassy-sync - I didn't change other dependencies, today.

Don't have a Minimal Repo, but my source is available here - note the tof branch.

Steps

• git clone [email protected]:lure23/Zoo-ESP32_Cx.git

  $ cd Zoo-*
  $ git checkout tof
  

• Read the README at the top and tof levels.

  This is likely beyond anyone wants to do, to repeat the problem. You'd need to set up tools, bring in vendor C sources... Oh, and you need VL53L5CX hardware! 🫠🎈🎈

• _

  $ cd tof/vl53l5cx
  $ make -f Makefile.dev m
  

Expected

The program runs and presents some results.

Actual

Above pasted exception.

Further info

If we compile it without --release (or not use probe-rs, see comments), all is fine:

$ make -f Makefile.dev m-dev
[...]
4.313160 [INFO ] Data #0: (30°C, 0ms)
4.313230 [INFO ] .target_status:    [[[Valid, Valid, Valid, SemiValid(9)], [Valid, Valid, SemiValid(9), Valid], [Valid, Valid, Valid, Valid], [Valid, Valid, Valid, Valid]]]
4.313369 [INFO ] .distance_mm:      [[[581, 685, 766, 1002], [850, 781, 997, 1038], [961, 641, 783, 1624], [645, 728, 780, 769]]]

[lure23]

The exception happens on this line, i.e. where the Watch is read/waited upon.

Using e.g. .get() instead of the .changed() gives the same behaviour.

---

What I would need to do is a minimum viable reproduction, but likely I'll just use non-release build and carry on. Perhaps later.

[lure23]

Updated the bug description on my own repo (a full rewrite).

The issue occurs with the combination of:

• probe-rs
• --release build
• embassy-sync - at least Watch mechanism, both .changed() and .get()

Running the code with espflash succeeds in both --release and default modes.

[peterkrull]

Hmm, I have not had any issues with extensive usage (e.g. RP2040, STM32, std, multi-core, interrupts). Could you try replacing the Watch with another channel, e.g. a PubSubChannel with a depth of 1? Just to see if the crash can be triggered without the Watch. Also as far as I can tell 'Load access fault' has something to do with incorrectly accessed memory, and your // Something brings in need for global allocator. Fake it!!! seems a bit spooky to me. Could you try using an actual allocator?