NUCLEO_L073RZ: wake up from deep sleep test fails when linking with microlib

[evedon]

Description of defect

Greentea SysTimer wake up from deep sleep test fails on NUCLEO_L073RZ when linking with microlib. See more details in #12014 (comment)

Target(s) affected by this defect ?

NUCLEO_L073RZ

Toolchain(s) (name and version) displaying this defect ?

ARM toolchain with C microlib.

What version of Mbed-os are you using (tag or sha) ?

What version(s) of tools are you using. List all that apply (E.g. mbed-cli)

How is this defect reproduced ?

To reproduce, you need #12068 in order to build greentea tests with microlib.
Create a test.json in TESTS\configs with the following:

{
  "target_overrides": {
    "*": {  
      "target.default_lib": "small"
    }
  }
}

and run the test:

mbed test -m NUCLEO_L073RZ -t ARM -n *systimer --app-config TESTS/configs/test.json

[ciarmcom]

Internal Jira reference: https://jira.arm.com/browse/MBOTRIAGE-2476

[MarceloSalazar]

@ARMmbed/team-st-mcd can you please have a look to this?

[LMESTM]

I cherry-picked #12608 and tried to reproduce, but I only get a build failure :

$ mbed test -m NUCLEO_L073RZ -t ARM -n *systimer --app-config TESTS/configs/test.json -v
Failed to build library

there must be something missing from #12608 or another dependency ...

[evedon]

I probably missed led you in my earlier comment:
Add

{
  "target_overrides": {
    "*": {  
      "target.default_lib": "small"
    }
  }
}

to https://github.com/ARMmbed/mbed-os/blob/master/TESTS/configs/baremetal.json

[LMESTM]

@evedon I still see the same.

• I'm on mbed master (commit b83d41eef5e0fa0b4ae46aa7a9a223badd9a2c07)
• I cherry picked#12068
• I modified the baremetal.json with "target.default_lib": "small"

then when I launch the test, I just get the below error message

$ mbed test -m NUCLEO_L073RZ -t ARM -n *systimer --app-config TESTS/configs/baremetal.json -v
Failed to build library
mbedgt: greentea test automation tool ver. 1.7.3
mbedgt: test specification file '.\BUILD\tests\NUCLEO_L073RZ\ARM\test_spec.json' (specified with --test-spec option)
mbedgt: using '.\BUILD\tests\NUCLEO_L073RZ\ARM\test_spec.json' from current directory!
("TestSpec::load('.\\BUILD\\tests\\NUCLEO_L073RZ\\ARM\\test_spec.json')", "[Errno 2] No such file or directory: '.\\\\BUILD\\\\tests\\\\NUCLEO_L073RZ\\\\ARM\\\\test_spec.json'")
mbedgt: detecting connected mbed-enabled devices...
mbedgt: detected 1 device
        | platform_name | platform_name_unique | serial_port | mount_point | target_id                |
        |---------------|----------------------|-------------|-------------|--------------------------|
        | NUCLEO_L073RZ | NUCLEO_L073RZ[0]     | COM66       | D:          | 07600221357F00130A3B964E |
mbedgt: shuffle seed: 0.7357669713
mbedgt: no platform/target matching tests were found!
mbedgt: no matching platforms were found!
mbedgt: completed in 0.13 sec
[mbed] Working path "C:\Data\Workspace\mbed" (program)
[mbed] Exec "c:\python27\python.exe -u C:\Data\Workspace\mbed\tools\test.py -D MBED_TEST_MODE -t ARM -m NUCLEO_L073RZ --source . --build .\BUILD\tests\NUCLEO_L073RZ\ARM --test-spec .\BUILD\tests\NUCLEO_L073RZ\ARM\test_spec.json --build-data .\BUILD\tests\NUCLEO_L073RZ\ARM\build_data.json -n *systimer -v --app-config TESTS/configs/baremetal.json --greentea" in "C:\Data\Workspace\mbed"

The test is working ok as long as I do not add the "target.default_lib": "small" config

Is there some extra change needed to the NUCLEO_L073RZ target ?

[LMESTM]

Ok I finally found out, targets.json also needs to be updated for NUCLEO_L073RZ with

        "supported_c_libs": {
                "arm": ["std"],
                "gcc_arm": ["std", "small"],
                "iar": ["std"]
        }

Why not applying the above for all the targets by default ?
I agree it may not have been tested for all yet, but what would prevent that a target to support one of the small configurations ?

[LMESTM]

@evedon
Actually compilation seems ok now, but the link fails with error below:

        [DEBUG] Errors: Error: L6218E: Undefined symbol __sync_fetch_and_add_4 (referred from C:\Keil_v5\ARM\ARMCLANG\bin\..\lib\libcxx\libcppabi_p.l(cxa_handlers.cpp.o)).
        [DEBUG] Errors: Finished: 0 information, 3 warning and 1 error messages.

[evedon]

Which version of ARMC6 are you running?
We build successfully with ARM Compiler 6.11 and 6.13

[LMESTM]

$ /c/Keil_v5/ARM/ARMCLANG/bin/armclang.exe --version
Product: MDK Professional 5.26
Component: ARM Compiler 6.10.1
Tool: armclang [5d143200]

[LMESTM]

I upgraded.

$ /c/Keil_v5/ARM/ARMCLANG/bin/armclang.exe --version
Product: MDK-ARM Lite 5.29
Component: ARM Compiler 6.13.1
Tool: armclang [5d895d00]

EDIT:
seems to be okay now - looks like the license manager took some time to make some check ...

[LMESTM]

@evedon
So I have reproduced the issue and I confirm that the deep sleep exit takes more time with microlib, but I don't know why ...

Below are a few measurements of the most time consuming part of the deep sleep exit (https://github.com/ARMmbed/mbed-os/blob/master/targets/TARGET_STM/sleep.c#L261)

ForceOscOutofDeepSleep();  // microlib 1ms VS. stnadard arm lib 0,640ms
ForcePeriphOutofDeepSleep(); // microlib 1,39ms VS. standard arm lib 0,7ms
SetSysClock(); // microlib 2,28ms VS. standard arm lib 1,21ms

When executed in microlib configuration the same code seems to be about 2X slower.
And all this code is executer under critical section ..

Is there a specific compiler optimization that is turned OFF with microlib ?
@kjbracey-arm any other idea ?