cpu/esp_common: fixes common CPU configurations

[gschorcht]

Contribution description

This PR fixes the handling of cpu_conf.h in cpu/esp_common. Furthermore, it removes the platform specific heap_stats function in cpu/esp_common if module esp_idf_heap is not used.

ESP CPUs share common definitions in cpu/esp_common. There is also a cpu_conf.h file in cpu/esp_common which should contain common CPU configurations. However, this file was not included in the CPU specific cpu_conf.h in cpu/esp32 and cpu/esp8266 so that the definitions in this file had no affect. Furthermore, to include this file in CPU specific cpu_conf.h, it has to be renamed.

Testing procedure

Compilation should succeed in Murdock.

Issues/PRs references

Prerequisite for PR #13517

[benpicco]

Tested on esp8266-esp-12x

[benpicco]

Hm, I ran tests/pkg_littlefs2 on both esp32-wroom-32 and esp8266-esp-12x but didn't see the crashes like on CI.

[gschorcht]

Hm, I ran tests/pkg_littlefs2 on both esp32-wroom-32 and esp8266-esp-12x but didn't see the crashes like on CI.

Yes, this happens sporadically. The test synchronization seems a little unstable. It's the same as with nrf52dk. I bet it won't happen next time.

[benpicco]

Yes, this happens sporadically. The test synchronization seems a little unstable.

Yea, but it prints a

EXCEPTION!! exccause=0 (IllegalInstructionCause) @8008783c excvaddr=00000000

.heap: 183848 (used 848, free 183000) [bytes]

register set
pc      : 400d96c5	ps      : 00060130	exccause: 00000000	excvaddr: 00000000
epc1    : 400d96c5	epc2    : 00000000	epc3    : 00000000	epc4    : 00000000
epc5    : 00000000	epc6    : 00000000	epc7    : 00000000	
a0      : 8008783c	a1      : 3ffb2e00	a2      : 3ffb30c4	a3      : 00000000
a4      : 00000000	a5      : 0000002b	a6      : 00000002	a7      : 3ffb2f00
a8      : 80082907	a9      : 3ffb2de0	a10     : 00000000	a11     : ffffffff
a12     : 00060120	a13     : 0000ff00	a14     : 00ff0000	a15     : ff000000
lbeg    : 40089f29	lend    : 40089f39	lcount  : fffffffc

Timeout in expect script at "child.expect(r'OK \(\d+ tests\)')" (tests/pkg_littlefs2/tests/01-run.py:14)

That's unusual.

[gschorcht]

Yea but this time it prints a

EXCEPTION!! exccause=0 (IllegalInstructionCause) @8008783c excvaddr=00000000

Wow, it's a crash. But, I don't see how the changes in this PR should affect the littlefs tests. Furthermore, they work locally for me. The last nightly build had also no problems, but these tests were executed on another node. Maybe node pi-6356c652 has a hardware problem with the flash.

[gschorcht]

Hm, this time the tests failed again but on node pi-e4c1d860. This is really weird. Now, I'm trying to compile and test with master.

[gschorcht]

@benpicco Obviously it was again a problem of a write access of the littlefs(2) file system to the flash memory. As explained in comment #12955 (comment), the cache of the flash memory is disabled during write accesses and the program code in the IROM is not available. Therefore, only program code from the IRAM can be executed. That it sometimes works and sometimes not is probably due to different delays during writing.

Although commit 5b4389c already moved the functions of the littlefs(2) filesystems to the IRAM, some of these functions in turn called system functions that were still in the IROM. Moving these functions to the IRAM seems to solve the problem.

The test runs are working now. The remaining failures are unrelated.

[gschorcht]

@kaspar030 tests/nordic_softdevice fails permanently on nrf52dk. Is anything known?

[benpicco]

Those tests have been failing quite reliably in the past…
CI was fine with the changes on esp, no need to bother Murdock again.