periph/pm: pm_layered never used

[immesys]

On SAMR21 the pm_set_lowest implemented in sys/pm_layered is never called because the weakly defined noop version in drivers/periph_common takes precedence. If I knew why, this would be a pull request not an issue.

To see this, you first have to fix #6640 which at the time of writing is not merged. So

git apply <<EOF
diff --git a/cpu/cortexm_common/periph/pm.c b/cpu/cortexm_common/periph/pm.c
index dacf1b6..1993074 100644
--- a/cpu/cortexm_common/periph/pm.c
+++ b/cpu/cortexm_common/periph/pm.c
@@ -23,7 +23,7 @@
 #include "cpu.h"
 #include "periph/pm.h"
 
-#ifndef FEATURES_PERIPH_PM
+#ifndef FEATURE_PERIPH_PM
 void pm_set_lowest(void)
 {
     cortexm_sleep(0);
EOF

Then compile a trivial example for samr21

cd examples/timer_periodic_wakeup
BOARD=samr21-xpro make 

Now look at the symbol for pm_set_lowest:

arm-none-eabi-readelf bin/samr21-xpro/timer_periodic_wakeup.elf -s | grep "pm_set"
   508: 00000439     2 FUNC    WEAK   DEFAULT    1 pm_set_lowest

It's the weak symbol that was defined in drivers/periph_common which does nothing.

Now remove the fallback weak symbol:

git apply <<EOF
diff --git a/drivers/periph_common/pm.c b/drivers/periph_common/pm.c
index afd31b7..37b03c3 100644
--- a/drivers/periph_common/pm.c
+++ b/drivers/periph_common/pm.c
@@ -24,7 +24,6 @@
 #define ENABLE_DEBUG (0)
 #include "debug.h"
 
-void __attribute__((weak)) pm_set_lowest(void) {}
 
 void __attribute__((weak)) pm_off(void)
 {
EOF

Now recompile the app and check the symbol. The proper one in sys/pm_layered is now there and it is called at runtime. Consequently the cpu's pm_set function is no longer stripped and actually appears too.

arm-none-eabi-readelf bin/samr21-xpro/timer_periodic_wakeup.elf -s | grep "pm_set"
   521: 00000fe5    68 FUNC    GLOBAL DEFAULT    1 pm_set_lowest
   531: 000012d5    68 FUNC    GLOBAL DEFAULT    1 pm_set

At present this bug means that the new power management system doesn't work and the SAMR21 stays at 5mA+ even when doing nothing. I'm not sure why testing did not detect this, but we should fix it soon.

[immesys]

In case it's relevant. I am compiling with arm-none-eabi-gcc 4.9.3 20150529 (prerelease) (15:4.9.3+svn231177-1).

[kYc0o]

Same as #6709, should we close one of them?

[miri64]

I would propose to close this one. It is referenced in #6709 so the information provided here is still accessible, but the discussion in #6709 went a little bit on already.

[roberthartung]

@immesys can you recheck with the 2017.10 branch?

[kaspar030]

The weak functions don't exist anymore since #7726.