stm32: unified up_perf initialization

[Gary-Hobson]

Summary

move up_perf_init from boards to arch

Impact

Testing

[ALTracer]

Why does this PR break my build when enabling CONFIG_SCHED_IRQMONITOR? This looks like a regression from 12.0.0, it used to work and show somewhat true numbers in ps. I was building for stm32f411-minimum with modified config.

chip/stm32_start.c: In function '__start':
chip/stm32_start.c:161:24: error: 'STM32_SYSCLK_FREQUENCY' undeclared (first use in this function)
  161 |   up_perf_init((void *)STM32_SYSCLK_FREQUENCY);
      |                        ^~~~~~~~~~~~~~~~~~~~~~
chip/stm32_start.c:161:24: note: each undeclared identifier is reported only once for each function it appears in
make[1]: *** [Makefile:157: stm32_start.o] Error 1

For example,

nuttx/boards/arm/stm32h7/nucleo-h743zi2/src/stm32_boot.c

Lines 53 to 55 in 1f7b49d

	#ifdef CONFIG_SCHED_IRQMONITOR
	up_perf_init((FAR void *)STM32_SYSCLK_FREQUENCY);
	#endif

had #include <arch/board/board.h> which provides STM32_SYSCLK_FREQUENCY macro, visible in this translation unit. Most STM32 boards simply use DWT_CYCCNT (see arch/arm/src/armv7-m/arm_perf.c) and thus need to know MCU frequency.
Does CI check builds for configs with IRQMONITOR, like nucleo-h743zi2:jumbo in arm-12.dat? Yes, it does.
arch/arm/src/stm32/stm32_start.c doesn't have this #include, but arch/arm/src/stm32h7/stm32_start.c does, hence no CI build failure.

diff --git a/arch/arm/src/stm32/stm32_start.c b/arch/arm/src/stm32/stm32_start.c
index 08b7d21f0..72332d731 100644
--- a/arch/arm/src/stm32/stm32_start.c
+++ b/arch/arm/src/stm32/stm32_start.c
@@ -29,6 +29,7 @@
 #include <debug.h>
 
 #include <nuttx/init.h>
+#include <arch/board/board.h>
 
 #include "arm_internal.h"
 #include "nvic.h"

[Gary-Hobson]

Why does this PR break my build when enabling CONFIG_SCHED_IRQMONITOR? This looks like a regression from 12.0.0, it > used to work and show somewhat true numbers in ps. I was building for stm32f411-minimum with modified config.

chip/stm32_start.c: In function '__start':
chip/stm32_start.c:161:24: error: 'STM32_SYSCLK_FREQUENCY' undeclared (first use in this function)
161 | up_perf_init((void *)STM32_SYSCLK_FREQUENCY);
| ^~~~~~~~~~~~~~~~~~~~~~
chip/stm32_start.c:161:24: note: each undeclared identifier is reported only once for each function it appears in
make[1]: *** [Makefile:157: stm32_start.o] Error 1
For example,
[nuttx/boards/arm/stm32h7/nucleo-h743zi2/src/stm32_boot.c]> (https://github.com/apache/nuttx/blob/1f7b49d7003c9bf27ade13be6fcd177b4352299d/boards/arm/stm32h7/nucleo-> h743zi2/src/stm32_boot.c#L53-L55)

Lines 53 to 55 in 1f7b49d

#ifdef CONFIG_SCHED_IRQMONITOR
up_perf_init((FAR void *)STM32_SYSCLK_FREQUENCY);
#endif
had #include <arch/board/board.h> which provides STM32_SYSCLK_FREQUENCY macro, visible in this translation unit. Most > STM32 boards simply use DWT_CYCCNT (see arch/arm/src/armv7-m/arm_perf.c) and thus need to know MCU frequency.
Does CI check builds for configs with IRQMONITOR, like nucleo-h743zi2:jumbo in arm-12.dat? Yes, it does.
arch/arm/src/stm32/stm32_start.c doesn't have this #include, but arch/arm/src/stm32h7/stm32_start.c does, hence no CI > build failure.
diff --git a/arch/arm/src/stm32/stm32_start.c b/arch/arm/src/stm32/stm32_start.c
index 08b7d21f0..72332d731 100644
--- a/arch/arm/src/stm32/stm32_start.c
+++ b/arch/arm/src/stm32/stm32_start.c
@@ -29,6 +29,7 @@
#include <debug.h>

#include <nuttx/init.h>
+#include <arch/board/board.h>

#include "arm_internal.h"
#include "nvic.h"

Already submitted a PR to fix it: #8948

[ALTracer]

Thanks, I guess, will move the discussion there.
Why is my text in your quote suddenly Chinese? I don't speak it... Auto translation maybe?

[Gary-Hobson]

Thanks, I guess, will move the discussion there. Why is my text in your quote suddenly Chinese? I don't speak it... Auto translation maybe?

yes， i fixed it