PANIC on TIM7 IRQ in timer example on armv6-m without FRAME_POINTERS #6624
Comments
|
stm32f072-disco:btimer defconfig diff with stm32f072-disco:nsh defconfig: Board timer initialization: diff --git a/boards/arm/stm32f0l0g0/stm32f072-discovery/src/stm32_bringup.c b/boards/arm/stm32f0l0g0/stm32f072-discovery/src/stm32_bringup.c
index f52f9d545..fb9981ba4 100644
--- a/boards/arm/stm32f0l0g0/stm32f072-discovery/src/stm32_bringup.c
+++ b/boards/arm/stm32f0l0g0/stm32f072-discovery/src/stm32_bringup.c
@@ -30,6 +30,11 @@
#include <syslog.h>
+#ifdef CONFIG_TIMER
+#include <nuttx/irq.h>
+#include "stm32_tim.h"
+#endif
+
#include "stm32f072-discovery.h"
#ifdef CONFIG_EXAMPLES_MODBUS
@@ -80,6 +85,17 @@ int stm32_bringup(void)
}
#endif
+#if defined(CONFIG_TIMER)
+ /* Initialize the timer, at this moment it's only Basic timers 6,7 */
+
+ #if defined(CONFIG_STM32F0L0G0_TIM6)
+ ret = stm32_timer_initialize("/dev/timer6", 6);
+ #endif
+ #if defined(CONFIG_STM32F0L0G0_TIM7)
+ ret = stm32_timer_initialize("/dev/timer7", 7);
+ #endif
+#endif
+
UNUSED(ret);
return OK;
}Hardware header aborting the build: diff --git a/arch/arm/src/stm32f0l0g0/hardware/stm32_tim.h b/arch/arm/src/stm32f0l0g0/hardware/stm32_tim.h [170/593]
index ceb057d57..8d69c3024 100644
--- a/arch/arm/src/stm32f0l0g0/hardware/stm32_tim.h
+++ b/arch/arm/src/stm32f0l0g0/hardware/stm32_tim.h
@@ -26,7 +26,7 @@
****************************************************************************/
#if defined(CONFIG_STM32F0L0G0_STM32F0) || defined(CONFIG_STM32F0L0G0_STM32L0)
-#error "Timer definitions should be validated for target before use"
+#warning "Timer definitions should be validated for target before use"
#endif
/* TODO Missing TIM2 definitions available on STM32G0x1 */
@@ -1101,7 +1101,9 @@
#define BTIM_CR1_URS (1 << 2) /* Bit 2: Update Request Source */
#define BTIM_CR1_OPM (1 << 3) /* Bit 3: One Pulse Mode */
#define BTIM_CR1_ARPE (1 << 7) /* Bit 7: Auto-Reload Preload enable */
+#if defined(CONFIG_STM32F0L0G0_STM32G0)
#define BTIM_CR1_UIFREMAP (1 << 11) /* Bit 11: UIF status bit remapping */
+#endif
/* Control register 2 */
@@ -1126,6 +1128,8 @@
/* Counter register */
+#if defined(CONFIG_STM32F0L0G0_STM32G0)
#define BTIM_CNT_UIFCPY (1 << 31) /* Bit 31: UIF copy */
+#endif
#endif /* __ARCH_ARM_SRC_STM32F0L0G0_HARDWARE_STM32_TIM_H */Basic timers work fine, and I don't need the rest yet. I also have backported DEBUG_OPT_UNUSED_SECTIONS and ARCH_HAVE_BACKTRACE for arm_backtrace_thumb. |
|
stm32f072-disco crash log, stackdump and backtrace: Notice it tries to jump into peripheral region ($pc=0x4000140a), that's TIM7 0x40001400-0x400017ff. |
|
Milandr mdr32f1qi stackdump/backtrace: This time, $pc is at 0x0, which is start of Flash. OS crashes in processing TIMER1 interrupt while trying to do the timer_notify callback to send a 17 signal to timer_main POSIX signal handler and switch tasks in up_block_task, (but g_readytorun.head->xcp.regs is bad? -- investigate). I can catch it with a hardware breakpoint at *0x0 and *0x2. |
|
It's definitely "-fomit-frame-pointers" option and not "-foptimize-sibling-calls" option. |
|
@ALTracer did this also happens on STM32F1/F4 or only STM32F0 ? |
|
BTW, never heard about Milandr MDR32 before and there is not much info about it on Internet. Are you planing to add support to it on mainline? |
|
@acassis I could verify timers on STM32F429I-disco board sometime next week, though it's armv7e-m. I don't have access to any STM32F1-based boards. Milandr has a line of MCUs produced since ~2010, see the official manufacturer site https://ic.milandr.ru/products/mikrokontrollery_i_protsessory/ and unofficial wiki http://startmilandr.codepx.ru/doku.php/prog:start (both sites in Russian). The LDM-Systems company sells evaluation boards https://ldm-systems.ru/catalog/helper?cat=milandr&scat=op I am not affiliated with them, we're only evaluating their chips for our next network-connected device with all silicon components originating from Russia. I thought about bringing mainline support, but I've been porting NuttX/10.3 for a month now, the only board is only 50% feature-complete, and I still rely on vendor's open-source (?) HAL for a couple of drivers. I will consult with my employer by the end of August 2022 if anything useful comes out of this project. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Summary
I get consistent crashes while testing apps/examples/timer with disabled CONFIG_FRAME_POINTERS.
With them enabled, it works, but .text consumes +4.5KiB flash.
Impact
armv6-m thumb MCUs using basic timer peripherals.
Testing
stm32f072b-disco with custom config, Milandr mdr32f1qi with custom config.
Details
After porting basic timer functionality from stm32f0l0g0 onto mdr32, I tried to launch some userspace demos. It worked from time to time. It crashed when I tried to reduce consumed flash space. I even enabled tmrinfo debug prints, but they didn't help much in case of crashes -- though they showed successfull signal passing when everything worked.
I think I found a single Kconfig option that breaks it: CONFIG_FRAME_POINTERS=n. It changes CFLAGS from
-fno-omit-frame-pointers -fno-optimize-sibling-callsto-fomit-frame-pointersand saves space. I couldn't get anything else to break at runtime.Stackdumps/backtraces, defconfigs and workdir patches will follow. I used release/10.3 branch.
The text was updated successfully, but these errors were encountered: