CH32V003 - Hardware Fault Interrupt when using TIM2 periodic interrupt

[quevedo]

Hello MCU lovers

I am starting to experiment with RISC-V, so I decided to get some devboards with the CH32V003 MCU. As I am used to work with Eclipse, I decided to try the official MounRiver Studio IDE. I started with some basic polling UART, and then I moved to a periodical interrupt in TIM2 toggling a GPIO output with a LED.

I found an example code at: https://pallavaggarwal.in/2023/09/21/ch32v003-programming-gpio-as-output/

I made a few changes to toggle a LED inside the Timer handler, but when trying to run the code (in debug mode), a "Hardware Fault" exception is thrown when the Handler is called. I even removed all GPIO config, leaving the program like the original one, just adding a variable increment inside the handler (see attached main.c).

I checked the functions to configure the timer and the ISR, as well as the flag cleaning function, they look OK.

I see 2 hypotheses for the problem: either there is a software bug in the functions that I could not find, or there is a hardware bug in my MCUs (I tried running the program in 2 boards, same result).

Any ideas?

[DavidAlfa]

Maybe?
https://www.reddit.com/r/RISCV/comments/126262j/notes_on_wch_fast_interrupts/

[rhodges]

I just added this to one of my programs:

Code: [Select]

void TIM2_IRQHandler(void)
{
}
And the resulting assembly language is:

Code: [Select]

00000276 <TIM2_IRQHandler>:
 276:   8082                    ret
You need to tell gcc that this is an interrupt handler. Something like this:

Code: [Select]

void SysTick_Handler(void) __attribute__((interrupt));
void SysTick_Handler(void)
{
    SysTick->SR = 0;

    timer_ms();
    clock_ms++;
    if (clock_ms < 100)
        return;
    clock_ms -= 100;

    timer_10();
}


[quevedo]

Prototyping the handler with the specification "__attribute__((interrupt))" just did the trick! Thank you!

[HwAoRrDk]

Note if you're using the WCH MounRiver IDE and compiler, you should really be using __attribute__((interrupt("WCH-Interrupt-fast"))) instead.

Otherwise, if you use regular __attribute__((interrupt), while it will work, if HPE is not turned off (default is on with WCH startup code) then you'll get the CPU registers saved/restored twice in your ISR (once by HPE, once by code generated by the compiler), which is a waste of time.

[rhodges]

Yes, I agree. In my case, I decided to use "any" gcc version and modified startup_ch32v00x.S to not enable HPE. This might be from code you posted.

Code: [Select]

/* Only enable nesting, not HPE */
    li t0, 0x2
    csrw 0x804, t0
 
I might enable HPE some day and use "naked" interrupt handlers, and end them with inline assembly "mret". This was discussed on another thread, and I think Bruce said it was dangerous. I would visually inspect the assembly output to verify them. But for now, I am staying with standard interrupt handlers.