Timing to be reflected __disable_irq() and __enable_irq()

[Fkasa.1]

We make below code.

The variable a is written from both main function and interrupt function.

When accessing variable a with the main function to realize exclusive control, interrupt disable and enable (__disable_irq(), __enable_irq()) are used.

How long does it take from the execution of the interrupt disable function to the reflection on the CPU at this time?

Or is there a mechanism to access the variable a after the interrupt disable is reflected in the structure of the CPU?

If there is a manual regarding the behavior after executing the interrupt prohibition function, please let me know.​

[TDK]

What chip family? You can find cycles per instruction on the M0 and M4 cores, but the M7 core is not as simple to calculate. Even on the M0/M4 cores it is not always straightforward.

https://developer.arm.com/documentation/ddi0439/b/CHDDIGAC

Keep in mind the compiler is free to rearrange/optimize as it sees fit. Use volatile if you don't want a variable optimized out.

[Fkasa.1]

@TDK​ 

Thank you for answering.

Currently We use M4 cores.

Even if interrupt disable is set in the register using __disable_irq(), We are concerned that variables will be accessed from the main function and interrupt function at the same time during the delay until the CPU actually disables interrupts.

[TDK]

You think there is a delay between __disable_irq and when interrupts get disabled? Why?
Gating access to a variable within those calls is sufficient to ensure no interrupts will be called or access it at the same time.

[Fkasa.1]

I think it takes a few CLKs to write the disable_irq register and it works.

If main function are interrupted between a few CLKs, program cause unexpected behavior.

>>Gating access to a variable within those calls is sufficient to ensure no >>interrupts will be called or access it at the same time.

We can rest assured that there is somewhere in the text that shows that the above is guaranteed.

[Nikita91]

If your application is as simple as your example code, you can't have the behavior you describe.

Are you sure you don't use a disable/enable pair in a function called from inside a protected code?

Using __disable_irq() / __enable_irq() is dangerous.

You must save the PRIMASK before you disable, and restore PRIMASK to enable again.

[Piranha · Accepted solution]

> We can rest assured that there is somewhere in the text that shows that the above is guaranteed.

https://developer.arm.com/documentation/dai0321/latest

Section: "4.8 Disabling interrupts using CPS and MSR instructions". Also I recommend reading at least sections 4.5-4.9, especially the 4.9.

[Fkasa.1]

Thank you for answering.

We read this section, and This is what we were looking for.

We will use __disable_irq() and __enable_irq() with confidence from now on

[Fkasa.1]

That's right.

We need to check the PriMask in the actual code.