fix(x86_64): don't set CR0Flags::TASK_SWITCHED

[sarahspberrypi]

As previously mentioned, the manual setting of this flag seems to be unnecessary, so I propose to remove it from the code.

[mkroening]

I have done some digging on why this was added:

kernel/librs/src/arch/x86_64/processor.rs

Lines 489 to 490 in 77bd5d1

	// Call the IRQ7 handler on the first FPU access.
	cr0.insert(CR0_TASK_SWITCHED);

kernel/librs/src/arch/x86_64/irq.rs

Lines 211 to 218 in 77bd5d1

	extern "x86-interrupt" fn device_not_available_exception(stack_frame: &mut ExceptionStackFrame) {
	// We set the CR0_TASK_SWITCHED flag every time we switch to a task.
	// This causes the "Device Not Available" Exception (int #7) to be thrown as soon as we use the FPU for the first time.
	// We have to clear the CR0_TASK_SWITCHED here and save the FPU context of the old task.
	unsafe { asm!("clts"); }
	panic!("FPU ToDo");
	}

If the processor executes a floating-point instruction while the TS flag in control register CR0 was set and the EM flag is clear, a device-not-available exception is generated.
I understand this was added intentionally to ensure the behavior for the initial task is the same as for subsequent tasks that were switched to and have set the TS flag implicitly.

I have not done enough digging to say whether it would be fine to remove this, but maybe you don't have a problem with the TS flag but with running interrupts in general.

[stlankes]

We should accept the PR. I have the same opinion like @sarahspberrypi

[stlankes]

@sarahspberrypi I talked to @mkroening We would like to set TASK_SWITCHED flag. Do you really need to remove the flag?

[sarahspberrypi]

@sarahspberrypi I talked to @mkroening We would like to set TASK_SWITCHED flag. Do you really need to remove the flag?

@stlankes The flag is set later during switch_to_task() call anyways (see here ), so I don't really see why we need to have it pre-enabled?
With SEV enabled, this apparently triggers an immediate hardware interrupt that cannot be masked. This interrupt can only be handled by hermit later when the IDT is loaded. But I'll talk with @mkroening in the new year about this.