[Qemu-devel] why do qem/arm not clear CPU_INTERRUPT_HARD bit of env->interrupt_request automatically?

[Qemu-devel] why do qem/arm not clear CPU_INTERRUPT_HARD bit of env->interrupt_request automatically?

[Donald Liew]

i'm reading qemu-system-arm code and trying to add some more
evaluation boards support for it. in the 0.8.2 source code i found
something i can't understand, when handling interrupts all other
targets clears this bit after calling do_interrupt, however the arm
target doesn't do this, why? won't this cause problems like redundant
interrupts? any special consideration about this?

donald

Re: [Qemu-devel] why do qem/arm not clear CPU_INTERRUPT_HARD bit of env->interrupt_request automatically?

[Paul Brook]

On Saturday 07 October 2006 19:15, Donald Liew wrote:
> i'm reading qemu-system-arm code and trying to add some more
> evaluation boards support for it. in the 0.8.2 source code i found
> something i can't understand, when handling interrupts all other
> targets clears this bit after calling do_interrupt, however the arm
> target doesn't do this, why? won't this cause problems like redundant
> interrupts? any special consideration about this?

I this this is the correct behavior. The nIRQ line is level triggered. 
spurious interrupts are avoided because do_interrupt sets the CPSR_I flag.

During normal operation theguest OS will clear the IRQ condition (by masking 
the interrupt on the PIC or device) before clearing the CPSE_I flag, so it 
doesn't matter what we do.

Consider the case where the guest OS were to clear CPSR_F without touching the 
IRQ line. In this case we would expect annother IRQ exception to be taken 
immediately. If (as you suggest above) we cleared CPU_INTERRUPT_HARD in 
cpu_exec then the IRQ would not be taken until something re-raised the IRQ 
line.

Paul

Re: [Qemu-devel] why do qem/arm not clear CPU_INTERRUPT_HARD bit of env->interrupt_request automatically?

[Donald Liew]

[-- Attachment #1: Type: text/plain, Size: 1222 bytes --]

i see. you are right, thanks very much

2006/10/8, Paul Brook <paul@codesourcery.com>:
>
> On Saturday 07 October 2006 19:15, Donald Liew wrote:
> > i'm reading qemu-system-arm code and trying to add some more
> > evaluation boards support for it. in the 0.8.2 source code i found
> > something i can't understand, when handling interrupts all other
> > targets clears this bit after calling do_interrupt, however the arm
> > target doesn't do this, why? won't this cause problems like redundant
> > interrupts? any special consideration about this?
>
> I this this is the correct behavior. The nIRQ line is level triggered.
> spurious interrupts are avoided because do_interrupt sets the CPSR_I flag.
>
> During normal operation theguest OS will clear the IRQ condition (by
> masking
> the interrupt on the PIC or device) before clearing the CPSE_I flag, so it
> doesn't matter what we do.
>
> Consider the case where the guest OS were to clear CPSR_F without touching
> the
> IRQ line. In this case we would expect annother IRQ exception to be taken
> immediately. If (as you suggest above) we cleared CPU_INTERRUPT_HARD in
> cpu_exec then the IRQ would not be taken until something re-raised the IRQ
> line.
>
> Paul
>

[-- Attachment #2: Type: text/html, Size: 1508 bytes --]