Re: [Qemu-devel] [PATCH v2 1/3] cpu: make CPU_INTERRUPT_RESET available on all targets

[Paolo Bonzini <pbonzini@redhat.com>]

Il 06/03/2013 00:23, Peter Maydell ha scritto:
> On 6 March 2013 03:00, Paolo Bonzini <pbonzini@redhat.com> wrote:
>> On the x86, some devices need access to the CPU reset pin (INIT#).
>> Provide a generic service to do this, using one of the internal
>> cpu_interrupt targets.  Generalize the PPC-specific code for
>> CPU_INTERRUPT_RESET to other targets, and provide a function that
>> will raise the interrupt on all CPUs.
> 
> Not sure this makes sense -- reset isn't an interrupt...

cpu_interrupt is not just for interrupts, CPU_INTERRUPT_TGT_INT_* is a
generic mechanism for adding events to the CPU that have to exit the
translation block (they do not even have to be input pins, though
CPU_INTERRUPT_RESET is).  This patch just takes one particular
CPU_INTERRUPT_TGT_INT_* value and makes it available to all targets.

It is important for the reset to exit the translation block, or the CPU
goes into the weeds.  The problem I was seeing is that the code looked like:

    mov $0xfe, %al
    out %al, $0x60
    jmp foo           // this is a relative jump
...
foo:
    cli
    hlt

Now, if the reset were synchronous (i.e. cpu_reset), it would modify the
stored PC to 0xfffffff0 without leaving the translation block.
Because the jump is relative, it would go to 0xfffffff0 + the offset
instead of jumping to foo.

This could also be implemented by something like this:

    run_on_cpu(cpu, cpu_reset);
    cpu_interrupt(cpu, CPU_INTERRUPT_EXITTB);

But I preferred to reuse the existing logic (there would be some
additional complication because the x86 INIT signal does _not_ reset a
couple of things that are reset at power up).

Paolo