On 20/02/2024 19:26, Philippe Mathieu-Daudé wrote:
Hi,
cpu_interrupt() doesn't scale well with heterogenous machines
because its mask is target specific (defined in target/ARCH/cpu.h).
While it is (often...) used by target-specific hw to notify cpu,
there is no restriction to use such target-specific hw in a
heterogeneous setup, where it'd still target the same kind of
target cpus.
The Alpha Typhoon HW is unlikely to be used heterogeneously,
but it is the simplest case I found to illustrate how I plan
to remove cpu_interrupt() calls from hw/: use the QDev GPIO API.
Does that sound good enough?
I think the basic mechanism of setting/resetting the interrupt using qdev GPIOs
should work, but I wonder if there isn't a better way of defining them to avoid more
#ifdeffery.
Is it possible to come up with some declarative syntax in either CPUClass or
CPUClass::sysemu_ops that would avoid the developer manually having to call
qdev_init_gpio_in_named() manually during CPU init() and/or create the various _irq()
callback functions by hand?
Thanks,
Phil.
Philippe Mathieu-Daudé (2):
target/alpha: Expose TMR and SMP IRQ lines via QDev
hw/alpha/typhoon: Set CPU IRQs using QDev API
hw/alpha/typhoon.c | 15 +++++++++------
target/alpha/cpu.c | 30 ++++++++++++++++++++++++++++++
2 files changed, 39 insertions(+), 6 deletions(-)
ATB,
Mark.
