On 21.03.2013, at 09:45, Jan Kiszka wrote:
> On 2013-03-21 09:34, Alexander Graf wrote:
>>
>> On 14.02.2013, at 07:32, Scott Wood wrote:
>>
>>> This allows platform code to register in-kernel irqchips that
>>> don't use the legacy KVM_CAP_IRQCHIP/KVM_CREATE_IRQCHIP interface.
>>>
>>> Signed-off-by: Scott Wood <scottw...@freescale.com>
>>> ---
>>> include/sysemu/kvm.h | 10 ++++++++++
>>> kvm-all.c | 11 +++++++++--
>>> 2 files changed, 19 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/include/sysemu/kvm.h b/include/sysemu/kvm.h
>>> index f2d97b5..b9a8701 100644
>>> --- a/include/sysemu/kvm.h
>>> +++ b/include/sysemu/kvm.h
>>> @@ -45,6 +45,7 @@ extern bool kvm_async_interrupts_allowed;
>>> extern bool kvm_irqfds_allowed;
>>> extern bool kvm_msi_via_irqfd_allowed;
>>> extern bool kvm_gsi_routing_allowed;
>>> +extern bool kvm_irqchip_wanted;
>>>
>>> #if defined CONFIG_KVM || !defined NEED_CPU_H
>>> #define kvm_enabled() (kvm_allowed)
>>> @@ -97,6 +98,14 @@ extern bool kvm_gsi_routing_allowed;
>>> */
>>> #define kvm_gsi_routing_enabled() (kvm_gsi_routing_allowed)
>>>
>>> +/**
>>> + * kvm_irqchip_wanted
>>> + *
>>> + * Returns: true if the user requested that an in-kernel IRQ chip be
>>> + * used, regardless of whether support has been detected.
>>> + */
>>> +#define kvm_irqchip_wanted() (kvm_irqchip_wanted)
>>> +
>>> #else
>>> #define kvm_enabled() (0)
>>> #define kvm_irqchip_in_kernel() (false)
>>> @@ -104,6 +113,7 @@ extern bool kvm_gsi_routing_allowed;
>>> #define kvm_irqfds_enabled() (false)
>>> #define kvm_msi_via_irqfd_enabled() (false)
>>> #define kvm_gsi_routing_allowed() (false)
>>> +#define kvm_irqchip_wanted() (false)
>>> #endif
>>>
>>> struct kvm_run;
>>> diff --git a/kvm-all.c b/kvm-all.c
>>> index 04ec2d5..13a628d 100644
>>> --- a/kvm-all.c
>>> +++ b/kvm-all.c
>>> @@ -109,6 +109,7 @@ bool kvm_async_interrupts_allowed;
>>> bool kvm_irqfds_allowed;
>>> bool kvm_msi_via_irqfd_allowed;
>>> bool kvm_gsi_routing_allowed;
>>> +bool kvm_irqchip_wanted;
>>>
>>> static const KVMCapabilityInfo kvm_required_capabilites[] = {
>>> KVM_CAP_INFO(USER_MEMORY),
>>> @@ -1205,8 +1206,14 @@ static int kvm_irqchip_create(KVMState *s)
>>>
>>> if (QTAILQ_EMPTY(&list->head) ||
>>> !qemu_opt_get_bool(QTAILQ_FIRST(&list->head),
>>> - "kernel_irqchip", true) ||
>>> - !kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
>>> + "kernel_irqchip", true)) {
>>> + return 0;
>>> + }
>>> +
>>> + kvm_irqchip_wanted = true;
>>> +
>>> + /* Platform code may have a different way of enabling an IRQ chip */
>>> + if (!kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
>>
>> Does x86 have the required checks then to make sure it has the CAP?
>
> To my understanding, x86 won't evaluate this new flag but continue to
> bail out from this service early.
>
> However, invoking something that is called "create" to just end up with
> a set flag "wanted" and then do the creation elsewhere is not a very
> beautiful design.
Yeah. This should probably be
kvm_irqchip_wanted(KVMState *s) {
if (!kvm_arch_irqchip_available(s))
return false;
return qemu_opt_get_bool("kernel_irqchip", true);
}
with kvm_arch_irqchip_available checking for KVM_CAP_IRQCHIP on x86 and the
respective device creation capability on ppc.
Alex
