Re: [PATCH v16 21/23] accel: introduce new accessor functions

2021-02-14 Thread Claudio Fontana 
On 2/8/21 1:54 PM, Philippe Mathieu-Daudé wrote:
> On 2/8/21 1:50 PM, Claudio Fontana wrote:
>> On 2/5/21 9:14 PM, Richard Henderson wrote:
>>> On 2/4/21 6:39 AM, Claudio Fontana wrote:
 @@ -6725,10 +6722,8 @@ static void x86_cpu_initfn(Object *obj)
  x86_cpu_load_model(cpu, xcc->model);
  }
  
 -/* if required, do the accelerator-specific cpu initialization */
 -if (cc->accel_cpu) {
 -cc->accel_cpu->cpu_instance_init(CPU(obj));
 -}
 +/* if required, do accelerator-specific cpu initializations */
 +accel_cpu_instance_init(CPU(obj));
  }
>>>
>>> Why is this only done for x86?
>>>
>>>
>>> r~
>>>
>>
>> It makes sense to include the other architectures.
>>
>> As the next step I would like to apply this to ARM, but to me it makes sense 
>> to first complete Philippe's series,
>> which reshuffles things so that TCG-only / KVM-only builds are both possible 
>> and error-free:
>>
>> https://www.mail-archive.com/qemu-devel@nongnu.org/msg777627.html
> 
> TBH this series is very unlikely to be merged before yours,
> so go ahead... (eventually you can cherry-pick what you need
> from it).
> 

Hi Philippe, Peter,

I am working on ARM right now,

splitting things between user/system, tcg/kvm.

One difficulty I found in particular is with the ARM PMU code.

Currently the PMU code is mixed in with the TCG helpers, in target/arm/helper.c.

Now, with KVM we should be using the KVM PMU, but still the KVM code currently 
calls pmu_init and registers the timer:

In target/arm/cpu.c initialization we see:

"
if (arm_feature(env, ARM_FEATURE_PMU)) {
pmu_init(cpu);

if (!kvm_enabled()) {
arm_register_pre_el_change_hook(cpu, _pre_el_change, 0);
arm_register_el_change_hook(cpu, _post_el_change, 0);
}

#ifndef CONFIG_USER_ONLY
cpu->pmu_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, arm_pmu_timer_cb,
cpu);
#endif
"

Now, even for KVM, pmu_init is called and the cpu->pmu_timer is registered,
but I don't see arm_pmu_timer_cb triggering.

Is the pmu_timer really necessary also for KVM builds, or should it actually be 
TCG-only?

Could you share your hints on which parts of the PMU code in helper.c should 
actually be shared if any?

Thanks a lot,

Claudio

Re: [PATCH v16 21/23] accel: introduce new accessor functions

2021-02-10 Thread Claudio Fontana 
On 2/8/21 1:54 PM, Philippe Mathieu-Daudé wrote:
> On 2/8/21 1:50 PM, Claudio Fontana wrote:
>> On 2/5/21 9:14 PM, Richard Henderson wrote:
>>> On 2/4/21 6:39 AM, Claudio Fontana wrote:
 @@ -6725,10 +6722,8 @@ static void x86_cpu_initfn(Object *obj)
  x86_cpu_load_model(cpu, xcc->model);
  }
  
 -/* if required, do the accelerator-specific cpu initialization */
 -if (cc->accel_cpu) {
 -cc->accel_cpu->cpu_instance_init(CPU(obj));
 -}
 +/* if required, do accelerator-specific cpu initializations */
 +accel_cpu_instance_init(CPU(obj));
  }
>>>
>>> Why is this only done for x86?
>>>
>>>
>>> r~
>>>
>>
>> It makes sense to include the other architectures.
>>
>> As the next step I would like to apply this to ARM, but to me it makes sense 
>> to first complete Philippe's series,
>> which reshuffles things so that TCG-only / KVM-only builds are both possible 
>> and error-free:
>>
>> https://www.mail-archive.com/qemu-devel@nongnu.org/msg777627.html
> 
> TBH this series is very unlikely to be merged before yours,
> so go ahead... (eventually you can cherry-pick what you need
> from it).
> 
> 

After taking a short look at the ARM side, I think I need first a detour to 
split more CONFIG_USER_ONLY code from the rest.

I'll experiment with using meson to add a target-only user_ss variable, and 
apply it to i386 and arm.

Ciao,

Claudio

Re: [PATCH v16 21/23] accel: introduce new accessor functions

2021-02-08 Thread Philippe Mathieu-Daudé 
On 2/8/21 1:50 PM, Claudio Fontana wrote:
> On 2/5/21 9:14 PM, Richard Henderson wrote:
>> On 2/4/21 6:39 AM, Claudio Fontana wrote:
>>> @@ -6725,10 +6722,8 @@ static void x86_cpu_initfn(Object *obj)
>>>  x86_cpu_load_model(cpu, xcc->model);
>>>  }
>>>  
>>> -/* if required, do the accelerator-specific cpu initialization */
>>> -if (cc->accel_cpu) {
>>> -cc->accel_cpu->cpu_instance_init(CPU(obj));
>>> -}
>>> +/* if required, do accelerator-specific cpu initializations */
>>> +accel_cpu_instance_init(CPU(obj));
>>>  }
>>
>> Why is this only done for x86?
>>
>>
>> r~
>>
> 
> It makes sense to include the other architectures.
> 
> As the next step I would like to apply this to ARM, but to me it makes sense 
> to first complete Philippe's series,
> which reshuffles things so that TCG-only / KVM-only builds are both possible 
> and error-free:
> 
> https://www.mail-archive.com/qemu-devel@nongnu.org/msg777627.html

TBH this series is very unlikely to be merged before yours,
so go ahead... (eventually you can cherry-pick what you need
from it).

Re: [PATCH v16 21/23] accel: introduce new accessor functions

2021-02-08 Thread Claudio Fontana 
On 2/5/21 9:14 PM, Richard Henderson wrote:
> On 2/4/21 6:39 AM, Claudio Fontana wrote:
>> @@ -6725,10 +6722,8 @@ static void x86_cpu_initfn(Object *obj)
>>  x86_cpu_load_model(cpu, xcc->model);
>>  }
>>  
>> -/* if required, do the accelerator-specific cpu initialization */
>> -if (cc->accel_cpu) {
>> -cc->accel_cpu->cpu_instance_init(CPU(obj));
>> -}
>> +/* if required, do accelerator-specific cpu initializations */
>> +accel_cpu_instance_init(CPU(obj));
>>  }
> 
> Why is this only done for x86?
> 
> 
> r~
> 

It makes sense to include the other architectures.

As the next step I would like to apply this to ARM, but to me it makes sense to 
first complete Philippe's series,
which reshuffles things so that TCG-only / KVM-only builds are both possible 
and error-free:

https://www.mail-archive.com/qemu-devel@nongnu.org/msg777627.html

Thanks,

Claudio

Re: [PATCH v16 21/23] accel: introduce new accessor functions

2021-02-05 Thread Richard Henderson 
On 2/4/21 6:39 AM, Claudio Fontana wrote:
> @@ -6725,10 +6722,8 @@ static void x86_cpu_initfn(Object *obj)
>  x86_cpu_load_model(cpu, xcc->model);
>  }
>  
> -/* if required, do the accelerator-specific cpu initialization */
> -if (cc->accel_cpu) {
> -cc->accel_cpu->cpu_instance_init(CPU(obj));
> -}
> +/* if required, do accelerator-specific cpu initializations */
> +accel_cpu_instance_init(CPU(obj));
>  }

Why is this only done for x86?


r~

[PATCH v16 21/23] accel: introduce new accessor functions

2021-02-04 Thread Claudio Fontana 
avoid open coding the accesses to cpu->accel_cpu interfaces,
and instead introduce:

accel_cpu_instance_init,
accel_cpu_realizefn

to be used by the targets/ initfn code,
and by cpu_exec_realizefn respectively.

Signed-off-by: Claudio Fontana 
---
 include/qemu/accel.h | 13 +
 accel/accel-common.c | 19 +++
 cpu.c|  6 +-
 target/i386/cpu.c|  9 ++---
 4 files changed, 35 insertions(+), 12 deletions(-)

diff --git a/include/qemu/accel.h b/include/qemu/accel.h
index b9d6d69eb8..da0c8ab523 100644
--- a/include/qemu/accel.h
+++ b/include/qemu/accel.h
@@ -78,4 +78,17 @@ int accel_init_machine(AccelState *accel, MachineState *ms);
 void accel_setup_post(MachineState *ms);
 #endif /* !CONFIG_USER_ONLY */
 
+/**
+ * accel_cpu_instance_init:
+ * @cpu: The CPU that needs to do accel-specific object initializations.
+ */
+void accel_cpu_instance_init(CPUState *cpu);
+
+/**
+ * accel_cpu_realizefn:
+ * @cpu: The CPU that needs to call accel-specific cpu realization.
+ * @errp: currently unused.
+ */
+void accel_cpu_realizefn(CPUState *cpu, Error **errp);
+
 #endif /* QEMU_ACCEL_H */
diff --git a/accel/accel-common.c b/accel/accel-common.c
index 9901b0531c..0f6fb4fb66 100644
--- a/accel/accel-common.c
+++ b/accel/accel-common.c
@@ -89,6 +89,25 @@ void accel_init_interfaces(AccelClass *ac)
 accel_init_cpu_interfaces(ac);
 }
 
+void accel_cpu_instance_init(CPUState *cpu)
+{
+CPUClass *cc = CPU_GET_CLASS(cpu);
+
+if (cc->accel_cpu && cc->accel_cpu->cpu_instance_init) {
+cc->accel_cpu->cpu_instance_init(cpu);
+}
+}
+
+void accel_cpu_realizefn(CPUState *cpu, Error **errp)
+{
+CPUClass *cc = CPU_GET_CLASS(cpu);
+
+if (cc->accel_cpu && cc->accel_cpu->cpu_realizefn) {
+/* NB: errp parameter is unused currently */
+cc->accel_cpu->cpu_realizefn(cpu, errp);
+}
+}
+
 static const TypeInfo accel_cpu_type = {
 .name = TYPE_ACCEL_CPU,
 .parent = TYPE_OBJECT,
diff --git a/cpu.c b/cpu.c
index ba5d272c1e..25e6fbfa2c 100644
--- a/cpu.c
+++ b/cpu.c
@@ -130,11 +130,7 @@ void cpu_exec_realizefn(CPUState *cpu, Error **errp)
 CPUClass *cc = CPU_GET_CLASS(cpu);
 
 cpu_list_add(cpu);
-
-if (cc->accel_cpu) {
-/* NB: errp parameter is unused currently */
-cc->accel_cpu->cpu_realizefn(cpu, errp);
-}
+accel_cpu_realizefn(cpu, errp);
 
 #ifdef CONFIG_TCG
 /* NB: errp parameter is unused currently */
diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index d7735f0ed3..be068fcc5e 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -28,7 +28,6 @@
 #include "sysemu/kvm.h"
 #include "sysemu/reset.h"
 #include "sysemu/hvf.h"
-#include "hw/core/accel-cpu.h"
 #include "sysemu/xen.h"
 #include "sysemu/whpx.h"
 #include "kvm/kvm_i386.h"
@@ -6675,8 +6674,6 @@ static void x86_cpu_initfn(Object *obj)
 {
 X86CPU *cpu = X86_CPU(obj);
 X86CPUClass *xcc = X86_CPU_GET_CLASS(obj);
-CPUClass *cc = CPU_CLASS(xcc);
-
 CPUX86State *env = >env;
 
 env->nr_dies = 1;
@@ -6725,10 +6722,8 @@ static void x86_cpu_initfn(Object *obj)
 x86_cpu_load_model(cpu, xcc->model);
 }
 
-/* if required, do the accelerator-specific cpu initialization */
-if (cc->accel_cpu) {
-cc->accel_cpu->cpu_instance_init(CPU(obj));
-}
+/* if required, do accelerator-specific cpu initializations */
+accel_cpu_instance_init(CPU(obj));
 }
 
 static int64_t x86_cpu_get_arch_id(CPUState *cs)
-- 
2.26.2