1. Patchew
  2. QEMU
  3. [Qemu-devel] [PATCH v8 0/7] trace: [tcg] Optimize per-vCPU tracing states
  4. View patch

[Qemu-devel] [PATCH v8 2/7] cpu: allocate cpu->trace_dstate in place

Emilio G. Cota posted 7 patches 8 years, 4 months ago
[Qemu-devel] [PATCH v8 2/7] cpu: allocate cpu->trace_dstate in place
Posted by Emilio G. Cota 8 years, 4 months ago 
There's little point in dynamically allocating the bitmap if we
know at compile-time the max number of events we want to support.
Thus, make room in the struct for the bitmap, which will make things
easier later: this paves the way for upcoming changes, in which
we'll use a u32 to fully capture cpu->trace_dstate.

This change also increases performance by saving a dereference and
improving locality--note that this is important since upcoming work
makes reading this bitmap fairly common.

Signed-off-by: Emilio G. Cota <cota@braap.org>
---
 include/qom/cpu.h | 9 +++------
 qom/cpu.c         | 8 --------
 trace/control.c   | 9 ++++++++-
 3 files changed, 11 insertions(+), 15 deletions(-)

diff --git a/include/qom/cpu.h b/include/qom/cpu.h
index 89ddb68..bc6e20f 100644
--- a/include/qom/cpu.h
+++ b/include/qom/cpu.h
@@ -259,6 +259,7 @@ typedef void (*run_on_cpu_func)(CPUState *cpu, run_on_cpu_data data);
 struct qemu_work_item;
 
 #define CPU_UNSET_NUMA_NODE_ID -1
+#define CPU_TRACE_DSTATE_MAX_EVENTS 32
 
 /**
  * CPUState:
@@ -373,12 +374,8 @@ struct CPUState {
     struct KVMState *kvm_state;
     struct kvm_run *kvm_run;
 
-    /*
-     * Used for events with 'vcpu' and *without* the 'disabled' properties.
-     * Dynamically allocated based on bitmap requried to hold up to
-     * trace_get_vcpu_event_count() entries.
-     */
-    unsigned long *trace_dstate;
+    /* Used for events with 'vcpu' and *without* the 'disabled' properties */
+    DECLARE_BITMAP(trace_dstate, CPU_TRACE_DSTATE_MAX_EVENTS);
 
     /* TODO Move common fields from CPUArchState here. */
     int cpu_index; /* used by alpha TCG */
diff --git a/qom/cpu.c b/qom/cpu.c
index 5069876..69fbb9c 100644
--- a/qom/cpu.c
+++ b/qom/cpu.c
@@ -382,7 +382,6 @@ static void cpu_common_unrealizefn(DeviceState *dev, Error **errp)
 
 static void cpu_common_initfn(Object *obj)
 {
-    uint32_t count;
     CPUState *cpu = CPU(obj);
     CPUClass *cc = CPU_GET_CLASS(obj);
 
@@ -397,18 +396,11 @@ static void cpu_common_initfn(Object *obj)
     QTAILQ_INIT(&cpu->breakpoints);
     QTAILQ_INIT(&cpu->watchpoints);
 
-    count = trace_get_vcpu_event_count();
-    if (count) {
-        cpu->trace_dstate = bitmap_new(count);
-    }
-
     cpu_exec_initfn(cpu);
 }
 
 static void cpu_common_finalize(Object *obj)
 {
-    CPUState *cpu = CPU(obj);
-    g_free(cpu->trace_dstate);
 }
 
 static int64_t cpu_common_get_arch_id(CPUState *cpu)
diff --git a/trace/control.c b/trace/control.c
index 9b157b0..83740aa 100644
--- a/trace/control.c
+++ b/trace/control.c
@@ -65,8 +65,15 @@ void trace_event_register_group(TraceEvent **events)
     size_t i;
     for (i = 0; events[i] != NULL; i++) {
         events[i]->id = next_id++;
-        if (events[i]->vcpu_id != TRACE_VCPU_EVENT_NONE) {
+        if (events[i]->vcpu_id == TRACE_VCPU_EVENT_NONE) {
+            continue;
+        }
+
+        if (likely(next_vcpu_id < CPU_TRACE_DSTATE_MAX_EVENTS)) {
             events[i]->vcpu_id = next_vcpu_id++;
+        } else {
+            error_report("WARNING: too many vcpu trace events; dropping '%s'",
+                         events[i]->name);
         }
     }
     event_groups = g_renew(TraceEventGroup, event_groups, nevent_groups + 1);
-- 
2.7.4
Re: [Qemu-devel] [PATCH v8 2/7] cpu: allocate cpu->trace_dstate in place
Posted by Lluís Vilanova 8 years, 4 months ago 
Emilio G Cota writes:

> There's little point in dynamically allocating the bitmap if we
> know at compile-time the max number of events we want to support.
> Thus, make room in the struct for the bitmap, which will make things
> easier later: this paves the way for upcoming changes, in which
> we'll use a u32 to fully capture cpu->trace_dstate.

> This change also increases performance by saving a dereference and
> improving locality--note that this is important since upcoming work
> makes reading this bitmap fairly common.

> Signed-off-by: Emilio G. Cota <cota@braap.org>

Reviewed-by: Lluís Vilanova <vilanova@ac.upc.edu>


> ---
>  include/qom/cpu.h | 9 +++------
>  qom/cpu.c         | 8 --------
>  trace/control.c   | 9 ++++++++-
>  3 files changed, 11 insertions(+), 15 deletions(-)

> diff --git a/include/qom/cpu.h b/include/qom/cpu.h
> index 89ddb68..bc6e20f 100644
> --- a/include/qom/cpu.h
> +++ b/include/qom/cpu.h
> @@ -259,6 +259,7 @@ typedef void (*run_on_cpu_func)(CPUState *cpu, run_on_cpu_data data);
>  struct qemu_work_item;
 
>  #define CPU_UNSET_NUMA_NODE_ID -1
> +#define CPU_TRACE_DSTATE_MAX_EVENTS 32
 
>  /**
>   * CPUState:
> @@ -373,12 +374,8 @@ struct CPUState {
>      struct KVMState *kvm_state;
>      struct kvm_run *kvm_run;
 
> -    /*
> -     * Used for events with 'vcpu' and *without* the 'disabled' properties.
> -     * Dynamically allocated based on bitmap requried to hold up to
> -     * trace_get_vcpu_event_count() entries.
> -     */
> -    unsigned long *trace_dstate;
> +    /* Used for events with 'vcpu' and *without* the 'disabled' properties */
> +    DECLARE_BITMAP(trace_dstate, CPU_TRACE_DSTATE_MAX_EVENTS);
 
>      /* TODO Move common fields from CPUArchState here. */
>      int cpu_index; /* used by alpha TCG */
> diff --git a/qom/cpu.c b/qom/cpu.c
> index 5069876..69fbb9c 100644
> --- a/qom/cpu.c
> +++ b/qom/cpu.c
> @@ -382,7 +382,6 @@ static void cpu_common_unrealizefn(DeviceState *dev, Error **errp)
 
>  static void cpu_common_initfn(Object *obj)
>  {
> -    uint32_t count;
>      CPUState *cpu = CPU(obj);
>      CPUClass *cc = CPU_GET_CLASS(obj);
 
> @@ -397,18 +396,11 @@ static void cpu_common_initfn(Object *obj)
>      QTAILQ_INIT(&cpu->breakpoints);
>      QTAILQ_INIT(&cpu->watchpoints);
 
> -    count = trace_get_vcpu_event_count();
> -    if (count) {
> -        cpu->trace_dstate = bitmap_new(count);
> -    }
> -
>      cpu_exec_initfn(cpu);
>  }
 
>  static void cpu_common_finalize(Object *obj)
>  {
> -    CPUState *cpu = CPU(obj);
> -    g_free(cpu->trace_dstate);
>  }
 
>  static int64_t cpu_common_get_arch_id(CPUState *cpu)
> diff --git a/trace/control.c b/trace/control.c
> index 9b157b0..83740aa 100644
> --- a/trace/control.c
> +++ b/trace/control.c
> @@ -65,8 +65,15 @@ void trace_event_register_group(TraceEvent **events)
>      size_t i;
>      for (i = 0; events[i] != NULL; i++) {
>          events[i]->id = next_id++;
> -        if (events[i]->vcpu_id != TRACE_VCPU_EVENT_NONE) {
> +        if (events[i]->vcpu_id == TRACE_VCPU_EVENT_NONE) {
> +            continue;
> +        }
> +
> +        if (likely(next_vcpu_id < CPU_TRACE_DSTATE_MAX_EVENTS)) {
>              events[i]->vcpu_id = next_vcpu_id++;
> +        } else {
> +            error_report("WARNING: too many vcpu trace events; dropping '%s'",
> +                         events[i]->name);
>          }
>      }
>      event_groups = g_renew(TraceEventGroup, event_groups, nevent_groups + 1);
> -- 
> 2.7.4
Re: [Qemu-devel] [PATCH v8 2/7] cpu: allocate cpu->trace_dstate in place
Posted by Lluís Vilanova 8 years, 4 months ago 
Lluís Vilanova writes:

> Emilio G Cota writes:
>> There's little point in dynamically allocating the bitmap if we
>> know at compile-time the max number of events we want to support.
>> Thus, make room in the struct for the bitmap, which will make things
>> easier later: this paves the way for upcoming changes, in which
>> we'll use a u32 to fully capture cpu->trace_dstate.

>> This change also increases performance by saving a dereference and
>> improving locality--note that this is important since upcoming work
>> makes reading this bitmap fairly common.

>> Signed-off-by: Emilio G. Cota <cota@braap.org>

> Reviewed-by: Lluís Vilanova <vilanova@ac.upc.edu>

BTW, I think this partially undoes Daniel's changes in
b7d48952c375842bd669460fd8384d90cc12286c.

You should check with him (CC'ed).

Lluis


>> ---
>> include/qom/cpu.h | 9 +++------
>> qom/cpu.c         | 8 --------
>> trace/control.c   | 9 ++++++++-
>> 3 files changed, 11 insertions(+), 15 deletions(-)

>> diff --git a/include/qom/cpu.h b/include/qom/cpu.h
>> index 89ddb68..bc6e20f 100644
>> --- a/include/qom/cpu.h
>> +++ b/include/qom/cpu.h
>> @@ -259,6 +259,7 @@ typedef void (*run_on_cpu_func)(CPUState *cpu, run_on_cpu_data data);
>> struct qemu_work_item;
 
>> #define CPU_UNSET_NUMA_NODE_ID -1
>> +#define CPU_TRACE_DSTATE_MAX_EVENTS 32
 
>> /**
>> * CPUState:
>> @@ -373,12 +374,8 @@ struct CPUState {
>> struct KVMState *kvm_state;
>> struct kvm_run *kvm_run;
 
>> -    /*
>> -     * Used for events with 'vcpu' and *without* the 'disabled' properties.
>> -     * Dynamically allocated based on bitmap requried to hold up to
>> -     * trace_get_vcpu_event_count() entries.
>> -     */
>> -    unsigned long *trace_dstate;
>> +    /* Used for events with 'vcpu' and *without* the 'disabled' properties */
>> +    DECLARE_BITMAP(trace_dstate, CPU_TRACE_DSTATE_MAX_EVENTS);
 
>> /* TODO Move common fields from CPUArchState here. */
>> int cpu_index; /* used by alpha TCG */
>> diff --git a/qom/cpu.c b/qom/cpu.c
>> index 5069876..69fbb9c 100644
>> --- a/qom/cpu.c
>> +++ b/qom/cpu.c
>> @@ -382,7 +382,6 @@ static void cpu_common_unrealizefn(DeviceState *dev, Error **errp)
 
>> static void cpu_common_initfn(Object *obj)
>> {
>> -    uint32_t count;
>> CPUState *cpu = CPU(obj);
>> CPUClass *cc = CPU_GET_CLASS(obj);
 
>> @@ -397,18 +396,11 @@ static void cpu_common_initfn(Object *obj)
>> QTAILQ_INIT(&cpu->breakpoints);
>> QTAILQ_INIT(&cpu->watchpoints);
 
>> -    count = trace_get_vcpu_event_count();
>> -    if (count) {
>> -        cpu->trace_dstate = bitmap_new(count);
>> -    }
>> -
>> cpu_exec_initfn(cpu);
>> }
 
>> static void cpu_common_finalize(Object *obj)
>> {
>> -    CPUState *cpu = CPU(obj);
>> -    g_free(cpu->trace_dstate);
>> }
 
>> static int64_t cpu_common_get_arch_id(CPUState *cpu)
>> diff --git a/trace/control.c b/trace/control.c
>> index 9b157b0..83740aa 100644
>> --- a/trace/control.c
>> +++ b/trace/control.c
>> @@ -65,8 +65,15 @@ void trace_event_register_group(TraceEvent **events)
>> size_t i;
>> for (i = 0; events[i] != NULL; i++) {
>> events[i]->id = next_id++;
>> -        if (events[i]->vcpu_id != TRACE_VCPU_EVENT_NONE) {
>> +        if (events[i]->vcpu_id == TRACE_VCPU_EVENT_NONE) {
>> +            continue;
>> +        }
>> +
>> +        if (likely(next_vcpu_id < CPU_TRACE_DSTATE_MAX_EVENTS)) {
>> events[i]->vcpu_id = next_vcpu_id++;
>> +        } else {
>> +            error_report("WARNING: too many vcpu trace events; dropping '%s'",
>> +                         events[i]->name);
>> }
>> }
>> event_groups = g_renew(TraceEventGroup, event_groups, nevent_groups + 1);
>> -- 
>> 2.7.4
Re: [Qemu-devel] [PATCH v8 2/7] cpu: allocate cpu->trace_dstate in place
Posted by Daniel P. Berrange 8 years, 4 months ago 
On Sun, Jun 25, 2017 at 12:41:57PM +0300, Lluís Vilanova wrote:
> Lluís Vilanova writes:
> 
> > Emilio G Cota writes:
> >> There's little point in dynamically allocating the bitmap if we
> >> know at compile-time the max number of events we want to support.
> >> Thus, make room in the struct for the bitmap, which will make things
> >> easier later: this paves the way for upcoming changes, in which
> >> we'll use a u32 to fully capture cpu->trace_dstate.
> 
> >> This change also increases performance by saving a dereference and
> >> improving locality--note that this is important since upcoming work
> >> makes reading this bitmap fairly common.
> 
> >> Signed-off-by: Emilio G. Cota <cota@braap.org>
> 
> > Reviewed-by: Lluís Vilanova <vilanova@ac.upc.edu>
> 
> BTW, I think this partially undoes Daniel's changes in
> b7d48952c375842bd669460fd8384d90cc12286c.
> 
> You should check with him (CC'ed).

That's ok - I only made those changes in order to remove the reliance on
the generated max vcpu event ID counter. Choosing to hardcode a fixed
limit on number of vcpu events instead is a fine alternative.

> 
> Lluis
> 
> 
> >> ---
> >> include/qom/cpu.h | 9 +++------
> >> qom/cpu.c         | 8 --------
> >> trace/control.c   | 9 ++++++++-
> >> 3 files changed, 11 insertions(+), 15 deletions(-)
> 
> >> diff --git a/include/qom/cpu.h b/include/qom/cpu.h
> >> index 89ddb68..bc6e20f 100644
> >> --- a/include/qom/cpu.h
> >> +++ b/include/qom/cpu.h
> >> @@ -259,6 +259,7 @@ typedef void (*run_on_cpu_func)(CPUState *cpu, run_on_cpu_data data);
> >> struct qemu_work_item;
>  
> >> #define CPU_UNSET_NUMA_NODE_ID -1
> >> +#define CPU_TRACE_DSTATE_MAX_EVENTS 32
>  
> >> /**
> >> * CPUState:
> >> @@ -373,12 +374,8 @@ struct CPUState {
> >> struct KVMState *kvm_state;
> >> struct kvm_run *kvm_run;
>  
> >> -    /*
> >> -     * Used for events with 'vcpu' and *without* the 'disabled' properties.
> >> -     * Dynamically allocated based on bitmap requried to hold up to
> >> -     * trace_get_vcpu_event_count() entries.
> >> -     */
> >> -    unsigned long *trace_dstate;
> >> +    /* Used for events with 'vcpu' and *without* the 'disabled' properties */
> >> +    DECLARE_BITMAP(trace_dstate, CPU_TRACE_DSTATE_MAX_EVENTS);
>  
> >> /* TODO Move common fields from CPUArchState here. */
> >> int cpu_index; /* used by alpha TCG */
> >> diff --git a/qom/cpu.c b/qom/cpu.c
> >> index 5069876..69fbb9c 100644
> >> --- a/qom/cpu.c
> >> +++ b/qom/cpu.c
> >> @@ -382,7 +382,6 @@ static void cpu_common_unrealizefn(DeviceState *dev, Error **errp)
>  
> >> static void cpu_common_initfn(Object *obj)
> >> {
> >> -    uint32_t count;
> >> CPUState *cpu = CPU(obj);
> >> CPUClass *cc = CPU_GET_CLASS(obj);
>  
> >> @@ -397,18 +396,11 @@ static void cpu_common_initfn(Object *obj)
> >> QTAILQ_INIT(&cpu->breakpoints);
> >> QTAILQ_INIT(&cpu->watchpoints);
>  
> >> -    count = trace_get_vcpu_event_count();
> >> -    if (count) {
> >> -        cpu->trace_dstate = bitmap_new(count);
> >> -    }
> >> -
> >> cpu_exec_initfn(cpu);
> >> }
>  
> >> static void cpu_common_finalize(Object *obj)
> >> {
> >> -    CPUState *cpu = CPU(obj);
> >> -    g_free(cpu->trace_dstate);
> >> }
>  
> >> static int64_t cpu_common_get_arch_id(CPUState *cpu)
> >> diff --git a/trace/control.c b/trace/control.c
> >> index 9b157b0..83740aa 100644
> >> --- a/trace/control.c
> >> +++ b/trace/control.c
> >> @@ -65,8 +65,15 @@ void trace_event_register_group(TraceEvent **events)
> >> size_t i;
> >> for (i = 0; events[i] != NULL; i++) {
> >> events[i]->id = next_id++;
> >> -        if (events[i]->vcpu_id != TRACE_VCPU_EVENT_NONE) {
> >> +        if (events[i]->vcpu_id == TRACE_VCPU_EVENT_NONE) {
> >> +            continue;
> >> +        }
> >> +
> >> +        if (likely(next_vcpu_id < CPU_TRACE_DSTATE_MAX_EVENTS)) {
> >> events[i]->vcpu_id = next_vcpu_id++;
> >> +        } else {
> >> +            error_report("WARNING: too many vcpu trace events; dropping '%s'",
> >> +                         events[i]->name);
> >> }

This should be an abort IMHO, as it would be considered a bug to have
added > 32 vcpu events.

I'd also suggest that the top level 'trace-events' file get a comment
added to the effect that we only support 32 events right now.

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

Patchew 2.1-devel-b67e4c2

© 2016 - 2025 Red Hat, Inc.