[PATCH] gdbstub: move update guest debug to accel ops

[PATCH] gdbstub: move update guest debug to accel ops

[Mads Ynddal]

From: Mads Ynddal <m.ynddal@samsung.com>

Continuing the refactor of a48e7d9e52 (gdbstub: move guest debug support
check to ops) by removing hardcoded kvm_enabled() from generic cpu.c
code, and replace it with a property of AccelOpsClass.

Signed-off-by: Mads Ynddal <m.ynddal@samsung.com>
---
 accel/kvm/kvm-accel-ops.c  |  1 +
 cpu.c                      | 10 +++++++---
 include/sysemu/accel-ops.h |  1 +
 3 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/accel/kvm/kvm-accel-ops.c b/accel/kvm/kvm-accel-ops.c
index fbf4fe3497..6ebf9a644f 100644
--- a/accel/kvm/kvm-accel-ops.c
+++ b/accel/kvm/kvm-accel-ops.c
@@ -99,6 +99,7 @@ static void kvm_accel_ops_class_init(ObjectClass *oc, void *data)
     ops->synchronize_pre_loadvm = kvm_cpu_synchronize_pre_loadvm;
 
 #ifdef KVM_CAP_SET_GUEST_DEBUG
+    ops->update_guest_debug = kvm_update_guest_debug;
     ops->supports_guest_debug = kvm_supports_guest_debug;
     ops->insert_breakpoint = kvm_insert_breakpoint;
     ops->remove_breakpoint = kvm_remove_breakpoint;
diff --git a/cpu.c b/cpu.c
index 2a09b05205..ef433a79e3 100644
--- a/cpu.c
+++ b/cpu.c
@@ -31,8 +31,8 @@
 #include "hw/core/sysemu-cpu-ops.h"
 #include "exec/address-spaces.h"
 #endif
+#include "sysemu/cpus.h"
 #include "sysemu/tcg.h"
-#include "sysemu/kvm.h"
 #include "sysemu/replay.h"
 #include "exec/cpu-common.h"
 #include "exec/exec-all.h"
@@ -378,10 +378,14 @@ void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
 void cpu_single_step(CPUState *cpu, int enabled)
 {
     if (cpu->singlestep_enabled != enabled) {
+        const AccelOpsClass *ops = cpus_get_accel();
+
         cpu->singlestep_enabled = enabled;
-        if (kvm_enabled()) {
-            kvm_update_guest_debug(cpu, 0);
+
+        if (ops->update_guest_debug) {
+            ops->update_guest_debug(cpu, 0);
         }
+
         trace_breakpoint_singlestep(cpu->cpu_index, enabled);
     }
 }
diff --git a/include/sysemu/accel-ops.h b/include/sysemu/accel-ops.h
index 8cc7996def..0a47a2f00c 100644
--- a/include/sysemu/accel-ops.h
+++ b/include/sysemu/accel-ops.h
@@ -48,6 +48,7 @@ struct AccelOpsClass {
 
     /* gdbstub hooks */
     bool (*supports_guest_debug)(void);
+    int (*update_guest_debug)(CPUState *cpu, unsigned long flags);
     int (*insert_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
     int (*remove_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
     void (*remove_all_breakpoints)(CPUState *cpu);
-- 
2.38.1

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Mads Ynddal]

> On 23 Nov 2022, at 13.17, Mads Ynddal <mads@ynddal.dk> wrote:
> 
> From: Mads Ynddal <m.ynddal@samsung.com>
> 
> Continuing the refactor of a48e7d9e52 (gdbstub: move guest debug support
> check to ops) by removing hardcoded kvm_enabled() from generic cpu.c
> code, and replace it with a property of AccelOpsClass.
> 
> Signed-off-by: Mads Ynddal <m.ynddal@samsung.com>
> ---
> accel/kvm/kvm-accel-ops.c  |  1 +
> cpu.c                      | 10 +++++++---
> include/sysemu/accel-ops.h |  1 +
> 3 files changed, 9 insertions(+), 3 deletions(-)
> 

+CC Alex Bennée

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Alex Bennée]

Mads Ynddal <mads@ynddal.dk> writes:

> From: Mads Ynddal <m.ynddal@samsung.com>
>
> Continuing the refactor of a48e7d9e52 (gdbstub: move guest debug support
> check to ops) by removing hardcoded kvm_enabled() from generic cpu.c
> code, and replace it with a property of AccelOpsClass.
>
> Signed-off-by: Mads Ynddal <m.ynddal@samsung.com>

Nice. Looks good to me but I'll have a proper look when I go through my
gdbstub/next queue. I don't think this is critical for 7.2.

> ---
>  accel/kvm/kvm-accel-ops.c  |  1 +
>  cpu.c                      | 10 +++++++---
>  include/sysemu/accel-ops.h |  1 +
>  3 files changed, 9 insertions(+), 3 deletions(-)
>
> diff --git a/accel/kvm/kvm-accel-ops.c b/accel/kvm/kvm-accel-ops.c
> index fbf4fe3497..6ebf9a644f 100644
> --- a/accel/kvm/kvm-accel-ops.c
> +++ b/accel/kvm/kvm-accel-ops.c
> @@ -99,6 +99,7 @@ static void kvm_accel_ops_class_init(ObjectClass *oc, void *data)
>      ops->synchronize_pre_loadvm = kvm_cpu_synchronize_pre_loadvm;
>  
>  #ifdef KVM_CAP_SET_GUEST_DEBUG
> +    ops->update_guest_debug = kvm_update_guest_debug;
>      ops->supports_guest_debug = kvm_supports_guest_debug;
>      ops->insert_breakpoint = kvm_insert_breakpoint;
>      ops->remove_breakpoint = kvm_remove_breakpoint;
> diff --git a/cpu.c b/cpu.c
> index 2a09b05205..ef433a79e3 100644
> --- a/cpu.c
> +++ b/cpu.c
> @@ -31,8 +31,8 @@
>  #include "hw/core/sysemu-cpu-ops.h"
>  #include "exec/address-spaces.h"
>  #endif
> +#include "sysemu/cpus.h"
>  #include "sysemu/tcg.h"
> -#include "sysemu/kvm.h"
>  #include "sysemu/replay.h"
>  #include "exec/cpu-common.h"
>  #include "exec/exec-all.h"
> @@ -378,10 +378,14 @@ void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
>  void cpu_single_step(CPUState *cpu, int enabled)
>  {
>      if (cpu->singlestep_enabled != enabled) {
> +        const AccelOpsClass *ops = cpus_get_accel();
> +
>          cpu->singlestep_enabled = enabled;
> -        if (kvm_enabled()) {
> -            kvm_update_guest_debug(cpu, 0);
> +
> +        if (ops->update_guest_debug) {
> +            ops->update_guest_debug(cpu, 0);
>          }
> +
>          trace_breakpoint_singlestep(cpu->cpu_index, enabled);
>      }
>  }
> diff --git a/include/sysemu/accel-ops.h b/include/sysemu/accel-ops.h
> index 8cc7996def..0a47a2f00c 100644
> --- a/include/sysemu/accel-ops.h
> +++ b/include/sysemu/accel-ops.h
> @@ -48,6 +48,7 @@ struct AccelOpsClass {
>  
>      /* gdbstub hooks */
>      bool (*supports_guest_debug)(void);
> +    int (*update_guest_debug)(CPUState *cpu, unsigned long flags);
>      int (*insert_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
>      int (*remove_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
>      void (*remove_all_breakpoints)(CPUState *cpu);


-- 
Alex Bennée

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Mads Ynddal]

> On 23 Nov 2022, at 15.05, Alex Bennée <alex.bennee@linaro.org> wrote:
> 
> Nice. Looks good to me but I'll have a proper look when I go through my
> gdbstub/next queue. I don't think this is critical for 7.2.
> 

Thanks, and I agree. It can easily wait.

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Philippe Mathieu-Daudé]

Hi,

On 23/11/22 13:17, Mads Ynddal wrote:
> From: Mads Ynddal <m.ynddal@samsung.com>
> 
> Continuing the refactor of a48e7d9e52 (gdbstub: move guest debug support
> check to ops) by removing hardcoded kvm_enabled() from generic cpu.c
> code, and replace it with a property of AccelOpsClass.
> 
> Signed-off-by: Mads Ynddal <m.ynddal@samsung.com>
> ---
>   accel/kvm/kvm-accel-ops.c  |  1 +
>   cpu.c                      | 10 +++++++---
>   include/sysemu/accel-ops.h |  1 +
>   3 files changed, 9 insertions(+), 3 deletions(-)
> 
> diff --git a/accel/kvm/kvm-accel-ops.c b/accel/kvm/kvm-accel-ops.c
> index fbf4fe3497..6ebf9a644f 100644
> --- a/accel/kvm/kvm-accel-ops.c
> +++ b/accel/kvm/kvm-accel-ops.c
> @@ -99,6 +99,7 @@ static void kvm_accel_ops_class_init(ObjectClass *oc, void *data)
>       ops->synchronize_pre_loadvm = kvm_cpu_synchronize_pre_loadvm;
>   
>   #ifdef KVM_CAP_SET_GUEST_DEBUG
> +    ops->update_guest_debug = kvm_update_guest_debug;
>       ops->supports_guest_debug = kvm_supports_guest_debug;
>       ops->insert_breakpoint = kvm_insert_breakpoint;
>       ops->remove_breakpoint = kvm_remove_breakpoint;
> diff --git a/cpu.c b/cpu.c
> index 2a09b05205..ef433a79e3 100644
> --- a/cpu.c
> +++ b/cpu.c
> @@ -31,8 +31,8 @@
>   #include "hw/core/sysemu-cpu-ops.h"
>   #include "exec/address-spaces.h"
>   #endif
> +#include "sysemu/cpus.h"
>   #include "sysemu/tcg.h"
> -#include "sysemu/kvm.h"
>   #include "sysemu/replay.h"
>   #include "exec/cpu-common.h"
>   #include "exec/exec-all.h"
> @@ -378,10 +378,14 @@ void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
>   void cpu_single_step(CPUState *cpu, int enabled)
>   {
>       if (cpu->singlestep_enabled != enabled) {
> +        const AccelOpsClass *ops = cpus_get_accel();
> +
>           cpu->singlestep_enabled = enabled;
> -        if (kvm_enabled()) {
> -            kvm_update_guest_debug(cpu, 0);
> +
> +        if (ops->update_guest_debug) {
> +            ops->update_guest_debug(cpu, 0);

Isn't this '0' flag here accelerator-specific? ...

>           }
> +
>           trace_breakpoint_singlestep(cpu->cpu_index, enabled);
>       }
>   }
> diff --git a/include/sysemu/accel-ops.h b/include/sysemu/accel-ops.h
> index 8cc7996def..0a47a2f00c 100644
> --- a/include/sysemu/accel-ops.h
> +++ b/include/sysemu/accel-ops.h
> @@ -48,6 +48,7 @@ struct AccelOpsClass {
>   
>       /* gdbstub hooks */
>       bool (*supports_guest_debug)(void);
> +    int (*update_guest_debug)(CPUState *cpu, unsigned long flags);

... if so the prototype should be:

        int (*update_guest_debug)(CPUState *cpu);

and the '0' value set within kvm-accel-ops.c handler implementation.

>       int (*insert_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
>       int (*remove_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
>       void (*remove_all_breakpoints)(CPUState *cpu);

Regards,

Phil.

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Mads Ynddal]

> Isn't this '0' flag here accelerator-specific? ...

> ... if so the prototype should be:
> 
>       int (*update_guest_debug)(CPUState *cpu);
> 
> and the '0' value set within kvm-accel-ops.c handler implementation.
> 

You're right, we can avoid the additional variable. We'll then have to wrap
`kvm_update_guest_debug`. Would the following be ok?

diff --git a/accel/kvm/kvm-accel-ops.c b/accel/kvm/kvm-accel-ops.c
index 6ebf9a644f..5e0fb42408 100644
--- a/accel/kvm/kvm-accel-ops.c
+++ b/accel/kvm/kvm-accel-ops.c
@@ -86,6 +86,10 @@ static bool kvm_cpus_are_resettable(void)
     return !kvm_enabled() || kvm_cpu_check_are_resettable();
 }
 
+static int kvm_update_guest_debug_ops(CPUState *cpu) {
+    return kvm_update_guest_debug(cpu, 0);
+}
+
 static void kvm_accel_ops_class_init(ObjectClass *oc, void *data)
 {
     AccelOpsClass *ops = ACCEL_OPS_CLASS(oc);
@@ -99,7 +103,7 @@ static void kvm_accel_ops_class_init(ObjectClass *oc, void *data)
     ops->synchronize_pre_loadvm = kvm_cpu_synchronize_pre_loadvm;
 
 #ifdef KVM_CAP_SET_GUEST_DEBUG
-    ops->update_guest_debug = kvm_update_guest_debug;
+    ops->update_guest_debug = kvm_update_guest_debug_ops;
     ops->supports_guest_debug = kvm_supports_guest_debug;
     ops->insert_breakpoint = kvm_insert_breakpoint;
     ops->remove_breakpoint = kvm_remove_breakpoint;
diff --git a/cpu.c b/cpu.c
index ef433a79e3..b2ade96caa 100644
--- a/cpu.c
+++ b/cpu.c
@@ -383,7 +383,7 @@ void cpu_single_step(CPUState *cpu, int enabled)
         cpu->singlestep_enabled = enabled;
 
         if (ops->update_guest_debug) {
-            ops->update_guest_debug(cpu, 0);
+            ops->update_guest_debug(cpu);
         }
 
         trace_breakpoint_singlestep(cpu->cpu_index, enabled);
diff --git a/include/sysemu/accel-ops.h b/include/sysemu/accel-ops.h
index 0a47a2f00c..cd6a4ef7a5 100644
--- a/include/sysemu/accel-ops.h
+++ b/include/sysemu/accel-ops.h
@@ -48,7 +48,7 @@ struct AccelOpsClass {
 
     /* gdbstub hooks */
     bool (*supports_guest_debug)(void);
-    int (*update_guest_debug)(CPUState *cpu, unsigned long flags);
+    int (*update_guest_debug)(CPUState *cpu);
     int (*insert_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
     int (*remove_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
     void (*remove_all_breakpoints)(CPUState *cpu);


If you have a better name for `kvm_update_guest_debug_ops`, I'm open for
suggestions.

On a side note. When compiling for an arch that isn't the same as the system
(i.e. aarch64 on x86_64), I'm getting a linker-error for cpu.c that
`cpus_get_accel` isn't defined. Do I need to move `cpus_get_accel` or somehow
#ifdef its use?

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Alex Bennée]

Mads Ynddal <mads@ynddal.dk> writes:

>> Isn't this '0' flag here accelerator-specific? ...
>
>> ... if so the prototype should be:
>> 
>>       int (*update_guest_debug)(CPUState *cpu);
>> 
>> and the '0' value set within kvm-accel-ops.c handler implementation.
>> 
>
> You're right, we can avoid the additional variable. We'll then have to wrap
> `kvm_update_guest_debug`. Would the following be ok?
>
> diff --git a/accel/kvm/kvm-accel-ops.c b/accel/kvm/kvm-accel-ops.c
> index 6ebf9a644f..5e0fb42408 100644
> --- a/accel/kvm/kvm-accel-ops.c
> +++ b/accel/kvm/kvm-accel-ops.c
> @@ -86,6 +86,10 @@ static bool kvm_cpus_are_resettable(void)
>      return !kvm_enabled() || kvm_cpu_check_are_resettable();
>  }
>  
> +static int kvm_update_guest_debug_ops(CPUState *cpu) {
> +    return kvm_update_guest_debug(cpu, 0);
> +}
> +
>  static void kvm_accel_ops_class_init(ObjectClass *oc, void *data)
>  {
>      AccelOpsClass *ops = ACCEL_OPS_CLASS(oc);
> @@ -99,7 +103,7 @@ static void kvm_accel_ops_class_init(ObjectClass *oc, void *data)
>      ops->synchronize_pre_loadvm = kvm_cpu_synchronize_pre_loadvm;
>  
>  #ifdef KVM_CAP_SET_GUEST_DEBUG
> -    ops->update_guest_debug = kvm_update_guest_debug;
> +    ops->update_guest_debug = kvm_update_guest_debug_ops;
>      ops->supports_guest_debug = kvm_supports_guest_debug;
>      ops->insert_breakpoint = kvm_insert_breakpoint;
>      ops->remove_breakpoint = kvm_remove_breakpoint;
> diff --git a/cpu.c b/cpu.c
> index ef433a79e3..b2ade96caa 100644
> --- a/cpu.c
> +++ b/cpu.c
> @@ -383,7 +383,7 @@ void cpu_single_step(CPUState *cpu, int enabled)
>          cpu->singlestep_enabled = enabled;
>  
>          if (ops->update_guest_debug) {
> -            ops->update_guest_debug(cpu, 0);
> +            ops->update_guest_debug(cpu);
>          }
>  
>          trace_breakpoint_singlestep(cpu->cpu_index, enabled);
> diff --git a/include/sysemu/accel-ops.h b/include/sysemu/accel-ops.h
> index 0a47a2f00c..cd6a4ef7a5 100644
> --- a/include/sysemu/accel-ops.h
> +++ b/include/sysemu/accel-ops.h
> @@ -48,7 +48,7 @@ struct AccelOpsClass {
>  
>      /* gdbstub hooks */
>      bool (*supports_guest_debug)(void);
> -    int (*update_guest_debug)(CPUState *cpu, unsigned long flags);
> +    int (*update_guest_debug)(CPUState *cpu);
>      int (*insert_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
>      int (*remove_breakpoint)(CPUState *cpu, int type, hwaddr addr, hwaddr len);
>      void (*remove_all_breakpoints)(CPUState *cpu);
>
>
> If you have a better name for `kvm_update_guest_debug_ops`, I'm open for
> suggestions.

It will do. You could just call it update_guest_debug as it is an
internal static function although I guess that makes grepping a bit of a
pain.

> On a side note. When compiling for an arch that isn't the same as the system
> (i.e. aarch64 on x86_64), I'm getting a linker-error for cpu.c that
> `cpus_get_accel` isn't defined. Do I need to move `cpus_get_accel` or somehow
> #ifdef its use?

Is something being accidentally linked with linux-user and softmmu?

-- 
Alex Bennée
Virtualisation Tech Lead @ Linaro

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Mads Ynddal]

> It will do. You could just call it update_guest_debug as it is an
> internal static function although I guess that makes grepping a bit of a
> pain.

I agree. It should preferably be something unique, to ease grep'ing.

> Is something being accidentally linked with linux-user and softmmu?

Good question. I'm not familiar enough with the code base to know.

I experimented with enabling/disabling linux-user when configuring, and it does
affect whether it compiles or not.

The following seems to fix it, and I can see the same approach is taken other
places in cpu.c. Would this be an acceptable solution?

diff --git a/cpu.c b/cpu.c
index 6effa5acc9..c9e8700691 100644
--- a/cpu.c
+++ b/cpu.c
@@ -386,6 +386,7 @@ void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
 void cpu_single_step(CPUState *cpu, int enabled)
 {
     if (cpu->singlestep_enabled != enabled) {
+#if !defined(CONFIG_USER_ONLY)
         const AccelOpsClass *ops = cpus_get_accel();

         cpu->singlestep_enabled = enabled;
@@ -393,6 +394,7 @@ void cpu_single_step(CPUState *cpu, int enabled)
         if (ops->update_guest_debug) {
             ops->update_guest_debug(cpu, 0);
         }
+#endif

         trace_breakpoint_singlestep(cpu->cpu_index, enabled);
     }

—
Mads Ynddal

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Alex Bennée]

Mads Ynddal <mads@ynddal.dk> writes:

>> It will do. You could just call it update_guest_debug as it is an
>> internal static function although I guess that makes grepping a bit of a
>> pain.
>
> I agree. It should preferably be something unique, to ease grep'ing.
>
>> Is something being accidentally linked with linux-user and softmmu?
>
> Good question. I'm not familiar enough with the code base to know.
>
> I experimented with enabling/disabling linux-user when configuring, and it does
> affect whether it compiles or not.
>
> The following seems to fix it, and I can see the same approach is taken other
> places in cpu.c. Would this be an acceptable solution?
>
> diff --git a/cpu.c b/cpu.c
> index 6effa5acc9..c9e8700691 100644
> --- a/cpu.c
> +++ b/cpu.c
> @@ -386,6 +386,7 @@ void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
>  void cpu_single_step(CPUState *cpu, int enabled)
>  {
>      if (cpu->singlestep_enabled != enabled) {
> +#if !defined(CONFIG_USER_ONLY)
>          const AccelOpsClass *ops = cpus_get_accel();
>
>          cpu->singlestep_enabled = enabled;
> @@ -393,6 +394,7 @@ void cpu_single_step(CPUState *cpu, int enabled)
>          if (ops->update_guest_debug) {
>              ops->update_guest_debug(cpu, 0);
>          }
> +#endif
>
>          trace_breakpoint_singlestep(cpu->cpu_index, enabled);
>      }

Sorry this dropped of my radar. Yes I think the ifdef will do. Are you
going to post a v2 with all the various updates?

-- 
Alex Bennée
Virtualisation Tech Lead @ Linaro

Re: [PATCH] gdbstub: move update guest debug to accel ops

[Mads Ynddal]

> 
> Sorry this dropped of my radar. Yes I think the ifdef will do. Are you
> going to post a v2 with all the various updates?
> 

No worries, I'll make a v2 with the changes.

—
Mads Ynddal