Skip to contentSkip to conversation

target/arm: Few accel cleanups

PATCH In review Philippe Mathieu-Daudé 6 participants 28 messages Jun 30, 2025 – Jul 7, 2025

lore
Share & tools
lore
#0 lore raw

cover target/arm: Few accel cleanups

Based-on: <[email protected]>
"arm: Fixes and preparatory cleanups for split-accel"

Philippe Mathieu-Daudé (6):
target/arm: Only set CPU_INTERRUPT_EXITTB for TCG target/arm: Only allow disabling NEON when using TCG target/arm: Better describe PMU depends on TCG or HVF target/arm: Re-use arm_is_psci_call() in HVF target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF target/arm: Allow arm_cpu_tlb_fill_align optionally set
CPUTLBEntryFull

 target/arm/cpu.c            |  4 ++--
 target/arm/helper.c         |  2 +-
 target/arm/hvf/hvf.c        |  7 ++++---
 target/arm/machine.c        | 10 +++++-----
 target/arm/tcg/psci.c       |  3 +++
 target/arm/tcg/tlb_helper.c |  4 +++-
 target/arm/trace-events     |  3 +++
 7 files changed, 21 insertions(+), 12 deletions(-)
Signature · 1 line
2.49.0
#1 lore raw

PATCH 1/6 target/arm: Only set CPU_INTERRUPT_EXITTB for TCG

Commit 34c45d53026 ("target-arm: kvm - re-inject guest debug exceptions") removed CPU_INTERRUPT_EXITTB from KVM, but it also appears on HVF. Better to restrict it to TCG.

---
 target/arm/helper.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target/arm/helper.c +1 -1
index 889d3088079..764b1f82dc2 100644
@@ -10929,7 +10929,7 @@ void arm_cpu_do_interrupt(CPUState *cs)
 
     arm_call_el_change_hook(cpu);
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled()) {
         cs->interrupt_request |= CPU_INTERRUPT_EXITTB;
     }
 }
Signature · 1 line
2.49.0
Reviewed-by on v1 1/6
#2 lore raw

Re: [PATCH 1/6] target/arm: Only set CPU_INTERRUPT_EXITTB for TCG

Quoted 9 lines · Philippe Mathieu-Daudé

Commit 34c45d53026 ("target-arm: kvm - re-inject guest debug exceptions") removed CPU_INTERRUPT_EXITTB from KVM, but it also appears on HVF. Better to restrict it to TCG.

---
  target/arm/helper.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

r~

Reviewed-by on v1 1/6
#3 lore raw

Re: [PATCH 1/6] target/arm: Only set CPU_INTERRUPT_EXITTB for TCG

In reply to Philippe Mathieu-Daudé
Quoted 9 lines · Philippe Mathieu-Daudé

Commit 34c45d53026 ("target-arm: kvm - re-inject guest debug exceptions") removed CPU_INTERRUPT_EXITTB from KVM, but it also appears on HVF. Better to restrict it to TCG.

---
  target/arm/helper.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
#4 lore raw

PATCH 2/6 target/arm: Only allow disabling NEON when using TCG

In reply to Philippe Mathieu-Daudé

Only allow disabling NEON when using TCG.

This avoids confusing user experience:

  $ qemu-system-aarch64 -M virt -accel hvf \
                        -cpu host,neon=off,vfp=off,vfp-d32=off
  qemu-system-aarch64: AArch64 CPUs must have both VFP and Neon or neither
  $ qemu-system-aarch64 -M virt -accel hvf \
                        -cpu host,neon=off,vfp=off,vfp-d32=off
  qemu-system-aarch64: ARM CPUs must have both VFP-D32 and Neon or neither
  $ qemu-system-aarch64 -M virt -accel hvf \
                        -cpu host,neon=off,vfp=off,vfp-d32=off
  qemu-system-aarch64: can't apply global host-arm-cpu.vfp-d32=off: Property 'host-arm-cpu.vfp-d32' not found
---
 target/arm/cpu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target/arm/cpu.c +1 -1
index ab5fbd9b40b..b6a8ba83a46 100644
@@ -1810,7 +1810,7 @@ static void arm_cpu_post_init(Object *obj)
 
     if (arm_feature(&cpu->env, ARM_FEATURE_NEON)) {
         cpu->has_neon = true;
-        if (!kvm_enabled()) {
+        if (tcg_enabled() || qtest_enabled()) {
             qdev_property_add_static(DEVICE(obj), &arm_cpu_has_neon_property);
         }
     }
Signature · 1 line
2.49.0
Reviewed-by on v1 2/6
#5 lore raw

Re: [PATCH 2/6] target/arm: Only allow disabling NEON when using TCG

Quoted 26 lines · Philippe Mathieu-Daudé

Only allow disabling NEON when using TCG.

This avoids confusing user experience:

   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: AArch64 CPUs must have both VFP and Neon or neither
   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: ARM CPUs must have both VFP-D32 and Neon or neither
   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: can't apply global host-arm-cpu.vfp-d32=off: Property 'host-arm-cpu.vfp-d32' not found
---
  target/arm/cpu.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
target/arm/cpu.c +0 -0
index ab5fbd9b40b..b6a8ba83a46 100644

target/arm/cpu.c static void arm_cpu_post_init(Object *obj) · line 1816

Quoted 8 lines
target/arm/cpu.c +1 -1
@@ -1810,7 +1810,7 @@ static void arm_cpu_post_init(Object *obj)
 
      if (arm_feature(&cpu->env, ARM_FEATURE_NEON)) {
          cpu->has_neon = true;
-        if (!kvm_enabled()) {
+        if (tcg_enabled() || qtest_enabled()) {
              qdev_property_add_static(DEVICE(obj), &arm_cpu_has_neon_property);
          }
      }

r~

Reviewed-by on v1 2/6
#6 lore raw

Re: [PATCH 2/6] target/arm: Only allow disabling NEON when using TCG

In reply to Philippe Mathieu-Daudé
Quoted 12 lines · Philippe Mathieu-Daudé

Only allow disabling NEON when using TCG.

This avoids confusing user experience:

   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: AArch64 CPUs must have both VFP and Neon or neither
   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: ARM CPUs must have both VFP-D32 and Neon or neither
Quoted 8 lines
   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: can't apply global host-arm-cpu.vfp-d32=off: Property 'host-arm-cpu.vfp-d32' not found
---
  target/arm/cpu.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
#7 lore raw

Re: [PATCH 2/6] target/arm: Only allow disabling NEON when using TCG

In reply to Philippe Mathieu-Daudé
Quoted 9 lines · Philippe Mathieu-Daudé

Only allow disabling NEON when using TCG.

This avoids confusing user experience:

  $ qemu-system-aarch64 -M virt -accel hvf \
                        -cpu host,neon=off,vfp=off,vfp-d32=off
  qemu-system-aarch64: AArch64 CPUs must have both VFP and Neon or neither
Quoted 7 lines
  $ qemu-system-aarch64 -M virt -accel hvf \
                        -cpu host,neon=off,vfp=off,vfp-d32=off
  qemu-system-aarch64: ARM CPUs must have both VFP-D32 and Neon or neither
  $ qemu-system-aarch64 -M virt -accel hvf \
                        -cpu host,neon=off,vfp=off,vfp-d32=off
  qemu-system-aarch64: can't apply global host-arm-cpu.vfp-d32=off: Property 'host-arm-cpu.vfp-d32' not found

I don't know about users, but I'm definitely confused. Aren't these three all the same command line ? What's the problem it's trying to show ?

-- PMM

#8 lore raw

Re: [PATCH 2/6] target/arm: Only allow disabling NEON when using TCG

Quoted 14 lines · Peter Maydell
Quoted 13 lines · Philippe Mathieu-Daudé

Only allow disabling NEON when using TCG.

This avoids confusing user experience:

   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: AArch64 CPUs must have both VFP and Neon or neither
   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: ARM CPUs must have both VFP-D32 and Neon or neither
Quoted 7 lines
Quoted 3 lines
   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: can't apply global host-arm-cpu.vfp-d32=off: Property 'host-arm-cpu.vfp-d32' not found

I don't know about users, but I'm definitely confused. Aren't these three all the same command line ? What's the problem it's trying to show ?

I should have only shared this confusing example:

$ qemu-system-aarch64 -M virt -accel hvf -cpu host,vfp-d32=off
qemu-system-aarch64: can't apply global host-arm-cpu.vfp-d32=off: Property 'host-arm-cpu.vfp-d32' not found
#9 lore raw

Re: [PATCH 2/6] target/arm: Only allow disabling NEON when using TCG

Quoted 24 lines · Philippe Mathieu-Daudé
Quoted 14 lines · Peter Maydell
Quoted 13 lines · Philippe Mathieu-Daudé

Only allow disabling NEON when using TCG.

This avoids confusing user experience:

   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: AArch64 CPUs must have both VFP and Neon or neither
   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: ARM CPUs must have both VFP-D32 and Neon or neither
Quoted 7 lines
Quoted 3 lines
   $ qemu-system-aarch64 -M virt -accel hvf \
                         -cpu host,neon=off,vfp=off,vfp-d32=off
   qemu-system-aarch64: can't apply global host-arm-cpu.vfp-d32=off: Property 'host-arm-cpu.vfp-d32' not found

I don't know about users, but I'm definitely confused. Aren't these three all the same command line ? What's the problem it's trying to show ?

Quoted 5 lines

I should have only shared this confusing example:

$ qemu-system-aarch64 -M virt -accel hvf -cpu host,vfp-d32=off
qemu-system-aarch64: can't apply global host-arm-cpu.vfp-d32=off:
Property 'host-arm-cpu.vfp-d32' not found

That seems correct, though ? We say
* Allow user to turn off VFP and Neon support, but only for TCG --
* KVM does not currently allow us to lie to the guest about its
* ID/feature registers, so the guest always sees what the host has.
and hvf is currently the same. The bug the patch is fixing looks like it is that -accel hvf lets us say '-cpu host,neon=off' but it won't actually do what the user might be intending. So we should avoid creating the property, as we already do for KVM.

(The other odd thing here is that hvf CPUs don't support AArch32, but ARM_FEATURE_NEON is about AArch32. But that's a separate thing and we'd need to cross-check that nothing aarch64 is incorrectly looking at that feature flag before we suppressed it for CPUs without any AArch32 support.)

-- PMM

#10 lore raw

PATCH 3/6 target/arm: Better describe PMU depends on TCG or HVF

In reply to Philippe Mathieu-Daudé

Simplify PMU logic by rewriting '!KVM' as 'TCG || HVF' (ignoring QTest, because vCPUs are not available there).

---
 target/arm/cpu.c     |  2 +-
 target/arm/machine.c | 10 +++++-----
 2 files changed, 6 insertions(+), 6 deletions(-)
target/arm/cpu.c +1 -1
index b6a8ba83a46..0311ff315fe 100644
@@ -2352,7 +2352,7 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
     if (arm_feature(env, ARM_FEATURE_PMU)) {
         pmu_init(cpu);
 
-        if (!kvm_enabled()) {
+        if (tcg_enabled() || hvf_enabled()) {
             arm_register_pre_el_change_hook(cpu, &pmu_pre_el_change, 0);
             arm_register_el_change_hook(cpu, &pmu_post_el_change, 0);
         }
target/arm/machine.c +5 -5
index e442d485241..baa7ad25ca9 100644
@@ -1,7 +1,7 @@
 #include "qemu/osdep.h"
 #include "cpu.h"
 #include "qemu/error-report.h"
-#include "system/kvm.h"
+#include "system/hvf.h"
 #include "system/tcg.h"
 #include "kvm_arm.h"
 #include "internals.h"
@@ -853,7 +853,7 @@ static int cpu_pre_save(void *opaque)
 {
     ARMCPU *cpu = opaque;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
         pmu_op_start(&cpu->env);
     }
 
@@ -888,7 +888,7 @@ static int cpu_post_save(void *opaque)
 {
     ARMCPU *cpu = opaque;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
         pmu_op_finish(&cpu->env);
     }
 
@@ -921,7 +921,7 @@ static int cpu_pre_load(void *opaque)
      */
     env->irq_line_state = UINT32_MAX;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
         pmu_op_start(env);
     }
 
@@ -1013,7 +1013,7 @@ static int cpu_post_load(void *opaque, int version_id)
         }
     }
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
         pmu_op_finish(env);
     }
 
Signature · 1 line
2.49.0
Reviewed-by on v1 3/6
#11 lore raw

Re: [PATCH 3/6] target/arm: Better describe PMU depends on TCG or HVF

Quoted 63 lines · Philippe Mathieu-Daudé

Simplify PMU logic by rewriting '!KVM' as 'TCG || HVF' (ignoring QTest, because vCPUs are not available there).

---
  target/arm/cpu.c     |  2 +-
  target/arm/machine.c | 10 +++++-----
  2 files changed, 6 insertions(+), 6 deletions(-)
target/arm/cpu.c +1 -1
index b6a8ba83a46..0311ff315fe 100644
@@ -2352,7 +2352,7 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
      if (arm_feature(env, ARM_FEATURE_PMU)) {
          pmu_init(cpu);
 
-        if (!kvm_enabled()) {
+        if (tcg_enabled() || hvf_enabled()) {
              arm_register_pre_el_change_hook(cpu, &pmu_pre_el_change, 0);
              arm_register_el_change_hook(cpu, &pmu_post_el_change, 0);
          }
target/arm/machine.c +4 -4
index e442d485241..baa7ad25ca9 100644
@@ -1,7 +1,7 @@
  #include "qemu/osdep.h"
  #include "cpu.h"
  #include "qemu/error-report.h"
-#include "system/kvm.h"
+#include "system/hvf.h"
  #include "system/tcg.h"
  #include "kvm_arm.h"
  #include "internals.h"
@@ -853,7 +853,7 @@ static int cpu_pre_save(void *opaque)
  {
      ARMCPU *cpu = opaque;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
          pmu_op_start(&cpu->env);
      }
 
@@ -888,7 +888,7 @@ static int cpu_post_save(void *opaque)
  {
      ARMCPU *cpu = opaque;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
          pmu_op_finish(&cpu->env);
      }
 
@@ -921,7 +921,7 @@ static int cpu_pre_load(void *opaque)
       */
      env->irq_line_state = UINT32_MAX;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
          pmu_op_start(env);
      }
 

target/arm/machine.c static int cpu_post_load(void *opaque, int version_id) · line 1019

Quoted 8 lines
target/arm/machine.c +1 -1
@@ -1013,7 +1013,7 @@ static int cpu_post_load(void *opaque, int version_id)
          }
      }
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
          pmu_op_finish(env);
      }
 

If I understand correctly, this is supported for tcg and hvf, but not kvm, right?
I'm just a bit confused by commit description mentioning "simplify logic", which is more "make it explicit".

Thanks,
Pierrick

Reviewed-by on v1 3/6
#12 lore raw

Re: [PATCH 3/6] target/arm: Better describe PMU depends on TCG or HVF

In reply to Philippe Mathieu-Daudé
Quoted 63 lines · Philippe Mathieu-Daudé

Simplify PMU logic by rewriting '!KVM' as 'TCG || HVF' (ignoring QTest, because vCPUs are not available there).

---
  target/arm/cpu.c     |  2 +-
  target/arm/machine.c | 10 +++++-----
  2 files changed, 6 insertions(+), 6 deletions(-)
target/arm/cpu.c +1 -1
index b6a8ba83a46..0311ff315fe 100644
@@ -2352,7 +2352,7 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
      if (arm_feature(env, ARM_FEATURE_PMU)) {
          pmu_init(cpu);
 
-        if (!kvm_enabled()) {
+        if (tcg_enabled() || hvf_enabled()) {
              arm_register_pre_el_change_hook(cpu, &pmu_pre_el_change, 0);
              arm_register_el_change_hook(cpu, &pmu_post_el_change, 0);
          }
target/arm/machine.c +4 -4
index e442d485241..baa7ad25ca9 100644
@@ -1,7 +1,7 @@
  #include "qemu/osdep.h"
  #include "cpu.h"
  #include "qemu/error-report.h"
-#include "system/kvm.h"
+#include "system/hvf.h"
  #include "system/tcg.h"
  #include "kvm_arm.h"
  #include "internals.h"
@@ -853,7 +853,7 @@ static int cpu_pre_save(void *opaque)
  {
      ARMCPU *cpu = opaque;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
          pmu_op_start(&cpu->env);
      }
 
@@ -888,7 +888,7 @@ static int cpu_post_save(void *opaque)
  {
      ARMCPU *cpu = opaque;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
          pmu_op_finish(&cpu->env);
      }
 
@@ -921,7 +921,7 @@ static int cpu_pre_load(void *opaque)
       */
      env->irq_line_state = UINT32_MAX;
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
          pmu_op_start(env);
      }
 

target/arm/machine.c static int cpu_post_load(void *opaque, int version_id) · line 1019

Quoted 8 lines
target/arm/machine.c +1 -1
@@ -1013,7 +1013,7 @@ static int cpu_post_load(void *opaque, int version_id)
          }
      }
 
-    if (!kvm_enabled()) {
+    if (tcg_enabled() || hvf_enabled()) {
          pmu_op_finish(env);
      }
 

If I understand correctly, this is supported for tcg and hvf, but not kvm, right?
I'm just a bit confused by commit description mentioning "simplify logic", which is more "make it explicit".

Thanks,
Pierrick

#13 lore raw

PATCH 4/6 target/arm: Re-use arm_is_psci_call() in HVF

In reply to Philippe Mathieu-Daudé

Re-use arm_is_psci_call() instead of open-coding it.

---
 target/arm/hvf/hvf.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
target/arm/hvf/hvf.c +2 -2
index 52199c4ff9d..7a99118c8c2 100644
@@ -2084,7 +2084,7 @@ int hvf_vcpu_exec(CPUState *cpu)
         break;
     case EC_AA64_HVC:
         cpu_synchronize_state(cpu);
-        if (arm_cpu->psci_conduit == QEMU_PSCI_CONDUIT_HVC) {
+        if (arm_is_psci_call(arm_cpu, EXCP_HVC)) {
             if (!hvf_handle_psci_call(cpu)) {
                 trace_hvf_unknown_hvc(env->pc, env->xregs[0]);
                 /* SMCCC 1.3 section 5.2 says every unknown SMCCC call returns -1 */
@@ -2097,7 +2097,7 @@ int hvf_vcpu_exec(CPUState *cpu)
         break;
     case EC_AA64_SMC:
         cpu_synchronize_state(cpu);
-        if (arm_cpu->psci_conduit == QEMU_PSCI_CONDUIT_SMC) {
+        if (arm_is_psci_call(arm_cpu, EXCP_SMC)) {
             advance_pc = true;
 
             if (!hvf_handle_psci_call(cpu)) {
Signature · 1 line
2.49.0
#16 lore raw

PATCH 5/6 target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF

In reply to Philippe Mathieu-Daudé

It is useful to compare PSCI calls of the same guest running under TCG or HVF.

---
 target/arm/hvf/hvf.c    | 3 ++-
 target/arm/tcg/psci.c   | 3 +++
 target/arm/trace-events | 3 +++
 3 files changed, 8 insertions(+), 1 deletion(-)
target/arm/hvf/hvf.c +2 -1
index 7a99118c8c2..6309c5b872e 100644
@@ -34,6 +34,7 @@
 #include "target/arm/multiprocessing.h"
 #include "target/arm/gtimer.h"
 #include "trace.h"
+#include "../trace.h"
 #include "migration/vmstate.h"
 
 #include "gdbstub/enums.h"
@@ -1149,7 +1150,7 @@ static bool hvf_handle_psci_call(CPUState *cpu)
     int target_el = 1;
     int32_t ret = 0;
 
-    trace_hvf_psci_call(param[0], param[1], param[2], param[3],
+    trace_arm_psci_call(param[0], param[1], param[2], param[3],
                         arm_cpu_mp_affinity(arm_cpu));
 
     switch (param[0]) {
target/arm/tcg/psci.c +3 -0
index cabed43e8a8..8df27ca123e 100644
@@ -25,6 +25,7 @@
 #include "internals.h"
 #include "arm-powerctl.h"
 #include "target/arm/multiprocessing.h"
+#include "../trace.h"
 
 bool arm_is_psci_call(ARMCPU *cpu, int excp_type)
 {
@@ -79,6 +80,8 @@ void arm_handle_psci_call(ARMCPU *cpu)
          */
         param[i] = is_a64(env) ? env->xregs[i] : env->regs[i];
     }
+    trace_arm_psci_call(param[0], param[1], param[2], param[3],
+                        arm_cpu_mp_affinity(cpu));
 
     if ((param[0] & QEMU_PSCI_0_2_64BIT) && !is_a64(env)) {
         ret = QEMU_PSCI_RET_NOT_SUPPORTED;
target/arm/trace-events +3 -0
index 4438dce7bec..a9cb5e0f5c6 100644
@@ -13,3 +13,6 @@ arm_gt_update_irq(int timer, int irqstate) "gt_update_irq: timer %d irqstate %d"
 
 # kvm.c
 kvm_arm_fixup_msi_route(uint64_t iova, uint64_t gpa) "MSI iova = 0x%"PRIx64" is translated into 0x%"PRIx64
+
+# tcg/psci.c and hvf/hvf.c
+arm_psci_call(uint64_t x0, uint64_t x1, uint64_t x2, uint64_t x3, uint32_t cpuid) "PSCI Call x0=0x%016"PRIx64" x1=0x%016"PRIx64" x2=0x%016"PRIx64" x3=0x%016"PRIx64" cpuid=0x%x"
Signature · 1 line
2.49.0
#17 lore raw

Re: [PATCH 5/6] target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF

Quoted 23 lines · Philippe Mathieu-Daudé

It is useful to compare PSCI calls of the same guest running under TCG or HVF.

---
  target/arm/hvf/hvf.c    | 3 ++-
  target/arm/tcg/psci.c   | 3 +++
  target/arm/trace-events | 3 +++
  3 files changed, 8 insertions(+), 1 deletion(-)
target/arm/hvf/hvf.c +1 -0
index 7a99118c8c2..6309c5b872e 100644
@@ -34,6 +34,7 @@
  #include "target/arm/multiprocessing.h"
  #include "target/arm/gtimer.h"
  #include "trace.h"
+#include "../trace.h"
  #include "migration/vmstate.h"
 
  #include "gdbstub/enums.h"

target/arm/hvf/hvf.c static bool hvf_handle_psci_call(CPUState *cpu) · line 1154

Quoted 6 lines
target/arm/hvf/hvf.c +1 -1
@@ -1149,7 +1150,7 @@ static bool hvf_handle_psci_call(CPUState *cpu)
      int target_el = 1;
      int32_t ret = 0;
 
-    trace_hvf_psci_call(param[0], param[1], param[2], param[3],
+    trace_arm_psci_call(param[0], param[1], param[2], param[3],
                          arm_cpu_mp_affinity(arm_cpu));

Lacks removal of the hvf trace?

r~

Quoted 34 lines
target/arm/hvf/hvf.c +0 -0
      switch (param[0]) {
target/arm/tcg/psci.c +3 -0
index cabed43e8a8..8df27ca123e 100644
@@ -25,6 +25,7 @@
  #include "internals.h"
  #include "arm-powerctl.h"
  #include "target/arm/multiprocessing.h"
+#include "../trace.h"
 
  bool arm_is_psci_call(ARMCPU *cpu, int excp_type)
  {
@@ -79,6 +80,8 @@ void arm_handle_psci_call(ARMCPU *cpu)
           */
          param[i] = is_a64(env) ? env->xregs[i] : env->regs[i];
      }
+    trace_arm_psci_call(param[0], param[1], param[2], param[3],
+                        arm_cpu_mp_affinity(cpu));
 
      if ((param[0] & QEMU_PSCI_0_2_64BIT) && !is_a64(env)) {
          ret = QEMU_PSCI_RET_NOT_SUPPORTED;
target/arm/trace-events +3 -0
index 4438dce7bec..a9cb5e0f5c6 100644
@@ -13,3 +13,6 @@ arm_gt_update_irq(int timer, int irqstate) "gt_update_irq: timer %d irqstate %d"
 
  # kvm.c
  kvm_arm_fixup_msi_route(uint64_t iova, uint64_t gpa) "MSI iova = 0x%"PRIx64" is translated into 0x%"PRIx64
+
+# tcg/psci.c and hvf/hvf.c
+arm_psci_call(uint64_t x0, uint64_t x1, uint64_t x2, uint64_t x3, uint32_t cpuid) "PSCI Call x0=0x%016"PRIx64" x1=0x%016"PRIx64" x2=0x%016"PRIx64" x3=0x%016"PRIx64" cpuid=0x%x"
#18 lore raw

Re: [PATCH 5/6] target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF

Quoted 22 lines · Richard Henderson
Quoted 21 lines · Philippe Mathieu-Daudé

It is useful to compare PSCI calls of the same guest running under TCG or HVF.

---
  target/arm/hvf/hvf.c    | 3 ++-
  target/arm/tcg/psci.c   | 3 +++
  target/arm/trace-events | 3 +++

  3 files changed, 8 insertions(+), 1 deletion(-)
target/arm/hvf/hvf.c +1 -0
index 7a99118c8c2..6309c5b872e 100644
@@ -34,6 +34,7 @@
   #include "target/arm/multiprocessing.h"
   #include "target/arm/gtimer.h"
   #include "trace.h"
+#include "../trace.h"
   #include "migration/vmstate.h"
   #include "gdbstub/enums.h"
Quoted 8 lines

target/arm/hvf/hvf.c static bool hvf_handle_psci_call(CPUState *cpu) · line 1153

Quoted 6 lines
target/arm/hvf/hvf.c +1 -1
@@ -1149,7 +1150,7 @@ static bool hvf_handle_psci_call(CPUState *cpu)
       int target_el = 1;
       int32_t ret = 0;
-    trace_hvf_psci_call(param[0], param[1], param[2], param[3],
+    trace_arm_psci_call(param[0], param[1], param[2], param[3],
                           arm_cpu_mp_affinity(arm_cpu));

Lacks removal of the hvf trace?

Oops indeed...

Reviewed-by on v1 5/6
#19 lore raw

Re: [PATCH 5/6] target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF

In reply to Philippe Mathieu-Daudé
Quoted 57 lines · Philippe Mathieu-Daudé

It is useful to compare PSCI calls of the same guest running under TCG or HVF.

---
  target/arm/hvf/hvf.c    | 3 ++-
  target/arm/tcg/psci.c   | 3 +++
  target/arm/trace-events | 3 +++
  3 files changed, 8 insertions(+), 1 deletion(-)
target/arm/hvf/hvf.c +2 -1
index 7a99118c8c2..6309c5b872e 100644
@@ -34,6 +34,7 @@
  #include "target/arm/multiprocessing.h"
  #include "target/arm/gtimer.h"
  #include "trace.h"
+#include "../trace.h"
  #include "migration/vmstate.h"
 
  #include "gdbstub/enums.h"
@@ -1149,7 +1150,7 @@ static bool hvf_handle_psci_call(CPUState *cpu)
      int target_el = 1;
      int32_t ret = 0;
 
-    trace_hvf_psci_call(param[0], param[1], param[2], param[3],
+    trace_arm_psci_call(param[0], param[1], param[2], param[3],
                          arm_cpu_mp_affinity(arm_cpu));
 
      switch (param[0]) {
target/arm/tcg/psci.c +3 -0
index cabed43e8a8..8df27ca123e 100644
@@ -25,6 +25,7 @@
  #include "internals.h"
  #include "arm-powerctl.h"
  #include "target/arm/multiprocessing.h"
+#include "../trace.h"
 
  bool arm_is_psci_call(ARMCPU *cpu, int excp_type)
  {
@@ -79,6 +80,8 @@ void arm_handle_psci_call(ARMCPU *cpu)
           */
          param[i] = is_a64(env) ? env->xregs[i] : env->regs[i];
      }
+    trace_arm_psci_call(param[0], param[1], param[2], param[3],
+                        arm_cpu_mp_affinity(cpu));
 
      if ((param[0] & QEMU_PSCI_0_2_64BIT) && !is_a64(env)) {
          ret = QEMU_PSCI_RET_NOT_SUPPORTED;
target/arm/trace-events +0 -0
index 4438dce7bec..a9cb5e0f5c6 100644

target/arm/trace-events arm_gt_update_irq(int timer, int irqstate) "gt_update_irq: timer %d irqstate %d" · line 18

Quoted 6 lines
target/arm/trace-events +3 -0
@@ -13,3 +13,6 @@ arm_gt_update_irq(int timer, int irqstate) "gt_update_irq: timer %d irqstate %d"
 
  # kvm.c
  kvm_arm_fixup_msi_route(uint64_t iova, uint64_t gpa) "MSI iova = 0x%"PRIx64" is translated into 0x%"PRIx64
+
+# tcg/psci.c and hvf/hvf.c
+arm_psci_call(uint64_t x0, uint64_t x1, uint64_t x2, uint64_t x3, uint32_t cpuid) "PSCI Call x0=0x%016"PRIx64" x1=0x%016"PRIx64" x2=0x%016"PRIx64" x3=0x%016"PRIx64" cpuid=0x%x"

Just a nit, using 'target/arm/trace.h' might be more readable than '../trace.h'.

#20 lore raw

Re: [PATCH 5/6] target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF

Quoted 12 lines · Pierrick Bouvier
Quoted 10 lines · Philippe Mathieu-Daudé

It is useful to compare PSCI calls of the same guest running under TCG or HVF.

---
  target/arm/hvf/hvf.c    | 3 ++-
  target/arm/tcg/psci.c   | 3 +++
  target/arm/trace-events | 3 +++
  3 files changed, 8 insertions(+), 1 deletion(-)
Quoted 13 lines

target/arm/hvf/hvf.c line 37

Quoted 9 lines
target/arm/hvf/hvf.c +1 -0
index 7a99118c8c2..6309c5b872e 100644
@@ -34,6 +34,7 @@
  #include "target/arm/multiprocessing.h"
  #include "target/arm/gtimer.h"
  #include "trace.h"
+#include "../trace.h"

Just a nit, using 'target/arm/trace.h' might be more readable than '../trace.h'.

Mmm. docs/devel/tracing.rst rather discourages this:

# While it is possible to include a trace.h file from outside a source file's own
# sub-directory, this is discouraged in general. It is strongly preferred that

# all events be declared directly in the sub-directory that uses them. The only
# exception is where there are some shared trace events defined in the top level
# directory trace-events file.

I don't know if we want to loosen that to permit events that are shared between multiple subdirs (cc'ing the trace subsystem maintainers for their view).

git grep 'include.*trace.h' | grep -v '"trace.h"'| grep -v 'trace.h:'|less

suggests that the only current place where we're including a trace.h not in the same directory is linux-user, where we opt to use the full linux-user/trace.h path. So probably for consistency we should use target/arm/trace.h here.

(That grep also shows up that hw/uefi is missing its trace.h header and the .c files are including trace-hw_uefi.h directly...)

-- PMM

#21 lore raw

Re: [PATCH 5/6] target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF

Quoted 33 lines · Peter Maydell
Quoted 12 lines · Pierrick Bouvier
Quoted 10 lines · Philippe Mathieu-Daudé

It is useful to compare PSCI calls of the same guest running under TCG or HVF.

---
  target/arm/hvf/hvf.c    | 3 ++-
  target/arm/tcg/psci.c   | 3 +++
  target/arm/trace-events | 3 +++
  3 files changed, 8 insertions(+), 1 deletion(-)
Quoted 13 lines

target/arm/hvf/hvf.c line 37

Quoted 9 lines
target/arm/hvf/hvf.c +1 -0
index 7a99118c8c2..6309c5b872e 100644
@@ -34,6 +34,7 @@
  #include "target/arm/multiprocessing.h"
  #include "target/arm/gtimer.h"
  #include "trace.h"
+#include "../trace.h"

Just a nit, using 'target/arm/trace.h' might be more readable than '../trace.h'.

Mmm. docs/devel/tracing.rst rather discourages this:

# While it is possible to include a trace.h file from outside a source file's own
# sub-directory, this is discouraged in general. It is strongly preferred that

Quoted 7 lines
# all events be declared directly in the sub-directory that uses them. The only
# exception is where there are some shared trace events defined in the top level
# directory trace-events file.

I don't know if we want to loosen that to permit events that are shared between multiple subdirs (cc'ing the trace subsystem maintainers for their view).

Code is generated from the trace-events files and my main concern is that the build dependencies on the generated code may not be obvious if a trace event from somewhere far away in the source tree hierarchy is used. You might hit linker errors because the .o files needed for the trace events are not being linked in.

I would try to stick with what's described in tracing.rst to avoid difficulties now or in the future.

Quoted 14 lines

git grep 'include.*trace.h' | grep -v '"trace.h"'| grep -v 'trace.h:'|less

suggests that the only current place where we're including a trace.h not in the same directory is linux-user, where we opt to use the full linux-user/trace.h path. So probably for consistency we should use target/arm/trace.h here.

(That grep also shows up that hw/uefi is missing its trace.h header and the .c files are including trace-hw_uefi.h directly...)

-- PMM

#22 lore raw

Re: [PATCH 5/6] target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF

Quoted 43 lines · Stefan Hajnoczi
Quoted 33 lines · Peter Maydell
Quoted 12 lines · Pierrick Bouvier
Quoted 10 lines · Philippe Mathieu-Daudé

It is useful to compare PSCI calls of the same guest running under TCG or HVF.

---
  target/arm/hvf/hvf.c    | 3 ++-
  target/arm/tcg/psci.c   | 3 +++
  target/arm/trace-events | 3 +++
  3 files changed, 8 insertions(+), 1 deletion(-)
Quoted 13 lines

target/arm/hvf/hvf.c line 37

Quoted 9 lines
target/arm/hvf/hvf.c +1 -0
index 7a99118c8c2..6309c5b872e 100644
@@ -34,6 +34,7 @@
  #include "target/arm/multiprocessing.h"
  #include "target/arm/gtimer.h"
  #include "trace.h"
+#include "../trace.h"

Just a nit, using 'target/arm/trace.h' might be more readable than '../trace.h'.

Mmm. docs/devel/tracing.rst rather discourages this:

# While it is possible to include a trace.h file from outside a source file's own
# sub-directory, this is discouraged in general. It is strongly preferred that

Quoted 7 lines
# all events be declared directly in the sub-directory that uses them. The only
# exception is where there are some shared trace events defined in the top level
# directory trace-events file.

I don't know if we want to loosen that to permit events that are shared between multiple subdirs (cc'ing the trace subsystem maintainers for their view).

Quoted 8 lines

Code is generated from the trace-events files and my main concern is that the build dependencies on the generated code may not be obvious if a trace event from somewhere far away in the source tree hierarchy is used. You might hit linker errors because the .o files needed for the trace events are not being linked in.

I would try to stick with what's described in tracing.rst to avoid difficulties now or in the future.

So how should we deal with "two different source files in different subdirectories both want to emit trace event X" ? Creating a utility function that can live in the parent dir just to wrap the trace event seems a bit inefficient.

I agree that using a trace event from a long way away across the source tree is probably a bad idea, but "immediate parent directory" doesn't seem too likely to cause issues ?

-- PMM

#23 lore raw

Re: [PATCH 5/6] target/arm: Share ARM_PSCI_CALL trace event between TCG and HVF

In reply to Peter Maydell
Quoted 41 lines · Peter Maydell
Quoted 12 lines · Pierrick Bouvier
Quoted 10 lines · Philippe Mathieu-Daudé

It is useful to compare PSCI calls of the same guest running under TCG or HVF.

---
  target/arm/hvf/hvf.c    | 3 ++-
  target/arm/tcg/psci.c   | 3 +++
  target/arm/trace-events | 3 +++
  3 files changed, 8 insertions(+), 1 deletion(-)
Quoted 13 lines

target/arm/hvf/hvf.c line 37

Quoted 9 lines
target/arm/hvf/hvf.c +1 -0
index 7a99118c8c2..6309c5b872e 100644
@@ -34,6 +34,7 @@
  #include "target/arm/multiprocessing.h"
  #include "target/arm/gtimer.h"
  #include "trace.h"
+#include "../trace.h"

Just a nit, using 'target/arm/trace.h' might be more readable than '../trace.h'.

Mmm. docs/devel/tracing.rst rather discourages this:

# While it is possible to include a trace.h file from outside a source file's own
# sub-directory, this is discouraged in general. It is strongly preferred that

# all events be declared directly in the sub-directory that uses them. The only
# exception is where there are some shared trace events defined in the top level
# directory trace-events file.

I don't know if we want to loosen that to permit events that are shared between multiple subdirs (cc'ing the trace subsystem maintainers for their view).

Quoted 6 lines

git grep 'include.*trace.h' | grep -v '"trace.h"'| grep -v 'trace.h:'|less

suggests that the only current place where we're including a trace.h not in the same directory is linux-user, where we opt to use the full linux-user/trace.h path. So probably for consistency we should use target/arm/trace.h here.

IMHO using the up-level relative paths is desirable, as it reinforces the intent that we shouldn't be pulling in trace events from arbitrary different sub-trees of the codebase, only the current dir & its (near) parents.

With regards,
Daniel

Signature · 3 lines
|: 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 :|
#24 lore raw

PATCH 6/6 target/arm: Allow arm_cpu_tlb_fill_align optionally set CPUTLBEntryFull

In reply to Philippe Mathieu-Daudé
---
 target/arm/tcg/tlb_helper.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
target/arm/tcg/tlb_helper.c +3 -1
index 23c72a99f5c..df04ef351d1 100644
@@ -349,7 +349,9 @@ bool arm_cpu_tlb_fill_align(CPUState *cs, CPUTLBEntryFull *out, vaddr address,
                               &res, fi)) {
         res.f.extra.arm.pte_attrs = res.cacheattrs.attrs;
         res.f.extra.arm.shareability = res.cacheattrs.shareability;
-        *out = res.f;
+        if (out) {
+            *out = res.f;
+        }
         return true;
     }
     if (probe) {
Signature · 1 line
2.49.0
#25 lore raw

Re: [PATCH 6/6] target/arm: Allow arm_cpu_tlb_fill_align optionally set CPUTLBEntryFull

Quoted 10 lines · Philippe Mathieu-Daudé
---
  target/arm/tcg/tlb_helper.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
target/arm/tcg/tlb_helper.c +0 -0
index 23c72a99f5c..df04ef351d1 100644

target/arm/tcg/tlb_helper.c bool arm_cpu_tlb_fill_align(CPUState *cs, CPUTLBEntryFull *out, vaddr address, · line 357

Quoted 10 lines
target/arm/tcg/tlb_helper.c +3 -1
@@ -349,7 +349,9 @@ bool arm_cpu_tlb_fill_align(CPUState *cs, CPUTLBEntryFull *out, vaddr address,
                                &res, fi)) {
          res.f.extra.arm.pte_attrs = res.cacheattrs.attrs;
          res.f.extra.arm.shareability = res.cacheattrs.shareability;
-        *out = res.f;
+        if (out) {
+            *out = res.f;
+        }
          return true;
      }
      if (probe) {

Why? There's no other way to get the phys addr result. Are you only calling this for the raise-exception side effect?

r~

#26 lore raw

Re: [PATCH 6/6] target/arm: Allow arm_cpu_tlb_fill_align optionally set CPUTLBEntryFull

Quoted 17 lines · Richard Henderson
Quoted 16 lines · Philippe Mathieu-Daudé

---
  target/arm/tcg/tlb_helper.c | 4 +++-

  1 file changed, 3 insertions(+), 1 deletion(-)
target/arm/tcg/tlb_helper.c +1 -1
index 23c72a99f5c..df04ef351d1 100644
@@ -349,7 +349,9 @@ bool arm_cpu_tlb_fill_align(CPUState *cs, 
 CPUTLBEntryFull *out, vaddr address,
                                 &res, fi)) {
           res.f.extra.arm.pte_attrs = res.cacheattrs.attrs;
           res.f.extra.arm.shareability = res.cacheattrs.shareability;
-        *out = res.f;
+        if (out) {
Quoted 8 lines
Quoted 5 lines
target/arm/tcg/tlb_helper.c +2 -0
+            *out = res.f;
+        }
           return true;
       }

      if (probe) {

Why?  There's no other way to get the phys addr result. Are you only calling this for the raise-exception side effect?

Yes:

-- >8 --
@@ -2121,6 +2121,25 @@ int hvf_vcpu_exec(CPUState *cpu)

             hvf_raise_exception(cpu, EXCP_UDEF, syn_uncategorized(), 1);
         }
         break;
+    case EC_INSNABORT: {
+        uint32_t set = (syndrome >> 12) & 3;
+        bool fnv = (syndrome >> 10) & 1;
+        bool ea = (syndrome >> 9) & 1;
+        bool s1ptw = (syndrome >> 7) & 1;
+        uint32_t ifsc = (syndrome >> 0) & 0x3f;

+
+ trace_hvf_insn_abort(env->pc, set, fnv, ea, s1ptw, ifsc);
+
+ cpu_synchronize_state(cpu);
+ if (tcg_enabled()) {
+ ret = EXCP_EMULATE;
+ arm_cpu_tlb_fill_align(cpu, NULL, env->pc, MMU_INST_FETCH,
+ arm_env_mmu_index(env), MO_32, 4, false, -1);
+ }
+ break;
+ }

     default:
         cpu_synchronize_state(cpu);
         trace_hvf_exit(syndrome, ec, env->pc);

---

I see probe_access_full_mmu() uses discard_tlb, I can use a similar stack variable if you rather.

#27 lore raw

Re: [PATCH 6/6] target/arm: Allow arm_cpu_tlb_fill_align optionally set CPUTLBEntryFull

In reply to Philippe Mathieu-Daudé
Quoted 10 lines · Philippe Mathieu-Daudé
---
  target/arm/tcg/tlb_helper.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
target/arm/tcg/tlb_helper.c +0 -0
index 23c72a99f5c..df04ef351d1 100644

target/arm/tcg/tlb_helper.c bool arm_cpu_tlb_fill_align(CPUState *cs, CPUTLBEntryFull *out, vaddr address, · line 357

Quoted 10 lines
target/arm/tcg/tlb_helper.c +3 -1
@@ -349,7 +349,9 @@ bool arm_cpu_tlb_fill_align(CPUState *cs, CPUTLBEntryFull *out, vaddr address,
                                &res, fi)) {
          res.f.extra.arm.pte_attrs = res.cacheattrs.attrs;
          res.f.extra.arm.shareability = res.cacheattrs.shareability;
-        *out = res.f;
+        if (out) {
+            *out = res.f;
+        }
          return true;
      }
      if (probe) {

Would that be possible to provide more context about why it's needed?

The goal of tlb_fill_align is precisely to return a new CPUTLBEntryFull, while checking all the protection to access this page and generating a page fault if needed.

In case you just want to check if an address is valid, arm_cpu_get_phys_page_attrs_debug might be a better choice (checking for return value -1 in case of error).

Thanks,
Pierrick