Skip to contentSkip to conversation

accel: Introduce current_accel_name()

PATCH In review Alexander Graf 3 participants 6 messages Jun 20, 2022 – Jun 21, 2022

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/2 accel: Introduce current_accel_name()

We need to fetch the name of the current accelerator in flexible error messages more going forward. Let's create a helper that gives it to us without casting in the target code.

---
 accel/accel-common.c | 8 ++++++++
 include/qemu/accel.h | 1 +
 softmmu/vl.c         | 3 +--
 3 files changed, 10 insertions(+), 2 deletions(-)
accel/accel-common.c +8 -0
index 7b8ec7e0f7..50035bda55 100644
@@ -49,6 +49,14 @@ AccelClass *accel_find(const char *opt_name)
     return ac;
 }
 
+/* Return the name of the current accelerator */
+const char *current_accel_name(void)
+{
+    AccelClass *ac = ACCEL_GET_CLASS(current_accel());
+
+    return ac->name;
+}
+
 static void accel_init_cpu_int_aux(ObjectClass *klass, void *opaque)
 {
     CPUClass *cc = CPU_CLASS(klass);
include/qemu/accel.h +1 -0
index 4f4c283f6f..be56da1b99 100644
@@ -68,6 +68,7 @@ typedef struct AccelClass {
 
 AccelClass *accel_find(const char *opt_name);
 AccelState *current_accel(void);
+const char *current_accel_name(void);
 
 void accel_init_interfaces(AccelClass *ac);
 
softmmu/vl.c +1 -2
index 54e920ada1..3dca5936c7 100644
@@ -2271,8 +2271,7 @@ static void configure_accelerators(const char *progname)
     }
 
     if (init_failed && !qtest_chrdev) {
-        AccelClass *ac = ACCEL_GET_CLASS(current_accel());
-        error_report("falling back to %s", ac->name);
+        error_report("falling back to %s", current_accel_name());
     }
 
     if (icount_enabled() && !tcg_enabled()) {
Signature · 1 line
2.32.1 (Apple Git-133)
#1 lore raw

PATCH v2 2/2 target/arm: Catch invalid kvm state also for hvf

Some features such as running in EL3 or running M profile code are incompatible with virtualization as QEMU implements it today. To prevent users from picking invalid configurations on other virt solutions like Hvf, let's run the same checks there too.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1073

---

v1 -> v2:

- Use current_accel_name()
- Use !tcg_enabled()

---
 target/arm/cpu.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
target/arm/cpu.c +7 -4
index 1b5d535788..0862dcd63c 100644
@@ -1490,7 +1490,7 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         }
     }
 
-    if (kvm_enabled()) {
+    if (!tcg_enabled()) {
         /*
          * Catch all the cases which might cause us to create more than one
          * address space for the CPU (otherwise we will assert() later in
@@ -1498,17 +1498,20 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
          */
         if (arm_feature(env, ARM_FEATURE_M)) {
             error_setg(errp,
-                       "Cannot enable KVM when using an M-profile guest CPU");
+                       "Cannot enable %s when using an M-profile guest CPU",
+                       current_accel_name());
             return;
         }
         if (cpu->has_el3) {
             error_setg(errp,
-                       "Cannot enable KVM when guest CPU has EL3 enabled");
+                       "Cannot enable %s when guest CPU has EL3 enabled",
+                       current_accel_name());
             return;
         }
         if (cpu->tag_memory) {
             error_setg(errp,
-                       "Cannot enable KVM when guest CPUs has MTE enabled");
+                       "Cannot enable %s when guest CPUs has MTE enabled",
+                       current_accel_name());
             return;
         }
     }
Signature · 1 line
2.32.1 (Apple Git-133)
Reviewed-by on v2 2/2
#2 lore raw

Re: [PATCH v2 2/2] target/arm: Catch invalid kvm state also for hvf

Quoted 10 lines · Alexander Graf

Some features such as running in EL3 or running M profile code are incompatible with virtualization as QEMU implements it today. To prevent users from picking invalid configurations on other virt solutions like Hvf, let's run the same checks there too.

Resolves:https://gitlab.com/qemu-project/qemu/-/issues/1073

---

Quoted 7 lines

v1 -> v2:

- Use current_accel_name()
- Use !tcg_enabled()

---
  target/arm/cpu.c | 11 +++++++----
  1 file changed, 7 insertions(+), 4 deletions(-)

r~

#3 lore raw

Re: [PATCH v2 2/2] target/arm: Catch invalid kvm state also for hvf

In reply to Alexander Graf
Quoted 13 lines · Alexander Graf

Some features such as running in EL3 or running M profile code are incompatible with virtualization as QEMU implements it today. To prevent users from picking invalid configurations on other virt solutions like Hvf, let's run the same checks there too.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1073

target/arm/cpu.c static void arm_cpu_realizefn(DeviceState *dev, Error **errp) · line 1493

Quoted 5 lines
target/arm/cpu.c +1 -1
@@ -1490,7 +1490,7 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         }
     }
 
-    if (kvm_enabled()) {
+    if (!tcg_enabled()) {

I'm a bit surprised we don't need to also have "&& !qtest_enabled()", but I guess if "make check" works then we're fine :-)

-- PMM

#4 lore raw

Re: [PATCH v2 2/2] target/arm: Catch invalid kvm state also for hvf

Quoted 18 lines · Peter Maydell
Quoted 16 lines · Alexander Graf

Some features such as running in EL3 or running M profile code are incompatible with virtualization as QEMU implements it today. To prevent users from picking invalid configurations on other virt solutions like Hvf, let's run the same checks there too.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1073

target/arm/cpu.c +0 -0
@@ -1490,7 +1490,7 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         }
     }
 
Quoted 5 lines

target/arm/cpu.c static void arm_cpu_realizefn(DeviceState *dev, Error **errp) · line 1493

Quoted 2 lines
target/arm/cpu.c +1 -1
-    if (kvm_enabled()) {
+    if (!tcg_enabled()) {

I'm a bit surprised we don't need to also have "&& !qtest_enabled()", but I guess if "make check" works then we're fine :-)

In fact you do need to handle qtest here too, otherwise lots of tests in 'make check' barf on the unexpected error...

I'm going to squash the following fix in and take the patchset into target-arm.next. (PS: you forgot the cover letter, I think ?)

target/arm/cpu.c +1 -0
index 0862dcd63cb..d9c4a9f56d2 100644
@@ -39,6 +39,7 @@
 #include "hw/boards.h"
 #endif
 #include "sysemu/tcg.h"
+#include "sysemu/qtest.h"
 #include "sysemu/hw_accel.h"
 #include "kvm_arm.h"
 #include "disas/capstone.h"
@@ -1490,8 +1491,12 @@ static void arm_cpu_realizefn(DeviceState *dev,

Error **errp)

         }
     }

- if (!tcg_enabled()) {

+    if (!tcg_enabled() && !qtest_enabled()) {
         /*
+         * We assume that no accelerator except TCG (and the "not really an
+         * accelerator" qtest) can handle these features, because Arm hardware
+         * virtualization can't virtualize them.
+         *

* Catch all the cases which might cause us to create more than one
* address space for the CPU (otherwise we will assert() later in
* cpu_address_space_init()).

thanks
-- PMM

Reviewed-by on v2 1/2
#5 lore raw
In reply to Alexander Graf
Quoted 10 lines · Alexander Graf

We need to fetch the name of the current accelerator in flexible error messages more going forward. Let's create a helper that gives it to us without casting in the target code.

---
  accel/accel-common.c | 8 ++++++++
  include/qemu/accel.h | 1 +
  softmmu/vl.c         | 3 +--
  3 files changed, 10 insertions(+), 2 deletions(-)

r~