Skip to contentSkip to conversation

hw/arm/virt, qtests: Fix make check-qtest-aarch64 when CONFIG_ARM_GIC_TCG is unset

PATCH Superseded Eric Auger 3 participants 10 messages Jan 31, 2022

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Eric Auger

author
#0 lore raw

cover hw/arm/virt, qtests: Fix make check-qtest-aarch64 when CONFIG_ARM_GIC_TCG is unset

When CONFIG_ARM_GIC_TCG is unset, qtests fail with ERROR:../qom/object.c:715:object_new_with_type: assertion failed: (type != NULL)

This is due to the fact a bunch of tests currently depend on GICv3 availability. qom-test and test-hmp try to launch sbsa-ref and xlnx-versal-virt which only support GICv3. Also a bunch of tests use gic-version=max which currectly selects GICv3, ignoring the fact this latter has been disabled.

This series proposes to ignore sbsa-ref and xlnx-versal-virt during qos-test and test-hmp execution if CONFIG_ARM_GIC_TCG is unset. Also it selects GICv2 if gic-version=max and CONFIG_ARM_GIC_TCG is unset, in TCG mode. With those fixes make check passes.

Best Regards

Eric

This series can be found at:
https://github.com/eauger/qemu/tree/config_arm_gic_tcg_v1

Eric Auger (2):
hw/arm/virt: Fix gic-version=max when CONFIG_ARM_GIC_TCG is unset tests/qtest: Special case sbsa-ref and xlnx-versal-virt if
!CONFIG_ARM_GIC_TCG

 hw/arm/virt.c          | 4 ++++
 tests/qtest/libqtest.c | 6 ++++++
 2 files changed, 10 insertions(+)
Signature · 1 line
2.26.3
author
#1 lore raw

PATCH 1/2 hw/arm/virt: Fix gic-version=max when CONFIG_ARM_GIC_TCG is unset

In TCG mode, if gic-version=max we always select GICv3 even if CONFIG_ARM_GIC_TCG is unset. We shall rather select GICv2. This also brings the benefit of fixing qos tests errors for tests using gic-version=max with CONFIG_ARM_GIC_TCG unset.

Fixes a8a5546798c3 ("hw/intc/arm_gicv3: Introduce CONFIG_ARM_GIC_TCG Kconfig selector")
---
 hw/arm/virt.c | 4 ++++
 1 file changed, 4 insertions(+)
hw/arm/virt.c +4 -0
index 141350bf21..2f1d4d0230 100644
@@ -1852,7 +1852,11 @@ static void finalize_gic_version(VirtMachineState *vms)
         vms->gic_version = VIRT_GIC_VERSION_2;
         break;
     case VIRT_GIC_VERSION_MAX:
+#ifdef CONFIG_ARM_GIC_TCG
         vms->gic_version = VIRT_GIC_VERSION_3;
+#else
+        vms->gic_version = VIRT_GIC_VERSION_2;
+#endif
         break;
     case VIRT_GIC_VERSION_HOST:
         error_report("gic-version=host requires KVM");
Signature · 1 line
2.26.3
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] hw/arm/virt: Fix gic-version=max when CONFIG_ARM_GIC_TCG is unset

Quoted 30 lines · Eric Auger

In TCG mode, if gic-version=max we always select GICv3 even if CONFIG_ARM_GIC_TCG is unset. We shall rather select GICv2. This also brings the benefit of fixing qos tests errors for tests using gic-version=max with CONFIG_ARM_GIC_TCG unset.

Fixes a8a5546798c3 ("hw/intc/arm_gicv3: Introduce CONFIG_ARM_GIC_TCG Kconfig selector")
---
 hw/arm/virt.c | 4 ++++
 1 file changed, 4 insertions(+)
hw/arm/virt.c +4 -0
index 141350bf21..2f1d4d0230 100644
@@ -1852,7 +1852,11 @@ static void finalize_gic_version(VirtMachineState *vms)
         vms->gic_version = VIRT_GIC_VERSION_2;
         break;
     case VIRT_GIC_VERSION_MAX:
+#ifdef CONFIG_ARM_GIC_TCG
         vms->gic_version = VIRT_GIC_VERSION_3;
+#else
+        vms->gic_version = VIRT_GIC_VERSION_2;
+#endif
         break;
     case VIRT_GIC_VERSION_HOST:
         error_report("gic-version=host requires KVM");
Signature · 1 line
2.26.3

Why is the config generically named "ARM_GIC_TCG" when it only controls Arm GICV3 TCG?

Otherwise,

Thanks,
drew

author
#3 lore raw

PATCH 2/2 tests/qtest: Special case sbsa-ref and xlnx-versal-virt if !CONFIG_ARM_GIC_TCG

In reply to Eric Auger

qom-test and test-hmp shall not run tests on sbsa-ref and xlnx-versal-virt if CONFIG_ARM_GIC_TCG is unset as those machines always instantiate GICv3.

Otherwise the tests fail with ERROR:../qom/object.c:715:object_new_with_type: assertion failed: (type != NULL)

Fixes a8a5546798c3 ("hw/intc/arm_gicv3: Introduce CONFIG_ARM_GIC_TCG Kconfig selector")
---
 tests/qtest/libqtest.c | 6 ++++++
 1 file changed, 6 insertions(+)
tests/qtest/libqtest.c +6 -0
index 41f4da4e54..f53983a28e 100644
@@ -1394,6 +1394,12 @@ void qtest_cb_for_every_machine(void (*cb)(const char *machine),
             g_str_equal("xenpv", machines[i].name)) {
             continue;
         }
+#ifndef CONFIG_ARM_GIC_TCG
+        if (!strncmp("sbsa-ref", machines[i].name, 8) ||
+            !strncmp("xlnx-versal-virt", machines[i].name, 16)) {
+            continue;
+        }
+#endif
         if (!skip_old_versioned ||
             !qtest_is_old_versioned_machine(machines[i].name)) {
             cb(machines[i].name);
Signature · 1 line
2.26.3
#4 lore raw

Re: [PATCH 2/2] tests/qtest: Special case sbsa-ref and xlnx-versal-virt if !CONFIG_ARM_GIC_TCG

Quoted 18 lines · Eric Auger

qom-test and test-hmp shall not run tests on sbsa-ref and xlnx-versal-virt if CONFIG_ARM_GIC_TCG is unset as those machines always instantiate GICv3.

Otherwise the tests fail with ERROR:../qom/object.c:715:object_new_with_type: assertion failed: (type != NULL)

Fixes a8a5546798c3 ("hw/intc/arm_gicv3: Introduce CONFIG_ARM_GIC_TCG Kconfig selector")
---
 tests/qtest/libqtest.c | 6 ++++++
 1 file changed, 6 insertions(+)
tests/qtest/libqtest.c +0 -0
index 41f4da4e54..f53983a28e 100644

tests/qtest/libqtest.c void qtest_cb_for_every_machine(void (*cb)(const char *machine), · line 1402

Quoted 9 lines
tests/qtest/libqtest.c +6 -0
@@ -1394,6 +1394,12 @@ void qtest_cb_for_every_machine(void (*cb)(const char *machine),
             g_str_equal("xenpv", machines[i].name)) {
             continue;
         }
+#ifndef CONFIG_ARM_GIC_TCG
+        if (!strncmp("sbsa-ref", machines[i].name, 8) ||
+            !strncmp("xlnx-versal-virt", machines[i].name, 16)) {
+            continue;
+        }
+#endif

Hmm, if these machine types completely depend on userspace gicv3 emulation, i.e. no way to use in-kernel gic or another tcg gic model, then I guess they shouldn't be built at all when ARM_GIC_TCG isn't configured. I.e.

hw/arm/Kconfig +2 -0
index 2e0049196d6c..d7cc028b049d 100644
@@ -209,6 +209,7 @@ config REALVIEW
 
 config SBSA_REF
     bool
+    depends on ARM_GIC_TCG
     imply PCI_DEVICES
     select AHCI
     select ARM_SMMUV3
@@ -378,6 +379,7 @@ config XLNX_ZYNQMP_ARM
 
 config XLNX_VERSAL
     bool
+    depends on ARM_GIC_TCG
     select ARM_GIC
     select PL011
     select CADENCE

Thanks,
drew

Quoted 6 lines
tests/qtest/libqtest.c +0 -0
         if (!skip_old_versioned ||
             !qtest_is_old_versioned_machine(machines[i].name)) {
             cb(machines[i].name);
Signature · 1 line
2.26.3
#5 lore raw

Re: [PATCH 2/2] tests/qtest: Special case sbsa-ref and xlnx-versal-virt if !CONFIG_ARM_GIC_TCG

Quoted 18 lines · Andrew Jones

Hmm, if these machine types completely depend on userspace gicv3 emulation, i.e. no way to use in-kernel gic or another tcg gic model, then I guess they shouldn't be built at all when ARM_GIC_TCG isn't configured. I.e.

hw/arm/Kconfig +1 -0
index 2e0049196d6c..d7cc028b049d 100644
@@ -209,6 +209,7 @@ config REALVIEW
 
 config SBSA_REF
     bool
+    depends on ARM_GIC_TCG
     imply PCI_DEVICES
     select AHCI
     select ARM_SMMUV3

hw/arm/Kconfig config XLNX_ZYNQMP_ARM · line 385

Quoted 7 lines
hw/arm/Kconfig +1 -0
@@ -378,6 +379,7 @@ config XLNX_ZYNQMP_ARM
 
 config XLNX_VERSAL
     bool
+    depends on ARM_GIC_TCG
     select ARM_GIC
     select PL011
     select CADENCE

I kind of agree, but isn't this kind of mixing two things?

(1) Both these machines require a GICv3 and a GICv2 won't do, so they should do something that says "if you want this machine type, you need a GICv3 device"

(2) Both these machines don't work with KVM or hvf, so if we're not building TCG then there's no point configuring in these machine models (a property they share with every other arm machine type except "virt", currently)

thanks
-- PMM

#6 lore raw

Re: [PATCH 2/2] tests/qtest: Special case sbsa-ref and xlnx-versal-virt if !CONFIG_ARM_GIC_TCG

Quoted 20 lines · Peter Maydell
Quoted 19 lines · Andrew Jones

Hmm, if these machine types completely depend on userspace gicv3 emulation, i.e. no way to use in-kernel gic or another tcg gic model, then I guess they shouldn't be built at all when ARM_GIC_TCG isn't configured. I.e.

hw/arm/Kconfig +1 -0
index 2e0049196d6c..d7cc028b049d 100644
@@ -209,6 +209,7 @@ config REALVIEW
 
 config SBSA_REF
     bool
+    depends on ARM_GIC_TCG
     imply PCI_DEVICES
     select AHCI
     select ARM_SMMUV3
@@ -378,6 +379,7 @@ config XLNX_ZYNQMP_ARM
 
Quoted 8 lines

hw/arm/Kconfig config XLNX_ZYNQMP_ARM · line 385

Quoted 6 lines
hw/arm/Kconfig +1 -0
 config XLNX_VERSAL
     bool
+    depends on ARM_GIC_TCG
     select ARM_GIC
     select PL011
     select CADENCE

I kind of agree, but isn't this kind of mixing two things?

How about two dependencies?

Quoted 4 lines

(1) Both these machines require a GICv3 and a GICv2 won't do, so they should do something that says "if you want this machine type, you need a GICv3 device"

depends on ARM_GIC_TCG (IMO, could use a rename to be gicv3 specific)

Quoted 5 lines

(2) Both these machines don't work with KVM or hvf, so if we're not building TCG then there's no point configuring in these machine models (a property they share with every other arm machine type except "virt", currently)

depends on TCG

Thanks,
drew

#7 lore raw

Re: [PATCH 2/2] tests/qtest: Special case sbsa-ref and xlnx-versal-virt if !CONFIG_ARM_GIC_TCG

Quoted 31 lines · Andrew Jones
Quoted 20 lines · Peter Maydell
Quoted 19 lines · Andrew Jones

Hmm, if these machine types completely depend on userspace gicv3 emulation, i.e. no way to use in-kernel gic or another tcg gic model, then I guess they shouldn't be built at all when ARM_GIC_TCG isn't configured. I.e.

hw/arm/Kconfig +1 -0
index 2e0049196d6c..d7cc028b049d 100644
@@ -209,6 +209,7 @@ config REALVIEW
 
 config SBSA_REF
     bool
+    depends on ARM_GIC_TCG
     imply PCI_DEVICES
     select AHCI
     select ARM_SMMUV3
@@ -378,6 +379,7 @@ config XLNX_ZYNQMP_ARM
 
Quoted 8 lines

hw/arm/Kconfig config XLNX_ZYNQMP_ARM · line 385

Quoted 6 lines
hw/arm/Kconfig +1 -0
 config XLNX_VERSAL
     bool
+    depends on ARM_GIC_TCG
     select ARM_GIC
     select PL011
     select CADENCE

I kind of agree, but isn't this kind of mixing two things?

Quoted 8 lines

How about two dependencies?

Quoted 4 lines

(1) Both these machines require a GICv3 and a GICv2 won't do, so they should do something that says "if you want this machine type, you need a GICv3 device"

depends on ARM_GIC_TCG (IMO, could use a rename to be gicv3 specific)

ARM_GIC_TCG has a "depends on ARM_GIC && TCG", though. "I need a GICv3" ought in principle to be satisfiable by the KVM GICv3.

Part of the problem here is that we've let ARM_GIC mean both GICv2 and GICv3.

Quoted 6 lines
Quoted 4 lines

(2) Both these machines don't work with KVM or hvf, so if we're not building TCG then there's no point configuring in these machine models (a property they share with every other arm machine type except "virt", currently)

depends on TCG

I would prefer a way of phrasing this that only required us to say it in the one machine that can handle not-TCG (virt) rather than in the large number of machines that cannot...

-- PMM

author
#8 lore raw

Re: [PATCH 2/2] tests/qtest: Special case sbsa-ref and xlnx-versal-virt if !CONFIG_ARM_GIC_TCG

In reply to Andrew Jones

Hi Drew,

Quoted 28 lines · Andrew Jones
Quoted 20 lines · Peter Maydell
Quoted 19 lines · Andrew Jones

Hmm, if these machine types completely depend on userspace gicv3 emulation, i.e. no way to use in-kernel gic or another tcg gic model, then I guess they shouldn't be built at all when ARM_GIC_TCG isn't configured. I.e.

hw/arm/Kconfig +1 -0
index 2e0049196d6c..d7cc028b049d 100644
@@ -209,6 +209,7 @@ config REALVIEW
 
 config SBSA_REF
     bool
+    depends on ARM_GIC_TCG
     imply PCI_DEVICES
     select AHCI
     select ARM_SMMUV3
@@ -378,6 +379,7 @@ config XLNX_ZYNQMP_ARM
 
Quoted 7 lines

hw/arm/Kconfig config XLNX_ZYNQMP_ARM · line 385

Quoted 6 lines
hw/arm/Kconfig +1 -0
 config XLNX_VERSAL
     bool
+    depends on ARM_GIC_TCG
     select ARM_GIC
     select PL011
     select CADENCE

I kind of agree, but isn't this kind of mixing two things?

Quoted 6 lines

How about two dependencies?

Quoted 3 lines

(1) Both these machines require a GICv3 and a GICv2 won't do, so they should do something that says "if you want this machine type, you need a GICv3 device"

depends on ARM_GIC_TCG (IMO, could use a rename to be gicv3 specific)

Yep I think it would be clearer to rename the CONFIG.

Quoted 6 lines
Quoted 5 lines

(2) Both these machines don't work with KVM or hvf, so if we're not building TCG then there's no point configuring in these machine models (a property they share with every other arm machine type except "virt", currently)

depends on TCG

That's what I would be inclined to do as well

Thanks

Eric

Quoted 4 lines

Thanks,
drew

author
#9 lore raw

Re: [PATCH 2/2] tests/qtest: Special case sbsa-ref and xlnx-versal-virt if !CONFIG_ARM_GIC_TCG

In reply to Andrew Jones

Hi Drew,

Quoted 24 lines · Andrew Jones
Quoted 23 lines · Eric Auger

qom-test and test-hmp shall not run tests on sbsa-ref and xlnx-versal-virt if CONFIG_ARM_GIC_TCG is unset as those machines always instantiate GICv3.

Otherwise the tests fail with ERROR:../qom/object.c:715:object_new_with_type: assertion failed: (type != NULL)

Fixes a8a5546798c3 ("hw/intc/arm_gicv3: Introduce CONFIG_ARM_GIC_TCG Kconfig selector")
---
 tests/qtest/libqtest.c | 6 ++++++
 1 file changed, 6 insertions(+)
tests/qtest/libqtest.c +2 -0
index 41f4da4e54..f53983a28e 100644
@@ -1394,6 +1394,12 @@ void qtest_cb_for_every_machine(void (*cb)(const char *machine),
             g_str_equal("xenpv", machines[i].name)) {
             continue;
         }
+#ifndef CONFIG_ARM_GIC_TCG
+        if (!strncmp("sbsa-ref", machines[i].name, 8) ||
Quoted 8 lines

tests/qtest/libqtest.c void qtest_cb_for_every_machine(void (*cb)(const char *machine), · line 1402

Quoted 4 lines
tests/qtest/libqtest.c +4 -0
+            !strncmp("xlnx-versal-virt", machines[i].name, 16)) {
+            continue;
+        }
+#endif

Hmm, if these machine types completely depend on userspace gicv3 emulation, i.e. no way to use in-kernel gic or another tcg gic model, then I guess they shouldn't be built at all when ARM_GIC_TCG isn't configured. I.e.

Yes your suggestion make sense to me, thatw would be a better fix indeed.

Eric

Quoted 32 lines
hw/arm/Kconfig +2 -0
index 2e0049196d6c..d7cc028b049d 100644
@@ -209,6 +209,7 @@ config REALVIEW
 
 config SBSA_REF
     bool
+    depends on ARM_GIC_TCG
     imply PCI_DEVICES
     select AHCI
     select ARM_SMMUV3
@@ -378,6 +379,7 @@ config XLNX_ZYNQMP_ARM
 
 config XLNX_VERSAL
     bool
+    depends on ARM_GIC_TCG
     select ARM_GIC
     select PL011
     select CADENCE

Thanks,
drew

Quoted 6 lines
tests/qtest/libqtest.c +0 -0
         if (!skip_old_versioned ||
             !qtest_is_old_versioned_machine(machines[i].name)) {
             cb(machines[i].name);
Signature · 1 line
2.26.3