Skip to contentSkip to conversation

target/arm/kvm: enable SVE in guests

PATCH Superseded Andrew Jones 4 participants 32 messages Sep 24, 2019 – Oct 1, 2019

v4 of 8 Superseded Latest is v8 →

lore
Share & tools
lore
#0 lore raw

v4 cover target/arm/kvm: enable SVE in guests

Since Linux kernel v5.2-rc1 KVM has support for enabling SVE in guests. This series provides the QEMU bits for that enablement. First, we select existing CPU properties representing features we want to advertise in addition to the SVE vector lengths and prepare them for a qmp query. Then we introduce the qmp query, applying it immediately to those selected features. We also document ARM CPU features at this time. We next add a qtest for the selected CPU features that uses the qmp query for its tests - and we continue to add tests as we add CPU features with the following patches. So then, once we have the support we need for CPU feature querying and testing, we add our first SVE CPU feature property, 'sve', which just allows SVE to be completely enabled/disabled. Following that feature property, we add all 16 vector length properties along with the input validation they need and tests to prove the validation works. At this point the SVE features are still only for TCG, so we provide some patches to prepare for KVM and then a patch that allows the 'max' CPU type to enable SVE with KVM, but at first without vector length properties. After a bit more preparation we add the SVE vector length properties to the KVM-enabled 'max' CPU type along with the additional input validation and tests that that needs. Finally we allow the 'host' CPU type to also enjoy these properties by simply sharing them with it.

v4:
- Integrated Richard Henderson's rework for the sve property validation, in order to do all validating at finalize time and save several lines of code.
- Fixed 'host' cpu SVE default. It was still off by default.
- Cleaned up #ifdef's for sve_bswap64()
- Removed redundant KVM_CAP_ARM_SVE extension check in

    kvm_arm_sve_get_vls()

- Improved the KVM SVE qtest
- Renamed sve<vl-bits> to sve<N> everywhere
- Renamed power-of-2 to power-of-two everywhere
- Picked up some more tags from Richard

Thanks!
drew

Andrew Jones (9):
target/arm/monitor: Introduce qmp_query_cpu_model_expansion
tests: arm: Introduce cpu feature tests
target/arm: Allow SVE to be disabled via a CPU property target/arm/cpu64: max cpu: Introduce sve<N> properties target/arm/kvm64: Add kvm_arch_get/put_sve target/arm/kvm64: max cpu: Enable SVE when available target/arm/kvm: scratch vcpu: Preserve input kvm_vcpu_init features target/arm/cpu64: max cpu: Support sve properties with KVM target/arm/kvm: host cpu: Add support for sve<N> properties

 docs/arm-cpu-features.rst | 301 +++++++++++++++++++++
 qapi/machine-target.json  |   6 +-
 target/arm/cpu.c          |  25 +-
 target/arm/cpu.h          |  21 ++
 target/arm/cpu64.c        | 341 +++++++++++++++++++++++-
 target/arm/helper.c       |  13 +-
 target/arm/kvm.c          |  25 +-
 target/arm/kvm32.c        |   6 +-
 target/arm/kvm64.c        | 277 +++++++++++++++++++-
 target/arm/kvm_arm.h      |  39 +++
 target/arm/monitor.c      | 157 +++++++++++
 tests/Makefile.include    |   5 +-
 tests/arm-cpu-features.c  | 537 ++++++++++++++++++++++++++++++++++++++
 13 files changed, 1718 insertions(+), 35 deletions(-)
 create mode 100644 docs/arm-cpu-features.rst
 create mode 100644 tests/arm-cpu-features.c
Signature · 1 line
2.20.1
#1 lore raw

PATCH v4 1/9 target/arm/monitor: Introduce qmp_query_cpu_model_expansion

Add support for the query-cpu-model-expansion QMP command to Arm. We do this selectively, only exposing CPU properties which represent optional CPU features which the user may want to enable/disable. Additionally we restrict the list of queryable cpu models to 'max', 'host', or the current type when KVM is in use. And, finally, we only implement expansion type 'full', as Arm does not yet have a "base" CPU type. More details and example queries are described in a new document (docs/arm-cpu-features.rst).

Note, certainly more features may be added to the list of advertised features, e.g. 'vfp' and 'neon'. The only requirement is that we can detect invalid configurations and emit failures at QMP query time. For 'vfp' and 'neon' this will require some refactoring to share a validation function between the QMP query and the CPU realize functions.

---
 docs/arm-cpu-features.rst | 137 +++++++++++++++++++++++++++++++++++
 qapi/machine-target.json  |   6 +-
 target/arm/monitor.c      | 145 ++++++++++++++++++++++++++++++++++++++
 3 files changed, 285 insertions(+), 3 deletions(-)
 create mode 100644 docs/arm-cpu-features.rst
docs/arm-cpu-features.rst new file +137 -0
new file mode 100644
index 000000000000..c79dcffb5556
@@ -0,0 +1,137 @@
+================
+ARM CPU Features
+================
+
+Examples of probing and using ARM CPU features
+
+Introduction
+============
+
+CPU features are optional features that a CPU of supporting type may
+choose to implement or not.  In QEMU, optional CPU features have
+corresponding boolean CPU proprieties that, when enabled, indicate
+that the feature is implemented, and, conversely, when disabled,
+indicate that it is not implemented. An example of an ARM CPU feature
+is the Performance Monitoring Unit (PMU).  CPU types such as the
+Cortex-A15 and the Cortex-A57, which respectively implement ARM
+architecture reference manuals ARMv7-A and ARMv8-A, may both optionally
+implement PMUs.  For example, if a user wants to use a Cortex-A15 without
+a PMU, then the `-cpu` parameter should contain `pmu=off` on the QEMU
+command line, i.e. `-cpu cortex-a15,pmu=off`.
+
+As not all CPU types support all optional CPU features, then whether or
+not a CPU property exists depends on the CPU type.  For example, CPUs
+that implement the ARMv8-A architecture reference manual may optionally
+support the AArch32 CPU feature, which may be enabled by disabling the
+`aarch64` CPU property.  A CPU type such as the Cortex-A15, which does
+not implement ARMv8-A, will not have the `aarch64` CPU property.
+
+QEMU's support may be limited for some CPU features, only partially
+supporting the feature or only supporting the feature under certain
+configurations.  For example, the `aarch64` CPU feature, which, when
+disabled, enables the optional AArch32 CPU feature, is only supported
+when using the KVM accelerator and when running on a host CPU type that
+supports the feature.
+
+CPU Feature Probing
+===================
+
+Determining which CPU features are available and functional for a given
+CPU type is possible with the `query-cpu-model-expansion` QMP command.
+Below are some examples where `scripts/qmp/qmp-shell` (see the top comment
+block in the script for usage) is used to issue the QMP commands.
+
+(1) Determine which CPU features are available for the `max` CPU type
+    (Note, we started QEMU with qemu-system-aarch64, so `max` is
+     implementing the ARMv8-A reference manual in this case)::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max"}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "pmu": true, "aarch64": true
+      }}}}
+
+We see that the `max` CPU type has the `pmu` and `aarch64` CPU features.
+We also see that the CPU features are enabled, as they are all `true`.
+
+(2) Let's try to disable the PMU::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"pmu":false}}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "pmu": false, "aarch64": true
+      }}}}
+
+We see it worked, as `pmu` is now `false`.
+
+(3) Let's try to disable `aarch64`, which enables the AArch32 CPU feature::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"aarch64":false}}
+      {"error": {
+       "class": "GenericError", "desc":
+       "'aarch64' feature cannot be disabled unless KVM is enabled and 32-bit EL1 is supported"
+      }}
+
+It looks like this feature is limited to a configuration we do not
+currently have.
+
+(4) Let's try probing CPU features for the Cortex-A15 CPU type::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"cortex-a15"}
+      {"return": {"model": {"name": "cortex-a15", "props": {"pmu": true}}}}
+
+Only the `pmu` CPU feature is available.
+
+A note about CPU feature dependencies
+-------------------------------------
+
+It's possible for features to have dependencies on other features. I.e.
+it may be possible to change one feature at a time without error, but
+when attempting to change all features at once an error could occur
+depending on the order they are processed.  It's also possible changing
+all at once doesn't generate an error, because a feature's dependencies
+are satisfied with other features, but the same feature cannot be changed
+independently without error.  For these reasons callers should always
+attempt to make their desired changes all at once in order to ensure the
+collection is valid.
+
+A note about CPU models and KVM
+-------------------------------
+
+Named CPU models generally do not work with KVM.  There are a few cases
+that do work, e.g. using the named CPU model `cortex-a57` with KVM on a
+seattle host, but mostly if KVM is enabled the `host` CPU type must be
+used.  This means the guest is provided all the same CPU features as the
+host CPU type has.  And, for this reason, the `host` CPU type should
+enable all CPU features that the host has by default.  Indeed it's even
+a bit strange to allow disabling CPU features that the host has when using
+the `host` CPU type, but in the absence of CPU models it's the best we can
+do if we want to launch guests without all the host's CPU features enabled.
+
+Enabling KVM also affects the `query-cpu-model-expansion` QMP command.  The
+affect is not only limited to specific features, as pointed out in example
+(3) of "CPU Feature Probing", but also to which CPU types may be expanded.
+When KVM is enabled, only the `max`, `host`, and current CPU type may be
+expanded.  This restriction is necessary as it's not possible to know all
+CPU types that may work with KVM, but it does impose a small risk of users
+experiencing unexpected errors.  For example on a seattle, as mentioned
+above, the `cortex-a57` CPU type is also valid when KVM is enabled.
+Therefore a user could use the `host` CPU type for the current type, but
+then attempt to query `cortex-a57`, however that query will fail with our
+restrictions.  This shouldn't be an issue though as management layers and
+users have been preferring the `host` CPU type for use with KVM for quite
+some time.  Additionally, if the KVM-enabled QEMU instance running on a
+seattle host is using the `cortex-a57` CPU type, then querying `cortex-a57`
+will work.
+
+Using CPU Features
+==================
+
+After determining which CPU features are available and supported for a
+given CPU type, then they may be selectively enabled or disabled on the
+QEMU command line with that CPU type::
+
+  $ qemu-system-aarch64 -M virt -cpu max,pmu=off
+
+The example above disables the PMU for the `max` CPU type.
+
qapi/machine-target.json +3 -3
index 55310a6aa226..04623224720d 100644
@@ -212,7 +212,7 @@
 ##
 { 'struct': 'CpuModelExpansionInfo',
   'data': { 'model': 'CpuModelInfo' },
-  'if': 'defined(TARGET_S390X) || defined(TARGET_I386)' }
+  'if': 'defined(TARGET_S390X) || defined(TARGET_I386) || defined(TARGET_ARM)' }
 
 ##
 # @query-cpu-model-expansion:
@@ -237,7 +237,7 @@
 #   query-cpu-model-expansion while using these is not advised.
 #
 # Some architectures may not support all expansion types. s390x supports
-# "full" and "static".
+# "full" and "static". Arm only supports "full".
 #
 # Returns: a CpuModelExpansionInfo. Returns an error if expanding CPU models is
 #          not supported, if the model cannot be expanded, if the model contains
@@ -251,7 +251,7 @@
   'data': { 'type': 'CpuModelExpansionType',
             'model': 'CpuModelInfo' },
   'returns': 'CpuModelExpansionInfo',
-  'if': 'defined(TARGET_S390X) || defined(TARGET_I386)' }
+  'if': 'defined(TARGET_S390X) || defined(TARGET_I386) || defined(TARGET_ARM)' }
 
 ##
 # @CpuDefinitionInfo:
target/arm/monitor.c +145 -0

156 lines not on this page · raw message

Signature · 1 line
2.20.1
Reviewed-by on v4 1/9
#2 lore raw

Re: [PATCH v4 1/9] target/arm/monitor: Introduce qmp_query_cpu_model_expansion

Hi Drew,

Quoted 16 lines · Andrew Jones

Add support for the query-cpu-model-expansion QMP command to Arm. We do this selectively, only exposing CPU properties which represent optional CPU features which the user may want to enable/disable. Additionally we restrict the list of queryable cpu models to 'max', 'host', or the current type when KVM is in use. And, finally, we only implement expansion type 'full', as Arm does not yet have a "base" CPU type. More details and example queries are described in a new document (docs/arm-cpu-features.rst).

Note, certainly more features may be added to the list of advertised features, e.g. 'vfp' and 'neon'. The only requirement is that we can detect invalid configurations and emit failures at QMP query time. For 'vfp' and 'neon' this will require some refactoring to share a validation function between the QMP query and the CPU realize functions.

Quoted 2 lines

Thanks

Eric

Quoted 344 lines
---
 docs/arm-cpu-features.rst | 137 +++++++++++++++++++++++++++++++++++
 qapi/machine-target.json  |   6 +-
 target/arm/monitor.c      | 145 ++++++++++++++++++++++++++++++++++++++
 3 files changed, 285 insertions(+), 3 deletions(-)
 create mode 100644 docs/arm-cpu-features.rst
docs/arm-cpu-features.rst new file +137 -0
new file mode 100644
index 000000000000..c79dcffb5556
@@ -0,0 +1,137 @@
+================
+ARM CPU Features
+================
+
+Examples of probing and using ARM CPU features
+
+Introduction
+============
+
+CPU features are optional features that a CPU of supporting type may
+choose to implement or not.  In QEMU, optional CPU features have
+corresponding boolean CPU proprieties that, when enabled, indicate
+that the feature is implemented, and, conversely, when disabled,
+indicate that it is not implemented. An example of an ARM CPU feature
+is the Performance Monitoring Unit (PMU).  CPU types such as the
+Cortex-A15 and the Cortex-A57, which respectively implement ARM
+architecture reference manuals ARMv7-A and ARMv8-A, may both optionally
+implement PMUs.  For example, if a user wants to use a Cortex-A15 without
+a PMU, then the `-cpu` parameter should contain `pmu=off` on the QEMU
+command line, i.e. `-cpu cortex-a15,pmu=off`.
+
+As not all CPU types support all optional CPU features, then whether or
+not a CPU property exists depends on the CPU type.  For example, CPUs
+that implement the ARMv8-A architecture reference manual may optionally
+support the AArch32 CPU feature, which may be enabled by disabling the
+`aarch64` CPU property.  A CPU type such as the Cortex-A15, which does
+not implement ARMv8-A, will not have the `aarch64` CPU property.
+
+QEMU's support may be limited for some CPU features, only partially
+supporting the feature or only supporting the feature under certain
+configurations.  For example, the `aarch64` CPU feature, which, when
+disabled, enables the optional AArch32 CPU feature, is only supported
+when using the KVM accelerator and when running on a host CPU type that
+supports the feature.
+
+CPU Feature Probing
+===================
+
+Determining which CPU features are available and functional for a given
+CPU type is possible with the `query-cpu-model-expansion` QMP command.
+Below are some examples where `scripts/qmp/qmp-shell` (see the top comment
+block in the script for usage) is used to issue the QMP commands.
+
+(1) Determine which CPU features are available for the `max` CPU type
+    (Note, we started QEMU with qemu-system-aarch64, so `max` is
+     implementing the ARMv8-A reference manual in this case)::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max"}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "pmu": true, "aarch64": true
+      }}}}
+
+We see that the `max` CPU type has the `pmu` and `aarch64` CPU features.
+We also see that the CPU features are enabled, as they are all `true`.
+
+(2) Let's try to disable the PMU::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"pmu":false}}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "pmu": false, "aarch64": true
+      }}}}
+
+We see it worked, as `pmu` is now `false`.
+
+(3) Let's try to disable `aarch64`, which enables the AArch32 CPU feature::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"aarch64":false}}
+      {"error": {
+       "class": "GenericError", "desc":
+       "'aarch64' feature cannot be disabled unless KVM is enabled and 32-bit EL1 is supported"
+      }}
+
+It looks like this feature is limited to a configuration we do not
+currently have.
+
+(4) Let's try probing CPU features for the Cortex-A15 CPU type::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"cortex-a15"}
+      {"return": {"model": {"name": "cortex-a15", "props": {"pmu": true}}}}
+
+Only the `pmu` CPU feature is available.
+
+A note about CPU feature dependencies
+-------------------------------------
+
+It's possible for features to have dependencies on other features. I.e.
+it may be possible to change one feature at a time without error, but
+when attempting to change all features at once an error could occur
+depending on the order they are processed.  It's also possible changing
+all at once doesn't generate an error, because a feature's dependencies
+are satisfied with other features, but the same feature cannot be changed
+independently without error.  For these reasons callers should always
+attempt to make their desired changes all at once in order to ensure the
+collection is valid.
+
+A note about CPU models and KVM
+-------------------------------
+
+Named CPU models generally do not work with KVM.  There are a few cases
+that do work, e.g. using the named CPU model `cortex-a57` with KVM on a
+seattle host, but mostly if KVM is enabled the `host` CPU type must be
+used.  This means the guest is provided all the same CPU features as the
+host CPU type has.  And, for this reason, the `host` CPU type should
+enable all CPU features that the host has by default.  Indeed it's even
+a bit strange to allow disabling CPU features that the host has when using
+the `host` CPU type, but in the absence of CPU models it's the best we can
+do if we want to launch guests without all the host's CPU features enabled.
+
+Enabling KVM also affects the `query-cpu-model-expansion` QMP command.  The
+affect is not only limited to specific features, as pointed out in example
+(3) of "CPU Feature Probing", but also to which CPU types may be expanded.
+When KVM is enabled, only the `max`, `host`, and current CPU type may be
+expanded.  This restriction is necessary as it's not possible to know all
+CPU types that may work with KVM, but it does impose a small risk of users
+experiencing unexpected errors.  For example on a seattle, as mentioned
+above, the `cortex-a57` CPU type is also valid when KVM is enabled.
+Therefore a user could use the `host` CPU type for the current type, but
+then attempt to query `cortex-a57`, however that query will fail with our
+restrictions.  This shouldn't be an issue though as management layers and
+users have been preferring the `host` CPU type for use with KVM for quite
+some time.  Additionally, if the KVM-enabled QEMU instance running on a
+seattle host is using the `cortex-a57` CPU type, then querying `cortex-a57`
+will work.
+
+Using CPU Features
+==================
+
+After determining which CPU features are available and supported for a
+given CPU type, then they may be selectively enabled or disabled on the
+QEMU command line with that CPU type::
+
+  $ qemu-system-aarch64 -M virt -cpu max,pmu=off
+
+The example above disables the PMU for the `max` CPU type.
+
qapi/machine-target.json +3 -3
index 55310a6aa226..04623224720d 100644
@@ -212,7 +212,7 @@
 ##
 { 'struct': 'CpuModelExpansionInfo',
   'data': { 'model': 'CpuModelInfo' },
-  'if': 'defined(TARGET_S390X) || defined(TARGET_I386)' }
+  'if': 'defined(TARGET_S390X) || defined(TARGET_I386) || defined(TARGET_ARM)' }
 
 ##
 # @query-cpu-model-expansion:
@@ -237,7 +237,7 @@
 #   query-cpu-model-expansion while using these is not advised.
 #
 # Some architectures may not support all expansion types. s390x supports
-# "full" and "static".
+# "full" and "static". Arm only supports "full".
 #
 # Returns: a CpuModelExpansionInfo. Returns an error if expanding CPU models is
 #          not supported, if the model cannot be expanded, if the model contains
@@ -251,7 +251,7 @@
   'data': { 'type': 'CpuModelExpansionType',
             'model': 'CpuModelInfo' },
   'returns': 'CpuModelExpansionInfo',
-  'if': 'defined(TARGET_S390X) || defined(TARGET_I386)' }
+  'if': 'defined(TARGET_S390X) || defined(TARGET_I386) || defined(TARGET_ARM)' }
 
 ##
 # @CpuDefinitionInfo:
target/arm/monitor.c +145 -0
index 6457c3c87f7c..edca8aa885f0 100644
@@ -21,8 +21,16 @@
  */
 
 #include "qemu/osdep.h"
+#include "hw/boards.h"
 #include "kvm_arm.h"
+#include "qapi/error.h"
+#include "qapi/visitor.h"
+#include "qapi/qobject-input-visitor.h"
+#include "qapi/qapi-commands-machine-target.h"
 #include "qapi/qapi-commands-misc-target.h"
+#include "qapi/qmp/qerror.h"
+#include "qapi/qmp/qdict.h"
+#include "qom/qom-qobject.h"
 
 static GICCapability *gic_cap_new(int version)
 {
@@ -81,3 +89,140 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
 
     return head;
 }
+
+/*
+ * These are cpu model features we want to advertise. The order here
+ * matters as this is the order in which qmp_query_cpu_model_expansion
+ * will attempt to set them. If there are dependencies between features,
+ * then the order that considers those dependencies must be used.
+ */
+static const char *cpu_model_advertised_features[] = {
+    "aarch64", "pmu",
+    NULL
+};
+
+CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
+                                                     CpuModelInfo *model,
+                                                     Error **errp)
+{
+    CpuModelExpansionInfo *expansion_info;
+    const QDict *qdict_in = NULL;
+    QDict *qdict_out;
+    ObjectClass *oc;
+    Object *obj;
+    const char *name;
+    int i;
+
+    if (type != CPU_MODEL_EXPANSION_TYPE_FULL) {
+        error_setg(errp, "The requested expansion type is not supported");
+        return NULL;
+    }
+
+    if (!kvm_enabled() && !strcmp(model->name, "host")) {
+        error_setg(errp, "The CPU type '%s' requires KVM", model->name);
+        return NULL;
+    }
+
+    oc = cpu_class_by_name(TYPE_ARM_CPU, model->name);
+    if (!oc) {
+        error_setg(errp, "The CPU type '%s' is not a recognized ARM CPU type",
+                   model->name);
+        return NULL;
+    }
+
+    if (kvm_enabled()) {
+        const char *cpu_type = current_machine->cpu_type;
+        int len = strlen(cpu_type) - strlen(ARM_CPU_TYPE_SUFFIX);
+        bool supported = false;
+
+        if (!strcmp(model->name, "host") || !strcmp(model->name, "max")) {
+            /* These are kvmarm's recommended cpu types */
+            supported = true;
+        } else if (strlen(model->name) == len &&
+                   !strncmp(model->name, cpu_type, len)) {
+            /* KVM is enabled and we're using this type, so it works. */
+            supported = true;
+        }
+        if (!supported) {
+            error_setg(errp, "We cannot guarantee the CPU type '%s' works "
+                             "with KVM on this host", model->name);
+            return NULL;
+        }
+    }
+
+    if (model->props) {
+        qdict_in = qobject_to(QDict, model->props);
+        if (!qdict_in) {
+            error_setg(errp, QERR_INVALID_PARAMETER_TYPE, "props", "dict");
+            return NULL;
+        }
+    }
+
+    obj = object_new(object_class_get_name(oc));
+
+    if (qdict_in) {
+        Visitor *visitor;
+        Error *err = NULL;
+
+        visitor = qobject_input_visitor_new(model->props);
+        visit_start_struct(visitor, NULL, NULL, 0, &err);
+        if (err) {
+            object_unref(obj);
+            error_propagate(errp, err);
+            return NULL;
+        }
+
+        i = 0;
+        while ((name = cpu_model_advertised_features[i++]) != NULL) {
+            if (qdict_get(qdict_in, name)) {
+                object_property_set(obj, visitor, name, &err);
+                if (err) {
+                    break;
+                }
+            }
+        }
+
+        if (!err) {
+            visit_check_struct(visitor, &err);
+        }
+        visit_end_struct(visitor, NULL);
+        visit_free(visitor);
+        if (err) {
+            object_unref(obj);
+            error_propagate(errp, err);
+            return NULL;
+        }
+    }
+
+    expansion_info = g_new0(CpuModelExpansionInfo, 1);
+    expansion_info->model = g_malloc0(sizeof(*expansion_info->model));
+    expansion_info->model->name = g_strdup(model->name);
+
+    qdict_out = qdict_new();
+
+    i = 0;
+    while ((name = cpu_model_advertised_features[i++]) != NULL) {
+        ObjectProperty *prop = object_property_find(obj, name, NULL);
+        if (prop) {
+            Error *err = NULL;
+            QObject *value;
+
+            assert(prop->get);
+            value = object_property_get_qobject(obj, name, &err);
+            assert(!err);
+
+            qdict_put_obj(qdict_out, name, value);
+        }
+    }
+
+    if (!qdict_size(qdict_out)) {
+        qobject_unref(qdict_out);
+    } else {
+        expansion_info->model->props = QOBJECT(qdict_out);
+        expansion_info->model->has_props = true;
+    }
+
+    object_unref(obj);
+
+    return expansion_info;
+}
#3 lore raw

PATCH v4 2/9 tests: arm: Introduce cpu feature tests

In reply to Andrew Jones

Now that Arm CPUs have advertised features lets add tests to ensure we maintain their expected availability with and without KVM.

---
 tests/Makefile.include   |   5 +-
 tests/arm-cpu-features.c | 242 +++++++++++++++++++++++++++++++++++++++
 2 files changed, 246 insertions(+), 1 deletion(-)
 create mode 100644 tests/arm-cpu-features.c
tests/Makefile.include +4 -1
index 793632ca72a5..3eee8fdf9637 100644
@@ -258,6 +258,7 @@ check-qtest-sparc64-$(CONFIG_ISA_TESTDEV) = tests/endianness-test$(EXESUF)
 check-qtest-sparc64-y += tests/prom-env-test$(EXESUF)
 check-qtest-sparc64-y += tests/boot-serial-test$(EXESUF)
 
+check-qtest-arm-y += tests/arm-cpu-features$(EXESUF)
 check-qtest-arm-y += tests/microbit-test$(EXESUF)
 check-qtest-arm-y += tests/m25p80-test$(EXESUF)
 check-qtest-arm-y += tests/test-arm-mptimer$(EXESUF)
@@ -265,7 +266,8 @@ check-qtest-arm-y += tests/boot-serial-test$(EXESUF)
 check-qtest-arm-y += tests/hexloader-test$(EXESUF)
 check-qtest-arm-$(CONFIG_PFLASH_CFI02) += tests/pflash-cfi02-test$(EXESUF)
 
-check-qtest-aarch64-y = tests/numa-test$(EXESUF)
+check-qtest-aarch64-y += tests/arm-cpu-features$(EXESUF)
+check-qtest-aarch64-y += tests/numa-test$(EXESUF)
 check-qtest-aarch64-y += tests/boot-serial-test$(EXESUF)
 check-qtest-aarch64-y += tests/migration-test$(EXESUF)
 # TODO: once aarch64 TCG is fixed on ARM 32 bit host, make test unconditional
@@ -828,6 +830,7 @@ tests/test-qapi-util$(EXESUF): tests/test-qapi-util.o $(test-util-obj-y)
 tests/numa-test$(EXESUF): tests/numa-test.o
 tests/vmgenid-test$(EXESUF): tests/vmgenid-test.o tests/boot-sector.o tests/acpi-utils.o
 tests/cdrom-test$(EXESUF): tests/cdrom-test.o tests/boot-sector.o $(libqos-obj-y)
+tests/arm-cpu-features$(EXESUF): tests/arm-cpu-features.o
 
 tests/migration/stress$(EXESUF): tests/migration/stress.o
 	$(call quiet-command, $(LINKPROG) -static -O3 $(PTHREAD_LIB) -o $@ $< ,"LINK","$(TARGET_DIR)$@")
tests/arm-cpu-features.c new file +242 -0

242 lines not on this page · raw message

Signature · 1 line
2.20.1
#4 lore raw

PATCH v4 3/9 target/arm: Allow SVE to be disabled via a CPU property

In reply to Andrew Jones

Since 97a28b0eeac14 ("target/arm: Allow VFP and Neon to be disabled via a CPU property") we can disable the 'max' cpu model's VFP and neon features, but there's no way to disable SVE. Add the 'sve=on|off' property to give it that flexibility. We also rename cpu_max_get/set_sve_vq to cpu_max_get/set_sve_max_vq in order for them to follow the typical *_get/set_<property-name> pattern.

---
 target/arm/cpu.c         |  3 ++-
 target/arm/cpu64.c       | 42 ++++++++++++++++++++++++++++++++++------
 target/arm/monitor.c     |  2 +-
 tests/arm-cpu-features.c |  1 +
 4 files changed, 40 insertions(+), 8 deletions(-)
target/arm/cpu.c +2 -1
index 2399c144718d..73be2ebfdd39 100644
@@ -200,7 +200,8 @@ static void arm_cpu_reset(CPUState *s)
         env->cp15.cpacr_el1 = deposit64(env->cp15.cpacr_el1, 16, 2, 3);
         env->cp15.cptr_el[3] |= CPTR_EZ;
         /* with maximum vector length */
-        env->vfp.zcr_el[1] = cpu->sve_max_vq - 1;
+        env->vfp.zcr_el[1] = cpu_isar_feature(aa64_sve, cpu) ?
+                             cpu->sve_max_vq - 1 : 0;
         env->vfp.zcr_el[2] = env->vfp.zcr_el[1];
         env->vfp.zcr_el[3] = env->vfp.zcr_el[1];
         /*
target/arm/cpu64.c +36 -6

65 lines not on this page · raw message

target/arm/monitor.c +1 -1
index edca8aa885f0..4fddb6c252a3 100644
@@ -97,7 +97,7 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
  * then the order that considers those dependencies must be used.
  */
 static const char *cpu_model_advertised_features[] = {
-    "aarch64", "pmu",
+    "aarch64", "pmu", "sve",
     NULL
 };
 
tests/arm-cpu-features.c +1 -0
index 198ff6d6b495..202bc0e3e823 100644
@@ -179,6 +179,7 @@ static void test_query_cpu_model_expansion(const void *data)
 
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
         assert_has_feature(qts, "max", "aarch64");
+        assert_has_feature(qts, "max", "sve");
         assert_has_feature(qts, "cortex-a57", "pmu");
         assert_has_feature(qts, "cortex-a57", "aarch64");
 
Signature · 1 line
2.20.1
Reviewed-by on v4 3/9
#5 lore raw

Re: [PATCH v4 3/9] target/arm: Allow SVE to be disabled via a CPU property

Hi Drew,

Quoted 9 lines · Andrew Jones

Since 97a28b0eeac14 ("target/arm: Allow VFP and Neon to be disabled via a CPU property") we can disable the 'max' cpu model's VFP and neon features, but there's no way to disable SVE. Add the 'sve=on|off' property to give it that flexibility. We also rename cpu_max_get/set_sve_vq to cpu_max_get/set_sve_max_vq in order for them to follow the typical *_get/set_<property-name> pattern.

Eric

Quoted 118 lines
---
 target/arm/cpu.c         |  3 ++-
 target/arm/cpu64.c       | 42 ++++++++++++++++++++++++++++++++++------
 target/arm/monitor.c     |  2 +-
 tests/arm-cpu-features.c |  1 +
 4 files changed, 40 insertions(+), 8 deletions(-)
target/arm/cpu.c +2 -1
index 2399c144718d..73be2ebfdd39 100644
@@ -200,7 +200,8 @@ static void arm_cpu_reset(CPUState *s)
         env->cp15.cpacr_el1 = deposit64(env->cp15.cpacr_el1, 16, 2, 3);
         env->cp15.cptr_el[3] |= CPTR_EZ;
         /* with maximum vector length */
-        env->vfp.zcr_el[1] = cpu->sve_max_vq - 1;
+        env->vfp.zcr_el[1] = cpu_isar_feature(aa64_sve, cpu) ?
+                             cpu->sve_max_vq - 1 : 0;
         env->vfp.zcr_el[2] = env->vfp.zcr_el[1];
         env->vfp.zcr_el[3] = env->vfp.zcr_el[1];
         /*
target/arm/cpu64.c +35 -6
index d7f5bf610a7d..8cdb0c79fa7a 100644
@@ -256,15 +256,15 @@ static void aarch64_a72_initfn(Object *obj)
     define_arm_cp_regs(cpu, cortex_a72_a57_a53_cp_reginfo);
 }
 
-static void cpu_max_get_sve_vq(Object *obj, Visitor *v, const char *name,
-                               void *opaque, Error **errp)
+static void cpu_max_get_sve_max_vq(Object *obj, Visitor *v, const char *name,
+                                   void *opaque, Error **errp)
 {
     ARMCPU *cpu = ARM_CPU(obj);
     visit_type_uint32(v, name, &cpu->sve_max_vq, errp);
 }
 
-static void cpu_max_set_sve_vq(Object *obj, Visitor *v, const char *name,
-                               void *opaque, Error **errp)
+static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,
+                                   void *opaque, Error **errp)
 {
     ARMCPU *cpu = ARM_CPU(obj);
     Error *err = NULL;
@@ -279,6 +279,34 @@ static void cpu_max_set_sve_vq(Object *obj, Visitor *v, const char *name,
     error_propagate(errp, err);
 }
 
+static void cpu_arm_get_sve(Object *obj, Visitor *v, const char *name,
+                            void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    bool value = cpu_isar_feature(aa64_sve, cpu);
+
+    visit_type_bool(v, name, &value, errp);
+}
+
+static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
+                            void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    Error *err = NULL;
+    bool value;
+    uint64_t t;
+
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
+
+    t = cpu->isar.id_aa64pfr0;> +    t = FIELD_DP64(t, ID_AA64PFR0, SVE, value);
+    cpu->isar.id_aa64pfr0 = t;
+}
+
 /* -cpu max: if KVM is enabled, like -cpu host (best possible with this host);
  * otherwise, a CPU with as many features enabled as our emulation supports.
  * The version of '-cpu max' for qemu-system-arm is defined in cpu.c;
@@ -391,8 +419,10 @@ static void aarch64_max_initfn(Object *obj)
 #endif
 
         cpu->sve_max_vq = ARM_MAX_VQ;
-        object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_vq,
-                            cpu_max_set_sve_vq, NULL, NULL, &error_fatal);
+        object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
+                            cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
+        object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
+                            cpu_arm_set_sve, NULL, NULL, &error_fatal);
     }
 }
 
target/arm/monitor.c +1 -1
index edca8aa885f0..4fddb6c252a3 100644
@@ -97,7 +97,7 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
  * then the order that considers those dependencies must be used.
  */
 static const char *cpu_model_advertised_features[] = {
-    "aarch64", "pmu",
+    "aarch64", "pmu", "sve",
     NULL
 };
 
tests/arm-cpu-features.c +1 -0
index 198ff6d6b495..202bc0e3e823 100644
@@ -179,6 +179,7 @@ static void test_query_cpu_model_expansion(const void *data)
 
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
         assert_has_feature(qts, "max", "aarch64");
+        assert_has_feature(qts, "max", "sve");
         assert_has_feature(qts, "cortex-a57", "pmu");
         assert_has_feature(qts, "cortex-a57", "aarch64");
 
#6 lore raw

PATCH v4 4/9 target/arm/cpu64: max cpu: Introduce sve<N> properties

In reply to Andrew Jones

Introduce cpu properties to give fine control over SVE vector lengths. We introduce a property for each valid length up to the current maximum supported, which is 2048-bits. The properties are named, e.g. sve128, sve256, sve384, sve512, ..., where the number is the number of bits. See the updates to docs/arm-cpu-features.rst for a description of the semantics and for example uses.

Note, as sve-max-vq is still present and we'd like to be able to support qmp_query_cpu_model_expansion with guests launched with e.g. -cpu max,sve-max-vq=8 on their command lines, then we do allow sve-max-vq and sve<N> properties to be provided at the same time, but this is not recommended, and is why sve-max-vq is not mentioned in the document. If sve-max-vq is provided then it enables all lengths smaller than and including the max and disables all lengths larger. It also has the side-effect that no larger lengths may be enabled and that the max itself cannot be disabled. Smaller non-power-of-two lengths may, however, be disabled, e.g. -cpu max,sve-max-vq=4,sve384=off provides a guest the vector lengths 128, 256, and 512 bits.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst | 157 +++++++++++++++++++++++++++++--
 target/arm/cpu.c          |  19 ++++
 target/arm/cpu.h          |  19 ++++
 target/arm/cpu64.c        | 190 +++++++++++++++++++++++++++++++++++++-
 target/arm/helper.c       |  13 ++-
 target/arm/monitor.c      |  12 +++
 tests/arm-cpu-features.c  | 189 +++++++++++++++++++++++++++++++++++++
 7 files changed, 587 insertions(+), 12 deletions(-)
docs/arm-cpu-features.rst +150 -7
index c79dcffb5556..1262fddc6201 100644
@@ -48,18 +48,28 @@ block in the script for usage) is used to issue the QMP commands.
       (QEMU) query-cpu-model-expansion type=full model={"name":"max"}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": true, "aarch64": true
+        "sve1664": true, "pmu": true, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true
       }}}}
 
-We see that the `max` CPU type has the `pmu` and `aarch64` CPU features.
-We also see that the CPU features are enabled, as they are all `true`.
+We see that the `max` CPU type has the `pmu`, `aarch64`, `sve`, and many
+`sve<N>` CPU features.  We also see that all the CPU features are
+enabled, as they are all `true`.  (The `sve<N>` CPU features are all
+optional SVE vector lengths.  See "SVE CPU Properties".)
 
 (2) Let's try to disable the PMU::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"pmu":false}}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": false, "aarch64": true
+        "sve1664": true, "pmu": false, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true
       }}}}
 
 We see it worked, as `pmu` is now `false`.
@@ -75,7 +85,22 @@ We see it worked, as `pmu` is now `false`.
 It looks like this feature is limited to a configuration we do not
 currently have.
 
-(4) Let's try probing CPU features for the Cortex-A15 CPU type::
+(4) Let's disable `sve` and see what happens to all the optional SVE
+    vector lengths::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"sve":false}}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "sve1664": false, "pmu": true, "sve1792": false, "sve1920": false,
+        "sve128": false, "aarch64": true, "sve1024": false, "sve": false,
+        "sve640": false, "sve768": false, "sve1408": false, "sve256": false,
+        "sve1152": false, "sve512": false, "sve384": false, "sve1536": false,
+        "sve896": false, "sve1280": false, "sve2048": false
+      }}}}
+
+As expected they are now all `false`.
+
+(5) Let's try probing CPU features for the Cortex-A15 CPU type::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"cortex-a15"}
       {"return": {"model": {"name": "cortex-a15", "props": {"pmu": true}}}}
@@ -131,7 +156,125 @@ After determining which CPU features are available and supported for a
 given CPU type, then they may be selectively enabled or disabled on the
 QEMU command line with that CPU type::
 
-  $ qemu-system-aarch64 -M virt -cpu max,pmu=off
+  $ qemu-system-aarch64 -M virt -cpu max,pmu=off,sve=on,sve128=on,sve256=on
+
+The example above disables the PMU and enables the first two SVE vector
+lengths for the `max` CPU type.  Note, the `sve=on` isn't actually
+necessary, because, as we observed above with our probe of the `max` CPU
+type, `sve` is already on by default.  Also, based on our probe of
+defaults, it would seem we need to disable many SVE vector lengths, rather
+than only enabling the two we want.  This isn't the case, because, as
+disabling many SVE vector lengths would be quite verbose, the `sve<N>` CPU
+properties have special semantics (see "SVE CPU Property Parsing
+Semantics").
+
+SVE CPU Properties
+==================
+
+There are two types of SVE CPU properties: `sve` and `sve<N>`.  The first
+is used to enable or disable the entire SVE feature, just as the `pmu`
+CPU property completely enables or disables the PMU.  The second type
+is used to enable or disable specific vector lengths, where `N` is the
+number of bits of the length.  The `sve<N>` CPU properties have special
+dependencies and constraints, see "SVE CPU Property Dependencies and
+Constraints" below.  Additionally, as we want all supported vector lengths
+to be enabled by default, then, in order to avoid overly verbose command
+lines (command lines full of `sve<N>=off`, for all `N` not wanted), we
+provide the parsing semantics listed in "SVE CPU Property Parsing
+Semantics".
+
+SVE CPU Property Dependencies and Constraints
+---------------------------------------------
+
+  1) At least one vector length must be enabled when `sve` is enabled.
+
+  2) If a vector length `N` is enabled, then all power-of-two vector
+     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
+     is enabled, then `sve128` and `sve256` must also be enabled,
+     but `sve384` is not required.
+
+SVE CPU Property Parsing Semantics
+----------------------------------
+
+  1) If SVE is disabled (`sve=off`), then which SVE vector lengths
+     are enabled or disabled is irrelevant to the guest, as the entire
+     SVE feature is disabled and that disables all vector lengths for
+     the guest.  However QEMU will still track any `sve<N>` CPU
+     properties provided by the user.  If later an `sve=on` is provided,
+     then the guest will get only the enabled lengths.
+
+  2) If SVE is enabled (`sve=on`), but no `sve<N>` CPU properties are
+     provided, then all supported vector lengths are enabled.
+
+  3) If SVE is enabled, then an error is generated when attempting to
+     disable the last enabled vector length (see constraint (1) of "SVE
+     CPU Property Dependencies and Constraints").
+
+  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+     CPU properties are set `on`, then the specified vector lengths are
+     disabled but the default for any unspecified lengths remains enabled.
+     Disabling a power-of-two vector length also disables all vector
+     lengths larger than the power-of-two length (see constraint (2) of
+     "SVE CPU Property Dependencies and Constraints").
+
+  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+     are enabled and all unspecified lengths default to disabled, except
+     for the required lengths per constraint (2) of "SVE CPU Property
+     Dependencies and Constraints", which will even be auto-enabled if
+     they were not explicitly enabled.
+
+  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+     explicitly disabled (i.e. avoiding the error specified in (3) of
+     "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
+     provided an error will be generated.  To avoid this error, one must
+     enable at least one vector length prior to enabling SVE.
+
+SVE CPU Property Examples
+-------------------------
+
+  1) Disable SVE::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off
+
+  2) Implicitly enable all vector lengths for the `max` CPU type::
+
+     $ qemu-system-aarch64 -M virt -cpu max
+
+  3) Only enable the 128-bit vector length::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on
+
+  4) Disable the 256-bit vector length and all larger vector lengths
+     since 256 is a power-of-two (this results in only the 128-bit length
+     being enabled)::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve256=off
+
+  5) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
+
+  6) The same as (5), but since the 128-bit and 256-bit vector
+     lengths are required for the 512-bit vector length to be enabled,
+     then allow them to be auto-enabled::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve512=on
+
+  7) Do the same as (6), but by first disabling SVE and then re-enabling it::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on
+
+  8) Force errors regarding the last vector length::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=off
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
+
+SVE CPU Property Recommendations
+--------------------------------
 
-The example above disables the PMU for the `max` CPU type.
+The examples in "SVE CPU Property Examples" exhibit many ways to select
+vector lengths which developers may find useful in order to avoid overly
+verbose command lines.  However, the recommended way to select vector
+lengths is to explicitly enable each desired length.  Therefore only
+example's (1), (3), and (5) exhibit recommended uses of the properties.
 
target/arm/cpu.c +19 -0
index 73be2ebfdd39..522fed95b339 100644
@@ -1199,6 +1199,19 @@ static void arm_cpu_finalizefn(Object *obj)
 #endif
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp)
+{
+    Error *local_err = NULL;
+
+    if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
+        arm_cpu_sve_finalize(cpu, &local_err);
+        if (local_err != NULL) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+}
+
 static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
 {
     CPUState *cs = CPU(dev);
@@ -1255,6 +1268,12 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         return;
     }
 
+    arm_cpu_finalize_features(cpu, &local_err);
+    if (local_err != NULL) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
     if (arm_feature(env, ARM_FEATURE_AARCH64) &&
         cpu->has_vfp != cpu->has_neon) {
         /*
target/arm/cpu.h +19 -0
index 297ad5e47ad8..11162484465a 100644
@@ -184,8 +184,13 @@ typedef struct {
 
 #ifdef TARGET_AARCH64
 # define ARM_MAX_VQ    16
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp);
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq);
 #else
 # define ARM_MAX_VQ    1
+static inline void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) { }
+static inline uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{ return 0; }
 #endif
 
 typedef struct ARMVectorReg {
@@ -915,6 +920,18 @@ struct ARMCPU {
 
     /* Used to set the maximum vector length the cpu will support.  */
     uint32_t sve_max_vq;
+
+    /*
+     * In sve_vq_map each set bit is a supported vector length of
+     * (bit-number + 1) * 16 bytes, i.e. each bit number + 1 is the vector
+     * length in quadwords.
+     *
+     * While processing properties during initialization, corresponding
+     * sve_vq_init bits are set for bits in sve_vq_map that have been
+     * set by properties.
+     */
+    DECLARE_BITMAP(sve_vq_map, ARM_MAX_VQ);
+    DECLARE_BITMAP(sve_vq_init, ARM_MAX_VQ);
 };
 
 void arm_cpu_post_init(Object *obj);
@@ -1834,6 +1851,8 @@ static inline int arm_feature(CPUARMState *env, int feature)
     return (env->features & (1ULL << feature)) != 0;
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp);
+
 #if !defined(CONFIG_USER_ONLY)
 /* Return true if exception levels below EL3 are in secure state,
  * or would be following an exception return to that level.
target/arm/cpu64.c +188 -2

222 lines not on this page · raw message

target/arm/helper.c +10 -3
index 507026c9154b..f33284c247d5 100644
@@ -5351,6 +5351,13 @@ int sve_exception_el(CPUARMState *env, int el)
     return 0;
 }
 
+static uint32_t sve_zcr_get_valid_len(ARMCPU *cpu, uint32_t start_len)
+{
+    uint32_t start_vq = (start_len & 0xf) + 1;
+
+    return arm_cpu_vq_map_next_smaller(cpu, start_vq + 1) - 1;
+}
+
 /*
  * Given that SVE is enabled, return the vector length for EL.
  */
@@ -5360,13 +5367,13 @@ uint32_t sve_zcr_len_for_el(CPUARMState *env, int el)
     uint32_t zcr_len = cpu->sve_max_vq - 1;
 
     if (el <= 1) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[1]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[1]);
     }
     if (el <= 2 && arm_feature(env, ARM_FEATURE_EL2)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[2]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[2]);
     }
     if (arm_feature(env, ARM_FEATURE_EL3)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[3]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[3]);
     }
     return zcr_len;
 }
target/arm/monitor.c +12 -0
index 4fddb6c252a3..e912ed2cefa0 100644
@@ -90,6 +90,8 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
     return head;
 }
 
+QEMU_BUILD_BUG_ON(ARM_MAX_VQ > 16);
+
 /*
  * These are cpu model features we want to advertise. The order here
  * matters as this is the order in which qmp_query_cpu_model_expansion
@@ -98,6 +100,9 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
  */
 static const char *cpu_model_advertised_features[] = {
     "aarch64", "pmu", "sve",
+    "sve128", "sve256", "sve384", "sve512",
+    "sve640", "sve768", "sve896", "sve1024", "sve1152", "sve1280",
+    "sve1408", "sve1536", "sve1664", "sve1792", "sve1920", "sve2048",
     NULL
 };
 
@@ -185,6 +190,9 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
         if (!err) {
             visit_check_struct(visitor, &err);
         }
+        if (!err) {
+            arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        }
         visit_end_struct(visitor, NULL);
         visit_free(visitor);
         if (err) {
@@ -192,6 +200,10 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
             error_propagate(errp, err);
             return NULL;
         }
+    } else {
+        Error *err = NULL;
+        arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        assert(err == NULL);
     }
 
     expansion_info = g_new0(CpuModelExpansionInfo, 1);
tests/arm-cpu-features.c +189 -0

216 lines not on this page · raw message

Signature · 1 line
2.20.1
#7 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 204

Quoted 9 lines · Andrew Jones
docs/arm-cpu-features.rst +9 -0
+SVE CPU Property Parsing Semantics
+----------------------------------
+
+  1) If SVE is disabled (`sve=off`), then which SVE vector lengths
+     are enabled or disabled is irrelevant to the guest, as the entire
+     SVE feature is disabled and that disables all vector lengths for
+     the guest.  However QEMU will still track any `sve<N>` CPU
+     properties provided by the user.  If later an `sve=on` is provided,
+     then the guest will get only the enabled lengths.

I just spoke with Andrea, who is doing the libvirt side of this, about this behavior. He suggests that we change it such that an error is generated if sve=off and there are explicitly enabled vector lengths, e.g. '-cpu max,sve=off,sve256=on' should generate an error. It would still be possible to have sve<N>=on and sve=off on the same command line, as long as the sve<N>=on is followed by sve<N>=off, like this -cpu max,sve=off,sve256=on,sve256=off,sve128=off, but I don't think we need to worry about that sort of thing. I'll add an error for the non-empty vector length map with SVE off case for v5.

Thanks,
drew

#8 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

In reply to Andrew Jones

Hi Drew,

Quoted 40 lines · Andrew Jones

Introduce cpu properties to give fine control over SVE vector lengths. We introduce a property for each valid length up to the current maximum supported, which is 2048-bits. The properties are named, e.g. sve128, sve256, sve384, sve512, ..., where the number is the number of bits. See the updates to docs/arm-cpu-features.rst for a description of the semantics and for example uses.

Note, as sve-max-vq is still present and we'd like to be able to support qmp_query_cpu_model_expansion with guests launched with e.g. -cpu max,sve-max-vq=8 on their command lines, then we do allow sve-max-vq and sve<N> properties to be provided at the same time, but this is not recommended, and is why sve-max-vq is not mentioned in the document. If sve-max-vq is provided then it enables all lengths smaller than and including the max and disables all lengths larger. It also has the side-effect that no larger lengths may be enabled and that the max itself cannot be disabled. Smaller non-power-of-two lengths may, however, be disabled, e.g. -cpu max,sve-max-vq=4,sve384=off provides a guest the vector lengths 128, 256, and 512 bits.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst | 157 +++++++++++++++++++++++++++++--
 target/arm/cpu.c          |  19 ++++
 target/arm/cpu.h          |  19 ++++
 target/arm/cpu64.c        | 190 +++++++++++++++++++++++++++++++++++++-
 target/arm/helper.c       |  13 ++-
 target/arm/monitor.c      |  12 +++
 tests/arm-cpu-features.c  | 189 +++++++++++++++++++++++++++++++++++++
 7 files changed, 587 insertions(+), 12 deletions(-)
docs/arm-cpu-features.rst +0 -0
index c79dcffb5556..1262fddc6201 100644

docs/arm-cpu-features.rst block in the script for usage) is used to issue the QMP commands. · line 55

Quoted 9 lines
docs/arm-cpu-features.rst +5 -1
@@ -48,18 +48,28 @@ block in the script for usage) is used to issue the QMP commands.
       (QEMU) query-cpu-model-expansion type=full model={"name":"max"}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": true, "aarch64": true
+        "sve1664": true, "pmu": true, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true

I don't really understand why this gets printed unsorted as you registered them in the logical order, any clue?

Note: This example is given with TCG or KVM with all those vectors supported by the host.

Quoted 48 lines
docs/arm-cpu-features.rst +25 -4
       }}}}
 
-We see that the `max` CPU type has the `pmu` and `aarch64` CPU features.
-We also see that the CPU features are enabled, as they are all `true`.
+We see that the `max` CPU type has the `pmu`, `aarch64`, `sve`, and many
+`sve<N>` CPU features.  We also see that all the CPU features are
+enabled, as they are all `true`.  (The `sve<N>` CPU features are all
+optional SVE vector lengths.  See "SVE CPU Properties".)
 
 (2) Let's try to disable the PMU::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"pmu":false}}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": false, "aarch64": true
+        "sve1664": true, "pmu": false, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true
       }}}}
 
 We see it worked, as `pmu` is now `false`.
@@ -75,7 +85,22 @@ We see it worked, as `pmu` is now `false`.
 It looks like this feature is limited to a configuration we do not
 currently have.
 
-(4) Let's try probing CPU features for the Cortex-A15 CPU type::
+(4) Let's disable `sve` and see what happens to all the optional SVE
+    vector lengths::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"sve":false}}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "sve1664": false, "pmu": true, "sve1792": false, "sve1920": false,
+        "sve128": false, "aarch64": true, "sve1024": false, "sve": false,
+        "sve640": false, "sve768": false, "sve1408": false, "sve256": false,
+        "sve1152": false, "sve512": false, "sve384": false, "sve1536": false,
+        "sve896": false, "sve1280": false, "sve2048": false
+      }}}}
+
+As expected they are now all `false`.
+
+(5) Let's try probing CPU features for the Cortex-A15 CPU type::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"cortex-a15"}
       {"return": {"model": {"name": "cortex-a15", "props": {"pmu": true}}}}

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 164

Quoted 10 lines
docs/arm-cpu-features.rst +6 -1
@@ -131,7 +156,125 @@ After determining which CPU features are available and supported for a
 given CPU type, then they may be selectively enabled or disabled on the
 QEMU command line with that CPU type::
 
-  $ qemu-system-aarch64 -M virt -cpu max,pmu=off
+  $ qemu-system-aarch64 -M virt -cpu max,pmu=off,sve=on,sve128=on,sve256=on
+
+The example above disables the PMU and enables the first two SVE vector
+lengths for the `max` CPU type.  Note, the `sve=on` isn't actually
+necessary, because, as we observed above with our probe of the `max` CPU
+type, `sve` is already on by default.

This holds with TCG and not KVM.

Also, based on our probe of

Quoted 19 lines
docs/arm-cpu-features.rst +19 -0
+defaults, it would seem we need to disable many SVE vector lengths, rather
+than only enabling the two we want.  This isn't the case, because, as
+disabling many SVE vector lengths would be quite verbose, the `sve<N>` CPU
+properties have special semantics (see "SVE CPU Property Parsing
+Semantics").
+
+SVE CPU Properties
+==================
+
+There are two types of SVE CPU properties: `sve` and `sve<N>`.  The first
+is used to enable or disable the entire SVE feature, just as the `pmu`
+CPU property completely enables or disables the PMU.  The second type
+is used to enable or disable specific vector lengths, where `N` is the
+number of bits of the length.  The `sve<N>` CPU properties have special
+dependencies and constraints, see "SVE CPU Property Dependencies and
+Constraints" below.  Additionally, as we want all supported vector lengths
+to be enabled by default, then, in order to avoid overly verbose command
+lines (command lines full of `sve<N>=off`, for all `N` not wanted), we
+provide the parsing semantics listed in "SVE CPU Property Parsing

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 194

Quoted 10 lines
docs/arm-cpu-features.rst +10 -0
+Semantics".
+
+SVE CPU Property Dependencies and Constraints
+---------------------------------------------
+
+  1) At least one vector length must be enabled when `sve` is enabled.> +
+  2) If a vector length `N` is enabled, then all power-of-two vector
+     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
+     is enabled, then `sve128` and `sve256` must also be enabled,
+     but `sve384` is not required.

I would remove the eg. part. reading that I tend to understand that the user must pass ,sve128=on, sve256=on and sve512=on whereas example 6 says only sve512=on can be set and other lower ^2 values are auto-enabled.

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 207

Quoted 13 lines
docs/arm-cpu-features.rst +13 -0
+
+SVE CPU Property Parsing Semantics
+----------------------------------
+
+  1) If SVE is disabled (`sve=off`), then which SVE vector lengths
+     are enabled or disabled is irrelevant to the guest, as the entire
+     SVE feature is disabled and that disables all vector lengths for
+     the guest.  However QEMU will still track any `sve<N>` CPU
+     properties provided by the user.  If later an `sve=on` is provided,
+     then the guest will get only the enabled lengths.
+
+  2) If SVE is enabled (`sve=on`), but no `sve<N>` CPU properties are
+     provided, then all supported vector lengths are enabled.

I understand from the code it also includes non ^2 values which is not obvious.

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 211

Quoted 4 lines
docs/arm-cpu-features.rst +4 -0
+
+  3) If SVE is enabled, then an error is generated when attempting to
+     disable the last enabled vector length (see constraint (1) of "SVE
+     CPU Property Dependencies and Constraints").

the same happens if you attempt to disabled any lower ^2 value

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 218

Quoted 7 lines
docs/arm-cpu-features.rst +7 -0
+
+  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+     CPU properties are set `on`, then the specified vector lengths are
+     disabled but the default for any unspecified lengths remains enabled.
+     Disabling a power-of-two vector length also disables all vector
+     lengths larger than the power-of-two length (see constraint (2) of
+     "SVE CPU Property Dependencies and Constraints").

So rephasing it:
disabling a non ^2 value only disable that one disabling a ^2 value disables all larger lengths too is that correct?

Quoted 32 lines
docs/arm-cpu-features.rst +32 -0
+
+  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+     are enabled and all unspecified lengths default to disabled, except
+     for the required lengths per constraint (2) of "SVE CPU Property
+     Dependencies and Constraints", which will even be auto-enabled if
+     they were not explicitly enabled.
+
+  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+     explicitly disabled (i.e. avoiding the error specified in (3) of
+     "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
+     provided an error will be generated.  To avoid this error, one must
+     enable at least one vector length prior to enabling SVE.
+
+SVE CPU Property Examples
+-------------------------
+
+  1) Disable SVE::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off
+
+  2) Implicitly enable all vector lengths for the `max` CPU type::
+
+     $ qemu-system-aarch64 -M virt -cpu max
+
+  3) Only enable the 128-bit vector length::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on
+
+  4) Disable the 256-bit vector length and all larger vector lengths
+     since 256 is a power-of-two (this results in only the 128-bit length
+     being enabled)::
+

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 261

Quoted 10 lines
docs/arm-cpu-features.rst +10 -0
+     $ qemu-system-aarch64 -M virt -cpu max,sve256=off
+
+  5) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
+
+  6) The same as (5), but since the 128-bit and 256-bit vector
+     lengths are required for the 512-bit vector length to be enabled,
+     then allow them to be auto-enabled::> +
+     $ qemu-system-aarch64 -M virt -cpu max,sve512=on

You should also document

$ qemu-system-aarch64 -M virt -cpu max,sve512=off
as in that case this result in 128, 256 and 384. the fact you get non ^2
values is not obvious here because if you enable a ^2 value you onlt get
<= ^2 values.

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 267

Quoted 5 lines
docs/arm-cpu-features.rst +5 -0
+
+  7) Do the same as (6), but by first disabling SVE and then re-enabling it::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on> +
+  8) Force errors regarding the last vector length::

You mean those commands will generate errors, right?

Quoted 19 lines
docs/arm-cpu-features.rst +11 -1
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=off
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
+
+SVE CPU Property Recommendations
+--------------------------------
 
-The example above disables the PMU for the `max` CPU type.
+The examples in "SVE CPU Property Examples" exhibit many ways to select
+vector lengths which developers may find useful in order to avoid overly
+verbose command lines.  However, the recommended way to select vector
+lengths is to explicitly enable each desired length.  Therefore only
+example's (1), (3), and (5) exhibit recommended uses of the properties.
 
target/arm/cpu.c +0 -0
index 73be2ebfdd39..522fed95b339 100644

target/arm/cpu.c static void arm_cpu_finalizefn(Object *obj) · line 1208

Quoted 10 lines
target/arm/cpu.c +7 -0
@@ -1199,6 +1199,19 @@ static void arm_cpu_finalizefn(Object *obj)
 #endif
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp)
+{
+    Error *local_err = NULL;
+
+    if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
+        arm_cpu_sve_finalize(cpu, &local_err);
+        if (local_err != NULL) {

nit: !local_err

target/arm/cpu.c static void arm_cpu_finalizefn(Object *obj) · line 1210

Quoted 2 lines
target/arm/cpu.c +2 -0
+            error_propagate(errp, local_err);
+            return;

not needed

target/arm/cpu.c static void arm_cpu_realizefn(DeviceState *dev, Error **errp) · line 1272

Quoted 13 lines
target/arm/cpu.c +6 -0
+        }
+    }
+}
+
 static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
 {
     CPUState *cs = CPU(dev);
@@ -1255,6 +1268,12 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         return;
     }
 
+    arm_cpu_finalize_features(cpu, &local_err);
+    if (local_err != NULL) {

same

Quoted 84 lines
target/arm/cpu.c +3 -0
+        error_propagate(errp, local_err);
+        return;> +    }
+
     if (arm_feature(env, ARM_FEATURE_AARCH64) &&
         cpu->has_vfp != cpu->has_neon) {
         /*
target/arm/cpu.h +19 -0
index 297ad5e47ad8..11162484465a 100644
@@ -184,8 +184,13 @@ typedef struct {
 
 #ifdef TARGET_AARCH64
 # define ARM_MAX_VQ    16
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp);
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq);
 #else
 # define ARM_MAX_VQ    1
+static inline void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) { }
+static inline uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{ return 0; }
 #endif
 
 typedef struct ARMVectorReg {
@@ -915,6 +920,18 @@ struct ARMCPU {
 
     /* Used to set the maximum vector length the cpu will support.  */
     uint32_t sve_max_vq;
+
+    /*
+     * In sve_vq_map each set bit is a supported vector length of
+     * (bit-number + 1) * 16 bytes, i.e. each bit number + 1 is the vector
+     * length in quadwords.
+     *
+     * While processing properties during initialization, corresponding
+     * sve_vq_init bits are set for bits in sve_vq_map that have been
+     * set by properties.
+     */
+    DECLARE_BITMAP(sve_vq_map, ARM_MAX_VQ);
+    DECLARE_BITMAP(sve_vq_init, ARM_MAX_VQ);
 };
 
 void arm_cpu_post_init(Object *obj);
@@ -1834,6 +1851,8 @@ static inline int arm_feature(CPUARMState *env, int feature)
     return (env->features & (1ULL << feature)) != 0;
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp);
+
 #if !defined(CONFIG_USER_ONLY)
 /* Return true if exception levels below EL3 are in secure state,
  * or would be following an exception return to that level.
target/arm/cpu64.c +24 -0
index 8cdb0c79fa7a..606e3eceb9c0 100644
@@ -256,11 +256,152 @@ static void aarch64_a72_initfn(Object *obj)
     define_arm_cp_regs(cpu, cortex_a72_a57_a53_cp_reginfo);
 }
 
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
+{
+    /*
+     * If any vector lengths are explicitly enabled with sve<N> properties,
+     * then all other lengths are implicitly disabled.  If sve-max-vq is
+     * specified then it is the same as explicitly enabling all lengths
+     * up to and including the specified maximum, which means all larger
+     * lengths will be implicitly disabled.  If no sve<N> properties
+     * are enabled and sve-max-vq is not specified, then all lengths not
+     * explicitly disabled will be enabled.  Additionally, all power-of-two
+     * vector lengths less than the maximum enabled length will be
+     * automatically enabled and all vector lengths larger than the largest
+     * disabled power-of-two vector length will be automatically disabled.
+     * Errors are generated if the user provided input that interferes with
+     * any of the above.  Finally, if SVE is not disabled, then at least one
+     * vector length must be enabled.
+     */
+    DECLARE_BITMAP(tmp, ARM_MAX_VQ);
+    uint32_t vq, max_vq = 0;
+
+    /*
+     * Process explicit sve<N> properties.
+     * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
+     * Check first for any sve<N> enabled.

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 292

Quoted 10 lines
target/arm/cpu64.c +10 -0
+     */
+    if (!bitmap_empty(cpu->sve_vq_map, ARM_MAX_VQ)) {
+        max_vq = find_last_bit(cpu->sve_vq_map, ARM_MAX_VQ) + 1;
+
+        if (cpu->sve_max_vq && max_vq > cpu->sve_max_vq) {
+            error_setg(errp, "cannot enable sve%d", max_vq * 128);
+            error_append_hint(errp, "sve%d is larger than the maximum vector "
+                              "length, sve-max-vq=%d (%d bits)\n",
+                              max_vq * 128, cpu->sve_max_vq,
+                              cpu->sve_max_vq * 128);

you could test directly if max_vq < cpu->sve_max_vq too which couldn't be correct either. But as far as I understand this is caught in *

Quoted 65 lines
target/arm/cpu64.c +65 -0
+            return;
+        }
+
+        /* Propagate enabled bits down through required powers-of-two. */
+        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                set_bit(vq - 1, cpu->sve_vq_map);
+            }
+        }
+    } else if (cpu->sve_max_vq == 0) {
+        /*
+         * No explicit bits enabled, and no implicit bits from sve-max-vq.
+         */
+        if (!cpu_isar_feature(aa64_sve, cpu)) {
+            /* SVE is disabled and so are all vector lengths.  Good. */
+            return;
+        }
+
+        /* Disabling a power-of-two disables all larger lengths. */
+        if (test_bit(0, cpu->sve_vq_init)) {> +            error_setg(errp, "cannot disable sve128");
+            error_append_hint(errp, "Disabling sve128 results in all vector "
+                              "lengths being disabled.\n");
+            error_append_hint(errp, "With SVE enabled, at least one vector "
+                              "length must be enabled.\n");
+            return;
+        }
+        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
+            if (test_bit(vq - 1, cpu->sve_vq_init)) {
+                break;
+            }
+        }
+        max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+
+        bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
+        max_vq = find_last_bit(cpu->sve_vq_map, max_vq) + 1;
+    }
+
+    /*
+     * Process the sve-max-vq property.
+     * Note that we know from the above that no bit above
+     * sve-max-vq is currently set.
+     */
+    if (cpu->sve_max_vq != 0) {
+        max_vq = cpu->sve_max_vq;
+
+        if (!test_bit(max_vq - 1, cpu->sve_vq_map) &&
+            test_bit(max_vq - 1, cpu->sve_vq_init)) {
+            error_setg(errp, "cannot disable sve%d", max_vq * 128);
+            error_append_hint(errp, "The maximum vector length must be "
+                              "enabled, sve-max-vq=%d (%d bits)\n",
+                              max_vq, max_vq * 128);
+            return;
+        }
+
+        /* Set all bits not explicitly set within sve-max-vq. */
+        bitmap_complement(tmp, cpu->sve_vq_init, max_vq);
+        bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+    }
+
+    /*
+     * We should know what max-vq is now.  Also, as we're done
+     * manipulating sve-vq-map, we ensure any bits above max-vq
+     * are clear, just in case anybody looks.
+     */
+    assert(max_vq != 0);

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 368

Quoted 10 lines
target/arm/cpu64.c +10 -0
+    bitmap_clear(cpu->sve_vq_map, max_vq, ARM_MAX_VQ - max_vq);
+
+    /* Ensure all required powers-of-two are enabled. */
+    for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+        if (!test_bit(vq - 1, cpu->sve_vq_map)) {
+            error_setg(errp, "cannot disable sve%d", vq * 128);
+            error_append_hint(errp, "sve%d is required as it "
+                              "is a power-of-two length smaller than "
+                              "the maximum, sve%d\n",
+                              vq * 128, max_vq * 128);

* I think this will also catches the case where ,sve512=on,sve-max-vq=5 where the end-user did not explicitly disabled anything

Quoted 19 lines
target/arm/cpu64.c +19 -0
+            return;
+        }
+    }
+
+    /* From now on sve_max_vq is the actual maximum supported length. */
+    cpu->sve_max_vq = max_vq;
+}
+
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{
+    uint32_t bitnum;
+
+    /*
+     * We allow vq == ARM_MAX_VQ + 1 to be input because the caller may want
+     * to find the maximum vq enabled, which may be ARM_MAX_VQ, but this
+     * function always returns the next smaller than the input.
+     */
+    assert(vq && vq <= ARM_MAX_VQ + 1);
+

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 396

Quoted 10 lines
target/arm/cpu64.c +5 -1
+    bitnum = find_last_bit(cpu->sve_vq_map, vq - 1);
+    return bitnum == vq - 1 ? 0 : bitnum + 1;
+}
+
 static void cpu_max_get_sve_max_vq(Object *obj, Visitor *v, const char *name,
                                    void *opaque, Error **errp)
 {
     ARMCPU *cpu = ARM_CPU(obj);
-    visit_type_uint32(v, name, &cpu->sve_max_vq, errp);
+    uint32_t value;

Shouldn't it be part of the previous patch?

target/arm/cpu64.c static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,

Quoted 12 lines
target/arm/cpu64.c +8 -0
+
+    /* All vector lengths are disabled when SVE is off. */
+    if (!cpu_isar_feature(aa64_sve, cpu)) {
+        value = 0;
+    } else {
+        value = cpu->sve_max_vq;
+    }
+    visit_type_uint32(v, name, &value, errp);
 }
 
 static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,
@@ -279,6 +420,44 @@ static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,

Independently on this patch I noticed that if sve_max_vq is out of scope, an error is propagated but still the cpu->sve-max_vq is set.

Quoted 306 lines
target/arm/cpu64.c +46 -1
     error_propagate(errp, err);
 }
 
+static void cpu_arm_get_sve_vq(Object *obj, Visitor *v, const char *name,
+                               void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq = atoi(&name[3]) / 128;
+    bool value;
+
+    /* All vector lengths are disabled when SVE is off. */
+    if (!cpu_isar_feature(aa64_sve, cpu)) {
+        value = false;
+    } else {
+        value = test_bit(vq - 1, cpu->sve_vq_map);
+    }
+    visit_type_bool(v, name, &value, errp);
+}
+
+static void cpu_arm_set_sve_vq(Object *obj, Visitor *v, const char *name,
+                               void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq = atoi(&name[3]) / 128;
+    Error *err = NULL;
+    bool value;
+
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
+
+    if (value) {
+        set_bit(vq - 1, cpu->sve_vq_map);
+    } else {
+        clear_bit(vq - 1, cpu->sve_vq_map);
+    }
+    set_bit(vq - 1, cpu->sve_vq_init);
+}
+
 static void cpu_arm_get_sve(Object *obj, Visitor *v, const char *name,
                             void *opaque, Error **errp)
 {
@@ -315,6 +494,7 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
 static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
@@ -418,11 +598,17 @@ static void aarch64_max_initfn(Object *obj)
         cpu->dcz_blocksize = 7; /*  512 bytes */
 #endif
 
-        cpu->sve_max_vq = ARM_MAX_VQ;
         object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                             cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
         object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
                             cpu_arm_set_sve, NULL, NULL, &error_fatal);
+
+        for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+            char name[8];
+            sprintf(name, "sve%d", vq * 128);
+            object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                                cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+        }
     }
 }
 
target/arm/helper.c +10 -3
index 507026c9154b..f33284c247d5 100644
@@ -5351,6 +5351,13 @@ int sve_exception_el(CPUARMState *env, int el)
     return 0;
 }
 
+static uint32_t sve_zcr_get_valid_len(ARMCPU *cpu, uint32_t start_len)
+{
+    uint32_t start_vq = (start_len & 0xf) + 1;
+
+    return arm_cpu_vq_map_next_smaller(cpu, start_vq + 1) - 1;
+}
+
 /*
  * Given that SVE is enabled, return the vector length for EL.
  */
@@ -5360,13 +5367,13 @@ uint32_t sve_zcr_len_for_el(CPUARMState *env, int el)
     uint32_t zcr_len = cpu->sve_max_vq - 1;
 
     if (el <= 1) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[1]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[1]);
     }
     if (el <= 2 && arm_feature(env, ARM_FEATURE_EL2)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[2]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[2]);
     }
     if (arm_feature(env, ARM_FEATURE_EL3)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[3]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[3]);
     }
     return zcr_len;
 }
target/arm/monitor.c +12 -0
index 4fddb6c252a3..e912ed2cefa0 100644
@@ -90,6 +90,8 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
     return head;
 }
 
+QEMU_BUILD_BUG_ON(ARM_MAX_VQ > 16);
+
 /*
  * These are cpu model features we want to advertise. The order here
  * matters as this is the order in which qmp_query_cpu_model_expansion
@@ -98,6 +100,9 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
  */
 static const char *cpu_model_advertised_features[] = {
     "aarch64", "pmu", "sve",
+    "sve128", "sve256", "sve384", "sve512",
+    "sve640", "sve768", "sve896", "sve1024", "sve1152", "sve1280",
+    "sve1408", "sve1536", "sve1664", "sve1792", "sve1920", "sve2048",
     NULL
 };
 
@@ -185,6 +190,9 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
         if (!err) {
             visit_check_struct(visitor, &err);
         }
+        if (!err) {
+            arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        }
         visit_end_struct(visitor, NULL);
         visit_free(visitor);
         if (err) {
@@ -192,6 +200,10 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
             error_propagate(errp, err);
             return NULL;
         }
+    } else {
+        Error *err = NULL;
+        arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        assert(err == NULL);
     }
 
     expansion_info = g_new0(CpuModelExpansionInfo, 1);
tests/arm-cpu-features.c +141 -0
index 202bc0e3e823..9a2dd402b769 100644
@@ -13,6 +13,18 @@
 #include "qapi/qmp/qdict.h"
 #include "qapi/qmp/qjson.h"
 
+#if __SIZEOF_LONG__ == 8
+#define BIT(n) (1UL << (n))
+#else
+#define BIT(n) (1ULL << (n))
+#endif
+
+/*
+ * We expect the SVE max-vq to be 16. Also it must be <= 64
+ * for our test code, otherwise 'vls' can't just be a uint64_t.
+ */
+#define SVE_MAX_VQ 16
+
 #define MACHINE    "-machine virt,gic-version=max "
 #define QUERY_HEAD "{ 'execute': 'query-cpu-model-expansion', " \
                      "'arguments': { 'type': 'full', "
@@ -157,6 +169,173 @@ static void assert_bad_props(QTestState *qts, const char *cpu_type)
     qobject_unref(resp);
 }
 
+static uint64_t resp_get_sve_vls(QDict *resp)
+{
+    QDict *props;
+    const QDictEntry *e;
+    uint64_t vls = 0;
+    int n = 0;
+
+    g_assert(resp);
+    g_assert(resp_has_props(resp));
+
+    props = resp_get_props(resp);
+
+    for (e = qdict_first(props); e; e = qdict_next(props, e)) {
+        if (strlen(e->key) > 3 && !strncmp(e->key, "sve", 3) &&
+            g_ascii_isdigit(e->key[3])) {
+            char *endptr;
+            int bits;
+
+            bits = g_ascii_strtoll(&e->key[3], &endptr, 10);
+            if (!bits || *endptr != '\0') {
+                continue;
+            }
+
+            if (qdict_get_bool(props, e->key)) {
+                vls |= BIT((bits / 128) - 1);
+            }
+            ++n;
+        }
+    }
+
+    g_assert(n == SVE_MAX_VQ);
+
+    return vls;
+}
+
+#define assert_sve_vls(qts, cpu_type, expected_vls, fmt, ...)          \
+({                                                                     \
+    QDict *_resp = do_query(qts, cpu_type, fmt, ##__VA_ARGS__);        \
+    g_assert(_resp);                                                   \
+    g_assert(resp_has_props(_resp));                                   \
+    g_assert(resp_get_sve_vls(_resp) == expected_vls);                 \
+    qobject_unref(_resp);                                              \
+})
+
+static void sve_tests_default(QTestState *qts, const char *cpu_type)
+{
+    /*
+     * With no sve-max-vq or sve<N> properties on the command line
+     * the default is to have all vector lengths enabled. This also
+     * tests that 'sve' is 'on' by default.
+     */
+    assert_sve_vls(qts, cpu_type, BIT(SVE_MAX_VQ) - 1, NULL);
+
+    /* With SVE off, all vector lengths should also be off. */
+    assert_sve_vls(qts, cpu_type, 0, "{ 'sve': false }");
+
+    /* With SVE on, we must have at least one vector length enabled. */
+    assert_error(qts, cpu_type, "cannot disable sve128", "{ 'sve128': false }");
+
+    /*
+     * ---------------------------------------------------------------------
+     *               power-of-two(vq)   all-power-            can      can
+     *                                  of-two(< vq)        enable   disable
+     * ---------------------------------------------------------------------
+     * vq < max_vq      no                MUST*              yes      yes
+     * vq < max_vq      yes               MUST*              yes      no
+     * ---------------------------------------------------------------------
+     * vq == max_vq     n/a               MUST*              yes**    yes**
+     * ---------------------------------------------------------------------
+     * vq > max_vq      n/a               no                 no       yes
+     * vq > max_vq      n/a               yes                yes      yes
+     * ---------------------------------------------------------------------
+     *
+     * [*] "MUST" means this requirement must already be satisfied,
+     *     otherwise 'max_vq' couldn't itself be enabled.
+     *
+     * [**] Not testable with the QMP interface, only with the command line.
+     */
+
+    /* max_vq := 8 */
+    assert_sve_vls(qts, cpu_type, 0x8b, "{ 'sve1024': true }");
+
+    /* max_vq := 8, vq < max_vq, !power-of-two(vq) */
+    assert_sve_vls(qts, cpu_type, 0x8f,
+                   "{ 'sve1024': true, 'sve384': true }");
+    assert_sve_vls(qts, cpu_type, 0x8b,
+                   "{ 'sve1024': true, 'sve384': false }");
+
+    /* max_vq := 8, vq < max_vq, power-of-two(vq) */
+    assert_sve_vls(qts, cpu_type, 0x8b,
+                   "{ 'sve1024': true, 'sve256': true }");
+    assert_error(qts, cpu_type, "cannot disable sve256",
+                 "{ 'sve1024': true, 'sve256': false }");
+
+    /* max_vq := 3, vq > max_vq, !all-power-of-two(< vq) */
+    assert_error(qts, cpu_type, "cannot disable sve512",
+                 "{ 'sve384': true, 'sve512': false, 'sve640': true }");
+
+    /*
+     * We can disable power-of-two vector lengths when all larger lengths
+     * are also disabled. We only need to disable the power-of-two length,
+     * as all non-enabled larger lengths will then be auto-disabled.
+     */
+    assert_sve_vls(qts, cpu_type, 0x7, "{ 'sve512': false }");
+
+    /* max_vq := 3, vq > max_vq, all-power-of-two(< vq) */
+    assert_sve_vls(qts, cpu_type, 0x1f,
+                   "{ 'sve384': true, 'sve512': true, 'sve640': true }");
+    assert_sve_vls(qts, cpu_type, 0xf,
+                   "{ 'sve384': true, 'sve512': true, 'sve640': false }");
+}
+
+static void sve_tests_sve_max_vq_8(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-cpu max,sve-max-vq=8");
+
+    assert_sve_vls(qts, "max", BIT(8) - 1, NULL);
+
+    /*
+     * Disabling the max-vq set by sve-max-vq is not allowed, but
+     * of course enabling it is OK.
+     */
+    assert_error(qts, "max", "cannot disable sve1024", "{ 'sve1024': false }");
+    assert_sve_vls(qts, "max", 0xff, "{ 'sve1024': true }");
+
+    /*
+     * Enabling anything larger than max-vq set by sve-max-vq is not

tests/arm-cpu-features.c static void assert_bad_props(QTestState *qts, const char *cpu_type) · line 310

Quoted 10 lines
tests/arm-cpu-features.c +10 -0
+     * allowed, but of course disabling everything larger is OK.
+     */
+    assert_error(qts, "max", "cannot enable sve1152", "{ 'sve1152': true }");
+    assert_sve_vls(qts, "max", 0xff, "{ 'sve1152': false }");
+
+    /*
+     * We can disable non power-of-two lengths smaller than the max-vq
+     * set by sve-max-vq, but not power-of-two lengths.
+     */
+    assert_sve_vls(qts, "max", 0xfb, "{ 'sve384': false }");

you can also test ,sve384=on => 0x7

Quoted 49 lines
tests/arm-cpu-features.c +31 -0
+    assert_error(qts, "max", "cannot disable sve256", "{ 'sve256': false }");
+
+    qtest_quit(qts);
+}
+
+static void sve_tests_sve_off(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-cpu max,sve=off");
+
+    /* SVE is off, so the map should be empty. */
+    assert_sve_vls(qts, "max", 0, NULL);
+
+    /* The map stays empty even if we turn lengths on or off. */
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': true }");
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+
+    /* With SVE re-enabled we should get all vector lengths enabled. */
+    assert_sve_vls(qts, "max", BIT(SVE_MAX_VQ) - 1, "{ 'sve': true }");
+
+    /* Or enable SVE with just specific vector lengths. */
+    assert_sve_vls(qts, "max", 0x3,
+                   "{ 'sve': true, 'sve128': true, 'sve256': true }");
+
+    qtest_quit(qts);
+}
+
 static void test_query_cpu_model_expansion(const void *data)
 {
     QTestState *qts;
@@ -180,9 +359,12 @@ static void test_query_cpu_model_expansion(const void *data)
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
         assert_has_feature(qts, "max", "aarch64");
         assert_has_feature(qts, "max", "sve");
+        assert_has_feature(qts, "max", "sve128");
         assert_has_feature(qts, "cortex-a57", "pmu");
         assert_has_feature(qts, "cortex-a57", "aarch64");
 
+        sve_tests_default(qts, "max");
+
         /* Test that features that depend on KVM generate errors without. */
         assert_error(qts, "max",
                      "'aarch64' feature cannot be disabled "
@@ -234,6 +416,13 @@ int main(int argc, char **argv)
     qtest_add_data_func("/arm/query-cpu-model-expansion",
                         NULL, test_query_cpu_model_expansion);
 

tests/arm-cpu-features.c int main(int argc, char **argv) · line 428

Quoted 10 lines
tests/arm-cpu-features.c +7 -0
+    if (g_str_equal(qtest_get_arch(), "aarch64")) {
+        qtest_add_data_func("/arm/max/query-cpu-model-expansion/sve-max-vq-8",
+                            NULL, sve_tests_sve_max_vq_8);
+        qtest_add_data_func("/arm/max/query-cpu-model-expansion/sve-off",
+                            NULL, sve_tests_sve_off);
+    }
+
     if (kvm_available) {
         qtest_add_data_func("/arm/kvm/query-cpu-model-expansion",
                             NULL, test_query_cpu_model_expansion_kvm);

Thanks

Eric

#9 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

Quoted 38 lines · Auger Eric

Hi Drew,

Quoted 35 lines · Andrew Jones

Introduce cpu properties to give fine control over SVE vector lengths. We introduce a property for each valid length up to the current maximum supported, which is 2048-bits. The properties are named, e.g. sve128, sve256, sve384, sve512, ..., where the number is the number of bits. See the updates to docs/arm-cpu-features.rst for a description of the semantics and for example uses.

Note, as sve-max-vq is still present and we'd like to be able to support qmp_query_cpu_model_expansion with guests launched with e.g. -cpu max,sve-max-vq=8 on their command lines, then we do allow sve-max-vq and sve<N> properties to be provided at the same time, but this is not recommended, and is why sve-max-vq is not mentioned in the document. If sve-max-vq is provided then it enables all lengths smaller than and including the max and disables all lengths larger. It also has the side-effect that no larger lengths may be enabled and that the max itself cannot be disabled. Smaller non-power-of-two lengths may, however, be disabled, e.g. -cpu max,sve-max-vq=4,sve384=off provides a guest the vector lengths 128, 256, and 512 bits.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst | 157 +++++++++++++++++++++++++++++--
 target/arm/cpu.c          |  19 ++++
 target/arm/cpu.h          |  19 ++++
 target/arm/cpu64.c        | 190 +++++++++++++++++++++++++++++++++++++-
 target/arm/helper.c       |  13 ++-
 target/arm/monitor.c      |  12 +++
 tests/arm-cpu-features.c  | 189 +++++++++++++++++++++++++++++++++++++
 7 files changed, 587 insertions(+), 12 deletions(-)
Quoted 16 lines
Quoted 5 lines
docs/arm-cpu-features.rst +0 -0
index c79dcffb5556..1262fddc6201 100644

docs/arm-cpu-features.rst block in the script for usage) is used to issue the QMP commands. · line 55

Quoted 9 lines
docs/arm-cpu-features.rst +5 -1
@@ -48,18 +48,28 @@ block in the script for usage) is used to issue the QMP commands.
       (QEMU) query-cpu-model-expansion type=full model={"name":"max"}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": true, "aarch64": true
+        "sve1664": true, "pmu": true, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true

I don't really understand why this gets printed unsorted as you registered them in the logical order, any clue?

Maybe qdict hashing? qmp-shell could probably be patched to sort them, but it's probably not worth it.

Quoted 3 lines

Note: This example is given with TCG or KVM with all those vectors supported by the host.

You want this note to be added? I don't think it's necessary, as it's just an example. A different configuration or different CPU model expansion will have different results, as we explain exhaustively in this document.

Quoted 51 lines
Quoted 51 lines
docs/arm-cpu-features.rst +25 -4
       }}}}
 
-We see that the `max` CPU type has the `pmu` and `aarch64` CPU features.
-We also see that the CPU features are enabled, as they are all `true`.
+We see that the `max` CPU type has the `pmu`, `aarch64`, `sve`, and many
+`sve<N>` CPU features.  We also see that all the CPU features are
+enabled, as they are all `true`.  (The `sve<N>` CPU features are all
+optional SVE vector lengths.  See "SVE CPU Properties".)
 
 (2) Let's try to disable the PMU::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"pmu":false}}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": false, "aarch64": true
+        "sve1664": true, "pmu": false, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true
       }}}}
 
 We see it worked, as `pmu` is now `false`.
@@ -75,7 +85,22 @@ We see it worked, as `pmu` is now `false`.
 It looks like this feature is limited to a configuration we do not
 currently have.
 
-(4) Let's try probing CPU features for the Cortex-A15 CPU type::
+(4) Let's disable `sve` and see what happens to all the optional SVE
+    vector lengths::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"sve":false}}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "sve1664": false, "pmu": true, "sve1792": false, "sve1920": false,
+        "sve128": false, "aarch64": true, "sve1024": false, "sve": false,
+        "sve640": false, "sve768": false, "sve1408": false, "sve256": false,
+        "sve1152": false, "sve512": false, "sve384": false, "sve1536": false,
+        "sve896": false, "sve1280": false, "sve2048": false
+      }}}}
+
+As expected they are now all `false`.
+
+(5) Let's try probing CPU features for the Cortex-A15 CPU type::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"cortex-a15"}
       {"return": {"model": {"name": "cortex-a15", "props": {"pmu": true}}}}
@@ -131,7 +156,125 @@ After determining which CPU features are available and supported for a
 given CPU type, then they may be selectively enabled or disabled on the
 QEMU command line with that CPU type::
 
Quoted 9 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a

Quoted 7 lines
docs/arm-cpu-features.rst +6 -1
-  $ qemu-system-aarch64 -M virt -cpu max,pmu=off
+  $ qemu-system-aarch64 -M virt -cpu max,pmu=off,sve=on,sve128=on,sve256=on
+
+The example above disables the PMU and enables the first two SVE vector
+lengths for the `max` CPU type.  Note, the `sve=on` isn't actually
+necessary, because, as we observed above with our probe of the `max` CPU
+type, `sve` is already on by default.

This holds with TCG and not KVM.

It also holds with KVM, as long as the host supports SVE. But that doesn't matter, because we're speaking specifically about the example above here, where we see that sve=on is the default.

Quoted 26 lines

Also, based on our probe of

Quoted 24 lines
docs/arm-cpu-features.rst +24 -0
+defaults, it would seem we need to disable many SVE vector lengths, rather
+than only enabling the two we want.  This isn't the case, because, as
+disabling many SVE vector lengths would be quite verbose, the `sve<N>` CPU
+properties have special semantics (see "SVE CPU Property Parsing
+Semantics").
+
+SVE CPU Properties
+==================
+
+There are two types of SVE CPU properties: `sve` and `sve<N>`.  The first
+is used to enable or disable the entire SVE feature, just as the `pmu`
+CPU property completely enables or disables the PMU.  The second type
+is used to enable or disable specific vector lengths, where `N` is the
+number of bits of the length.  The `sve<N>` CPU properties have special
+dependencies and constraints, see "SVE CPU Property Dependencies and
+Constraints" below.  Additionally, as we want all supported vector lengths
+to be enabled by default, then, in order to avoid overly verbose command
+lines (command lines full of `sve<N>=off`, for all `N` not wanted), we
+provide the parsing semantics listed in "SVE CPU Property Parsing
+Semantics".
+
+SVE CPU Property Dependencies and Constraints
+---------------------------------------------
+
Quoted 8 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 194

Quoted 5 lines
docs/arm-cpu-features.rst +5 -0
+  1) At least one vector length must be enabled when `sve` is enabled.> +
+  2) If a vector length `N` is enabled, then all power-of-two vector
+     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
+     is enabled, then `sve128` and `sve256` must also be enabled,
+     but `sve384` is not required.

I would remove the eg. part. reading that I tend to understand that the user must pass ,sve128=on, sve256=on and sve512=on whereas example 6 says only sve512=on can be set and other lower ^2 values are auto-enabled.

I'll keep the example, but instead of using the property names, I'll change to using "the 128-bit vector length", 256-bit ..., etc.

Quoted 7 lines
Quoted 7 lines
docs/arm-cpu-features.rst +7 -0
+
+SVE CPU Property Parsing Semantics
+----------------------------------
+
+  1) If SVE is disabled (`sve=off`), then which SVE vector lengths
+     are enabled or disabled is irrelevant to the guest, as the entire
+     SVE feature is disabled and that disables all vector lengths for
Quoted 8 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 207

Quoted 6 lines
docs/arm-cpu-features.rst +6 -0
+     the guest.  However QEMU will still track any `sve<N>` CPU
+     properties provided by the user.  If later an `sve=on` is provided,
+     then the guest will get only the enabled lengths.
+
+  2) If SVE is enabled (`sve=on`), but no `sve<N>` CPU properties are
+     provided, then all supported vector lengths are enabled.

I understand from the code it also includes non ^2 values which is not obvious.

Yes, the default is to have *all* supported vector lengths enabled, not just the power-of-two. As we already use the word 'all' here, then how can we further clarify this?

Quoted 5 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 211

Quoted 4 lines
docs/arm-cpu-features.rst +4 -0
+
+  3) If SVE is enabled, then an error is generated when attempting to
+     disable the last enabled vector length (see constraint (1) of "SVE
+     CPU Property Dependencies and Constraints").

the same happens if you attempt to disabled any lower ^2 value

That should be another bullet, which I appear to be missing, and will add.

Quoted 11 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 218

Quoted 7 lines
docs/arm-cpu-features.rst +7 -0
+
+  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+     CPU properties are set `on`, then the specified vector lengths are
+     disabled but the default for any unspecified lengths remains enabled.
+     Disabling a power-of-two vector length also disables all vector
+     lengths larger than the power-of-two length (see constraint (2) of
+     "SVE CPU Property Dependencies and Constraints").

So rephasing it:
disabling a non ^2 value only disable that one disabling a ^2 value disables all larger lengths too is that correct?

That's correct. We auto-disable all uninitialized larger lengths when a power-of-two dependency length is explicitly disabled.

Quoted 45 lines
Quoted 33 lines
docs/arm-cpu-features.rst +32 -0
+
+  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+     are enabled and all unspecified lengths default to disabled, except
+     for the required lengths per constraint (2) of "SVE CPU Property
+     Dependencies and Constraints", which will even be auto-enabled if
+     they were not explicitly enabled.
+
+  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+     explicitly disabled (i.e. avoiding the error specified in (3) of
+     "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
+     provided an error will be generated.  To avoid this error, one must
+     enable at least one vector length prior to enabling SVE.
+
+SVE CPU Property Examples
+-------------------------
+
+  1) Disable SVE::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off
+
+  2) Implicitly enable all vector lengths for the `max` CPU type::
+
+     $ qemu-system-aarch64 -M virt -cpu max
+
+  3) Only enable the 128-bit vector length::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on
+
+  4) Disable the 256-bit vector length and all larger vector lengths
+     since 256 is a power-of-two (this results in only the 128-bit length
+     being enabled)::
+

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 261

Quoted 10 lines
docs/arm-cpu-features.rst +10 -0
+     $ qemu-system-aarch64 -M virt -cpu max,sve256=off
+
+  5) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
+
+  6) The same as (5), but since the 128-bit and 256-bit vector
+     lengths are required for the 512-bit vector length to be enabled,
+     then allow them to be auto-enabled::> +
+     $ qemu-system-aarch64 -M virt -cpu max,sve512=on

You should also document

Quoted 4 lines
$ qemu-system-aarch64 -M virt -cpu max,sve512=off
as in that case this result in 128, 256 and 384. the fact you get non ^2
values is not obvious here because if you enable a ^2 value you onlt get
<= ^2 values.

I can change (4) above to use 512 instead of 256.

Quoted 6 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 267

Quoted 5 lines
docs/arm-cpu-features.rst +5 -0
+
+  7) Do the same as (6), but by first disabling SVE and then re-enabling it::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on> +
+  8) Force errors regarding the last vector length::

You mean those commands will generate errors, right?

Yes

Quoted 30 lines
Quoted 19 lines
docs/arm-cpu-features.rst +11 -1
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=off
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
+
+SVE CPU Property Recommendations
+--------------------------------
 
-The example above disables the PMU for the `max` CPU type.
+The examples in "SVE CPU Property Examples" exhibit many ways to select
+vector lengths which developers may find useful in order to avoid overly
+verbose command lines.  However, the recommended way to select vector
+lengths is to explicitly enable each desired length.  Therefore only
+example's (1), (3), and (5) exhibit recommended uses of the properties.
 
target/arm/cpu.c +0 -0
index 73be2ebfdd39..522fed95b339 100644

target/arm/cpu.c static void arm_cpu_finalizefn(Object *obj) · line 1208

Quoted 10 lines
target/arm/cpu.c +7 -0
@@ -1199,6 +1199,19 @@ static void arm_cpu_finalizefn(Object *obj)
 #endif
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp)
+{
+    Error *local_err = NULL;
+
+    if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
+        arm_cpu_sve_finalize(cpu, &local_err);
+        if (local_err != NULL) {

nit: !local_err

Quoted 3 lines

target/arm/cpu.c static void arm_cpu_finalizefn(Object *obj) · line 1210

Quoted 2 lines
target/arm/cpu.c +2 -0
+            error_propagate(errp, local_err);
+            return;

not needed

Yet. I expect a chain of finalizers to go here, so we'll want the 'return' when we get the next one. As it doesn't hurt, it might as well go in now.

Quoted 101 lines

target/arm/cpu.c static void arm_cpu_realizefn(DeviceState *dev, Error **errp) · line 1272

Quoted 13 lines
target/arm/cpu.c +6 -0
+        }
+    }
+}
+
 static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
 {
     CPUState *cs = CPU(dev);
@@ -1255,6 +1268,12 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         return;
     }
 
+    arm_cpu_finalize_features(cpu, &local_err);
+    if (local_err != NULL) {

same

Quoted 87 lines
target/arm/cpu.c +3 -0
+        error_propagate(errp, local_err);
+        return;> +    }
+
     if (arm_feature(env, ARM_FEATURE_AARCH64) &&
         cpu->has_vfp != cpu->has_neon) {
         /*
target/arm/cpu.h +19 -0
index 297ad5e47ad8..11162484465a 100644
@@ -184,8 +184,13 @@ typedef struct {
 
 #ifdef TARGET_AARCH64
 # define ARM_MAX_VQ    16
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp);
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq);
 #else
 # define ARM_MAX_VQ    1
+static inline void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) { }
+static inline uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{ return 0; }
 #endif
 
 typedef struct ARMVectorReg {
@@ -915,6 +920,18 @@ struct ARMCPU {
 
     /* Used to set the maximum vector length the cpu will support.  */
     uint32_t sve_max_vq;
+
+    /*
+     * In sve_vq_map each set bit is a supported vector length of
+     * (bit-number + 1) * 16 bytes, i.e. each bit number + 1 is the vector
+     * length in quadwords.
+     *
+     * While processing properties during initialization, corresponding
+     * sve_vq_init bits are set for bits in sve_vq_map that have been
+     * set by properties.
+     */
+    DECLARE_BITMAP(sve_vq_map, ARM_MAX_VQ);
+    DECLARE_BITMAP(sve_vq_init, ARM_MAX_VQ);
 };
 
 void arm_cpu_post_init(Object *obj);
@@ -1834,6 +1851,8 @@ static inline int arm_feature(CPUARMState *env, int feature)
     return (env->features & (1ULL << feature)) != 0;
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp);
+
 #if !defined(CONFIG_USER_ONLY)
 /* Return true if exception levels below EL3 are in secure state,
  * or would be following an exception return to that level.
target/arm/cpu64.c +27 -0
index 8cdb0c79fa7a..606e3eceb9c0 100644
@@ -256,11 +256,152 @@ static void aarch64_a72_initfn(Object *obj)
     define_arm_cp_regs(cpu, cortex_a72_a57_a53_cp_reginfo);
 }
 
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
+{
+    /*
+     * If any vector lengths are explicitly enabled with sve<N> properties,
+     * then all other lengths are implicitly disabled.  If sve-max-vq is
+     * specified then it is the same as explicitly enabling all lengths
+     * up to and including the specified maximum, which means all larger
+     * lengths will be implicitly disabled.  If no sve<N> properties
+     * are enabled and sve-max-vq is not specified, then all lengths not
+     * explicitly disabled will be enabled.  Additionally, all power-of-two
+     * vector lengths less than the maximum enabled length will be
+     * automatically enabled and all vector lengths larger than the largest
+     * disabled power-of-two vector length will be automatically disabled.
+     * Errors are generated if the user provided input that interferes with
+     * any of the above.  Finally, if SVE is not disabled, then at least one
+     * vector length must be enabled.
+     */
+    DECLARE_BITMAP(tmp, ARM_MAX_VQ);
+    uint32_t vq, max_vq = 0;
+
+    /*
+     * Process explicit sve<N> properties.
+     * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
+     * Check first for any sve<N> enabled.
+     */
+    if (!bitmap_empty(cpu->sve_vq_map, ARM_MAX_VQ)) {
+        max_vq = find_last_bit(cpu->sve_vq_map, ARM_MAX_VQ) + 1;
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 292

Quoted 7 lines
target/arm/cpu64.c +7 -0
+
+        if (cpu->sve_max_vq && max_vq > cpu->sve_max_vq) {
+            error_setg(errp, "cannot enable sve%d", max_vq * 128);
+            error_append_hint(errp, "sve%d is larger than the maximum vector "
+                              "length, sve-max-vq=%d (%d bits)\n",
+                              max_vq * 128, cpu->sve_max_vq,
+                              cpu->sve_max_vq * 128);

you could test directly if max_vq < cpu->sve_max_vq too which couldn't

That's not an error. Here max_vq is the maximum enabled with sve<N> properties. We haven't considered cpu->sve_max_vq yet, except for ensuring we don't enable something larger than it.

Quoted 70 lines

be correct either. But as far as I understand this is caught in *

Quoted 69 lines
target/arm/cpu64.c +69 -0
+            return;
+        }
+
+        /* Propagate enabled bits down through required powers-of-two. */
+        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                set_bit(vq - 1, cpu->sve_vq_map);
+            }
+        }
+    } else if (cpu->sve_max_vq == 0) {
+        /*
+         * No explicit bits enabled, and no implicit bits from sve-max-vq.
+         */
+        if (!cpu_isar_feature(aa64_sve, cpu)) {
+            /* SVE is disabled and so are all vector lengths.  Good. */
+            return;
+        }
+
+        /* Disabling a power-of-two disables all larger lengths. */
+        if (test_bit(0, cpu->sve_vq_init)) {> +            error_setg(errp, "cannot disable sve128");
+            error_append_hint(errp, "Disabling sve128 results in all vector "
+                              "lengths being disabled.\n");
+            error_append_hint(errp, "With SVE enabled, at least one vector "
+                              "length must be enabled.\n");
+            return;
+        }
+        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
+            if (test_bit(vq - 1, cpu->sve_vq_init)) {
+                break;
+            }
+        }
+        max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+
+        bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
+        max_vq = find_last_bit(cpu->sve_vq_map, max_vq) + 1;
+    }
+
+    /*
+     * Process the sve-max-vq property.
+     * Note that we know from the above that no bit above
+     * sve-max-vq is currently set.
+     */
+    if (cpu->sve_max_vq != 0) {
+        max_vq = cpu->sve_max_vq;
+
+        if (!test_bit(max_vq - 1, cpu->sve_vq_map) &&
+            test_bit(max_vq - 1, cpu->sve_vq_init)) {
+            error_setg(errp, "cannot disable sve%d", max_vq * 128);
+            error_append_hint(errp, "The maximum vector length must be "
+                              "enabled, sve-max-vq=%d (%d bits)\n",
+                              max_vq, max_vq * 128);
+            return;
+        }
+
+        /* Set all bits not explicitly set within sve-max-vq. */
+        bitmap_complement(tmp, cpu->sve_vq_init, max_vq);
+        bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+    }
+
+    /*
+     * We should know what max-vq is now.  Also, as we're done
+     * manipulating sve-vq-map, we ensure any bits above max-vq
+     * are clear, just in case anybody looks.
+     */
+    assert(max_vq != 0);
+    bitmap_clear(cpu->sve_vq_map, max_vq, ARM_MAX_VQ - max_vq);
+
+    /* Ensure all required powers-of-two are enabled. */
+    for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 368

Quoted 6 lines
target/arm/cpu64.c +6 -0
+        if (!test_bit(vq - 1, cpu->sve_vq_map)) {
+            error_setg(errp, "cannot disable sve%d", vq * 128);
+            error_append_hint(errp, "sve%d is required as it "
+                              "is a power-of-two length smaller than "
+                              "the maximum, sve%d\n",
+                              vq * 128, max_vq * 128);

* I think this will also catches the case where ,sve512=on,sve-max-vq=5 where the end-user did not explicitly disabled anything

As stated above, 'sve512=on,sve-max-vq=5' isn't an error. It doesn't hurt to enable vector lengths explicitly that would be enabled automatically anyway with the sve-max-vq property.

Quoted 22 lines
Quoted 22 lines
target/arm/cpu64.c +22 -0
+            return;
+        }
+    }
+
+    /* From now on sve_max_vq is the actual maximum supported length. */
+    cpu->sve_max_vq = max_vq;
+}
+
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{
+    uint32_t bitnum;
+
+    /*
+     * We allow vq == ARM_MAX_VQ + 1 to be input because the caller may want
+     * to find the maximum vq enabled, which may be ARM_MAX_VQ, but this
+     * function always returns the next smaller than the input.
+     */
+    assert(vq && vq <= ARM_MAX_VQ + 1);
+
+    bitnum = find_last_bit(cpu->sve_vq_map, vq - 1);
+    return bitnum == vq - 1 ? 0 : bitnum + 1;
+}
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 396

Quoted 7 lines
target/arm/cpu64.c +2 -1
+
 static void cpu_max_get_sve_max_vq(Object *obj, Visitor *v, const char *name,
                                    void *opaque, Error **errp)
 {
     ARMCPU *cpu = ARM_CPU(obj);
-    visit_type_uint32(v, name, &cpu->sve_max_vq, errp);
+    uint32_t value;

Shouldn't it be part of the previous patch?

Yes. Will fix for v5.

Quoted 10 lines
Quoted 10 lines
target/arm/cpu64.c +8 -0
+
+    /* All vector lengths are disabled when SVE is off. */
+    if (!cpu_isar_feature(aa64_sve, cpu)) {
+        value = 0;
+    } else {
+        value = cpu->sve_max_vq;
+    }
+    visit_type_uint32(v, name, &value, errp);
 }
 
Quoted 4 lines

target/arm/cpu64.c static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,

Quoted 2 lines
target/arm/cpu64.c +0 -0
 static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,
@@ -279,6 +420,44 @@ static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,

Independently on this patch I noticed that if sve_max_vq is out of scope, an error is propagated but still the cpu->sve-max_vq is set.

I'll sneak that fix into patch 8/9 where we touch the function anyway.

Quoted 309 lines
Quoted 309 lines
target/arm/cpu64.c +46 -1
     error_propagate(errp, err);
 }
 
+static void cpu_arm_get_sve_vq(Object *obj, Visitor *v, const char *name,
+                               void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq = atoi(&name[3]) / 128;
+    bool value;
+
+    /* All vector lengths are disabled when SVE is off. */
+    if (!cpu_isar_feature(aa64_sve, cpu)) {
+        value = false;
+    } else {
+        value = test_bit(vq - 1, cpu->sve_vq_map);
+    }
+    visit_type_bool(v, name, &value, errp);
+}
+
+static void cpu_arm_set_sve_vq(Object *obj, Visitor *v, const char *name,
+                               void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq = atoi(&name[3]) / 128;
+    Error *err = NULL;
+    bool value;
+
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
+
+    if (value) {
+        set_bit(vq - 1, cpu->sve_vq_map);
+    } else {
+        clear_bit(vq - 1, cpu->sve_vq_map);
+    }
+    set_bit(vq - 1, cpu->sve_vq_init);
+}
+
 static void cpu_arm_get_sve(Object *obj, Visitor *v, const char *name,
                             void *opaque, Error **errp)
 {
@@ -315,6 +494,7 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
 static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
@@ -418,11 +598,17 @@ static void aarch64_max_initfn(Object *obj)
         cpu->dcz_blocksize = 7; /*  512 bytes */
 #endif
 
-        cpu->sve_max_vq = ARM_MAX_VQ;
         object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                             cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
         object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
                             cpu_arm_set_sve, NULL, NULL, &error_fatal);
+
+        for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+            char name[8];
+            sprintf(name, "sve%d", vq * 128);
+            object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                                cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+        }
     }
 }
 
target/arm/helper.c +10 -3
index 507026c9154b..f33284c247d5 100644
@@ -5351,6 +5351,13 @@ int sve_exception_el(CPUARMState *env, int el)
     return 0;
 }
 
+static uint32_t sve_zcr_get_valid_len(ARMCPU *cpu, uint32_t start_len)
+{
+    uint32_t start_vq = (start_len & 0xf) + 1;
+
+    return arm_cpu_vq_map_next_smaller(cpu, start_vq + 1) - 1;
+}
+
 /*
  * Given that SVE is enabled, return the vector length for EL.
  */
@@ -5360,13 +5367,13 @@ uint32_t sve_zcr_len_for_el(CPUARMState *env, int el)
     uint32_t zcr_len = cpu->sve_max_vq - 1;
 
     if (el <= 1) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[1]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[1]);
     }
     if (el <= 2 && arm_feature(env, ARM_FEATURE_EL2)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[2]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[2]);
     }
     if (arm_feature(env, ARM_FEATURE_EL3)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[3]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[3]);
     }
     return zcr_len;
 }
target/arm/monitor.c +12 -0
index 4fddb6c252a3..e912ed2cefa0 100644
@@ -90,6 +90,8 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
     return head;
 }
 
+QEMU_BUILD_BUG_ON(ARM_MAX_VQ > 16);
+
 /*
  * These are cpu model features we want to advertise. The order here
  * matters as this is the order in which qmp_query_cpu_model_expansion
@@ -98,6 +100,9 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
  */
 static const char *cpu_model_advertised_features[] = {
     "aarch64", "pmu", "sve",
+    "sve128", "sve256", "sve384", "sve512",
+    "sve640", "sve768", "sve896", "sve1024", "sve1152", "sve1280",
+    "sve1408", "sve1536", "sve1664", "sve1792", "sve1920", "sve2048",
     NULL
 };
 
@@ -185,6 +190,9 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
         if (!err) {
             visit_check_struct(visitor, &err);
         }
+        if (!err) {
+            arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        }
         visit_end_struct(visitor, NULL);
         visit_free(visitor);
         if (err) {
@@ -192,6 +200,10 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
             error_propagate(errp, err);
             return NULL;
         }
+    } else {
+        Error *err = NULL;
+        arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        assert(err == NULL);
     }
 
     expansion_info = g_new0(CpuModelExpansionInfo, 1);
tests/arm-cpu-features.c +144 -0
index 202bc0e3e823..9a2dd402b769 100644
@@ -13,6 +13,18 @@
 #include "qapi/qmp/qdict.h"
 #include "qapi/qmp/qjson.h"
 
+#if __SIZEOF_LONG__ == 8
+#define BIT(n) (1UL << (n))
+#else
+#define BIT(n) (1ULL << (n))
+#endif
+
+/*
+ * We expect the SVE max-vq to be 16. Also it must be <= 64
+ * for our test code, otherwise 'vls' can't just be a uint64_t.
+ */
+#define SVE_MAX_VQ 16
+
 #define MACHINE    "-machine virt,gic-version=max "
 #define QUERY_HEAD "{ 'execute': 'query-cpu-model-expansion', " \
                      "'arguments': { 'type': 'full', "
@@ -157,6 +169,173 @@ static void assert_bad_props(QTestState *qts, const char *cpu_type)
     qobject_unref(resp);
 }
 
+static uint64_t resp_get_sve_vls(QDict *resp)
+{
+    QDict *props;
+    const QDictEntry *e;
+    uint64_t vls = 0;
+    int n = 0;
+
+    g_assert(resp);
+    g_assert(resp_has_props(resp));
+
+    props = resp_get_props(resp);
+
+    for (e = qdict_first(props); e; e = qdict_next(props, e)) {
+        if (strlen(e->key) > 3 && !strncmp(e->key, "sve", 3) &&
+            g_ascii_isdigit(e->key[3])) {
+            char *endptr;
+            int bits;
+
+            bits = g_ascii_strtoll(&e->key[3], &endptr, 10);
+            if (!bits || *endptr != '\0') {
+                continue;
+            }
+
+            if (qdict_get_bool(props, e->key)) {
+                vls |= BIT((bits / 128) - 1);
+            }
+            ++n;
+        }
+    }
+
+    g_assert(n == SVE_MAX_VQ);
+
+    return vls;
+}
+
+#define assert_sve_vls(qts, cpu_type, expected_vls, fmt, ...)          \
+({                                                                     \
+    QDict *_resp = do_query(qts, cpu_type, fmt, ##__VA_ARGS__);        \
+    g_assert(_resp);                                                   \
+    g_assert(resp_has_props(_resp));                                   \
+    g_assert(resp_get_sve_vls(_resp) == expected_vls);                 \
+    qobject_unref(_resp);                                              \
+})
+
+static void sve_tests_default(QTestState *qts, const char *cpu_type)
+{
+    /*
+     * With no sve-max-vq or sve<N> properties on the command line
+     * the default is to have all vector lengths enabled. This also
+     * tests that 'sve' is 'on' by default.
+     */
+    assert_sve_vls(qts, cpu_type, BIT(SVE_MAX_VQ) - 1, NULL);
+
+    /* With SVE off, all vector lengths should also be off. */
+    assert_sve_vls(qts, cpu_type, 0, "{ 'sve': false }");
+
+    /* With SVE on, we must have at least one vector length enabled. */
+    assert_error(qts, cpu_type, "cannot disable sve128", "{ 'sve128': false }");
+
+    /*
+     * ---------------------------------------------------------------------
+     *               power-of-two(vq)   all-power-            can      can
+     *                                  of-two(< vq)        enable   disable
+     * ---------------------------------------------------------------------
+     * vq < max_vq      no                MUST*              yes      yes
+     * vq < max_vq      yes               MUST*              yes      no
+     * ---------------------------------------------------------------------
+     * vq == max_vq     n/a               MUST*              yes**    yes**
+     * ---------------------------------------------------------------------
+     * vq > max_vq      n/a               no                 no       yes
+     * vq > max_vq      n/a               yes                yes      yes
+     * ---------------------------------------------------------------------
+     *
+     * [*] "MUST" means this requirement must already be satisfied,
+     *     otherwise 'max_vq' couldn't itself be enabled.
+     *
+     * [**] Not testable with the QMP interface, only with the command line.
+     */
+
+    /* max_vq := 8 */
+    assert_sve_vls(qts, cpu_type, 0x8b, "{ 'sve1024': true }");
+
+    /* max_vq := 8, vq < max_vq, !power-of-two(vq) */
+    assert_sve_vls(qts, cpu_type, 0x8f,
+                   "{ 'sve1024': true, 'sve384': true }");
+    assert_sve_vls(qts, cpu_type, 0x8b,
+                   "{ 'sve1024': true, 'sve384': false }");
+
+    /* max_vq := 8, vq < max_vq, power-of-two(vq) */
+    assert_sve_vls(qts, cpu_type, 0x8b,
+                   "{ 'sve1024': true, 'sve256': true }");
+    assert_error(qts, cpu_type, "cannot disable sve256",
+                 "{ 'sve1024': true, 'sve256': false }");
+
+    /* max_vq := 3, vq > max_vq, !all-power-of-two(< vq) */
+    assert_error(qts, cpu_type, "cannot disable sve512",
+                 "{ 'sve384': true, 'sve512': false, 'sve640': true }");
+
+    /*
+     * We can disable power-of-two vector lengths when all larger lengths
+     * are also disabled. We only need to disable the power-of-two length,
+     * as all non-enabled larger lengths will then be auto-disabled.
+     */
+    assert_sve_vls(qts, cpu_type, 0x7, "{ 'sve512': false }");
+
+    /* max_vq := 3, vq > max_vq, all-power-of-two(< vq) */
+    assert_sve_vls(qts, cpu_type, 0x1f,
+                   "{ 'sve384': true, 'sve512': true, 'sve640': true }");
+    assert_sve_vls(qts, cpu_type, 0xf,
+                   "{ 'sve384': true, 'sve512': true, 'sve640': false }");
+}
+
+static void sve_tests_sve_max_vq_8(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-cpu max,sve-max-vq=8");
+
+    assert_sve_vls(qts, "max", BIT(8) - 1, NULL);
+
+    /*
+     * Disabling the max-vq set by sve-max-vq is not allowed, but
+     * of course enabling it is OK.
+     */
+    assert_error(qts, "max", "cannot disable sve1024", "{ 'sve1024': false }");
+    assert_sve_vls(qts, "max", 0xff, "{ 'sve1024': true }");
+
+    /*
+     * Enabling anything larger than max-vq set by sve-max-vq is not
+     * allowed, but of course disabling everything larger is OK.
+     */
+    assert_error(qts, "max", "cannot enable sve1152", "{ 'sve1152': true }");
Quoted 8 lines

tests/arm-cpu-features.c static void assert_bad_props(QTestState *qts, const char *cpu_type) · line 310

Quoted 7 lines
tests/arm-cpu-features.c +7 -0
+    assert_sve_vls(qts, "max", 0xff, "{ 'sve1152': false }");
+
+    /*
+     * We can disable non power-of-two lengths smaller than the max-vq
+     * set by sve-max-vq, but not power-of-two lengths.
+     */
+    assert_sve_vls(qts, "max", 0xfb, "{ 'sve384': false }");

you can also test ,sve384=on => 0x7

We wouldn't get 0x7, we'd get 0xff. I could still add the test to ensure we can enable smaller vector lengths without an error though.

Quoted 61 lines
Quoted 49 lines
tests/arm-cpu-features.c +31 -0
+    assert_error(qts, "max", "cannot disable sve256", "{ 'sve256': false }");
+
+    qtest_quit(qts);
+}
+
+static void sve_tests_sve_off(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-cpu max,sve=off");
+
+    /* SVE is off, so the map should be empty. */
+    assert_sve_vls(qts, "max", 0, NULL);
+
+    /* The map stays empty even if we turn lengths on or off. */
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': true }");
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+
+    /* With SVE re-enabled we should get all vector lengths enabled. */
+    assert_sve_vls(qts, "max", BIT(SVE_MAX_VQ) - 1, "{ 'sve': true }");
+
+    /* Or enable SVE with just specific vector lengths. */
+    assert_sve_vls(qts, "max", 0x3,
+                   "{ 'sve': true, 'sve128': true, 'sve256': true }");
+
+    qtest_quit(qts);
+}
+
 static void test_query_cpu_model_expansion(const void *data)
 {
     QTestState *qts;
@@ -180,9 +359,12 @@ static void test_query_cpu_model_expansion(const void *data)
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
         assert_has_feature(qts, "max", "aarch64");
         assert_has_feature(qts, "max", "sve");
+        assert_has_feature(qts, "max", "sve128");
         assert_has_feature(qts, "cortex-a57", "pmu");
         assert_has_feature(qts, "cortex-a57", "aarch64");
 
+        sve_tests_default(qts, "max");
+
         /* Test that features that depend on KVM generate errors without. */
         assert_error(qts, "max",
                      "'aarch64' feature cannot be disabled "
@@ -234,6 +416,13 @@ int main(int argc, char **argv)
     qtest_add_data_func("/arm/query-cpu-model-expansion",
                         NULL, test_query_cpu_model_expansion);
 

tests/arm-cpu-features.c int main(int argc, char **argv) · line 428

Quoted 10 lines
tests/arm-cpu-features.c +7 -0
+    if (g_str_equal(qtest_get_arch(), "aarch64")) {
+        qtest_add_data_func("/arm/max/query-cpu-model-expansion/sve-max-vq-8",
+                            NULL, sve_tests_sve_max_vq_8);
+        qtest_add_data_func("/arm/max/query-cpu-model-expansion/sve-off",
+                            NULL, sve_tests_sve_off);
+    }
+
     if (kvm_available) {
         qtest_add_data_func("/arm/kvm/query-cpu-model-expansion",
                             NULL, test_query_cpu_model_expansion_kvm);

Thanks

Quoted 2 lines

Eric

Thanks,
drew

#10 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

Quoted 56 lines · Andrew Jones
Quoted 38 lines · Auger Eric

Hi Drew,

Quoted 35 lines · Andrew Jones

Introduce cpu properties to give fine control over SVE vector lengths. We introduce a property for each valid length up to the current maximum supported, which is 2048-bits. The properties are named, e.g. sve128, sve256, sve384, sve512, ..., where the number is the number of bits. See the updates to docs/arm-cpu-features.rst for a description of the semantics and for example uses.

Note, as sve-max-vq is still present and we'd like to be able to support qmp_query_cpu_model_expansion with guests launched with e.g. -cpu max,sve-max-vq=8 on their command lines, then we do allow sve-max-vq and sve<N> properties to be provided at the same time, but this is not recommended, and is why sve-max-vq is not mentioned in the document. If sve-max-vq is provided then it enables all lengths smaller than and including the max and disables all lengths larger. It also has the side-effect that no larger lengths may be enabled and that the max itself cannot be disabled. Smaller non-power-of-two lengths may, however, be disabled, e.g. -cpu max,sve-max-vq=4,sve384=off provides a guest the vector lengths 128, 256, and 512 bits.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst | 157 +++++++++++++++++++++++++++++--
 target/arm/cpu.c          |  19 ++++
 target/arm/cpu.h          |  19 ++++
 target/arm/cpu64.c        | 190 +++++++++++++++++++++++++++++++++++++-
 target/arm/helper.c       |  13 ++-
 target/arm/monitor.c      |  12 +++
 tests/arm-cpu-features.c  | 189 +++++++++++++++++++++++++++++++++++++
 7 files changed, 587 insertions(+), 12 deletions(-)
Quoted 16 lines
Quoted 5 lines
docs/arm-cpu-features.rst +0 -0
index c79dcffb5556..1262fddc6201 100644

docs/arm-cpu-features.rst block in the script for usage) is used to issue the QMP commands. · line 55

Quoted 9 lines
docs/arm-cpu-features.rst +5 -1
@@ -48,18 +48,28 @@ block in the script for usage) is used to issue the QMP commands.
       (QEMU) query-cpu-model-expansion type=full model={"name":"max"}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": true, "aarch64": true
+        "sve1664": true, "pmu": true, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true

I don't really understand why this gets printed unsorted as you registered them in the logical order, any clue?

Quoted 2 lines

Maybe qdict hashing? qmp-shell could probably be patched to sort them, but it's probably not worth it.

OK. Anyway beyond the scope of the series but that's unpractical.

Quoted 8 lines
Quoted 4 lines

Note: This example is given with TCG or KVM with all those vectors supported by the host.

You want this note to be added? I don't think it's necessary, as it's just an example. A different configuration or different CPU model expansion will have different results, as we explain exhaustively in this document.

a reader may understand that along with max cpu type whatever the state of KVM all sve<N> are supported by default, which is not true in KVM case. Maybe add TCG case.

Quoted 63 lines
Quoted 52 lines
Quoted 52 lines
docs/arm-cpu-features.rst +25 -4
       }}}}
 
-We see that the `max` CPU type has the `pmu` and `aarch64` CPU features.
-We also see that the CPU features are enabled, as they are all `true`.
+We see that the `max` CPU type has the `pmu`, `aarch64`, `sve`, and many
+`sve<N>` CPU features.  We also see that all the CPU features are
+enabled, as they are all `true`.  (The `sve<N>` CPU features are all
+optional SVE vector lengths.  See "SVE CPU Properties".)
 
 (2) Let's try to disable the PMU::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"pmu":false}}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": false, "aarch64": true
+        "sve1664": true, "pmu": false, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true
       }}}}
 
 We see it worked, as `pmu` is now `false`.
@@ -75,7 +85,22 @@ We see it worked, as `pmu` is now `false`.
 It looks like this feature is limited to a configuration we do not
 currently have.
 
-(4) Let's try probing CPU features for the Cortex-A15 CPU type::
+(4) Let's disable `sve` and see what happens to all the optional SVE
+    vector lengths::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"sve":false}}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "sve1664": false, "pmu": true, "sve1792": false, "sve1920": false,
+        "sve128": false, "aarch64": true, "sve1024": false, "sve": false,
+        "sve640": false, "sve768": false, "sve1408": false, "sve256": false,
+        "sve1152": false, "sve512": false, "sve384": false, "sve1536": false,
+        "sve896": false, "sve1280": false, "sve2048": false
+      }}}}
+
+As expected they are now all `false`.
+
+(5) Let's try probing CPU features for the Cortex-A15 CPU type::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"cortex-a15"}
       {"return": {"model": {"name": "cortex-a15", "props": {"pmu": true}}}}
@@ -131,7 +156,125 @@ After determining which CPU features are available and supported for a
 given CPU type, then they may be selectively enabled or disabled on the
 QEMU command line with that CPU type::
 
Quoted 10 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a

Quoted 7 lines
docs/arm-cpu-features.rst +6 -1
-  $ qemu-system-aarch64 -M virt -cpu max,pmu=off
+  $ qemu-system-aarch64 -M virt -cpu max,pmu=off,sve=on,sve128=on,sve256=on
+
+The example above disables the PMU and enables the first two SVE vector
+lengths for the `max` CPU type.  Note, the `sve=on` isn't actually
+necessary, because, as we observed above with our probe of the `max` CPU
+type, `sve` is already on by default.

This holds with TCG and not KVM.

Quoted 1 line

It also holds with KVM, as long as the host supports SVE.

as long as the hist supports SVE: Yes that's what I meant

maybe add, as we observed abovewith our probe of the max CPU

Quoted 1 line
Quoted 1 line

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 165

Quoted 1 line
docs/arm-cpu-features.rst +1 -0
+type, in this TCG case, `sve` is already on by default

But that doesn't

Quoted 38 lines

matter, because we're speaking specifically about the example above here, where we see that sve=on is the default.

Quoted 26 lines

Also, based on our probe of

Quoted 24 lines
docs/arm-cpu-features.rst +24 -0
+defaults, it would seem we need to disable many SVE vector lengths, rather
+than only enabling the two we want.  This isn't the case, because, as
+disabling many SVE vector lengths would be quite verbose, the `sve<N>` CPU
+properties have special semantics (see "SVE CPU Property Parsing
+Semantics").
+
+SVE CPU Properties
+==================
+
+There are two types of SVE CPU properties: `sve` and `sve<N>`.  The first
+is used to enable or disable the entire SVE feature, just as the `pmu`
+CPU property completely enables or disables the PMU.  The second type
+is used to enable or disable specific vector lengths, where `N` is the
+number of bits of the length.  The `sve<N>` CPU properties have special
+dependencies and constraints, see "SVE CPU Property Dependencies and
+Constraints" below.  Additionally, as we want all supported vector lengths
+to be enabled by default, then, in order to avoid overly verbose command
+lines (command lines full of `sve<N>=off`, for all `N` not wanted), we
+provide the parsing semantics listed in "SVE CPU Property Parsing
+Semantics".
+
+SVE CPU Property Dependencies and Constraints
+---------------------------------------------
+
Quoted 8 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 194

Quoted 5 lines
docs/arm-cpu-features.rst +5 -0
+  1) At least one vector length must be enabled when `sve` is enabled.> +
+  2) If a vector length `N` is enabled, then all power-of-two vector
+     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
+     is enabled, then `sve128` and `sve256` must also be enabled,
+     but `sve384` is not required.

I would remove the eg. part. reading that I tend to understand that the user must pass ,sve128=on, sve256=on and sve512=on whereas example 6 says only sve512=on can be set and other lower ^2 values are auto-enabled.

Quoted 2 lines

I'll keep the example, but instead of using the property names, I'll change to using "the 128-bit vector length", 256-bit ..., etc.

ok

Quoted 17 lines
Quoted 16 lines
Quoted 4 lines
docs/arm-cpu-features.rst +3 -0
+
+SVE CPU Property Parsing Semantics
+----------------------------------

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 207

Quoted 10 lines
docs/arm-cpu-features.rst +10 -0
+
+  1) If SVE is disabled (`sve=off`), then which SVE vector lengths
+     are enabled or disabled is irrelevant to the guest, as the entire
+     SVE feature is disabled and that disables all vector lengths for
+     the guest.  However QEMU will still track any `sve<N>` CPU
+     properties provided by the user.  If later an `sve=on` is provided,
+     then the guest will get only the enabled lengths.
+
+  2) If SVE is enabled (`sve=on`), but no `sve<N>` CPU properties are
+     provided, then all supported vector lengths are enabled.

I understand from the code it also includes non ^2 values which is not obvious.

Quoted 3 lines

Yes, the default is to have *all* supported vector lengths enabled, not just the power-of-two. As we already use the word 'all' here, then how can we further clarify this?

I would simply add "including non ^2 values".

Quoted 16 lines
Quoted 6 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 211

Quoted 5 lines
docs/arm-cpu-features.rst +4 -0
+
+  3) If SVE is enabled, then an error is generated when attempting to
+     disable the last enabled vector length (see constraint (1) of "SVE
+     CPU Property Dependencies and Constraints").

the same happens if you attempt to disabled any lower ^2 value

That should be another bullet, which I appear to be missing, and will add.

Quoted 7 lines
Quoted 7 lines
docs/arm-cpu-features.rst +7 -0
+
+  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+     CPU properties are set `on`, then the specified vector lengths are
+     disabled but the default for any unspecified lengths remains enabled.
+     Disabling a power-of-two vector length also disables all vector
+     lengths larger than the power-of-two length (see constraint (2) of
+     "SVE CPU Property Dependencies and Constraints").
Quoted 57 lines
Quoted 4 lines

So rephasing it:
disabling a non ^2 value only disable that one disabling a ^2 value disables all larger lengths too is that correct?

That's correct. We auto-disable all uninitialized larger lengths when a power-of-two dependency length is explicitly disabled.

Quoted 49 lines
Quoted 33 lines
docs/arm-cpu-features.rst +32 -0
+
+  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+     are enabled and all unspecified lengths default to disabled, except
+     for the required lengths per constraint (2) of "SVE CPU Property
+     Dependencies and Constraints", which will even be auto-enabled if
+     they were not explicitly enabled.
+
+  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+     explicitly disabled (i.e. avoiding the error specified in (3) of
+     "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
+     provided an error will be generated.  To avoid this error, one must
+     enable at least one vector length prior to enabling SVE.
+
+SVE CPU Property Examples
+-------------------------
+
+  1) Disable SVE::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off
+
+  2) Implicitly enable all vector lengths for the `max` CPU type::
+
+     $ qemu-system-aarch64 -M virt -cpu max
+
+  3) Only enable the 128-bit vector length::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on
+
+  4) Disable the 256-bit vector length and all larger vector lengths
+     since 256 is a power-of-two (this results in only the 128-bit length
+     being enabled)::
+

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 261

Quoted 10 lines
docs/arm-cpu-features.rst +10 -0
+     $ qemu-system-aarch64 -M virt -cpu max,sve256=off
+
+  5) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
+
+  6) The same as (5), but since the 128-bit and 256-bit vector
+     lengths are required for the 512-bit vector length to be enabled,
+     then allow them to be auto-enabled::> +
+     $ qemu-system-aarch64 -M virt -cpu max,sve512=on

You should also document

$ qemu-system-aarch64 -M virt -cpu max,sve512=off
as in that case this result in 128, 256 and 384. the fact you get non ^2
values is not obvious here because if you enable a ^2 value you onlt get
<= ^2 values.

yep.

Quoted 159 lines

I can change (4) above to use 512 instead of 256.

Quoted 6 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 267

Quoted 5 lines
docs/arm-cpu-features.rst +5 -0
+
+  7) Do the same as (6), but by first disabling SVE and then re-enabling it::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on> +
+  8) Force errors regarding the last vector length::

You mean those commands will generate errors, right?

Yes

Quoted 30 lines
Quoted 19 lines
docs/arm-cpu-features.rst +11 -1
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=off
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
+
+SVE CPU Property Recommendations
+--------------------------------
 
-The example above disables the PMU for the `max` CPU type.
+The examples in "SVE CPU Property Examples" exhibit many ways to select
+vector lengths which developers may find useful in order to avoid overly
+verbose command lines.  However, the recommended way to select vector
+lengths is to explicitly enable each desired length.  Therefore only
+example's (1), (3), and (5) exhibit recommended uses of the properties.
 
target/arm/cpu.c +0 -0
index 73be2ebfdd39..522fed95b339 100644

target/arm/cpu.c static void arm_cpu_finalizefn(Object *obj) · line 1208

Quoted 10 lines
target/arm/cpu.c +7 -0
@@ -1199,6 +1199,19 @@ static void arm_cpu_finalizefn(Object *obj)
 #endif
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp)
+{
+    Error *local_err = NULL;
+
+    if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
+        arm_cpu_sve_finalize(cpu, &local_err);
+        if (local_err != NULL) {

nit: !local_err

Quoted 3 lines

target/arm/cpu.c static void arm_cpu_finalizefn(Object *obj) · line 1210

Quoted 2 lines
target/arm/cpu.c +2 -0
+            error_propagate(errp, local_err);
+            return;

not needed

Yet. I expect a chain of finalizers to go here, so we'll want the 'return' when we get the next one. As it doesn't hurt, it might as well go in now.

Quoted 101 lines

target/arm/cpu.c static void arm_cpu_realizefn(DeviceState *dev, Error **errp) · line 1272

Quoted 13 lines
target/arm/cpu.c +6 -0
+        }
+    }
+}
+
 static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
 {
     CPUState *cs = CPU(dev);
@@ -1255,6 +1268,12 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         return;
     }
 
+    arm_cpu_finalize_features(cpu, &local_err);
+    if (local_err != NULL) {

same

Quoted 87 lines
target/arm/cpu.c +3 -0
+        error_propagate(errp, local_err);
+        return;> +    }
+
     if (arm_feature(env, ARM_FEATURE_AARCH64) &&
         cpu->has_vfp != cpu->has_neon) {
         /*
target/arm/cpu.h +19 -0
index 297ad5e47ad8..11162484465a 100644
@@ -184,8 +184,13 @@ typedef struct {
 
 #ifdef TARGET_AARCH64
 # define ARM_MAX_VQ    16
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp);
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq);
 #else
 # define ARM_MAX_VQ    1
+static inline void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) { }
+static inline uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{ return 0; }
 #endif
 
 typedef struct ARMVectorReg {
@@ -915,6 +920,18 @@ struct ARMCPU {
 
     /* Used to set the maximum vector length the cpu will support.  */
     uint32_t sve_max_vq;
+
+    /*
+     * In sve_vq_map each set bit is a supported vector length of
+     * (bit-number + 1) * 16 bytes, i.e. each bit number + 1 is the vector
+     * length in quadwords.
+     *
+     * While processing properties during initialization, corresponding
+     * sve_vq_init bits are set for bits in sve_vq_map that have been
+     * set by properties.
+     */
+    DECLARE_BITMAP(sve_vq_map, ARM_MAX_VQ);
+    DECLARE_BITMAP(sve_vq_init, ARM_MAX_VQ);
 };
 
 void arm_cpu_post_init(Object *obj);
@@ -1834,6 +1851,8 @@ static inline int arm_feature(CPUARMState *env, int feature)
     return (env->features & (1ULL << feature)) != 0;
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp);
+
 #if !defined(CONFIG_USER_ONLY)
 /* Return true if exception levels below EL3 are in secure state,
  * or would be following an exception return to that level.
target/arm/cpu64.c +27 -0
index 8cdb0c79fa7a..606e3eceb9c0 100644
@@ -256,11 +256,152 @@ static void aarch64_a72_initfn(Object *obj)
     define_arm_cp_regs(cpu, cortex_a72_a57_a53_cp_reginfo);
 }
 
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
+{
+    /*
+     * If any vector lengths are explicitly enabled with sve<N> properties,
+     * then all other lengths are implicitly disabled.  If sve-max-vq is
+     * specified then it is the same as explicitly enabling all lengths
+     * up to and including the specified maximum, which means all larger
+     * lengths will be implicitly disabled.  If no sve<N> properties
+     * are enabled and sve-max-vq is not specified, then all lengths not
+     * explicitly disabled will be enabled.  Additionally, all power-of-two
+     * vector lengths less than the maximum enabled length will be
+     * automatically enabled and all vector lengths larger than the largest
+     * disabled power-of-two vector length will be automatically disabled.
+     * Errors are generated if the user provided input that interferes with
+     * any of the above.  Finally, if SVE is not disabled, then at least one
+     * vector length must be enabled.
+     */
+    DECLARE_BITMAP(tmp, ARM_MAX_VQ);
+    uint32_t vq, max_vq = 0;
+
+    /*
+     * Process explicit sve<N> properties.
+     * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
+     * Check first for any sve<N> enabled.
+     */
+    if (!bitmap_empty(cpu->sve_vq_map, ARM_MAX_VQ)) {
+        max_vq = find_last_bit(cpu->sve_vq_map, ARM_MAX_VQ) + 1;
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 292

Quoted 7 lines
target/arm/cpu64.c +7 -0
+
+        if (cpu->sve_max_vq && max_vq > cpu->sve_max_vq) {
+            error_setg(errp, "cannot enable sve%d", max_vq * 128);
+            error_append_hint(errp, "sve%d is larger than the maximum vector "
+                              "length, sve-max-vq=%d (%d bits)\n",
+                              max_vq * 128, cpu->sve_max_vq,
+                              cpu->sve_max_vq * 128);

you could test directly if max_vq < cpu->sve_max_vq too which couldn't

Quoted 3 lines

That's not an error. Here max_vq is the maximum enabled with sve<N> properties. We haven't considered cpu->sve_max_vq yet, except for ensuring we don't enable something larger than it.

Oh ok I had in mind sve512=on,sve-max-vq=5 was an error but reading the code again it is indeed value. My original thought was setting sve<n> was disabling larger values but that's only valid with sve-max-vq unset.

Quoted 80 lines
Quoted 71 lines

be correct either. But as far as I understand this is caught in *

Quoted 69 lines
target/arm/cpu64.c +69 -0
+            return;
+        }
+
+        /* Propagate enabled bits down through required powers-of-two. */
+        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                set_bit(vq - 1, cpu->sve_vq_map);
+            }
+        }
+    } else if (cpu->sve_max_vq == 0) {
+        /*
+         * No explicit bits enabled, and no implicit bits from sve-max-vq.
+         */
+        if (!cpu_isar_feature(aa64_sve, cpu)) {
+            /* SVE is disabled and so are all vector lengths.  Good. */
+            return;
+        }
+
+        /* Disabling a power-of-two disables all larger lengths. */
+        if (test_bit(0, cpu->sve_vq_init)) {> +            error_setg(errp, "cannot disable sve128");
+            error_append_hint(errp, "Disabling sve128 results in all vector "
+                              "lengths being disabled.\n");
+            error_append_hint(errp, "With SVE enabled, at least one vector "
+                              "length must be enabled.\n");
+            return;
+        }
+        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
+            if (test_bit(vq - 1, cpu->sve_vq_init)) {
+                break;
+            }
+        }
+        max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+
+        bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
+        max_vq = find_last_bit(cpu->sve_vq_map, max_vq) + 1;
+    }
+
+    /*
+     * Process the sve-max-vq property.
+     * Note that we know from the above that no bit above
+     * sve-max-vq is currently set.
+     */
+    if (cpu->sve_max_vq != 0) {
+        max_vq = cpu->sve_max_vq;
+
+        if (!test_bit(max_vq - 1, cpu->sve_vq_map) &&
+            test_bit(max_vq - 1, cpu->sve_vq_init)) {
+            error_setg(errp, "cannot disable sve%d", max_vq * 128);
+            error_append_hint(errp, "The maximum vector length must be "
+                              "enabled, sve-max-vq=%d (%d bits)\n",
+                              max_vq, max_vq * 128);
+            return;
+        }
+
+        /* Set all bits not explicitly set within sve-max-vq. */
+        bitmap_complement(tmp, cpu->sve_vq_init, max_vq);
+        bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+    }
+
+    /*
+     * We should know what max-vq is now.  Also, as we're done
+     * manipulating sve-vq-map, we ensure any bits above max-vq
+     * are clear, just in case anybody looks.
+     */
+    assert(max_vq != 0);
+    bitmap_clear(cpu->sve_vq_map, max_vq, ARM_MAX_VQ - max_vq);
+
+    /* Ensure all required powers-of-two are enabled. */
+    for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 368

Quoted 6 lines
target/arm/cpu64.c +6 -0
+        if (!test_bit(vq - 1, cpu->sve_vq_map)) {
+            error_setg(errp, "cannot disable sve%d", vq * 128);
+            error_append_hint(errp, "sve%d is required as it "
+                              "is a power-of-two length smaller than "
+                              "the maximum, sve%d\n",
+                              vq * 128, max_vq * 128);

* I think this will also catches the case where ,sve512=on,sve-max-vq=5 where the end-user did not explicitly disabled anything

Quoted 3 lines

As stated above, 'sve512=on,sve-max-vq=5' isn't an error. It doesn't hurt to enable vector lengths explicitly that would be enabled automatically anyway with the sve-max-vq property.

Now understood.

Quoted 370 lines
Quoted 24 lines
Quoted 24 lines
target/arm/cpu64.c +22 -0
+            return;
+        }
+    }
+
+    /* From now on sve_max_vq is the actual maximum supported length. */
+    cpu->sve_max_vq = max_vq;
+}
+
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{
+    uint32_t bitnum;
+
+    /*
+     * We allow vq == ARM_MAX_VQ + 1 to be input because the caller may want
+     * to find the maximum vq enabled, which may be ARM_MAX_VQ, but this
+     * function always returns the next smaller than the input.
+     */
+    assert(vq && vq <= ARM_MAX_VQ + 1);
+
+    bitnum = find_last_bit(cpu->sve_vq_map, vq - 1);
+    return bitnum == vq - 1 ? 0 : bitnum + 1;
+}
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 396

Quoted 7 lines
target/arm/cpu64.c +2 -1
+
 static void cpu_max_get_sve_max_vq(Object *obj, Visitor *v, const char *name,
                                    void *opaque, Error **errp)
 {
     ARMCPU *cpu = ARM_CPU(obj);
-    visit_type_uint32(v, name, &cpu->sve_max_vq, errp);
+    uint32_t value;

Shouldn't it be part of the previous patch?

Yes. Will fix for v5.

Quoted 10 lines
Quoted 10 lines
target/arm/cpu64.c +8 -0
+
+    /* All vector lengths are disabled when SVE is off. */
+    if (!cpu_isar_feature(aa64_sve, cpu)) {
+        value = 0;
+    } else {
+        value = cpu->sve_max_vq;
+    }
+    visit_type_uint32(v, name, &value, errp);
 }
 
Quoted 4 lines

target/arm/cpu64.c static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,

Quoted 2 lines
target/arm/cpu64.c +0 -0
 static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,
@@ -279,6 +420,44 @@ static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,

Independently on this patch I noticed that if sve_max_vq is out of scope, an error is propagated but still the cpu->sve-max_vq is set.

I'll sneak that fix into patch 8/9 where we touch the function anyway.

Quoted 309 lines
Quoted 309 lines
target/arm/cpu64.c +46 -1
     error_propagate(errp, err);
 }
 
+static void cpu_arm_get_sve_vq(Object *obj, Visitor *v, const char *name,
+                               void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq = atoi(&name[3]) / 128;
+    bool value;
+
+    /* All vector lengths are disabled when SVE is off. */
+    if (!cpu_isar_feature(aa64_sve, cpu)) {
+        value = false;
+    } else {
+        value = test_bit(vq - 1, cpu->sve_vq_map);
+    }
+    visit_type_bool(v, name, &value, errp);
+}
+
+static void cpu_arm_set_sve_vq(Object *obj, Visitor *v, const char *name,
+                               void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq = atoi(&name[3]) / 128;
+    Error *err = NULL;
+    bool value;
+
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
+
+    if (value) {
+        set_bit(vq - 1, cpu->sve_vq_map);
+    } else {
+        clear_bit(vq - 1, cpu->sve_vq_map);
+    }
+    set_bit(vq - 1, cpu->sve_vq_init);
+}
+
 static void cpu_arm_get_sve(Object *obj, Visitor *v, const char *name,
                             void *opaque, Error **errp)
 {
@@ -315,6 +494,7 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
 static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
@@ -418,11 +598,17 @@ static void aarch64_max_initfn(Object *obj)
         cpu->dcz_blocksize = 7; /*  512 bytes */
 #endif
 
-        cpu->sve_max_vq = ARM_MAX_VQ;
         object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                             cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
         object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
                             cpu_arm_set_sve, NULL, NULL, &error_fatal);
+
+        for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+            char name[8];
+            sprintf(name, "sve%d", vq * 128);
+            object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                                cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+        }
     }
 }
 
target/arm/helper.c +10 -3
index 507026c9154b..f33284c247d5 100644
@@ -5351,6 +5351,13 @@ int sve_exception_el(CPUARMState *env, int el)
     return 0;
 }
 
+static uint32_t sve_zcr_get_valid_len(ARMCPU *cpu, uint32_t start_len)
+{
+    uint32_t start_vq = (start_len & 0xf) + 1;
+
+    return arm_cpu_vq_map_next_smaller(cpu, start_vq + 1) - 1;
+}
+
 /*
  * Given that SVE is enabled, return the vector length for EL.
  */
@@ -5360,13 +5367,13 @@ uint32_t sve_zcr_len_for_el(CPUARMState *env, int el)
     uint32_t zcr_len = cpu->sve_max_vq - 1;
 
     if (el <= 1) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[1]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[1]);
     }
     if (el <= 2 && arm_feature(env, ARM_FEATURE_EL2)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[2]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[2]);
     }
     if (arm_feature(env, ARM_FEATURE_EL3)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[3]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[3]);
     }
     return zcr_len;
 }
target/arm/monitor.c +12 -0
index 4fddb6c252a3..e912ed2cefa0 100644
@@ -90,6 +90,8 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
     return head;
 }
 
+QEMU_BUILD_BUG_ON(ARM_MAX_VQ > 16);
+
 /*
  * These are cpu model features we want to advertise. The order here
  * matters as this is the order in which qmp_query_cpu_model_expansion
@@ -98,6 +100,9 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
  */
 static const char *cpu_model_advertised_features[] = {
     "aarch64", "pmu", "sve",
+    "sve128", "sve256", "sve384", "sve512",
+    "sve640", "sve768", "sve896", "sve1024", "sve1152", "sve1280",
+    "sve1408", "sve1536", "sve1664", "sve1792", "sve1920", "sve2048",
     NULL
 };
 
@@ -185,6 +190,9 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
         if (!err) {
             visit_check_struct(visitor, &err);
         }
+        if (!err) {
+            arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        }
         visit_end_struct(visitor, NULL);
         visit_free(visitor);
         if (err) {
@@ -192,6 +200,10 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
             error_propagate(errp, err);
             return NULL;
         }
+    } else {
+        Error *err = NULL;
+        arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        assert(err == NULL);
     }
 
     expansion_info = g_new0(CpuModelExpansionInfo, 1);
tests/arm-cpu-features.c +144 -0
index 202bc0e3e823..9a2dd402b769 100644
@@ -13,6 +13,18 @@
 #include "qapi/qmp/qdict.h"
 #include "qapi/qmp/qjson.h"
 
+#if __SIZEOF_LONG__ == 8
+#define BIT(n) (1UL << (n))
+#else
+#define BIT(n) (1ULL << (n))
+#endif
+
+/*
+ * We expect the SVE max-vq to be 16. Also it must be <= 64
+ * for our test code, otherwise 'vls' can't just be a uint64_t.
+ */
+#define SVE_MAX_VQ 16
+
 #define MACHINE    "-machine virt,gic-version=max "
 #define QUERY_HEAD "{ 'execute': 'query-cpu-model-expansion', " \
                      "'arguments': { 'type': 'full', "
@@ -157,6 +169,173 @@ static void assert_bad_props(QTestState *qts, const char *cpu_type)
     qobject_unref(resp);
 }
 
+static uint64_t resp_get_sve_vls(QDict *resp)
+{
+    QDict *props;
+    const QDictEntry *e;
+    uint64_t vls = 0;
+    int n = 0;
+
+    g_assert(resp);
+    g_assert(resp_has_props(resp));
+
+    props = resp_get_props(resp);
+
+    for (e = qdict_first(props); e; e = qdict_next(props, e)) {
+        if (strlen(e->key) > 3 && !strncmp(e->key, "sve", 3) &&
+            g_ascii_isdigit(e->key[3])) {
+            char *endptr;
+            int bits;
+
+            bits = g_ascii_strtoll(&e->key[3], &endptr, 10);
+            if (!bits || *endptr != '\0') {
+                continue;
+            }
+
+            if (qdict_get_bool(props, e->key)) {
+                vls |= BIT((bits / 128) - 1);
+            }
+            ++n;
+        }
+    }
+
+    g_assert(n == SVE_MAX_VQ);
+
+    return vls;
+}
+
+#define assert_sve_vls(qts, cpu_type, expected_vls, fmt, ...)          \
+({                                                                     \
+    QDict *_resp = do_query(qts, cpu_type, fmt, ##__VA_ARGS__);        \
+    g_assert(_resp);                                                   \
+    g_assert(resp_has_props(_resp));                                   \
+    g_assert(resp_get_sve_vls(_resp) == expected_vls);                 \
+    qobject_unref(_resp);                                              \
+})
+
+static void sve_tests_default(QTestState *qts, const char *cpu_type)
+{
+    /*
+     * With no sve-max-vq or sve<N> properties on the command line
+     * the default is to have all vector lengths enabled. This also
+     * tests that 'sve' is 'on' by default.
+     */
+    assert_sve_vls(qts, cpu_type, BIT(SVE_MAX_VQ) - 1, NULL);
+
+    /* With SVE off, all vector lengths should also be off. */
+    assert_sve_vls(qts, cpu_type, 0, "{ 'sve': false }");
+
+    /* With SVE on, we must have at least one vector length enabled. */
+    assert_error(qts, cpu_type, "cannot disable sve128", "{ 'sve128': false }");
+
+    /*
+     * ---------------------------------------------------------------------
+     *               power-of-two(vq)   all-power-            can      can
+     *                                  of-two(< vq)        enable   disable
+     * ---------------------------------------------------------------------
+     * vq < max_vq      no                MUST*              yes      yes
+     * vq < max_vq      yes               MUST*              yes      no
+     * ---------------------------------------------------------------------
+     * vq == max_vq     n/a               MUST*              yes**    yes**
+     * ---------------------------------------------------------------------
+     * vq > max_vq      n/a               no                 no       yes
+     * vq > max_vq      n/a               yes                yes      yes
+     * ---------------------------------------------------------------------
+     *
+     * [*] "MUST" means this requirement must already be satisfied,
+     *     otherwise 'max_vq' couldn't itself be enabled.
+     *
+     * [**] Not testable with the QMP interface, only with the command line.
+     */
+
+    /* max_vq := 8 */
+    assert_sve_vls(qts, cpu_type, 0x8b, "{ 'sve1024': true }");
+
+    /* max_vq := 8, vq < max_vq, !power-of-two(vq) */
+    assert_sve_vls(qts, cpu_type, 0x8f,
+                   "{ 'sve1024': true, 'sve384': true }");
+    assert_sve_vls(qts, cpu_type, 0x8b,
+                   "{ 'sve1024': true, 'sve384': false }");
+
+    /* max_vq := 8, vq < max_vq, power-of-two(vq) */
+    assert_sve_vls(qts, cpu_type, 0x8b,
+                   "{ 'sve1024': true, 'sve256': true }");
+    assert_error(qts, cpu_type, "cannot disable sve256",
+                 "{ 'sve1024': true, 'sve256': false }");
+
+    /* max_vq := 3, vq > max_vq, !all-power-of-two(< vq) */
+    assert_error(qts, cpu_type, "cannot disable sve512",
+                 "{ 'sve384': true, 'sve512': false, 'sve640': true }");
+
+    /*
+     * We can disable power-of-two vector lengths when all larger lengths
+     * are also disabled. We only need to disable the power-of-two length,
+     * as all non-enabled larger lengths will then be auto-disabled.
+     */
+    assert_sve_vls(qts, cpu_type, 0x7, "{ 'sve512': false }");
+
+    /* max_vq := 3, vq > max_vq, all-power-of-two(< vq) */
+    assert_sve_vls(qts, cpu_type, 0x1f,
+                   "{ 'sve384': true, 'sve512': true, 'sve640': true }");
+    assert_sve_vls(qts, cpu_type, 0xf,
+                   "{ 'sve384': true, 'sve512': true, 'sve640': false }");
+}
+
+static void sve_tests_sve_max_vq_8(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-cpu max,sve-max-vq=8");
+
+    assert_sve_vls(qts, "max", BIT(8) - 1, NULL);
+
+    /*
+     * Disabling the max-vq set by sve-max-vq is not allowed, but
+     * of course enabling it is OK.
+     */
+    assert_error(qts, "max", "cannot disable sve1024", "{ 'sve1024': false }");
+    assert_sve_vls(qts, "max", 0xff, "{ 'sve1024': true }");
+
+    /*
+     * Enabling anything larger than max-vq set by sve-max-vq is not
+     * allowed, but of course disabling everything larger is OK.
+     */
+    assert_error(qts, "max", "cannot enable sve1152", "{ 'sve1152': true }");
Quoted 8 lines

tests/arm-cpu-features.c static void assert_bad_props(QTestState *qts, const char *cpu_type) · line 310

Quoted 7 lines
tests/arm-cpu-features.c +7 -0
+    assert_sve_vls(qts, "max", 0xff, "{ 'sve1152': false }");
+
+    /*
+     * We can disable non power-of-two lengths smaller than the max-vq
+     * set by sve-max-vq, but not power-of-two lengths.
+     */
+    assert_sve_vls(qts, "max", 0xfb, "{ 'sve384': false }");

you can also test ,sve384=on => 0x7

Quoted 2 lines

We wouldn't get 0x7, we'd get 0xff. I could still add the test to ensure we can enable smaller vector lengths without an error though.

Oh yes because you have the sve-max-vq=8
I rather meant in sve_tests_default then

Eric

Quoted 68 lines
Quoted 62 lines
Quoted 50 lines
tests/arm-cpu-features.c +31 -0
+    assert_error(qts, "max", "cannot disable sve256", "{ 'sve256': false }");
+
+    qtest_quit(qts);
+}
+
+static void sve_tests_sve_off(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-cpu max,sve=off");
+
+    /* SVE is off, so the map should be empty. */
+    assert_sve_vls(qts, "max", 0, NULL);
+
+    /* The map stays empty even if we turn lengths on or off. */
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': true }");
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+
+    /* With SVE re-enabled we should get all vector lengths enabled. */
+    assert_sve_vls(qts, "max", BIT(SVE_MAX_VQ) - 1, "{ 'sve': true }");
+
+    /* Or enable SVE with just specific vector lengths. */
+    assert_sve_vls(qts, "max", 0x3,
+                   "{ 'sve': true, 'sve128': true, 'sve256': true }");
+
+    qtest_quit(qts);
+}
+
 static void test_query_cpu_model_expansion(const void *data)
 {
     QTestState *qts;
@@ -180,9 +359,12 @@ static void test_query_cpu_model_expansion(const void *data)
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
         assert_has_feature(qts, "max", "aarch64");
         assert_has_feature(qts, "max", "sve");
+        assert_has_feature(qts, "max", "sve128");
         assert_has_feature(qts, "cortex-a57", "pmu");
         assert_has_feature(qts, "cortex-a57", "aarch64");
 
+        sve_tests_default(qts, "max");
+
         /* Test that features that depend on KVM generate errors without. */
         assert_error(qts, "max",
                      "'aarch64' feature cannot be disabled "
@@ -234,6 +416,13 @@ int main(int argc, char **argv)
     qtest_add_data_func("/arm/query-cpu-model-expansion",
                         NULL, test_query_cpu_model_expansion);
 

tests/arm-cpu-features.c int main(int argc, char **argv) · line 428

Quoted 10 lines
tests/arm-cpu-features.c +7 -0
+    if (g_str_equal(qtest_get_arch(), "aarch64")) {
+        qtest_add_data_func("/arm/max/query-cpu-model-expansion/sve-max-vq-8",
+                            NULL, sve_tests_sve_max_vq_8);
+        qtest_add_data_func("/arm/max/query-cpu-model-expansion/sve-off",
+                            NULL, sve_tests_sve_off);
+    }
+
     if (kvm_available) {
         qtest_add_data_func("/arm/kvm/query-cpu-model-expansion",
                             NULL, test_query_cpu_model_expansion_kvm);

Thanks

Quoted 2 lines

Eric

Thanks,
drew

#11 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

Quoted 67 lines · Auger Eric
Quoted 56 lines · Andrew Jones
Quoted 38 lines · Auger Eric

Hi Drew,

Quoted 35 lines · Andrew Jones

Introduce cpu properties to give fine control over SVE vector lengths. We introduce a property for each valid length up to the current maximum supported, which is 2048-bits. The properties are named, e.g. sve128, sve256, sve384, sve512, ..., where the number is the number of bits. See the updates to docs/arm-cpu-features.rst for a description of the semantics and for example uses.

Note, as sve-max-vq is still present and we'd like to be able to support qmp_query_cpu_model_expansion with guests launched with e.g. -cpu max,sve-max-vq=8 on their command lines, then we do allow sve-max-vq and sve<N> properties to be provided at the same time, but this is not recommended, and is why sve-max-vq is not mentioned in the document. If sve-max-vq is provided then it enables all lengths smaller than and including the max and disables all lengths larger. It also has the side-effect that no larger lengths may be enabled and that the max itself cannot be disabled. Smaller non-power-of-two lengths may, however, be disabled, e.g. -cpu max,sve-max-vq=4,sve384=off provides a guest the vector lengths 128, 256, and 512 bits.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst | 157 +++++++++++++++++++++++++++++--
 target/arm/cpu.c          |  19 ++++
 target/arm/cpu.h          |  19 ++++
 target/arm/cpu64.c        | 190 +++++++++++++++++++++++++++++++++++++-
 target/arm/helper.c       |  13 ++-
 target/arm/monitor.c      |  12 +++
 tests/arm-cpu-features.c  | 189 +++++++++++++++++++++++++++++++++++++
 7 files changed, 587 insertions(+), 12 deletions(-)
Quoted 16 lines
Quoted 5 lines
docs/arm-cpu-features.rst +0 -0
index c79dcffb5556..1262fddc6201 100644

docs/arm-cpu-features.rst block in the script for usage) is used to issue the QMP commands. · line 55

Quoted 9 lines
docs/arm-cpu-features.rst +5 -1
@@ -48,18 +48,28 @@ block in the script for usage) is used to issue the QMP commands.
       (QEMU) query-cpu-model-expansion type=full model={"name":"max"}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": true, "aarch64": true
+        "sve1664": true, "pmu": true, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true

I don't really understand why this gets printed unsorted as you registered them in the logical order, any clue?

Quoted 2 lines

Maybe qdict hashing? qmp-shell could probably be patched to sort them, but it's probably not worth it.

OK. Anyway beyond the scope of the series but that's unpractical.

Quoted 5 lines
Quoted 4 lines

Note: This example is given with TCG or KVM with all those vectors supported by the host.

Quoted 6 lines
Quoted 3 lines

You want this note to be added? I don't think it's necessary, as it's just an example. A different configuration or different CPU model expansion will have different results, as we explain exhaustively in this document.

a reader may understand that along with max cpu type whatever the state of KVM all sve<N> are supported by default, which is not true in KVM case. Maybe add TCG case.

I'll mention above where we have "we started QEMU with qemu-system-aarch64" that it's using TCG. If I can find some place where it doesn't break the flow too much, then I'll add a statement that says all lengths being 'true' is unlikely when KVM is in use.

Quoted 63 lines
Quoted 63 lines
Quoted 62 lines
Quoted 49 lines
docs/arm-cpu-features.rst +25 -4
       }}}}
 
-We see that the `max` CPU type has the `pmu` and `aarch64` CPU features.
-We also see that the CPU features are enabled, as they are all `true`.
+We see that the `max` CPU type has the `pmu`, `aarch64`, `sve`, and many
+`sve<N>` CPU features.  We also see that all the CPU features are
+enabled, as they are all `true`.  (The `sve<N>` CPU features are all
+optional SVE vector lengths.  See "SVE CPU Properties".)
 
 (2) Let's try to disable the PMU::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"pmu":false}}
       { "return": {
         "model": { "name": "max", "props": {
-        "pmu": false, "aarch64": true
+        "sve1664": true, "pmu": false, "sve1792": true, "sve1920": true,
+        "sve128": true, "aarch64": true, "sve1024": true, "sve": true,
+        "sve640": true, "sve768": true, "sve1408": true, "sve256": true,
+        "sve1152": true, "sve512": true, "sve384": true, "sve1536": true,
+        "sve896": true, "sve1280": true, "sve2048": true
       }}}}
 
 We see it worked, as `pmu` is now `false`.
@@ -75,7 +85,22 @@ We see it worked, as `pmu` is now `false`.
 It looks like this feature is limited to a configuration we do not
 currently have.
 
-(4) Let's try probing CPU features for the Cortex-A15 CPU type::
+(4) Let's disable `sve` and see what happens to all the optional SVE
+    vector lengths::
+
+      (QEMU) query-cpu-model-expansion type=full model={"name":"max","props":{"sve":false}}
+      { "return": {
+        "model": { "name": "max", "props": {
+        "sve1664": false, "pmu": true, "sve1792": false, "sve1920": false,
+        "sve128": false, "aarch64": true, "sve1024": false, "sve": false,
+        "sve640": false, "sve768": false, "sve1408": false, "sve256": false,
+        "sve1152": false, "sve512": false, "sve384": false, "sve1536": false,
+        "sve896": false, "sve1280": false, "sve2048": false
+      }}}}
+
+As expected they are now all `false`.
+
+(5) Let's try probing CPU features for the Cortex-A15 CPU type::
 
       (QEMU) query-cpu-model-expansion type=full model={"name":"cortex-a15"}
       {"return": {"model": {"name": "cortex-a15", "props": {"pmu": true}}}}

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 164

Quoted 10 lines
docs/arm-cpu-features.rst +6 -1
@@ -131,7 +156,125 @@ After determining which CPU features are available and supported for a
 given CPU type, then they may be selectively enabled or disabled on the
 QEMU command line with that CPU type::
 
-  $ qemu-system-aarch64 -M virt -cpu max,pmu=off
+  $ qemu-system-aarch64 -M virt -cpu max,pmu=off,sve=on,sve128=on,sve256=on
+
+The example above disables the PMU and enables the first two SVE vector
+lengths for the `max` CPU type.  Note, the `sve=on` isn't actually
+necessary, because, as we observed above with our probe of the `max` CPU
+type, `sve` is already on by default.

This holds with TCG and not KVM.

Quoted 5 lines
Quoted 1 line

It also holds with KVM, as long as the host supports SVE.

as long as the hist supports SVE: Yes that's what I meant

maybe add, as we observed abovewith our probe of the max CPU

Quoted 1 line
Quoted 1 line
Quoted 1 line
docs/arm-cpu-features.rst +1 -0
+type, in this TCG case, `sve` is already on by default

But "in this TCG case, sve is already on by default" implies that it won't be on by default with KVM. However it will be, when KVM supports the feature. I don't like adding "but ..., if ..."'s everywhere because then readers can't get through a paragraph without having to double think everything. We're presenting a single example here, so we shouldn't try to introduce another example (the same thing, but with KVM, when KVM supports the feature) in its description. If you think we should add another example for KVM, then we should do it separately. I'd argue it's not really necessary to do though. Although a couple sentences at the top that emphasize this is a single TCG example, and with KVM things are different, might be good.

Quoted 60 lines

But that doesn't

Quoted 38 lines

matter, because we're speaking specifically about the example above here, where we see that sve=on is the default.

Quoted 26 lines

Also, based on our probe of

Quoted 24 lines
docs/arm-cpu-features.rst +24 -0
+defaults, it would seem we need to disable many SVE vector lengths, rather
+than only enabling the two we want.  This isn't the case, because, as
+disabling many SVE vector lengths would be quite verbose, the `sve<N>` CPU
+properties have special semantics (see "SVE CPU Property Parsing
+Semantics").
+
+SVE CPU Properties
+==================
+
+There are two types of SVE CPU properties: `sve` and `sve<N>`.  The first
+is used to enable or disable the entire SVE feature, just as the `pmu`
+CPU property completely enables or disables the PMU.  The second type
+is used to enable or disable specific vector lengths, where `N` is the
+number of bits of the length.  The `sve<N>` CPU properties have special
+dependencies and constraints, see "SVE CPU Property Dependencies and
+Constraints" below.  Additionally, as we want all supported vector lengths
+to be enabled by default, then, in order to avoid overly verbose command
+lines (command lines full of `sve<N>=off`, for all `N` not wanted), we
+provide the parsing semantics listed in "SVE CPU Property Parsing
+Semantics".
+
+SVE CPU Property Dependencies and Constraints
+---------------------------------------------
+
Quoted 8 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 194

Quoted 5 lines
docs/arm-cpu-features.rst +5 -0
+  1) At least one vector length must be enabled when `sve` is enabled.> +
+  2) If a vector length `N` is enabled, then all power-of-two vector
+     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
+     is enabled, then `sve128` and `sve256` must also be enabled,
+     but `sve384` is not required.

I would remove the eg. part. reading that I tend to understand that the user must pass ,sve128=on, sve256=on and sve512=on whereas example 6 says only sve512=on can be set and other lower ^2 values are auto-enabled.

Quoted 2 lines

I'll keep the example, but instead of using the property names, I'll change to using "the 128-bit vector length", 256-bit ..., etc.

ok

Quoted 17 lines
Quoted 16 lines
Quoted 4 lines
docs/arm-cpu-features.rst +3 -0
+
+SVE CPU Property Parsing Semantics
+----------------------------------

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 207

Quoted 10 lines
docs/arm-cpu-features.rst +10 -0
+
+  1) If SVE is disabled (`sve=off`), then which SVE vector lengths
+     are enabled or disabled is irrelevant to the guest, as the entire
+     SVE feature is disabled and that disables all vector lengths for
+     the guest.  However QEMU will still track any `sve<N>` CPU
+     properties provided by the user.  If later an `sve=on` is provided,
+     then the guest will get only the enabled lengths.
+
+  2) If SVE is enabled (`sve=on`), but no `sve<N>` CPU properties are
+     provided, then all supported vector lengths are enabled.

I understand from the code it also includes non ^2 values which is not obvious.

Quoted 4 lines
Quoted 3 lines

Yes, the default is to have *all* supported vector lengths enabled, not just the power-of-two. As we already use the word 'all' here, then how can we further clarify this?

I would simply add "including non ^2 values".

OK, but these precise descriptions get pretty long, especially when you add KVM. For TCG it's just 'including the non power-of-two lengths', but then when we add KVM, which from QEMU's point of view decides what's supported and power-of-two's have nothing to do with it, we get

...all supported vector lengths, which when KVM is not in use means including the non power-of-two lengths, and, when KVM is in use, it means all vector lengths supported by the host processor.

I'll add that for v5. I just hope people don't end up always 'tl;dr'ing this document...

Quoted 693 lines
Quoted 16 lines
Quoted 6 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 211

Quoted 5 lines
docs/arm-cpu-features.rst +4 -0
+
+  3) If SVE is enabled, then an error is generated when attempting to
+     disable the last enabled vector length (see constraint (1) of "SVE
+     CPU Property Dependencies and Constraints").

the same happens if you attempt to disabled any lower ^2 value

That should be another bullet, which I appear to be missing, and will add.

Quoted 7 lines
Quoted 7 lines
docs/arm-cpu-features.rst +7 -0
+
+  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+     CPU properties are set `on`, then the specified vector lengths are
+     disabled but the default for any unspecified lengths remains enabled.
+     Disabling a power-of-two vector length also disables all vector
+     lengths larger than the power-of-two length (see constraint (2) of
+     "SVE CPU Property Dependencies and Constraints").
Quoted 57 lines
Quoted 4 lines

So rephasing it:
disabling a non ^2 value only disable that one disabling a ^2 value disables all larger lengths too is that correct?

That's correct. We auto-disable all uninitialized larger lengths when a power-of-two dependency length is explicitly disabled.

Quoted 49 lines
Quoted 33 lines
docs/arm-cpu-features.rst +32 -0
+
+  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+     are enabled and all unspecified lengths default to disabled, except
+     for the required lengths per constraint (2) of "SVE CPU Property
+     Dependencies and Constraints", which will even be auto-enabled if
+     they were not explicitly enabled.
+
+  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+     explicitly disabled (i.e. avoiding the error specified in (3) of
+     "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
+     provided an error will be generated.  To avoid this error, one must
+     enable at least one vector length prior to enabling SVE.
+
+SVE CPU Property Examples
+-------------------------
+
+  1) Disable SVE::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off
+
+  2) Implicitly enable all vector lengths for the `max` CPU type::
+
+     $ qemu-system-aarch64 -M virt -cpu max
+
+  3) Only enable the 128-bit vector length::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on
+
+  4) Disable the 256-bit vector length and all larger vector lengths
+     since 256 is a power-of-two (this results in only the 128-bit length
+     being enabled)::
+

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 261

Quoted 10 lines
docs/arm-cpu-features.rst +10 -0
+     $ qemu-system-aarch64 -M virt -cpu max,sve256=off
+
+  5) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
+
+  6) The same as (5), but since the 128-bit and 256-bit vector
+     lengths are required for the 512-bit vector length to be enabled,
+     then allow them to be auto-enabled::> +
+     $ qemu-system-aarch64 -M virt -cpu max,sve512=on

You should also document

$ qemu-system-aarch64 -M virt -cpu max,sve512=off
as in that case this result in 128, 256 and 384. the fact you get non ^2
values is not obvious here because if you enable a ^2 value you onlt get
<= ^2 values.

yep.

Quoted 159 lines

I can change (4) above to use 512 instead of 256.

Quoted 6 lines

docs/arm-cpu-features.rst After determining which CPU features are available and supported for a · line 267

Quoted 5 lines
docs/arm-cpu-features.rst +5 -0
+
+  7) Do the same as (6), but by first disabling SVE and then re-enabling it::
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on> +
+  8) Force errors regarding the last vector length::

You mean those commands will generate errors, right?

Yes

Quoted 30 lines
Quoted 19 lines
docs/arm-cpu-features.rst +11 -1
+
+     $ qemu-system-aarch64 -M virt -cpu max,sve128=off
+     $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
+
+SVE CPU Property Recommendations
+--------------------------------
 
-The example above disables the PMU for the `max` CPU type.
+The examples in "SVE CPU Property Examples" exhibit many ways to select
+vector lengths which developers may find useful in order to avoid overly
+verbose command lines.  However, the recommended way to select vector
+lengths is to explicitly enable each desired length.  Therefore only
+example's (1), (3), and (5) exhibit recommended uses of the properties.
 
target/arm/cpu.c +0 -0
index 73be2ebfdd39..522fed95b339 100644

target/arm/cpu.c static void arm_cpu_finalizefn(Object *obj) · line 1208

Quoted 10 lines
target/arm/cpu.c +7 -0
@@ -1199,6 +1199,19 @@ static void arm_cpu_finalizefn(Object *obj)
 #endif
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp)
+{
+    Error *local_err = NULL;
+
+    if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
+        arm_cpu_sve_finalize(cpu, &local_err);
+        if (local_err != NULL) {

nit: !local_err

Quoted 3 lines

target/arm/cpu.c static void arm_cpu_finalizefn(Object *obj) · line 1210

Quoted 2 lines
target/arm/cpu.c +2 -0
+            error_propagate(errp, local_err);
+            return;

not needed

Yet. I expect a chain of finalizers to go here, so we'll want the 'return' when we get the next one. As it doesn't hurt, it might as well go in now.

Quoted 101 lines

target/arm/cpu.c static void arm_cpu_realizefn(DeviceState *dev, Error **errp) · line 1272

Quoted 13 lines
target/arm/cpu.c +6 -0
+        }
+    }
+}
+
 static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
 {
     CPUState *cs = CPU(dev);
@@ -1255,6 +1268,12 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         return;
     }
 
+    arm_cpu_finalize_features(cpu, &local_err);
+    if (local_err != NULL) {

same

Quoted 87 lines
target/arm/cpu.c +3 -0
+        error_propagate(errp, local_err);
+        return;> +    }
+
     if (arm_feature(env, ARM_FEATURE_AARCH64) &&
         cpu->has_vfp != cpu->has_neon) {
         /*
target/arm/cpu.h +19 -0
index 297ad5e47ad8..11162484465a 100644
@@ -184,8 +184,13 @@ typedef struct {
 
 #ifdef TARGET_AARCH64
 # define ARM_MAX_VQ    16
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp);
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq);
 #else
 # define ARM_MAX_VQ    1
+static inline void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) { }
+static inline uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{ return 0; }
 #endif
 
 typedef struct ARMVectorReg {
@@ -915,6 +920,18 @@ struct ARMCPU {
 
     /* Used to set the maximum vector length the cpu will support.  */
     uint32_t sve_max_vq;
+
+    /*
+     * In sve_vq_map each set bit is a supported vector length of
+     * (bit-number + 1) * 16 bytes, i.e. each bit number + 1 is the vector
+     * length in quadwords.
+     *
+     * While processing properties during initialization, corresponding
+     * sve_vq_init bits are set for bits in sve_vq_map that have been
+     * set by properties.
+     */
+    DECLARE_BITMAP(sve_vq_map, ARM_MAX_VQ);
+    DECLARE_BITMAP(sve_vq_init, ARM_MAX_VQ);
 };
 
 void arm_cpu_post_init(Object *obj);
@@ -1834,6 +1851,8 @@ static inline int arm_feature(CPUARMState *env, int feature)
     return (env->features & (1ULL << feature)) != 0;
 }
 
+void arm_cpu_finalize_features(ARMCPU *cpu, Error **errp);
+
 #if !defined(CONFIG_USER_ONLY)
 /* Return true if exception levels below EL3 are in secure state,
  * or would be following an exception return to that level.
target/arm/cpu64.c +27 -0
index 8cdb0c79fa7a..606e3eceb9c0 100644
@@ -256,11 +256,152 @@ static void aarch64_a72_initfn(Object *obj)
     define_arm_cp_regs(cpu, cortex_a72_a57_a53_cp_reginfo);
 }
 
+void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
+{
+    /*
+     * If any vector lengths are explicitly enabled with sve<N> properties,
+     * then all other lengths are implicitly disabled.  If sve-max-vq is
+     * specified then it is the same as explicitly enabling all lengths
+     * up to and including the specified maximum, which means all larger
+     * lengths will be implicitly disabled.  If no sve<N> properties
+     * are enabled and sve-max-vq is not specified, then all lengths not
+     * explicitly disabled will be enabled.  Additionally, all power-of-two
+     * vector lengths less than the maximum enabled length will be
+     * automatically enabled and all vector lengths larger than the largest
+     * disabled power-of-two vector length will be automatically disabled.
+     * Errors are generated if the user provided input that interferes with
+     * any of the above.  Finally, if SVE is not disabled, then at least one
+     * vector length must be enabled.
+     */
+    DECLARE_BITMAP(tmp, ARM_MAX_VQ);
+    uint32_t vq, max_vq = 0;
+
+    /*
+     * Process explicit sve<N> properties.
+     * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
+     * Check first for any sve<N> enabled.
+     */
+    if (!bitmap_empty(cpu->sve_vq_map, ARM_MAX_VQ)) {
+        max_vq = find_last_bit(cpu->sve_vq_map, ARM_MAX_VQ) + 1;
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 292

Quoted 7 lines
target/arm/cpu64.c +7 -0
+
+        if (cpu->sve_max_vq && max_vq > cpu->sve_max_vq) {
+            error_setg(errp, "cannot enable sve%d", max_vq * 128);
+            error_append_hint(errp, "sve%d is larger than the maximum vector "
+                              "length, sve-max-vq=%d (%d bits)\n",
+                              max_vq * 128, cpu->sve_max_vq,
+                              cpu->sve_max_vq * 128);

you could test directly if max_vq < cpu->sve_max_vq too which couldn't

Quoted 3 lines

That's not an error. Here max_vq is the maximum enabled with sve<N> properties. We haven't considered cpu->sve_max_vq yet, except for ensuring we don't enable something larger than it.

Oh ok I had in mind sve512=on,sve-max-vq=5 was an error but reading the code again it is indeed value. My original thought was setting sve<n> was disabling larger values but that's only valid with sve-max-vq unset.

Quoted 80 lines
Quoted 71 lines

be correct either. But as far as I understand this is caught in *

Quoted 69 lines
target/arm/cpu64.c +69 -0
+            return;
+        }
+
+        /* Propagate enabled bits down through required powers-of-two. */
+        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                set_bit(vq - 1, cpu->sve_vq_map);
+            }
+        }
+    } else if (cpu->sve_max_vq == 0) {
+        /*
+         * No explicit bits enabled, and no implicit bits from sve-max-vq.
+         */
+        if (!cpu_isar_feature(aa64_sve, cpu)) {
+            /* SVE is disabled and so are all vector lengths.  Good. */
+            return;
+        }
+
+        /* Disabling a power-of-two disables all larger lengths. */
+        if (test_bit(0, cpu->sve_vq_init)) {> +            error_setg(errp, "cannot disable sve128");
+            error_append_hint(errp, "Disabling sve128 results in all vector "
+                              "lengths being disabled.\n");
+            error_append_hint(errp, "With SVE enabled, at least one vector "
+                              "length must be enabled.\n");
+            return;
+        }
+        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
+            if (test_bit(vq - 1, cpu->sve_vq_init)) {
+                break;
+            }
+        }
+        max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+
+        bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
+        max_vq = find_last_bit(cpu->sve_vq_map, max_vq) + 1;
+    }
+
+    /*
+     * Process the sve-max-vq property.
+     * Note that we know from the above that no bit above
+     * sve-max-vq is currently set.
+     */
+    if (cpu->sve_max_vq != 0) {
+        max_vq = cpu->sve_max_vq;
+
+        if (!test_bit(max_vq - 1, cpu->sve_vq_map) &&
+            test_bit(max_vq - 1, cpu->sve_vq_init)) {
+            error_setg(errp, "cannot disable sve%d", max_vq * 128);
+            error_append_hint(errp, "The maximum vector length must be "
+                              "enabled, sve-max-vq=%d (%d bits)\n",
+                              max_vq, max_vq * 128);
+            return;
+        }
+
+        /* Set all bits not explicitly set within sve-max-vq. */
+        bitmap_complement(tmp, cpu->sve_vq_init, max_vq);
+        bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+    }
+
+    /*
+     * We should know what max-vq is now.  Also, as we're done
+     * manipulating sve-vq-map, we ensure any bits above max-vq
+     * are clear, just in case anybody looks.
+     */
+    assert(max_vq != 0);
+    bitmap_clear(cpu->sve_vq_map, max_vq, ARM_MAX_VQ - max_vq);
+
+    /* Ensure all required powers-of-two are enabled. */
+    for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 368

Quoted 6 lines
target/arm/cpu64.c +6 -0
+        if (!test_bit(vq - 1, cpu->sve_vq_map)) {
+            error_setg(errp, "cannot disable sve%d", vq * 128);
+            error_append_hint(errp, "sve%d is required as it "
+                              "is a power-of-two length smaller than "
+                              "the maximum, sve%d\n",
+                              vq * 128, max_vq * 128);

* I think this will also catches the case where ,sve512=on,sve-max-vq=5 where the end-user did not explicitly disabled anything

Quoted 3 lines

As stated above, 'sve512=on,sve-max-vq=5' isn't an error. It doesn't hurt to enable vector lengths explicitly that would be enabled automatically anyway with the sve-max-vq property.

Now understood.

Quoted 370 lines
Quoted 24 lines
Quoted 24 lines
target/arm/cpu64.c +22 -0
+            return;
+        }
+    }
+
+    /* From now on sve_max_vq is the actual maximum supported length. */
+    cpu->sve_max_vq = max_vq;
+}
+
+uint32_t arm_cpu_vq_map_next_smaller(ARMCPU *cpu, uint32_t vq)
+{
+    uint32_t bitnum;
+
+    /*
+     * We allow vq == ARM_MAX_VQ + 1 to be input because the caller may want
+     * to find the maximum vq enabled, which may be ARM_MAX_VQ, but this
+     * function always returns the next smaller than the input.
+     */
+    assert(vq && vq <= ARM_MAX_VQ + 1);
+
+    bitnum = find_last_bit(cpu->sve_vq_map, vq - 1);
+    return bitnum == vq - 1 ? 0 : bitnum + 1;
+}
Quoted 8 lines

target/arm/cpu64.c static void aarch64_a72_initfn(Object *obj) · line 396

Quoted 7 lines
target/arm/cpu64.c +2 -1
+
 static void cpu_max_get_sve_max_vq(Object *obj, Visitor *v, const char *name,
                                    void *opaque, Error **errp)
 {
     ARMCPU *cpu = ARM_CPU(obj);
-    visit_type_uint32(v, name, &cpu->sve_max_vq, errp);
+    uint32_t value;

Shouldn't it be part of the previous patch?

Yes. Will fix for v5.

Quoted 10 lines
Quoted 10 lines
target/arm/cpu64.c +8 -0
+
+    /* All vector lengths are disabled when SVE is off. */
+    if (!cpu_isar_feature(aa64_sve, cpu)) {
+        value = 0;
+    } else {
+        value = cpu->sve_max_vq;
+    }
+    visit_type_uint32(v, name, &value, errp);
 }
 
Quoted 4 lines

target/arm/cpu64.c static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,

Quoted 2 lines
target/arm/cpu64.c +0 -0
 static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,
@@ -279,6 +420,44 @@ static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,

Independently on this patch I noticed that if sve_max_vq is out of scope, an error is propagated but still the cpu->sve-max_vq is set.

I'll sneak that fix into patch 8/9 where we touch the function anyway.

Quoted 317 lines
Quoted 306 lines
target/arm/cpu64.c +46 -1
     error_propagate(errp, err);
 }
 
+static void cpu_arm_get_sve_vq(Object *obj, Visitor *v, const char *name,
+                               void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq = atoi(&name[3]) / 128;
+    bool value;
+
+    /* All vector lengths are disabled when SVE is off. */
+    if (!cpu_isar_feature(aa64_sve, cpu)) {
+        value = false;
+    } else {
+        value = test_bit(vq - 1, cpu->sve_vq_map);
+    }
+    visit_type_bool(v, name, &value, errp);
+}
+
+static void cpu_arm_set_sve_vq(Object *obj, Visitor *v, const char *name,
+                               void *opaque, Error **errp)
+{
+    ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq = atoi(&name[3]) / 128;
+    Error *err = NULL;
+    bool value;
+
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
+
+    if (value) {
+        set_bit(vq - 1, cpu->sve_vq_map);
+    } else {
+        clear_bit(vq - 1, cpu->sve_vq_map);
+    }
+    set_bit(vq - 1, cpu->sve_vq_init);
+}
+
 static void cpu_arm_get_sve(Object *obj, Visitor *v, const char *name,
                             void *opaque, Error **errp)
 {
@@ -315,6 +494,7 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
 static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
+    uint32_t vq;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
@@ -418,11 +598,17 @@ static void aarch64_max_initfn(Object *obj)
         cpu->dcz_blocksize = 7; /*  512 bytes */
 #endif
 
-        cpu->sve_max_vq = ARM_MAX_VQ;
         object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                             cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
         object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
                             cpu_arm_set_sve, NULL, NULL, &error_fatal);
+
+        for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+            char name[8];
+            sprintf(name, "sve%d", vq * 128);
+            object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                                cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+        }
     }
 }
 
target/arm/helper.c +10 -3
index 507026c9154b..f33284c247d5 100644
@@ -5351,6 +5351,13 @@ int sve_exception_el(CPUARMState *env, int el)
     return 0;
 }
 
+static uint32_t sve_zcr_get_valid_len(ARMCPU *cpu, uint32_t start_len)
+{
+    uint32_t start_vq = (start_len & 0xf) + 1;
+
+    return arm_cpu_vq_map_next_smaller(cpu, start_vq + 1) - 1;
+}
+
 /*
  * Given that SVE is enabled, return the vector length for EL.
  */
@@ -5360,13 +5367,13 @@ uint32_t sve_zcr_len_for_el(CPUARMState *env, int el)
     uint32_t zcr_len = cpu->sve_max_vq - 1;
 
     if (el <= 1) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[1]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[1]);
     }
     if (el <= 2 && arm_feature(env, ARM_FEATURE_EL2)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[2]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[2]);
     }
     if (arm_feature(env, ARM_FEATURE_EL3)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[3]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[3]);
     }
     return zcr_len;
 }
target/arm/monitor.c +12 -0
index 4fddb6c252a3..e912ed2cefa0 100644
@@ -90,6 +90,8 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
     return head;
 }
 
+QEMU_BUILD_BUG_ON(ARM_MAX_VQ > 16);
+
 /*
  * These are cpu model features we want to advertise. The order here
  * matters as this is the order in which qmp_query_cpu_model_expansion
@@ -98,6 +100,9 @@ GICCapabilityList *qmp_query_gic_capabilities(Error **errp)
  */
 static const char *cpu_model_advertised_features[] = {
     "aarch64", "pmu", "sve",
+    "sve128", "sve256", "sve384", "sve512",
+    "sve640", "sve768", "sve896", "sve1024", "sve1152", "sve1280",
+    "sve1408", "sve1536", "sve1664", "sve1792", "sve1920", "sve2048",
     NULL
 };
 
@@ -185,6 +190,9 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
         if (!err) {
             visit_check_struct(visitor, &err);
         }
+        if (!err) {
+            arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        }
         visit_end_struct(visitor, NULL);
         visit_free(visitor);
         if (err) {
@@ -192,6 +200,10 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
             error_propagate(errp, err);
             return NULL;
         }
+    } else {
+        Error *err = NULL;
+        arm_cpu_finalize_features(ARM_CPU(obj), &err);
+        assert(err == NULL);
     }
 
     expansion_info = g_new0(CpuModelExpansionInfo, 1);
tests/arm-cpu-features.c +141 -0
index 202bc0e3e823..9a2dd402b769 100644
@@ -13,6 +13,18 @@
 #include "qapi/qmp/qdict.h"
 #include "qapi/qmp/qjson.h"
 
+#if __SIZEOF_LONG__ == 8
+#define BIT(n) (1UL << (n))
+#else
+#define BIT(n) (1ULL << (n))
+#endif
+
+/*
+ * We expect the SVE max-vq to be 16. Also it must be <= 64
+ * for our test code, otherwise 'vls' can't just be a uint64_t.
+ */
+#define SVE_MAX_VQ 16
+
 #define MACHINE    "-machine virt,gic-version=max "
 #define QUERY_HEAD "{ 'execute': 'query-cpu-model-expansion', " \
                      "'arguments': { 'type': 'full', "
@@ -157,6 +169,173 @@ static void assert_bad_props(QTestState *qts, const char *cpu_type)
     qobject_unref(resp);
 }
 
+static uint64_t resp_get_sve_vls(QDict *resp)
+{
+    QDict *props;
+    const QDictEntry *e;
+    uint64_t vls = 0;
+    int n = 0;
+
+    g_assert(resp);
+    g_assert(resp_has_props(resp));
+
+    props = resp_get_props(resp);
+
+    for (e = qdict_first(props); e; e = qdict_next(props, e)) {
+        if (strlen(e->key) > 3 && !strncmp(e->key, "sve", 3) &&
+            g_ascii_isdigit(e->key[3])) {
+            char *endptr;
+            int bits;
+
+            bits = g_ascii_strtoll(&e->key[3], &endptr, 10);
+            if (!bits || *endptr != '\0') {
+                continue;
+            }
+
+            if (qdict_get_bool(props, e->key)) {
+                vls |= BIT((bits / 128) - 1);
+            }
+            ++n;
+        }
+    }
+
+    g_assert(n == SVE_MAX_VQ);
+
+    return vls;
+}
+
+#define assert_sve_vls(qts, cpu_type, expected_vls, fmt, ...)          \
+({                                                                     \
+    QDict *_resp = do_query(qts, cpu_type, fmt, ##__VA_ARGS__);        \
+    g_assert(_resp);                                                   \
+    g_assert(resp_has_props(_resp));                                   \
+    g_assert(resp_get_sve_vls(_resp) == expected_vls);                 \
+    qobject_unref(_resp);                                              \
+})
+
+static void sve_tests_default(QTestState *qts, const char *cpu_type)
+{
+    /*
+     * With no sve-max-vq or sve<N> properties on the command line
+     * the default is to have all vector lengths enabled. This also
+     * tests that 'sve' is 'on' by default.
+     */
+    assert_sve_vls(qts, cpu_type, BIT(SVE_MAX_VQ) - 1, NULL);
+
+    /* With SVE off, all vector lengths should also be off. */
+    assert_sve_vls(qts, cpu_type, 0, "{ 'sve': false }");
+
+    /* With SVE on, we must have at least one vector length enabled. */
+    assert_error(qts, cpu_type, "cannot disable sve128", "{ 'sve128': false }");
+
+    /*
+     * ---------------------------------------------------------------------
+     *               power-of-two(vq)   all-power-            can      can
+     *                                  of-two(< vq)        enable   disable
+     * ---------------------------------------------------------------------
+     * vq < max_vq      no                MUST*              yes      yes
+     * vq < max_vq      yes               MUST*              yes      no
+     * ---------------------------------------------------------------------
+     * vq == max_vq     n/a               MUST*              yes**    yes**
+     * ---------------------------------------------------------------------
+     * vq > max_vq      n/a               no                 no       yes
+     * vq > max_vq      n/a               yes                yes      yes
+     * ---------------------------------------------------------------------
+     *
+     * [*] "MUST" means this requirement must already be satisfied,
+     *     otherwise 'max_vq' couldn't itself be enabled.
+     *
+     * [**] Not testable with the QMP interface, only with the command line.
+     */
+
+    /* max_vq := 8 */
+    assert_sve_vls(qts, cpu_type, 0x8b, "{ 'sve1024': true }");
+
+    /* max_vq := 8, vq < max_vq, !power-of-two(vq) */
+    assert_sve_vls(qts, cpu_type, 0x8f,
+                   "{ 'sve1024': true, 'sve384': true }");
+    assert_sve_vls(qts, cpu_type, 0x8b,
+                   "{ 'sve1024': true, 'sve384': false }");
+
+    /* max_vq := 8, vq < max_vq, power-of-two(vq) */
+    assert_sve_vls(qts, cpu_type, 0x8b,
+                   "{ 'sve1024': true, 'sve256': true }");
+    assert_error(qts, cpu_type, "cannot disable sve256",
+                 "{ 'sve1024': true, 'sve256': false }");
+
+    /* max_vq := 3, vq > max_vq, !all-power-of-two(< vq) */
+    assert_error(qts, cpu_type, "cannot disable sve512",
+                 "{ 'sve384': true, 'sve512': false, 'sve640': true }");
+
+    /*
+     * We can disable power-of-two vector lengths when all larger lengths
+     * are also disabled. We only need to disable the power-of-two length,
+     * as all non-enabled larger lengths will then be auto-disabled.
+     */
+    assert_sve_vls(qts, cpu_type, 0x7, "{ 'sve512': false }");
+
+    /* max_vq := 3, vq > max_vq, all-power-of-two(< vq) */
+    assert_sve_vls(qts, cpu_type, 0x1f,
+                   "{ 'sve384': true, 'sve512': true, 'sve640': true }");
+    assert_sve_vls(qts, cpu_type, 0xf,
+                   "{ 'sve384': true, 'sve512': true, 'sve640': false }");
+}
+
+static void sve_tests_sve_max_vq_8(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-cpu max,sve-max-vq=8");
+
+    assert_sve_vls(qts, "max", BIT(8) - 1, NULL);
+
+    /*
+     * Disabling the max-vq set by sve-max-vq is not allowed, but
+     * of course enabling it is OK.
+     */
+    assert_error(qts, "max", "cannot disable sve1024", "{ 'sve1024': false }");
+    assert_sve_vls(qts, "max", 0xff, "{ 'sve1024': true }");
+
+    /*
+     * Enabling anything larger than max-vq set by sve-max-vq is not

tests/arm-cpu-features.c static void assert_bad_props(QTestState *qts, const char *cpu_type) · line 310

Quoted 10 lines
tests/arm-cpu-features.c +10 -0
+     * allowed, but of course disabling everything larger is OK.
+     */
+    assert_error(qts, "max", "cannot enable sve1152", "{ 'sve1152': true }");
+    assert_sve_vls(qts, "max", 0xff, "{ 'sve1152': false }");
+
+    /*
+     * We can disable non power-of-two lengths smaller than the max-vq
+     * set by sve-max-vq, but not power-of-two lengths.
+     */
+    assert_sve_vls(qts, "max", 0xfb, "{ 'sve384': false }");

you can also test ,sve384=on => 0x7

Quoted 4 lines
Quoted 2 lines

We wouldn't get 0x7, we'd get 0xff. I could still add the test to ensure we can enable smaller vector lengths without an error though.

Oh yes because you have the sve-max-vq=8
I rather meant in sve_tests_default then

I can add to both.

Thanks,
drew

#12 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

In reply to Andrew Jones
Quoted 14 lines · Andrew Jones
target/arm/helper.c +7 -0
+static uint32_t sve_zcr_get_valid_len(ARMCPU *cpu, uint32_t start_len)
+{
+    uint32_t start_vq = (start_len & 0xf) + 1;
+
+    return arm_cpu_vq_map_next_smaller(cpu, start_vq + 1) - 1;
+}
+
 /*
  * Given that SVE is enabled, return the vector length for EL.
  */
@@ -5360,13 +5367,13 @@ uint32_t sve_zcr_len_for_el(CPUARMState *env, int el)
     uint32_t zcr_len = cpu->sve_max_vq - 1;
 
     if (el <= 1) {

target/arm/helper.c uint32_t sve_zcr_len_for_el(CPUARMState *env, int el) · line 5376

Quoted 10 lines
target/arm/helper.c +3 -3
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[1]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[1]);
     }
     if (el <= 2 && arm_feature(env, ARM_FEATURE_EL2)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[2]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[2]);
     }
     if (arm_feature(env, ARM_FEATURE_EL3)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[3]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[3]);

This has lost the MIN relation between these 3 registers. One possible solution is to keep these 3 if statements as-is, but make one call

    zcr_len = arm_cpu_vq_map_next_smaller(cpu, zcr_len);

at the end.

tests/arm-cpu-features.c line 20

Quoted 5 lines
tests/arm-cpu-features.c +5 -0
+#if __SIZEOF_LONG__ == 8
+#define BIT(n) (1UL << (n))
+#else
+#define BIT(n) (1ULL << (n))
+#endif

There's no reason not to always use 1ULL is there?

r~

#13 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

Quoted 1 line · Richard Henderson

On 9/24/19 4:31 AM, Andrew Jones wrote:

<snip>

Quoted 8 lines

tests/arm-cpu-features.c line 20

Quoted 6 lines
tests/arm-cpu-features.c +5 -0
+#if __SIZEOF_LONG__ == 8
+#define BIT(n) (1UL << (n))
+#else
+#define BIT(n) (1ULL << (n))
+#endif

There's no reason not to always use 1ULL is there?

Also we already have this helper in bitops.h so should just use that.

--
Alex Bennée

#14 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

Quoted 14 lines · Alex Bennée
Quoted 1 line · Richard Henderson

On 9/24/19 4:31 AM, Andrew Jones wrote:

<snip>

Quoted 8 lines

tests/arm-cpu-features.c line 20

Quoted 6 lines
tests/arm-cpu-features.c +5 -0
+#if __SIZEOF_LONG__ == 8
+#define BIT(n) (1UL << (n))
+#else
+#define BIT(n) (1ULL << (n))
+#endif

There's no reason not to always use 1ULL is there?

Quoted 1 line

Also we already have this helper in bitops.h so should just use that.

bitops.h BIT() uses UL, but we need ULL. FWIW, the kernel has a BIT_ULL() that we could mimic in QEMU. I guess I could even add that for this patch, if you'd like.

Thanks,
drew

#15 lore raw

Re: [PATCH v4 4/9] target/arm/cpu64: max cpu: Introduce sve<N> properties

In reply to Richard Henderson
Quoted 30 lines · Richard Henderson
Quoted 14 lines · Andrew Jones
target/arm/helper.c +7 -0
+static uint32_t sve_zcr_get_valid_len(ARMCPU *cpu, uint32_t start_len)
+{
+    uint32_t start_vq = (start_len & 0xf) + 1;
+
+    return arm_cpu_vq_map_next_smaller(cpu, start_vq + 1) - 1;
+}
+
 /*
  * Given that SVE is enabled, return the vector length for EL.
  */
@@ -5360,13 +5367,13 @@ uint32_t sve_zcr_len_for_el(CPUARMState *env, int el)
     uint32_t zcr_len = cpu->sve_max_vq - 1;
 
     if (el <= 1) {

target/arm/helper.c uint32_t sve_zcr_len_for_el(CPUARMState *env, int el) · line 5376

Quoted 10 lines
target/arm/helper.c +3 -3
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[1]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[1]);
     }
     if (el <= 2 && arm_feature(env, ARM_FEATURE_EL2)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[2]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[2]);
     }
     if (arm_feature(env, ARM_FEATURE_EL3)) {
-        zcr_len = MIN(zcr_len, 0xf & (uint32_t)env->vfp.zcr_el[3]);
+        zcr_len = sve_zcr_get_valid_len(cpu, env->vfp.zcr_el[3]);

This has lost the MIN relation between these 3 registers. One possible solution is to keep these 3 if statements as-is, but make one call

Quoted 3 lines
    zcr_len = arm_cpu_vq_map_next_smaller(cpu, zcr_len);

at the end.

Argh. I obviously didn't look closely enough at this function when I was modifying it. I must have interpreted it as an if-else-if sequence...

I'll fix this as you suggest.

Quoted 8 lines

tests/arm-cpu-features.c line 20

Quoted 6 lines
tests/arm-cpu-features.c +5 -0
+#if __SIZEOF_LONG__ == 8
+#define BIT(n) (1UL << (n))
+#else
+#define BIT(n) (1ULL << (n))
+#endif

There's no reason not to always use 1ULL is there?

Indeed ULL == UL for AArch64. I'll drop the #if... stuff.

Thanks,
drew

#16 lore raw

PATCH v4 5/9 target/arm/kvm64: Add kvm_arch_get/put_sve

In reply to Andrew Jones

These are the SVE equivalents to kvm_arch_get/put_fpsimd. Note, the swabbing is different than it is for fpsmid because the vector format is a little-endian stream of words.

---
 target/arm/kvm64.c | 137 +++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 133 insertions(+), 4 deletions(-)
target/arm/kvm64.c +133 -4
index 28f6db57d5ee..ea454c613919 100644
@@ -671,11 +671,12 @@ int kvm_arch_destroy_vcpu(CPUState *cs)
 bool kvm_arm_reg_syncs_via_cpreg_list(uint64_t regidx)
 {
     /* Return true if the regidx is a register we should synchronize
-     * via the cpreg_tuples array (ie is not a core reg we sync by
-     * hand in kvm_arch_get/put_registers())
+     * via the cpreg_tuples array (ie is not a core or sve reg that
+     * we sync by hand in kvm_arch_get/put_registers())
      */
     switch (regidx & KVM_REG_ARM_COPROC_MASK) {
     case KVM_REG_ARM_CORE:
+    case KVM_REG_ARM64_SVE:
         return false;
     default:
         return true;
@@ -761,6 +762,78 @@ static int kvm_arch_put_fpsimd(CPUState *cs)
     return 0;
 }
 
+/*
+ * SVE registers are encoded in KVM's memory in an endianness-invariant format.
+ * The byte at offset i from the start of the in-memory representation contains
+ * the bits [(7 + 8 * i) : (8 * i)] of the register value. As this means the
+ * lowest offsets are stored in the lowest memory addresses, then that nearly
+ * matches QEMU's representation, which is to use an array of host-endian
+ * uint64_t's, where the lower offsets are at the lower indices. To complete
+ * the translation we just need to byte swap the uint64_t's on big-endian hosts.
+ */
+static uint64_t *sve_bswap64(uint64_t *dst, uint64_t *src, int nr)
+{
+#ifdef HOST_WORDS_BIGENDIAN
+    int i;
+
+    for (i = 0; i < nr; ++i) {
+        dst[i] = bswap64(src[i]);
+    }
+
+    return dst;
+#else
+    return src;
+#endif
+}
+
+/*
+ * KVM SVE registers come in slices where ZREGs have a slice size of 2048 bits
+ * and PREGS and the FFR have a slice size of 256 bits. However we simply hard
+ * code the slice index to zero for now as it's unlikely we'll need more than
+ * one slice for quite some time.
+ */
+static int kvm_arch_put_sve(CPUState *cs)
+{
+    ARMCPU *cpu = ARM_CPU(cs);
+    CPUARMState *env = &cpu->env;
+    uint64_t tmp[ARM_MAX_VQ * 2];
+    uint64_t *r;
+    struct kvm_one_reg reg;
+    int n, ret;
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_ZREGS; ++n) {
+        r = sve_bswap64(tmp, &env->vfp.zregs[n].d[0], cpu->sve_max_vq * 2);
+        reg.addr = (uintptr_t)r;
+        reg.id = KVM_REG_ARM64_SVE_ZREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+    }
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_PREGS; ++n) {
+        r = sve_bswap64(tmp, r = &env->vfp.pregs[n].p[0],
+                        DIV_ROUND_UP(cpu->sve_max_vq, 8));
+        reg.addr = (uintptr_t)r;
+        reg.id = KVM_REG_ARM64_SVE_PREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+    }
+
+    r = sve_bswap64(tmp, &env->vfp.pregs[FFR_PRED_NUM].p[0],
+                    DIV_ROUND_UP(cpu->sve_max_vq, 8));
+    reg.addr = (uintptr_t)r;
+    reg.id = KVM_REG_ARM64_SVE_FFR(0);
+    ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+    if (ret) {
+        return ret;
+    }
+
+    return 0;
+}
+
 int kvm_arch_put_registers(CPUState *cs, int level)
 {
     struct kvm_one_reg reg;
@@ -855,7 +928,11 @@ int kvm_arch_put_registers(CPUState *cs, int level)
         }
     }
 
-    ret = kvm_arch_put_fpsimd(cs);
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arch_put_sve(cs);
+    } else {
+        ret = kvm_arch_put_fpsimd(cs);
+    }
     if (ret) {
         return ret;
     }
@@ -918,6 +995,54 @@ static int kvm_arch_get_fpsimd(CPUState *cs)
     return 0;
 }
 
+/*
+ * KVM SVE registers come in slices where ZREGs have a slice size of 2048 bits
+ * and PREGS and the FFR have a slice size of 256 bits. However we simply hard
+ * code the slice index to zero for now as it's unlikely we'll need more than
+ * one slice for quite some time.
+ */
+static int kvm_arch_get_sve(CPUState *cs)
+{
+    ARMCPU *cpu = ARM_CPU(cs);
+    CPUARMState *env = &cpu->env;
+    struct kvm_one_reg reg;
+    uint64_t *r;
+    int n, ret;
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_ZREGS; ++n) {
+        r = &env->vfp.zregs[n].d[0];
+        reg.addr = (uintptr_t)r;
+        reg.id = KVM_REG_ARM64_SVE_ZREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+        sve_bswap64(r, r, cpu->sve_max_vq * 2);
+    }
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_PREGS; ++n) {
+        r = &env->vfp.pregs[n].p[0];
+        reg.addr = (uintptr_t)r;
+        reg.id = KVM_REG_ARM64_SVE_PREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+        sve_bswap64(r, r, DIV_ROUND_UP(cpu->sve_max_vq, 8));
+    }
+
+    r = &env->vfp.pregs[FFR_PRED_NUM].p[0];
+    reg.addr = (uintptr_t)r;
+    reg.id = KVM_REG_ARM64_SVE_FFR(0);
+    ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
+    if (ret) {
+        return ret;
+    }
+    sve_bswap64(r, r, DIV_ROUND_UP(cpu->sve_max_vq, 8));
+
+    return 0;
+}
+
 int kvm_arch_get_registers(CPUState *cs)
 {
     struct kvm_one_reg reg;
@@ -1012,7 +1137,11 @@ int kvm_arch_get_registers(CPUState *cs)
         env->spsr = env->banked_spsr[i];
     }
 
-    ret = kvm_arch_get_fpsimd(cs);
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arch_get_sve(cs);
+    } else {
+        ret = kvm_arch_get_fpsimd(cs);
+    }
     if (ret) {
         return ret;
     }
Signature · 1 line
2.20.1
Reviewed-by on v4 5/9
#17 lore raw

Re: [PATCH v4 5/9] target/arm/kvm64: Add kvm_arch_get/put_sve

Hi Drew,

Quoted 6 lines · Andrew Jones

These are the SVE equivalents to kvm_arch_get/put_fpsimd. Note, the swabbing is different than it is for fpsmid because the vector format is a little-endian stream of words.

Eric

Quoted 184 lines
---
 target/arm/kvm64.c | 137 +++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 133 insertions(+), 4 deletions(-)
target/arm/kvm64.c +133 -4
index 28f6db57d5ee..ea454c613919 100644
@@ -671,11 +671,12 @@ int kvm_arch_destroy_vcpu(CPUState *cs)
 bool kvm_arm_reg_syncs_via_cpreg_list(uint64_t regidx)
 {
     /* Return true if the regidx is a register we should synchronize
-     * via the cpreg_tuples array (ie is not a core reg we sync by
-     * hand in kvm_arch_get/put_registers())
+     * via the cpreg_tuples array (ie is not a core or sve reg that
+     * we sync by hand in kvm_arch_get/put_registers())
      */
     switch (regidx & KVM_REG_ARM_COPROC_MASK) {
     case KVM_REG_ARM_CORE:
+    case KVM_REG_ARM64_SVE:
         return false;
     default:
         return true;
@@ -761,6 +762,78 @@ static int kvm_arch_put_fpsimd(CPUState *cs)
     return 0;
 }
 
+/*
+ * SVE registers are encoded in KVM's memory in an endianness-invariant format.
+ * The byte at offset i from the start of the in-memory representation contains
+ * the bits [(7 + 8 * i) : (8 * i)] of the register value. As this means the
+ * lowest offsets are stored in the lowest memory addresses, then that nearly
+ * matches QEMU's representation, which is to use an array of host-endian
+ * uint64_t's, where the lower offsets are at the lower indices. To complete
+ * the translation we just need to byte swap the uint64_t's on big-endian hosts.
+ */
+static uint64_t *sve_bswap64(uint64_t *dst, uint64_t *src, int nr)
+{
+#ifdef HOST_WORDS_BIGENDIAN
+    int i;
+
+    for (i = 0; i < nr; ++i) {
+        dst[i] = bswap64(src[i]);
+    }
+
+    return dst;
+#else
+    return src;
+#endif
+}
+
+/*
+ * KVM SVE registers come in slices where ZREGs have a slice size of 2048 bits
+ * and PREGS and the FFR have a slice size of 256 bits. However we simply hard
+ * code the slice index to zero for now as it's unlikely we'll need more than
+ * one slice for quite some time.
+ */
+static int kvm_arch_put_sve(CPUState *cs)
+{
+    ARMCPU *cpu = ARM_CPU(cs);
+    CPUARMState *env = &cpu->env;
+    uint64_t tmp[ARM_MAX_VQ * 2];
+    uint64_t *r;
+    struct kvm_one_reg reg;
+    int n, ret;
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_ZREGS; ++n) {
+        r = sve_bswap64(tmp, &env->vfp.zregs[n].d[0], cpu->sve_max_vq * 2);
+        reg.addr = (uintptr_t)r;
+        reg.id = KVM_REG_ARM64_SVE_ZREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+    }
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_PREGS; ++n) {
+        r = sve_bswap64(tmp, r = &env->vfp.pregs[n].p[0],
+                        DIV_ROUND_UP(cpu->sve_max_vq, 8));
+        reg.addr = (uintptr_t)r;
+        reg.id = KVM_REG_ARM64_SVE_PREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+    }
+
+    r = sve_bswap64(tmp, &env->vfp.pregs[FFR_PRED_NUM].p[0],
+                    DIV_ROUND_UP(cpu->sve_max_vq, 8));
+    reg.addr = (uintptr_t)r;
+    reg.id = KVM_REG_ARM64_SVE_FFR(0);
+    ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+    if (ret) {
+        return ret;
+    }
+
+    return 0;
+}
+
 int kvm_arch_put_registers(CPUState *cs, int level)
 {
     struct kvm_one_reg reg;
@@ -855,7 +928,11 @@ int kvm_arch_put_registers(CPUState *cs, int level)
         }
     }
 
-    ret = kvm_arch_put_fpsimd(cs);
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arch_put_sve(cs);
+    } else {
+        ret = kvm_arch_put_fpsimd(cs);
+    }
     if (ret) {
         return ret;
     }
@@ -918,6 +995,54 @@ static int kvm_arch_get_fpsimd(CPUState *cs)
     return 0;
 }
 
+/*
+ * KVM SVE registers come in slices where ZREGs have a slice size of 2048 bits
+ * and PREGS and the FFR have a slice size of 256 bits. However we simply hard
+ * code the slice index to zero for now as it's unlikely we'll need more than
+ * one slice for quite some time.
+ */
+static int kvm_arch_get_sve(CPUState *cs)
+{
+    ARMCPU *cpu = ARM_CPU(cs);
+    CPUARMState *env = &cpu->env;
+    struct kvm_one_reg reg;
+    uint64_t *r;
+    int n, ret;
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_ZREGS; ++n) {
+        r = &env->vfp.zregs[n].d[0];
+        reg.addr = (uintptr_t)r;
+        reg.id = KVM_REG_ARM64_SVE_ZREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+        sve_bswap64(r, r, cpu->sve_max_vq * 2);
+    }
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_PREGS; ++n) {
+        r = &env->vfp.pregs[n].p[0];
+        reg.addr = (uintptr_t)r;
+        reg.id = KVM_REG_ARM64_SVE_PREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+        sve_bswap64(r, r, DIV_ROUND_UP(cpu->sve_max_vq, 8));
+    }
+
+    r = &env->vfp.pregs[FFR_PRED_NUM].p[0];
+    reg.addr = (uintptr_t)r;
+    reg.id = KVM_REG_ARM64_SVE_FFR(0);
+    ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
+    if (ret) {
+        return ret;
+    }
+    sve_bswap64(r, r, DIV_ROUND_UP(cpu->sve_max_vq, 8));
+
+    return 0;
+}
+
 int kvm_arch_get_registers(CPUState *cs)
 {
     struct kvm_one_reg reg;
@@ -1012,7 +1137,11 @@ int kvm_arch_get_registers(CPUState *cs)
         env->spsr = env->banked_spsr[i];
     }
 
-    ret = kvm_arch_get_fpsimd(cs);
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arch_get_sve(cs);
+    } else {
+        ret = kvm_arch_get_fpsimd(cs);
+    }
     if (ret) {
         return ret;
     }
#18 lore raw

Re: [PATCH v4 5/9] target/arm/kvm64: Add kvm_arch_get/put_sve

In reply to Andrew Jones
Quoted 10 lines · Andrew Jones

These are the SVE equivalents to kvm_arch_get/put_fpsimd. Note, the swabbing is different than it is for fpsmid because the vector format is a little-endian stream of words.

---
 target/arm/kvm64.c | 137 +++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 133 insertions(+), 4 deletions(-)

It looks like I need to add the below changes to this patch as well, since FPSR and FPCR are still in use with SVE.

Thanks,
drew

target/arm/kvm64.c +36 -38
index 2da366ba113e..be31e56a1ab0 100644
@@ -854,10 +854,8 @@ int kvm_arm_cpreg_level(uint64_t regidx)
 
 static int kvm_arch_put_fpsimd(CPUState *cs)
 {
-    ARMCPU *cpu = ARM_CPU(cs);
-    CPUARMState *env = &cpu->env;
+    CPUARMState *env = &ARM_CPU(cs)->env;
     struct kvm_one_reg reg;
-    uint32_t fpr;
     int i, ret;
 
     for (i = 0; i < 32; i++) {
@@ -875,22 +873,6 @@ static int kvm_arch_put_fpsimd(CPUState *cs)
         }
     }
 
-    reg.addr = (uintptr_t)(&fpr);
-    fpr = vfp_get_fpsr(env);
-    reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpsr);
-    ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
-    if (ret) {
-        return ret;
-    }
-
-    reg.addr = (uintptr_t)(&fpr);
-    fpr = vfp_get_fpcr(env);
-    reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpcr);
-    ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
-    if (ret) {
-        return ret;
-    }
-
     return 0;
 }
 
@@ -970,6 +952,7 @@ int kvm_arch_put_registers(CPUState *cs, int level)
 {
     struct kvm_one_reg reg;
     uint64_t val;
+    uint32_t fpr;
     int i, ret;
     unsigned int el;
 
@@ -1069,6 +1052,22 @@ int kvm_arch_put_registers(CPUState *cs, int level)
         return ret;
     }
 
+    reg.addr = (uintptr_t)(&fpr);
+    fpr = vfp_get_fpsr(env);
+    reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpsr);
+    ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+    if (ret) {
+        return ret;
+    }
+
+    reg.addr = (uintptr_t)(&fpr);
+    fpr = vfp_get_fpcr(env);
+    reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpcr);
+    ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+    if (ret) {
+        return ret;
+    }
+
     ret = kvm_put_vcpu_events(cpu);
     if (ret) {
         return ret;
@@ -1087,10 +1086,8 @@ int kvm_arch_put_registers(CPUState *cs, int level)
 
 static int kvm_arch_get_fpsimd(CPUState *cs)
 {
-    ARMCPU *cpu = ARM_CPU(cs);
-    CPUARMState *env = &cpu->env;
+    CPUARMState *env = &ARM_CPU(cs)->env;
     struct kvm_one_reg reg;
-    uint32_t fpr;
     int i, ret;
 
     for (i = 0; i < 32; i++) {
@@ -1108,22 +1105,6 @@ static int kvm_arch_get_fpsimd(CPUState *cs)
         }
     }
 
-    reg.addr = (uintptr_t)(&fpr);
-    reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpsr);
-    ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
-    if (ret) {
-        return ret;
-    }
-    vfp_set_fpsr(env, fpr);
-
-    reg.addr = (uintptr_t)(&fpr);
-    reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpcr);
-    ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
-    if (ret) {
-        return ret;
-    }
-    vfp_set_fpcr(env, fpr);
-
     return 0;
 }
 
@@ -1180,6 +1161,7 @@ int kvm_arch_get_registers(CPUState *cs)
     struct kvm_one_reg reg;
     uint64_t val;
     unsigned int el;
+    uint32_t fpr;
     int i, ret;
 
     ARMCPU *cpu = ARM_CPU(cs);
@@ -1278,6 +1260,22 @@ int kvm_arch_get_registers(CPUState *cs)
         return ret;
     }
 
+    reg.addr = (uintptr_t)(&fpr);
+    reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpsr);
+    ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
+    if (ret) {
+        return ret;
+    }
+    vfp_set_fpsr(env, fpr);
+
+    reg.addr = (uintptr_t)(&fpr);
+    reg.id = AARCH64_SIMD_CTRL_REG(fp_regs.fpcr);
+    ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
+    if (ret) {
+        return ret;
+    }
+    vfp_set_fpcr(env, fpr);
+
     ret = kvm_get_vcpu_events(cpu);
     if (ret) {
         return ret;
#19 lore raw

Re: [PATCH v4 5/9] target/arm/kvm64: Add kvm_arch_get/put_sve

In reply to Andrew Jones
Quoted 75 lines · Andrew Jones

These are the SVE equivalents to kvm_arch_get/put_fpsimd. Note, the swabbing is different than it is for fpsmid because the vector format is a little-endian stream of words.

---
 target/arm/kvm64.c | 137 +++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 133 insertions(+), 4 deletions(-)
target/arm/kvm64.c +45 -2
index 28f6db57d5ee..ea454c613919 100644
@@ -671,11 +671,12 @@ int kvm_arch_destroy_vcpu(CPUState *cs)
 bool kvm_arm_reg_syncs_via_cpreg_list(uint64_t regidx)
 {
     /* Return true if the regidx is a register we should synchronize
-     * via the cpreg_tuples array (ie is not a core reg we sync by
-     * hand in kvm_arch_get/put_registers())
+     * via the cpreg_tuples array (ie is not a core or sve reg that
+     * we sync by hand in kvm_arch_get/put_registers())
      */
     switch (regidx & KVM_REG_ARM_COPROC_MASK) {
     case KVM_REG_ARM_CORE:
+    case KVM_REG_ARM64_SVE:
         return false;
     default:
         return true;
@@ -761,6 +762,78 @@ static int kvm_arch_put_fpsimd(CPUState *cs)
     return 0;
 }
 
+/*
+ * SVE registers are encoded in KVM's memory in an endianness-invariant format.
+ * The byte at offset i from the start of the in-memory representation contains
+ * the bits [(7 + 8 * i) : (8 * i)] of the register value. As this means the
+ * lowest offsets are stored in the lowest memory addresses, then that nearly
+ * matches QEMU's representation, which is to use an array of host-endian
+ * uint64_t's, where the lower offsets are at the lower indices. To complete
+ * the translation we just need to byte swap the uint64_t's on big-endian hosts.
+ */
+static uint64_t *sve_bswap64(uint64_t *dst, uint64_t *src, int nr)
+{
+#ifdef HOST_WORDS_BIGENDIAN
+    int i;
+
+    for (i = 0; i < nr; ++i) {
+        dst[i] = bswap64(src[i]);
+    }
+
+    return dst;
+#else
+    return src;
+#endif
+}
+
+/*
+ * KVM SVE registers come in slices where ZREGs have a slice size of 2048 bits
+ * and PREGS and the FFR have a slice size of 256 bits. However we simply hard
+ * code the slice index to zero for now as it's unlikely we'll need more than
+ * one slice for quite some time.
+ */
+static int kvm_arch_put_sve(CPUState *cs)
+{
+    ARMCPU *cpu = ARM_CPU(cs);
+    CPUARMState *env = &cpu->env;
+    uint64_t tmp[ARM_MAX_VQ * 2];
+    uint64_t *r;
+    struct kvm_one_reg reg;
+    int n, ret;
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_ZREGS; ++n) {
+        r = sve_bswap64(tmp, &env->vfp.zregs[n].d[0], cpu->sve_max_vq * 2);
+        reg.addr = (uintptr_t)r;

target/arm/kvm64.c static int kvm_arch_put_fpsimd(CPUState *cs) · line 816

Quoted 10 lines
target/arm/kvm64.c +10 -0
+        reg.id = KVM_REG_ARM64_SVE_ZREG(n, 0);
+        ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+        if (ret) {
+            return ret;
+        }
+    }
+
+    for (n = 0; n < KVM_ARM64_SVE_NUM_PREGS; ++n) {
+        r = sve_bswap64(tmp, r = &env->vfp.pregs[n].p[0],
+                        DIV_ROUND_UP(cpu->sve_max_vq, 8));

I see a bug here that I introduced between v2 and v3 when I switched to DIV_ROUND_UP. I dropped the '* 2's on all of these. They should be DIV_ROUND_UP(cpu->sve_max_vq * 2, 8). I'll fix for v5, which I'll be posting later today.

Thanks,
drew

#20 lore raw

PATCH v4 6/9 target/arm/kvm64: max cpu: Enable SVE when available

In reply to Andrew Jones

Enable SVE in the KVM guest when the 'max' cpu type is configured and KVM supports it. KVM SVE requires use of the new finalize vcpu ioctl, so we add that now too. For starters SVE can only be turned on or off, getting all vector lengths the host CPU supports when on. We'll add the other SVE CPU properties in later patches.

---
 target/arm/cpu64.c       | 17 ++++++++++++++---
 target/arm/kvm.c         |  5 +++++
 target/arm/kvm64.c       | 20 +++++++++++++++++++-
 target/arm/kvm_arm.h     | 27 +++++++++++++++++++++++++++
 tests/arm-cpu-features.c |  1 +
 5 files changed, 66 insertions(+), 4 deletions(-)
target/arm/cpu64.c +14 -3
index 606e3eceb9c0..b7eff4e1e107 100644
@@ -481,6 +481,11 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
         return;
     }
 
+    if (value && kvm_enabled() && !kvm_arm_sve_supported(CPU(cpu))) {
+        error_setg(errp, "'sve' feature not supported by KVM on this host");
+        return;
+    }
+
     t = cpu->isar.id_aa64pfr0;
     t = FIELD_DP64(t, ID_AA64PFR0, SVE, value);
     cpu->isar.id_aa64pfr0 = t;
@@ -495,11 +500,16 @@ static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
     uint32_t vq;
+    uint64_t t;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
+        if (kvm_arm_sve_supported(CPU(cpu))) {
+            t = cpu->isar.id_aa64pfr0;
+            t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
+            cpu->isar.id_aa64pfr0 = t;
+        }
     } else {
-        uint64_t t;
         uint32_t u;
         aarch64_a57_initfn(obj);
 
@@ -600,8 +610,6 @@ static void aarch64_max_initfn(Object *obj)
 
         object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                             cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
-        object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
-                            cpu_arm_set_sve, NULL, NULL, &error_fatal);
 
         for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
             char name[8];
@@ -610,6 +618,9 @@ static void aarch64_max_initfn(Object *obj)
                                 cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
         }
     }
+
+    object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
+                        cpu_arm_set_sve, NULL, NULL, &error_fatal);
 }
 
 struct ARMCPUInfo {
target/arm/kvm.c +5 -0
index b2eaa50b8df9..72569f71236e 100644
@@ -51,6 +51,11 @@ int kvm_arm_vcpu_init(CPUState *cs)
     return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_INIT, &init);
 }
 
+int kvm_arm_vcpu_finalize(CPUState *cs, int feature)
+{
+    return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_FINALIZE, &feature);
+}
+
 void kvm_arm_init_serror_injection(CPUState *cs)
 {
     cap_has_inject_serror_esr = kvm_check_extension(cs->kvm_state,
target/arm/kvm64.c +19 -1
index ea454c613919..fc62bab8684e 100644
@@ -602,6 +602,13 @@ bool kvm_arm_aarch32_supported(CPUState *cpu)
     return kvm_check_extension(s, KVM_CAP_ARM_EL1_32BIT);
 }
 
+bool kvm_arm_sve_supported(CPUState *cpu)
+{
+    KVMState *s = KVM_STATE(current_machine->accelerator);
+
+    return kvm_check_extension(s, KVM_CAP_ARM_SVE);
+}
+
 #define ARM_CPU_ID_MPIDR       3, 0, 0, 0, 5
 
 int kvm_arch_init_vcpu(CPUState *cs)
@@ -630,13 +637,17 @@ int kvm_arch_init_vcpu(CPUState *cs)
         cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_EL1_32BIT;
     }
     if (!kvm_check_extension(cs->kvm_state, KVM_CAP_ARM_PMU_V3)) {
-            cpu->has_pmu = false;
+        cpu->has_pmu = false;
     }
     if (cpu->has_pmu) {
         cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_PMU_V3;
     } else {
         unset_feature(&env->features, ARM_FEATURE_PMU);
     }
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        assert(kvm_arm_sve_supported(cs));
+        cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_SVE;
+    }
 
     /* Do KVM_ARM_VCPU_INIT ioctl */
     ret = kvm_arm_vcpu_init(cs);
@@ -644,6 +655,13 @@ int kvm_arch_init_vcpu(CPUState *cs)
         return ret;
     }
 
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arm_vcpu_finalize(cs, KVM_ARM_VCPU_SVE);
+        if (ret) {
+            return ret;
+        }
+    }
+
     /*
      * When KVM is in use, PSCI is emulated in-kernel and not by qemu.
      * Currently KVM has its own idea about MPIDR assignment, so we
target/arm/kvm_arm.h +27 -0
index b3106c8600af..1151877f97ea 100644
@@ -27,6 +27,20 @@
  */
 int kvm_arm_vcpu_init(CPUState *cs);
 
+/**
+ * kvm_arm_vcpu_finalize
+ * @cs: CPUState
+ * @feature: int
+ *
+ * Finalizes the configuration of the specified VCPU feature by
+ * invoking the KVM_ARM_VCPU_FINALIZE ioctl. Features requiring
+ * this are documented in the "KVM_ARM_VCPU_FINALIZE" section of
+ * KVM's API documentation.
+ *
+ * Returns: 0 if success else < 0 error code
+ */
+int kvm_arm_vcpu_finalize(CPUState *cs, int feature);
+
 /**
  * kvm_arm_register_device:
  * @mr: memory region for this device
@@ -225,6 +239,14 @@ bool kvm_arm_aarch32_supported(CPUState *cs);
  */
 bool kvm_arm_pmu_supported(CPUState *cs);
 
+/**
+ * bool kvm_arm_sve_supported:
+ * @cs: CPUState
+ *
+ * Returns true if the KVM VCPU can enable SVE and false otherwise.
+ */
+bool kvm_arm_sve_supported(CPUState *cs);
+
 /**
  * kvm_arm_get_max_vm_ipa_size - Returns the number of bits in the
  * IPA address space supported by KVM
@@ -275,6 +297,11 @@ static inline bool kvm_arm_pmu_supported(CPUState *cs)
     return false;
 }
 
+static inline bool kvm_arm_sve_supported(CPUState *cs)
+{
+    return false;
+}
+
 static inline int kvm_arm_get_max_vm_ipa_size(MachineState *ms)
 {
     return -ENOENT;
tests/arm-cpu-features.c +1 -0
index 9a2dd402b769..d50f98cb6aea 100644
@@ -386,6 +386,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
 
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
         assert_has_feature(qts, "host", "aarch64");
+        assert_has_feature(qts, "max", "sve");
 
         assert_error(qts, "cortex-a15",
             "We cannot guarantee the CPU type 'cortex-a15' works "
Signature · 1 line
2.20.1
Reviewed-by on v4 6/9
#21 lore raw

Re: [PATCH v4 6/9] target/arm/kvm64: max cpu: Enable SVE when available

Hi Drew,

Quoted 8 lines · Andrew Jones

Enable SVE in the KVM guest when the 'max' cpu type is configured and KVM supports it. KVM SVE requires use of the new finalize vcpu ioctl, so we add that now too. For starters SVE can only be turned on or off, getting all vector lengths the host CPU supports when on. We'll add the other SVE CPU properties in later patches.

Eric

Quoted 193 lines
---
 target/arm/cpu64.c       | 17 ++++++++++++++---
 target/arm/kvm.c         |  5 +++++
 target/arm/kvm64.c       | 20 +++++++++++++++++++-
 target/arm/kvm_arm.h     | 27 +++++++++++++++++++++++++++
 tests/arm-cpu-features.c |  1 +
 5 files changed, 66 insertions(+), 4 deletions(-)
target/arm/cpu64.c +14 -3
index 606e3eceb9c0..b7eff4e1e107 100644
@@ -481,6 +481,11 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
         return;
     }
 
+    if (value && kvm_enabled() && !kvm_arm_sve_supported(CPU(cpu))) {
+        error_setg(errp, "'sve' feature not supported by KVM on this host");
+        return;
+    }
+
     t = cpu->isar.id_aa64pfr0;
     t = FIELD_DP64(t, ID_AA64PFR0, SVE, value);
     cpu->isar.id_aa64pfr0 = t;
@@ -495,11 +500,16 @@ static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
     uint32_t vq;
+    uint64_t t;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
+        if (kvm_arm_sve_supported(CPU(cpu))) {
+            t = cpu->isar.id_aa64pfr0;
+            t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
+            cpu->isar.id_aa64pfr0 = t;
+        }
     } else {
-        uint64_t t;
         uint32_t u;
         aarch64_a57_initfn(obj);
 
@@ -600,8 +610,6 @@ static void aarch64_max_initfn(Object *obj)
 
         object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                             cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
-        object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
-                            cpu_arm_set_sve, NULL, NULL, &error_fatal);
 
         for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
             char name[8];
@@ -610,6 +618,9 @@ static void aarch64_max_initfn(Object *obj)
                                 cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
         }
     }
+
+    object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
+                        cpu_arm_set_sve, NULL, NULL, &error_fatal);
 }
 
 struct ARMCPUInfo {
target/arm/kvm.c +5 -0
index b2eaa50b8df9..72569f71236e 100644
@@ -51,6 +51,11 @@ int kvm_arm_vcpu_init(CPUState *cs)
     return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_INIT, &init);
 }
 
+int kvm_arm_vcpu_finalize(CPUState *cs, int feature)
+{
+    return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_FINALIZE, &feature);
+}
+
 void kvm_arm_init_serror_injection(CPUState *cs)
 {
     cap_has_inject_serror_esr = kvm_check_extension(cs->kvm_state,
target/arm/kvm64.c +19 -1
index ea454c613919..fc62bab8684e 100644
@@ -602,6 +602,13 @@ bool kvm_arm_aarch32_supported(CPUState *cpu)
     return kvm_check_extension(s, KVM_CAP_ARM_EL1_32BIT);
 }
 
+bool kvm_arm_sve_supported(CPUState *cpu)
+{
+    KVMState *s = KVM_STATE(current_machine->accelerator);
+
+    return kvm_check_extension(s, KVM_CAP_ARM_SVE);
+}
+
 #define ARM_CPU_ID_MPIDR       3, 0, 0, 0, 5
 
 int kvm_arch_init_vcpu(CPUState *cs)
@@ -630,13 +637,17 @@ int kvm_arch_init_vcpu(CPUState *cs)
         cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_EL1_32BIT;
     }
     if (!kvm_check_extension(cs->kvm_state, KVM_CAP_ARM_PMU_V3)) {
-            cpu->has_pmu = false;
+        cpu->has_pmu = false;
     }
     if (cpu->has_pmu) {
         cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_PMU_V3;
     } else {
         unset_feature(&env->features, ARM_FEATURE_PMU);
     }
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        assert(kvm_arm_sve_supported(cs));
+        cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_SVE;
+    }
 
     /* Do KVM_ARM_VCPU_INIT ioctl */
     ret = kvm_arm_vcpu_init(cs);
@@ -644,6 +655,13 @@ int kvm_arch_init_vcpu(CPUState *cs)
         return ret;
     }
 
+    if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arm_vcpu_finalize(cs, KVM_ARM_VCPU_SVE);
+        if (ret) {
+            return ret;
+        }
+    }
+
     /*
      * When KVM is in use, PSCI is emulated in-kernel and not by qemu.
      * Currently KVM has its own idea about MPIDR assignment, so we
target/arm/kvm_arm.h +27 -0
index b3106c8600af..1151877f97ea 100644
@@ -27,6 +27,20 @@
  */
 int kvm_arm_vcpu_init(CPUState *cs);
 
+/**
+ * kvm_arm_vcpu_finalize
+ * @cs: CPUState
+ * @feature: int
+ *
+ * Finalizes the configuration of the specified VCPU feature by
+ * invoking the KVM_ARM_VCPU_FINALIZE ioctl. Features requiring
+ * this are documented in the "KVM_ARM_VCPU_FINALIZE" section of
+ * KVM's API documentation.
+ *
+ * Returns: 0 if success else < 0 error code
+ */
+int kvm_arm_vcpu_finalize(CPUState *cs, int feature);
+
 /**
  * kvm_arm_register_device:
  * @mr: memory region for this device
@@ -225,6 +239,14 @@ bool kvm_arm_aarch32_supported(CPUState *cs);
  */
 bool kvm_arm_pmu_supported(CPUState *cs);
 
+/**
+ * bool kvm_arm_sve_supported:
+ * @cs: CPUState
+ *
+ * Returns true if the KVM VCPU can enable SVE and false otherwise.
+ */
+bool kvm_arm_sve_supported(CPUState *cs);
+
 /**
  * kvm_arm_get_max_vm_ipa_size - Returns the number of bits in the
  * IPA address space supported by KVM
@@ -275,6 +297,11 @@ static inline bool kvm_arm_pmu_supported(CPUState *cs)
     return false;
 }
 
+static inline bool kvm_arm_sve_supported(CPUState *cs)
+{
+    return false;
+}
+
 static inline int kvm_arm_get_max_vm_ipa_size(MachineState *ms)
 {
     return -ENOENT;
tests/arm-cpu-features.c +1 -0
index 9a2dd402b769..d50f98cb6aea 100644
@@ -386,6 +386,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
 
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
         assert_has_feature(qts, "host", "aarch64");
+        assert_has_feature(qts, "max", "sve");
 
         assert_error(qts, "cortex-a15",
             "We cannot guarantee the CPU type 'cortex-a15' works "
#22 lore raw

PATCH v4 7/9 target/arm/kvm: scratch vcpu: Preserve input kvm_vcpu_init features

In reply to Andrew Jones

kvm_arm_create_scratch_host_vcpu() takes a struct kvm_vcpu_init parameter. Rather than just using it as an output parameter to pass back the preferred target, use it also as an input parameter, allowing a caller to pass a selected target if they wish and to also pass cpu features. If the caller doesn't want to select a target they can pass -1 for the target which indicates they want to use the preferred target and have it passed back like before.

---
 target/arm/kvm.c   | 20 +++++++++++++++-----
 target/arm/kvm32.c |  6 +++++-
 target/arm/kvm64.c |  6 +++++-
 3 files changed, 25 insertions(+), 7 deletions(-)
target/arm/kvm.c +15 -5
index 72569f71236e..7a540d9591f9 100644
@@ -66,7 +66,7 @@ bool kvm_arm_create_scratch_host_vcpu(const uint32_t *cpus_to_try,
                                       int *fdarray,
                                       struct kvm_vcpu_init *init)
 {
-    int ret, kvmfd = -1, vmfd = -1, cpufd = -1;
+    int ret = 0, kvmfd = -1, vmfd = -1, cpufd = -1;
 
     kvmfd = qemu_open("/dev/kvm", O_RDWR);
     if (kvmfd < 0) {
@@ -86,7 +86,14 @@ bool kvm_arm_create_scratch_host_vcpu(const uint32_t *cpus_to_try,
         goto finish;
     }
 
-    ret = ioctl(vmfd, KVM_ARM_PREFERRED_TARGET, init);
+    if (init->target == -1) {
+        struct kvm_vcpu_init preferred;
+
+        ret = ioctl(vmfd, KVM_ARM_PREFERRED_TARGET, &preferred);
+        if (!ret) {
+            init->target = preferred.target;
+        }
+    }
     if (ret >= 0) {
         ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init);
         if (ret < 0) {
@@ -98,10 +105,12 @@ bool kvm_arm_create_scratch_host_vcpu(const uint32_t *cpus_to_try,
          * creating one kind of guest CPU which is its preferred
          * CPU type.
          */
+        struct kvm_vcpu_init try;
+
         while (*cpus_to_try != QEMU_KVM_ARM_TARGET_NONE) {
-            init->target = *cpus_to_try++;
-            memset(init->features, 0, sizeof(init->features));
-            ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init);
+            try.target = *cpus_to_try++;
+            memcpy(try.features, init->features, sizeof(init->features));
+            ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, &try);
             if (ret >= 0) {
                 break;
             }
@@ -109,6 +118,7 @@ bool kvm_arm_create_scratch_host_vcpu(const uint32_t *cpus_to_try,
         if (ret < 0) {
             goto err;
         }
+        init->target = try.target;
     } else {
         /* Treat a NULL cpus_to_try argument the same as an empty
          * list, which means we will fail the call since this must
target/arm/kvm32.c +5 -1
index 2451a2d4bbef..32bf8d6757c4 100644
@@ -53,7 +53,11 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
         QEMU_KVM_ARM_TARGET_CORTEX_A15,
         QEMU_KVM_ARM_TARGET_NONE
     };
-    struct kvm_vcpu_init init;
+    /*
+     * target = -1 informs kvm_arm_create_scratch_host_vcpu()
+     * to use the preferred target
+     */
+    struct kvm_vcpu_init init = { .target = -1, };
 
     if (!kvm_arm_create_scratch_host_vcpu(cpus_to_try, fdarray, &init)) {
         return false;
target/arm/kvm64.c +5 -1
index fc62bab8684e..f96649ae0349 100644
@@ -502,7 +502,11 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
         KVM_ARM_TARGET_CORTEX_A57,
         QEMU_KVM_ARM_TARGET_NONE
     };
-    struct kvm_vcpu_init init;
+    /*
+     * target = -1 informs kvm_arm_create_scratch_host_vcpu()
+     * to use the preferred target
+     */
+    struct kvm_vcpu_init init = { .target = -1, };
 
     if (!kvm_arm_create_scratch_host_vcpu(cpus_to_try, fdarray, &init)) {
         return false;
Signature · 1 line
2.20.1
#23 lore raw

PATCH v4 8/9 target/arm/cpu64: max cpu: Support sve properties with KVM

In reply to Andrew Jones

Extend the SVE vq map initialization and validation with KVM's supported vector lengths when KVM is enabled. In order to determine and select supported lengths we add two new KVM functions for getting and setting the KVM_REG_ARM64_SVE_VLS pseudo-register.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst |  36 +++++---
 target/arm/cpu64.c        | 167 +++++++++++++++++++++++++++++---------
 target/arm/kvm64.c        | 100 ++++++++++++++++++++++-
 target/arm/kvm_arm.h      |  12 +++
 tests/arm-cpu-features.c  | 105 +++++++++++++++++++++++-
 5 files changed, 368 insertions(+), 52 deletions(-)
docs/arm-cpu-features.rst +26 -10
index 1262fddc6201..939366f959cf 100644
@@ -188,10 +188,17 @@ SVE CPU Property Dependencies and Constraints
 
   1) At least one vector length must be enabled when `sve` is enabled.
 
-  2) If a vector length `N` is enabled, then all power-of-two vector
-     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
-     is enabled, then `sve128` and `sve256` must also be enabled,
-     but `sve384` is not required.
+  2) If a vector length `N` is enabled, then, when KVM is enabled, all
+     smaller, host supported vector lengths must also be enabled.  If
+     KVM is not enabled, then only all the smaller, power-of-two vector
+     lengths must be enabled.  E.g. with KVM if the host supports all
+     vector lengths up to 512-bits (128, 256, 384, 512), then if
+     `sve512` is enabled, `sve128`, `sve256`, and `sve384` must also
+     be enabled. Without KVM, `sve384` would not be required.
+
+  3) If KVM is enabled then only vector lengths that the host CPU type
+     support may be enabled.  If SVE is not supported by the host, then
+     no `sve*` properties may be enabled.
 
 SVE CPU Property Parsing Semantics
 ----------------------------------
@@ -210,20 +217,29 @@ SVE CPU Property Parsing Semantics
      disable the last enabled vector length (see constraint (1) of "SVE
      CPU Property Dependencies and Constraints").
 
-  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+  4) When KVM is enabled, if the host does not support SVE, then an error
+     is generated when attempting to enable any `sve*` properties.
+
+  5) When KVM is enabled, if the host does support SVE, then an error is
+     generated when attempting to enable any vector lengths not supported
+     by the host.
+
+  6) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
      CPU properties are set `on`, then the specified vector lengths are
      disabled but the default for any unspecified lengths remains enabled.
-     Disabling a power-of-two vector length also disables all vector
-     lengths larger than the power-of-two length (see constraint (2) of
-     "SVE CPU Property Dependencies and Constraints").
+     When KVM is not enabled, disabling a power-of-two vector length also
+     disables all vector lengths larger than the power-of-two length.
+     When KVM is enabled, then disabling any supported vector length also
+     disables all larger vector lengths (see constraint (2) of "SVE CPU
+     Property Dependencies and Constraints").
 
-  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+  7) If one or more `sve<N>` CPU properties are set to `on`, then they
      are enabled and all unspecified lengths default to disabled, except
      for the required lengths per constraint (2) of "SVE CPU Property
      Dependencies and Constraints", which will even be auto-enabled if
      they were not explicitly enabled.
 
-  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+  8) If SVE was disabled (`sve=off`), allowing all vector lengths to be
      explicitly disabled (i.e. avoiding the error specified in (3) of
      "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
      provided an error will be generated.  To avoid this error, one must
target/arm/cpu64.c +127 -40

224 lines not on this page · raw message

target/arm/kvm64.c +99 -1

118 lines not on this page · raw message

target/arm/kvm_arm.h +12 -0
index 1151877f97ea..a1cc6513f72b 100644
@@ -212,6 +212,17 @@ typedef struct ARMHostCPUFeatures {
  */
 bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf);
 
+/**
+ * kvm_arm_sve_get_vls:
+ * @cs: CPUState
+ * @map: bitmap to fill in
+ *
+ * Get all the SVE vector lengths supported by the KVM host, setting
+ * the bits corresponding to their length in quadwords minus one
+ * (vq - 1) in @map up to ARM_MAX_VQ.
+ */
+void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map);
+
 /**
  * kvm_arm_set_cpu_features_from_host:
  * @cpu: ARMCPU to set the features for
@@ -315,6 +326,7 @@ static inline int kvm_arm_vgic_probe(void)
 static inline void kvm_arm_pmu_set_irq(CPUState *cs, int irq) {}
 static inline void kvm_arm_pmu_init(CPUState *cs) {}
 
+static inline void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map) {}
 #endif
 
 static inline const char *gic_class_name(void)
tests/arm-cpu-features.c +104 -1

134 lines not on this page · raw message

Signature · 1 line
2.20.1
Reviewed-by on v4 8/9
#24 lore raw

Re: [PATCH v4 8/9] target/arm/cpu64: max cpu: Support sve properties with KVM

Hi Drew,

Quoted 89 lines · Andrew Jones

Extend the SVE vq map initialization and validation with KVM's supported vector lengths when KVM is enabled. In order to determine and select supported lengths we add two new KVM functions for getting and setting the KVM_REG_ARM64_SVE_VLS pseudo-register.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst |  36 +++++---
 target/arm/cpu64.c        | 167 +++++++++++++++++++++++++++++---------
 target/arm/kvm64.c        | 100 ++++++++++++++++++++++-
 target/arm/kvm_arm.h      |  12 +++
 tests/arm-cpu-features.c  | 105 +++++++++++++++++++++++-
 5 files changed, 368 insertions(+), 52 deletions(-)
docs/arm-cpu-features.rst +26 -10
index 1262fddc6201..939366f959cf 100644
@@ -188,10 +188,17 @@ SVE CPU Property Dependencies and Constraints
 
   1) At least one vector length must be enabled when `sve` is enabled.
 
-  2) If a vector length `N` is enabled, then all power-of-two vector
-     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
-     is enabled, then `sve128` and `sve256` must also be enabled,
-     but `sve384` is not required.
+  2) If a vector length `N` is enabled, then, when KVM is enabled, all
+     smaller, host supported vector lengths must also be enabled.  If
+     KVM is not enabled, then only all the smaller, power-of-two vector
+     lengths must be enabled.  E.g. with KVM if the host supports all
+     vector lengths up to 512-bits (128, 256, 384, 512), then if
+     `sve512` is enabled, `sve128`, `sve256`, and `sve384` must also
+     be enabled. Without KVM, `sve384` would not be required.
+
+  3) If KVM is enabled then only vector lengths that the host CPU type
+     support may be enabled.  If SVE is not supported by the host, then
+     no `sve*` properties may be enabled.
 
 SVE CPU Property Parsing Semantics
 ----------------------------------
@@ -210,20 +217,29 @@ SVE CPU Property Parsing Semantics
      disable the last enabled vector length (see constraint (1) of "SVE
      CPU Property Dependencies and Constraints").
 
-  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+  4) When KVM is enabled, if the host does not support SVE, then an error
+     is generated when attempting to enable any `sve*` properties.
+
+  5) When KVM is enabled, if the host does support SVE, then an error is
+     generated when attempting to enable any vector lengths not supported
+     by the host.
+
+  6) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
      CPU properties are set `on`, then the specified vector lengths are
      disabled but the default for any unspecified lengths remains enabled.
-     Disabling a power-of-two vector length also disables all vector
-     lengths larger than the power-of-two length (see constraint (2) of
-     "SVE CPU Property Dependencies and Constraints").
+     When KVM is not enabled, disabling a power-of-two vector length also
+     disables all vector lengths larger than the power-of-two length.
+     When KVM is enabled, then disabling any supported vector length also
+     disables all larger vector lengths (see constraint (2) of "SVE CPU
+     Property Dependencies and Constraints").
 
-  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+  7) If one or more `sve<N>` CPU properties are set to `on`, then they
      are enabled and all unspecified lengths default to disabled, except
      for the required lengths per constraint (2) of "SVE CPU Property
      Dependencies and Constraints", which will even be auto-enabled if
      they were not explicitly enabled.
 
-  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+  8) If SVE was disabled (`sve=off`), allowing all vector lengths to be
      explicitly disabled (i.e. avoiding the error specified in (3) of
      "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
      provided an error will be generated.  To avoid this error, one must
target/arm/cpu64.c +0 -0
index b7eff4e1e107..18dd5e24ec61 100644
@@ -273,9 +273,18 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
      * any of the above.  Finally, if SVE is not disabled, then at least one
      * vector length must be enabled.
      */

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 285

Quoted 10 lines
target/arm/cpu64.c +7 -0
+    DECLARE_BITMAP(kvm_supported, ARM_MAX_VQ);
     DECLARE_BITMAP(tmp, ARM_MAX_VQ);
     uint32_t vq, max_vq = 0;
 
+    /* Collect the set of vector lengths supported by KVM. */
+    bitmap_zero(kvm_supported, ARM_MAX_VQ);
+    if (kvm_enabled() && kvm_arm_sve_supported(CPU(cpu))) {
+        kvm_arm_sve_get_vls(CPU(cpu), kvm_supported);
+    } else if (kvm_enabled()) {
+        assert(!cpu_isar_feature(aa64_sve, cpu));

why not set an error and propagate it instead?

Quoted 49 lines
target/arm/cpu64.c +19 -16
+    }
+
     /*
      * Process explicit sve<N> properties.
      * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
@@ -293,10 +302,19 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Propagate enabled bits down through required powers-of-two. */
-        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
-            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
-                set_bit(vq - 1, cpu->sve_vq_map);
+        if (kvm_enabled()) {
+            /*
+             * For KVM we have to automatically enable all supported unitialized
+             * lengths, even when the smaller lengths are not all powers-of-two.
+             */
+            bitmap_andnot(tmp, kvm_supported, cpu->sve_vq_init, max_vq);
+            bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+        } else {
+            /* Propagate enabled bits down through required powers-of-two. */
+            for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+                if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                    set_bit(vq - 1, cpu->sve_vq_map);
+                }
             }
         }
     } else if (cpu->sve_max_vq == 0) {
@@ -308,23 +326,46 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Disabling a power-of-two disables all larger lengths. */
-        if (test_bit(0, cpu->sve_vq_init)) {
-            error_setg(errp, "cannot disable sve128");
-            error_append_hint(errp, "Disabling sve128 results in all vector "
-                              "lengths being disabled.\n");
-            error_append_hint(errp, "With SVE enabled, at least one vector "
-                              "length must be enabled.\n");
-            return;
-        }
-        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
-            if (test_bit(vq - 1, cpu->sve_vq_init)) {
-                break;
+        if (kvm_enabled()) {
+            /* Disabling a supported length disables all larger lengths. */
+            for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+                if (test_bit(vq - 1, cpu->sve_vq_init) &&

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 342

Quoted 10 lines
target/arm/cpu64.c +10 -0
+                    test_bit(vq - 1, kvm_supported)) {
+                    break;
+                }
+            }
+            max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+            bitmap_andnot(cpu->sve_vq_map, kvm_supported,
+                          cpu->sve_vq_init, max_vq);
+            if (max_vq == 0 || bitmap_empty(cpu->sve_vq_map, max_vq)) {
+                vq = find_next_bit(kvm_supported, ARM_MAX_VQ, 0) + 1;
+                error_setg(errp, "cannot disable sve%d", vq * 128);

isn't the one disabled max_vq? Do you really need to re-compute vq?

Quoted 55 lines
target/arm/cpu64.c +35 -10
+                error_append_hint(errp, "Disabling sve%d results in all "
+                                  "vector lengths being disabled.\n",
+                                  vq * 128);
+                error_append_hint(errp, "With SVE enabled, at least one "
+                                  "vector length must be enabled.\n");
+                return;> +            }
+        } else {
+            /* Disabling a power-of-two disables all larger lengths. */
+            if (test_bit(0, cpu->sve_vq_init)) {
+                error_setg(errp, "cannot disable sve128");
+                error_append_hint(errp, "Disabling sve128 results in all "
+                                  "vector lengths being disabled.\n");
+                error_append_hint(errp, "With SVE enabled, at least one "
+                                  "vector length must be enabled.\n");
+                return;
+            }
+            for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
+                if (test_bit(vq - 1, cpu->sve_vq_init)) {
+                    break;
+                }
             }
+            max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+            bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
         }
-        max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
 
-        bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
         max_vq = find_last_bit(cpu->sve_vq_map, max_vq) + 1;
     }
 
@@ -358,16 +399,48 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
     assert(max_vq != 0);
     bitmap_clear(cpu->sve_vq_map, max_vq, ARM_MAX_VQ - max_vq);
 
-    /* Ensure all required powers-of-two are enabled. */
-    for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
-        if (!test_bit(vq - 1, cpu->sve_vq_map)) {
-            error_setg(errp, "cannot disable sve%d", vq * 128);
-            error_append_hint(errp, "sve%d is required as it "
-                              "is a power-of-two length smaller than "
-                              "the maximum, sve%d\n",
-                              vq * 128, max_vq * 128);
+    if (kvm_enabled()) {
+        /* Ensure the set of lengths matches what KVM supports. */
+        bitmap_xor(tmp, cpu->sve_vq_map, kvm_supported, max_vq);
+        if (!bitmap_empty(tmp, max_vq)) {
+            vq = find_last_bit(tmp, max_vq) + 1;
+            if (test_bit(vq - 1, cpu->sve_vq_map)) {
+                if (cpu->sve_max_vq) {
+                    error_setg(errp, "cannot set sve-max-vq=%d",
+                               cpu->sve_max_vq);
+                    error_append_hint(errp, "This KVM host does not support "
+                                      "the vector length %d-bits.\n",
+                                      vq * 128);
+                    error_append_hint(errp, "It may not be possible to use "

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 424

Quoted 10 lines
target/arm/cpu64.c +10 -0
+                                      "sve-max-vq with this KVM host. Try "
+                                      "using only sve<N> properties.\n");
+                } else {
+                    error_setg(errp, "cannot enable sve%d", vq * 128);
+                    error_append_hint(errp, "This KVM host does not support "
+                                      "the vector length %d-bits.\n",
+                                      vq * 128);
+                }
+            } else {
+                error_setg(errp, "cannot disable sve%d", vq * 128);

I am not sure about the above message. what is the vq value supposed to be? For instance if we previously entered the
"if (!bitmap_empty(cpu->sve_vq_map, ARM_MAX_VQ)) {" path

Quoted 380 lines
target/arm/cpu64.c +45 -14
+                error_append_hint(errp, "The KVM host requires all "
+                                  "supported vector lengths smaller "
+                                  "than %d bits to also be enabled.\n",
+                                  max_vq * 128);
+            }
             return;
         }
+    } else {
+        /* Ensure all required powers-of-two are enabled. */
+        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+            if (!test_bit(vq - 1, cpu->sve_vq_map)) {
+                error_setg(errp, "cannot disable sve%d", vq * 128);
+                error_append_hint(errp, "sve%d is required as it "
+                                  "is a power-of-two length smaller than "
+                                  "the maximum, sve%d\n",
+                                  vq * 128, max_vq * 128);
+                return;
+            }
+        }
     }
 
     /* From now on sve_max_vq is the actual maximum supported length. */
@@ -411,13 +484,22 @@ static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,
     Error *err = NULL;
 
     visit_type_uint32(v, name, &cpu->sve_max_vq, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
 
-    if (!err && (cpu->sve_max_vq == 0 || cpu->sve_max_vq > ARM_MAX_VQ)) {
-        error_setg(&err, "unsupported SVE vector length");
-        error_append_hint(&err, "Valid sve-max-vq in range [1-%d]\n",
+    if (kvm_enabled() && !kvm_arm_sve_supported(CPU(cpu))) {
+        error_setg(errp, "cannot set sve-max-vq");
+        error_append_hint(errp, "SVE not supported by KVM on this host\n");
+        return;
+    }
+
+    if (cpu->sve_max_vq == 0 || cpu->sve_max_vq > ARM_MAX_VQ) {
+        error_setg(errp, "unsupported SVE vector length");
+        error_append_hint(errp, "Valid sve-max-vq in range [1-%d]\n",
                           ARM_MAX_VQ);
     }
-    error_propagate(errp, err);
 }
 
 static void cpu_arm_get_sve_vq(Object *obj, Visitor *v, const char *name,
@@ -450,6 +532,12 @@ static void cpu_arm_set_sve_vq(Object *obj, Visitor *v, const char *name,
         return;
     }
 
+    if (value && kvm_enabled() && !kvm_arm_sve_supported(CPU(cpu))) {
+        error_setg(errp, "cannot enable %s", name);
+        error_append_hint(errp, "SVE not supported by KVM on this host\n");
+        return;
+    }
+
     if (value) {
         set_bit(vq - 1, cpu->sve_vq_map);
     } else {
@@ -607,20 +695,19 @@ static void aarch64_max_initfn(Object *obj)
         cpu->ctr = 0x80038003; /* 32 byte I and D cacheline size, VIPT icache */
         cpu->dcz_blocksize = 7; /*  512 bytes */
 #endif
-
-        object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
-                            cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
-
-        for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
-            char name[8];
-            sprintf(name, "sve%d", vq * 128);
-            object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
-                                cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
-        }
     }
 
     object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
                         cpu_arm_set_sve, NULL, NULL, &error_fatal);
+    object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
+                        cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
+
+    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+        char name[8];
+        sprintf(name, "sve%d", vq * 128);
+        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+    }
 }
 
 struct ARMCPUInfo {
target/arm/kvm64.c +99 -1
index f96649ae0349..cff4217a8469 100644
@@ -613,6 +613,100 @@ bool kvm_arm_sve_supported(CPUState *cpu)
     return kvm_check_extension(s, KVM_CAP_ARM_SVE);
 }
 
+QEMU_BUILD_BUG_ON(KVM_ARM64_SVE_VQ_MIN != 1);
+
+void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map)
+{
+    /* Only call this function if kvm_arm_sve_supported() returns true. */
+    static uint64_t vls[KVM_ARM64_SVE_VLS_WORDS];
+    static bool probed;
+    uint32_t vq = 0;
+    int i, j;
+
+    bitmap_clear(map, 0, ARM_MAX_VQ);
+
+    /*
+     * KVM ensures all host CPUs support the same set of vector lengths.
+     * So we only need to create the scratch VCPUs once and then cache
+     * the results.
+     */
+    if (!probed) {
+        struct kvm_vcpu_init init = {
+            .target = -1,
+            .features[0] = (1 << KVM_ARM_VCPU_SVE),
+        };
+        struct kvm_one_reg reg = {
+            .id = KVM_REG_ARM64_SVE_VLS,
+            .addr = (uint64_t)&vls[0],
+        };
+        int fdarray[3], ret;
+
+        probed = true;
+
+        if (!kvm_arm_create_scratch_host_vcpu(NULL, fdarray, &init)) {
+            error_report("failed to create scratch VCPU with SVE enabled");
+            abort();
+        }
+        ret = ioctl(fdarray[2], KVM_GET_ONE_REG, &reg);
+        kvm_arm_destroy_scratch_host_vcpu(fdarray);
+        if (ret) {
+            error_report("failed to get KVM_REG_ARM64_SVE_VLS: %s",
+                         strerror(errno));
+            abort();
+        }
+
+        for (i = KVM_ARM64_SVE_VLS_WORDS - 1; i >= 0; --i) {
+            if (vls[i]) {
+                vq = 64 - clz64(vls[i]) + i * 64;
+                break;
+            }
+        }
+        if (vq > ARM_MAX_VQ) {
+            warn_report("KVM supports vector lengths larger than "
+                        "QEMU can enable");
+        }
+    }
+
+    for (i = 0; i < KVM_ARM64_SVE_VLS_WORDS; ++i) {
+        if (!vls[i]) {
+            continue;
+        }
+        for (j = 1; j <= 64; ++j) {
+            vq = j + i * 64;
+            if (vq > ARM_MAX_VQ) {
+                return;
+            }
+            if (vls[i] & (1UL << (j - 1))) {
+                set_bit(vq - 1, map);
+            }
+        }
+    }
+}
+
+static int kvm_arm_sve_set_vls(CPUState *cs)
+{
+    uint64_t vls[KVM_ARM64_SVE_VLS_WORDS] = {0};
+    struct kvm_one_reg reg = {
+        .id = KVM_REG_ARM64_SVE_VLS,
+        .addr = (uint64_t)&vls[0],
+    };
+    ARMCPU *cpu = ARM_CPU(cs);
+    uint32_t vq;
+    int i, j;
+
+    assert(cpu->sve_max_vq <= KVM_ARM64_SVE_VQ_MAX);
+
+    for (vq = 1; vq <= cpu->sve_max_vq; ++vq) {
+        if (test_bit(vq - 1, cpu->sve_vq_map)) {
+            i = (vq - 1) / 64;
+            j = (vq - 1) % 64;
+            vls[i] |= 1UL << j;
+        }
+    }
+
+    return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+}
+
 #define ARM_CPU_ID_MPIDR       3, 0, 0, 0, 5
 
 int kvm_arch_init_vcpu(CPUState *cs)
@@ -624,7 +718,7 @@ int kvm_arch_init_vcpu(CPUState *cs)
 
     if (cpu->kvm_target == QEMU_KVM_ARM_TARGET_NONE ||
         !object_dynamic_cast(OBJECT(cpu), TYPE_AARCH64_CPU)) {
-        fprintf(stderr, "KVM is not supported for this guest CPU type\n");
+        error_report("KVM is not supported for this guest CPU type");
         return -EINVAL;
     }
 
@@ -660,6 +754,10 @@ int kvm_arch_init_vcpu(CPUState *cs)
     }
 
     if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arm_sve_set_vls(cs);
+        if (ret) {
+            return ret;
+        }
         ret = kvm_arm_vcpu_finalize(cs, KVM_ARM_VCPU_SVE);
         if (ret) {
             return ret;
target/arm/kvm_arm.h +12 -0
index 1151877f97ea..a1cc6513f72b 100644
@@ -212,6 +212,17 @@ typedef struct ARMHostCPUFeatures {
  */
 bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf);
 
+/**
+ * kvm_arm_sve_get_vls:
+ * @cs: CPUState
+ * @map: bitmap to fill in
+ *
+ * Get all the SVE vector lengths supported by the KVM host, setting
+ * the bits corresponding to their length in quadwords minus one
+ * (vq - 1) in @map up to ARM_MAX_VQ.
+ */
+void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map);
+
 /**
  * kvm_arm_set_cpu_features_from_host:
  * @cpu: ARMCPU to set the features for
@@ -315,6 +326,7 @@ static inline int kvm_arm_vgic_probe(void)
 static inline void kvm_arm_pmu_set_irq(CPUState *cs, int irq) {}
 static inline void kvm_arm_pmu_init(CPUState *cs) {}
 
+static inline void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map) {}
 #endif
 
 static inline const char *gic_class_name(void)
tests/arm-cpu-features.c +100 -1
index d50f98cb6aea..a0129aebf409 100644
@@ -117,6 +117,17 @@ static QDict *resp_get_props(QDict *resp)
     return qdict;
 }
 
+static bool resp_get_feature(QDict *resp, const char *feature)
+{
+    QDict *props;
+
+    g_assert(resp);
+    g_assert(resp_has_props(resp));
+    props = resp_get_props(resp);
+    g_assert(qdict_get(props, feature));
+    return qdict_get_bool(props, feature);
+}
+
 #define assert_has_feature(qts, cpu_type, feature)                     \
 ({                                                                     \
     QDict *_resp = do_query_no_props(qts, cpu_type);                   \
@@ -336,6 +347,25 @@ static void sve_tests_sve_off(const void *data)
     qtest_quit(qts);
 }
 
+static void sve_tests_sve_off_kvm(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-accel kvm -cpu max,sve=off");
+
+    /*
+     * We don't know if this host supports SVE so we don't
+     * attempt to test enabling anything. We only test that
+     * everything is disabled (as it should be with sve=off)
+     * and that using sve<N>=off to explicitly disable vector
+     * lengths is OK too.
+     */
+    assert_sve_vls(qts, "max", 0, NULL);
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+
+    qtest_quit(qts);
+}
+
 static void test_query_cpu_model_expansion(const void *data)
 {
     QTestState *qts;
@@ -385,12 +415,81 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
     assert_has_feature(qts, "host", "pmu");
 
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
+        bool kvm_supports_sve;
+        char max_name[8], name[8];
+        uint32_t max_vq, vq;
+        uint64_t vls;
+        QDict *resp;
+        char *error;
+
         assert_has_feature(qts, "host", "aarch64");
-        assert_has_feature(qts, "max", "sve");
 
         assert_error(qts, "cortex-a15",
             "We cannot guarantee the CPU type 'cortex-a15' works "
             "with KVM on this host", NULL);
+
+        assert_has_feature(qts, "max", "sve");
+        resp = do_query_no_props(qts, "max");
+        kvm_supports_sve = resp_get_feature(resp, "sve");
+        vls = resp_get_sve_vls(resp);
+        qobject_unref(resp);
+
+        if (kvm_supports_sve) {
+            g_assert(vls != 0);
+            max_vq = 64 - __builtin_clzll(vls);
+            sprintf(max_name, "sve%d", max_vq * 128);
+
+            /* Enabling a supported length is of course fine. */
+            assert_sve_vls(qts, "max", vls, "{ %s: true }", max_name);
+
+            /* Get the next supported length smaller than max-vq. */
+            vq = 64 - __builtin_clzll(vls & ~BIT(max_vq - 1));
+            if (vq) {
+                /*
+                 * We have at least one length smaller than max-vq,
+                 * so we can disable max-vq.
+                 */
+                assert_sve_vls(qts, "max", (vls & ~BIT(max_vq - 1)),
+                               "{ %s: false }", max_name);
+
+                /*
+                 * Smaller, supported vector lengths cannot be disabled
+                 * unless all larger, supported vector lengths are also
+                 * disabled.
+                 */
+                sprintf(name, "sve%d", vq * 128);
+                error = g_strdup_printf("cannot disable %s", name);
+                assert_error(qts, "max", error,
+                             "{ %s: true, %s: false }",
+                             max_name, name);
+                g_free(error);
+            }
+
+            /*
+             * The smallest, supported vector length is required, because
+             * we need at least one vector length enabled.
+             */
+            vq = __builtin_ffsll(vls);
+            sprintf(name, "sve%d", vq * 128);
+            error = g_strdup_printf("cannot disable %s", name);
+            assert_error(qts, "max", error, "{ %s: false }", name);
+            g_free(error);
+
+            /* Get an unsupported length. */
+            for (vq = 1; vq <= max_vq; ++vq) {
+                if (!(vls & BIT(vq - 1))) {
+                    break;
+                }
+            }
+            if (vq <= SVE_MAX_VQ) {
+                sprintf(name, "sve%d", vq * 128);
+                error = g_strdup_printf("cannot enable %s", name);
+                assert_error(qts, "max", error, "{ %s: true }", name);
+                g_free(error);
+            }
+        } else {
+            g_assert(vls == 0);
+        }
     } else {
         assert_error(qts, "host",
                      "'pmu' feature not supported by KVM on this host",

tests/arm-cpu-features.c int main(int argc, char **argv) · line 535

Quoted 10 lines
tests/arm-cpu-features.c +4 -0
@@ -427,6 +526,10 @@ int main(int argc, char **argv)
     if (kvm_available) {
         qtest_add_data_func("/arm/kvm/query-cpu-model-expansion",
                             NULL, test_query_cpu_model_expansion_kvm);
+        if (g_str_equal(qtest_get_arch(), "aarch64")) {
+            qtest_add_data_func("/arm/kvm/query-cpu-model-expansion/sve-off",
+                                NULL, sve_tests_sve_off_kvm);
+        }
     }
 
     return g_test_run();

Besides,

Thanks

Eric

#25 lore raw

Re: [PATCH v4 8/9] target/arm/cpu64: max cpu: Support sve properties with KVM

Quoted 96 lines · Auger Eric

Hi Drew,

Quoted 93 lines · Andrew Jones

Extend the SVE vq map initialization and validation with KVM's supported vector lengths when KVM is enabled. In order to determine and select supported lengths we add two new KVM functions for getting and setting the KVM_REG_ARM64_SVE_VLS pseudo-register.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst |  36 +++++---
 target/arm/cpu64.c        | 167 +++++++++++++++++++++++++++++---------
 target/arm/kvm64.c        | 100 ++++++++++++++++++++++-
 target/arm/kvm_arm.h      |  12 +++
 tests/arm-cpu-features.c  | 105 +++++++++++++++++++++++-
 5 files changed, 368 insertions(+), 52 deletions(-)
docs/arm-cpu-features.rst +26 -10
index 1262fddc6201..939366f959cf 100644
@@ -188,10 +188,17 @@ SVE CPU Property Dependencies and Constraints
 
   1) At least one vector length must be enabled when `sve` is enabled.
 
-  2) If a vector length `N` is enabled, then all power-of-two vector
-     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
-     is enabled, then `sve128` and `sve256` must also be enabled,
-     but `sve384` is not required.
+  2) If a vector length `N` is enabled, then, when KVM is enabled, all
+     smaller, host supported vector lengths must also be enabled.  If
+     KVM is not enabled, then only all the smaller, power-of-two vector
+     lengths must be enabled.  E.g. with KVM if the host supports all
+     vector lengths up to 512-bits (128, 256, 384, 512), then if
+     `sve512` is enabled, `sve128`, `sve256`, and `sve384` must also
+     be enabled. Without KVM, `sve384` would not be required.
+
+  3) If KVM is enabled then only vector lengths that the host CPU type
+     support may be enabled.  If SVE is not supported by the host, then
+     no `sve*` properties may be enabled.
 
 SVE CPU Property Parsing Semantics
 ----------------------------------
@@ -210,20 +217,29 @@ SVE CPU Property Parsing Semantics
      disable the last enabled vector length (see constraint (1) of "SVE
      CPU Property Dependencies and Constraints").
 
-  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+  4) When KVM is enabled, if the host does not support SVE, then an error
+     is generated when attempting to enable any `sve*` properties.
+
+  5) When KVM is enabled, if the host does support SVE, then an error is
+     generated when attempting to enable any vector lengths not supported
+     by the host.
+
+  6) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
      CPU properties are set `on`, then the specified vector lengths are
      disabled but the default for any unspecified lengths remains enabled.
-     Disabling a power-of-two vector length also disables all vector
-     lengths larger than the power-of-two length (see constraint (2) of
-     "SVE CPU Property Dependencies and Constraints").
+     When KVM is not enabled, disabling a power-of-two vector length also
+     disables all vector lengths larger than the power-of-two length.
+     When KVM is enabled, then disabling any supported vector length also
+     disables all larger vector lengths (see constraint (2) of "SVE CPU
+     Property Dependencies and Constraints").
 
-  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+  7) If one or more `sve<N>` CPU properties are set to `on`, then they
      are enabled and all unspecified lengths default to disabled, except
      for the required lengths per constraint (2) of "SVE CPU Property
      Dependencies and Constraints", which will even be auto-enabled if
      they were not explicitly enabled.
 
-  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+  8) If SVE was disabled (`sve=off`), allowing all vector lengths to be
      explicitly disabled (i.e. avoiding the error specified in (3) of
      "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
      provided an error will be generated.  To avoid this error, one must
target/arm/cpu64.c +1 -0
index b7eff4e1e107..18dd5e24ec61 100644
@@ -273,9 +273,18 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
      * any of the above.  Finally, if SVE is not disabled, then at least one
      * vector length must be enabled.
      */
+    DECLARE_BITMAP(kvm_supported, ARM_MAX_VQ);
     DECLARE_BITMAP(tmp, ARM_MAX_VQ);
     uint32_t vq, max_vq = 0;
 
Quoted 7 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 285

Quoted 6 lines
target/arm/cpu64.c +6 -0
+    /* Collect the set of vector lengths supported by KVM. */
+    bitmap_zero(kvm_supported, ARM_MAX_VQ);
+    if (kvm_enabled() && kvm_arm_sve_supported(CPU(cpu))) {
+        kvm_arm_sve_get_vls(CPU(cpu), kvm_supported);
+    } else if (kvm_enabled()) {
+        assert(!cpu_isar_feature(aa64_sve, cpu));

why not set an error and propagate it instead?

This should never happen. We shouldn't be here if KVM is enabled and SVE isn't supported. The question is how defensive do we want QEMU code? We could just drop the check altogether if we don't want the assert, but I'd rather keep it.

Quoted 52 lines
Quoted 52 lines
target/arm/cpu64.c +22 -16
+    }
+
     /*
      * Process explicit sve<N> properties.
      * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
@@ -293,10 +302,19 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Propagate enabled bits down through required powers-of-two. */
-        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
-            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
-                set_bit(vq - 1, cpu->sve_vq_map);
+        if (kvm_enabled()) {
+            /*
+             * For KVM we have to automatically enable all supported unitialized
+             * lengths, even when the smaller lengths are not all powers-of-two.
+             */
+            bitmap_andnot(tmp, kvm_supported, cpu->sve_vq_init, max_vq);
+            bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+        } else {
+            /* Propagate enabled bits down through required powers-of-two. */
+            for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+                if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                    set_bit(vq - 1, cpu->sve_vq_map);
+                }
             }
         }
     } else if (cpu->sve_max_vq == 0) {
@@ -308,23 +326,46 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Disabling a power-of-two disables all larger lengths. */
-        if (test_bit(0, cpu->sve_vq_init)) {
-            error_setg(errp, "cannot disable sve128");
-            error_append_hint(errp, "Disabling sve128 results in all vector "
-                              "lengths being disabled.\n");
-            error_append_hint(errp, "With SVE enabled, at least one vector "
-                              "length must be enabled.\n");
-            return;
-        }
-        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
-            if (test_bit(vq - 1, cpu->sve_vq_init)) {
-                break;
+        if (kvm_enabled()) {
+            /* Disabling a supported length disables all larger lengths. */
+            for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+                if (test_bit(vq - 1, cpu->sve_vq_init) &&
+                    test_bit(vq - 1, kvm_supported)) {
+                    break;
+                }
Quoted 8 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 342

Quoted 7 lines
target/arm/cpu64.c +7 -0
+            }
+            max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+            bitmap_andnot(cpu->sve_vq_map, kvm_supported,
+                          cpu->sve_vq_init, max_vq);
+            if (max_vq == 0 || bitmap_empty(cpu->sve_vq_map, max_vq)) {
+                vq = find_next_bit(kvm_supported, ARM_MAX_VQ, 0) + 1;
+                error_setg(errp, "cannot disable sve%d", vq * 128);

isn't the one disabled max_vq? Do you really need to re-compute vq?

We're not looking for max_vq with find_next_bit(), we're looking for "min_vq". Here we're checking that we always have at least one vector length enabled. You can't disable the minimum, because that's the last one.

Quoted 60 lines
Quoted 60 lines
target/arm/cpu64.c +40 -10
+                error_append_hint(errp, "Disabling sve%d results in all "
+                                  "vector lengths being disabled.\n",
+                                  vq * 128);
+                error_append_hint(errp, "With SVE enabled, at least one "
+                                  "vector length must be enabled.\n");
+                return;> +            }
+        } else {
+            /* Disabling a power-of-two disables all larger lengths. */
+            if (test_bit(0, cpu->sve_vq_init)) {
+                error_setg(errp, "cannot disable sve128");
+                error_append_hint(errp, "Disabling sve128 results in all "
+                                  "vector lengths being disabled.\n");
+                error_append_hint(errp, "With SVE enabled, at least one "
+                                  "vector length must be enabled.\n");
+                return;
+            }
+            for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
+                if (test_bit(vq - 1, cpu->sve_vq_init)) {
+                    break;
+                }
             }
+            max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+            bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
         }
-        max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
 
-        bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
         max_vq = find_last_bit(cpu->sve_vq_map, max_vq) + 1;
     }
 
@@ -358,16 +399,48 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
     assert(max_vq != 0);
     bitmap_clear(cpu->sve_vq_map, max_vq, ARM_MAX_VQ - max_vq);
 
-    /* Ensure all required powers-of-two are enabled. */
-    for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
-        if (!test_bit(vq - 1, cpu->sve_vq_map)) {
-            error_setg(errp, "cannot disable sve%d", vq * 128);
-            error_append_hint(errp, "sve%d is required as it "
-                              "is a power-of-two length smaller than "
-                              "the maximum, sve%d\n",
-                              vq * 128, max_vq * 128);
+    if (kvm_enabled()) {
+        /* Ensure the set of lengths matches what KVM supports. */
+        bitmap_xor(tmp, cpu->sve_vq_map, kvm_supported, max_vq);
+        if (!bitmap_empty(tmp, max_vq)) {
+            vq = find_last_bit(tmp, max_vq) + 1;
+            if (test_bit(vq - 1, cpu->sve_vq_map)) {
+                if (cpu->sve_max_vq) {
+                    error_setg(errp, "cannot set sve-max-vq=%d",
+                               cpu->sve_max_vq);
+                    error_append_hint(errp, "This KVM host does not support "
+                                      "the vector length %d-bits.\n",
+                                      vq * 128);
+                    error_append_hint(errp, "It may not be possible to use "
+                                      "sve-max-vq with this KVM host. Try "
+                                      "using only sve<N> properties.\n");
+                } else {
+                    error_setg(errp, "cannot enable sve%d", vq * 128);
+                    error_append_hint(errp, "This KVM host does not support "
Quoted 8 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 424

Quoted 5 lines
target/arm/cpu64.c +5 -0
+                                      "the vector length %d-bits.\n",
+                                      vq * 128);
+                }
+            } else {
+                error_setg(errp, "cannot disable sve%d", vq * 128);

I am not sure about the above message. what is the vq value supposed to be? For instance if we previously entered the
"if (!bitmap_empty(cpu->sve_vq_map, ARM_MAX_VQ)) {" path

'vq' here should be enabled because it's a dependency for the max-vq. If the user had enabled it or left it uninitialized (allowing it to be auto-enabled) then we wouldn't be here. To get here the user must have explicitly disabled it. Hence the 'cannot disable' message.

Quoted 388 lines
Quoted 388 lines
target/arm/cpu64.c +45 -14
+                error_append_hint(errp, "The KVM host requires all "
+                                  "supported vector lengths smaller "
+                                  "than %d bits to also be enabled.\n",
+                                  max_vq * 128);
+            }
             return;
         }
+    } else {
+        /* Ensure all required powers-of-two are enabled. */
+        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+            if (!test_bit(vq - 1, cpu->sve_vq_map)) {
+                error_setg(errp, "cannot disable sve%d", vq * 128);
+                error_append_hint(errp, "sve%d is required as it "
+                                  "is a power-of-two length smaller than "
+                                  "the maximum, sve%d\n",
+                                  vq * 128, max_vq * 128);
+                return;
+            }
+        }
     }
 
     /* From now on sve_max_vq is the actual maximum supported length. */
@@ -411,13 +484,22 @@ static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,
     Error *err = NULL;
 
     visit_type_uint32(v, name, &cpu->sve_max_vq, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
 
-    if (!err && (cpu->sve_max_vq == 0 || cpu->sve_max_vq > ARM_MAX_VQ)) {
-        error_setg(&err, "unsupported SVE vector length");
-        error_append_hint(&err, "Valid sve-max-vq in range [1-%d]\n",
+    if (kvm_enabled() && !kvm_arm_sve_supported(CPU(cpu))) {
+        error_setg(errp, "cannot set sve-max-vq");
+        error_append_hint(errp, "SVE not supported by KVM on this host\n");
+        return;
+    }
+
+    if (cpu->sve_max_vq == 0 || cpu->sve_max_vq > ARM_MAX_VQ) {
+        error_setg(errp, "unsupported SVE vector length");
+        error_append_hint(errp, "Valid sve-max-vq in range [1-%d]\n",
                           ARM_MAX_VQ);
     }
-    error_propagate(errp, err);
 }
 
 static void cpu_arm_get_sve_vq(Object *obj, Visitor *v, const char *name,
@@ -450,6 +532,12 @@ static void cpu_arm_set_sve_vq(Object *obj, Visitor *v, const char *name,
         return;
     }
 
+    if (value && kvm_enabled() && !kvm_arm_sve_supported(CPU(cpu))) {
+        error_setg(errp, "cannot enable %s", name);
+        error_append_hint(errp, "SVE not supported by KVM on this host\n");
+        return;
+    }
+
     if (value) {
         set_bit(vq - 1, cpu->sve_vq_map);
     } else {
@@ -607,20 +695,19 @@ static void aarch64_max_initfn(Object *obj)
         cpu->ctr = 0x80038003; /* 32 byte I and D cacheline size, VIPT icache */
         cpu->dcz_blocksize = 7; /*  512 bytes */
 #endif
-
-        object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
-                            cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
-
-        for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
-            char name[8];
-            sprintf(name, "sve%d", vq * 128);
-            object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
-                                cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
-        }
     }
 
     object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
                         cpu_arm_set_sve, NULL, NULL, &error_fatal);
+    object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
+                        cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
+
+    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+        char name[8];
+        sprintf(name, "sve%d", vq * 128);
+        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+    }
 }
 
 struct ARMCPUInfo {
target/arm/kvm64.c +99 -1
index f96649ae0349..cff4217a8469 100644
@@ -613,6 +613,100 @@ bool kvm_arm_sve_supported(CPUState *cpu)
     return kvm_check_extension(s, KVM_CAP_ARM_SVE);
 }
 
+QEMU_BUILD_BUG_ON(KVM_ARM64_SVE_VQ_MIN != 1);
+
+void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map)
+{
+    /* Only call this function if kvm_arm_sve_supported() returns true. */
+    static uint64_t vls[KVM_ARM64_SVE_VLS_WORDS];
+    static bool probed;
+    uint32_t vq = 0;
+    int i, j;
+
+    bitmap_clear(map, 0, ARM_MAX_VQ);
+
+    /*
+     * KVM ensures all host CPUs support the same set of vector lengths.
+     * So we only need to create the scratch VCPUs once and then cache
+     * the results.
+     */
+    if (!probed) {
+        struct kvm_vcpu_init init = {
+            .target = -1,
+            .features[0] = (1 << KVM_ARM_VCPU_SVE),
+        };
+        struct kvm_one_reg reg = {
+            .id = KVM_REG_ARM64_SVE_VLS,
+            .addr = (uint64_t)&vls[0],
+        };
+        int fdarray[3], ret;
+
+        probed = true;
+
+        if (!kvm_arm_create_scratch_host_vcpu(NULL, fdarray, &init)) {
+            error_report("failed to create scratch VCPU with SVE enabled");
+            abort();
+        }
+        ret = ioctl(fdarray[2], KVM_GET_ONE_REG, &reg);
+        kvm_arm_destroy_scratch_host_vcpu(fdarray);
+        if (ret) {
+            error_report("failed to get KVM_REG_ARM64_SVE_VLS: %s",
+                         strerror(errno));
+            abort();
+        }
+
+        for (i = KVM_ARM64_SVE_VLS_WORDS - 1; i >= 0; --i) {
+            if (vls[i]) {
+                vq = 64 - clz64(vls[i]) + i * 64;
+                break;
+            }
+        }
+        if (vq > ARM_MAX_VQ) {
+            warn_report("KVM supports vector lengths larger than "
+                        "QEMU can enable");
+        }
+    }
+
+    for (i = 0; i < KVM_ARM64_SVE_VLS_WORDS; ++i) {
+        if (!vls[i]) {
+            continue;
+        }
+        for (j = 1; j <= 64; ++j) {
+            vq = j + i * 64;
+            if (vq > ARM_MAX_VQ) {
+                return;
+            }
+            if (vls[i] & (1UL << (j - 1))) {
+                set_bit(vq - 1, map);
+            }
+        }
+    }
+}
+
+static int kvm_arm_sve_set_vls(CPUState *cs)
+{
+    uint64_t vls[KVM_ARM64_SVE_VLS_WORDS] = {0};
+    struct kvm_one_reg reg = {
+        .id = KVM_REG_ARM64_SVE_VLS,
+        .addr = (uint64_t)&vls[0],
+    };
+    ARMCPU *cpu = ARM_CPU(cs);
+    uint32_t vq;
+    int i, j;
+
+    assert(cpu->sve_max_vq <= KVM_ARM64_SVE_VQ_MAX);
+
+    for (vq = 1; vq <= cpu->sve_max_vq; ++vq) {
+        if (test_bit(vq - 1, cpu->sve_vq_map)) {
+            i = (vq - 1) / 64;
+            j = (vq - 1) % 64;
+            vls[i] |= 1UL << j;
+        }
+    }
+
+    return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+}
+
 #define ARM_CPU_ID_MPIDR       3, 0, 0, 0, 5
 
 int kvm_arch_init_vcpu(CPUState *cs)
@@ -624,7 +718,7 @@ int kvm_arch_init_vcpu(CPUState *cs)
 
     if (cpu->kvm_target == QEMU_KVM_ARM_TARGET_NONE ||
         !object_dynamic_cast(OBJECT(cpu), TYPE_AARCH64_CPU)) {
-        fprintf(stderr, "KVM is not supported for this guest CPU type\n");
+        error_report("KVM is not supported for this guest CPU type");
         return -EINVAL;
     }
 
@@ -660,6 +754,10 @@ int kvm_arch_init_vcpu(CPUState *cs)
     }
 
     if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arm_sve_set_vls(cs);
+        if (ret) {
+            return ret;
+        }
         ret = kvm_arm_vcpu_finalize(cs, KVM_ARM_VCPU_SVE);
         if (ret) {
             return ret;
target/arm/kvm_arm.h +12 -0
index 1151877f97ea..a1cc6513f72b 100644
@@ -212,6 +212,17 @@ typedef struct ARMHostCPUFeatures {
  */
 bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf);
 
+/**
+ * kvm_arm_sve_get_vls:
+ * @cs: CPUState
+ * @map: bitmap to fill in
+ *
+ * Get all the SVE vector lengths supported by the KVM host, setting
+ * the bits corresponding to their length in quadwords minus one
+ * (vq - 1) in @map up to ARM_MAX_VQ.
+ */
+void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map);
+
 /**
  * kvm_arm_set_cpu_features_from_host:
  * @cpu: ARMCPU to set the features for
@@ -315,6 +326,7 @@ static inline int kvm_arm_vgic_probe(void)
 static inline void kvm_arm_pmu_set_irq(CPUState *cs, int irq) {}
 static inline void kvm_arm_pmu_init(CPUState *cs) {}
 
+static inline void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map) {}
 #endif
 
 static inline const char *gic_class_name(void)
tests/arm-cpu-features.c +104 -1
index d50f98cb6aea..a0129aebf409 100644
@@ -117,6 +117,17 @@ static QDict *resp_get_props(QDict *resp)
     return qdict;
 }
 
+static bool resp_get_feature(QDict *resp, const char *feature)
+{
+    QDict *props;
+
+    g_assert(resp);
+    g_assert(resp_has_props(resp));
+    props = resp_get_props(resp);
+    g_assert(qdict_get(props, feature));
+    return qdict_get_bool(props, feature);
+}
+
 #define assert_has_feature(qts, cpu_type, feature)                     \
 ({                                                                     \
     QDict *_resp = do_query_no_props(qts, cpu_type);                   \
@@ -336,6 +347,25 @@ static void sve_tests_sve_off(const void *data)
     qtest_quit(qts);
 }
 
+static void sve_tests_sve_off_kvm(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-accel kvm -cpu max,sve=off");
+
+    /*
+     * We don't know if this host supports SVE so we don't
+     * attempt to test enabling anything. We only test that
+     * everything is disabled (as it should be with sve=off)
+     * and that using sve<N>=off to explicitly disable vector
+     * lengths is OK too.
+     */
+    assert_sve_vls(qts, "max", 0, NULL);
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+
+    qtest_quit(qts);
+}
+
 static void test_query_cpu_model_expansion(const void *data)
 {
     QTestState *qts;
@@ -385,12 +415,81 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
     assert_has_feature(qts, "host", "pmu");
 
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
+        bool kvm_supports_sve;
+        char max_name[8], name[8];
+        uint32_t max_vq, vq;
+        uint64_t vls;
+        QDict *resp;
+        char *error;
+
         assert_has_feature(qts, "host", "aarch64");
-        assert_has_feature(qts, "max", "sve");
 
         assert_error(qts, "cortex-a15",
             "We cannot guarantee the CPU type 'cortex-a15' works "
             "with KVM on this host", NULL);
+
+        assert_has_feature(qts, "max", "sve");
+        resp = do_query_no_props(qts, "max");
+        kvm_supports_sve = resp_get_feature(resp, "sve");
+        vls = resp_get_sve_vls(resp);
+        qobject_unref(resp);
+
+        if (kvm_supports_sve) {
+            g_assert(vls != 0);
+            max_vq = 64 - __builtin_clzll(vls);
+            sprintf(max_name, "sve%d", max_vq * 128);
+
+            /* Enabling a supported length is of course fine. */
+            assert_sve_vls(qts, "max", vls, "{ %s: true }", max_name);
+
+            /* Get the next supported length smaller than max-vq. */
+            vq = 64 - __builtin_clzll(vls & ~BIT(max_vq - 1));
+            if (vq) {
+                /*
+                 * We have at least one length smaller than max-vq,
+                 * so we can disable max-vq.
+                 */
+                assert_sve_vls(qts, "max", (vls & ~BIT(max_vq - 1)),
+                               "{ %s: false }", max_name);
+
+                /*
+                 * Smaller, supported vector lengths cannot be disabled
+                 * unless all larger, supported vector lengths are also
+                 * disabled.
+                 */
+                sprintf(name, "sve%d", vq * 128);
+                error = g_strdup_printf("cannot disable %s", name);
+                assert_error(qts, "max", error,
+                             "{ %s: true, %s: false }",
+                             max_name, name);
+                g_free(error);
+            }
+
+            /*
+             * The smallest, supported vector length is required, because
+             * we need at least one vector length enabled.
+             */
+            vq = __builtin_ffsll(vls);
+            sprintf(name, "sve%d", vq * 128);
+            error = g_strdup_printf("cannot disable %s", name);
+            assert_error(qts, "max", error, "{ %s: false }", name);
+            g_free(error);
+
+            /* Get an unsupported length. */
+            for (vq = 1; vq <= max_vq; ++vq) {
+                if (!(vls & BIT(vq - 1))) {
+                    break;
+                }
+            }
+            if (vq <= SVE_MAX_VQ) {
+                sprintf(name, "sve%d", vq * 128);
+                error = g_strdup_printf("cannot enable %s", name);
+                assert_error(qts, "max", error, "{ %s: true }", name);
+                g_free(error);
+            }
+        } else {
+            g_assert(vls == 0);
+        }
     } else {
         assert_error(qts, "host",
                      "'pmu' feature not supported by KVM on this host",
@@ -427,6 +526,10 @@ int main(int argc, char **argv)
     if (kvm_available) {
         qtest_add_data_func("/arm/kvm/query-cpu-model-expansion",
                             NULL, test_query_cpu_model_expansion_kvm);
+        if (g_str_equal(qtest_get_arch(), "aarch64")) {
+            qtest_add_data_func("/arm/kvm/query-cpu-model-expansion/sve-off",
+                                NULL, sve_tests_sve_off_kvm);
+        }
     }
 
Quoted 6 lines

tests/arm-cpu-features.c int main(int argc, char **argv) · line 535

Quoted 2 lines
tests/arm-cpu-features.c +0 -0
     return g_test_run();

Besides,

Thanks!

drew

#26 lore raw

Re: [PATCH v4 8/9] target/arm/cpu64: max cpu: Support sve properties with KVM

Quoted 154 lines · Andrew Jones
Quoted 96 lines · Auger Eric

Hi Drew,

Quoted 93 lines · Andrew Jones

Extend the SVE vq map initialization and validation with KVM's supported vector lengths when KVM is enabled. In order to determine and select supported lengths we add two new KVM functions for getting and setting the KVM_REG_ARM64_SVE_VLS pseudo-register.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst |  36 +++++---
 target/arm/cpu64.c        | 167 +++++++++++++++++++++++++++++---------
 target/arm/kvm64.c        | 100 ++++++++++++++++++++++-
 target/arm/kvm_arm.h      |  12 +++
 tests/arm-cpu-features.c  | 105 +++++++++++++++++++++++-
 5 files changed, 368 insertions(+), 52 deletions(-)
docs/arm-cpu-features.rst +26 -10
index 1262fddc6201..939366f959cf 100644
@@ -188,10 +188,17 @@ SVE CPU Property Dependencies and Constraints
 
   1) At least one vector length must be enabled when `sve` is enabled.
 
-  2) If a vector length `N` is enabled, then all power-of-two vector
-     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
-     is enabled, then `sve128` and `sve256` must also be enabled,
-     but `sve384` is not required.
+  2) If a vector length `N` is enabled, then, when KVM is enabled, all
+     smaller, host supported vector lengths must also be enabled.  If
+     KVM is not enabled, then only all the smaller, power-of-two vector
+     lengths must be enabled.  E.g. with KVM if the host supports all
+     vector lengths up to 512-bits (128, 256, 384, 512), then if
+     `sve512` is enabled, `sve128`, `sve256`, and `sve384` must also
+     be enabled. Without KVM, `sve384` would not be required.
+
+  3) If KVM is enabled then only vector lengths that the host CPU type
+     support may be enabled.  If SVE is not supported by the host, then
+     no `sve*` properties may be enabled.
 
 SVE CPU Property Parsing Semantics
 ----------------------------------
@@ -210,20 +217,29 @@ SVE CPU Property Parsing Semantics
      disable the last enabled vector length (see constraint (1) of "SVE
      CPU Property Dependencies and Constraints").
 
-  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+  4) When KVM is enabled, if the host does not support SVE, then an error
+     is generated when attempting to enable any `sve*` properties.
+
+  5) When KVM is enabled, if the host does support SVE, then an error is
+     generated when attempting to enable any vector lengths not supported
+     by the host.
+
+  6) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
      CPU properties are set `on`, then the specified vector lengths are
      disabled but the default for any unspecified lengths remains enabled.
-     Disabling a power-of-two vector length also disables all vector
-     lengths larger than the power-of-two length (see constraint (2) of
-     "SVE CPU Property Dependencies and Constraints").
+     When KVM is not enabled, disabling a power-of-two vector length also
+     disables all vector lengths larger than the power-of-two length.
+     When KVM is enabled, then disabling any supported vector length also
+     disables all larger vector lengths (see constraint (2) of "SVE CPU
+     Property Dependencies and Constraints").
 
-  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+  7) If one or more `sve<N>` CPU properties are set to `on`, then they
      are enabled and all unspecified lengths default to disabled, except
      for the required lengths per constraint (2) of "SVE CPU Property
      Dependencies and Constraints", which will even be auto-enabled if
      they were not explicitly enabled.
 
-  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+  8) If SVE was disabled (`sve=off`), allowing all vector lengths to be
      explicitly disabled (i.e. avoiding the error specified in (3) of
      "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
      provided an error will be generated.  To avoid this error, one must
target/arm/cpu64.c +1 -0
index b7eff4e1e107..18dd5e24ec61 100644
@@ -273,9 +273,18 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
      * any of the above.  Finally, if SVE is not disabled, then at least one
      * vector length must be enabled.
      */
+    DECLARE_BITMAP(kvm_supported, ARM_MAX_VQ);
     DECLARE_BITMAP(tmp, ARM_MAX_VQ);
     uint32_t vq, max_vq = 0;
 
Quoted 7 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 285

Quoted 6 lines
target/arm/cpu64.c +6 -0
+    /* Collect the set of vector lengths supported by KVM. */
+    bitmap_zero(kvm_supported, ARM_MAX_VQ);
+    if (kvm_enabled() && kvm_arm_sve_supported(CPU(cpu))) {
+        kvm_arm_sve_get_vls(CPU(cpu), kvm_supported);
+    } else if (kvm_enabled()) {
+        assert(!cpu_isar_feature(aa64_sve, cpu));

why not set an error and propagate it instead?

This should never happen. We shouldn't be here if KVM is enabled and SVE isn't supported. The question is how defensive do we want QEMU code? We could just drop the check altogether if we don't want the assert, but I'd rather keep it.

Quoted 44 lines
Quoted 44 lines
target/arm/cpu64.c +15 -15
+    }
+
     /*
      * Process explicit sve<N> properties.
      * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
@@ -293,10 +302,19 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Propagate enabled bits down through required powers-of-two. */
-        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
-            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
-                set_bit(vq - 1, cpu->sve_vq_map);
+        if (kvm_enabled()) {
+            /*
+             * For KVM we have to automatically enable all supported unitialized
+             * lengths, even when the smaller lengths are not all powers-of-two.
+             */
+            bitmap_andnot(tmp, kvm_supported, cpu->sve_vq_init, max_vq);
+            bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+        } else {
+            /* Propagate enabled bits down through required powers-of-two. */
+            for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+                if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                    set_bit(vq - 1, cpu->sve_vq_map);
+                }
             }
         }
     } else if (cpu->sve_max_vq == 0) {
@@ -308,23 +326,46 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Disabling a power-of-two disables all larger lengths. */
-        if (test_bit(0, cpu->sve_vq_init)) {
-            error_setg(errp, "cannot disable sve128");
-            error_append_hint(errp, "Disabling sve128 results in all vector "
-                              "lengths being disabled.\n");
-            error_append_hint(errp, "With SVE enabled, at least one vector "
-                              "length must be enabled.\n");
-            return;
-        }
-        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
-            if (test_bit(vq - 1, cpu->sve_vq_init)) {
Quoted 8 lines
Quoted 8 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 335

Quoted 8 lines
target/arm/cpu64.c +7 -1
-                break;
+        if (kvm_enabled()) {
+            /* Disabling a supported length disables all larger lengths. */
+            for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+                if (test_bit(vq - 1, cpu->sve_vq_init) &&
+                    test_bit(vq - 1, kvm_supported)) {
+                    break;
+                }

the above loop looks for the 1st disabled vq that is also supported, right?

Quoted 5 lines
Quoted 5 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 340

Quoted 5 lines
target/arm/cpu64.c +5 -0
+            }
+            max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+            bitmap_andnot(cpu->sve_vq_map, kvm_supported,
+                          cpu->sve_vq_init, max_vq);
+            if (max_vq == 0 || bitmap_empty(cpu->sve_vq_map, max_vq)) {

here we don't have anything enabled below the disabled one. So don't we have the culprit already?

Quoted 78 lines
Quoted 3 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 342

Quoted 2 lines
target/arm/cpu64.c +2 -0
+                vq = find_next_bit(kvm_supported, ARM_MAX_VQ, 0) + 1;
+                error_setg(errp, "cannot disable sve%d", vq * 128);

isn't the one disabled max_vq? Do you really need to re-compute vq?

We're not looking for max_vq with find_next_bit(), we're looking for "min_vq". Here we're checking that we always have at least one vector length enabled. You can't disable the minimum, because that's the last one.

Quoted 60 lines
Quoted 60 lines
target/arm/cpu64.c +40 -10
+                error_append_hint(errp, "Disabling sve%d results in all "
+                                  "vector lengths being disabled.\n",
+                                  vq * 128);
+                error_append_hint(errp, "With SVE enabled, at least one "
+                                  "vector length must be enabled.\n");
+                return;> +            }
+        } else {
+            /* Disabling a power-of-two disables all larger lengths. */
+            if (test_bit(0, cpu->sve_vq_init)) {
+                error_setg(errp, "cannot disable sve128");
+                error_append_hint(errp, "Disabling sve128 results in all "
+                                  "vector lengths being disabled.\n");
+                error_append_hint(errp, "With SVE enabled, at least one "
+                                  "vector length must be enabled.\n");
+                return;
+            }
+            for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
+                if (test_bit(vq - 1, cpu->sve_vq_init)) {
+                    break;
+                }
             }
+            max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+            bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
         }
-        max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
 
-        bitmap_complement(cpu->sve_vq_map, cpu->sve_vq_init, max_vq);
         max_vq = find_last_bit(cpu->sve_vq_map, max_vq) + 1;
     }
 
@@ -358,16 +399,48 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
     assert(max_vq != 0);
     bitmap_clear(cpu->sve_vq_map, max_vq, ARM_MAX_VQ - max_vq);
 
-    /* Ensure all required powers-of-two are enabled. */
-    for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
-        if (!test_bit(vq - 1, cpu->sve_vq_map)) {
-            error_setg(errp, "cannot disable sve%d", vq * 128);
-            error_append_hint(errp, "sve%d is required as it "
-                              "is a power-of-two length smaller than "
-                              "the maximum, sve%d\n",
-                              vq * 128, max_vq * 128);
+    if (kvm_enabled()) {
+        /* Ensure the set of lengths matches what KVM supports. */
+        bitmap_xor(tmp, cpu->sve_vq_map, kvm_supported, max_vq);
+        if (!bitmap_empty(tmp, max_vq)) {
+            vq = find_last_bit(tmp, max_vq) + 1;
+            if (test_bit(vq - 1, cpu->sve_vq_map)) {
+                if (cpu->sve_max_vq) {
+                    error_setg(errp, "cannot set sve-max-vq=%d",
+                               cpu->sve_max_vq);
+                    error_append_hint(errp, "This KVM host does not support "
+                                      "the vector length %d-bits.\n",
+                                      vq * 128);
+                    error_append_hint(errp, "It may not be possible to use "
+                                      "sve-max-vq with this KVM host. Try "
+                                      "using only sve<N> properties.\n");
+                } else {
+                    error_setg(errp, "cannot enable sve%d", vq * 128);
+                    error_append_hint(errp, "This KVM host does not support "
Quoted 8 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 424

Quoted 5 lines
target/arm/cpu64.c +5 -0
+                                      "the vector length %d-bits.\n",
+                                      vq * 128);
+                }
+            } else {
+                error_setg(errp, "cannot disable sve%d", vq * 128);

I am not sure about the above message. what is the vq value supposed to be? For instance if we previously entered the
"if (!bitmap_empty(cpu->sve_vq_map, ARM_MAX_VQ)) {" path

Quoted 4 lines

'vq' here should be enabled because it's a dependency for the max-vq. If the user had enabled it or left it uninitialized (allowing it to be auto-enabled) then we wouldn't be here. To get here the user must have explicitly disabled it. Hence the 'cannot disable' message.

OK you're right. Got it.

Eric

Quoted 400 lines
Quoted 389 lines
Quoted 389 lines
target/arm/cpu64.c +45 -14
+                error_append_hint(errp, "The KVM host requires all "
+                                  "supported vector lengths smaller "
+                                  "than %d bits to also be enabled.\n",
+                                  max_vq * 128);
+            }
             return;
         }
+    } else {
+        /* Ensure all required powers-of-two are enabled. */
+        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+            if (!test_bit(vq - 1, cpu->sve_vq_map)) {
+                error_setg(errp, "cannot disable sve%d", vq * 128);
+                error_append_hint(errp, "sve%d is required as it "
+                                  "is a power-of-two length smaller than "
+                                  "the maximum, sve%d\n",
+                                  vq * 128, max_vq * 128);
+                return;
+            }
+        }
     }
 
     /* From now on sve_max_vq is the actual maximum supported length. */
@@ -411,13 +484,22 @@ static void cpu_max_set_sve_max_vq(Object *obj, Visitor *v, const char *name,
     Error *err = NULL;
 
     visit_type_uint32(v, name, &cpu->sve_max_vq, &err);
+    if (err) {
+        error_propagate(errp, err);
+        return;
+    }
 
-    if (!err && (cpu->sve_max_vq == 0 || cpu->sve_max_vq > ARM_MAX_VQ)) {
-        error_setg(&err, "unsupported SVE vector length");
-        error_append_hint(&err, "Valid sve-max-vq in range [1-%d]\n",
+    if (kvm_enabled() && !kvm_arm_sve_supported(CPU(cpu))) {
+        error_setg(errp, "cannot set sve-max-vq");
+        error_append_hint(errp, "SVE not supported by KVM on this host\n");
+        return;
+    }
+
+    if (cpu->sve_max_vq == 0 || cpu->sve_max_vq > ARM_MAX_VQ) {
+        error_setg(errp, "unsupported SVE vector length");
+        error_append_hint(errp, "Valid sve-max-vq in range [1-%d]\n",
                           ARM_MAX_VQ);
     }
-    error_propagate(errp, err);
 }
 
 static void cpu_arm_get_sve_vq(Object *obj, Visitor *v, const char *name,
@@ -450,6 +532,12 @@ static void cpu_arm_set_sve_vq(Object *obj, Visitor *v, const char *name,
         return;
     }
 
+    if (value && kvm_enabled() && !kvm_arm_sve_supported(CPU(cpu))) {
+        error_setg(errp, "cannot enable %s", name);
+        error_append_hint(errp, "SVE not supported by KVM on this host\n");
+        return;
+    }
+
     if (value) {
         set_bit(vq - 1, cpu->sve_vq_map);
     } else {
@@ -607,20 +695,19 @@ static void aarch64_max_initfn(Object *obj)
         cpu->ctr = 0x80038003; /* 32 byte I and D cacheline size, VIPT icache */
         cpu->dcz_blocksize = 7; /*  512 bytes */
 #endif
-
-        object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
-                            cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
-
-        for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
-            char name[8];
-            sprintf(name, "sve%d", vq * 128);
-            object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
-                                cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
-        }
     }
 
     object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
                         cpu_arm_set_sve, NULL, NULL, &error_fatal);
+    object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
+                        cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
+
+    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+        char name[8];
+        sprintf(name, "sve%d", vq * 128);
+        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+    }
 }
 
 struct ARMCPUInfo {
target/arm/kvm64.c +99 -1
index f96649ae0349..cff4217a8469 100644
@@ -613,6 +613,100 @@ bool kvm_arm_sve_supported(CPUState *cpu)
     return kvm_check_extension(s, KVM_CAP_ARM_SVE);
 }
 
+QEMU_BUILD_BUG_ON(KVM_ARM64_SVE_VQ_MIN != 1);
+
+void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map)
+{
+    /* Only call this function if kvm_arm_sve_supported() returns true. */
+    static uint64_t vls[KVM_ARM64_SVE_VLS_WORDS];
+    static bool probed;
+    uint32_t vq = 0;
+    int i, j;
+
+    bitmap_clear(map, 0, ARM_MAX_VQ);
+
+    /*
+     * KVM ensures all host CPUs support the same set of vector lengths.
+     * So we only need to create the scratch VCPUs once and then cache
+     * the results.
+     */
+    if (!probed) {
+        struct kvm_vcpu_init init = {
+            .target = -1,
+            .features[0] = (1 << KVM_ARM_VCPU_SVE),
+        };
+        struct kvm_one_reg reg = {
+            .id = KVM_REG_ARM64_SVE_VLS,
+            .addr = (uint64_t)&vls[0],
+        };
+        int fdarray[3], ret;
+
+        probed = true;
+
+        if (!kvm_arm_create_scratch_host_vcpu(NULL, fdarray, &init)) {
+            error_report("failed to create scratch VCPU with SVE enabled");
+            abort();
+        }
+        ret = ioctl(fdarray[2], KVM_GET_ONE_REG, &reg);
+        kvm_arm_destroy_scratch_host_vcpu(fdarray);
+        if (ret) {
+            error_report("failed to get KVM_REG_ARM64_SVE_VLS: %s",
+                         strerror(errno));
+            abort();
+        }
+
+        for (i = KVM_ARM64_SVE_VLS_WORDS - 1; i >= 0; --i) {
+            if (vls[i]) {
+                vq = 64 - clz64(vls[i]) + i * 64;
+                break;
+            }
+        }
+        if (vq > ARM_MAX_VQ) {
+            warn_report("KVM supports vector lengths larger than "
+                        "QEMU can enable");
+        }
+    }
+
+    for (i = 0; i < KVM_ARM64_SVE_VLS_WORDS; ++i) {
+        if (!vls[i]) {
+            continue;
+        }
+        for (j = 1; j <= 64; ++j) {
+            vq = j + i * 64;
+            if (vq > ARM_MAX_VQ) {
+                return;
+            }
+            if (vls[i] & (1UL << (j - 1))) {
+                set_bit(vq - 1, map);
+            }
+        }
+    }
+}
+
+static int kvm_arm_sve_set_vls(CPUState *cs)
+{
+    uint64_t vls[KVM_ARM64_SVE_VLS_WORDS] = {0};
+    struct kvm_one_reg reg = {
+        .id = KVM_REG_ARM64_SVE_VLS,
+        .addr = (uint64_t)&vls[0],
+    };
+    ARMCPU *cpu = ARM_CPU(cs);
+    uint32_t vq;
+    int i, j;
+
+    assert(cpu->sve_max_vq <= KVM_ARM64_SVE_VQ_MAX);
+
+    for (vq = 1; vq <= cpu->sve_max_vq; ++vq) {
+        if (test_bit(vq - 1, cpu->sve_vq_map)) {
+            i = (vq - 1) / 64;
+            j = (vq - 1) % 64;
+            vls[i] |= 1UL << j;
+        }
+    }
+
+    return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+}
+
 #define ARM_CPU_ID_MPIDR       3, 0, 0, 0, 5
 
 int kvm_arch_init_vcpu(CPUState *cs)
@@ -624,7 +718,7 @@ int kvm_arch_init_vcpu(CPUState *cs)
 
     if (cpu->kvm_target == QEMU_KVM_ARM_TARGET_NONE ||
         !object_dynamic_cast(OBJECT(cpu), TYPE_AARCH64_CPU)) {
-        fprintf(stderr, "KVM is not supported for this guest CPU type\n");
+        error_report("KVM is not supported for this guest CPU type");
         return -EINVAL;
     }
 
@@ -660,6 +754,10 @@ int kvm_arch_init_vcpu(CPUState *cs)
     }
 
     if (cpu_isar_feature(aa64_sve, cpu)) {
+        ret = kvm_arm_sve_set_vls(cs);
+        if (ret) {
+            return ret;
+        }
         ret = kvm_arm_vcpu_finalize(cs, KVM_ARM_VCPU_SVE);
         if (ret) {
             return ret;
target/arm/kvm_arm.h +12 -0
index 1151877f97ea..a1cc6513f72b 100644
@@ -212,6 +212,17 @@ typedef struct ARMHostCPUFeatures {
  */
 bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf);
 
+/**
+ * kvm_arm_sve_get_vls:
+ * @cs: CPUState
+ * @map: bitmap to fill in
+ *
+ * Get all the SVE vector lengths supported by the KVM host, setting
+ * the bits corresponding to their length in quadwords minus one
+ * (vq - 1) in @map up to ARM_MAX_VQ.
+ */
+void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map);
+
 /**
  * kvm_arm_set_cpu_features_from_host:
  * @cpu: ARMCPU to set the features for
@@ -315,6 +326,7 @@ static inline int kvm_arm_vgic_probe(void)
 static inline void kvm_arm_pmu_set_irq(CPUState *cs, int irq) {}
 static inline void kvm_arm_pmu_init(CPUState *cs) {}
 
+static inline void kvm_arm_sve_get_vls(CPUState *cs, unsigned long *map) {}
 #endif
 
 static inline const char *gic_class_name(void)
tests/arm-cpu-features.c +104 -1
index d50f98cb6aea..a0129aebf409 100644
@@ -117,6 +117,17 @@ static QDict *resp_get_props(QDict *resp)
     return qdict;
 }
 
+static bool resp_get_feature(QDict *resp, const char *feature)
+{
+    QDict *props;
+
+    g_assert(resp);
+    g_assert(resp_has_props(resp));
+    props = resp_get_props(resp);
+    g_assert(qdict_get(props, feature));
+    return qdict_get_bool(props, feature);
+}
+
 #define assert_has_feature(qts, cpu_type, feature)                     \
 ({                                                                     \
     QDict *_resp = do_query_no_props(qts, cpu_type);                   \
@@ -336,6 +347,25 @@ static void sve_tests_sve_off(const void *data)
     qtest_quit(qts);
 }
 
+static void sve_tests_sve_off_kvm(const void *data)
+{
+    QTestState *qts;
+
+    qts = qtest_init(MACHINE "-accel kvm -cpu max,sve=off");
+
+    /*
+     * We don't know if this host supports SVE so we don't
+     * attempt to test enabling anything. We only test that
+     * everything is disabled (as it should be with sve=off)
+     * and that using sve<N>=off to explicitly disable vector
+     * lengths is OK too.
+     */
+    assert_sve_vls(qts, "max", 0, NULL);
+    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+
+    qtest_quit(qts);
+}
+
 static void test_query_cpu_model_expansion(const void *data)
 {
     QTestState *qts;
@@ -385,12 +415,81 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
     assert_has_feature(qts, "host", "pmu");
 
     if (g_str_equal(qtest_get_arch(), "aarch64")) {
+        bool kvm_supports_sve;
+        char max_name[8], name[8];
+        uint32_t max_vq, vq;
+        uint64_t vls;
+        QDict *resp;
+        char *error;
+
         assert_has_feature(qts, "host", "aarch64");
-        assert_has_feature(qts, "max", "sve");
 
         assert_error(qts, "cortex-a15",
             "We cannot guarantee the CPU type 'cortex-a15' works "
             "with KVM on this host", NULL);
+
+        assert_has_feature(qts, "max", "sve");
+        resp = do_query_no_props(qts, "max");
+        kvm_supports_sve = resp_get_feature(resp, "sve");
+        vls = resp_get_sve_vls(resp);
+        qobject_unref(resp);
+
+        if (kvm_supports_sve) {
+            g_assert(vls != 0);
+            max_vq = 64 - __builtin_clzll(vls);
+            sprintf(max_name, "sve%d", max_vq * 128);
+
+            /* Enabling a supported length is of course fine. */
+            assert_sve_vls(qts, "max", vls, "{ %s: true }", max_name);
+
+            /* Get the next supported length smaller than max-vq. */
+            vq = 64 - __builtin_clzll(vls & ~BIT(max_vq - 1));
+            if (vq) {
+                /*
+                 * We have at least one length smaller than max-vq,
+                 * so we can disable max-vq.
+                 */
+                assert_sve_vls(qts, "max", (vls & ~BIT(max_vq - 1)),
+                               "{ %s: false }", max_name);
+
+                /*
+                 * Smaller, supported vector lengths cannot be disabled
+                 * unless all larger, supported vector lengths are also
+                 * disabled.
+                 */
+                sprintf(name, "sve%d", vq * 128);
+                error = g_strdup_printf("cannot disable %s", name);
+                assert_error(qts, "max", error,
+                             "{ %s: true, %s: false }",
+                             max_name, name);
+                g_free(error);
+            }
+
+            /*
+             * The smallest, supported vector length is required, because
+             * we need at least one vector length enabled.
+             */
+            vq = __builtin_ffsll(vls);
+            sprintf(name, "sve%d", vq * 128);
+            error = g_strdup_printf("cannot disable %s", name);
+            assert_error(qts, "max", error, "{ %s: false }", name);
+            g_free(error);
+
+            /* Get an unsupported length. */
+            for (vq = 1; vq <= max_vq; ++vq) {
+                if (!(vls & BIT(vq - 1))) {
+                    break;
+                }
+            }
+            if (vq <= SVE_MAX_VQ) {
+                sprintf(name, "sve%d", vq * 128);
+                error = g_strdup_printf("cannot enable %s", name);
+                assert_error(qts, "max", error, "{ %s: true }", name);
+                g_free(error);
+            }
+        } else {
+            g_assert(vls == 0);
+        }
     } else {
         assert_error(qts, "host",
                      "'pmu' feature not supported by KVM on this host",
@@ -427,6 +526,10 @@ int main(int argc, char **argv)
     if (kvm_available) {
         qtest_add_data_func("/arm/kvm/query-cpu-model-expansion",
                             NULL, test_query_cpu_model_expansion_kvm);
+        if (g_str_equal(qtest_get_arch(), "aarch64")) {
+            qtest_add_data_func("/arm/kvm/query-cpu-model-expansion/sve-off",
+                                NULL, sve_tests_sve_off_kvm);
+        }
     }
 
Quoted 6 lines

tests/arm-cpu-features.c int main(int argc, char **argv) · line 535

Quoted 2 lines
tests/arm-cpu-features.c +0 -0
     return g_test_run();

Besides,

Thanks!

drew

#27 lore raw

Re: [PATCH v4 8/9] target/arm/cpu64: max cpu: Support sve properties with KVM

Quoted 158 lines · Auger Eric
Quoted 155 lines · Andrew Jones
Quoted 96 lines · Auger Eric

Hi Drew,

Quoted 93 lines · Andrew Jones

Extend the SVE vq map initialization and validation with KVM's supported vector lengths when KVM is enabled. In order to determine and select supported lengths we add two new KVM functions for getting and setting the KVM_REG_ARM64_SVE_VLS pseudo-register.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst |  36 +++++---
 target/arm/cpu64.c        | 167 +++++++++++++++++++++++++++++---------
 target/arm/kvm64.c        | 100 ++++++++++++++++++++++-
 target/arm/kvm_arm.h      |  12 +++
 tests/arm-cpu-features.c  | 105 +++++++++++++++++++++++-
 5 files changed, 368 insertions(+), 52 deletions(-)
docs/arm-cpu-features.rst +26 -10
index 1262fddc6201..939366f959cf 100644
@@ -188,10 +188,17 @@ SVE CPU Property Dependencies and Constraints
 
   1) At least one vector length must be enabled when `sve` is enabled.
 
-  2) If a vector length `N` is enabled, then all power-of-two vector
-     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
-     is enabled, then `sve128` and `sve256` must also be enabled,
-     but `sve384` is not required.
+  2) If a vector length `N` is enabled, then, when KVM is enabled, all
+     smaller, host supported vector lengths must also be enabled.  If
+     KVM is not enabled, then only all the smaller, power-of-two vector
+     lengths must be enabled.  E.g. with KVM if the host supports all
+     vector lengths up to 512-bits (128, 256, 384, 512), then if
+     `sve512` is enabled, `sve128`, `sve256`, and `sve384` must also
+     be enabled. Without KVM, `sve384` would not be required.
+
+  3) If KVM is enabled then only vector lengths that the host CPU type
+     support may be enabled.  If SVE is not supported by the host, then
+     no `sve*` properties may be enabled.
 
 SVE CPU Property Parsing Semantics
 ----------------------------------
@@ -210,20 +217,29 @@ SVE CPU Property Parsing Semantics
      disable the last enabled vector length (see constraint (1) of "SVE
      CPU Property Dependencies and Constraints").
 
-  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+  4) When KVM is enabled, if the host does not support SVE, then an error
+     is generated when attempting to enable any `sve*` properties.
+
+  5) When KVM is enabled, if the host does support SVE, then an error is
+     generated when attempting to enable any vector lengths not supported
+     by the host.
+
+  6) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
      CPU properties are set `on`, then the specified vector lengths are
      disabled but the default for any unspecified lengths remains enabled.
-     Disabling a power-of-two vector length also disables all vector
-     lengths larger than the power-of-two length (see constraint (2) of
-     "SVE CPU Property Dependencies and Constraints").
+     When KVM is not enabled, disabling a power-of-two vector length also
+     disables all vector lengths larger than the power-of-two length.
+     When KVM is enabled, then disabling any supported vector length also
+     disables all larger vector lengths (see constraint (2) of "SVE CPU
+     Property Dependencies and Constraints").
 
-  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+  7) If one or more `sve<N>` CPU properties are set to `on`, then they
      are enabled and all unspecified lengths default to disabled, except
      for the required lengths per constraint (2) of "SVE CPU Property
      Dependencies and Constraints", which will even be auto-enabled if
      they were not explicitly enabled.
 
-  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+  8) If SVE was disabled (`sve=off`), allowing all vector lengths to be
      explicitly disabled (i.e. avoiding the error specified in (3) of
      "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
      provided an error will be generated.  To avoid this error, one must
target/arm/cpu64.c +1 -0
index b7eff4e1e107..18dd5e24ec61 100644
@@ -273,9 +273,18 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
      * any of the above.  Finally, if SVE is not disabled, then at least one
      * vector length must be enabled.
      */
+    DECLARE_BITMAP(kvm_supported, ARM_MAX_VQ);
     DECLARE_BITMAP(tmp, ARM_MAX_VQ);
     uint32_t vq, max_vq = 0;
 
Quoted 7 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 285

Quoted 6 lines
target/arm/cpu64.c +6 -0
+    /* Collect the set of vector lengths supported by KVM. */
+    bitmap_zero(kvm_supported, ARM_MAX_VQ);
+    if (kvm_enabled() && kvm_arm_sve_supported(CPU(cpu))) {
+        kvm_arm_sve_get_vls(CPU(cpu), kvm_supported);
+    } else if (kvm_enabled()) {
+        assert(!cpu_isar_feature(aa64_sve, cpu));

why not set an error and propagate it instead?

This should never happen. We shouldn't be here if KVM is enabled and SVE isn't supported. The question is how defensive do we want QEMU code? We could just drop the check altogether if we don't want the assert, but I'd rather keep it.

Quoted 45 lines
Quoted 45 lines
target/arm/cpu64.c +15 -16
+    }
+
     /*
      * Process explicit sve<N> properties.
      * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
@@ -293,10 +302,19 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Propagate enabled bits down through required powers-of-two. */
-        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
-            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
-                set_bit(vq - 1, cpu->sve_vq_map);
+        if (kvm_enabled()) {
+            /*
+             * For KVM we have to automatically enable all supported unitialized
+             * lengths, even when the smaller lengths are not all powers-of-two.
+             */
+            bitmap_andnot(tmp, kvm_supported, cpu->sve_vq_init, max_vq);
+            bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+        } else {
+            /* Propagate enabled bits down through required powers-of-two. */
+            for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+                if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                    set_bit(vq - 1, cpu->sve_vq_map);
+                }
             }
         }
     } else if (cpu->sve_max_vq == 0) {
@@ -308,23 +326,46 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Disabling a power-of-two disables all larger lengths. */
-        if (test_bit(0, cpu->sve_vq_init)) {
-            error_setg(errp, "cannot disable sve128");
-            error_append_hint(errp, "Disabling sve128 results in all vector "
-                              "lengths being disabled.\n");
-            error_append_hint(errp, "With SVE enabled, at least one vector "
-                              "length must be enabled.\n");
-            return;
-        }
-        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
-            if (test_bit(vq - 1, cpu->sve_vq_init)) {
-                break;
Quoted 8 lines
Quoted 7 lines
Quoted 7 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 335

Quoted 7 lines
target/arm/cpu64.c +7 -0
+        if (kvm_enabled()) {
+            /* Disabling a supported length disables all larger lengths. */
+            for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+                if (test_bit(vq - 1, cpu->sve_vq_init) &&
+                    test_bit(vq - 1, kvm_supported)) {
+                    break;
+                }

the above loop looks for the 1st disabled vq that is also supported, right?

Right

Quoted 7 lines
Quoted 5 lines
Quoted 5 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 340

Quoted 5 lines
target/arm/cpu64.c +5 -0
+            }
+            max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+            bitmap_andnot(cpu->sve_vq_map, kvm_supported,
+                          cpu->sve_vq_init, max_vq);
+            if (max_vq == 0 || bitmap_empty(cpu->sve_vq_map, max_vq)) {

here we don't have anything enabled below the disabled one. So don't we have the culprit already?

Oh, you're right. We can drop the find_next_bit call. Thanks for catching that.

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 342

Quoted 2 lines
target/arm/cpu64.c +2 -0
+                vq = find_next_bit(kvm_supported, ARM_MAX_VQ, 0) + 1;
+                error_setg(errp, "cannot disable sve%d", vq * 128);

isn't the one disabled max_vq? Do you really need to re-compute vq?

vq != max_vq here. max_vq is one smaller, even 0 if vq=1. So while vq is already correct, as you've pointed out, we need to use specifically that, not max_vq.

Thanks,
drew

#28 lore raw

Re: [PATCH v4 8/9] target/arm/cpu64: max cpu: Support sve properties with KVM

Hi Drew,

Quoted 181 lines · Andrew Jones
Quoted 158 lines · Auger Eric
Quoted 155 lines · Andrew Jones
Quoted 96 lines · Auger Eric

Hi Drew,

Quoted 93 lines · Andrew Jones

Extend the SVE vq map initialization and validation with KVM's supported vector lengths when KVM is enabled. In order to determine and select supported lengths we add two new KVM functions for getting and setting the KVM_REG_ARM64_SVE_VLS pseudo-register.

This patch has been co-authored with Richard Henderson, who reworked the target/arm/cpu64.c changes in order to push all the validation and auto-enabling/disabling steps into the finalizer, resulting in a nice LOC reduction.

---
 docs/arm-cpu-features.rst |  36 +++++---
 target/arm/cpu64.c        | 167 +++++++++++++++++++++++++++++---------
 target/arm/kvm64.c        | 100 ++++++++++++++++++++++-
 target/arm/kvm_arm.h      |  12 +++
 tests/arm-cpu-features.c  | 105 +++++++++++++++++++++++-
 5 files changed, 368 insertions(+), 52 deletions(-)
docs/arm-cpu-features.rst +26 -10
index 1262fddc6201..939366f959cf 100644
@@ -188,10 +188,17 @@ SVE CPU Property Dependencies and Constraints
 
   1) At least one vector length must be enabled when `sve` is enabled.
 
-  2) If a vector length `N` is enabled, then all power-of-two vector
-     lengths smaller than `N` must also be enabled.  E.g. if `sve512`
-     is enabled, then `sve128` and `sve256` must also be enabled,
-     but `sve384` is not required.
+  2) If a vector length `N` is enabled, then, when KVM is enabled, all
+     smaller, host supported vector lengths must also be enabled.  If
+     KVM is not enabled, then only all the smaller, power-of-two vector
+     lengths must be enabled.  E.g. with KVM if the host supports all
+     vector lengths up to 512-bits (128, 256, 384, 512), then if
+     `sve512` is enabled, `sve128`, `sve256`, and `sve384` must also
+     be enabled. Without KVM, `sve384` would not be required.
+
+  3) If KVM is enabled then only vector lengths that the host CPU type
+     support may be enabled.  If SVE is not supported by the host, then
+     no `sve*` properties may be enabled.
 
 SVE CPU Property Parsing Semantics
 ----------------------------------
@@ -210,20 +217,29 @@ SVE CPU Property Parsing Semantics
      disable the last enabled vector length (see constraint (1) of "SVE
      CPU Property Dependencies and Constraints").
 
-  4) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
+  4) When KVM is enabled, if the host does not support SVE, then an error
+     is generated when attempting to enable any `sve*` properties.
+
+  5) When KVM is enabled, if the host does support SVE, then an error is
+     generated when attempting to enable any vector lengths not supported
+     by the host.
+
+  6) If one or more `sve<N>` CPU properties are set `off`, but no `sve<N>`,
      CPU properties are set `on`, then the specified vector lengths are
      disabled but the default for any unspecified lengths remains enabled.
-     Disabling a power-of-two vector length also disables all vector
-     lengths larger than the power-of-two length (see constraint (2) of
-     "SVE CPU Property Dependencies and Constraints").
+     When KVM is not enabled, disabling a power-of-two vector length also
+     disables all vector lengths larger than the power-of-two length.
+     When KVM is enabled, then disabling any supported vector length also
+     disables all larger vector lengths (see constraint (2) of "SVE CPU
+     Property Dependencies and Constraints").
 
-  5) If one or more `sve<N>` CPU properties are set to `on`, then they
+  7) If one or more `sve<N>` CPU properties are set to `on`, then they
      are enabled and all unspecified lengths default to disabled, except
      for the required lengths per constraint (2) of "SVE CPU Property
      Dependencies and Constraints", which will even be auto-enabled if
      they were not explicitly enabled.
 
-  6) If SVE was disabled (`sve=off`), allowing all vector lengths to be
+  8) If SVE was disabled (`sve=off`), allowing all vector lengths to be
      explicitly disabled (i.e. avoiding the error specified in (3) of
      "SVE CPU Property Parsing Semantics"), then if later an `sve=on` is
      provided an error will be generated.  To avoid this error, one must
target/arm/cpu64.c +1 -0
index b7eff4e1e107..18dd5e24ec61 100644
@@ -273,9 +273,18 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
      * any of the above.  Finally, if SVE is not disabled, then at least one
      * vector length must be enabled.
      */
+    DECLARE_BITMAP(kvm_supported, ARM_MAX_VQ);
     DECLARE_BITMAP(tmp, ARM_MAX_VQ);
     uint32_t vq, max_vq = 0;
 
Quoted 7 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 285

Quoted 6 lines
target/arm/cpu64.c +6 -0
+    /* Collect the set of vector lengths supported by KVM. */
+    bitmap_zero(kvm_supported, ARM_MAX_VQ);
+    if (kvm_enabled() && kvm_arm_sve_supported(CPU(cpu))) {
+        kvm_arm_sve_get_vls(CPU(cpu), kvm_supported);
+    } else if (kvm_enabled()) {
+        assert(!cpu_isar_feature(aa64_sve, cpu));

why not set an error and propagate it instead?

This should never happen. We shouldn't be here if KVM is enabled and SVE isn't supported. The question is how defensive do we want QEMU code? We could just drop the check altogether if we don't want the assert, but I'd rather keep it.

Quoted 45 lines
Quoted 45 lines
target/arm/cpu64.c +15 -16
+    }
+
     /*
      * Process explicit sve<N> properties.
      * From the properties, sve_vq_map<N> implies sve_vq_init<N>.
@@ -293,10 +302,19 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Propagate enabled bits down through required powers-of-two. */
-        for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
-            if (!test_bit(vq - 1, cpu->sve_vq_init)) {
-                set_bit(vq - 1, cpu->sve_vq_map);
+        if (kvm_enabled()) {
+            /*
+             * For KVM we have to automatically enable all supported unitialized
+             * lengths, even when the smaller lengths are not all powers-of-two.
+             */
+            bitmap_andnot(tmp, kvm_supported, cpu->sve_vq_init, max_vq);
+            bitmap_or(cpu->sve_vq_map, cpu->sve_vq_map, tmp, max_vq);
+        } else {
+            /* Propagate enabled bits down through required powers-of-two. */
+            for (vq = pow2floor(max_vq); vq >= 1; vq >>= 1) {
+                if (!test_bit(vq - 1, cpu->sve_vq_init)) {
+                    set_bit(vq - 1, cpu->sve_vq_map);
+                }
             }
         }
     } else if (cpu->sve_max_vq == 0) {
@@ -308,23 +326,46 @@ void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp)
             return;
         }
 
-        /* Disabling a power-of-two disables all larger lengths. */
-        if (test_bit(0, cpu->sve_vq_init)) {
-            error_setg(errp, "cannot disable sve128");
-            error_append_hint(errp, "Disabling sve128 results in all vector "
-                              "lengths being disabled.\n");
-            error_append_hint(errp, "With SVE enabled, at least one vector "
-                              "length must be enabled.\n");
-            return;
-        }
-        for (vq = 2; vq <= ARM_MAX_VQ; vq <<= 1) {
-            if (test_bit(vq - 1, cpu->sve_vq_init)) {
-                break;
Quoted 8 lines
Quoted 7 lines
Quoted 7 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 335

Quoted 7 lines
target/arm/cpu64.c +7 -0
+        if (kvm_enabled()) {
+            /* Disabling a supported length disables all larger lengths. */
+            for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+                if (test_bit(vq - 1, cpu->sve_vq_init) &&
+                    test_bit(vq - 1, kvm_supported)) {
+                    break;
+                }

the above loop looks for the 1st disabled vq that is also supported, right?

Right

Quoted 7 lines
Quoted 5 lines
Quoted 5 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 340

Quoted 5 lines
target/arm/cpu64.c +5 -0
+            }
+            max_vq = vq <= ARM_MAX_VQ ? vq - 1 : ARM_MAX_VQ;
+            bitmap_andnot(cpu->sve_vq_map, kvm_supported,
+                          cpu->sve_vq_init, max_vq);
+            if (max_vq == 0 || bitmap_empty(cpu->sve_vq_map, max_vq)) {

here we don't have anything enabled below the disabled one. So don't we have the culprit already?

Oh, you're right. We can drop the find_next_bit call. Thanks for catching that.

Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

target/arm/cpu64.c void arm_cpu_sve_finalize(ARMCPU *cpu, Error **errp) · line 342

Quoted 2 lines
target/arm/cpu64.c +2 -0
+                vq = find_next_bit(kvm_supported, ARM_MAX_VQ, 0) + 1;
+                error_setg(errp, "cannot disable sve%d", vq * 128);

isn't the one disabled max_vq? Do you really need to re-compute vq?

vq != max_vq here. max_vq is one smaller, even 0 if vq=1. So while vq is already correct, as you've pointed out, we need to use specifically that, not max_vq.

OK

Thanks

Eric

Quoted 4 lines

Thanks,
drew

#29 lore raw

PATCH v4 9/9 target/arm/kvm: host cpu: Add support for sve<N> properties

In reply to Andrew Jones

Allow cpu 'host' to enable SVE when it's available, unless the user chooses to disable it with the added 'sve=off' cpu property. Also give the user the ability to select vector lengths with the sve<N> properties. We don't adopt 'max' cpu's other sve property, sve-max-vq, because that property is difficult to use with KVM. That property assumes all vector lengths in the range from 1 up to and including the specified maximum length are supported, but there may be optional lengths not supported by the host in that range. With KVM one must be more specific when enabling vector lengths.

---
 docs/arm-cpu-features.rst | 19 ++++++++++++-------
 target/arm/cpu.c          |  3 +++
 target/arm/cpu.h          |  2 ++
 target/arm/cpu64.c        | 33 +++++++++++++++++----------------
 target/arm/kvm64.c        | 14 +++++++++++++-
 tests/arm-cpu-features.c  | 21 +++++++++++----------
 6 files changed, 58 insertions(+), 34 deletions(-)
docs/arm-cpu-features.rst +12 -7
index 939366f959cf..0c16eef7c8c2 100644
@@ -256,31 +256,36 @@ SVE CPU Property Examples
 
      $ qemu-system-aarch64 -M virt -cpu max
 
-  3) Only enable the 128-bit vector length::
+  3) When KVM is enabled, implicitly enable all host CPU supported vector
+     lengths with the `host` CPU type::
+
+     $ qemu-system-aarch64 -M virt,accel=kvm -cpu host
+
+  4) Only enable the 128-bit vector length::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=on
 
-  4) Disable the 256-bit vector length and all larger vector lengths
+  5) Disable the 256-bit vector length and all larger vector lengths
      since 256 is a power-of-two (this results in only the 128-bit length
      being enabled)::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve256=off
 
-  5) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
+  6) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
 
-  6) The same as (5), but since the 128-bit and 256-bit vector
+  7) The same as (6), but since the 128-bit and 256-bit vector
      lengths are required for the 512-bit vector length to be enabled,
      then allow them to be auto-enabled::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve512=on
 
-  7) Do the same as (6), but by first disabling SVE and then re-enabling it::
+  8) Do the same as (7), but by first disabling SVE and then re-enabling it::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on
 
-  8) Force errors regarding the last vector length::
+  9) Force errors regarding the last vector length::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=off
      $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
@@ -292,5 +297,5 @@ The examples in "SVE CPU Property Examples" exhibit many ways to select
 vector lengths which developers may find useful in order to avoid overly
 verbose command lines.  However, the recommended way to select vector
 lengths is to explicitly enable each desired length.  Therefore only
-example's (1), (3), and (5) exhibit recommended uses of the properties.
+example's (1), (4), and (6) exhibit recommended uses of the properties.
 
target/arm/cpu.c +3 -0
index 522fed95b339..7695ae551218 100644
@@ -2671,6 +2671,9 @@ static void arm_host_initfn(Object *obj)
     ARMCPU *cpu = ARM_CPU(obj);
 
     kvm_arm_set_cpu_features_from_host(cpu);
+    if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
+        aarch64_add_sve_properties(obj);
+    }
     arm_cpu_post_init(obj);
 }
 
target/arm/cpu.h +2 -0
index 11162484465a..5b9c3e4cd73d 100644
@@ -974,11 +974,13 @@ int aarch64_cpu_gdb_write_register(CPUState *cpu, uint8_t *buf, int reg);
 void aarch64_sve_narrow_vq(CPUARMState *env, unsigned vq);
 void aarch64_sve_change_el(CPUARMState *env, int old_el,
                            int new_el, bool el0_a64);
+void aarch64_add_sve_properties(Object *obj);
 #else
 static inline void aarch64_sve_narrow_vq(CPUARMState *env, unsigned vq) { }
 static inline void aarch64_sve_change_el(CPUARMState *env, int o,
                                          int n, bool a)
 { }
+static inline void aarch64_add_sve_properties(Object *obj) { }
 #endif
 
 #if !defined(CONFIG_TCG)
target/arm/cpu64.c +17 -16
index 18dd5e24ec61..e7cd420faa9d 100644
@@ -579,6 +579,21 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
     cpu->isar.id_aa64pfr0 = t;
 }
 
+void aarch64_add_sve_properties(Object *obj)
+{
+    uint32_t vq;
+
+    object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
+                        cpu_arm_set_sve, NULL, NULL, &error_fatal);
+
+    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+        char name[8];
+        sprintf(name, "sve%d", vq * 128);
+        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+    }
+}
+
 /* -cpu max: if KVM is enabled, like -cpu host (best possible with this host);
  * otherwise, a CPU with as many features enabled as our emulation supports.
  * The version of '-cpu max' for qemu-system-arm is defined in cpu.c;
@@ -587,17 +602,11 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
 static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
-    uint32_t vq;
-    uint64_t t;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
-        if (kvm_arm_sve_supported(CPU(cpu))) {
-            t = cpu->isar.id_aa64pfr0;
-            t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
-            cpu->isar.id_aa64pfr0 = t;
-        }
     } else {
+        uint64_t t;
         uint32_t u;
         aarch64_a57_initfn(obj);
 
@@ -697,17 +706,9 @@ static void aarch64_max_initfn(Object *obj)
 #endif
     }
 
-    object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
-                        cpu_arm_set_sve, NULL, NULL, &error_fatal);
+    aarch64_add_sve_properties(obj);
     object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                         cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
-
-    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
-        char name[8];
-        sprintf(name, "sve%d", vq * 128);
-        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
-                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
-    }
 }
 
 struct ARMCPUInfo {
target/arm/kvm64.c +13 -1
index cff4217a8469..2da366ba113e 100644
@@ -488,7 +488,9 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
      * and then query that CPU for the relevant ID registers.
      */
     int fdarray[3];
+    bool sve_supported;
     uint64_t features = 0;
+    uint64_t t;
     int err;
 
     /* Old kernels may not know about the PREFERRED_TARGET ioctl: however
@@ -578,13 +580,23 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
                               ARM64_SYS_REG(3, 0, 0, 3, 2));
     }
 
+    sve_supported = ioctl(fdarray[0], KVM_CHECK_EXTENSION, KVM_CAP_ARM_SVE) > 0;
+
     kvm_arm_destroy_scratch_host_vcpu(fdarray);
 
     if (err < 0) {
         return false;
     }
 
-   /* We can assume any KVM supporting CPU is at least a v8
+    /* Add feature bits that can't appear until after VCPU init. */
+    if (sve_supported) {
+        t = ahcf->isar.id_aa64pfr0;
+        t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
+        ahcf->isar.id_aa64pfr0 = t;
+    }
+
+    /*
+     * We can assume any KVM supporting CPU is at least a v8
      * with VFPv4+Neon; this in turn implies most of the other
      * feature bits.
      */
tests/arm-cpu-features.c +11 -10
index a0129aebf409..4dc067fb5558 100644
@@ -351,7 +351,7 @@ static void sve_tests_sve_off_kvm(const void *data)
 {
     QTestState *qts;
 
-    qts = qtest_init(MACHINE "-accel kvm -cpu max,sve=off");
+    qts = qtest_init(MACHINE "-accel kvm -cpu host,sve=off");
 
     /*
      * We don't know if this host supports SVE so we don't
@@ -360,8 +360,8 @@ static void sve_tests_sve_off_kvm(const void *data)
      * and that using sve<N>=off to explicitly disable vector
      * lengths is OK too.
      */
-    assert_sve_vls(qts, "max", 0, NULL);
-    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+    assert_sve_vls(qts, "host", 0, NULL);
+    assert_sve_vls(qts, "host", 0, "{ 'sve128': false }");
 
     qtest_quit(qts);
 }
@@ -428,8 +428,8 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             "We cannot guarantee the CPU type 'cortex-a15' works "
             "with KVM on this host", NULL);
 
-        assert_has_feature(qts, "max", "sve");
-        resp = do_query_no_props(qts, "max");
+        assert_has_feature(qts, "host", "sve");
+        resp = do_query_no_props(qts, "host");
         kvm_supports_sve = resp_get_feature(resp, "sve");
         vls = resp_get_sve_vls(resp);
         qobject_unref(resp);
@@ -440,7 +440,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             sprintf(max_name, "sve%d", max_vq * 128);
 
             /* Enabling a supported length is of course fine. */
-            assert_sve_vls(qts, "max", vls, "{ %s: true }", max_name);
+            assert_sve_vls(qts, "host", vls, "{ %s: true }", max_name);
 
             /* Get the next supported length smaller than max-vq. */
             vq = 64 - __builtin_clzll(vls & ~BIT(max_vq - 1));
@@ -449,7 +449,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
                  * We have at least one length smaller than max-vq,
                  * so we can disable max-vq.
                  */
-                assert_sve_vls(qts, "max", (vls & ~BIT(max_vq - 1)),
+                assert_sve_vls(qts, "host", (vls & ~BIT(max_vq - 1)),
                                "{ %s: false }", max_name);
 
                 /*
@@ -459,7 +459,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
                  */
                 sprintf(name, "sve%d", vq * 128);
                 error = g_strdup_printf("cannot disable %s", name);
-                assert_error(qts, "max", error,
+                assert_error(qts, "host", error,
                              "{ %s: true, %s: false }",
                              max_name, name);
                 g_free(error);
@@ -472,7 +472,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             vq = __builtin_ffsll(vls);
             sprintf(name, "sve%d", vq * 128);
             error = g_strdup_printf("cannot disable %s", name);
-            assert_error(qts, "max", error, "{ %s: false }", name);
+            assert_error(qts, "host", error, "{ %s: false }", name);
             g_free(error);
 
             /* Get an unsupported length. */
@@ -484,13 +484,14 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             if (vq <= SVE_MAX_VQ) {
                 sprintf(name, "sve%d", vq * 128);
                 error = g_strdup_printf("cannot enable %s", name);
-                assert_error(qts, "max", error, "{ %s: true }", name);
+                assert_error(qts, "host", error, "{ %s: true }", name);
                 g_free(error);
             }
         } else {
             g_assert(vls == 0);
         }
     } else {
+        assert_has_not_feature(qts, "host", "sve");
         assert_error(qts, "host",
                      "'pmu' feature not supported by KVM on this host",
                      "{ 'pmu': true }");
Signature · 1 line
2.20.1
Reviewed-by on v4 9/9
#30 lore raw

Re: [PATCH v4 9/9] target/arm/kvm: host cpu: Add support for sve<N> properties

Hi Drew,

Quoted 114 lines · Andrew Jones

Allow cpu 'host' to enable SVE when it's available, unless the user chooses to disable it with the added 'sve=off' cpu property. Also give the user the ability to select vector lengths with the sve<N> properties. We don't adopt 'max' cpu's other sve property, sve-max-vq, because that property is difficult to use with KVM. That property assumes all vector lengths in the range from 1 up to and including the specified maximum length are supported, but there may be optional lengths not supported by the host in that range. With KVM one must be more specific when enabling vector lengths.

---
 docs/arm-cpu-features.rst | 19 ++++++++++++-------
 target/arm/cpu.c          |  3 +++
 target/arm/cpu.h          |  2 ++
 target/arm/cpu64.c        | 33 +++++++++++++++++----------------
 target/arm/kvm64.c        | 14 +++++++++++++-
 tests/arm-cpu-features.c  | 21 +++++++++++----------
 6 files changed, 58 insertions(+), 34 deletions(-)
docs/arm-cpu-features.rst +12 -7
index 939366f959cf..0c16eef7c8c2 100644
@@ -256,31 +256,36 @@ SVE CPU Property Examples
 
      $ qemu-system-aarch64 -M virt -cpu max
 
-  3) Only enable the 128-bit vector length::
+  3) When KVM is enabled, implicitly enable all host CPU supported vector
+     lengths with the `host` CPU type::
+
+     $ qemu-system-aarch64 -M virt,accel=kvm -cpu host
+
+  4) Only enable the 128-bit vector length::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=on
 
-  4) Disable the 256-bit vector length and all larger vector lengths
+  5) Disable the 256-bit vector length and all larger vector lengths
      since 256 is a power-of-two (this results in only the 128-bit length
      being enabled)::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve256=off
 
-  5) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
+  6) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
 
-  6) The same as (5), but since the 128-bit and 256-bit vector
+  7) The same as (6), but since the 128-bit and 256-bit vector
      lengths are required for the 512-bit vector length to be enabled,
      then allow them to be auto-enabled::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve512=on
 
-  7) Do the same as (6), but by first disabling SVE and then re-enabling it::
+  8) Do the same as (7), but by first disabling SVE and then re-enabling it::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on
 
-  8) Force errors regarding the last vector length::
+  9) Force errors regarding the last vector length::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=off
      $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
@@ -292,5 +297,5 @@ The examples in "SVE CPU Property Examples" exhibit many ways to select
 vector lengths which developers may find useful in order to avoid overly
 verbose command lines.  However, the recommended way to select vector
 lengths is to explicitly enable each desired length.  Therefore only
-example's (1), (3), and (5) exhibit recommended uses of the properties.
+example's (1), (4), and (6) exhibit recommended uses of the properties.
 
target/arm/cpu.c +3 -0
index 522fed95b339..7695ae551218 100644
@@ -2671,6 +2671,9 @@ static void arm_host_initfn(Object *obj)
     ARMCPU *cpu = ARM_CPU(obj);
 
     kvm_arm_set_cpu_features_from_host(cpu);
+    if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
+        aarch64_add_sve_properties(obj);
+    }
     arm_cpu_post_init(obj);
 }
 
target/arm/cpu.h +2 -0
index 11162484465a..5b9c3e4cd73d 100644
@@ -974,11 +974,13 @@ int aarch64_cpu_gdb_write_register(CPUState *cpu, uint8_t *buf, int reg);
 void aarch64_sve_narrow_vq(CPUARMState *env, unsigned vq);
 void aarch64_sve_change_el(CPUARMState *env, int old_el,
                            int new_el, bool el0_a64);
+void aarch64_add_sve_properties(Object *obj);
 #else
 static inline void aarch64_sve_narrow_vq(CPUARMState *env, unsigned vq) { }
 static inline void aarch64_sve_change_el(CPUARMState *env, int o,
                                          int n, bool a)
 { }
+static inline void aarch64_add_sve_properties(Object *obj) { }
 #endif
 
 #if !defined(CONFIG_TCG)
target/arm/cpu64.c +0 -0
index 18dd5e24ec61..e7cd420faa9d 100644
@@ -579,6 +579,21 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
     cpu->isar.id_aa64pfr0 = t;
 }

target/arm/cpu64.c static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name, · line 590

Quoted 10 lines
target/arm/cpu64.c +9 -0
 
+void aarch64_add_sve_properties(Object *obj)
+{
+    uint32_t vq;
+
+    object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
+                        cpu_arm_set_sve, NULL, NULL, &error_fatal);
+
+    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+        char name[8];

line needed

Quoted 52 lines
target/arm/cpu64.c +8 -16
+        sprintf(name, "sve%d", vq * 128);
+        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+    }
+}
+
 /* -cpu max: if KVM is enabled, like -cpu host (best possible with this host);
  * otherwise, a CPU with as many features enabled as our emulation supports.
  * The version of '-cpu max' for qemu-system-arm is defined in cpu.c;
@@ -587,17 +602,11 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
 static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
-    uint32_t vq;
-    uint64_t t;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
-        if (kvm_arm_sve_supported(CPU(cpu))) {
-            t = cpu->isar.id_aa64pfr0;
-            t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
-            cpu->isar.id_aa64pfr0 = t;
-        }
     } else {
+        uint64_t t;
         uint32_t u;
         aarch64_a57_initfn(obj);
 
@@ -697,17 +706,9 @@ static void aarch64_max_initfn(Object *obj)
 #endif
     }
 
-    object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
-                        cpu_arm_set_sve, NULL, NULL, &error_fatal);
+    aarch64_add_sve_properties(obj);
     object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                         cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
-
-    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
-        char name[8];
-        sprintf(name, "sve%d", vq * 128);
-        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
-                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
-    }
 }
 
 struct ARMCPUInfo {
target/arm/kvm64.c +0 -0
index cff4217a8469..2da366ba113e 100644

target/arm/kvm64.c bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf) · line 493

Quoted 6 lines
target/arm/kvm64.c +2 -0
@@ -488,7 +488,9 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
      * and then query that CPU for the relevant ID registers.
      */
     int fdarray[3];
+    bool sve_supported;
     uint64_t features = 0;
+    uint64_t t;

t can stay local to the block below

target/arm/kvm64.c bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf) · line 583

Quoted 8 lines
target/arm/kvm64.c +1 -0
     int err;
 
     /* Old kernels may not know about the PREFERRED_TARGET ioctl: however
@@ -578,13 +580,23 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
                               ARM64_SYS_REG(3, 0, 0, 3, 2));
     }
 
+    sve_supported = ioctl(fdarray[0], KVM_CHECK_EXTENSION, KVM_CAP_ARM_SVE) > 0;

!ioctl?

Quoted 98 lines
target/arm/kvm64.c +10 -1
+
     kvm_arm_destroy_scratch_host_vcpu(fdarray);
 
     if (err < 0) {
         return false;
     }
 
-   /* We can assume any KVM supporting CPU is at least a v8
+    /* Add feature bits that can't appear until after VCPU init. */
+    if (sve_supported) {
+        t = ahcf->isar.id_aa64pfr0;
+        t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
+        ahcf->isar.id_aa64pfr0 = t;
+    }
+
+    /*
+     * We can assume any KVM supporting CPU is at least a v8
      * with VFPv4+Neon; this in turn implies most of the other
      * feature bits.
      */
tests/arm-cpu-features.c +10 -10
index a0129aebf409..4dc067fb5558 100644
@@ -351,7 +351,7 @@ static void sve_tests_sve_off_kvm(const void *data)
 {
     QTestState *qts;
 
-    qts = qtest_init(MACHINE "-accel kvm -cpu max,sve=off");
+    qts = qtest_init(MACHINE "-accel kvm -cpu host,sve=off");
 
     /*
      * We don't know if this host supports SVE so we don't
@@ -360,8 +360,8 @@ static void sve_tests_sve_off_kvm(const void *data)
      * and that using sve<N>=off to explicitly disable vector
      * lengths is OK too.
      */
-    assert_sve_vls(qts, "max", 0, NULL);
-    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+    assert_sve_vls(qts, "host", 0, NULL);
+    assert_sve_vls(qts, "host", 0, "{ 'sve128': false }");
 
     qtest_quit(qts);
 }
@@ -428,8 +428,8 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             "We cannot guarantee the CPU type 'cortex-a15' works "
             "with KVM on this host", NULL);
 
-        assert_has_feature(qts, "max", "sve");
-        resp = do_query_no_props(qts, "max");
+        assert_has_feature(qts, "host", "sve");
+        resp = do_query_no_props(qts, "host");
         kvm_supports_sve = resp_get_feature(resp, "sve");
         vls = resp_get_sve_vls(resp);
         qobject_unref(resp);
@@ -440,7 +440,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             sprintf(max_name, "sve%d", max_vq * 128);
 
             /* Enabling a supported length is of course fine. */
-            assert_sve_vls(qts, "max", vls, "{ %s: true }", max_name);
+            assert_sve_vls(qts, "host", vls, "{ %s: true }", max_name);
 
             /* Get the next supported length smaller than max-vq. */
             vq = 64 - __builtin_clzll(vls & ~BIT(max_vq - 1));
@@ -449,7 +449,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
                  * We have at least one length smaller than max-vq,
                  * so we can disable max-vq.
                  */
-                assert_sve_vls(qts, "max", (vls & ~BIT(max_vq - 1)),
+                assert_sve_vls(qts, "host", (vls & ~BIT(max_vq - 1)),
                                "{ %s: false }", max_name);
 
                 /*
@@ -459,7 +459,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
                  */
                 sprintf(name, "sve%d", vq * 128);
                 error = g_strdup_printf("cannot disable %s", name);
-                assert_error(qts, "max", error,
+                assert_error(qts, "host", error,
                              "{ %s: true, %s: false }",
                              max_name, name);
                 g_free(error);
@@ -472,7 +472,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             vq = __builtin_ffsll(vls);
             sprintf(name, "sve%d", vq * 128);
             error = g_strdup_printf("cannot disable %s", name);
-            assert_error(qts, "max", error, "{ %s: false }", name);
+            assert_error(qts, "host", error, "{ %s: false }", name);
             g_free(error);
 
             /* Get an unsupported length. */
@@ -484,13 +484,14 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             if (vq <= SVE_MAX_VQ) {
                 sprintf(name, "sve%d", vq * 128);
                 error = g_strdup_printf("cannot enable %s", name);
-                assert_error(qts, "max", error, "{ %s: true }", name);
+                assert_error(qts, "host", error, "{ %s: true }", name);

tests/arm-cpu-features.c static void test_query_cpu_model_expansion_kvm(const void *data) · line 497

Quoted 10 lines
tests/arm-cpu-features.c +1 -0
                 g_free(error);
             }
         } else {
             g_assert(vls == 0);
         }
     } else {
+        assert_has_not_feature(qts, "host", "sve");
         assert_error(qts, "host",
                      "'pmu' feature not supported by KVM on this host",
                      "{ 'pmu': true }");

Besides

Thanks

Eric

#31 lore raw

Re: [PATCH v4 9/9] target/arm/kvm: host cpu: Add support for sve<N> properties

Quoted 118 lines · Auger Eric

Hi Drew,

Quoted 115 lines · Andrew Jones

Allow cpu 'host' to enable SVE when it's available, unless the user chooses to disable it with the added 'sve=off' cpu property. Also give the user the ability to select vector lengths with the sve<N> properties. We don't adopt 'max' cpu's other sve property, sve-max-vq, because that property is difficult to use with KVM. That property assumes all vector lengths in the range from 1 up to and including the specified maximum length are supported, but there may be optional lengths not supported by the host in that range. With KVM one must be more specific when enabling vector lengths.

---
 docs/arm-cpu-features.rst | 19 ++++++++++++-------
 target/arm/cpu.c          |  3 +++
 target/arm/cpu.h          |  2 ++
 target/arm/cpu64.c        | 33 +++++++++++++++++----------------
 target/arm/kvm64.c        | 14 +++++++++++++-
 tests/arm-cpu-features.c  | 21 +++++++++++----------
 6 files changed, 58 insertions(+), 34 deletions(-)
docs/arm-cpu-features.rst +12 -7
index 939366f959cf..0c16eef7c8c2 100644
@@ -256,31 +256,36 @@ SVE CPU Property Examples
 
      $ qemu-system-aarch64 -M virt -cpu max
 
-  3) Only enable the 128-bit vector length::
+  3) When KVM is enabled, implicitly enable all host CPU supported vector
+     lengths with the `host` CPU type::
+
+     $ qemu-system-aarch64 -M virt,accel=kvm -cpu host
+
+  4) Only enable the 128-bit vector length::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=on
 
-  4) Disable the 256-bit vector length and all larger vector lengths
+  5) Disable the 256-bit vector length and all larger vector lengths
      since 256 is a power-of-two (this results in only the 128-bit length
      being enabled)::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve256=off
 
-  5) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
+  6) Enable the 128-bit, 256-bit, and 512-bit vector lengths::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=on,sve256=on,sve512=on
 
-  6) The same as (5), but since the 128-bit and 256-bit vector
+  7) The same as (6), but since the 128-bit and 256-bit vector
      lengths are required for the 512-bit vector length to be enabled,
      then allow them to be auto-enabled::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve512=on
 
-  7) Do the same as (6), but by first disabling SVE and then re-enabling it::
+  8) Do the same as (7), but by first disabling SVE and then re-enabling it::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve512=on,sve=on
 
-  8) Force errors regarding the last vector length::
+  9) Force errors regarding the last vector length::
 
      $ qemu-system-aarch64 -M virt -cpu max,sve128=off
      $ qemu-system-aarch64 -M virt -cpu max,sve=off,sve128=off,sve=on
@@ -292,5 +297,5 @@ The examples in "SVE CPU Property Examples" exhibit many ways to select
 vector lengths which developers may find useful in order to avoid overly
 verbose command lines.  However, the recommended way to select vector
 lengths is to explicitly enable each desired length.  Therefore only
-example's (1), (3), and (5) exhibit recommended uses of the properties.
+example's (1), (4), and (6) exhibit recommended uses of the properties.
 
target/arm/cpu.c +3 -0
index 522fed95b339..7695ae551218 100644
@@ -2671,6 +2671,9 @@ static void arm_host_initfn(Object *obj)
     ARMCPU *cpu = ARM_CPU(obj);
 
     kvm_arm_set_cpu_features_from_host(cpu);
+    if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
+        aarch64_add_sve_properties(obj);
+    }
     arm_cpu_post_init(obj);
 }
 
target/arm/cpu.h +2 -0
index 11162484465a..5b9c3e4cd73d 100644
@@ -974,11 +974,13 @@ int aarch64_cpu_gdb_write_register(CPUState *cpu, uint8_t *buf, int reg);
 void aarch64_sve_narrow_vq(CPUARMState *env, unsigned vq);
 void aarch64_sve_change_el(CPUARMState *env, int old_el,
                            int new_el, bool el0_a64);
+void aarch64_add_sve_properties(Object *obj);
 #else
 static inline void aarch64_sve_narrow_vq(CPUARMState *env, unsigned vq) { }
 static inline void aarch64_sve_change_el(CPUARMState *env, int o,
                                          int n, bool a)
 { }
+static inline void aarch64_add_sve_properties(Object *obj) { }
 #endif
 
 #if !defined(CONFIG_TCG)
target/arm/cpu64.c +0 -0
index 18dd5e24ec61..e7cd420faa9d 100644
@@ -579,6 +579,21 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
     cpu->isar.id_aa64pfr0 = t;
 }
 
Quoted 10 lines

target/arm/cpu64.c static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name, · line 590

Quoted 9 lines
target/arm/cpu64.c +9 -0
+void aarch64_add_sve_properties(Object *obj)
+{
+    uint32_t vq;
+
+    object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
+                        cpu_arm_set_sve, NULL, NULL, &error_fatal);
+
+    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
+        char name[8];

line needed

checkpatch doesn't complain and I prefer this simple block to be more dense.

Quoted 46 lines
Quoted 46 lines
target/arm/cpu64.c +8 -16
+        sprintf(name, "sve%d", vq * 128);
+        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
+                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
+    }
+}
+
 /* -cpu max: if KVM is enabled, like -cpu host (best possible with this host);
  * otherwise, a CPU with as many features enabled as our emulation supports.
  * The version of '-cpu max' for qemu-system-arm is defined in cpu.c;
@@ -587,17 +602,11 @@ static void cpu_arm_set_sve(Object *obj, Visitor *v, const char *name,
 static void aarch64_max_initfn(Object *obj)
 {
     ARMCPU *cpu = ARM_CPU(obj);
-    uint32_t vq;
-    uint64_t t;
 
     if (kvm_enabled()) {
         kvm_arm_set_cpu_features_from_host(cpu);
-        if (kvm_arm_sve_supported(CPU(cpu))) {
-            t = cpu->isar.id_aa64pfr0;
-            t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
-            cpu->isar.id_aa64pfr0 = t;
-        }
     } else {
+        uint64_t t;
         uint32_t u;
         aarch64_a57_initfn(obj);
 
@@ -697,17 +706,9 @@ static void aarch64_max_initfn(Object *obj)
 #endif
     }
 
-    object_property_add(obj, "sve", "bool", cpu_arm_get_sve,
-                        cpu_arm_set_sve, NULL, NULL, &error_fatal);
+    aarch64_add_sve_properties(obj);
     object_property_add(obj, "sve-max-vq", "uint32", cpu_max_get_sve_max_vq,
                         cpu_max_set_sve_max_vq, NULL, NULL, &error_fatal);
-
-    for (vq = 1; vq <= ARM_MAX_VQ; ++vq) {
-        char name[8];
-        sprintf(name, "sve%d", vq * 128);
-        object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
-                            cpu_arm_set_sve_vq, NULL, NULL, &error_fatal);
-    }
 }
 
Quoted 13 lines

target/arm/kvm64.c bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf) · line 493

Quoted 12 lines
target/arm/cpu64.c +0 -0
 struct ARMCPUInfo {
target/arm/kvm64.c +2 -0
index cff4217a8469..2da366ba113e 100644
@@ -488,7 +488,9 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
      * and then query that CPU for the relevant ID registers.
      */
     int fdarray[3];
+    bool sve_supported;
     uint64_t features = 0;
+    uint64_t t;

t can stay local to the block below

I made its scope function-wide because it's unlikely this will be the only feature we enable this way. We could start it with a smaller scope and move it when necessary, but as it doesn't hurt to make it function- wide now, I did it now.

Quoted 9 lines

target/arm/kvm64.c bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf) · line 583

Quoted 8 lines
target/arm/kvm64.c +1 -0
     int err;
 
     /* Old kernels may not know about the PREFERRED_TARGET ioctl: however
@@ -578,13 +580,23 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
                               ARM64_SYS_REG(3, 0, 0, 3, 2));
     }
 
+    sve_supported = ioctl(fdarray[0], KVM_CHECK_EXTENSION, KVM_CAP_ARM_SVE) > 0;

!ioctl?

!ioctl should be safe with KVM_CHECK_EXTENSION (according to the API documentation), but nowhere in QEMU do we assume that. We always check for the possibility of a negative (error) return value.

Quoted 109 lines
Quoted 98 lines
target/arm/kvm64.c +10 -1
+
     kvm_arm_destroy_scratch_host_vcpu(fdarray);
 
     if (err < 0) {
         return false;
     }
 
-   /* We can assume any KVM supporting CPU is at least a v8
+    /* Add feature bits that can't appear until after VCPU init. */
+    if (sve_supported) {
+        t = ahcf->isar.id_aa64pfr0;
+        t = FIELD_DP64(t, ID_AA64PFR0, SVE, 1);
+        ahcf->isar.id_aa64pfr0 = t;
+    }
+
+    /*
+     * We can assume any KVM supporting CPU is at least a v8
      * with VFPv4+Neon; this in turn implies most of the other
      * feature bits.
      */
tests/arm-cpu-features.c +10 -10
index a0129aebf409..4dc067fb5558 100644
@@ -351,7 +351,7 @@ static void sve_tests_sve_off_kvm(const void *data)
 {
     QTestState *qts;
 
-    qts = qtest_init(MACHINE "-accel kvm -cpu max,sve=off");
+    qts = qtest_init(MACHINE "-accel kvm -cpu host,sve=off");
 
     /*
      * We don't know if this host supports SVE so we don't
@@ -360,8 +360,8 @@ static void sve_tests_sve_off_kvm(const void *data)
      * and that using sve<N>=off to explicitly disable vector
      * lengths is OK too.
      */
-    assert_sve_vls(qts, "max", 0, NULL);
-    assert_sve_vls(qts, "max", 0, "{ 'sve128': false }");
+    assert_sve_vls(qts, "host", 0, NULL);
+    assert_sve_vls(qts, "host", 0, "{ 'sve128': false }");
 
     qtest_quit(qts);
 }
@@ -428,8 +428,8 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             "We cannot guarantee the CPU type 'cortex-a15' works "
             "with KVM on this host", NULL);
 
-        assert_has_feature(qts, "max", "sve");
-        resp = do_query_no_props(qts, "max");
+        assert_has_feature(qts, "host", "sve");
+        resp = do_query_no_props(qts, "host");
         kvm_supports_sve = resp_get_feature(resp, "sve");
         vls = resp_get_sve_vls(resp);
         qobject_unref(resp);
@@ -440,7 +440,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             sprintf(max_name, "sve%d", max_vq * 128);
 
             /* Enabling a supported length is of course fine. */
-            assert_sve_vls(qts, "max", vls, "{ %s: true }", max_name);
+            assert_sve_vls(qts, "host", vls, "{ %s: true }", max_name);
 
             /* Get the next supported length smaller than max-vq. */
             vq = 64 - __builtin_clzll(vls & ~BIT(max_vq - 1));
@@ -449,7 +449,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
                  * We have at least one length smaller than max-vq,
                  * so we can disable max-vq.
                  */
-                assert_sve_vls(qts, "max", (vls & ~BIT(max_vq - 1)),
+                assert_sve_vls(qts, "host", (vls & ~BIT(max_vq - 1)),
                                "{ %s: false }", max_name);
 
                 /*
@@ -459,7 +459,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
                  */
                 sprintf(name, "sve%d", vq * 128);
                 error = g_strdup_printf("cannot disable %s", name);
-                assert_error(qts, "max", error,
+                assert_error(qts, "host", error,
                              "{ %s: true, %s: false }",
                              max_name, name);
                 g_free(error);
@@ -472,7 +472,7 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             vq = __builtin_ffsll(vls);
             sprintf(name, "sve%d", vq * 128);
             error = g_strdup_printf("cannot disable %s", name);
-            assert_error(qts, "max", error, "{ %s: false }", name);
+            assert_error(qts, "host", error, "{ %s: false }", name);
             g_free(error);
 
             /* Get an unsupported length. */
@@ -484,13 +484,14 @@ static void test_query_cpu_model_expansion_kvm(const void *data)
             if (vq <= SVE_MAX_VQ) {
                 sprintf(name, "sve%d", vq * 128);
                 error = g_strdup_printf("cannot enable %s", name);
-                assert_error(qts, "max", error, "{ %s: true }", name);
+                assert_error(qts, "host", error, "{ %s: true }", name);

tests/arm-cpu-features.c static void test_query_cpu_model_expansion_kvm(const void *data) · line 497

Quoted 10 lines
tests/arm-cpu-features.c +1 -0
                 g_free(error);
             }
         } else {
             g_assert(vls == 0);
         }
     } else {
+        assert_has_not_feature(qts, "host", "sve");
         assert_error(qts, "host",
                      "'pmu' feature not supported by KVM on this host",
                      "{ 'pmu': true }");

Besides

Quoted 2 lines

Thanks!

drew