Software Design Guidelines

Error Detection and Error Handling

Workflow

Error detection and error handling workflow in the ACRN hypervisor is shown in Figure 247.

../_images/work_flow_of_error_detection_and_error_handling.png

Figure 247 Error Detection and Error Handling Workflow

Design Assumption

There are three types of design assumptions in the ACRN hypervisor, as shown below:

Pre-condition

Pre-conditions shall be defined right before the definition/declaration of the corresponding function in the C source file or header file. All pre-conditions shall be guaranteed by the caller of the function. Error checking of the pre-conditions are not needed in release version of the function. Developers could use ASSERT to catch design errors in a debug version for some cases. Verification of the hypervisor shall check whether each caller guarantees all pre-conditions of the callee (or not).

This design assumption applies to the following cases:

  • Input parameters of the function.
  • Global state, such as hypervisor operation mode.
Post-condition

Post-conditions shall be defined right before the definition/declaration of the corresponding function in the C source file or header file. All post-conditions shall be guaranteed by the function. All callers of the function should trust these post-conditions are met. Error checking of the post-conditions are not needed in release version of each caller. Developers could use ASSERT to catch design errors in a debug version for some cases. Verification of the hypervisor shall check whether the function guarantees all post-conditions (or not).

This design assumption applies to the following case:

  • Return value of the function

    It is used to guarantee that the return value is valid, such as the return pointer is not NULL, the return value is within a valid range, or the members of the return structure are valid.

Application Constraints

Application constraints of the hypervisor shall be defined in design document and safety manual. All application constraints shall be guaranteed by external safety applications, such as Board Support Package, firmware, safety VM, or Hardware. The verification of application integration shall check whether the safety application meets all application constraints. These constraints must be verified during hypervisor validation test. It is optional to do error checking for application constraints at hypervisor boot time.

This design assumption applies to the following cases:

  • Configuration data defined by external safety application, such as physical PCI device information specific for each board design.
  • Input data which is only specified by external safety application.

Note

If input data can be specified by both a non-safety VM and a safety VM, the application constraint isn’t applicable to these data. Related error checking and handling shall be done during hypervisor design.

Refer to the C Programming Language Coding Guidelines to document these design assumptions with doxygen-style comments.

Architecture Level

Functional Safety Consideration

The hypervisor will do range check in hypercalls and HW capability checks according to Table A.2 of FuSa Standards [IEC_61508-3_2010] .

Error Handling Methods

The error handling methods used in the ACRN hypervisor on an architecture level are shown below.

Invoke default fatal error handler

The hypervisor shall invoke the default fatal error handler when the below cases occur. Customers can define platform-specific handlers, allowing them to implement additional error reporting (mostly to hardware) if required. The default fatal error handler will invoke platform-specific handlers defined by users at first, then it will panic the system.

This method applies to the following cases:

  • Related hardware resources are unavailable.
  • Boot information is invalid during platform initialization.
  • Unexpected exception occurs in root mode due to hardware failures.
  • Failures occur in the VM dedicated for error handling.
Return error code

The hypervisor shall return an error code to the VM when the below cases occur. The error code shall indicate the error type detected (e.g. invalid parameter, device not found, device busy, resource unavailable, etc).

This method applies to the following case:

  • The hypercall parameter from the VM is invalid.
Inform the safety VM through specific register or memory area

The hypervisor shall inform the safety VM through a specific register or memory area when the below cases occur. The VM will decide how to handle the related error. This shall only be done after the VM (Safety OS or Service OS) dedicated to error handling has started.

This method applies to the following cases:

  • Machine check errors occur due to hardware failures.
  • Unexpected VM entry failures occur, where the VM is not the one dedicated for error handling.
Panic the system via ASSERT

The hypervisor can panic the system when the below cases occur. It shall only be used for debug and used to check pre-conditions and post-conditions to catch design errors.

This method applies to the following case:

  • Software design errors occur.

Rules of Error Detection and Error Handling

The rules of error detection and error handling on an architecture level are shown in Table 11 below.

Table 11 Rules of Error Detection and Error Handling on Architecture Level
Resource Class Failure Mode Error Detection via Hypervisor Error Handling Policy Example
External resource provided by VM Invalid register/memory state on VM exit Yes Follow SDM strictly, or state any deviation to the document explicitly Unsupported MSR or invalid CPU ID
Invalid hypercall parameter Yes The hypervisor shall return related error code to the VM Invalid hypercall parameter provided by any VM
Invalid data in the sharing memory area Yes Case by case depending on the data Invalid data in memory shared with all VMs, such as IO request buffers and sbuf for debug
External resource provided by bootloader (UEFI or SBL) Invalid E820 table or invalid boot information Yes The hypervisor shall panic during platform initialization Invalid E820 table or invalid boot information
Physical resource used by the hypervisor 1GB page is not available on the platform or invalid physical CPU ID Yes The hypervisor shall panic during platform initialization 1GB page is not available on the platform or invalid physical CPU ID

Examples

Here is an example to illustrate when error handling codes are required on an architecture level.

There are two pre-condition statements of vcpu_from_vid. It indicates that it’s the caller’s responsibility to guarantee these pre-conditions.

/**
 * @pre vcpu_id < CONFIG_MAX_VCPUS_PER_VM
 * @pre &(vm->hw.vcpu_array[vcpu_id])->state != VCPU_OFFLINE
 */
static inline struct acrn_vcpu *vcpu_from_vid(struct acrn_vm *vm, uint16_t vcpu_id)
{
        return &(vm->hw.vcpu_array[vcpu_id]);
}

vcpu_from_vid is called by hcall_set_vcpu_regs, which is a hypercall. hcall_set_vcpu_regs is an external interface and vcpu_id is provided by VM. In this case, we shall add the error checking codes before calling vcpu_from_vid to make sure that the passed parameters are valid and the pre-conditions are guaranteed.

Here is the sample codes for error checking before calling vcpu_from_vid:

status = 0;

if (vcpu_id >= CONFIG_MAX_VCPUS_PER_VM) {
        pr_err("vcpu id is out of range \r\n");
        status = -EINVAL;
} else if ((&(vm->hw.vcpu_array[vcpu_id]))->state == VCPU_OFFLINE) {
        pr_err("vcpu is offline \r\n");
        status = -EINVAL;
}

if (status == 0) {
        vcpu = vcpu_from_vid(vm, vcpu_id);
        ...
}

Module Level

Functional Safety Consideration

Data verification, and explicit specification of pre-conditions and post-conditions are applied for internal functions of the hypervisor according to Table A.4 of FuSa Standards [IEC_61508-3_2010] .

Error Handling Methods

The error handling methods used in the ACRN hypervisor on a module level are shown below.

Panic the system via ASSERT

The hypervisor can panic the system when the below cases occur. It shall only be used for debugging, used to check pre-conditions and post-conditions to catch design errors.

This method applies to the following case:

  • Software design errors occur.

Rules of Error Detection and Error Handling

The rules of error detection and error handling on a module level are shown in Table 12 below.

Table 12 Rules of Error Detection and Error Handling on Module Level
Resource Class Failure Mode Error Detection via Hypervisor Error Handling Policy Example
Internal data of the hypervisor N/A Partial. The related pre-conditions are required. The design will guarantee the correctness and the test cases will verify the related pre-conditions. If the design can not guarantee the correctness, the related error handling codes need to be added. Note: Some examples of pre-conditions are listed, like non-empty array, valid array size and non-null pointer. The hypervisor shall use the internal resource/data directly. virtual PCI device information, defined with array ‘pci_vdevs[]’ through static allocation.
Configuration data of the VM Corrupted VM config No. The related pre-conditions are required. Note: VM configuration data are auto generated based on different board configs, they are defined as static structure. The bootloader initializes hypervisor (including code, data, and bss) and verifies the integrity of hypervisor image in which VM configurations are. Thus hypervisor does not need any additional mechanism. ‘vm_config->pci_devs’ is configured statically.
Configuration data of the hypervisor N/A No. The related pre-conditions are required. The design will guarantee the correctness and this shall be verified manually. The hypervisor shall use the internal resource/data directly. The maximum number of PCI devices in the VM, defined with CONFIG_MAX_PCI_DEV_NUM through configuration.

Examples

Here are some examples to illustrate when error handling codes are required on a module level.

Example_1: Analyze the function ‘partition_mode_vpci_init’

/**
 * @pre vm != NULL
 * @pre vm->vpci->pci_vdev_cnt <= CONFIG_MAX_PCI_DEV_NUM
 */
static int32_t partition_mode_vpci_init(const struct acrn_vm *vm)
{
        struct acrn_vpci *vpci = (struct acrn_vpci *)&(vm->vpci);
        struct pci_vdev *vdev;
        struct acrn_vm_config *vm_config = get_vm_config(vm->vm_id);
        struct acrn_vm_pci_dev_config *pci_dev_config;
        uint32_t i;

        vpci->pci_vdev_cnt = vm_config->pci_dev_num;

        for (i = 0U; i < vpci->pci_vdev_cnt; i++) {
                vdev = &vpci->pci_vdevs[i];
                vdev->vpci = vpci;
                pci_dev_config = &vm_config->pci_devs[i];
                vdev->vbdf.value = pci_dev_config->vbdf.value;

                if (vdev->vbdf.value != 0U) {
                        partition_mode_pdev_init(vdev, pci_dev_config->pbdf);
                        vdev->ops = &pci_ops_vdev_pt;
                } else {
                        vdev->ops = &pci_ops_vdev_hostbridge;
                }

                if (vdev->ops->init != NULL) {
                        if (vdev->ops->init(vdev) != 0) {
                                pr_err("%s() failed at PCI device (vbdf %x)!",
                                        __func__, vdev->vbdf);
                        }
                }
        }

        return 0;
}

get_vm_config is called by partition_mode_vpci_init. There are one pre-condition and two post-conditions of get_vm_config. It indicates that the caller of get_vm_config shall guarantee these pre-conditions and get_vm_config itself shall guarantee the post-condition.

/**
 * @pre vm_id < CONFIG_MAX_VM_NUM
 * @post retval != NULL
 * @post retval->pci_dev_num <= MAX_PCI_DEV_NUM
 */
struct acrn_vm_config *get_vm_config(uint16_t vm_id)
{
        return &vm_configs[vm_id];
}

Question_1: Is error checking required for ‘vm_config’?

No. Because ‘vm_config’ is getting data from get_vm_config and the post-condition of get_vm_config guarantees that the return value is not NULL.

Question_2: Is error checking required for ‘vdev’?

No. Here are the reasons:

  1. The pre-condition of partition_mode_vpci_init guarantees that ‘vm’ is not NULL. It indicates that ‘vpci’ is not NULL. Since ‘vdev’ is getting data from the array ‘pci_vdevs[]’ via indexing, ‘vdev’ is not NULL as long as the index is valid.
  2. The post-condition of get_vm_config guarantees that ‘vpci->pci_vdev_cnt’ is less than or equal to ‘CONFIG_MAX_PCI_DEV_NUM’, which is the array size of ‘pci_vdevs[]’. It indicates that the index used to get ‘vdev’ is always valid.

Given the two reasons above, ‘vdev’ is always not NULL. So, the error checking codes are not required for ‘vdev’.

Question_3: Is error checking required for ‘pci_dev_config’?

No. ‘pci_dev_config’ is getting data from the array ‘pci_vdevs[]’, which is the physical PCI device information coming from Board Support Package and firmware. For physical PCI device information, the related application constraints shall be defined in the design document or safety manual. For debug purpose, developers could use ASSERT here to catch the Board Support Package or firmware failures, which does not guarantee these application constraints.

Question_4: Is error checking required for ‘vdev->ops->init’?

No. Here are the reasons:

  1. Question_2 proves that ‘vdev’ is always not NULL.
  2. ‘vdev->ops’ is fully initialized before ‘vdev->ops->init’ is called.

Given the two reasons above, ‘vdev->ops->init’ is always not NULL. So, the error checking codes are not required for ‘vdev->ops->init’.

Question_5: How to handle the case when ‘vdev->ops->init(vdev)’ returns non-zero?

This case indicates that the initialization of specific virtual device fails. Investigation has to be done to figure out the root-cause. Default fatal error handler shall be invoked here if it is caused by a hardware failure or invalid boot information.

Example_2: Analyze the function ‘partition_mode_vpci_deinit’

/**
 * @pre vdev != NULL
 * @pre vm->vpci->pci_vdev_cnt <= CONFIG_MAX_PCI_DEV_NUM
 */
static void partition_mode_vpci_deinit(const struct acrn_vm *vm)
{
        struct pci_vdev *vdev;
        uint32_t i;

        for (i = 0U; i < vm->vpci.pci_vdev_cnt; i++) {
                vdev = (struct pci_vdev *) &(vm->vpci.pci_vdevs[i]);
                if ((vdev->ops != NULL) && (vdev->ops->deinit != NULL)) {
                        if (vdev->ops->deinit(vdev) != 0) {
                                pr_err("vdev->ops->deinit failed!");
                        }
                }
                /* TODO: implement the deinit of 'vdev->ops' */
        }
}

Question_6: Is error checking required for ‘vdev->ops’ and ‘vdev->ops->init’?

Yes. Because ‘vdev->ops’ and ‘vdev->ops->init’ can not be guaranteed to be not NULL. If the VM called partition_mode_vpci_deinit twice, it may be NULL.

Module Level Configuration Design Guidelines

Design Goals

There are two goals for module level configuration design, as shown below:

  1. In order to make the hypervisor more flexible, one source code and binary is preferred for different platforms with different configurations;
  2. If one module is not used by a specific project, the module source code is treated as dead code. The effort to configure it in/out shall be minimized.

Hypervisor Operation Modes

The hypervisor operation modes are shown in Table 13 below.

Table 13 Hypervisor Operation Modes
Operation Modes Sub-modes Description
INIT mode DETECT mode The hypervisor detects firmware, detects hardware resource, and reads configuration data.
STARTUP mode The hypervisor initializes hardware resources, creates virtual resources like VCPU and VM, and executes VMLAUNCH instruction(the very first VM entry).
OPERATIONAL mode N/A After the first VM entry, the hypervisor runs in VMX root mode and guest OS runs in VMX non-root mode.
TERMINATION mode N/A If any fatal error is detected, the hypervisor will enter TERMINATION mode. In this mode, a default fatal error handler will be invoked to handle the fatal error.

Configurable Module Properties

The properties of configurable modules are shown below:

  • The functionality of the module depends on platform configurations;
  • Corresponding platform configurations can be detected in DETECT mode;
  • The module APIs shall be configured in DETECT mode;
  • The module APIs shall be used in modes other than DETECT mode.

Platform configurations include:

  • Features depending on hardware or firmware
  • Configuration data provided by firmware
  • Configuration data provided by BSP

Design Rules

The module level configuration design rules are shown below:

  1. The platform configurations shall be detectable by hypervisor in DETECT mode;
  2. Configurable module APIs shall be abstracted as operations which are implemented through a set of function pointers in the operations data structure;
  3. Every function pointer in the operations data structure shall be instantiated as one module API in DETECT mode and the API is allowed to be implemented as empty function for some specific configurations;
  4. The operations data structure shall be read-only in STARTUP mode, OPERATIONAL mode, and TERMINATION mode;
  5. The configurable module shall only be accessed via APIs in the operations data structure in STARTUP mode or OPERATIONAL mode;
  6. In order to guarantee that the function pointer in the operations data structure is dereferenced after it has been instantiated, the pre-condition shall be added for the function which dereferences the function pointer, instead of checking the pointer for NULL.

Note

The third rule shall be double checked during code review.

Use Cases

The following table shows some use cases of module level configuration design:

Table 14 Module Level Configuration Design Use Cases
Platform Configuration Configurable Module Prerequisite
Features depending on hardware or firmware This module is used to virtualize part of LAPIC functionalities. It can be done via APICv or software emulation depending on CPU capabilities. For example, KBL NUC doesn’t support virtual-interrupt delivery, while other platforms support it. If a function pointer is used, the prerequisite is “hv_operation_mode == OPERATIONAL”.
Configuration data provided by firmware This module is used to interact with firmware (UEFI or SBL), and the configuration data is provided by firmware. For example, UP2 uses SBL and KBL NUC uses UEFI. If a function pointer is used, the prerequisite is “hv_operation_mode != DETECT”.
Configuration data provided by BSP This module is used to virtualize LAPIC, and the configuration data is provided by BSP. For example, some VMs use LAPIC pass-through and the other VMs use vLAPIC. If a function pointer is used, the prerequisite is “hv_operation_mode == OPERATIONAL”.

Note

Prerequisite is used to guarantee that the function pointer used for configuration is dereferenced after it has been instantiated.

Examples

Take the module for parsing boot information as an example to illustrate the idea of module level configuration design.

../_images/boot_information_parsing_module.png

Figure 248 Boot Information Parsing Module

As shown in the source code below, ‘struct firmware_operations’ is an operations data structure that contains a set of function pointers. Different firmware may have different implementations:

  • ‘firmware_uefi_ops’ is for UEFI platform;
  • ‘firmware_sbl_ops’ is for SBL platform.
struct firmware_operations {
        void (*init)(void);
        uint64_t (*get_ap_trampoline)(void);
        void *(*get_rsdp)(void);
        void (*init_irq)(void);
        int32_t (*init_vm_boot_info)(struct acrn_vm *vm);
};

static struct firmware_operations firmware_uefi_ops = {
        .init = uefi_init,
        .get_ap_trampoline = uefi_get_ap_trampoline,
        .get_rsdp = uefi_get_rsdp,
        .init_irq = uefi_init_irq,
        .init_vm_boot_info = uefi_init_vm_boot_info,
};

static struct firmware_operations firmware_sbl_ops = {
        .init = sbl_init,
        .get_ap_trampoline = sbl_get_ap_trampoline,
        .get_rsdp = sbl_get_rsdp,
        .init_irq = sbl_init_irq,
        .init_vm_boot_info = sbl_init_vm_boot_info,
};

‘firmware_ops’ is the operations set that is dereferenced and takes effect.

‘init_firmware_operations’ is called when the hypervisor is in DETECT mode and ‘firmware_ops’ is instantiated here to either ‘firmware_uefi_ops’ or ‘firmware_sbl_ops’ depending on the platform.

Note

All the other exported interfaces using ‘firmware_ops’ shall be called after the instantiation.

static struct firmware_operations *firmware_ops;

struct firmware_operations* uefi_get_firmware_operations(void)
{
        return &firmware_uefi_ops;
}

struct firmware_operations* sbl_get_firmware_operations(void)
{
        return &firmware_sbl_ops;
}

void init_firmware_operations(void)
{
        if (is_firmware_sbl()) {
                firmware_ops = sbl_get_firmware_operations();
        } else {
                firmware_ops = uefi_get_firmware_operations();
        }
}

For example, when the hypervisor needs to initialize the VM boot information, it calls ‘firmware_init_vm_boot_info’ and ‘firmware_ops->init_vm_boot_info’ is dereferenced here with correct API being called.

/**
 * @pre firmware_ops->init_vm_boot_info != NULL
 */
int32_t firmware_init_vm_boot_info(struct acrn_vm *vm)
{
        return firmware_ops->init_vm_boot_info(vm);
}

References

[IEC_61508-3_2010](1, 2) IEC 61508-3:2010, Functional safety of electrical/electronic/programmable electronic safety-related systems - Part 3: Software requirements