Moving the whole codebase to rely on the AccessPlatform definition from
vm-virtio so that we can fully remove it from virtio-queue crate.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This table is listed as required in the ARM Base Boot Requirements
document. The particular need arises to make the serial debugging of
Windows guest functional.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>
If the underlying kernel is old PTY resize is disabled and this is
represented by the use of None in the provided Option<File> type. In the
virtio-console PTY path don't blindly unwrap() the value that will be
preserved across a reboot.
Fixes: #3496
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Setting up the SIGWINCH handler requires at least Linux 5.7. However
this functionality is not required for basic PTY operation.
Fixes: #3456
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Advertise the PCI MMIO config spaces here so that the MMIO config space
is correctly recognised.
Tested by: --platform num_pci_segments=1 or 16 hotplug NVMe vfio-user device
works correctly with hypervisor-fw & OVMF and direct kernel boot.
Fixes: #3432
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
When restoring replace the internal value of the device tree rather than
replacing the Arc<Mutex<DeviceTree>> itself. This is fixes an issue
where the AddressManager has a copy of the the original
Arc<Mutex<DeviceTree>> from when the DeviceManager was created. The
original restore path only replaced the DeviceManager's version of the
Arc<Mutex<DeviceTree>>. Instead replace the contents of the
Arc<Mutex<DeviceTree>> so all users see the updated version.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Currently, a tuple containing PCI space start address and PCI space
size is used to pass the PCI space information to the FDT creator.
In order to support the multiple PCI segment for FDT, more information
such as the PCI segment ID should be passed to the FDT creator. If we
still use a tuple to store these information, the code flexibility and
readablity will be harmed.
To address this issue, this commit replaces the tuple containing the
PCI space information to a structure `PciSpaceInfo` and uses a vector
of `PciSpaceInfo` to store PCI space information for each segment, so
that multiple PCI segment information can be passed to the FDT together.
Note that the scope of this commit will only contain the refactor of
original code, the actual multiple PCI segments support will be in
following series, and for now `--platform num_pci_segments` should only
be 1.
Signed-off-by: Henry Wang <Henry.Wang@arm.com>
That function call can return -1 when it fails. Wrapping -1 into File
causes the code to panic when the File is dropped.
Signed-off-by: Wei Liu <liuwe@microsoft.com>
I encountered some trouble trying to use a virtio-console hooked up to
a PTY. Reading from the PTY would produce stuff like this
"\n\nsh-5.1# \n\nsh-5.1# " (where I'm just pressing enter at a shell
prompt), and a terminal would render that like this:
----------------------------------------------------------------
sh-5.1#
sh-5.1#
----------------------------------------------------------------
This was because we weren't disabling the ICRNL termios iflag, which
turns carriage returns (\r) into line feeds (\n). Other raw mode
implementations (like QEMU's) set this flag, and don't have this
problem.
Instead of fixing our raw mode implementation to just disable ICRNL,
or copy the flags from QEMU's, though, here I've changed it to use the
raw mode implementation in libc. It seems to work correctly in my
testing, and means we don't have to worry about what exactly raw mode
looks like under the hood any more.
Signed-off-by: Alyssa Ross <hi@alyssa.is>
This allocator allocates 64-bit MMIO addresses for use with platform
devices e.g. ACPI control devices and ensures there is no overlap with
PCI address space ranges which can cause issues with PCI device
remapping.
Use this allocator the ACPI platform devices.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
With the segment id now encoded in the bdf it is not necessary to have
the separate field for it.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
In particular use the accessor for getting the device id from the bdf.
As a side effect the VIOT table is now segment aware.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Since each segment must have a non-overlapping memory range associated
with it the device memory must be equally divided amongst all segments.
A new allocator is used for each segment to ensure that BARs are
allocated from the correct address ranges. This requires changes to
PciDevice::allocate/free_bars to take that allocator and when
reallocating BARs the correct allocator must be identified from the
ranges.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
For all the devices that support being hotplugged (disk, net, pmem, fs
and vsock) add "pci_segment" option and propagate that through to the
addition onto the PCI busses.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Move the decision on whether to use a 64-bit bar up to the DeviceManager
so that it can use both the device type (e.g. block) and the PCI segment
ID to decide what size bar should be used.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Generate a set of 8 IRQs and round-robin distribute those over all the
slots for a bus. This same set of IRQs is then used for all PCI
segments.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
The platform config may specify a number of PCI segments to use, if this
greater than 1 then we add supplemental PCI segments as well as the
default segment.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
For the bus scanning the GED AML code now calls into a PSCN method that
scans all buses. This approach was chosen since it handles the case
correctly where one GED interrupt is services for two hotplugs on
distinct segments.
The PCIU and PCID field values are now determined by the PSEG field that
is uses to select which segment those values should be used for.
Similarly _EJ0 will notify based on the value of _SEG.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Replace the hardcoded zero PCI segment id when adding devices to the bus
and extend the DeviceTree to hold the PCI segment id.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
For now this still contains just one segment but is expanding in
preparation for more segments.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
This commit moves the code that generates the DSDT data for the PCI bus
into PciSegment making no functional changes to the generated AML.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Relying on the vm-virtio/virtio-queue crate from rust-vmm which has been
copied inside the Cloud Hypervisor tree, the entire codebase is moved to
the new definition of a Queue and other related structures.
The reason for this move is to follow the upstream until we get some
agreement for the patches that we need on top of that to make it
properly work with Cloud Hypervisor.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
When allocating PCI MMIO BARs they should always be naturally aligned
(i.e. aligned to the size of the BAR itself.)
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
The allocate_bars method has a side effect which collates the BARs used
for the device and stores them internally. Ensure that any use of this
internal state is after the state is created otherwise no MMIO regions
will be seen and so none will be mapped.
Fixes: #3237
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Move the PciSegment struct and the associated code to a new file. This
will allow some clearer separation between the core DeviceManager and
PCI handling.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Move the PCI related state from the DeviceManager struct to a PciSegment
struct inside the DeviceManager. This is in preparation for multiple
segment support. Currently this state is just the bus itself, the MMIO
and PIO config devices and hotplug related state.
The main change that this required is using the Arc<Mutex<PciBus>> in
the device addition logic in order to ensure that
the bus could be created earlier.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
By creating the BlocksState object in the MemoryManager, we can directly
provide it to the virtio-mem device when being created. This will allow
the MemoryManager through each VirtioMemZone to have a handle onto the
blocks that are plugged at any point in time.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Adding the snapshot/restore support along with migration as well,
allowing a VM with virtio-mem devices attached to be properly
migrated.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Since we only support a single PCI bus right now advertise only a single
bus in the ACPI tables. This reduces the number of VM exits from probing
substantially.
Number of PCI config I/O port exits: 17871 -> 1551 (91% reduction) with
direct kernel boot.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Use a simpler method for extracting the affected slot on the eject
command. Also update the terminology to reflect that this a slot rather
than a bdf (which is what device id refers to elsewhere.)
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Refactor the serial buffer handling in order to write the serial
buffer's output to a PTY connected after the serial device stops being
written to by the guest.
This change moves the serial buffer initialization inside the serial
manager. That is done to allow the serial buffer to be made aware of
the PTY and epoll fds needed in order to modify the
EpollDispatch::File trigger. These are then used by the serial buffer
to trigger an epoll event when the PTY fd is writable and the buffer
has content in it. They are also used to remove the trigger when the
buffer is emptied in order to avoid unnecessary wake-ups.
Signed-off-by: William Douglas <william.douglas@intel.com>
Implement the infrastructure that lets a virtio-mem device map the guest
memory into the device. This is necessary since with virtio-mem zones
memory can be added or removed and the vfio-user device must be
informed.
Fixes: #3025
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
By moving this from the VfioUserPciDevice to DeviceManager the client
can be reused for handling DMA mapping behind an IOMMU.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
For vfio-user the mapping handler is per device and needs to be removed
when the device in unplugged.
For VFIO the mapping handler is for the default VFIO container (used
when no vIOMMU is used - using a vIOMMU does not require mappings with
virtio-mem)
To represent these two use cases use an enum for the handlers that are
stored.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Looking up devices on the port I/O bus is time consuming during the
boot at there is an O(lg n) tree lookup and the overhead from taking a
lock on the bus contents.
Avoid this by adding a fast path uses the hardcoded port address and
size and directs PCI config requests directly to the device.
Command line:
target/release/cloud-hypervisor --kernel ~/src/linux/vmlinux --cmdline "root=/dev/vda1 console=ttyS0" --serial tty --console off --disk path=~/workloads/focal-server-cloudimg-amd64-custom-20210609-0.raw --api-socket /tmp/api
PIO exit: 17913
PCI fast path: 17871
Percentage on fast path: 99.8%
perf before:
marvin:~/src/cloud-hypervisor (main *)$ perf report -g | grep resolve
6.20% 6.20% vcpu0 cloud-hypervisor [.] vm_device:🚌:Bus::resolve
perf after:
marvin:~/src/cloud-hypervisor (2021-09-17-ioapic-fast-path *)$ perf report -g | grep resolve
0.08% 0.08% vcpu0 cloud-hypervisor [.] vm_device:🚌:Bus::resolve
The compromise required to implement this fast path is bringing the
creation of the PciConfigIo device into the DeviceManager::new() so that
it can be used in the VmmOps struct which is created before
DeviceManager::create_devices() is called.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
The MSI IOVA address on X86 and AArch64 is different.
This commit refactored the code to receive the MSI IOVA address and size
from device_manager, which provides the actual IOVA space data for both
architectures.
Signed-off-by: Michael Zhao <michael.zhao@arm.com>
This change switches from handling serial input in the VMM thread to
its own thread controlled by the SerialManager.
The motivation for this change is to avoid the VMM thread being unable
to process events while serial input is happening and vice versa.
The change also makes future work flushing the serial buffer on PTY
connections easier.
Signed-off-by: William Douglas <william.douglas@intel.com>
The clone method for PtyPair should have been an impl of the Clone
trait but the method ended up not being used. Future work will make
use of the trait however so correct the missing trait implementation.
Signed-off-by: William Douglas <william.douglas@intel.com>
For most use cases, there is no need to create multiple VFIO containers
as it causes unwanted behaviors. Especially when passing multiple
devices from the same IOMMU group, we need to use the same container so
that it can properly list the groups that have been already opened. The
correct logic was already there in vfio-ioctls, but it was incorrectly
used from our VMM implementation.
For the special case where we put a VFIO device behind a vIOMMU, we must
create one container per device, as we need to control the DMA mappings
per device, which is performed at the container level. Because we must
keep one container per device, the vIOMMU use case prevents multiple
devices attached to the same IOMMU group to be passed through the VM.
But this is a limitation that we are fine with, especially since the
vIOMMU doesn't let us group multiple devices in the same group from a
guest perspective.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
When a pty is resized (using the TIOCSWINSZ ioctl -- see ioctl_tty(2)),
the kernel will send a SIGWINCH signal to the pty's foreground process
group to notify it of the resize. This is the only way to be notified
by the kernel of a pty resize.
We can't just make the cloud-hypervisor process's process group the
foreground process group though, because a process can only set the
foreground process group of its controlling terminal, and
cloud-hypervisor's controlling terminal will often be the terminal the
user is running it in. To work around this, we fork a subprocess in a
new process group, and set its process group to be the foreground
process group of the pty. The subprocess additionally must be running
in a new session so that it can have a different controlling
terminal. This subprocess writes a byte to a pipe every time the pty
is resized, and the virtio-console device can listen for this in its
epoll loop.
Alternatives I considered were to have the subprocess just send
SIGWINCH to its parent, and to use an eventfd instead of a pipe.
I decided against the signal approach because re-purposing a signal
that has a very specific meaning (even if this use was only slightly
different to its normal meaning) felt unclean, and because it would
have required using pidfds to avoid race conditions if
cloud-hypervisor had terminated, which added complexity. I decided
against using an eventfd because using a pipe instead allows the child
to be notified (via poll(2)) when nothing is reading from the pipe any
more, meaning it can be reliably notified of parent death and
terminate itself immediately.
I used clone3(2) instead of fork(2) because without
CLONE_CLEAR_SIGHAND the subprocess would inherit signal-hook's signal
handlers, and there's no other straightforward way to restore all signal
handlers to their defaults in the child process. The only way to do
it would be to iterate through all possible signals, or maintain a
global list of monitored signals ourselves (vmm:vm::HANDLED_SIGNALS is
insufficient because it doesn't take into account e.g. the SIGSYS
signal handler that catches seccomp violations).
Signed-off-by: Alyssa Ross <hi@alyssa.is>
This prepares us to be able to handle console resizes in the console
device's epoll loop, which we'll have to do if the output is a pty,
since we won't get SIGWINCH from it.
Signed-off-by: Alyssa Ross <hi@alyssa.is>
This concept ends up being broken with multiple types on input connected
e.g. console on TTY and serial on PTY. Already the code for checking for
injecting into the serial device checks that the serial is configured.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Introduce a common solution for spawning the virtio threads which will
make it easier to add the panic handling.
During this effort I discovered that there were no seccomp filters
registered for the vhost-user-net thread nor the vhost-user-block
thread. This change also incorporates basic seccomp filters for those as
part of the refactoring.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Current AArch64 power button is only for device tree using a PL061
GPIO controller device. Since AArch64 now supports ACPI, this
commit extend the power button on AArch64 to:
- Using GED for ACPI+UEFI boot.
- Using PL061 for device tree boot.
Signed-off-by: Henry Wang <Henry.Wang@arm.com>
Move the processing of the input from stdin, PTY or file from the VMM
thread to the existing virtio-console thread. The handling of the resize
of a virtio-console has not changed but the name of the struct used to
support that has been renamed to reflect its usage.
Fixes: #3060
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Downcasting of GicDevice trait might fail. Therefore we try to
downcast the trait first and only if the downcasting succeeded we
can then use the object to call methods. Otherwise, do nothing and
log the failure.
Signed-off-by: Henry Wang <Henry.Wang@arm.com>
This commit implements the GIC (including both GICv3 and GICv3ITS)
Pausable trait. The pause of device manager will trigger a "pause"
of GIC, where we flush GIC pending tables and ITS tables to the
guest RAM.
Signed-off-by: Henry Wang <Henry.Wang@arm.com>
This prevents the boot of the guest kernel from being blocked by
blocking I/O on the serial output since the data will be buffered into
the SerialBuffer.
Fixes: #3004
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Introduce a dynamic buffer for storing output from the serial port. The
SerialBuffer implements std::io::Write and can be used in place of the
direct output for the serial device.
The internals of the buffer is a vector that grows dynamically based on
demand up to a fixed size at which point old data will be overwritten.
Currently the buffer is only flushed upon writes.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Check the config to find out which device is attached to the tty and
then send the input from the user into that device (serial or
virtio-console.)
Fixes: #3005
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
vhdx_sync.rs in block_util implements traits to represent the vhdx
crate as a supported block device in the cloud hypervisor. The vhdx
is added to the block device list in device_manager.rs at the vmm
crate so that it can automatically detect a vhdx disk and invoke the
corresponding crate.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Signed-off-by: Fazla Mehrab <akm.fazla.mehrab@intel.com>
This is to make sure the NUMA node data structures can be accessed
both from the `vmm` crate and `arch` crate.
Signed-off-by: Henry Wang <Henry.Wang@arm.com>
The AArch64 platform provides a NUMA binding for the device tree,
which means on AArch64 platform, the NUMA setup can be extended to
more than the ACPI feature.
Based on above, this commit extends the NUMA setup and data
structures to following scenarios:
- All AArch64 platform
- x86_64 platform with ACPI feature enabled
Signed-off-by: Henry Wang <Henry.Wang@arm.com>
Signed-off-by: Michael Zhao <Michael.Zhao@arm.com>
Instead of panicking with an expect() function, the QcowDiskSync::new
function now propagates the error properly. This ensures the VMM will
not panic, which might be the source of weird errors if only one thread
exits while the VMM continues to run.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
We cannot let vhost-user devices connect to the backend when the Block,
Fs or Net object is being created during a restore/migration. The reason
is we can't have two VMs (source and destination) connected to the same
backend at the same time. That's why we must delay the connection with
the vhost-user backend until the restoration is performed.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
In anticipation for creating vhost-user devices in a different way when
being restored compared to a fresh start, this commit introduces a new
boolean created by the Vm depending on the use case, and passed down to
the DeviceManager. In the future, the DeviceManager will use this flag
to assess how vhost-user devices should be created.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Create the vfio-user / user devices from the config. Currently hotplug
of the devices is not supported nor can they be placed behind the
(virt-)iommu.
Removal of the coldplugged device is however supported.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
This patch connects the dots between the vm.rs code and each Migratable
device, in order to make sure Migratable methods are correctly invoked
when migration happens.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
It's totally acceptable to snapshot and restore a virtio-fs device that
has no cache region, since this is a valid mode of functioning for
virtio-fs itself.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
As we are now using an global control to start/stop dirty pages log from
the `hypervisor` crate, we need to explicitly tell the hypervisor (KVM)
whether a region needs dirty page tracking when it is created.
This reverts commit f063346de3.
Signed-off-by: Bo Chen <chen.bo@intel.com>
With the support of dynamically turning on/off dirty-pages-log during
live-migration (only for guest RAM regions), we now can create guest
memory regions without dirty-pages-log by default both for guest RAM
regions and other regions backed by file/device.
Signed-off-by: Bo Chen <chen.bo@intel.com>
When running a TDX guest, we need the virtio drivers to use the DMA API
to share specific memory pages with the VMM on the host. The point is to
let the VMM get access to the pages related to the buffers pointed by
the virtqueues.
The way to force the virtio drivers to use the DMA API is by exposing
the virtio devices with the feature VIRTIO_F_IOMMU_PLATFORM. This is a
feature indicating the device will require some address translation, as
it will not deal directly with physical addresses.
Cloud Hypervisor takes care of this requirement by adding a generic
parameter called "force_iommu". This parameter value is decided based on
the "tdx" feature gate, and then passed to the DeviceManager. It's up to
the DeviceManager to use this parameter on every virtio device creation,
which will imply setting the VIRTIO_F_IOMMU_PLATFORM feature.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Previously the same function was used to both create and remove regions.
This worked on KVM because it uses size 0 to indicate removal.
MSHV has two calls -- one for creation and one for removal. It also
requires having the size field available because it is not slot based.
Split set_user_memory_region to {create/remove}_user_memory_region. For
KVM they still use set_user_memory_region underneath, but for MSHV they
map to different functions.
This fixes user memory region removal on MSHV.
Signed-off-by: Wei Liu <liuwe@microsoft.com>
We have been building Cloud Hypervisor with command like:
`cargo build --no-default-features --features ...`.
After implementing ACPI, we donot have to use specify all features
explicitly. Default build command `cargo build` can work.
This commit fixed some build warnings with default build option and
changed github workflow correspondingly.
Signed-off-by: Michael Zhao <michael.zhao@arm.com>
Issue from beta verion of clippy:
Error: --> vm-virtio/src/queue.rs:700:59
|
700 | if let Some(used_event) = self.get_used_event(&mem) {
| ^^^^ help: change this to: `mem`
|
= note: `-D clippy::needless-borrow` implied by `-D warnings`
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#needless_borrow
Signed-off-by: Bo Chen <chen.bo@intel.com>
The Linux kernel expects that any PCI devices that advertise I/O bars
have use an address that is within the range advertised by the bus
itself. Unfortunately we were not advertising any I/O ports associated
with the PCI bus in the ACPI tables.
Signed-off-by: Rob Bradford <robert.bradford@intel.com>
Make sure the unique PCI bus is tied to the default NUMA node 0, and
update the documentation to let the users know about this special case.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Sometimes we need balloon deflate automatically to give memory
back to guest, especially for some low priority guest processes
under memory pressure. Enable deflate_on_oom to support this.
Usage: --balloon "size=0,deflate_on_oom=on" \
Signed-off-by: Fei Li <lifei.shirley@bytedance.com>
Since using the VIRTIO configuration to expose the virtual IOMMU
topology has been deprecated, the virtio-iommu implementation must be
updated.
In order to follow the latest patchset that is about to be merged in the
upstream Linux kernel, it must rely on ACPI, and in particular the newly
introduced VIOT table to expose the information about the list of PCI
devices attached to the virtual IOMMU.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
_DSM (Device Specific Method) is a control method that enables devices
to provide device specific control functions. Linux kernel will evaluate
this device then initialize preserve_config in acpi pci initialization.
Signed-off-by: Jianyong Wu <jianyong.wu@arm.com>
As the first step to complete live-migration with tracking dirty-pages
written by the VMM, this commit patches the dependent vm-memory crate to
the upstream version with the dirty-page-tracking capability. Most
changes are due to the updated `GuestMemoryMmap`, `GuestRegionMmap`, and
`MmapRegion` structs which are taking an additional generic type
parameter to specify what 'bitmap backend' is used.
The above changes should be transparent to the rest of the code base,
e.g. all unit/integration tests should pass without additional changes.
Signed-off-by: Bo Chen <chen.bo@intel.com>
After adding "get_interrupt_controller()" function in DeviceManager,
"enable_interrupt_controller()" became redundant, because the latter
one is the a simple wrapper on the interrupt controller.
Signed-off-by: Michael Zhao <michael.zhao@arm.com>
The function used to calculate "gicr-typer" value has nothing with
DeviceManager. Now it is moved to AArch64 specific files.
Signed-off-by: Michael Zhao <michael.zhao@arm.com>
