It still tries to access the gpu->vdpy_handel after the gpu
is free. This is incorrect.
Tracked-On: #7337
Acked-by: Wang Yu <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
When virtio-gpu tries to submit the framebuffer based on
VIRTIO_GPU_CMD_2D, one pixman_image will be created. When the
sdl rendering_thread is terminated, the pixman_image will be
released. But its pointer is set to NULL. In the next reboot,
its access in vdpy_surface_set is incorrect.
Tracked-On: #7337
Acked-by: Wang Yu <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Now all the 3D ops are handled in one dedicated thread. As 3D ops are
not safe in multi-thread env, some checks are added so that it can indicate
that it is not in the expected code path.
Tracked-On: #7296
Acked-by: Wang Yu1 <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Otherwise the memory related with vga_context is leaked.
v1->v2: Use the pthread_join instead of usleep to wait for
the termination of vga_thread.
Tracked-On: #7296
Acked-by: Wang Yu1 <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
This is not needed as it is reconfigured in course of vga_thread_loop.
At the same time it is not thread-safe as the 3D api is involved
in vdpy_surface_set.
v2->v3: move the initialization of vga_surface width/height into
vga_thread.
Tracked-On: #7296
Acked-by: Wang Yu1 <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
In order to support the VGA op, one vga_thread is created. But
in course of virtio_gpu_reset, it will also create one vga_thread.
This is incorrect.
v1->v2: the thread_id of vga_thread is treated as read-only after
the thread is created.
One thread_status based on atomic type indicates the life_cycle
of vga_thread.
v2->v3: Add the vga_thread_mtx in course of changing life_cycle of vga_thread
Tracked-On: #7296
Acked-by: Wang Yu1 <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
The guest memory is allocated through hugetlb. Huge page fault is
triggered by writing a byte to the starting address for each huge page.
In hugetlb_fault(), The physical pages are allocated and *clear*.
So no need to erasing the memory again after hugetlb setup. This will
bring about 130ms (changes with platform) latency for each 1G memory.
Notice: this means we depends on kernel to erase the memory in huge page
fault!
Tracked-On: #7298
Signed-off-by: Conghui <conghui.chen@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Add license information to virtio_gpu.c
Fix some grammar issues.
Tracked-On: #7241
Signed-off-by: Sun Peng <peng.p.sun@linux.intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Many of the license and Intel copyright headers include the "All rights
reserved" string. It is not relevant in the context of the BSD-3-Clause
license that the code is released under. This patch removes those strings
throughout the code (hypervisor, devicemodel and misc).
Tracked-On: #7254
Signed-off-by: Geoffroy Van Cutsem <geoffroy.vancutsem@intel.com>
The DMABuf sharing is based on the BLOB feature. This will
check the attribute of system memory for guest_vm and udmabuf.list_limit
parameter. If it meets with the requirement, export the BLOB feature
so that the FE driver in guest_vm can send the blob cmd.
VIRTIO_GPU_CMD_CREATE_BLOB:
VIRTIO_GPU_CMD_SET_SCANOUT_BLOB:
VIRTIO_GPU_CMD_SET_FLUSH(BLOB)
Tracked-On: #7210
Acked-by: Wang Yu <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Now Virtio-gpu FE is using the below cmds to submit the framebuffer that
needs to be displayed.
VIRTIO_GPU_CMD_TRANSFER_TO_HOST_2D
VIRTIO_GPU_CMD_SET_SCANOUT
VIRTIO_GPU_CMD_FLUSH
This will introduce twice copy. In order to optimize the twice copy, the below
cmds are added and the virtio-gpu BE can use the dma-buf sharing.
VIRTIO_GPU_CMD_CREATE_BLOB
VIRTIO_GPU_CMD_SET_SCANOUT_BLOB
VIRTIO_GPU_CMD_FLUSH
V2->V3: Remove the inlined header file of devicemodel/include/public/udmabuf.h
to avoid the license issue
Tracked-On: #7210
Acked-by: Wang Yu <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Now the Opengl rendering is used implicitly in SDL backend. But sometimes
it needs to create the texture explicitly based on DMABuf so that it can
render the framebuffer submitted from virtio-gpu.
Tracked-On: #7210
Acked-by: Wang Yu <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
After using the memfd to allocate memory for guest vm, it needs to
query the offset in memfd for the given GPA. In order to search the
memfd/offset, the added memory_region from memfd needs to be kept in
one arrary.
Tracked-On: #7210
Acked-by: Wang Yu <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
The ACRN-DM uses the explicit hugetlb file to allocate the memory
for guest and the 2M/1G hugetlb can be used. This is based on mounting
hugetlb filesystem.
Now the Linux provides one flexbile memfd mechanism to allocate the memory
and this can make it possible to create the udmabuf for sharing. And the
memfd still can be based on Hugetlb 1G/2M policy. This can bring the below
benefits besides the 1G/2M hugetlb support.
a. memfd is based on anonymous file
b. possible dmabuf mechanism
v2->v3: After completing the mmap for memfd, it will call the ftruncate to
set the desired size and then add the F_SEAL_SEAL flag.
Tracked-On: #7210
Acked-by: Wang Yu <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Now the memory region is aligned down to 2M if 2M hugetlb is supported.
Maybe some contents are dropped if it is not aligned.
Tracked-On: #7210
Acked-by: Wang Yu <yu1.wang@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
- VGA interface of virtio-gpu adapter support
- Ovmf:ACRN: Fix the building error
- acrn: fix how to get memory below 4G
Tracked-On: #7231
Signed-off-by: Peter Fang <peter.fang@intel.com>
The virtio-gpu needs to include some system header files.
In order to support the cross-compiler building, the header
path is added.
Tracked-On: #7210
Reported-by: Naveen Kumar Saine <naveen.kumar.saini@intel.com>
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Legacy VGA & VBE interface as a common interface is supported by
many legacy and modern OS. Many installer of OS distribution use
this interface to display the GUI of installer when setup a refresh
new installation on bare-metal. Besides, Windows OS always use this
interface to display it's BSOD, recovery mode & safe mode GUI. It
is need because Windows don't include virtio-gpu driver as their
in-box driver, VGA interface will be used before the virtio-gpu
driver been installed.
To be compatiable with the PCI bar layout of legacy VGA, the layout
is refined to meet with the requirement of legacy VGA and modern
virtio-gpu.
BAR0: VGA Framebuffer memory, 16 MB in size.
BAR2: MMIO Space
[0x0000~0x03ff] EDID data blob
[0x0400~0x041f] VGA ioports registers
[0x0500~0x0516] bochs display interface registers
[0x1000~0x17ff] Virtio common configuration registers
[0x1800~0x1fff] Virtio ISR state registers
[0x2000~0x2fff] Virtio device configuration registers
[0x3000~0x3fff] Virtio notification registers
BAR4: MSI/MSI-X
BAR5: Virtio port io
Tracked-On: #7210
Signed-off-by: Sun Peng <peng.p.sun@linux.intel.com>
Reviewed-by: Zhao, yakui <yakui.zhao@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Hardware cursor emulation of virtio-gpu video adapter. Guest vm can
show its own cursor in virtual display, not share the cursor with
service vm. It also accelerated by SDL(OpenGL ES 2.0 backend) API
with hardware acceleration based on the GPU hardware own by service
vm.
Tracked-On: #7210
Signed-off-by: Sun, Peng <peng.p.sun@linux.intel.com>
Reviewed-by: Zhao, yakui <yakui.zhao@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Implement 2D mode of virtio-gpu adapter which can transfer outputs
from guest vm framebuffer and store them to host 2D scanout buffers.
These 2D scanout buffer can be shown by calling vdpy_surface_set/update
API provided device/include/vdisplay.h.
Virtio-gpu 2D commands calling sequence:
1) VIRTIO_GPU_CMD_GET_EDID
2) VIRTIO_GPU_CMD_GET_DISPLAY_INFO
3) VIRTIO_GPU_CMD_RESOURCE_CREATE_2D
4) VIRTIO_GPU_CMD_RESOURCE_ATTACH_BACKING
5) VIRTIO_GPU_CMD_SET_SCANOUT
6) VIRTIO_GPU_CMD_TRANSFER_TO_HOST_2D
7) VIRTIO_GPU_CMD_RESOURCE_FLUSH
Tracked-On: #7210
Signed-off-by: Sun, Peng <peng.p.sun@linux.intel.com>
Reviewed-by: Zhao, yakui <yakui.zhao@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Extended display identification data(EDID) is a data to store
display timings which are supported by ACRN virtual monitor.
Virtio-gpu FE driver will request it to config crtc for display
resolutions.
Tracked-On: #7210
Signed-off-by: Sun, Peng <peng.p.sun@linux.intel.com>
Reviewed-by: Zhao, yakui <yakui.zhao@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
ACRN virtual monitor shown as a GUI system window of service vm.
It can display graphic outputs of guest vm which is stored in
service vm's buffers by virtio-gpu. Display operation is accelerated
by Intel GPU PF(SRIOV) device with SDL(OpenGL ES 2.0 backend) API.
This provides one generic display solution. When the virtio-gpu is
added, it will firstly try to setup the connection to graphics system
and then display the framebuffer from the guest vm in the created
window region.
Tracked-On: #7210
Signed-off-by: Sun, Peng <peng.p.sun@linux.intel.com>
Reviewed-by: Zhao, yakui <yakui.zhao@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Virtio-gpu is a virtio based graphic adapter. It supports 2D mode
which can transfer the guest vm's framebuffer into service vm's
buffers for displaying. It can co-work with Intel GPU VF(SRIOV)
device and provide thransport for the accelerated contents. With
this, guest vm can benefit Intel GPU hardware to accelerate media
coding, 3D rendering and compute.
Tracked-On: #7210
Signed-off-by: Sun, Peng <peng.p.sun@linux.intel.com>
Reviewed-by: Zhao, yakui <yakui.zhao@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
This reverts commit 6750d5a277.
the bug is root caused and fixed, so recovery the original patch.
Tracked-On: #6966
Signed-off-by: Minggui Cao <minggui.cao@intel.com>
i2c bus counter 'acpi_i2c_adapter_num' shall be updated
when virtio-i2c device is de-initialized, else in user VM
reboot case, this global counter will not be reset and
keep the value last user VM boot, which is not expected
and cause iasl(building ACPI data) program crash finally
because the i2c device name space overflow.
Tracked-On: #7208
Signed-off-by: Yonghua Huang <yonghua.huang@intel.com>
Now, the DM assumes the device name entered by user is dependent on
the type.
For example, the device of tap type is named tap_TAAG, this patch
remove the above assumption.
User could set any name, no matter the code of DM or actually created,
it is represented by user input.
Tracked-On: #6690
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Signed-off-by: Chenli Wei<chenli.wei@linux.intel.com>
DM dynamic param '-A' has been removed.
For compability, add a warning for `-A` instead of terminate the launch
process.
Tracked-On:#6690
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Signed-off-by: Yuanyuan Zhao <yuanyuan.zhao@linux.intel.com>
This reverts commit 811992ee2b.
There could be some conflict with current configure tool. will fix it
later.
Tracked-On: #6966
Signed-off-by: Minggui Cao <minggui.cao@intel.com>
for CPU core partition VM, like RTVM, set PMU passthrough
flag for vtune/perf to run in guest VM.
Tracked-On: #6966
Acked-by: Anthony Xu <anthony.xu@intel.com>
Signed-off-by: Minggui Cao <minggui.cao@intel.com>
HMAC_*, MD5_* and SHA256_* are deprecated since openssl3.0, replace them with the corresponding equivalents.
Tracked-On: #6743
Signed-off-by: Tw <wei.tan@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Previous commit introduced the "use physical APIC IDs as vLAPIC IDs for VMs" change,
but it didn't update the apic_id in mptable. Changed the mptable code to also
set apic_id to physical APIC ID to fix the bug
Tracked-On: #7146
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
When the virtio-XXX pci devices fall back to legacy PCI intx,
the pci_irq_assert is called to inject the interrupt and then
the pci_irq_deassert is used to mark the completion of PCI interrupt.
Currently the HV vIOAPIC uses the pin_state for the interrupt injection
of legacy PCI intx. In such case it will fail to inject the PCI legacy
intx and the guest system fails to be booted when adding the boot option
of "pci=nomsi".
PCI legacy INTx usually use active low level trigger mode as it is Open-Drain
state and allows multitple interrupt signals to share a single line.
https://wiki.osdev.org/PCI_Local_Bus_Signals
In such case DM needs to set the correct state for the PCI device so that the
HV vIOAPIC can help to inject the PCI legacy intx.
BTW: When the MSI/MSIX is used for PCI device, it uses another mechanism
to inject the interrupt. It is harmless to configure the initial state.
Tracked-On: #7124
Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
'mevent_lmutex' is initialized as default type,
while attempting to recursively lock on this
kind of mutext results in undefined behaviour.
Recursively lock on 'mevent_lmutex' can be detected
in mevent thread when user tries to trigger system
reset from user VM, in this case, user VM reboot hang.
The backtrace for this issue:
#1 in mevent_qlock () at core/mevent.c:93
#2 in mevent_delete_even at core/mevent.c:357
===>Recursively LOCK
#3 in mevent_delete_close at core/mevent.c:387
#4 in acrn_timer_deinit at core/timer.c:106
#5 in virtio_reset_dev at hw/pci/virtio/virtio.c:171
#6 in virtio_console_reset at
hw/pci/virtio/virtio_console.c:196
#7 in virtio_console_destroy at
hw/pci/virtio/virtio_console.c:1015
#8 in virtio_console_teardown_backend at
hw/pci/virtio/virtio_console.c:1042
#9 in mevent_drain_del_list () at
core/mevent.c:348 ===> 1st LOCK
#10 in mevent_dispatch () at core/mevent.c:472
#11 in main at core/main.c:1110
So the root cause is:
mevent_mutex lock is recursively locked by mevent thread
itself (#9 for this first lock and #2 for recursively lock),
which is not allowed for mutex with default attribute.
This patch changes the mutex type of 'mevent_lmutex'
from default to "PTHREAD_MUTEX_RECURSIVE", because
recrusively lock shall be allowed as user of mevent
may call mevent functions (where mutex lock maybe required)
in teardown callbacks.
Tracked-On: #7133
Signed-off-by: Yonghua Huang <yonghua.huang@intel.com>
Acked-by: Yu Wang <yu1.wang@intel.com>
The current code does not always check the return value of function
ioctl called in ACRN userspace, and lack of error message printing
to help debug.
This code fixes it by checking the return value of ioctl, and adding
function errormsg to return a string describing of the error code.
Tracked-On: #7029
Signed-off-by: Wen Qian <qian.wen@intel.com>
Signed-off-by: Li Fei <fei1.li@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Libvirt or kata container needs to send some commands
(such as VM destory command) to the DM instance of User VM
through command monitor socket, they will specify the socket
path and pass this path name to DM instance through DM parameter.
In this patch, add new DM parameter (cmd_monitor) to get socket
path from libvirt or kata container. If cmd_monitor is specified,
it initialize and deinitialize command monitor in DM main loop.
v2-->v3:
Include command monitor initialization and deinitialization.
Tracked-On: #5921
Signed-off-by: Xiangyang Wu <xiangyang.wu@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
This monitor module is to initialize socket intance, register
handlers to handle command from socket message, close socket,
free socket instance:
init_cmd_monitor: initialize socket intance and register handlers
to handle command.
deinit_cmd_monitor: close socket and free socket instance.
In this patch DM makefile is updated to build command monitor.
v1--v2:
Update socket path and update log message format.
Parse JSON format command message using libcjson lib APIs.
v2-->v3:
Use socket path length MACRO.
Update JSON format command message to {"command": "xxx"}.
Tracked-On: #5921
Signed-off-by: Xiangyang Wu <xiangyang.wu@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
The command handler is to implement handlers for each command,
currently, two handler is implemented for command monitor:
user_vm_destroy_handler: the handler is for user VM destroy command,
which shuts down to standard post-launched user VM forcefully.
user_vm_blkrescan_handler: the handler is for user VM blkrescan
command, which rescan virtio-blk device to revalidate and update
the backend file for user VM.
v1--v2:
Update log message format.
Generate JSON format ack message using libcjson lib APIs.
v2-->v3:
Update ACK message to {"ack": 0} or {"ack": -1}
Tracked-On: #5921
Signed-off-by: Xiangyang Wu <xiangyang.wu@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
The command module provides interfaces to find a command
intance, register handler for specified command, dispatch
command and invoke related handler.
find_command: find a command instance by name.
register_command_handler: register the handler for one
command instance.
dispatch_command_handlers: dispatch the command and invoke
registered handler.
v1-->v2:
Only support single handler for one command instance.
v2-->v3:
Remove command id.
Add error check to avoid regiter handler to command
instance which has handler.
Update the second parameter type of register_command_handler
to avoid unnecessary type conversion.
Tracked-On: #5921
Signed-off-by: Xiangyang Wu <xiangyang.wu@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
The socket module is to provide interfaces below, the unix domain
socket can be created by command monitor for the communication
between DM instance and libvirt daemon or kata container:
init_socket: allocate a new socket instance according to socket path
deinit_socket: free a socket instance
open_socket: open one unix domain socket server, initialize a
socket, create one thread to listen to client, another thread to
poll message from client.
close_socket: close one unix domain socket server
find_socket_client: find socket client instance according to fd
write_socket_char: send message through unix domain socket server
v1--v2:
Update some log message format and copyright format.
v2-->v3:
Update SOCKET_MAX_CLIENT to 1H since the socket instance of
command monitor only have one client (libvirt or kata container).
Tracked-On: #5921
Signed-off-by: Xiangyang Wu <xiangyang.wu@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Enable `lapic_pt` automatically for rtvm for better performance.
Reserve `--lapic_pt` for future use. If VM is not in realtime mode,
`--lapic_pt` will cause a warning.
Tracked-On: #6690
Signed-off-by: Yuanyuan Zhao <yuanyuan.zhao@linux.intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
In function probe_disk_log_file(), handler 'dir'
returned by opendir() is not released before
function return and results in memory leakage.
Tracked-On: #7098
Signed-off-by: Yonghua Huang <yonghua.huang@intel.com>
By default, pCPU is allowed to request software SRAM buffer
from given region in TCC buffer driver only if this pCPU is
set in the target region's CPU affinity configuration.
This check shall be disabled for software SRAM virtualization
usage in ACRN service VM, because software SRAM buffers are
requested by ACRN DM on behalf of user VM, but ACRN DM and
user VM may run on different CPUs while the target software
SRAM region may be configured only for pCPUs that user VM runs on.
This patch turns off such affinity check in TCC driver when
initializing vSSRAM for user VM.
Tracked-On: #7010
Signed-off-by: Yonghua Huang <yonghua.huang@intel.com>
Reviewed-by: Geoffroy Van Cutsem <geoffroy.vancutsem@intel.com>
Acked-by: Yu Wang <yu1.wang@intel.com>
The current mac detection logic only check whether there are parameters
after tap opt. Don't care whether the parameter is MAC or not, then
mac_provided will be set to 1, which will disable the mac_seed setting
when using sub parameters
This patch will check opt and fix the above issue.
Tracked-On: #6690
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Signed-off-by: Chenli Wei <chenli.wei@linux.intel.com>
This API depends on GET_PLATFORM_INFO ioctl command,
which will not be supported from ACRN HSM anymore.
This patch removes the definition of it and
also cleans related data structure.
Tracked-On: #6690
Signed-off-by: Yonghua Huang <yonghua.huang@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
GET_PLATFORM_INFO IOCTL syscall will not be supported
from ACRN HSM driver, while vm_get_config() API depends
on it hence shall be removed.
Without vm_get_config(), vm_get_cpu_affinity_dm() is used
to get guest CPU bitmask.
Tracked-On: #6690
Signed-off-by: Yonghua Huang <yonghua.huang@intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Cpu affinty was set by pcpu id which can't be obtained
explictly by user. Use lapic id instead which can be easily
read from `/proc/cpuinfo` as `apicid`.
Tracked-On: #6690
Signed-off-by: Yuanyuan Zhao <yuanyuan.zhao@linux.intel.com>
Acked-by: Wang Yu1 <yu1.wang@intel.com>
The GET_PLATFORM_INFO will be removed from hypervisor. The acrn-dm can
only refer to Service VM's resources from now, all the resources out of
Service VM are not awared by acrn-dm. The original info got from
GET_PLATFORM_INFO needs to changed to Service VM's perspective like lapic id.
The pcpu_id is the index of lapic instance in MADT table. This patch
parses the Service VM's MADT table to convert the pcpu_id to lapic_id instead
of GET_PLATFORM_INFO.
Tracked-On: #6690
Signed-off-by: Yuanyuan Zhao <yuanyuan.zhao@linux.intel.com>
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Now the acpidev_pt module only use the hid to check the device,it can't
work well if there are more then one instance.
So this patch add UID to identify same type device to fix these issue.
Tracked-On: #6690
Acked-by: Wang, Yu1 <yu1.wang@intel.com>
Signed-off-by: Chenli Wei <chenli.wei@linux.intel.com>
Zhao Yakui