- when init, cr0 & cr4 should read from VMCS
- when world switch, cr0/cr4 read shadow should also be save/restore
Signed-off-by: Jason Chen CJ <jason.cj.chen@intel.com>
Current code is missing to handle the following race condition:
Before:
sti
|
V
handle softirq with
a while loop
| <----- if a new softirq generated at this point,
V it will be missed to be handled.
cli
|
V
return with unclear softirq bitmap
After:
again:
sti
|
V
handle softirq with
a while loop
|
V
cli
|
V
check if a new softirq generated, if yes,
jump back to above again lable point,
and do a while loop again.
|
V
return with clear softirq bitmap
Signed-off-by: Zheng, Gen <gen.zheng@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Now, trampline code is used by both AP start and BSP s3 resume.
For s3 resume, ACPI (5.2.10 Firmware ACPI Control Structure (FACS).
table 5-37) defines the real mode address should be set to:
realmode address = CS(waking_vec >> 4): IP(wakeing_vec & 0x000F)
But not all bootloader (like ABL) follow ACPI definition about the
CS:IP setup before jump to trampline code for S3 resume.
To handle all these cases, a long jmp is issued at very beginning
of trampline code to fixup the CS:IP setup. After the fixup, the
CS is set to: (waking_vect >> 4), the IP is set to: (the_address_
of_next_instruction_of_long_jmp & 0xF). Which is aligned with ACPI
definition.
Another thing is that we can't calculate the fixup CS and IP value.
The reason is related with limitations of real mode (can't get
current ip address without stack). So we calculate the CS and IP
when preparing the trampline code.
Signed-off-by: Zheng Gen <gen.zheng@intel.com>
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Acked-by: Eddie Dong <Eddie.dong@intel.com>
Cleanup "cpu_secondary_xx" in the symbols/section/functions/variables
name in trampline code.
There is item left: the default C entry is Ap start c entry. Before
ACRN enter S3, the c entry will be updated to high level S3 C entry.
So s3 resume will go s3 resume path instead of AP startup path.
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Signed-off-by: Zheng Gen <gen.zheng@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Acked-by: Eddie Dong <Eddie.dong@intel.com>
We will reuse this part of code for:
- AP bootup
- BSP wakeup from S3
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Acked-by: Eddie Dong <Eddie.dong@intel.com>
- unmask MTRR from guest CPUID to enable MTRR
- MTRR virtualization can be disabled by commenting out CONFIG_MTRR_ENABLED
Signed-off-by: bliu11 <baohong.liu@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
-rdmsr: emulate all MTRR registers besides variable range MTRRs
-wrmsr: emulate all MTRR registers besides variable range MTRRs and MTRRCAP
Signed-off-by: bliu11 <baohong.liu@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
MTRRs have no effect on the memory type used for an access to GPA.
This patch updates the EPT MT field according to the memory type
selected by MTRRs, which combines with the PAT, PCD and PWT bits
from the guest paging structures to determine the effective memory type
Signed-off-by: bliu11 <baohong.liu@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Linux commit edfe63ec97ed ("x86/mtrr: Fix Xorg crashes in Qemu sessions")
disables PAT feature if MTRR is not enabled. This patch does partial
emulation of MTRR to prevent this from happening: enable fixed-range
MTRRs and disable virable range MTRRs
By default IA32_PAT MSR (SDM Vol3 11.12.4, Table 11-12) doesn't include
'WC' type. If MTRR is disabled from the guests, Linux doesn't allow
writing IA32_PAT MSR so WC type can't be enabled. This creates some
performance issues for certian applications that rely on WC memory type.
Implementation summary:
- Enable MTRR feature: MTRRdefType.E=1
- Enable fixed range MTRRs: MTRRCAP.fix=1, MTRRdefType.FE=1
- For simplicity, disable variable range MTRRs: MTRRCAP.vcnt=0.
It's expected that this bit is honored by the guests and they won't
change the guest memory type through variable MTRRs.
Signed-off-by: bliu11 <baohong.liu@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
- Add PAGING_REQUEST_TYPE_MODIFY_MT memory map request type
- Update map_mem_region() to allow modifying the memory type related
fields in a page table entry
- Add ept_update_mt()
- add modify_mem_mt() for both EPT and MMU
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
In current code, on uefi platform, vmcs will be overwritten according to
uefi context, using the exec_vmwrite directly.
This patch use vmx_write_cr<#> interface to init control registers.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
- Ajust the order of control registers just for more clear in logic.
Will not change the guest init state after the ajustment.
- Add a comment to point out the CR4 should be inited before CR0.
The value of CR4 will be used during CR0 set.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
- export start_cpus to start/online APs.
- Add stop_cpus to offline APs.
- Update cpu_dead to decrement running cpus number and do cleanup
for AP down
Signed-off-by: Yin Fegnwei <fengwei.yin@intel.com>
Acked-by: Eddie Dong <Eddie.dong@intel.com>
To handle cpu down/up dynamically, arcn needs to support vmx off/on
dynamically. Following changes is introduced:
vmx_off will be used when down AP. It does:
- vmclear the mapped vcpu
- off vmx.
exec_vmxon_instr is updated to handle start and up AP both. It does
- if vmx was on on AP, load the vmxon_region saved. Otherwise,
allocate vmxon_region.
- if there is mapped vcpu, vmptrld mapped vcpu.
Signed-off-by: Zheng Gen <gen.zheng@intel.com>
Signed-off-by: Yin Fegnwei <fengwei.yin@intel.com>
Acked-by: Eddie Dong <Eddie.dong@intel.com>
Per performance consideration, we don't flush vcpu context when doing
vcpu swithing (because it's only swithing between vcpu and idle).
But when enter S3, we need to call vmclear against all vcpus attached
to APs. We need to know which vcpu is attached with which pcpu.
This patch introduced API to get vcpu mapped to specific pcpu.
Signed-off-by: Yin Fegnwei <fengwei.yin@intel.com>
Acked-by: Eddie Dong <Eddie.dong@intel.com>
current_vcpu is not correct when there are multi vcpus in one VM,
using it is in-correct, so remove it.
Signed-off-by: Jason Chen CJ <jason.cj.chen@intel.com>
Retrieve dseed from SeedList HOB(Hand-Off-Block).
SBL passes SeedList HOB to ACRN by MBI modules.
Signed-off-by: Qi Yadong <yadong.qi@intel.com>
Reviewed-by: Zhu Bing <bing.zhu@intel.com>
Reviewed-by: Wang Kai <kai.z.wang@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
keep the global variables access exclusive in vcpu pause & resume.
Signed-]off-by: Minggui Cao <minggui.cao@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
The current implementation of per_cpu relies on several non-c99 features,
and in additional involves arbitrary pointer arithmetic which is not MIS-
RA C friendly.
This patch introduces struct per_cpu_region which holds all the per_cpu
variables. Allocation of per_cpu data regions and access to per_cpu vari-
ables are greatly simplified, at the cost of making all per_cpu varaibl-
es accessible in files.
Signed-off-by: Huihuang Shi <huihuang.shi@intel.com>
Enable VMX vpid ctrl and assign an unique vpid to each vcpu
so that VMX transitions are not required to invalidate any
linear mappings or combined mappings.
SDM Vol 3 - 28.3.3.3
If EPT is in use, the logical processor associates all mappings
it creates with the value of bits 51:12 of current EPTP.
If a VMM uses different EPTP values for different guests, it may
use the same VPID for those guests. Doing so cannot result in one
guest using translations that pertain to the other.
In our UOS, the trusty world and normal world are using different
EPTP. So we can use the same VPID for it.
Signed-off-by: Li, Fei1 <fei1.li@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
We need know which tlb to flush: ept or vpid.
1. error handle for invept.
it's the same with invvpid error handle.
change its name to compatible with vpid.
2. the macro name for flush ept tlb request.
Signed-off-by: Li, Fei1 <fei1.li@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Copy CR0 / CR4 value from normal world context to secure world context
when init secure world.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Acked-by: Wang Kai <kai.z.wang@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
In current code, sos/uos bsp can only start from 64bit mode.
For sbl platform:
This patch start sos bsp from protected mode by default.
CONFIG_START_VM0_BSP_64BIT is defined to allow start sos bsp
from 64bit mode. If a config CONFIG_START_VM0_BSP_64BIT
defined in config file, then sos bsp will start from 64bit mode.
This patch start uos bsp from real mode, which needs the integration
of virtual bootloader (vsbl).
For uefi platform:
This patch sets sos bsp vcpu mode according to the uefi context.
This patch starts uos bsp from protected mode, because vsbl is not ready
to publish for uefi platform yet. After vsbl is ready, can change to
start uos bsp from real mode.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Reviewed-by: Eddie Dong <eddie.dong@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Acked-by: Xu, Anthony <anthony.xu@intel.com>
In current implementation, a vcpu can only start from real mode
or 64bit mode.
This patch adds support to start a vcpu from protected mode.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Reviewed-by: Eddie Dong <eddie.dong@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Acked-by: Xu, Anthony <anthony.xu@intel.com>
In current implementation, on sbl platform, vm0 bsp
starts from 64bit mode. And hv need to prepare init
page table for it.
In this patch series, on sbl platform, vm0 bsp starts
from non-paging protected mode.
This patch prepares an init gdt for vm0 bsp on sbl
platform.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Reviewed-by: Eddie Dong <eddie.dong@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Acked-by: Xu, Anthony <anthony.xu@intel.com>
Translate gva2gpa in different paging modes.
Change the definition of gva2gpa.
- return value for error status
- Add a parameter for error code when paging fault.
Change the definition of vm_gva2gpa.
- return value for error status
- Add a parameter for error code when paing fault.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Xu, Anthony <anthony.xu@intel.com>
Use # of paging level to identify paging mode
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Eddie Dong <eddie.dong@intel.com>
Acked-by: Xu, Anthony <anthony.xu@intel.com>
In current implemenation, cr0/cr4 host mask value are set
according to the value from fixed0/fixed1 values of cr0/cr4.
In fact, host mask can be set to the bits, which need to be trapped.
This patch, add code to support exiting long mode in CR0 write handling.
Add some check when modify CR0/CR4.
- CR0_PG, CR0_PE, CR0_WP, CR0_NE are trapped for CR0.
PG, PE are trapped to track vcpu mode switch.
WP is trapped for info of protection when paing walk.
NE is always on bit.
- CR4_PSE, CR4_PAE, CR4_VMXE are trapped for CR4.
PSE, PAE are trapped to track paging mode.
VMXE is always on bit.
- Reserved bits and always off bits are not allow to be set by guest.
If guest try to set these bits when vmexit, a #GP will be injected.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Xu, Anthony <anthony.xu@intel.com>
_Static_assert is supported in C11 standard.
Please see N1570(C11 mannual) 6.4.1.
replace _Static_assert with ASSERT.
Signed-off-by: huihuang shi <huihuang.shi@intel.com>
- 'dispatch' will be null and be referenced if
'basic_exit_reason < ARRAY_SIZE(dispatch_table)'
if false.
Signed-off-by: Yonghua Huang <yonghua.huang@intel.com>
The following massages is not fatal error but should be print to serial and
sbuf at the same time. pr_fatal is not a good choice. pr_acrnlog is designed
to deal with the situation. So replace the following misused pr_fatal with
pr_acrnlog.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Hypervisor shouldn't deal with errno return by hypercall.
The SOS should do this. Here just return the errno to SOS.
Signed-off-by: Li, Fei1 <fei1.li@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
vm_state_info in struct vm_arch is not used, remove it
Signed-off-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Xu, Anthony <anthony.xu@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Guest CR3 read/write operations are not trapped.
Remove CR3 handling in cr_access_vmexit_handler.
Also remove unused API vmx_read_cr3.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Move from vmexit.c to vmx.c
Declare the functions in vmx.h
Rename the functions' name with prefix vmx_.
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
There are some massages which is not fatal error but should to print
to serial and sbuf(hvlog) at the same time. pr_fatal is for fatal error
massages and it is not good choice for the situation above.
Introduce a new API pr_acrnlog to deal with the situation. And replace the
following printf with pr_acrnlog for massages should be print to sbuf and
serial. Then developers can get those massages on serial and BTM(Boot Time
Measurement) can use acrnlog to get those massages from sbuf.
BTM refers to Boot Time Measurement which will read acrnlog file to get
timestamps of steps we want.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
When we create an UOS, we didn't indicate the vmid.
Thus we can't get the vm description for the vm
description array.
Instead we use a temporary vm description to save data to
fill the vm structure when crate an UOS. It's uselesss once
UOS has created. So we don't need to maintain vm description
array here for UOS.
Signed-off-by: Li, Fei1 <fei1.li@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Don't panic system when execute vmxon instruction failed.
And let's follow that only print error info when error return
from library function.
Signed-off-by: Li, Fei1 <fei1.li@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
We cleanup ASSERT. This serial try to only panic when create
SOS failed.
Signed-off-by: Li, Fei1 <fei1.li@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
