We inject invalid opcode if instruction decode fails.
We don't support many instruction. If new type guest hit
the invalid opcode and it's necessary to emulate that
instruction, we could add new instruction then.
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
TO inject the
- Invalid Opcode exception
- Stack Fault exception
- Alignment Check exception
to guest.
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
The DPL from SS access right field is always correct according
to SDM. We use it instead of using CS selector.
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
When doing copy_to/from_gva, it's possible the guest no page
happens on none-first page. In this case, we need get correct
fault address from gva2gpa.
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Acked-by: Eddie Dong <Eddie.dong@intel.com>
It is necessary to notify the VHM and hypervisor on the completion of a VHM
request even when the UOS is in suspend or system reset mode because the VHM and
hypervisor rely on the notification to reset their own states on the
request.
Currently the VHM request state is checked against REQ_STATE_PROCESSING instead
of REQ_STATE_COMPLETE when handling system reset or suspend/resume, leading to a
completed request unnotified, and causing the HV to complain on an occupied VHM
request when it raises a new one.
This patch fixes this issue by properly notifying completed requests to the VHM
& hypervisor. Some concerns are raised during a discussion on the potential
races which does not hurt for now but may in the future. These considerations
and potential solutions are documented as comments for future reference.
Tracked-On: #895
Signed-off-by: Junjie Mao <junjie.mao@intel.com>
Acked-by: Yin Fengwei <fengwei.yin@intel.com>
there is no git repo, which download from github release,
however, acrn-dm -v will get tag info from this repo, in
such case, the tag info was null.
this patch will fix nul tag, which get it from CL mock build.
Tracked-On: #676
Signed-off-by: Wei Liu <weix.w.liu@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
This patch add some comments after the default and before the break
in the switch statement based on MISRA-C requirement.
Signed-off-by: Shiqing Gao <shiqing.gao@intel.com>
Reviewed-by: Junjie Mao <junjie.mao@intel.com>
some exceptions can happen if westnon or network name service
not started before launch uos guest, like gvt and network exceptions,
which can cause SOS reboot. so add the dependency.
add data partiton check in launch script, for it can be called in every
clearlinux platform.
Signed-off-by: Minggui Cao <minggui.cao@intel.com>
Reviewed-by: Binbin Wu <binbin.wu@intel.com>
Added prepare_vm function as a wrapper function on top of prepare_vm0.
This makes adding support for partition mode ACRN to boot multiple
VMs from HV cleaner.
Signed-off-by: Sainath Grandhi <sainath.grandhi@intel.com>
This is the counterpart in DM to the VHM request state update in the
hypervisor. Major changes include:
* Remove accesses to the obsolete 'valid' member.
* Access the 'processed' member using atomic operations.
* Sync the documentation on vhm_request.
In addition, the new state transition also requires a VHM request to be always
handled properly, as there is no 'FAILED' state any more. Instead of crashing
the device model (and thus the UOS as well), the device model should return all
1s or ignore the request when it is to load from or store to an invalid address,
respectively.
Note: there is an issue in vm_system_reset() and vm_suspend_resume() where
completed VHM requests are not properly notified, causing the hypervisor to
complain as it sees uncompleted requests while trying to create a new one. This
issue will be resolved in a separate patch.
v1 -> v2:
* Use macro-defined constants for the default values for invalid PIO/MMIO
reads.
* Change the return type of vmexit_handler_t in DM to void as the return
values are no longer necessary.
* Remove VM_EXITCODE that are no longer used.
Tracked-On: #875
Signed-off-by: Junjie Mao <junjie.mao@intel.com>
Acked-by: Yu Wang <yu1.wang@intel.com>
This patch introduces wrappers to the built-in atomic operations provided by
gcc.
There are two sets of built-in atomic functions available. Before gcc 4.7.0 only
the __sync built-in functions are available, while since 4.7.0 a new set of
built-in functions with the __atomic prefix is introduced as a
replacement. Since the __sync functions will eventually be deprecated, the
__atomic ones are preferred whenever available.
The interfaces provided are listed below, mostly following the naming of the
underlying built-in functions which explain themselves.
atomic_load
atomic_store
atomic_xchg
atomic_cmpxchg
atomic_add_fetch
atomic_sub_fetch
atomic_and_fetch
atomic_xor_fetch
atomic_or_fetch
atomic_nand_fetch
atomic_fetch_add
atomic_fetch_sub
atomic_fetch_and
atomic_fetch_xor
atomic_fetch_or
atomic_fetch_nand
atomic_test_and_set
atomic_clear
atomic_thread_fence
atomic_signal_fence
Tracked-On: #875
Signed-off-by: Junjie Mao <junjie.mao@intel.com>
Acked-by: Yu Wang <yu1.wang@intel.com>
MISRA-C requirement:
The value of an expression shall be the same under any order of
evaluation.
The order in which side effects take place is unspecified and may lead
to unexpected results.
This patch add a temporary variable for temporary storage to avoid the
side effects of evaluation order.
Signed-off-by: Shiqing Gao <shiqing.gao@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
add wrap function cpu_do_idle in default_idle to handle arch cpu specific
idle operation.
Signed-off-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
vm_update_register calls vm_get/set_register to update register and vm_update_rflags
calls vm_update_register to update RFLAGS.
We have make vm_get/set_register as non-failed function in previous patch.
So, this patch make the vm_update_register/rflags as void.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Reviewed-by: Eddie Dong <eddie.dong@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
vie_read/write_bytereg call vm_get/set_register to get/set byteregs.
We have make vm_get/set_register as non-failed function in previous patch.
So, this patch make the vie_read/write_bytereg as non-failed function too.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Originally, vm_set/get_register return -EINVAL when "vcpu == NULL" or reg is invalid.
But, we don't check the return value actually and there is no chance we get an
null-vcpu and invalid reg in current implementation.
This patch add pre-assumptions about valid parameters before the function and make
them as non-failed functions.
- static uint64_t vm_get_register(struct vcpu *vcpu, enum cpu_reg_name reg)
- static void vm_set_register(struct vcpu *vcpu, enum cpu_reg_name reg, uint64_t val)
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Originally, vm_get_seg_desc return -EINVAL when "vcpu == NULL" or seg is invalid.
But, we don't check the return value actually and there is no chance we get an
null-vcpu and invalid seg in current implementation.
This patch adds pre-assumptions and makes the function as void.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
This patch just removes some dead codes related to Instruction Emulation.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Originally, there is cross-references between instr_emul.h and instr_emul_wrapper.h.
User must include both of them when calling instruction emulation functions. This
will raise up some confusion and inconvenience.
So we rearrange the logic of instruction emulation code as following:
- External API -- defined in instr_emul.h
* decode_instruction(struct vcpu *vcpu)
* emulate_instruction(struct vcpu *vcpu)
- Make all other functions as static in instr_emul.c
- Remove instr_emul_wrapper.c/h
No functional change.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Reviewed-by: Eddie Dong <eddie.dong@intel.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
ASSERT is too strict for HV when error happens during emulating instruction.
This patch remove all ASSERT and return a negative error code when failing to
emulate instruction.
Originally, getcc will return -EINVAL when opsize are not one of (1, 2, 4, 8).
But theoretically, opsize in current implementation can only be one of (1, 2, 4, 8).
So, we will always get valid "cc".
This patch add a pre-assumption and make sure that getcc always return valid value.
For the current code, #GP will be injected to guest if something goes wrong with
instruction emulation.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
There are so many __unused prefix spaning the emulate_xxx functions. This patch
unify the parameters of emulate_xxx to avoid it.
- All emulate_xxx functions are defined as emulate_xxx(struct vcpu *vcpu, instr_emul_vie *vie)
or emulate_xxx(struct vcpu *vcpu, instr_emul_vie *vie, struct vm_guest_paging *paging).
- Move mmio_read/write to instr_emul.c and call them directly.
Signed-off-by: Kaige Fu <kaige.fu@intel.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
This patch resolves one arcn-dm crash issue when rebooting UOS.
The rootcause is that uart releases unopened backend tty.
One reproduced case is that the board does not support IOC but IOC
feature is enabled in the acrn-dm. After rebooting UOS, crash will
happen.
NOTE: This issue is not related to IOC, it also can be reproduced
with NON-IOC scenario. Just set one invalid PTY to the lpc, then
this issue should be reproduced.
We need re-visit the whole policy for such scenario in future.
Signed-off-by: Yuan Liu <yuan1.liu@intel.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
Acked-by: Yu Wang <yu1.wang@intel.com>
The formal solution has merged,revert this patch.
This reverts commit a9d04cc5ce.
Signed-off-by: Mingqiang Chi <mingqiang.chi@intel.com>
Reviewed-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
move most of fields from run_context into ext_context for world switch.
these fields do not need doing runtime save/restore during vm exit/entry.
v3:
- update cr0/cr4 registers switch method
v2:
- use struct name ext_context instead of saved_context
- updated according to previous v2 patch
Signed-off-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
there will be 3 types of vcpu runtime contexts:
- runtime contexts always saved/restored during VM exit/entry, which
include general registers rax/rcx/rdx/rbx/rbp/rsi/rdi/r8~r15, cr2 and
msr for spectre control (ia32_spec_ctrl)
- runtime contexts on-demand cached/updated during VM exit/entry, which
include frequently used registers rsp, rip, efer, rflags, cr0 and cr4
- runtime contexts always read/write from/to VMCS, which include left
registers not in above
this patch add get/set register APIs for vcpu runtime contexts, and unified
the save/restore method for them according to above description.
v3:
- update vcpu_get/set_cr0/4 as unified interface to get/set guest cr0/cr4,
use on-demand cache for reading, but always write to VMCS for writing.
v2:
- use reg_cached/reg_updated for on-demand runtime contexts
- always read/write cr3 from/to VMCS
Signed-off-by: Jason Chen CJ <jason.cj.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
There are potential value outside range of underlying type
in some assignment expressions. This violates Rule 10.3 or
Rule 10.4 of MISRA C:2012.
BTW, all operations shall be conducted in exactly the same
arithmetic (underlying) type, otherwise, there is a
value outside range violation.
Update related assignment expressions.
V1-->V2:
* Fix potential overflow in "pit_calibrate_tsc";
* Move PTDEV_INVALID_PIN definition before
get_entry_info since this MACRO is only used by
debug function.
Signed-off-by: Xiangyang Wu <xiangyang.wu@linux.intel.com>
Reviewed-by: Junjie Mao <junjie.mao@intel.com>
Since global vector table is being used on all pcpus, it's not necessary to request timer irq
at each cpu init. With this change, per_cpu timer nodes are removed, and only BSP registers
and unregisters timer irq.
Signed-off-by: Yan, Like <like.yan@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Fix the parameter name mismatch between API declaration and definition.
v2 -> v3:
* Fix two more violations which are missed in previous report.
shell_puts and console_write
v1 -> v2:
* Replace 'ret_desc' with 'desc'
Signed-off-by: Shiqing Gao <shiqing.gao@intel.com>
Reviewed-by: Junjie Mao <junjie.mao@intel.com>
Two improvements:
Firstly, disable EOI VMEXIT for all vectors by default. Only the level
sensitive irq need to set it due to need emulate send EOI to vioapic to
clear Remote IRR bit.
Secondly, to clear RVI(Requesting virtual interrupt) and SVI(Servicing
virtual interrupt) bits.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
In some special scenarios, the LAPIC somehow hasn't send EOI to IOAPIC
which cause the Remote IRR bit can't be clear. To clear it, some OSes
will use EOI Register to clear it for 0x20 version IOAPIC, otherwise
use switch Trigger Mode to Edge Sensitive to clear it.
This patch emulate this IOAPIC behavior to satisfy this requirement.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
The IOAPIC specification defined EOI register for 0x20 version. The
original vioapic code implemented this register but the version was
still export as 0x11.
Tried to set 0x20, the Linux kernel has't access this EOI register,
still rely on lapic to send eoi.
From Linux ioapic driver comments, it says that only send EOI via EOI
register when met IOAPIC hardware bug.
This patch removes all 0x20 IOAPIC code to reduce the code size.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
To make vioapic simpler, avoid lots of code to convert variable type
between uint8_t and uint32_t.
This patch changes all variable type of pin related variables to
uint32_t instead of original uint_8.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
During ioapic reset, some registers need to be set to the default value
which defined in ioapic spec.
So far, the vioapic function only be called by ioapic itself in
ioapic_init. And just invoked after calloc the vioapic object, so all
the content are already set to zero. But this vioapic_reset function be
exported as one API which maybe invoked by other scenarios in future.
So this patch resolves this potential issue.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
The APIC ID field should be bit 24:27 in IOAPIC Identification register.
The bits 28:31 are reserved bits which need to be avoid touched.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
For level sensitive trigger mode, the vioapic should not deliver extra
interrupt to vlapic if the previous one hasn't received EOI.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Merge multiple if to switch-case. And set 0xFFFFFFFFU as the default
value of undefined address.
And the IOREGSEL register only bits 7:0 are defined, so mask the other
bits for read operation.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
1, Remove vioapic_mmio_read/vioapic_mmio_write, and combine them to
vioapic_mmio_rw.
2, The vioapic_read/write are used for emulate the ioapic indirect
access logic. So change their name with vioapic_indirect as the prefix.
Signed-off-by: Yu Wang <yu1.wang@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
This patch changes vm resuming condition from VM_STARTED to VM_PAUSED
Signed-off-by: Yuan Liu <yuan1.liu@intel.com>
Reviewed-by: Yuhong Tao <yuhong.tao@intel.com>
Reviewed-by: Like Yan <like.yan@intel.com>
This patch resolves vm state mismatch between vm and acrnd
which causes vm resuming failure
Signed-off-by: Yuan Liu <yuan1.liu@intel.com>
Reviewed-by: Yuhong Tao <yuhong.tao@intel.com>
Reviewed-by: Like Yan <like.yan@intel.com>
This patch resolves acrnd does not notify vm stop state(suspend/shutdown) to
cbc_lifecycle after SOS enters suspend/shutdown, that causes cbc_lifecycle
cannot trigger SOS suspend/shutdown flow.
Signed-off-by: Yuan Liu <yuan1.liu@intel.com>
Reviewed-by: Like Yan <like.yan@intel.com>
The issue is "acrnctl list" command always returns error with "no vmname
provided" information.
The rootcause is that vmname is an invalid parameter, which does not contain
any vm information.
Signed-off-by: Yuan Liu <yuan1.liu@intel.com>
Reviewed-by: Yuhong Tao <yuhong.tao@intel.com>
Reviewed-by: Like Yan <like.yan@intel.com>
Remove structure vm_attr
Wrap two APIs alloc_vm_id and free_vm_id
Signed-off-by: Mingqiang Chi <mingqiang.chi@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
