CPUM,IEM,HM: Enabled XSAVE and AVX for AMD-V too.

VMM/HMVMXR0: nit.

VMM: renamed HM_[DISABLE|RESTORE]_PREEMPT_IF_NEEDED() macro.

HM,CPUM,IEM: XSETBV fixes and adjustments.

HM: Save/Load/Restore XCR0 handling during world switching. Implemented XSETBV for VT-x.

HM/VMX: Added new exit codes.

HMR0VMX,EM,IEM: Moved the CRx related interpretation methods from EM to IEM (VT-x only) to avoid tinkering with the EMAll.cpp stuff any more.

VMM/GIM: Allow dynamic enabling of #UD traps and per-VCPU hypercalls.

VMM: GIM raw-mode support.

VMM/GIM: Add hypercall support and KVM spinlock support.

VMM: Rename variable to better reflect its purpose.

VMM: Clear idHostCpu mapping while deregistering the context-hooks rather than in HMR0Leave(). Simplified HMR0_[DISABLE|RESTORE]_PREEMPT_IF_NEEDED() and made VMMR0ThreadCtxHooksDeregister() check before deregistering the hook rather than asking the caller to do it each time.

SUPDrv/VMMR0: properly handle the CR4 shadow register on Linux >= 4.0

VMM/HMVMXR0: Fix debug assertion on 32-bit hosts. Host-state is already saved in HMR0Enter() while entering from ring-3/longjmp.

VMM,SUP: Apply the tsc delta where it matters. Made sense out of the paravirt-tsc-mode enable/disable code.

VMM/HM: stat adjustments.

VMM/HMVMXR0: Fix host hang if VMX-preemption timer is disabled.

VMM/HMVMXR0: comment update.

VMM: doxygen nits.

VMM/HMVMXR0: spaces.

VMM/HMVMXR0: nit.

VMM/HMVMXR0: Try fix regression caused by r98028 (single-stepping in dbx)

VMM/HMVMXR0: Fixed unintentional losing #DB intercepts while loading guest CR0. Also removed #MF from the real-on-v86 mode exception mask, as it too like #NM is handled separately. Improved some comments.

VMM: Implemented TM TSC-mode switching with paravirtualized guests.

VMM/HMVMXR0: Fixed incorrect setting of the block NMI force-flag.

vmexit trace points.

no else after break

HMVMXR0.cpp: Assert that EFLAGS are valid.

HMVMXR0.cpp: hmR0VmxCheckExitDueToEventDelivery returns error codes in at least three places, so the user need to check for more than just VINF_EM_RESET and VINF_EM_DOUBLE_FAULT. Also, check for positive result and then do return code massaging in RT_UNLIKELY branch, instead of multiple negative results checks.

HMVMXR0.cpp: no else after return.

HMVMXR0.cpp: Use assertions better.

VMM: Fix sign inconsistency in the RealUseTSC case.

VMM: Fixed incorrect pure TSC-offsetting mode switch with paravirt. VMs when the host/TM cannot do offsetting.

VMM/HMVMXR0: comestics.

VMM/HMVMXR0: Windows build fix, doesn't like negating unsigned numbers with unary.

VMM/HMVMXR0: Remove assertion in hmR0VmxExitRdtsc() for when CR4.TSD is set and we most probably should #GP.

VMM/HMVMXR0: Fix incorrect sign and calculation on guest TSC offset in VMCS.

VMM/HMVMXR0: nits.

VMM/HMVMXR0: nits.

HMVMXR0.cpp: Fixed several bugs in the real-mode-on-top-of-v86 instruction emulation: STI always set the inhibit interrupts. IRET used didn't keep enough flags in the high word, but incorrectly (I believe) kept the resume flags (RF). The RF flag was only cleared in by a few instructions, it should normally be cleared when an instruction is retired (with a couple of exception). Made the real-mode direct event injection code return VINF_EM_DBG_STEPPED when stepping (currently passing fStepping around as that clearer).

HMVMXR0.cpp: Fixed bug in hmR0VmxInjectEventVmcs where cs.ValidSel wasn't updated in the real-mode case. Fixed a few single stepping issues. Some cleanups like CS:RIP formatting, unnecessary {} and else after return.

VMM/HMVMXR0: logging fix.

VMM/HMVMXR0: Fix checking PAT MSR validity in hmR0VmxCheckGuestState().

VMM/HMVMXR0,HMSVMR0: oops, forgot bugref number.

VMM/HMVMXR0,VMM/HMSVMR0: Fix incorrect disassembly of instruction caused by a #PF caused due to an ext. int/NMI while dealing with shadow #PFs. Fixes the warcraft 2 testcase.

VMM/HMVMXR0: Added ApicAccessToR3 stat and addressed a 4.3 todo.

VMM/HMVMXR0: Logging.

VMM/HMVMXR0: Fix a wrong invalid-guest-state check with regards to EFER.LMA and LME.

VMM/HMVMXR0: Use unordered atomics for VMMRZCallRing3[Enable|Disable]() and some VT-x nits.

VMM/HMVMXR0: Fix strict assertion on older CPUs; check for secondary processor-based execution controls before reading it off the VMCS.

VMM/HMVMXR0: Renaming nits.

HostDrivers/Support, VMM: support CONFIG_PAX_KERNEXEC Linux kernels

VMM/HMVMXR0: Wrong, EMT preemption doesn't lose VMCS state.

VMM/HMVMXR0: Clear inhibit-interrupts force-flag when only NMI blocking is in effect. Don't start VM with NMI-window exiting control set.

VMM/HMVMXR0: Move comment to the right place.

VMM/HMVMXR0: Cleanup.

VMM/HMVMXR0: Nits.

VMM/HMVMXR0: Make hmR0VmxGetGuestIntrState() side-effect free in terms of force-flags.

VMM/HMVMXR0: comment.

VMM/HMVMXR0: Don't lose interruptibility-state when we get preempted early!

VMM/HMVMXR0: Less confusing variable name.

VMM/HMVMXR0: Fix NMI redelivery when the original delivery caused an exception VM-exit.

VMM/HM: Add tracking of last 30 VM-exits. Can be useful for guru situations even on release builds. Should not add much overhead to each exit.

VMM/HMVMXR0: Comment.

VMM/HMVMXR0: Fix NMI injection.

VMM/HMVMXR0: Convert NMIs to TRPM as TRPM_TRAP and not TRPM_HARDWARE_INT.

VMM: Fixed potential bug in AMD-V NMI injection when interrupt shadowing is in effect, renamed INHIBIT_NMIS to BLOCK_NMIS to match Intel specs.

VMM/HMVMXR0: Fix NMI re-injection when NMI injection caused a faulting VM-exit and blocking NMIs until IRET completes if the IRET caused a fault.

Grammar.

VMM/HM: nits.

VMM/HMVMXR0: nit.

VMM/HMVMXR0: Improve SMP logging a bit.

VMM: Fix NMI handling.

Comments and nits.

VMM: VT-x and AMD-V support for making GIM hypercalls.

VMM/GIM, VMM/HM: Add support for Hyper-V hypervisor capabilities reporting through CPUID. Updated saved states as well, will break old states.

VMM/GIM: Fix Hyper-V TSC offset scaling to 100ns units.

VMM/HMVMXR0: Trial; don't adjust TSC offset with paravirt TSC.

VMM: Change error codes to be consistent.

VMM/HMVMXR0: Temporary instrumentation for event injection issue.

VMM/HMVMXR0: Load guest-state after injection of events; simplifies state-handling a tiny bit.

VMM/HMVMXR0: Nit.

VMM/HMVMXR0: Use VMCS offsetting until we fix TM, makes Hyper-V guests more stable.

VMM/GIM: More bits for Hyper-V implementation.

VMM/HMVMXR0: Relax EFER consistency/corruption checks for the fixes done in r93674.

VMM/HMVMXR0: Intel likes hyphenation.

VMM/HMVMXR0: Intel likes hyphenation.

VMM/HMVMXR0: Fixed invalid EFER swapping when guest is using PAE paging (when VT-x UX is not in effect). Also fixes swapping EFER for 32-bit PAE hosts and 32-bit PAE guests when EFER.NXE differs when guest is using PAE paging.

VMM/HMVMXR0,HMSVMR0: debug build fixes.

VMM/HMVMXR0: Do EFER swapping using auto-load/store area in the VMCS when required.

VMM/HMVMXR0: Implemented EFER swapping using VMCS controls.

VMM/IEM: Implemented hardware task-switches, code path disabled.

VMM/HMVMXR0: Assertion is a little too strict (global vs local init).

VMM/HMVMXR0: Another assertion to try catch uCurrentAsid mismatch.

VMM/HMVMXR0: STAM counter typo.

VMM/HMVMXR0: Also intercept #TS exceptions while debugging all exception intercepts, logging.

VMM/HM: Fix some selector limit checks, comments.

VMM/HMVMXR0: nit.

VMM/HMVMXR0: Added hmR0VmxSetPendingXcptGP() and some doxygen fixes.

VMM: Assert 4K alignment on global VT-x and AMD-V physical pages. Change some release assertions to debug ones.

VMM/HMVMXR0: When only all-context VPID flush is supported by the CPU, avoid reflushing -all- VPIDs for every run after hitting VPID limit. It is sufficient to flush all VPIDs once per host CPU when the limit is reached.

VMM/HMVMXR0: undo temporary change from r92738.

VMM/HMVMXR0: Temporary enable of HMVMX_ALWAYS_FLUSH_TLB for a test build.

VMM/HMVMXR0: Assert.

VMM/HMVMXR0: More info. on assert failures.

VMM/HMVMXR0: More assertions in TLB flushing.

VMM/HMVMXR0: First entry into VT-x may already have saved the host-context, skip that case for the assertion checking for guest mode changes..

VMM/HMVMXR0: Avoid one extra VM-exit while injecting interrupts if interrupt-window exiting was previously there but exited for other reasons.

VMM/HMVMXR0: Undo r92473. We -do- need it, better comments.

VMM/HMVMXR0: This should not be necessary, while in preemption hook we're already non-preemptible and no-long-jump guaranteed.

VMM/HMVMXR0: Build fix.

VMM/HMVMXR0: Fix for workaround done in r92215.

HMR0VMX.cpp: Hacked hmR0VmxSaveHostSegmentRegs to call VMXRestoreHostState when necessary so we don't corrupt the host state should HM_CHANGED_HOST_CONTEXT be set and executed without leaving the HM ring-0 context. Also restricted the dont-bother-to-restore-host-Idtr.limit condition to only include Linux and Solaris as this is known to cause trouble on Windows and Darwin already (HMR0Mixed.mac needs tweaking and testing to complete this change.)

VMM/HMVMXR0: Fix single-stepping over IO instructions (string I/O is a todo).

VMM/HMVMXR0: Fix single-stepping over IRET and POPF in real-on-v86 mode.

VMM/HMVMXR0: Fix single-stepping in real-on-v86 mode for certain emulated instructions. Added todo for the rest.

VMM/HMVMXR0: Comment clarification.

VMM/HMVMXR0: Fix single-stepping with IRET on real-on-v86 mode while not breaking single-stepping of emulated instructions.

VMM/HMVMXR0: comment copy & paste error.

VMM/HMVMXR0: Todo comment.

VMM/HMVMXR0: nit.

MSR rewrite: initial hacking - half disabled.

VMM/HMVMXR0, HMSVMR0: Comments, be consistent.

VMM/HMVMXR0: Comments.

VMM/HMVMXR0: Four consecutive NOT operators is a tad too redundant.

VMM/HMVMXR0: Minor nit while checking PAE PDPTEs in hmR0VmxCheckGuestState().

VMM/HMVMXR0: Save guest-control registers on CRX-read exits.

VMM/HMVMXR0: Add guest-state checks for PAE PDPTEs.

VMM/HMVMXR0: Comment.

More granular assertions.

VMM/HMVMXR0: Handle #DB from ICEBP instruction which is undocumented by Intel. This causes the undocumented "privileged software exception" interruption-type exception exit.

VMM/HMVMXR0: comment nit.

VMM/HMVMXR0: Safer assumptions while updating guest-state on fully preemptible kernels.

VMM/HMVMXR0: Drop 64-bit MSR optimizations on hybrid darwin kernels.

VMM/HMVMXR0: Hybrid fixes.

VMM/HMVMXR0: bugref Comment, spaces.

VMM/HM: VMCPU_HMCF -> HMCPU_CF macro renaming.

VMM/HMVMXR0: Now that we have atomic flag updates, reenable WRMSR optimization while loading guest MSR.

HMVMX: Do not fall back on FERR emulation (see #6117).

VMM/HMVMXR0: Comment, macros.

VMM/HMVMXR0: Simplified the MSR optimization code.

VMM/HMVMXR0: Build fix.

VMM/HMVMXR0: Update guest TSC_AUX before auto-load/store.

VMM/HMVMXR0: MSR auto-load/store fix for hybrid darwin.

VMM/HMVMXR0: Assertions.

VMM/HMVMXR0: Comment.

VMM/HMVMXR0: Comments.

VMM/HMVMXR0: Added a couple of missing Unsupported-Feature Combos errors.

VMM/HMVMXR0: Comment.

VMM/HMVMXR0: Fix darwin 32-bit/PAE kernel case.

VMM/HMVMXR0: Careful with preemption!

VMM/HMVMXR0: MSR optimizations for world-switch of 64-bit guests.

VMM/HMVMXR0: Better fix for r90778.

VMM/HMVMXR0: Don't cause unnecessary host-state update in case we're not swapping MSRs.

VMM/HMVMXR0: Build fix.

VMM: Retire VBOX_WITH_AUTO_MSR_LOAD_RESTORE define.

VMM/HMVMXR0: spaces.

VMM: MSR auto-load/store handling rewrite for VT-x.

VMM/HMVMXR0: Fixed single-stepping in the guest to generate #DB correctly.

VMM/HMVMXR0: Comment.

VMM/HMVMXR0: Fix occasional 32-bit Haiku guest hang during boot.

VMM/HMVMXR0, HMSVMR0: Use IPRT.

VMM/HMVMXR0: Fixed DPC latency issue on Windows hosts.

VMM/HMVMXR0: Fix for MOV DRx intercepts on 32-bit hosts with 64-bit guests.

VMM/HMVMXR0: nits.

VMM/HMVMXR0: Don't forget to load guest debug state when not intercepting #DB and Mov DRx.

VMM/HMVMXR0: Macro naming fix and use RT_BOOL.

VMM/HMVMXR0, HMSVMR0: Guest-FPU pre-load optimization.

VMM/HMVMXR0: Incorrect assertion for SMP.

VMM/HMVMXR0: Fix incorrect masking of TPR bits with TPR-shadowing.

VMM/HMVMXR0: More paranoid assertions cannot hurt.

VMM/HMVMXR0: Introduced HMVMX_ALWAYS_FLUSH_TLB debug switch.

VMM/HMVMXR0: INS/OUTS with REP prefix update RFLAGS. Fix incorrect assertion in hmR0VmxCheckGuestState().

VMM/HMVMXR0: Adding functions for dynamic MSR swapping.

VMM/HMVMXR0: Fix MSR_TSC_AUX, restore host MSR.

VMM/HMVMXR0: Fix MSR_TSC_AUX, use the right check to be doubly sure.

VMM/HMVMXR0: Fix MSR_TSC_AUX auto-load/store when not using TSC offsetted mode when not compiling with VBOX_WITH_AUTO_MSR_LOAD_RESTORE.

VMM/HMVMXR0: Fix MSR_TSC_AUX auto-load/store when not using TSC offsetted mode.

HMR0VMX.cpp: Clear the TF bit in the single step loop, instead on later in the common exit-to-ring3 code.

VMM/HMVMXR0: Partially revert r89778. Back to disassembling and interpreting string IO instructions, doesn't yet work.

VMM: Some @thread doxygen corrections.

VMM: FPU cleanup.

VMM/HMVMXR0: Comment update.

VMM/HMVMXR0, HMSVMR0: Comment out unused code paths in TLB shootdown.

VMM/HMVMXR0: Use instruction info. for string IO VM-exits on CPUs that support it, rather than disassembling the instruction.

VMM: IntrInfo to IntInfo renaming.

VMM/HMVMXR0: Propagate all unexpected exits to release log, slight changes to error macro names.

VMM/HMVMXR0: Make sure host segments are present in the LDT if we're going to try restore it.

VMM: Whitespace cleanups by scm.

VMM/HMVMXR0: Don't assume CR8 load/store exiting is available on 32-bit CPUs. Only needed for 64-bit guests.

VMM/HMVMXR0, HMSVMR0: Alignment and assert.

VMM/HMVMXR0: Checking this once outside the loop is sufficient.

VMM/HMVMXR0: VMCS size is bits 44:32 not 43:32. Assert what the Intel spec. claims as well.

VMM/HMVMXR0: Fix switcher verification bits to be compiled to match what the switcher actually does.

VMM/HMVMXR0, HMSVMR0: Use HMCF macros.

VMM/VMMR0: Don't disable interrupts for a long time when disabling preemption is sufficient. Other minor fixes.

VMM/HMVMXR0, HMSVMR0: Restore hyper/guest DR7 in accordance with what we load.

VMM/HM: Added total VM-exits STAM.

VMM/HMVMXR0, HMSVMR0: Fixes for shared FPU/debug state handling with thread-context hooks.

HMVMX: Look at the right guest long mode indicator.

VMM/HMVMXR0: Debug macro switch HMVMX_ALWAYS_SWAP_FPU_STATE.

VMM/HMVMXR0: Fix assertions and stale comments.

VMM/HMVMXR0: Fix invalid assertion on 32-bit hosts.

VMM/HMVMXR0: Try restore state properly on assertions in longjmps contexts. Reevaluate FPU state after injecting #NM. Other cosmetics.

VMM/HMVMXR0: Fix check in hmR0VmxCheckGuestState(), CR0.PE MB1 unless it's UX.

Build fix.

VMM/HMVMXR0: Nits.

VMM/HMVMXR0: oops.

VMM/HMVMXR0: Check VMCS values rather than fudged pCtx values in hmR0VmxCheckGuestState().

VMM/HMVMXR0: Leave VT-x properly for VERR_VMX_INVALID_GUEST_STATE and VERR_VMX_INVALID_VMCS_PTR gurus.

VMM/HMVMXR0: Temporary workaround, intercept CR0.PE changes with UX.

VMM: Fixed strict build assertion assumption with thread-context hooks. More assertions, slight cleanup.

VMM/HMVMXR0: Clear interrupt-window exiting control on premature returns to ring-3 before injecting an event.

VMM/HMVMXR0: Attempt to fix guru in the switcher case with MOV DRx intercepts.

debug build fix.

VMM/HMVMXR0: Assertion.

VMM/HMVMXR0: Address possible deadlock situation with the PGM lock. Reenabled thread-context hooks again.

EM,HM: MWait fix, clearing the active bit too early.

VMM/HMVMXR0: Nits.

VMM/HMVMXR0: Clarify that it's guest-CPU context CR0 and not VMCS CR0.

VMM/HMVMXR0: comment cosmetic.

HMVMXR0.cpp: Made hmR0VmxSetupVMRunHandler trigger the necessary host (and guest) field updates, and moved it up so the HM_CHANGED_VMX_EXIT_CTLS/HM_CHANGED_VMX_ENTRY_CTLS would take affect before executing guest code again.

HMVMXR0.cpp: Only the 64 -> 32 needs the exit controls update, I believe.

VMM/HMVMXR0: No need to ask VMX exit ctls to be reloaded when we're going back to ring-3.

VMM: nits.

VMM: Nits.

VMM/HMVMXR0: Avoid calling PGM twice for reading just 4 consecutive bytes.

VMM/HM: Preempt stats and minor optimizations to VT-x when thread-context hooks are used.

VMM: More descriptive host-NMI stat name.

VMM/HMVMXR0: Undo injected event on premature returns to ring-3. Should fix longjmp disallowed ring-0 assertions on real-on-v86 event injection

Comments.

More paranoia regarding pfnStartVM changes on 32-bit hosts (doesn't help).

VMM/HMVMXR0: Nits.

VMM: Propagate errors properly while leaving HM context.

VMM: Ensure longjmps after hmR0[Vmx|Svm]LeaveSession() don't happen.

HMVMXR0.cpp: When the guest switches from long mode to one of the legacy modes, we may end up with the wrong host-resume values from what I can tell. Attempting a quick fix to see if that helps with the host tripple faults I'm having.

VMM/HMVMXR0: Hungarian.

VMM: Addressed a rare corner case stale TLB entry issue.

VMM/HM: Cleanup.

VMM/HMVMXR0: Cleanup.

VMM/HM: Naming fixes.

VMM/HMVMXR0: Make code more robust regarding longjmps in awkward places.

VMM/HMVMXR0: Even without preemption hooks, we can now load the guest-state before disabling interrupts. Better handling of triple faults conditions (don't leave with interrupts disabled).

HM: Adjusted some CPUMIsXxxDebugStateActive() assertion for 64-on-32-bit conditions.

VMM/HMVMXR0: Update comment and add assertion. We no longer have gaps between leaving and thread-context hook deregisteration.

VMM/HM: More dead code path elimination.

VMM/HMVMXR0: Safer longjmp coverage, only restore host state when altered by VT-x. VMM/HMSVMR0: Exit normally while getting errors during the world-switch.

VMM/HM: Eliminate dead code paths.

VMM: Ditch VBOX_WITH_VMMR0_DISABLE_PREEMPTION.

VMM: Enable thread-context hooks. Reimplemented event injection logic for VT-x and AMD-V.

VMM/HMVMXR0: Fix invalid VMCS launch-state when thread-context hooks are used.

VMM/HMVMXR0: Comments.

VMM/HM: Thread-context hooks, disable preemption while touching host state on VM-exits. AMD-V still todo.

VMM/HMVMXR0: Don't play safe when we have assertions to catch things.

VMM/HMVMXR0: Avoid loading the host state earlier when we can get preempted (when thread-context hooks are used).

VMM/HMVMXR0: Burn fix.

VMM/VMMR0, HMVMXR0: Minor fixes relating to error paths.

VMM/HM: Preemption hooks, work in progress. Hopefully I didn't break the non-hook case.

VMM/HMVMXR0: More log adjustments.

HMVMX: Log a bit more.

VMM/HMVMXR0: Asserted by the caller, don't re-assert here.

VMM/HMVMXR0: Additional logging to track VMCS state changes.

VMM/HM: Better error reporting for unsupported VT-x feature combos.

VMM/HMVMXR0: Try fix stale VMCS launch state while getting preempted.

VMM/VMMR0: Preemption hooks implemented and enabled for Solaris and Linux hosts.

VMM: X2APIC + NMI. Only tested on AMD64.

VMM/HM: Preemption hoooks, work in progress.

VMM/HM: Redundant updates of flags is not necessary.

Warnings and space.

VMM/HM: RunGuestCodeStep() Load should be Save, and some typos, cosmetics.

VMM/HM: One extra char. to make typing without autocomplete easier.

VMM/HMVMXR0: Hungarian, hex prefixes.

VMM/HM: Preemption hooks.

VMM/HM: Preemption hooks. Some common structural changes and cleanup, and initial imlementation of VT-x/AMD-V specific hook functionality.. Work in progress.

HMR0VMX.cpp: missing fSingleInstruction test.

adjusted my assertion.

VMM/HMVMXR0: Avoid OR'ing status codes when we AssertRCReturn check it afterwards.

HNVMXR0.cpp: Separate run-loop for single stepping to catch exceptions and such without harming performance of the normal execution (requires RIP and CS to be saved). Simplified the UNUSABLE handling in hmR0VmxReadSegmentReg to just make sure P and a few irrelevant other bits are cleared. Always clearing the G would cause the limit to change on reload. If having bits set here is causing trouble, we'll fix the other code making the trouble, not work around it here.

No MMIO2 aliasing when doing full IEM verification runs.

More single stepping work.

VMM/HMVMXR0: Added segment checks to hmR0VmxCheckGuestState() while using unrestricted execution.

build fix + adj.

adjustments.

VMM: I/O breakpoints.

VMM/HMVMXR0: Added missing TR guest-state checks.

HMSVMXR0.cpp: DR6.Bx values are cleared on #DB it seems, while the others aren't.

VMM/HMVMXR0: Some missed checks and fixes to segment checks.

HMVMXR0: Fixed copy&past from SVM code with fDr6=false in CPUMR0DebugStateMaybeSaveGuestAndRestoreHost call, causing DR6 to be lost upon ring-3 returns. I also think we must intercept #DB when doing lazy DRx loading, otherwise the #DB will update the host DR6 and the guess will be missing the debug event.

VMM: More debugging related stuff.

VMM/HMVMXR0, HMSVMR0: Fix typo in checking invalid guest state, nits.

build fix.

VMM: Debug register handling redo. (only partly tested on AMD-V so far.)

VMM/HMVMXR0: Fixed a bug in checking invalid guest-state logic.

VMM: Removed all VBOX_WITH_OLD_[VTX|AMDV]_CODE bits.

VMM/HMVMXR0: 130 columns.

VMM/HMVMXR0: spaces.

VMM/HMVMXR0: Typo.

VMM/HMVMXR0: Build fix.

VMM/HM: Implemented VMX guest-state checks in accordance with the Intel spec. Any errors founds now flow to the release log.

EM: Started on HM single stepping for IEM verification purposes. Trying to fix the HM debugging in the proccess. VT-x only atm.

VMM/HMVMXR0: Naming.

VMM: make the StatExitHostNmi release

VMM/VMMR0: HW(VMX|SVM)R0.h -> HM(VMX|SVM)R0.h. Now that the refactor is done.

Space.

VMM/HMVMXR0: Make save-DR7 function name more clear (addressed a todo).

shadow warning.

IEM,HM,PGM: Started on string I/O optimizations using IEM (disabled). Cleaned up confusing status code handling in hmR0VmxCheckForceFlags (involving PGM) as well as some use of incorrect doxygen groups (@name).

Too lazy to do: pushfq; cli; ... popfq ?

Too lazy to do: pushfq; cli; ... popfq ?

Too lazy to do: pushfq; cli; ... popfq ?

HMR0VMX.cpp: INS/OUTS todo update.

HMR0VMX.cpp: Attempt to fix incorrect DR7 and DR[0-3] checks in I/O port path.

VMM/HMVMXR0: Typos and make re-checking of CR4.VMXE on VMXR0Enter() only for strict builds.

VMM/HMVMXR0: nits.

CPUM,++: Fix DR6 and DR7 read-as-1 (RA1) and read-as-zero (RAZ) values on load since REM didn't set them right for years. Introduced constants for these values.

VMM/HMVMXR0: Minor clarification.

HMVMX: preserve SS.DPL and CS.L/D/G when the 'unusable' bit is set. SS.DPL = CPL, so we absolutely must preserve that.

hmR0VmxReadSegmentReg: Update on the unused stuff, prepared fix but needs some discussion+testing before enabling.

VMM/HM: comments.

VMM/HM: Dispatch host NMIs on Intel. Added separate STAM counter for host NMIs with the necessary changes to old, new VT-x, AMD-V code.

VMM/HMVMXR0: Assert that the valid bit is set on Xcpt exits.

VMM: XCPT_NMI -> XCPT_OR_NMI. Name it like Intel does in the spec.

VMM/HMVMXR0: Added a todo.

VMM/HMVMXR0: Log exit reasons for more rare/improbable gurus.

VMM: Naming convention. lasterror->LastError.

VMM/HM(VMX|SVM)R0: Easier to toggle statements.

VMM/HMVMXR0: More validations for the error code valid bit.

HMVMX: Look at the right exit info to figure out error code.

VMM: spaces.

VMM: nits.

VMM/HMVMXR0: Fail more gracefully when trying to run 64-bit guests on Darwin 32-bit/PAE kernels.

VMM: Fix LDTR restoration to be done dynamically. This is required for 64-bit Darwin hosts.

VMM: Optimized world-switch with lazy restoration LDTR and TR on Intel.

VMM/HM: Spaces, comment.

VMM/HM: Alignment and fix an assertion in the new AMD-V code.

VMM/HMVMXR0: Go back to ring-3 on host interrupts unconditionally if there's no preemption timer supported.

VMM/HMVMXR0: Avoid reloading guest state on multiple reschedules on longjmp trips.

VMM/HM: Cleanup event injection statistics.

VMM/VMMR0: HM profile counters.

VMM/HM: Doxygen.

VMM/HM: Refactor the state-flags for proper sharing between old, new VT-x and AMD-V code combinations.

VMM/HMVMXR0: Bits.

VMM/HMVMXR0: Incorrect debug assertion.

VMM/HMVMXR0: prefer typedef to macros when possible.

VMM/HMVMXR0: More debug assertion fixes, comments.

VMM/HMVMXR0: Fix guru with re-injecting software interrupts during event delivery for EPT violation exits (non-exception exits).

VMM/HMVMXR0: comments.

VMM/HMVMXR0: Reflect any exception to the guest if event delivery of that exception caused an EPT violation/misconfig. of APIC-access VM-exit. In this case the VM-exit interruption-information field won't be valid because the resulting VM-exit is -not- an exception.

VMM/HMVMXR0: grr.

VMM/HMVMXR0: Disable flushing of the log while loading the guest-state (only while entering VT-x from ring-3, debug builds).

VMM/HMVMXR0: SMP logging.

VMM/HMVMXR0: SMP logging.

VMM/HMSVMR0, HMVMXR0: AMD-V bits, VT-x minor change.

VMM/HMVMXR0: TPR patching is not needed on Intel, probably never was. The virtual-APIC feature has been documented since the first VT-x specification (Jan 2005).

VMM: AMD-V bits and VT-x comment nit.

VMM/HMVMXR0: comment.

VMM/HMSVMR0: AMD-V bits, TPR shadowing, other adjustments.

VMM/HMVMXR0: Don't need to update guest APIC state (i.e. TPR) for nested page faults that are not MMIO.

VMM/HMVMXR0: bits.

VMM/HMVMXR0: comment.

VMM/HMVMXR0: doxygen.

VMM/HMVMXR0: Comments.

VMM/HMVMXR0: Comments and tidying.

VMM/HMVMXR0: Skip redundant VMWRITES during VM startup.

VMM/HMVMXR0: Use UINT64_C only when needed.

VMM/HMVMXR0: Stats and skip calling the flush function unnecessarily.

VMM/HMVMXR0: Flush-by-EPT on VCPU reschedule rather than just VPID, another corner case.

VMM/HMVMXR0: Don't miss flushing-by-EPT if we the VCPU gets rescheduled to another host CPU before the flush happens.

VMM/HMVMXR0: comment.

VMM: HM bits, add changed-flag for CR2 as AMD-V has CR2 caching.

nit.

VMM, recompiler: Purge deprecated macros.

VMM/HMVMXR0: Flush all EPTP tagged-TLB entries (when possible) while enabling VMX on the host CPU. Not perfect yet.

VMM/HMVMXR0: Fix invalidating EPT-tagged TLB-entries when VPID+EPT is in use.

VMM/HMVMXR0: Avoid re-reading CR4 as it cannot be changed by vmxoff and some comments.

VMM/HMVMXR0: Log4Func to keep things at level 4 and not flow or level 1 logging.

VMM/HMVMXR0: Use log level 4 by default so that instrumenting logging (with lower levels) is much easier.

VMM/HM: AMD-V and VT-x state-change bits.

VMM/HM: Comment nits.

VMM/HMVMXR0: Added a couple of comments in the tagged-TLB flushing code.

VMM/HMVMXR0: Cleanup, and fixme todo.

VMM/HMVMXR0: debug build fix.

VMM/HMVMXR0: Avoid saving/restoring EFER whenever possible on every VM-entry/VM-exit.

VMM/HMVMXR0: 32-bit host fix.

VMM/HM: macro rename and comment update.

VMM: Optimized 64-bit host VT-x world-switch.

VMM/HM: CR0 and CR4 masks can be 32-bit. Plus hungarian naming fixes.

VMM/HMVMXR0: TR too, shouldn't be in the LDT and must have RPL 0.

VMM/HMVMXR0: Make single-stepping work in the hypervisor debugger when the monitor-trap flag is not available.

Started populating the RC address space in the debugger.

VMM/HMVMXR0: Log CS during loading the guest state.

VMM/HMVMXR0: nits.

VMM/HMVMXR0: Be more explicit about type sizes. It's rather critical in this area.

VMM/HMVMXR0: nit.

VMM/HMVMXR0: 32-bit hybrid darwin kernel fixes.

VMM/HMVMXR0: nit.

VMM/HMVMXR0: Fix Vista 64-bit guests to work again.

VMM/HMVMXR0: DR7 is 32-bit in reality, upper bits are reserved MBZ, so avoid complicated guest-natural width writes and going through the VMCS cache unnecessarily.

VMM/HMVMXR0: nit.

VMM/HMVMXR0: Kernel GS Base is no longer manually saved/restore across the world switch. If MSR-bitmaps are not supported, changing Kernel GS Base through WRMSR should reload the new value into the auto-load/store area in the VMCS.

VMM/HMVMXR0: In case we don't have TPR-shadow or CR8_LOAD exit is forced to always be set by the CPU, update the APIC state on CR8 writes.

VMM/HMVMXR0: comment.

VMM/HMVMXR0: Only cause a VM-exit when the highest-priority pending interrupt can be delivered to the guest and not whenever the guest lowers its TPR.

VMM: Facility for getting the highest-priority pending interrupt from the APIC device.

VMM: nits.

VMM: Name fixes for non-static functions.

VMM/HMVMXR0: spacing.

VMM/HMVMXR0: Don't mix up store and load. This is horrible terminology.

VMM: HM cleanup for CTRL, CONTROLS in symbolic names.

VMM/HMVMXR0: copy, paste typo.

VMM/HMVMXR0: paranoidroid.

VMM/HMVMXR0: Don't intercept CR3 reads with unrestricted guests.

VMM/HMVMXR0: Fix overly-delayed update of PGM's view of CR3. More longjmp fun.

VMM/HMVMXR0: Easier to read when reading huge logs.

VMM/HMVMXR0: State-save ordering enforcement assertion, comment and don't save autoload/store MSRs for getting EFER as it's always up to date.

VMM/HMVMXR0: lines.

VMM/HMVMXR0: Return of the AssertRCReturns!

Don't OR status codes together, just AssertRCReturn immediately. If two methods fails with different statuses, we'll be *very* confused. Also, for (human) analysis, return VINF_SUCCESS when you mean no informational statuses are used. int rc = VINF_SUCCESS; at the top and return rc; at the bottom *might* save an xor eax,eax; instruction in the slow path, but it makes it harder to read. (It also forces the compiler to remember the last rc value, which might not come without a cost on register starved 32-bit.)

VMM/HMVMXR0: uint64->32 fix.

VMM/HMVMXR0: Return proper status code while leaving root-mode not just VINF_SUCCESS always.

VMM/HMVMXR0: Fixes for pure 32-bit darwin hosts (not hybrid) with better identification and separation of guest-natural, host-natural fields.

VMM/HMVMXR0: limit macro scope.

VMM/HMVMXR0: Remove redundant and slightly misleading comment.

VMM: Use RT_BIT which uses unsigned to avoid nasty sign-extension issues with shifting. Also means less typing and looks cleaner.

VMM/HMVMXR0: Fix sign extension issue, use uint32_t.

VMM/HMVMXR0: Some clarification regarding the unusable bit.

VMM/HMVMXR0: Avoid syncing TRPM and Pending HM events back and forth when teasing to enter VT-x but we keep returning to ring-3 for one reason or another. Now do the actual conversion later during event injection when, currently, it's too late to go back to ring-3.

VMM/HMVMXR0: todo for thinking about tomorrow.

VMM/HMVMXR0: A few extra segment selector checks.

VMM/HMVMXR0: More checks on strict builds.

VMM/HMVMXR0: More aggressive state sync.: don't save IDTR, LDTR, GDTR, TR unless needed.

VMM/HMVMXR0: nit.

VMM: Fix for HM_PROFILE_EXIT_DISPATCH.

VMM/HMVMXR0: Assertion and clarified todo comment.

VMM/VMMR0: Fix debug assertion regarding leaving HM context with guest/hyper FPU/debug bits active.

VMM/HMVMXR0: Eliminated some unnecessary DECLCALLBACK calling convention.

VMM: STAM counter descriptions, cleanup. Fixed a few erroneous stats in the new VT-x code.

VMM/HMVMXR0: Remove assertion for now as it's an enum not a pfn.

VMMR0: build fix

VMM/HMVMXR0: wrap to 130 columns.

VMM/HMVMXR0: nits.

VMM/HMVMXR0: Redundant.

VMM/HMVMX0: We're already in non-preemptible context. Too paranoid to assume CR4.VMXE can be changed by interrupt handlers. Do less work, assert instead.

VMM/HMVMXR0: Restore TSC_AUX properly when not using auto-load/store MSR feature.

VMM/VMMR0: Load kernel-gs base MSR as part of the auto-load/store MSR feature. VMM/HMVMXR0: Fixed the code to also deal with the case when auto-load/store is not defined. VMM/VMMR0: Fixed LegacyandAmd64.mac to restore the MSRs from the stack in the right order, not sure how it could have worked previously.

VMM/HMVMXR0: Resetup preemption timer when the preemption timer expires unconditionally, and other HM bits.

VMM/HMVMXR0: Update TSC offsetting when the guest write TSC. This fixes the Windows 7 guest BSODs and hangs. Fixes also re-updating TSC offsetting when we get rescheduled to another CPU due to a longjmp.

Move HMRCA.asm into the switcher code so we don't need VMMRC.rc.

VMM/HMVMXR0: paranoia and comments.

VMM/HM: Simplified real/protected mode transition tracking logic.

VMM/HMVMXR0: unintentional commit of debugging stuff.

VMM/HMVMXR0: Darwin hybrid-kernel fixes.

VMM: Introduced VMCPUSTATE_STARTED_HM for indicating that we're between HMR3Enter and HMR3Leave. Added HMIsInHwVirtCtx and VMMIsLongJumpArmed methods/macros.

VMM/HMVMXR0: nit.

VMM/HMVMXR0: spaces.

VMM/HMVMXR0: Don't trash CR2 for VM-exits caused due to a #PF delivery when not intercepting #PFs (i.e. Nested Paging).

VMM/HMVMXR0: nits.

VMM/HMVMXR0: Theoretically possibly to miss a CR2 update when we should.

VMM/HMVMXR0: Alignment.

VMM/HMVMXR0: Build fix for HMVMX_ALWAYS_TRAP_PF.

VMM/HMVMXR0: Fix macro renaming breakage.

VMM/HMVMXR0: Alignment.

VMM: Introducing VINF_EM_TRIPLE_FAULT status code for triple-fault conditions in HM, to result in a guru meditation.

VMM/HMVMXR0: Safer to update CR2 while injecting exceptions and not earlier while converting from TRPM traps (from ring-3).

VMM/HMVMXR0: 32-bit fixes, round 2.

VMM/HMVMXR0: RFlags is 32-bit with upper half as always 0, avoid going through the VMCS read-cache for it.

VMM/HMVMXR0: Fix for 32-bit hosts, round 1.

VMMR0/HMVMXR0: Don't intercept CR0.WP when we have nested paging.

VMMR0/HMVMXR0: typo.

VMM/HMVMXR0: removed some superfluous comments.

VMMR0/HM: Profiling longjmp round trips are less useful, adjusted STAM counters to reflect true time spent on VMXR0RunGuestCode. - Removed excessive inlining in HMVMXR0. - Added exit-to-ring3 stat to HWVMXR0.

VMM/VMMR0: HM cleanup.

VMM/HMVMXR0: Actually we -do- need to intercept #DB (hypervisor debugging) but it's not handled in the #DB handler properly.

VMMR0/HMVMXR0: Fix single-stepping over STI during guest debugging and inadvertently intercepting #DB recursively.

VMMR0/HMVMXR0: Clear interrupt-inhibition if we're single-stepping in the hyperdebugger case.

VMMR0/HMVMXR0: Implemented setting of pending debug exceptions for single-stepping over STI and MOV SS.

VMMR0/HMVMXR0: bits.

VMMR0/HMVMXR0: Simultaneous hyper and guest single-stepping support when there is no Monitor Trap Flag capability on the CPU.

VMM/VMMR0: HM macro prefix cleanup a bit.

HMVMX: HMVMX_USE_FUNCTION_TABLE (enabled, as before) and HM_PROFILE_EXIT_DISPATCH (disabled).

VMM/VMMR0: HM bits.

VMMR0/HMVMXR0: Fix MOV DRx intercepts. Haiku now works.

VMMR0/HMVMXR0: fix pending events on task switches. Fixes Warcraft 2 for DOS with NestedPaging.

VMMR0/HMVMXR0: reflect #DE and others back to the guest on a vectoring #PF. This case is not covered by the Intel manuals.

VMMR0/HMVMXR0: Fix regression with EFER MSR write exits. Fixes NetBSD again.

VMMR0/HMVMXR0: oops

VMM/VMMR0: Update EFER on VM-reentry when WRMSR changes it. Fixes Solaris SMP guest panic.

VMM/VMMR0: bits.

VMM/VMMR0: HM bits.

VMM/VMMR0: HM bits.

HMVMX: Windows build fix.

VMM/VMMR0: HM bits.

VMM/VMMR0: More vectoring event fun.

VMM/VMMR0: More vectoring event handling.

VMM/VMMR0: Fixes to vectoring #PF, and reflecting exceptions to the guest.

VMM/VMMR0: HM bits.

VMM/VMMR0: HM bits.

VMM: use of symbolic names and asserts.

VMMR0/HMVMXR0: HM bits.

VMMR0/HMVMXR0: Redo of VT-x event injection. Some STAM cleanup. VMM: TRPM enhancements to include instruction-length information for software interrupts and exceptions (#BP, #OF).

VMMR0/HMVMXR0: Inject software int from TRPM

VMM: Clean up STAM counters and hungarian fixes.

VMMR0/HMVMXR0: Event injection, don't need to re-clear the field everytime.

VMMR0/HMVMXR0: type fixes.

HM: Alignment fixes. s/u32InstrLen/cbInstr/g. A couple of doxygen fixes. Changed a few 'unsigned' structure members to 'uint32_t' - we stopped believing in 'unsigned' 6-7 years ago.

VMMR0/HMVMXR0: Redo the event injection logic to be done only on the entry side. This is cleaner and any returns the recompiler should not lose any events.

VMM/VMMR0: HM bits.

VMMR0/HMVMXR0: leaner, part 2.

VMMR0/HMVMXR0: leaner, less parameter passing, part 1.

VMMR0/HMVMXR0: Fix LDTR usable bit, fixes NetBSD guests.

VMM/HMVMXR0: More assertions.

VMM/VMMR0: more consistent names.

VMM/VMMR0/HM: comment and todo for problem earlier seen today with assert screwing up VMX state somehow.

VMM: Remove unused error codes, use more specific error codes. Restore CR4 on VMXON failures in the new VT-x code.

VMM/VMMR0/HM: bits.

HM: WBINVD exiting must be conditional.

HM: 64-bit Windows build fix.

VMM/VMMR0/HM: build fix.

VMM/VMMR0/HM: build fixes for 32-bit hosts

VMM/VMMR0: HM bits.

VMM/VMMR0: obsolete comments.

VMM/VMMR0: HM bits, stam counter and switch experiment.

VMM/VMMR0: HM bits.

VMM/VMMR0: HM bits.

VMM/VMMR0: HM bits. Save preemption timer value on every VM-exit and reset it up when required.

VMM/VMMR0: HM bits. Fixed issue with MOV-SS and interrupt inhibition. Use interrupt-window exiting for interrupt-inhibited state as well for better guest interrupt latency.

VMM/VMMR0: HM bits.

VMM/VMMR0: HM bits.

VMM/VMMR0: comment.

VMM/VMMR0: HM bits, optimizations, be more aggressive with saving state for hot exit paths. Do less on every exit, moved pending event delivery to only the required exits.

VMM/VMMR0: HM bits.

VMM/VMMR0: HM bits, remove unused flags.

VMM/VMMR0: Very important stats.

VMM/VMMR0: HM bits.

VMM/VMMR0: HMVMXR0 properties.