2 * Copyright (c) 2012 ARM Limited
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions are
16 * met: redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer;
18 * redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution;
21 * neither the name of the copyright holders nor the names of its
22 * contributors may be used to endorse or promote products derived from
23 * this software without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
26 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
27 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
28 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
29 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
30 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
31 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
32 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
33 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
34 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
35 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 * Authors: Andreas Sandberg
40 #include <linux/kvm.h>
41 #include <sys/ioctl.h>
49 #include "arch/mmapped_ipr.hh"
50 #include "arch/utility.hh"
51 #include "cpu/kvm/base.hh"
52 #include "debug/Checkpoint.hh"
53 #include "debug/Drain.hh"
54 #include "debug/Kvm.hh"
55 #include "debug/KvmIO.hh"
56 #include "debug/KvmRun.hh"
57 #include "params/BaseKvmCPU.hh"
58 #include "sim/process.hh"
59 #include "sim/system.hh"
63 /* Used by some KVM macros */
64 #define PAGE_SIZE pageSize
66 volatile bool timerOverflowed
= false;
69 onTimerOverflow(int signo
, siginfo_t
*si
, void *data
)
71 timerOverflowed
= true;
74 BaseKvmCPU::BaseKvmCPU(BaseKvmCPUParams
*params
)
78 dataPort(name() + ".dcache_port", this),
79 instPort(name() + ".icache_port", this),
80 threadContextDirty(true),
82 vcpuID(vm
.allocVCPUID()), vcpuFD(-1), vcpuMMapSize(0),
83 _kvmRun(NULL
), mmioRing(NULL
),
84 pageSize(sysconf(_SC_PAGE_SIZE
)),
86 perfControlledByTimer(params
->usePerfOverflow
),
87 hostFreq(params
->hostFreq
),
88 hostFactor(params
->hostFactor
),
93 panic("KVM: Failed to determine host page size (%i)\n",
96 thread
= new SimpleThread(this, 0, params
->system
,
97 params
->itb
, params
->dtb
, params
->isa
[0]);
98 thread
->setStatus(ThreadContext::Halted
);
100 threadContexts
.push_back(tc
);
104 if (params
->usePerfOverflow
)
105 runTimer
.reset(new PerfKvmTimer(hwCycles
,
110 runTimer
.reset(new PosixKvmTimer(KVM_TIMER_SIGNAL
, CLOCK_MONOTONIC
,
115 BaseKvmCPU::~BaseKvmCPU()
118 munmap(_kvmRun
, vcpuMMapSize
);
128 fatal("KVM: Multithreading not supported");
130 tc
->initMemProxies(tc
);
132 // initialize CPU, including PC
133 if (FullSystem
&& !switchedOut())
134 TheISA::initCPU(tc
, tc
->contextId());
136 mmio_req
.setThreadContext(tc
->contextId(), 0);
140 BaseKvmCPU::startup()
142 const BaseKvmCPUParams
* const p(
143 dynamic_cast<const BaseKvmCPUParams
*>(params()));
149 assert(vcpuFD
== -1);
151 // Tell the VM that a CPU is about to start.
154 // We can't initialize KVM CPUs in BaseKvmCPU::init() since we are
155 // not guaranteed that the parent KVM VM has initialized at that
156 // point. Initialize virtual CPUs here instead.
157 vcpuFD
= vm
.createVCPU(vcpuID
);
159 // Setup signal handlers. This has to be done after the vCPU is
160 // created since it manipulates the vCPU signal mask.
161 setupSignalHandler();
163 // Map the KVM run structure */
164 vcpuMMapSize
= kvm
.getVCPUMMapSize();
165 _kvmRun
= (struct kvm_run
*)mmap(0, vcpuMMapSize
,
166 PROT_READ
| PROT_WRITE
, MAP_SHARED
,
168 if (_kvmRun
== MAP_FAILED
)
169 panic("KVM: Failed to map run data structure\n");
171 // Setup a pointer to the MMIO ring buffer if coalesced MMIO is
172 // available. The offset into the KVM's communication page is
173 // provided by the coalesced MMIO capability.
174 int mmioOffset(kvm
.capCoalescedMMIO());
175 if (!p
->useCoalescedMMIO
) {
176 inform("KVM: Coalesced MMIO disabled by config.\n");
177 } else if (mmioOffset
) {
178 inform("KVM: Coalesced IO available\n");
179 mmioRing
= (struct kvm_coalesced_mmio_ring
*)(
180 (char *)_kvmRun
+ (mmioOffset
* pageSize
));
182 inform("KVM: Coalesced not supported by host OS\n");
189 BaseKvmCPU::regStats()
191 using namespace Stats
;
196 .name(name() + ".committedInsts")
197 .desc("Number of instructions committed")
201 .name(name() + ".numVMExits")
202 .desc("total number of KVM exits")
206 .name(name() + ".numVMHalfEntries")
207 .desc("number of KVM entries to finalize pending operations")
211 .name(name() + ".numExitSignal")
212 .desc("exits due to signal delivery")
216 .name(name() + ".numMMIO")
217 .desc("number of VM exits due to memory mapped IO")
221 .name(name() + ".numCoalescedMMIO")
222 .desc("number of coalesced memory mapped IO requests")
226 .name(name() + ".numIO")
227 .desc("number of VM exits due to legacy IO")
231 .name(name() + ".numHalt")
232 .desc("number of VM exits due to wait for interrupt instructions")
236 .name(name() + ".numInterrupts")
237 .desc("number of interrupts delivered")
241 .name(name() + ".numHypercalls")
242 .desc("number of hypercalls")
247 BaseKvmCPU::serializeThread(std::ostream
&os
, ThreadID tid
)
249 if (DTRACE(Checkpoint
)) {
250 DPRINTF(Checkpoint
, "KVM: Serializing thread %i:\n", tid
);
255 assert(_status
== Idle
);
256 thread
->serialize(os
);
260 BaseKvmCPU::unserializeThread(Checkpoint
*cp
, const std::string
§ion
,
263 DPRINTF(Checkpoint
, "KVM: Unserialize thread %i:\n", tid
);
266 assert(_status
== Idle
);
267 thread
->unserialize(cp
, section
);
268 threadContextDirty
= true;
272 BaseKvmCPU::drain(DrainManager
*dm
)
277 DPRINTF(Drain
, "BaseKvmCPU::drain\n");
280 // The base KVM code is normally ready when it is in the
281 // Running state, but the architecture specific code might be
282 // of a different opinion. This may happen when the CPU been
283 // notified of an event that hasn't been accepted by the vCPU
285 if (!archIsDrained()) {
290 // The state of the CPU is consistent, so we don't need to do
291 // anything special to drain it. We simply de-schedule the
292 // tick event and enter the Idle state to prevent nasty things
293 // like MMIOs from happening.
294 if (tickEvent
.scheduled())
295 deschedule(tickEvent
);
300 // Idle, no need to drain
301 assert(!tickEvent
.scheduled());
303 // Sync the thread context here since we'll need it when we
304 // switch CPUs or checkpoint the CPU.
309 case RunningServiceCompletion
:
310 // The CPU has just requested a service that was handled in
311 // the RunningService state, but the results have still not
312 // been reported to the CPU. Now, we /could/ probably just
313 // update the register state ourselves instead of letting KVM
314 // handle it, but that would be tricky. Instead, we enter KVM
315 // and let it do its stuff.
318 DPRINTF(Drain
, "KVM CPU is waiting for service completion, "
319 "requesting drain.\n");
323 // We need to drain since the CPU is waiting for service (e.g., MMIOs)
326 DPRINTF(Drain
, "KVM CPU is waiting for service, requesting drain.\n");
330 panic("KVM: Unhandled CPU state in drain()\n");
336 BaseKvmCPU::drainResume()
338 assert(!tickEvent
.scheduled());
340 // We might have been switched out. In that case, we don't need to
345 DPRINTF(Kvm
, "drainResume\n");
348 // The tick event is de-scheduled as a part of the draining
349 // process. Re-schedule it if the thread context is active.
350 if (tc
->status() == ThreadContext::Active
) {
351 schedule(tickEvent
, nextCycle());
359 BaseKvmCPU::switchOut()
361 DPRINTF(Kvm
, "switchOut\n");
363 BaseCPU::switchOut();
365 // We should have drained prior to executing a switchOut, which
366 // means that the tick event shouldn't be scheduled and the CPU is
368 assert(!tickEvent
.scheduled());
369 assert(_status
== Idle
);
373 BaseKvmCPU::takeOverFrom(BaseCPU
*cpu
)
375 DPRINTF(Kvm
, "takeOverFrom\n");
377 BaseCPU::takeOverFrom(cpu
);
379 // We should have drained prior to executing a switchOut, which
380 // means that the tick event shouldn't be scheduled and the CPU is
382 assert(!tickEvent
.scheduled());
383 assert(_status
== Idle
);
384 assert(threadContexts
.size() == 1);
386 // Force an update of the KVM state here instead of flagging the
387 // TC as dirty. This is not ideal from a performance point of
388 // view, but it makes debugging easier as it allows meaningful KVM
389 // state to be dumped before and after a takeover.
391 threadContextDirty
= false;
395 BaseKvmCPU::verifyMemoryMode() const
397 if (!(system
->isAtomicMode() && system
->bypassCaches())) {
398 fatal("The KVM-based CPUs requires the memory system to be in the "
399 "'atomic_noncaching' mode.\n");
406 DPRINTF(Kvm
, "wakeup()\n");
408 if (thread
->status() != ThreadContext::Suspended
)
415 BaseKvmCPU::activateContext(ThreadID thread_num
, Cycles delay
)
417 DPRINTF(Kvm
, "ActivateContext %d (%d cycles)\n", thread_num
, delay
);
419 assert(thread_num
== 0);
422 assert(_status
== Idle
);
423 assert(!tickEvent
.scheduled());
425 numCycles
+= ticksToCycles(thread
->lastActivate
- thread
->lastSuspend
);
427 schedule(tickEvent
, clockEdge(delay
));
433 BaseKvmCPU::suspendContext(ThreadID thread_num
)
435 DPRINTF(Kvm
, "SuspendContext %d\n", thread_num
);
437 assert(thread_num
== 0);
443 assert(_status
== Running
);
445 // The tick event may no be scheduled if the quest has requested
446 // the monitor to wait for interrupts. The normal CPU models can
447 // get their tick events descheduled by quiesce instructions, but
448 // that can't happen here.
449 if (tickEvent
.scheduled())
450 deschedule(tickEvent
);
456 BaseKvmCPU::deallocateContext(ThreadID thread_num
)
458 // for now, these are equivalent
459 suspendContext(thread_num
);
463 BaseKvmCPU::haltContext(ThreadID thread_num
)
465 // for now, these are equivalent
466 suspendContext(thread_num
);
470 BaseKvmCPU::getContext(int tn
)
479 BaseKvmCPU::totalInsts() const
485 BaseKvmCPU::totalOps() const
487 hack_once("Pretending totalOps is equivalent to totalInsts()\n");
494 inform("State dumping not implemented.");
501 assert(_status
!= Idle
);
505 // handleKvmExit() will determine the next state of the CPU
506 delay
= handleKvmExit();
512 case RunningServiceCompletion
:
514 Tick
ticksToExecute(mainEventQueue
.nextTick() - curTick());
516 // We might need to update the KVM state.
519 DPRINTF(KvmRun
, "Entering KVM...\n");
521 // Force an immediate exit from KVM after completing
522 // pending operations. The architecture-specific code
523 // takes care to run until it is in a state where it can
524 // safely be drained.
525 delay
= kvmRunDrain();
527 delay
= kvmRun(ticksToExecute
);
530 // Entering into KVM implies that we'll have to reload the thread
531 // context from KVM if we want to access it. Flag the KVM state as
532 // dirty with respect to the cached thread context.
533 kvmStateDirty
= true;
535 // Enter into the RunningService state unless the
536 // simulation was stopped by a timer.
537 if (_kvmRun
->exit_reason
!= KVM_EXIT_INTR
) {
538 _status
= RunningService
;
549 panic("BaseKvmCPU entered tick() in an illegal state (%i)\n",
553 // Schedule a new tick if we are still running
555 schedule(tickEvent
, clockEdge(ticksToCycles(delay
)));
559 BaseKvmCPU::kvmRunDrain()
561 // By default, the only thing we need to drain is a pending IO
562 // operation which assumes that we are in the
563 // RunningServiceCompletion state.
564 assert(_status
== RunningServiceCompletion
);
566 // Deliver the data from the pending IO operation and immediately
572 BaseKvmCPU::getHostCycles() const
574 return hwCycles
.read();
578 BaseKvmCPU::kvmRun(Tick ticks
)
581 DPRINTF(KvmRun
, "KVM: Executing for %i ticks\n", ticks
);
582 timerOverflowed
= false;
585 // Settings ticks == 0 is a special case which causes an entry
586 // into KVM that finishes pending operations (e.g., IO) and
587 // then immediately exits.
588 DPRINTF(KvmRun
, "KVM: Delivering IO without full guest entry\n");
592 // This signal is always masked while we are executing in gem5
593 // and gets unmasked temporarily as soon as we enter into
594 // KVM. See setSignalMask() and setupSignalHandler().
595 raise(KVM_TIMER_SIGNAL
);
597 // Enter into KVM. KVM will check for signals after completing
598 // pending operations (IO). Since the KVM_TIMER_SIGNAL is
599 // pending, this forces an immediate exit into gem5 again. We
600 // don't bother to setup timers since this shouldn't actually
601 // execute any code in the guest.
604 // We always execute at least one cycle to prevent the
605 // BaseKvmCPU::tick() to be rescheduled on the same tick
607 ticksExecuted
= clockPeriod();
609 if (ticks
< runTimer
->resolution()) {
610 DPRINTF(KvmRun
, "KVM: Adjusting tick count (%i -> %i)\n",
611 ticks
, runTimer
->resolution());
612 ticks
= runTimer
->resolution();
615 // Get hardware statistics after synchronizing contexts. The KVM
616 // state update might affect guest cycle counters.
617 uint64_t baseCycles(getHostCycles());
618 uint64_t baseInstrs(hwInstructions
.read());
620 // Arm the run timer and start the cycle timer if it isn't
621 // controlled by the overflow timer. Starting/stopping the cycle
622 // timer automatically starts the other perf timers as they are in
623 // the same counter group.
624 runTimer
->arm(ticks
);
625 if (!perfControlledByTimer
)
631 if (!perfControlledByTimer
)
634 // The timer signal may have been delivered after we exited
635 // from KVM. It will be pending in that case since it is
636 // masked when we aren't executing in KVM. Discard it to make
637 // sure we don't deliver it immediately next time we try to
639 discardPendingSignal(KVM_TIMER_SIGNAL
);
641 const uint64_t hostCyclesExecuted(getHostCycles() - baseCycles
);
642 const uint64_t simCyclesExecuted(hostCyclesExecuted
* hostFactor
);
643 const uint64_t instsExecuted(hwInstructions
.read() - baseInstrs
);
644 ticksExecuted
= runTimer
->ticksFromHostCycles(hostCyclesExecuted
);
646 if (ticksExecuted
< ticks
&&
648 _kvmRun
->exit_reason
== KVM_EXIT_INTR
) {
649 // TODO: We should probably do something clever here...
650 warn("KVM: Early timer event, requested %i ticks but got %i ticks.\n",
651 ticks
, ticksExecuted
);
654 /* Update statistics */
655 numCycles
+= simCyclesExecuted
;;
656 numInsts
+= instsExecuted
;
657 ctrInsts
+= instsExecuted
;
658 system
->totalNumInsts
+= instsExecuted
;
661 "KVM: Executed %i instructions in %i cycles "
662 "(%i ticks, sim cycles: %i).\n",
663 instsExecuted
, hostCyclesExecuted
, ticksExecuted
, simCyclesExecuted
);
668 return ticksExecuted
+ flushCoalescedMMIO();
672 BaseKvmCPU::kvmNonMaskableInterrupt()
675 if (ioctl(KVM_NMI
) == -1)
676 panic("KVM: Failed to deliver NMI to virtual CPU\n");
680 BaseKvmCPU::kvmInterrupt(const struct kvm_interrupt
&interrupt
)
683 if (ioctl(KVM_INTERRUPT
, (void *)&interrupt
) == -1)
684 panic("KVM: Failed to deliver interrupt to virtual CPU\n");
688 BaseKvmCPU::getRegisters(struct kvm_regs
®s
) const
690 if (ioctl(KVM_GET_REGS
, ®s
) == -1)
691 panic("KVM: Failed to get guest registers\n");
695 BaseKvmCPU::setRegisters(const struct kvm_regs
®s
)
697 if (ioctl(KVM_SET_REGS
, (void *)®s
) == -1)
698 panic("KVM: Failed to set guest registers\n");
702 BaseKvmCPU::getSpecialRegisters(struct kvm_sregs
®s
) const
704 if (ioctl(KVM_GET_SREGS
, ®s
) == -1)
705 panic("KVM: Failed to get guest special registers\n");
709 BaseKvmCPU::setSpecialRegisters(const struct kvm_sregs
®s
)
711 if (ioctl(KVM_SET_SREGS
, (void *)®s
) == -1)
712 panic("KVM: Failed to set guest special registers\n");
716 BaseKvmCPU::getFPUState(struct kvm_fpu
&state
) const
718 if (ioctl(KVM_GET_FPU
, &state
) == -1)
719 panic("KVM: Failed to get guest FPU state\n");
723 BaseKvmCPU::setFPUState(const struct kvm_fpu
&state
)
725 if (ioctl(KVM_SET_FPU
, (void *)&state
) == -1)
726 panic("KVM: Failed to set guest FPU state\n");
731 BaseKvmCPU::setOneReg(uint64_t id
, const void *addr
)
733 #ifdef KVM_SET_ONE_REG
734 struct kvm_one_reg reg
;
736 reg
.addr
= (uint64_t)addr
;
738 if (ioctl(KVM_SET_ONE_REG
, ®
) == -1) {
739 panic("KVM: Failed to set register (0x%x) value (errno: %i)\n",
743 panic("KVM_SET_ONE_REG is unsupported on this platform.\n");
748 BaseKvmCPU::getOneReg(uint64_t id
, void *addr
) const
750 #ifdef KVM_GET_ONE_REG
751 struct kvm_one_reg reg
;
753 reg
.addr
= (uint64_t)addr
;
755 if (ioctl(KVM_GET_ONE_REG
, ®
) == -1) {
756 panic("KVM: Failed to get register (0x%x) value (errno: %i)\n",
760 panic("KVM_GET_ONE_REG is unsupported on this platform.\n");
765 BaseKvmCPU::getAndFormatOneReg(uint64_t id
) const
767 #ifdef KVM_GET_ONE_REG
768 std::ostringstream ss
;
770 ss
.setf(std::ios::hex
, std::ios::basefield
);
771 ss
.setf(std::ios::showbase
);
772 #define HANDLE_INTTYPE(len) \
773 case KVM_REG_SIZE_U ## len: { \
774 uint ## len ## _t value; \
775 getOneReg(id, &value); \
779 #define HANDLE_ARRAY(len) \
780 case KVM_REG_SIZE_U ## len: { \
781 uint8_t value[len / 8]; \
782 getOneReg(id, value); \
783 ss << "[" << value[0]; \
784 for (int i = 1; i < len / 8; ++i) \
785 ss << ", " << value[i]; \
789 switch (id
& KVM_REG_SIZE_MASK
) {
802 #undef HANDLE_INTTYPE
807 panic("KVM_GET_ONE_REG is unsupported on this platform.\n");
812 BaseKvmCPU::syncThreadContext()
817 assert(!threadContextDirty
);
819 updateThreadContext();
820 kvmStateDirty
= false;
824 BaseKvmCPU::syncKvmState()
826 if (!threadContextDirty
)
829 assert(!kvmStateDirty
);
832 threadContextDirty
= false;
836 BaseKvmCPU::handleKvmExit()
838 DPRINTF(KvmRun
, "handleKvmExit (exit_reason: %i)\n", _kvmRun
->exit_reason
);
839 assert(_status
== RunningService
);
841 // Switch into the running state by default. Individual handlers
842 // can override this.
844 switch (_kvmRun
->exit_reason
) {
845 case KVM_EXIT_UNKNOWN
:
846 return handleKvmExitUnknown();
848 case KVM_EXIT_EXCEPTION
:
849 return handleKvmExitException();
852 _status
= RunningServiceCompletion
;
854 return handleKvmExitIO();
856 case KVM_EXIT_HYPERCALL
:
858 return handleKvmExitHypercall();
861 /* The guest has halted and is waiting for interrupts */
862 DPRINTF(Kvm
, "handleKvmExitHalt\n");
865 // Suspend the thread until the next interrupt arrives
868 // This is actually ignored since the thread is suspended.
872 _status
= RunningServiceCompletion
;
873 /* Service memory mapped IO requests */
874 DPRINTF(KvmIO
, "KVM: Handling MMIO (w: %u, addr: 0x%x, len: %u)\n",
875 _kvmRun
->mmio
.is_write
,
876 _kvmRun
->mmio
.phys_addr
, _kvmRun
->mmio
.len
);
879 return doMMIOAccess(_kvmRun
->mmio
.phys_addr
, _kvmRun
->mmio
.data
,
880 _kvmRun
->mmio
.len
, _kvmRun
->mmio
.is_write
);
882 case KVM_EXIT_IRQ_WINDOW_OPEN
:
883 return handleKvmExitIRQWindowOpen();
885 case KVM_EXIT_FAIL_ENTRY
:
886 return handleKvmExitFailEntry();
889 /* KVM was interrupted by a signal, restart it in the next
893 case KVM_EXIT_INTERNAL_ERROR
:
894 panic("KVM: Internal error (suberror: %u)\n",
895 _kvmRun
->internal
.suberror
);
899 panic("KVM: Unexpected exit (exit_reason: %u)\n", _kvmRun
->exit_reason
);
904 BaseKvmCPU::handleKvmExitIO()
906 panic("KVM: Unhandled guest IO (dir: %i, size: %i, port: 0x%x, count: %i)\n",
907 _kvmRun
->io
.direction
, _kvmRun
->io
.size
,
908 _kvmRun
->io
.port
, _kvmRun
->io
.count
);
912 BaseKvmCPU::handleKvmExitHypercall()
914 panic("KVM: Unhandled hypercall\n");
918 BaseKvmCPU::handleKvmExitIRQWindowOpen()
920 warn("KVM: Unhandled IRQ window.\n");
926 BaseKvmCPU::handleKvmExitUnknown()
929 panic("KVM: Unknown error when starting vCPU (hw reason: 0x%llx)\n",
930 _kvmRun
->hw
.hardware_exit_reason
);
934 BaseKvmCPU::handleKvmExitException()
937 panic("KVM: Got exception when starting vCPU "
938 "(exception: %u, error_code: %u)\n",
939 _kvmRun
->ex
.exception
, _kvmRun
->ex
.error_code
);
943 BaseKvmCPU::handleKvmExitFailEntry()
946 panic("KVM: Failed to enter virtualized mode (hw reason: 0x%llx)\n",
947 _kvmRun
->fail_entry
.hardware_entry_failure_reason
);
951 BaseKvmCPU::doMMIOAccess(Addr paddr
, void *data
, int size
, bool write
)
953 ThreadContext
*tc(thread
->getTC());
956 mmio_req
.setPhys(paddr
, size
, Request::UNCACHEABLE
, dataMasterId());
957 // Some architectures do need to massage physical addresses a bit
958 // before they are inserted into the memory system. This enables
959 // APIC accesses on x86 and m5ops where supported through a MMIO
961 BaseTLB::Mode
tlb_mode(write
? BaseTLB::Write
: BaseTLB::Read
);
962 Fault
fault(tc
->getDTBPtr()->finalizePhysical(&mmio_req
, tc
, tlb_mode
));
963 if (fault
!= NoFault
)
964 warn("Finalization of MMIO address failed: %s\n", fault
->name());
967 const MemCmd
cmd(write
? MemCmd::WriteReq
: MemCmd::ReadReq
);
968 Packet
pkt(&mmio_req
, cmd
);
969 pkt
.dataStatic(data
);
971 if (mmio_req
.isMmappedIpr()) {
972 const Cycles
ipr_delay(write
?
973 TheISA::handleIprWrite(tc
, &pkt
) :
974 TheISA::handleIprRead(tc
, &pkt
));
975 return clockEdge(ipr_delay
);
977 return dataPort
.sendAtomic(&pkt
);
982 BaseKvmCPU::setSignalMask(const sigset_t
*mask
)
984 std::unique_ptr
<struct kvm_signal_mask
> kvm_mask
;
987 kvm_mask
.reset((struct kvm_signal_mask
*)operator new(
988 sizeof(struct kvm_signal_mask
) + sizeof(*mask
)));
989 // The kernel and the user-space headers have different ideas
990 // about the size of sigset_t. This seems like a massive hack,
991 // but is actually what qemu does.
992 assert(sizeof(*mask
) >= 8);
994 memcpy(kvm_mask
->sigset
, mask
, kvm_mask
->len
);
997 if (ioctl(KVM_SET_SIGNAL_MASK
, (void *)kvm_mask
.get()) == -1)
998 panic("KVM: Failed to set vCPU signal mask (errno: %i)\n",
1003 BaseKvmCPU::ioctl(int request
, long p1
) const
1006 panic("KVM: CPU ioctl called before initialization\n");
1008 return ::ioctl(vcpuFD
, request
, p1
);
1012 BaseKvmCPU::flushCoalescedMMIO()
1017 DPRINTF(KvmIO
, "KVM: Flushing the coalesced MMIO ring buffer\n");
1019 // TODO: We might need to do synchronization when we start to
1020 // support multiple CPUs
1022 while (mmioRing
->first
!= mmioRing
->last
) {
1023 struct kvm_coalesced_mmio
&ent(
1024 mmioRing
->coalesced_mmio
[mmioRing
->first
]);
1026 DPRINTF(KvmIO
, "KVM: Handling coalesced MMIO (addr: 0x%x, len: %u)\n",
1027 ent
.phys_addr
, ent
.len
);
1030 ticks
+= doMMIOAccess(ent
.phys_addr
, ent
.data
, ent
.len
, true);
1032 mmioRing
->first
= (mmioRing
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
1039 BaseKvmCPU::setupSignalHandler()
1041 struct sigaction sa
;
1043 memset(&sa
, 0, sizeof(sa
));
1044 sa
.sa_sigaction
= onTimerOverflow
;
1045 sa
.sa_flags
= SA_SIGINFO
| SA_RESTART
;
1046 if (sigaction(KVM_TIMER_SIGNAL
, &sa
, NULL
) == -1)
1047 panic("KVM: Failed to setup vCPU signal handler\n");
1050 if (sigprocmask(SIG_BLOCK
, NULL
, &sigset
) == -1)
1051 panic("KVM: Failed get signal mask\n");
1053 // Request KVM to setup the same signal mask as we're currently
1054 // running with. We'll sometimes need to mask the KVM_TIMER_SIGNAL
1055 // to cause immediate exits from KVM after servicing IO
1056 // requests. See kvmRun().
1057 setSignalMask(&sigset
);
1059 // Mask the KVM_TIMER_SIGNAL so it isn't delivered unless we're
1060 // actually executing inside KVM.
1061 sigaddset(&sigset
, KVM_TIMER_SIGNAL
);
1062 if (sigprocmask(SIG_SETMASK
, &sigset
, NULL
) == -1)
1063 panic("KVM: Failed mask the KVM timer signal\n");
1067 BaseKvmCPU::discardPendingSignal(int signum
) const
1069 int discardedSignal
;
1071 // Setting the timeout to zero causes sigtimedwait to return
1073 struct timespec timeout
;
1075 timeout
.tv_nsec
= 0;
1078 sigemptyset(&sigset
);
1079 sigaddset(&sigset
, signum
);
1082 discardedSignal
= sigtimedwait(&sigset
, NULL
, &timeout
);
1083 } while (discardedSignal
== -1 && errno
== EINTR
);
1085 if (discardedSignal
== signum
)
1087 else if (discardedSignal
== -1 && errno
== EAGAIN
)
1090 panic("Unexpected return value from sigtimedwait: %i (errno: %i)\n",
1091 discardedSignal
, errno
);
1095 BaseKvmCPU::setupCounters()
1097 DPRINTF(Kvm
, "Attaching cycle counter...\n");
1098 PerfKvmCounterConfig
cfgCycles(PERF_TYPE_HARDWARE
,
1099 PERF_COUNT_HW_CPU_CYCLES
);
1100 cfgCycles
.disabled(true)
1103 if (perfControlledByTimer
) {
1104 // We need to configure the cycles counter to send overflows
1105 // since we are going to use it to trigger timer signals that
1106 // trap back into m5 from KVM. In practice, this means that we
1107 // need to set some non-zero sample period that gets
1108 // overridden when the timer is armed.
1109 cfgCycles
.wakeupEvents(1)
1113 hwCycles
.attach(cfgCycles
,
1114 0); // TID (0 => currentThread)
1116 DPRINTF(Kvm
, "Attaching instruction counter...\n");
1117 PerfKvmCounterConfig
cfgInstructions(PERF_TYPE_HARDWARE
,
1118 PERF_COUNT_HW_INSTRUCTIONS
);
1119 hwInstructions
.attach(cfgInstructions
,
1120 0, // TID (0 => currentThread)
1125 BaseKvmCPU::tryDrain()
1130 if (!archIsDrained()) {
1131 DPRINTF(Drain
, "tryDrain: Architecture code is not ready.\n");
1135 if (_status
== Idle
|| _status
== Running
) {
1137 "tryDrain: CPU transitioned into the Idle state, drain done\n");
1138 drainManager
->signalDrainDone();
1139 drainManager
= NULL
;
1142 DPRINTF(Drain
, "tryDrain: CPU not ready.\n");
1148 BaseKvmCPU::ioctlRun()
1150 if (ioctl(KVM_RUN
) == -1) {
1152 panic("KVM: Failed to start virtual CPU (errno: %i)\n",