1/*
2 * Copyright (c) 2012 ARM Limited
3 * All rights reserved
4 *
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.
13 *
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.
24 *
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.
36 *
37 * Authors: Andreas Sandberg
38 */
39
40#ifndef __CPU_KVM_BASE_HH__
41#define __CPU_KVM_BASE_HH__
42
43#include <memory>
44
45#include "base/statistics.hh"
46#include "cpu/kvm/perfevent.hh"
47#include "cpu/kvm/timer.hh"
48#include "cpu/kvm/vm.hh"
49#include "cpu/base.hh"
50#include "cpu/simple_thread.hh"
51
52/** Signal to use to trigger time-based exits from KVM */
53#define KVM_TIMER_SIGNAL SIGRTMIN
54
55// forward declarations
56class ThreadContext;
57struct BaseKvmCPUParams;
58
59/**
60 * Base class for KVM based CPU models
61 *
62 * All architecture specific KVM implementation should inherit from
63 * this class. The most basic CPU models only need to override the
64 * updateKvmState() and updateThreadContext() methods to implement
65 * state synchronization between gem5 and KVM.
66 *
67 * The architecture specific implementation is also responsible for
68 * delivering interrupts into the VM. This is typically done by
69 * overriding tick() and checking the thread context before entering
70 * into the VM. In order to deliver an interrupt, the implementation
71 * then calls KvmVM::setIRQLine() or BaseKvmCPU::kvmInterrupt()
72 * depending on the specifics of the underlying hardware/drivers.
73 */
74class BaseKvmCPU : public BaseCPU
75{
76 public:
77 BaseKvmCPU(BaseKvmCPUParams *params);
78 virtual ~BaseKvmCPU();
79
80 void init();
81 void startup();
82 void regStats();
83
84 void serializeThread(std::ostream &os, ThreadID tid);
85 void unserializeThread(Checkpoint *cp, const std::string &section,
86 ThreadID tid);
87
88 unsigned int drain(DrainManager *dm);
89 void drainResume();
90
91 void switchOut();
92 void takeOverFrom(BaseCPU *cpu);
93
94 void verifyMemoryMode() const;
95
96 MasterPort &getDataPort() { return dataPort; }
97 MasterPort &getInstPort() { return instPort; }
98
99 void wakeup();
100 void activateContext(ThreadID thread_num, Cycles delay);
101 void suspendContext(ThreadID thread_num);
102 void deallocateContext(ThreadID thread_num);
103 void haltContext(ThreadID thread_num);
104
105 ThreadContext *getContext(int tn);
106
107 Counter totalInsts() const;
108 Counter totalOps() const;
109
110 /** Dump the internal state to the terminal. */
111 virtual void dump();
112
113 /**
114 * A cached copy of a thread's state in the form of a SimpleThread
115 * object.
116 *
117 * Normally the actual thread state is stored in the KVM vCPU. If KVM has
118 * been running this copy is will be out of date. If we recently handled
119 * some events within gem5 that required state to be updated this could be
120 * the most up-to-date copy. When getContext() or updateThreadContext() is
121 * called this copy gets updated. The method syncThreadContext can
122 * be used within a KVM CPU to update the thread context if the
123 * KVM state is dirty (i.e., the vCPU has been run since the last
124 * update).
125 */
126 SimpleThread *thread;
127
128 /** ThreadContext object, provides an interface for external
129 * objects to modify this thread's state.
130 */
131 ThreadContext *tc;
132
133 KvmVM &vm;
134
135 protected:
136 enum Status {
137 /** Context not scheduled in KVM */
138 Idle,
139 /** Running normally */
140 Running,
141 };
142
143 /** CPU run state */
144 Status _status;
145
146 /**
147 * Execute the CPU until the next event in the main event queue or
148 * until the guest needs service from gem5.
149 *
150 * @note This method is virtual in order to allow implementations
151 * to check for architecture specific events (e.g., interrupts)
152 * before entering the VM.
153 */
154 virtual void tick();
155
156 /**
157 * Get the value of the hardware cycle counter in the guest.
158 *
159 * This method is supposed to return the total number of cycles
160 * executed in hardware mode relative to some arbitrary point in
161 * the past. It's mainly used when estimating the number of cycles
162 * actually executed by the CPU in kvmRun(). The default behavior
163 * of this method is to use the cycles performance counter, but
164 * some architectures may want to use internal registers instead.
165 *
166 * @return Number of host cycles executed relative to an undefined
167 * point in the past.
168 */
169 virtual uint64_t getHostCycles() const;
170
171 /**
172 * Request KVM to run the guest for a given number of ticks. The
173 * method returns the approximate number of ticks executed.
174 *
175 * @note The returned number of ticks can be both larger or
176 * smaller than the requested number of ticks. A smaller number
177 * can, for example, occur when the guest executes MMIO. A larger
178 * number is typically due to performance counter inaccuracies.
179 *
180 * @param ticks Number of ticks to execute
181 * @return Number of ticks executed (see note)
182 */
183 Tick kvmRun(Tick ticks);
184
185 /**
186 * Get a pointer to the kvm_run structure containing all the input
187 * and output parameters from kvmRun().
188 */
189 struct kvm_run *getKvmRunState() { return _kvmRun; };
190
191 /**
192 * Retrieve a pointer to guest data stored at the end of the
193 * kvm_run structure. This is mainly used for PIO operations
194 * (KVM_EXIT_IO).
195 *
196 * @param offset Offset as specified by the kvm_run structure
197 * @return Pointer to guest data
198 */
199 uint8_t *getGuestData(uint64_t offset) const {
200 return (uint8_t *)_kvmRun + offset;
201 };
202
203 /**
204 * @addtogroup KvmInterrupts
205 * @{
206 */
207 /**
208 * Send a non-maskable interrupt to the guest
209 *
210 * @note The presence of this call depends on Kvm::capUserNMI().
211 */
212 void kvmNonMaskableInterrupt();
213
214 /**
215 * Send a normal interrupt to the guest
216 *
217 * @note Make sure that ready_for_interrupt_injection in kvm_run
218 * is set prior to calling this function. If not, an interrupt
219 * window must be requested by setting request_interrupt_window in
220 * kvm_run to 1 and restarting the guest.
221 *
222 * @param interrupt Structure describing the interrupt to send
223 */
224 void kvmInterrupt(const struct kvm_interrupt &interrupt);
225
226 /** @} */
227
228 /** @{ */
229 /**
230 * Get/Set the register state of the guest vCPU
231 *
232 * KVM has two different interfaces for accessing the state of the
233 * guest CPU. One interface updates 'normal' registers and one
234 * updates 'special' registers. The distinction between special
235 * and normal registers isn't very clear and is architecture
236 * dependent.
237 */
238 void getRegisters(struct kvm_regs &regs) const;
239 void setRegisters(const struct kvm_regs &regs);
240 void getSpecialRegisters(struct kvm_sregs &regs) const;
241 void setSpecialRegisters(const struct kvm_sregs &regs);
242 /** @} */
243
244 /** @{ */
245 /**
246 * Get/Set the guest FPU/vector state
247 */
248 void getFPUState(struct kvm_fpu &state) const;
249 void setFPUState(const struct kvm_fpu &state);
250 /** @} */
251
252 /** @{ */
253 /**
254 * Get/Set single register using the KVM_(SET|GET)_ONE_REG API.
255 *
256 * @note The presence of this call depends on Kvm::capOneReg().
257 */
258 void setOneReg(uint64_t id, const void *addr);
259 void setOneReg(uint64_t id, uint64_t value) { setOneReg(id, &value); }
260 void setOneReg(uint64_t id, uint32_t value) { setOneReg(id, &value); }
261 void getOneReg(uint64_t id, void *addr) const;
262 uint64_t getOneRegU64(uint64_t id) const {
263 uint64_t value;
264 getOneReg(id, &value);
265 return value;
266 }
267 uint32_t getOneRegU32(uint64_t id) const {
268 uint32_t value;
269 getOneReg(id, &value);
270 return value;
271 }
272 /** @} */
273
274 /**
275 * Get and format one register for printout.
276 *
277 * This function call getOneReg() to retrieve the contents of one
278 * register and automatically formats it for printing.
279 *
280 * @note The presence of this call depends on Kvm::capOneReg().
281 */
282 std::string getAndFormatOneReg(uint64_t id) const;
283
284 /** @{ */
285 /**
286 * Update the KVM state from the current thread context
287 *
288 * The base CPU calls this method before starting the guest CPU
289 * when the contextDirty flag is set. The architecture dependent
290 * CPU implementation is expected to update all guest state
291 * (registers, special registers, and FPU state).
292 */
293 virtual void updateKvmState() = 0;
294
295 /**
296 * Update the current thread context with the KVM state
297 *
298 * The base CPU after the guest updates any of the KVM state. In
299 * practice, this happens after kvmRun is called. The architecture
300 * dependent code is expected to read the state of the guest CPU
301 * and update gem5's thread state.
302 */
303 virtual void updateThreadContext() = 0;
304
305 /**
306 * Update a thread context if the KVM state is dirty with respect
307 * to the cached thread context.
308 */
309 void syncThreadContext();
310
311 /**
312 * Update the KVM if the thread context is dirty.
313 */
314 void syncKvmState();
315 /** @} */
316
317 /** @{ */
318 /**
319 * Main kvmRun exit handler, calls the relevant handleKvmExit*
320 * depending on exit type.
321 *
322 * @return Number of ticks spent servicing the exit request
323 */
324 virtual Tick handleKvmExit();
325
326 /**
327 * The guest performed a legacy IO request (out/inp on x86)
328 *
329 * @return Number of ticks spent servicing the IO request
330 */
331 virtual Tick handleKvmExitIO();
332
333 /**
334 * The guest requested a monitor service using a hypercall
335 *
336 * @return Number of ticks spent servicing the hypercall
337 */
338 virtual Tick handleKvmExitHypercall();
339
340 /**
341 * The guest exited because an interrupt window was requested
342 *
343 * The guest exited because an interrupt window was requested
344 * (request_interrupt_window in the kvm_run structure was set to 1
345 * before calling kvmRun) and it is now ready to receive
346 *
347 * @return Number of ticks spent servicing the IRQ
348 */
349 virtual Tick handleKvmExitIRQWindowOpen();
350
351 /**
352 * An unknown architecture dependent error occurred when starting
353 * the vCPU
354 *
355 * The kvm_run data structure contains the hardware error
356 * code. The defaults behavior of this method just prints the HW
357 * error code and panics. Architecture dependent implementations
358 * may want to override this method to provide better,
359 * hardware-aware, error messages.
360 *
361 * @return Number of ticks delay the next CPU tick
362 */
363 virtual Tick handleKvmExitUnknown();
364
365 /**
366 * An unhandled virtualization exception occured
367 *
368 * Some KVM virtualization drivers return unhandled exceptions to
369 * the user-space monitor. This interface is currently only used
370 * by the Intel VMX KVM driver.
371 *
372 * @return Number of ticks delay the next CPU tick
373 */
374 virtual Tick handleKvmExitException();
375
376 /**
377 * KVM failed to start the virtualized CPU
378 *
379 * The kvm_run data structure contains the hardware-specific error
380 * code.
381 *
382 * @return Number of ticks delay the next CPU tick
383 */
384 virtual Tick handleKvmExitFailEntry();
385 /** @} */
386
387 /**
388 * Inject a memory mapped IO request into gem5
389 *
390 * @param paddr Physical address
391 * @param data Pointer to the source/destination buffer
392 * @param size Memory access size
393 * @param write True if write, False if read
394 * @return Number of ticks spent servicing the memory access
395 */
396 Tick doMMIOAccess(Addr paddr, void *data, int size, bool write);
397
398
399 /**
400 * @addtogroup KvmIoctl
401 * @{
402 */
403 /**
404 * vCPU ioctl interface.
405 *
406 * @param request KVM vCPU request
407 * @param p1 Optional request parameter
408 *
409 * @return -1 on error (error number in errno), ioctl dependent
410 * value otherwise.
411 */
412 int ioctl(int request, long p1) const;
413 int ioctl(int request, void *p1) const {
414 return ioctl(request, (long)p1);
415 }
416 int ioctl(int request) const {
417 return ioctl(request, 0L);
418 }
419 /** @} */
420
421
422 /**
423 * KVM memory port. Uses the default MasterPort behavior, but
424 * panics on timing accesses.
425 */
426 class KVMCpuPort : public MasterPort
427 {
428
429 public:
430 KVMCpuPort(const std::string &_name, BaseKvmCPU *_cpu)
431 : MasterPort(_name, _cpu)
432 { }
433
434 protected:
435 bool recvTimingResp(PacketPtr pkt)
436 {
437 panic("The KVM CPU doesn't expect recvTimingResp!\n");
438 return true;
439 }
440
441 void recvRetry()
442 {
443 panic("The KVM CPU doesn't expect recvRetry!\n");
444 }
445
446 };
447
448 /** Port for data requests */
449 KVMCpuPort dataPort;
450
451 /** Unused dummy port for the instruction interface */
452 KVMCpuPort instPort;
453
454 /** Pre-allocated MMIO memory request */
455 Request mmio_req;
456
457 /**
458 * Is the gem5 context dirty? Set to true to force an update of
459 * the KVM vCPU state upon the next call to kvmRun().
460 */
461 bool threadContextDirty;
462
463 /**
464 * Is the KVM state dirty? Set to true to force an update of
465 * the KVM vCPU state upon the next call to kvmRun().
466 */
467 bool kvmStateDirty;
468
469 /** KVM internal ID of the vCPU */
470 const long vcpuID;
471
472 private:
473 struct TickEvent : public Event
474 {
475 BaseKvmCPU &cpu;
476
477 TickEvent(BaseKvmCPU &c)
478 : Event(CPU_Tick_Pri), cpu(c) {}
479
480 void process() { cpu.tick(); }
481
482 const char *description() const {
483 return "BaseKvmCPU tick";
484 }
485 };
486
487 /**
488 * Service MMIO requests in the mmioRing.
489 *
490 *
491 * @return Number of ticks spent servicing the MMIO requests in
492 * the MMIO ring buffer
493 */
494 Tick flushCoalescedMMIO();
495
496 /**
497 * Setup a signal handler to catch the timer signal used to
498 * switch back to the monitor.
499 */
500 void setupSignalHandler();
501
502 /** Setup hardware performance counters */
503 void setupCounters();
504
505 /** KVM vCPU file descriptor */
506 int vcpuFD;
507 /** Size of MMAPed kvm_run area */
508 int vcpuMMapSize;
509 /**
510 * Pointer to the kvm_run structure used to communicate parameters
511 * with KVM.
512 *
513 * @note This is the base pointer of the MMAPed KVM region. The
514 * first page contains the kvm_run structure. Subsequent pages may
515 * contain other data such as the MMIO ring buffer.
516 */
517 struct kvm_run *_kvmRun;
518 /**
519 * Coalesced MMIO ring buffer. NULL if coalesced MMIO is not
520 * supported.
521 */
522 struct kvm_coalesced_mmio_ring *mmioRing;
523 /** Cached page size of the host */
524 const long pageSize;
525
526 TickEvent tickEvent;
527
528 /** @{ */
529 /** Guest performance counters */
530 PerfKvmCounter hwCycles;
531 PerfKvmCounter hwInstructions;
532 /** @} */
533
534 /**
535 * Does the runTimer control the performance counters?
536 *
537 * The run timer will automatically enable and disable performance
538 * counters if a PerfEvent-based timer is used to control KVM
539 * exits.
540 */
541 bool perfControlledByTimer;
542
543 /**
544 * Timer used to force execution into the monitor after a
545 * specified number of simulation tick equivalents have executed
546 * in the guest. This counter generates the signal specified by
547 * KVM_TIMER_SIGNAL.
548 */
549 std::unique_ptr<BaseKvmTimer> runTimer;
550
551 float hostFactor;
552
553 public:
554 /* @{ */
555 Stats::Scalar numInsts;
556 Stats::Scalar numVMExits;
557 Stats::Scalar numMMIO;
558 Stats::Scalar numCoalescedMMIO;
559 Stats::Scalar numIO;
560 Stats::Scalar numHalt;
561 Stats::Scalar numInterrupts;
562 Stats::Scalar numHypercalls;
563 /* @} */
564
565 /** Number of instructions executed by the CPU */
566 Counter ctrInsts;
567};
568
569#endif