VirtualBox

source: vbox/trunk/src/VBox/VMM/VMMAll/GIMAllKvm.cpp@ 68226

Last change on this file since 68226 was 65792, checked in by vboxsync, 8 years ago

EM,GIM: Let EM be the only one to mess with VMCPU_FF_UNHALT and related waking up.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 16.1 KB
Line 
1/* $Id: GIMAllKvm.cpp 65792 2017-02-15 19:23:46Z vboxsync $ */
2/** @file
3 * GIM - Guest Interface Manager, KVM, All Contexts.
4 */
5
6/*
7 * Copyright (C) 2015-2016 Oracle Corporation
8 *
9 * This file is part of VirtualBox Open Source Edition (OSE), as
10 * available from http://www.virtualbox.org. This file is free software;
11 * you can redistribute it and/or modify it under the terms of the GNU
12 * General Public License (GPL) as published by the Free Software
13 * Foundation, in version 2 as it comes in the "COPYING" file of the
14 * VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15 * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16 */
17
18
19/*********************************************************************************************************************************
20* Header Files *
21*********************************************************************************************************************************/
22#define LOG_GROUP LOG_GROUP_GIM
23#include <VBox/vmm/gim.h>
24#include <VBox/vmm/hm.h>
25#include <VBox/vmm/em.h>
26#include <VBox/vmm/tm.h>
27#include <VBox/vmm/pgm.h>
28#include <VBox/vmm/pdmdev.h>
29#include <VBox/vmm/pdmapi.h>
30#include "GIMKvmInternal.h"
31#include "GIMInternal.h"
32#include <VBox/vmm/vm.h>
33
34#include <VBox/dis.h>
35#include <VBox/err.h>
36#include <VBox/sup.h>
37
38#include <iprt/asm-amd64-x86.h>
39#include <iprt/time.h>
40
41
42/**
43 * Handles the KVM hypercall.
44 *
45 * @returns Strict VBox status code.
46 * @retval VINF_SUCCESS if the hypercall succeeded (even if its operation
47 * failed).
48 * @retval VINF_GIM_R3_HYPERCALL re-start the hypercall from ring-3.
49 * @retval VERR_GIM_HYPERCALL_ACCESS_DENIED CPL is insufficient.
50 *
51 * @param pVCpu The cross context virtual CPU structure.
52 * @param pCtx Pointer to the guest-CPU context.
53 *
54 * @thread EMT(pVCpu).
55 */
56VMM_INT_DECL(VBOXSTRICTRC) gimKvmHypercall(PVMCPU pVCpu, PCPUMCTX pCtx)
57{
58 VMCPU_ASSERT_EMT(pVCpu);
59
60 PVM pVM = pVCpu->CTX_SUFF(pVM);
61 STAM_REL_COUNTER_INC(&pVM->gim.s.StatHypercalls);
62
63 /*
64 * Get the hypercall operation and arguments.
65 */
66 bool const fIs64BitMode = CPUMIsGuestIn64BitCodeEx(pCtx);
67 uint64_t uHyperOp = pCtx->rax;
68 uint64_t uHyperArg0 = pCtx->rbx;
69 uint64_t uHyperArg1 = pCtx->rcx;
70 uint64_t uHyperArg2 = pCtx->rdi;
71 uint64_t uHyperArg3 = pCtx->rsi;
72 uint64_t uHyperRet = KVM_HYPERCALL_RET_ENOSYS;
73 uint64_t uAndMask = UINT64_C(0xffffffffffffffff);
74 if (!fIs64BitMode)
75 {
76 uAndMask = UINT64_C(0xffffffff);
77 uHyperOp &= UINT64_C(0xffffffff);
78 uHyperArg0 &= UINT64_C(0xffffffff);
79 uHyperArg1 &= UINT64_C(0xffffffff);
80 uHyperArg2 &= UINT64_C(0xffffffff);
81 uHyperArg3 &= UINT64_C(0xffffffff);
82 uHyperRet &= UINT64_C(0xffffffff);
83 }
84
85 /*
86 * Verify that guest ring-0 is the one making the hypercall.
87 */
88 uint32_t uCpl = CPUMGetGuestCPL(pVCpu);
89 if (RT_UNLIKELY(uCpl))
90 {
91 pCtx->rax = KVM_HYPERCALL_RET_EPERM & uAndMask;
92 return VERR_GIM_HYPERCALL_ACCESS_DENIED;
93 }
94
95 /*
96 * Do the work.
97 */
98 int rc = VINF_SUCCESS;
99 switch (uHyperOp)
100 {
101 case KVM_HYPERCALL_OP_KICK_CPU:
102 {
103 if (uHyperArg1 < pVM->cCpus)
104 {
105 PVMCPU pVCpuDst = &pVM->aCpus[uHyperArg1]; /* ASSUMES pVCpu index == ApicId of the VCPU. */
106 EMUnhaltAndWakeUp(pVM, pVCpuDst);
107 uHyperRet = KVM_HYPERCALL_RET_SUCCESS;
108 }
109 else
110 {
111 /* Shouldn't ever happen! If it does, throw a guru, as otherwise it'll lead to deadlocks in the guest anyway! */
112 rc = VERR_GIM_HYPERCALL_FAILED;
113 }
114 break;
115 }
116
117 case KVM_HYPERCALL_OP_VAPIC_POLL_IRQ:
118 uHyperRet = KVM_HYPERCALL_RET_SUCCESS;
119 break;
120
121 default:
122 break;
123 }
124
125 /*
126 * Place the result in rax/eax.
127 */
128 pCtx->rax = uHyperRet & uAndMask;
129 return rc;
130}
131
132
133/**
134 * Returns whether the guest has configured and enabled the use of KVM's
135 * hypercall interface.
136 *
137 * @returns true if hypercalls are enabled, false otherwise.
138 * @param pVCpu The cross context virtual CPU structure.
139 */
140VMM_INT_DECL(bool) gimKvmAreHypercallsEnabled(PVMCPU pVCpu)
141{
142 NOREF(pVCpu);
143 /* KVM paravirt interface doesn't have hypercall control bits (like Hyper-V does)
144 that guests can control, i.e. hypercalls are always enabled. */
145 return true;
146}
147
148
149/**
150 * Returns whether the guest has configured and enabled the use of KVM's
151 * paravirtualized TSC.
152 *
153 * @returns true if paravirt. TSC is enabled, false otherwise.
154 * @param pVM The cross context VM structure.
155 */
156VMM_INT_DECL(bool) gimKvmIsParavirtTscEnabled(PVM pVM)
157{
158 uint32_t cCpus = pVM->cCpus;
159 for (uint32_t i = 0; i < cCpus; i++)
160 {
161 PVMCPU pVCpu = &pVM->aCpus[i];
162 PGIMKVMCPU pGimKvmCpu = &pVCpu->gim.s.u.KvmCpu;
163 if (MSR_GIM_KVM_SYSTEM_TIME_IS_ENABLED(pGimKvmCpu->u64SystemTimeMsr))
164 return true;
165 }
166 return false;
167}
168
169
170/**
171 * MSR read handler for KVM.
172 *
173 * @returns Strict VBox status code like CPUMQueryGuestMsr().
174 * @retval VINF_CPUM_R3_MSR_READ
175 * @retval VERR_CPUM_RAISE_GP_0
176 *
177 * @param pVCpu The cross context virtual CPU structure.
178 * @param idMsr The MSR being read.
179 * @param pRange The range this MSR belongs to.
180 * @param puValue Where to store the MSR value read.
181 */
182VMM_INT_DECL(VBOXSTRICTRC) gimKvmReadMsr(PVMCPU pVCpu, uint32_t idMsr, PCCPUMMSRRANGE pRange, uint64_t *puValue)
183{
184 NOREF(pRange);
185 PVM pVM = pVCpu->CTX_SUFF(pVM);
186 PGIMKVM pKvm = &pVM->gim.s.u.Kvm;
187 PGIMKVMCPU pKvmCpu = &pVCpu->gim.s.u.KvmCpu;
188
189 switch (idMsr)
190 {
191 case MSR_GIM_KVM_SYSTEM_TIME:
192 case MSR_GIM_KVM_SYSTEM_TIME_OLD:
193 *puValue = pKvmCpu->u64SystemTimeMsr;
194 return VINF_SUCCESS;
195
196 case MSR_GIM_KVM_WALL_CLOCK:
197 case MSR_GIM_KVM_WALL_CLOCK_OLD:
198 *puValue = pKvm->u64WallClockMsr;
199 return VINF_SUCCESS;
200
201 default:
202 {
203#ifdef IN_RING3
204 static uint32_t s_cTimes = 0;
205 if (s_cTimes++ < 20)
206 LogRel(("GIM: KVM: Unknown/invalid RdMsr (%#x) -> #GP(0)\n", idMsr));
207#endif
208 LogFunc(("Unknown/invalid RdMsr (%#RX32) -> #GP(0)\n", idMsr));
209 break;
210 }
211 }
212
213 return VERR_CPUM_RAISE_GP_0;
214}
215
216
217/**
218 * MSR write handler for KVM.
219 *
220 * @returns Strict VBox status code like CPUMSetGuestMsr().
221 * @retval VINF_CPUM_R3_MSR_WRITE
222 * @retval VERR_CPUM_RAISE_GP_0
223 *
224 * @param pVCpu The cross context virtual CPU structure.
225 * @param idMsr The MSR being written.
226 * @param pRange The range this MSR belongs to.
227 * @param uRawValue The raw value with the ignored bits not masked.
228 */
229VMM_INT_DECL(VBOXSTRICTRC) gimKvmWriteMsr(PVMCPU pVCpu, uint32_t idMsr, PCCPUMMSRRANGE pRange, uint64_t uRawValue)
230{
231 NOREF(pRange);
232 PVM pVM = pVCpu->CTX_SUFF(pVM);
233 PGIMKVMCPU pKvmCpu = &pVCpu->gim.s.u.KvmCpu;
234
235 switch (idMsr)
236 {
237 case MSR_GIM_KVM_SYSTEM_TIME:
238 case MSR_GIM_KVM_SYSTEM_TIME_OLD:
239 {
240 bool fEnable = RT_BOOL(uRawValue & MSR_GIM_KVM_SYSTEM_TIME_ENABLE_BIT);
241#ifdef IN_RING0
242 NOREF(fEnable); NOREF(pKvmCpu);
243 gimR0KvmUpdateSystemTime(pVM, pVCpu);
244 return VINF_CPUM_R3_MSR_WRITE;
245#elif defined(IN_RC)
246 Assert(pVM->cCpus == 1);
247 if (fEnable)
248 {
249 RTCCUINTREG fEFlags = ASMIntDisableFlags();
250 pKvmCpu->uTsc = TMCpuTickGetNoCheck(pVCpu) | UINT64_C(1);
251 pKvmCpu->uVirtNanoTS = TMVirtualGetNoCheck(pVM) | UINT64_C(1);
252 ASMSetFlags(fEFlags);
253 }
254 return VINF_CPUM_R3_MSR_WRITE;
255#else /* IN_RING3 */
256 if (!fEnable)
257 {
258 gimR3KvmDisableSystemTime(pVM);
259 pKvmCpu->u64SystemTimeMsr = uRawValue;
260 return VINF_SUCCESS;
261 }
262
263 /* Is the system-time struct. already enabled? If so, get flags that need preserving. */
264 GIMKVMSYSTEMTIME SystemTime;
265 RT_ZERO(SystemTime);
266 if ( MSR_GIM_KVM_SYSTEM_TIME_IS_ENABLED(pKvmCpu->u64SystemTimeMsr)
267 && MSR_GIM_KVM_SYSTEM_TIME_GUEST_GPA(uRawValue) == pKvmCpu->GCPhysSystemTime)
268 {
269 int rc2 = PGMPhysSimpleReadGCPhys(pVM, &SystemTime, pKvmCpu->GCPhysSystemTime, sizeof(GIMKVMSYSTEMTIME));
270 if (RT_SUCCESS(rc2))
271 pKvmCpu->fSystemTimeFlags = (SystemTime.fFlags & GIM_KVM_SYSTEM_TIME_FLAGS_GUEST_PAUSED);
272 }
273
274 /* Enable and populate the system-time struct. */
275 pKvmCpu->u64SystemTimeMsr = uRawValue;
276 pKvmCpu->GCPhysSystemTime = MSR_GIM_KVM_SYSTEM_TIME_GUEST_GPA(uRawValue);
277 pKvmCpu->u32SystemTimeVersion += 2;
278 int rc = gimR3KvmEnableSystemTime(pVM, pVCpu);
279 if (RT_FAILURE(rc))
280 {
281 pKvmCpu->u64SystemTimeMsr = 0;
282 /* We shouldn't throw a #GP(0) here for buggy guests (neither does KVM apparently), see @bugref{8627}. */
283 }
284 return VINF_SUCCESS;
285#endif
286 }
287
288 case MSR_GIM_KVM_WALL_CLOCK:
289 case MSR_GIM_KVM_WALL_CLOCK_OLD:
290 {
291#ifndef IN_RING3
292 return VINF_CPUM_R3_MSR_WRITE;
293#else
294 /* Enable the wall-clock struct. */
295 RTGCPHYS GCPhysWallClock = MSR_GIM_KVM_WALL_CLOCK_GUEST_GPA(uRawValue);
296 if (RT_LIKELY(RT_ALIGN_64(GCPhysWallClock, 4) == GCPhysWallClock))
297 {
298 int rc = gimR3KvmEnableWallClock(pVM, GCPhysWallClock);
299 if (RT_SUCCESS(rc))
300 {
301 PGIMKVM pKvm = &pVM->gim.s.u.Kvm;
302 pKvm->u64WallClockMsr = uRawValue;
303 return VINF_SUCCESS;
304 }
305 }
306 return VERR_CPUM_RAISE_GP_0;
307#endif /* IN_RING3 */
308 }
309
310 default:
311 {
312#ifdef IN_RING3
313 static uint32_t s_cTimes = 0;
314 if (s_cTimes++ < 20)
315 LogRel(("GIM: KVM: Unknown/invalid WrMsr (%#x,%#x`%08x) -> #GP(0)\n", idMsr,
316 uRawValue & UINT64_C(0xffffffff00000000), uRawValue & UINT64_C(0xffffffff)));
317#endif
318 LogFunc(("Unknown/invalid WrMsr (%#RX32,%#RX64) -> #GP(0)\n", idMsr, uRawValue));
319 break;
320 }
321 }
322
323 return VERR_CPUM_RAISE_GP_0;
324}
325
326
327/**
328 * Whether we need to trap \#UD exceptions in the guest.
329 *
330 * On AMD-V we need to trap them because paravirtualized Linux/KVM guests use
331 * the Intel VMCALL instruction to make hypercalls and we need to trap and
332 * optionally patch them to the AMD-V VMMCALL instruction and handle the
333 * hypercall.
334 *
335 * I guess this was done so that guest teleporation between an AMD and an Intel
336 * machine would working without any changes at the time of teleporation.
337 * However, this also means we -always- need to intercept \#UD exceptions on one
338 * of the two CPU models (Intel or AMD). Hyper-V solves this problem more
339 * elegantly by letting the hypervisor supply an opaque hypercall page.
340 *
341 * For raw-mode VMs, this function will always return true. See gimR3KvmInit().
342 *
343 * @param pVCpu The cross context virtual CPU structure.
344 */
345VMM_INT_DECL(bool) gimKvmShouldTrapXcptUD(PVMCPU pVCpu)
346{
347 PVM pVM = pVCpu->CTX_SUFF(pVM);
348 return pVM->gim.s.u.Kvm.fTrapXcptUD;
349}
350
351
352/**
353 * Checks the currently disassembled instruction and executes the hypercall if
354 * it's a hypercall instruction.
355 *
356 * @returns Strict VBox status code.
357 * @param pVCpu The cross context virtual CPU structure.
358 * @param pCtx Pointer to the guest-CPU context.
359 * @param pDis Pointer to the disassembled instruction state at RIP.
360 *
361 * @thread EMT(pVCpu).
362 *
363 * @todo Make this function static when @bugref{7270#c168} is addressed.
364 */
365VMM_INT_DECL(VBOXSTRICTRC) gimKvmExecHypercallInstr(PVMCPU pVCpu, PCPUMCTX pCtx, PDISCPUSTATE pDis)
366{
367 Assert(pVCpu);
368 Assert(pCtx);
369 Assert(pDis);
370 VMCPU_ASSERT_EMT(pVCpu);
371
372 /*
373 * If the instruction at RIP is the Intel VMCALL instruction or
374 * the AMD VMMCALL instruction handle it as a hypercall.
375 *
376 * Linux/KVM guests always uses the Intel VMCALL instruction but we patch
377 * it to the host-native one whenever we encounter it so subsequent calls
378 * will not require disassembly (when coming from HM).
379 */
380 if ( pDis->pCurInstr->uOpcode == OP_VMCALL
381 || pDis->pCurInstr->uOpcode == OP_VMMCALL)
382 {
383 /*
384 * Perform the hypercall.
385 *
386 * For HM, we can simply resume guest execution without performing the hypercall now and
387 * do it on the next VMCALL/VMMCALL exit handler on the patched instruction.
388 *
389 * For raw-mode we need to do this now anyway. So we do it here regardless with an added
390 * advantage is that it saves one world-switch for the HM case.
391 */
392 VBOXSTRICTRC rcStrict = gimKvmHypercall(pVCpu, pCtx);
393 if (rcStrict == VINF_SUCCESS)
394 {
395 /*
396 * Patch the instruction to so we don't have to spend time disassembling it each time.
397 * Makes sense only for HM as with raw-mode we will be getting a #UD regardless.
398 */
399 PVM pVM = pVCpu->CTX_SUFF(pVM);
400 PCGIMKVM pKvm = &pVM->gim.s.u.Kvm;
401 if ( pDis->pCurInstr->uOpcode != pKvm->uOpCodeNative
402 && HMIsEnabled(pVM))
403 {
404 /** @todo r=ramshankar: we probably should be doing this in an
405 * EMT rendezvous. */
406 uint8_t abHypercall[3];
407 size_t cbWritten = 0;
408 int rc = VMMPatchHypercall(pVM, &abHypercall, sizeof(abHypercall), &cbWritten);
409 AssertRC(rc);
410 Assert(sizeof(abHypercall) == pDis->cbInstr);
411 Assert(sizeof(abHypercall) == cbWritten);
412
413 rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->rip, &abHypercall, sizeof(abHypercall));
414 AssertRC(rc);
415
416 /** @todo Add stats for patching. */
417 }
418 }
419 else
420 {
421 /* The KVM provider doesn't have any concept of continuing hypercalls. */
422 Assert(rcStrict != VINF_GIM_HYPERCALL_CONTINUING);
423#ifdef IN_RING3
424 Assert(rcStrict != VINF_GIM_R3_HYPERCALL);
425#endif
426 }
427 return rcStrict;
428 }
429
430 return VERR_GIM_INVALID_HYPERCALL_INSTR;
431}
432
433
434/**
435 * Exception handler for \#UD.
436 *
437 * @returns Strict VBox status code.
438 * @retval VINF_SUCCESS if the hypercall succeeded (even if its operation
439 * failed).
440 * @retval VINF_GIM_R3_HYPERCALL re-start the hypercall from ring-3.
441 * @retval VERR_GIM_HYPERCALL_ACCESS_DENIED CPL is insufficient.
442 * @retval VERR_GIM_INVALID_HYPERCALL_INSTR instruction at RIP is not a valid
443 * hypercall instruction.
444 *
445 * @param pVCpu The cross context virtual CPU structure.
446 * @param pCtx Pointer to the guest-CPU context.
447 * @param pDis Pointer to the disassembled instruction state at RIP.
448 * Optional, can be NULL.
449 * @param pcbInstr Where to store the instruction length of the hypercall
450 * instruction. Optional, can be NULL.
451 *
452 * @thread EMT(pVCpu).
453 */
454VMM_INT_DECL(VBOXSTRICTRC) gimKvmXcptUD(PVMCPU pVCpu, PCPUMCTX pCtx, PDISCPUSTATE pDis, uint8_t *pcbInstr)
455{
456 VMCPU_ASSERT_EMT(pVCpu);
457
458 /*
459 * If we didn't ask for #UD to be trapped, bail.
460 */
461 PVM pVM = pVCpu->CTX_SUFF(pVM);
462 PCGIMKVM pKvm = &pVM->gim.s.u.Kvm;
463 if (RT_UNLIKELY(!pKvm->fTrapXcptUD))
464 return VERR_GIM_IPE_3;
465
466 if (!pDis)
467 {
468 unsigned cbInstr;
469 DISCPUSTATE Dis;
470 int rc = EMInterpretDisasCurrent(pVM, pVCpu, &Dis, &cbInstr);
471 if (RT_SUCCESS(rc))
472 {
473 if (pcbInstr)
474 *pcbInstr = (uint8_t)cbInstr;
475 return gimKvmExecHypercallInstr(pVCpu, pCtx, &Dis);
476 }
477
478 Log(("GIM: KVM: Failed to disassemble instruction at CS:RIP=%04x:%08RX64. rc=%Rrc\n", pCtx->cs.Sel, pCtx->rip, rc));
479 return rc;
480 }
481
482 return gimKvmExecHypercallInstr(pVCpu, pCtx, pDis);
483}
484
Note: See TracBrowser for help on using the repository browser.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette