GIMAllKvm.cpp@ 59747

Last change on this file since 59747 was 58123, checked in by vboxsync, 9 years ago
VMM: Made @param pVCpu more uniform and to the point.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 14.2 KB

Line
1	/* $Id: GIMAllKvm.cpp 58123 2015-10-08 18:09:45Z vboxsync $ */
2	/** @file
3	* GIM - Guest Interface Manager, KVM, All Contexts.
4	*/
5
6	/*
7	* Copyright (C) 2015 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 "GIMKvmInternal.h"
24	#include "GIMInternal.h"
25
26	#include <VBox/err.h>
27	#include <VBox/dis.h>
28	#include <VBox/vmm/hm.h>
29	#include <VBox/vmm/em.h>
30	#include <VBox/vmm/tm.h>
31	#include <VBox/vmm/vm.h>
32	#include <VBox/vmm/pgm.h>
33	#include <VBox/vmm/pdmdev.h>
34	#include <VBox/vmm/pdmapi.h>
35	#include <VBox/sup.h>
36
37	#include <iprt/asm-amd64-x86.h>
38	#include <iprt/time.h>
39
40
41	/**
42	* Handles the KVM hypercall.
43	*
44	* @returns VBox status code.
45	* @param pVCpu The cross context virtual CPU structure.
46	* @param pCtx Pointer to the guest-CPU context.
47	*
48	* @thread EMT.
49	*/
50	VMM_INT_DECL(int) gimKvmHypercall(PVMCPU pVCpu, PCPUMCTX pCtx)
51	{
52	/*
53	* Get the hypercall operation and arguments.
54	*/
55	bool const fIs64BitMode = CPUMIsGuestIn64BitCodeEx(pCtx);
56	uint64_t uHyperOp = pCtx->rax;
57	uint64_t uHyperArg0 = pCtx->rbx;
58	uint64_t uHyperArg1 = pCtx->rcx;
59	uint64_t uHyperArg2 = pCtx->rdi;
60	uint64_t uHyperArg3 = pCtx->rsi;
61	uint64_t uHyperRet = KVM_HYPERCALL_RET_ENOSYS;
62	uint64_t uAndMask = UINT64_C(0xffffffffffffffff);
63	if (!fIs64BitMode)
64	{
65	uAndMask = UINT64_C(0xffffffff);
66	uHyperOp &= UINT64_C(0xffffffff);
67	uHyperArg0 &= UINT64_C(0xffffffff);
68	uHyperArg1 &= UINT64_C(0xffffffff);
69	uHyperArg2 &= UINT64_C(0xffffffff);
70	uHyperArg3 &= UINT64_C(0xffffffff);
71	uHyperRet &= UINT64_C(0xffffffff);
72	}
73
74	/*
75	* Verify that guest ring-0 is the one making the hypercall.
76	*/
77	uint32_t uCpl = CPUMGetGuestCPL(pVCpu);
78	if (uCpl)
79	{
80	pCtx->rax = KVM_HYPERCALL_RET_EPERM & uAndMask;
81	return VINF_SUCCESS;
82	}
83
84	/*
85	* Do the work.
86	*/
87	switch (uHyperOp)
88	{
89	case KVM_HYPERCALL_OP_KICK_CPU:
90	{
91	PVM pVM = pVCpu->CTX_SUFF(pVM);
92	if (uHyperArg1 < pVM->cCpus)
93	{
94	PVMCPU pVCpuTarget = &pVM->aCpus[uHyperArg1]; /** ASSUMES pVCpu index == ApicId of the VCPU. */
95	VMCPU_FF_SET(pVCpuTarget, VMCPU_FF_UNHALT);
96	#ifdef IN_RING0
97	/*
98	* We might be here with preemption disabled or enabled (i.e. depending on thread-context hooks
99	* being used), so don't try obtaining the GVMMR0 used lock here. See @bugref{7270#c148}.
100	*/
101	GVMMR0SchedWakeUpEx(pVM, pVCpuTarget->idCpu, false /* fTakeUsedLock */);
102	#elif defined(IN_RING3)
103	int rc2 = SUPR3CallVMMR0(pVM->pVMR0, pVCpuTarget->idCpu, VMMR0_DO_GVMM_SCHED_WAKE_UP, NULL);
104	AssertRC(rc2);
105	#elif defined(IN_RC)
106	/* Nothing to do for raw-mode, shouldn't really be used by raw-mode guests anyway. */
107	Assert(pVM->cCpus == 1);
108	#endif
109	uHyperRet = KVM_HYPERCALL_RET_SUCCESS;
110	}
111	break;
112	}
113
114	case KVM_HYPERCALL_OP_VAPIC_POLL_IRQ:
115	uHyperRet = KVM_HYPERCALL_RET_SUCCESS;
116	break;
117
118	default:
119	break;
120	}
121
122	/*
123	* Place the result in rax/eax.
124	*/
125	pCtx->rax = uHyperRet & uAndMask;
126	return VINF_SUCCESS;
127	}
128
129
130	/**
131	* Returns whether the guest has configured and enabled the use of KVM's
132	* hypercall interface.
133	*
134	* @returns true if hypercalls are enabled, false otherwise.
135	* @param pVCpu The cross context virtual CPU structure.
136	*/
137	VMM_INT_DECL(bool) gimKvmAreHypercallsEnabled(PVMCPU pVCpu)
138	{
139	NOREF(pVCpu);
140	/* KVM paravirt interface doesn't have hypercall control bits (like Hyper-V does)
141	that guests can control, i.e. hypercalls are always enabled. */
142	return true;
143	}
144
145
146	/**
147	* Returns whether the guest has configured and enabled the use of KVM's
148	* paravirtualized TSC.
149	*
150	* @returns true if paravirt. TSC is enabled, false otherwise.
151	* @param pVM The cross context VM structure.
152	*/
153	VMM_INT_DECL(bool) gimKvmIsParavirtTscEnabled(PVM pVM)
154	{
155	uint32_t cCpus = pVM->cCpus;
156	for (uint32_t i = 0; i < cCpus; i++)
157	{
158	PVMCPU pVCpu = &pVM->aCpus[i];
159	PGIMKVMCPU pGimKvmCpu = &pVCpu->gim.s.u.KvmCpu;
160	if (MSR_GIM_KVM_SYSTEM_TIME_IS_ENABLED(pGimKvmCpu->u64SystemTimeMsr))
161	return true;
162	}
163	return false;
164	}
165
166
167	/**
168	* MSR read handler for KVM.
169	*
170	* @returns Strict VBox status code like CPUMQueryGuestMsr().
171	* @retval VINF_CPUM_R3_MSR_READ
172	* @retval VERR_CPUM_RAISE_GP_0
173	*
174	* @param pVCpu The cross context virtual CPU structure.
175	* @param idMsr The MSR being read.
176	* @param pRange The range this MSR belongs to.
177	* @param puValue Where to store the MSR value read.
178	*/
179	VMM_INT_DECL(VBOXSTRICTRC) gimKvmReadMsr(PVMCPU pVCpu, uint32_t idMsr, PCCPUMMSRRANGE pRange, uint64_t *puValue)
180	{
181	NOREF(pRange);
182	PVM pVM = pVCpu->CTX_SUFF(pVM);
183	PGIMKVM pKvm = &pVM->gim.s.u.Kvm;
184	PGIMKVMCPU pKvmCpu = &pVCpu->gim.s.u.KvmCpu;
185
186	switch (idMsr)
187	{
188	case MSR_GIM_KVM_SYSTEM_TIME:
189	case MSR_GIM_KVM_SYSTEM_TIME_OLD:
190	*puValue = pKvmCpu->u64SystemTimeMsr;
191	return VINF_SUCCESS;
192
193	case MSR_GIM_KVM_WALL_CLOCK:
194	case MSR_GIM_KVM_WALL_CLOCK_OLD:
195	*puValue = pKvm->u64WallClockMsr;
196	return VINF_SUCCESS;
197
198	default:
199	{
200	#ifdef IN_RING3
201	static uint32_t s_cTimes = 0;
202	if (s_cTimes++ < 20)
203	LogRel(("GIM: KVM: Unknown/invalid RdMsr (%#x) -> #GP(0)\n", idMsr));
204	#endif
205	LogFunc(("Unknown/invalid RdMsr (%#RX32) -> #GP(0)\n", idMsr));
206	break;
207	}
208	}
209
210	return VERR_CPUM_RAISE_GP_0;
211	}
212
213
214	/**
215	* MSR write handler for KVM.
216	*
217	* @returns Strict VBox status code like CPUMSetGuestMsr().
218	* @retval VINF_CPUM_R3_MSR_WRITE
219	* @retval VERR_CPUM_RAISE_GP_0
220	*
221	* @param pVCpu The cross context virtual CPU structure.
222	* @param idMsr The MSR being written.
223	* @param pRange The range this MSR belongs to.
224	* @param uRawValue The raw value with the ignored bits not masked.
225	*/
226	VMM_INT_DECL(VBOXSTRICTRC) gimKvmWriteMsr(PVMCPU pVCpu, uint32_t idMsr, PCCPUMMSRRANGE pRange, uint64_t uRawValue)
227	{
228	NOREF(pRange);
229	PVM pVM = pVCpu->CTX_SUFF(pVM);
230	PGIMKVMCPU pKvmCpu = &pVCpu->gim.s.u.KvmCpu;
231
232	switch (idMsr)
233	{
234	case MSR_GIM_KVM_SYSTEM_TIME:
235	case MSR_GIM_KVM_SYSTEM_TIME_OLD:
236	{
237	bool fEnable = RT_BOOL(uRawValue & MSR_GIM_KVM_SYSTEM_TIME_ENABLE_BIT);
238	#ifdef IN_RING0
239	NOREF(fEnable); NOREF(pKvmCpu);
240	gimR0KvmUpdateSystemTime(pVM, pVCpu);
241	return VINF_CPUM_R3_MSR_WRITE;
242	#elif defined(IN_RC)
243	Assert(pVM->cCpus == 1);
244	if (fEnable)
245	{
246	RTCCUINTREG fEFlags = ASMIntDisableFlags();
247	pKvmCpu->uTsc = TMCpuTickGetNoCheck(pVCpu) \| UINT64_C(1);
248	pKvmCpu->uVirtNanoTS = TMVirtualGetNoCheck(pVM) \| UINT64_C(1);
249	ASMSetFlags(fEFlags);
250	}
251	return VINF_CPUM_R3_MSR_WRITE;
252	#else /* IN_RING3 */
253	if (!fEnable)
254	{
255	gimR3KvmDisableSystemTime(pVM);
256	pKvmCpu->u64SystemTimeMsr = uRawValue;
257	return VINF_SUCCESS;
258	}
259
260	/* Is the system-time struct. already enabled? If so, get flags that need preserving. */
261	uint8_t fFlags = 0;
262	GIMKVMSYSTEMTIME SystemTime;
263	RT_ZERO(SystemTime);
264	if ( MSR_GIM_KVM_SYSTEM_TIME_IS_ENABLED(pKvmCpu->u64SystemTimeMsr)
265	&& MSR_GIM_KVM_SYSTEM_TIME_GUEST_GPA(uRawValue) == pKvmCpu->GCPhysSystemTime)
266	{
267	int rc2 = PGMPhysSimpleReadGCPhys(pVM, &SystemTime, pKvmCpu->GCPhysSystemTime, sizeof(GIMKVMSYSTEMTIME));
268	if (RT_SUCCESS(rc2))
269	pKvmCpu->fSystemTimeFlags = (SystemTime.fFlags & GIM_KVM_SYSTEM_TIME_FLAGS_GUEST_PAUSED);
270	}
271
272	/* Enable and populate the system-time struct. */
273	pKvmCpu->u64SystemTimeMsr = uRawValue;
274	pKvmCpu->GCPhysSystemTime = MSR_GIM_KVM_SYSTEM_TIME_GUEST_GPA(uRawValue);
275	pKvmCpu->u32SystemTimeVersion += 2;
276	int rc = gimR3KvmEnableSystemTime(pVM, pVCpu);
277	if (RT_FAILURE(rc))
278	{
279	pKvmCpu->u64SystemTimeMsr = 0;
280	return VERR_CPUM_RAISE_GP_0;
281	}
282	return VINF_SUCCESS;
283	#endif
284	}
285
286	case MSR_GIM_KVM_WALL_CLOCK:
287	case MSR_GIM_KVM_WALL_CLOCK_OLD:
288	{
289	#ifndef IN_RING3
290	return VINF_CPUM_R3_MSR_WRITE;
291	#else
292	/* Enable the wall-clock struct. */
293	RTGCPHYS GCPhysWallClock = MSR_GIM_KVM_WALL_CLOCK_GUEST_GPA(uRawValue);
294	if (RT_LIKELY(RT_ALIGN_64(GCPhysWallClock, 4) == GCPhysWallClock))
295	{
296	int rc = gimR3KvmEnableWallClock(pVM, GCPhysWallClock);
297	if (RT_SUCCESS(rc))
298	{
299	PGIMKVM pKvm = &pVM->gim.s.u.Kvm;
300	pKvm->u64WallClockMsr = uRawValue;
301	return VINF_SUCCESS;
302	}
303	}
304	return VERR_CPUM_RAISE_GP_0;
305	#endif /* IN_RING3 */
306	}
307
308	default:
309	{
310	#ifdef IN_RING3
311	static uint32_t s_cTimes = 0;
312	if (s_cTimes++ < 20)
313	LogRel(("GIM: KVM: Unknown/invalid WrMsr (%#x,%#x`%08x) -> #GP(0)\n", idMsr,
314	uRawValue & UINT64_C(0xffffffff00000000), uRawValue & UINT64_C(0xffffffff)));
315	#endif
316	LogFunc(("Unknown/invalid WrMsr (%#RX32,%#RX64) -> #GP(0)\n", idMsr, uRawValue));
317	break;
318	}
319	}
320
321	return VERR_CPUM_RAISE_GP_0;
322	}
323
324
325	/**
326	* Whether we need to trap \#UD exceptions in the guest.
327	*
328	* On AMD-V we need to trap them because paravirtualized Linux/KVM guests use
329	* the Intel VMCALL instruction to make hypercalls and we need to trap and
330	* optionally patch them to the AMD-V VMMCALL instruction and handle the
331	* hypercall.
332	*
333	* I guess this was done so that guest teleporation between an AMD and an Intel
334	* machine would working without any changes at the time of teleporation.
335	* However, this also means we -always- need to intercept \#UD exceptions on one
336	* of the two CPU models (Intel or AMD). Hyper-V solves this problem more
337	* elegantly by letting the hypervisor supply an opaque hypercall page.
338	*
339	* For raw-mode VMs, this function will always return true. See gimR3KvmInit().
340	*
341	* @param pVCpu The cross context virtual CPU structure.
342	*/
343	VMM_INT_DECL(bool) gimKvmShouldTrapXcptUD(PVMCPU pVCpu)
344	{
345	PVM pVM = pVCpu->CTX_SUFF(pVM);
346	return pVM->gim.s.u.Kvm.fTrapXcptUD;
347	}
348
349
350	/**
351	* Exception handler for \#UD.
352	*
353	* @param pVCpu The cross context virtual CPU structure.
354	* @param pCtx Pointer to the guest-CPU context.
355	* @param pDis Pointer to the disassembled instruction state at RIP.
356	* Optional, can be NULL.
357	*
358	* @thread EMT.
359	*/
360	VMM_INT_DECL(int) gimKvmXcptUD(PVMCPU pVCpu, PCPUMCTX pCtx, PDISCPUSTATE pDis)
361	{
362	/*
363	* If we didn't ask for #UD to be trapped, bail.
364	*/
365	PVM pVM = pVCpu->CTX_SUFF(pVM);
366	PGIMKVM pKvm = &pVM->gim.s.u.Kvm;
367	if (RT_UNLIKELY(!pVM->gim.s.u.Kvm.fTrapXcptUD))
368	return VERR_GIM_OPERATION_FAILED;
369
370	int rc = VINF_SUCCESS;
371	if (!pDis)
372	{
373	/*
374	* Disassemble the instruction at RIP to figure out if it's the Intel VMCALL instruction
375	* or the AMD VMMCALL instruction and if so, handle it as a hypercall.
376	*/
377	DISCPUSTATE Dis;
378	rc = EMInterpretDisasCurrent(pVM, pVCpu, &Dis, NULL /* pcbInstr */);
379	pDis = &Dis;
380	}
381
382	if (RT_SUCCESS(rc))
383	{
384	/*
385	* Patch the instruction to so we don't have to spend time disassembling it each time.
386	* Makes sense only for HM as with raw-mode we will be getting a #UD regardless.
387	*/
388	if ( pDis->pCurInstr->uOpcode == OP_VMCALL
389	\|\| pDis->pCurInstr->uOpcode == OP_VMMCALL)
390	{
391	/*
392	* Make sure guest ring-0 is the one making the hypercall.
393	*/
394	if (CPUMGetGuestCPL(pVCpu))
395	return VERR_GIM_HYPERCALL_ACCESS_DENIED;
396
397	if ( pDis->pCurInstr->uOpcode != pKvm->uOpCodeNative
398	&& HMIsEnabled(pVM))
399	{
400	uint8_t abHypercall[3];
401	size_t cbWritten = 0;
402	rc = VMMPatchHypercall(pVM, &abHypercall, sizeof(abHypercall), &cbWritten);
403	AssertRC(rc);
404	Assert(sizeof(abHypercall) == pDis->cbInstr);
405	Assert(sizeof(abHypercall) == cbWritten);
406
407	rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->rip, &abHypercall, sizeof(abHypercall));
408	}
409
410	/*
411	* Update RIP and perform the hypercall.
412	*
413	* For HM, we can simply resume guest execution without performing the hypercall now and
414	* do it on the next VMCALL/VMMCALL exit handler on the patched instruction.
415	*
416	* For raw-mode we need to do this now anyway. So we do it here regardless with an added
417	* advantage is that it saves one world-switch for the HM case.
418	*/
419	if (RT_SUCCESS(rc))
420	{
421	pCtx->rip += pDis->cbInstr;
422	rc = gimKvmHypercall(pVCpu, pCtx);
423	}
424	}
425	else
426	rc = VERR_GIM_OPERATION_FAILED;
427	}
428	return rc;
429	}
430

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/VMM/VMMAll/GIMAllKvm.cpp@ 59747

Download in other formats: