Virtio.cpp@ 86997

Last change on this file since 86997 was 82968, checked in by vboxsync, 5 years ago
Copyright year updates by scm.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 34.5 KB

Line
1	/* $Id: Virtio.cpp 82968 2020-02-04 10:35:17Z vboxsync $ */
2	/** @file
3	* Virtio - Virtio Common Functions (VRing, VQueue, Virtio PCI)
4	*/
5
6	/*
7	* Copyright (C) 2009-2020 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_DEV_VIRTIO
23
24	#include <iprt/param.h>
25	#include <iprt/uuid.h>
26	#include <VBox/vmm/pdmdev.h>
27	#include <VBox/AssertGuest.h>
28	#include "Virtio.h"
29
30
31	/*********************************************************************************************************************************
32	* Defined Constants And Macros *
33	*********************************************************************************************************************************/
34	#define INSTANCE(pThis) (pThis->szInstance)
35
36
37	static void vqueueReset(PVQUEUE pQueue)
38	{
39	pQueue->VRing.addrDescriptors = 0;
40	pQueue->VRing.addrAvail = 0;
41	pQueue->VRing.addrUsed = 0;
42	pQueue->uNextAvailIndex = 0;
43	pQueue->uNextUsedIndex = 0;
44	pQueue->uPageNumber = 0;
45	}
46
47	static void vqueueInit(PVQUEUE pQueue, uint32_t uPageNumber)
48	{
49	pQueue->VRing.addrDescriptors = (uint64_t)uPageNumber << PAGE_SHIFT;
50	pQueue->VRing.addrAvail = pQueue->VRing.addrDescriptors + sizeof(VRINGDESC) * pQueue->VRing.uSize;
51	pQueue->VRing.addrUsed = RT_ALIGN(pQueue->VRing.addrAvail + RT_UOFFSETOF_DYN(VRINGAVAIL, auRing[pQueue->VRing.uSize]),
52	PAGE_SIZE); /* The used ring must start from the next page. */
53	pQueue->uNextAvailIndex = 0;
54	pQueue->uNextUsedIndex = 0;
55	}
56
57	// void vqueueElemFree(PVQUEUEELEM pElem)
58	// {
59	// }
60
61	static void vringReadDesc(PPDMDEVINS pDevIns, PVRING pVRing, uint32_t uIndex, PVRINGDESC pDesc)
62	{
63	//Log(("%s vringReadDesc: ring=%p idx=%u\n", INSTANCE(pThis), pVRing, uIndex));
64	PDMDevHlpPhysRead(pDevIns,
65	pVRing->addrDescriptors + sizeof(VRINGDESC) * (uIndex % pVRing->uSize),
66	pDesc, sizeof(VRINGDESC));
67	/** @todo r=bird: Why exactly are we sometimes using PDMDevHlpPhysRead rather
68	* than PDMDevHlpPCIPhysRead? */
69	}
70
71	static uint16_t vringReadAvail(PPDMDEVINS pDevIns, PVRING pVRing, uint32_t uIndex)
72	{
73	uint16_t tmp = 0;
74	PDMDevHlpPhysRead(pDevIns, pVRing->addrAvail + RT_UOFFSETOF_DYN(VRINGAVAIL, auRing[uIndex % pVRing->uSize]),
75	&tmp, sizeof(tmp));
76	return tmp;
77	}
78
79	static uint16_t vringReadAvailFlags(PPDMDEVINS pDevIns, PVRING pVRing)
80	{
81	uint16_t tmp = 0;
82	PDMDevHlpPhysRead(pDevIns, pVRing->addrAvail + RT_UOFFSETOF(VRINGAVAIL, uFlags), &tmp, sizeof(tmp));
83	return tmp;
84	}
85
86	void vringSetNotification(PPDMDEVINS pDevIns, PVRING pVRing, bool fEnabled)
87	{
88	uint16_t fState = 0;
89	PDMDevHlpPhysRead(pDevIns, pVRing->addrUsed + RT_UOFFSETOF(VRINGUSED, uFlags), &fState, sizeof(fState));
90
91	if (fEnabled)
92	fState &= ~ VRINGUSED_F_NO_NOTIFY;
93	else
94	fState \|= VRINGUSED_F_NO_NOTIFY;
95
96	PDMDevHlpPCIPhysWrite(pDevIns, pVRing->addrUsed + RT_UOFFSETOF(VRINGUSED, uFlags), &fState, sizeof(fState));
97	}
98
99	bool vqueueSkip(PPDMDEVINS pDevIns, PVPCISTATE pThis, PVQUEUE pQueue)
100	{
101	if (vqueueIsEmpty(pDevIns, pQueue))
102	return false;
103
104	Log2(("%s vqueueSkip: %s avail_idx=%u\n", INSTANCE(pThis), pQueue->szName, pQueue->uNextAvailIndex));
105	RT_NOREF(pThis);
106	pQueue->uNextAvailIndex++;
107	return true;
108	}
109
110	bool vqueueGet(PPDMDEVINS pDevIns, PVPCISTATE pThis, PVQUEUE pQueue, PVQUEUEELEM pElem, bool fRemove)
111	{
112	if (vqueueIsEmpty(pDevIns, pQueue))
113	return false;
114
115	pElem->cIn = pElem->cOut = 0;
116
117	Log2(("%s vqueueGet: %s avail_idx=%u\n", INSTANCE(pThis), pQueue->szName, pQueue->uNextAvailIndex));
118
119	VRINGDESC desc;
120	uint16_t idx = vringReadAvail(pDevIns, &pQueue->VRing, pQueue->uNextAvailIndex);
121	if (fRemove)
122	pQueue->uNextAvailIndex++;
123	pElem->uIndex = idx;
124	do
125	{
126	VQUEUESEG *pSeg;
127
128	/*
129	* Malicious guests may try to trick us into writing beyond aSegsIn or
130	* aSegsOut boundaries by linking several descriptors into a loop. We
131	* cannot possibly get a sequence of linked descriptors exceeding the
132	* total number of descriptors in the ring (see @bugref{8620}).
133	*/
134	if (pElem->cIn + pElem->cOut >= VRING_MAX_SIZE)
135	{
136	static volatile uint32_t s_cMessages = 0;
137	static volatile uint32_t s_cThreshold = 1;
138	if (ASMAtomicIncU32(&s_cMessages) == ASMAtomicReadU32(&s_cThreshold))
139	{
140	LogRel(("%s: too many linked descriptors; check if the guest arranges descriptors in a loop.\n",
141	INSTANCE(pThis)));
142	if (ASMAtomicReadU32(&s_cMessages) != 1)
143	LogRel(("%s: (the above error has occured %u times so far)\n",
144	INSTANCE(pThis), ASMAtomicReadU32(&s_cMessages)));
145	ASMAtomicWriteU32(&s_cThreshold, ASMAtomicReadU32(&s_cThreshold) * 10);
146	}
147	break;
148	}
149	RT_UNTRUSTED_VALIDATED_FENCE();
150
151	vringReadDesc(pDevIns, &pQueue->VRing, idx, &desc);
152	if (desc.u16Flags & VRINGDESC_F_WRITE)
153	{
154	Log2(("%s vqueueGet: %s IN seg=%u desc_idx=%u addr=%p cb=%u\n", INSTANCE(pThis),
155	pQueue->szName, pElem->cIn, idx, desc.u64Addr, desc.uLen));
156	pSeg = &pElem->aSegsIn[pElem->cIn++];
157	}
158	else
159	{
160	Log2(("%s vqueueGet: %s OUT seg=%u desc_idx=%u addr=%p cb=%u\n", INSTANCE(pThis),
161	pQueue->szName, pElem->cOut, idx, desc.u64Addr, desc.uLen));
162	pSeg = &pElem->aSegsOut[pElem->cOut++];
163	}
164
165	pSeg->addr = desc.u64Addr;
166	pSeg->cb = desc.uLen;
167	pSeg->pv = NULL;
168
169	idx = desc.u16Next;
170	} while (desc.u16Flags & VRINGDESC_F_NEXT);
171
172	Log2(("%s vqueueGet: %s head_desc_idx=%u nIn=%u nOut=%u\n", INSTANCE(pThis),
173	pQueue->szName, pElem->uIndex, pElem->cIn, pElem->cOut));
174	return true;
175	}
176
177	#ifdef LOG_ENABLED
178	static uint16_t vringReadUsedIndex(PPDMDEVINS pDevIns, PVRING pVRing)
179	{
180	uint16_t tmp = 0;
181	PDMDevHlpPhysRead(pDevIns, pVRing->addrUsed + RT_UOFFSETOF(VRINGUSED, uIndex), &tmp, sizeof(tmp));
182	return tmp;
183	}
184	#endif
185
186	static void vringWriteUsedIndex(PPDMDEVINS pDevIns, PVRING pVRing, uint16_t u16Value)
187	{
188	PDMDevHlpPCIPhysWrite(pDevIns,
189	pVRing->addrUsed + RT_UOFFSETOF(VRINGUSED, uIndex),
190	&u16Value, sizeof(u16Value));
191	}
192
193	static void vringWriteUsedElem(PPDMDEVINS pDevIns, PVRING pVRing, uint32_t uIndex, uint32_t uId, uint32_t uLen)
194	{
195	VRINGUSEDELEM elem;
196
197	elem.uId = uId;
198	elem.uLen = uLen;
199	PDMDevHlpPCIPhysWrite(pDevIns,
200	pVRing->addrUsed + RT_UOFFSETOF_DYN(VRINGUSED, aRing[uIndex % pVRing->uSize]),
201	&elem, sizeof(elem));
202	}
203
204	void vqueuePut(PPDMDEVINS pDevIns, PVPCISTATE pThis, PVQUEUE pQueue, PVQUEUEELEM pElem, uint32_t uTotalLen, uint32_t uReserved)
205	{
206	Log2(("%s vqueuePut: %s desc_idx=%u acb=%u (%u)\n", INSTANCE(pThis), pQueue->szName, pElem->uIndex, uTotalLen, uReserved));
207	RT_NOREF(pThis);
208
209	Assert(uReserved < uTotalLen);
210
211	uint32_t cbLen = uTotalLen - uReserved;
212	uint32_t cbSkip = uReserved;
213
214	for (unsigned i = 0; i < pElem->cIn && cbLen > 0; ++i)
215	{
216	if (cbSkip >= pElem->aSegsIn[i].cb) /* segment completely skipped? */
217	{
218	cbSkip -= pElem->aSegsIn[i].cb;
219	continue;
220	}
221
222	uint32_t cbSegLen = pElem->aSegsIn[i].cb - cbSkip;
223	if (cbSegLen > cbLen) /* last segment only partially used? */
224	cbSegLen = cbLen;
225
226	/*
227	* XXX: We should assert pv != NULL, but we need to check and
228	* fix all callers first.
229	*/
230	if (pElem->aSegsIn[i].pv != NULL)
231	{
232	Log2(("%s vqueuePut: %s used_idx=%u seg=%u addr=%RGp pv=%p cb=%u acb=%u\n", INSTANCE(pThis), pQueue->szName,
233	pQueue->uNextUsedIndex, i, pElem->aSegsIn[i].addr, pElem->aSegsIn[i].pv, pElem->aSegsIn[i].cb, cbSegLen));
234
235	PDMDevHlpPCIPhysWrite(pDevIns,
236	pElem->aSegsIn[i].addr + cbSkip,
237	pElem->aSegsIn[i].pv,
238	cbSegLen);
239	}
240
241	cbSkip = 0;
242	cbLen -= cbSegLen;
243	}
244
245	Log2(("%s vqueuePut: %s used_idx=%u guest_used_idx=%u id=%u len=%u\n", INSTANCE(pThis), pQueue->szName,
246	pQueue->uNextUsedIndex, vringReadUsedIndex(pDevIns, &pQueue->VRing), pElem->uIndex, uTotalLen));
247
248	vringWriteUsedElem(pDevIns, &pQueue->VRing,
249	pQueue->uNextUsedIndex++,
250	pElem->uIndex, uTotalLen);
251	}
252
253	static void vqueueNotify(PPDMDEVINS pDevIns, PVPCISTATE pThis, PVQUEUE pQueue)
254	{
255	uint16_t const fAvail = vringReadAvailFlags(pDevIns, &pQueue->VRing);
256	LogFlow(("%s vqueueNotify: %s availFlags=%x guestFeatures=%x vqueue is %sempty\n", INSTANCE(pThis), pQueue->szName,
257	fAvail, pThis->uGuestFeatures, vqueueIsEmpty(pDevIns, pQueue)?"":"not "));
258	if ( !(fAvail & VRINGAVAIL_F_NO_INTERRUPT)
259	\|\| ((pThis->uGuestFeatures & VPCI_F_NOTIFY_ON_EMPTY) && vqueueIsEmpty(pDevIns, pQueue)))
260	{
261	int rc = vpciRaiseInterrupt(pDevIns, pThis, VERR_INTERNAL_ERROR, VPCI_ISR_QUEUE);
262	if (RT_FAILURE(rc))
263	Log(("%s vqueueNotify: Failed to raise an interrupt (%Rrc).\n", INSTANCE(pThis), rc));
264	}
265	else
266	STAM_REL_COUNTER_INC(&pThis->StatIntsSkipped);
267
268	}
269
270	void vqueueSync(PPDMDEVINS pDevIns, PVPCISTATE pThis, PVQUEUE pQueue)
271	{
272	Log2(("%s vqueueSync: %s old_used_idx=%u new_used_idx=%u\n", INSTANCE(pThis),
273	pQueue->szName, vringReadUsedIndex(pDevIns, &pQueue->VRing), pQueue->uNextUsedIndex));
274	vringWriteUsedIndex(pDevIns, &pQueue->VRing, pQueue->uNextUsedIndex);
275	vqueueNotify(pDevIns, pThis, pQueue);
276	}
277
278
279	/**
280	* Raise interrupt.
281	*
282	* @param pDevIns The device instance.
283	* @param pThis The shared virtio core instance data.
284	* @param rcBusy Status code to return when the critical section is busy.
285	* @param u8IntCause Interrupt cause bit mask to set in PCI ISR port.
286	*/
287	int vpciRaiseInterrupt(PPDMDEVINS pDevIns, PVPCISTATE pThis, int rcBusy, uint8_t u8IntCause)
288	{
289	RT_NOREF_PV(rcBusy);
290	// int rc = vpciCsEnter(pThis, rcBusy);
291	// if (RT_UNLIKELY(rc != VINF_SUCCESS))
292	// return rc;
293
294	STAM_REL_COUNTER_INC(&pThis->StatIntsRaised);
295	LogFlow(("%s vpciRaiseInterrupt: u8IntCause=%x\n", INSTANCE(pThis), u8IntCause));
296
297	pThis->uISR \|= u8IntCause;
298	PDMDevHlpPCISetIrq(pDevIns, 0, 1);
299	// vpciCsLeave(pThis);
300	return VINF_SUCCESS;
301	}
302
303	/**
304	* Lower interrupt.
305	*
306	* @param pDevIns The device instance.
307	* @param pThis The shared virtio core instance data.
308	*/
309	static void vpciLowerInterrupt(PPDMDEVINS pDevIns, PVPCISTATE pThis)
310	{
311	LogFlow(("%s vpciLowerInterrupt\n", INSTANCE(pThis)));
312	RT_NOREF(pThis);
313	PDMDevHlpPCISetIrq(pDevIns, 0, 0);
314	}
315
316
317	void vpciReset(PPDMDEVINS pDevIns, PVPCISTATE pThis)
318	{
319	/* No interrupts should survive device reset, see @bugref(9556). */
320	if (pThis->uISR)
321	vpciLowerInterrupt(pDevIns, pThis);
322
323	pThis->uGuestFeatures = 0;
324	pThis->uQueueSelector = 0;
325	pThis->uStatus = 0;
326	pThis->uISR = 0;
327
328	for (unsigned i = 0; i < pThis->cQueues; i++)
329	vqueueReset(&pThis->Queues[i]);
330	}
331
332
333	DECLINLINE(uint32_t) vpciGetHostFeatures(PVPCISTATE pThis, PCVPCIIOCALLBACKS pCallbacks)
334	{
335	return pCallbacks->pfnGetHostFeatures(pThis) \| VPCI_F_NOTIFY_ON_EMPTY;
336	}
337
338	/**
339	* Port I/O Handler for IN operations.
340	*
341	* @returns VBox status code.
342	*
343	* @param pDevIns The device instance.
344	* @param pThis The shared virtio core instance data.
345	* @param offPort The offset into the I/O range of the port being read.
346	* @param pu32 Where to store the result.
347	* @param cb Number of bytes read.
348	* @param pCallbacks Pointer to the callbacks.
349	* @thread EMT
350	*/
351	int vpciIOPortIn(PPDMDEVINS pDevIns,
352	PVPCISTATE pThis,
353	RTIOPORT offPort,
354	uint32_t *pu32,
355	unsigned cb,
356	PCVPCIIOCALLBACKS pCallbacks)
357	{
358	STAM_PROFILE_ADV_START(&pThis->CTX_SUFF(StatIORead), a);
359
360	/*
361	* We probably do not need to enter critical section when reading registers
362	* as the most of them are either constant or being changed during
363	* initialization only, the exception being ISR which can be raced by all
364	* threads but I see no big harm in it. It also happens to be the most read
365	* register as it gets read in interrupt handler. By dropping cs protection
366	* here we gain the ability to deliver RX packets to the guest while TX is
367	* holding cs transmitting queued packets.
368	*
369	int rc = vpciCsEnter(pThis, VINF_IOM_R3_IOPORT_READ);
370	if (RT_UNLIKELY(rc != VINF_SUCCESS))
371	{
372	STAM_PROFILE_ADV_STOP(&pThis->CTX_SUFF(StatIORead), a);
373	return rc;
374	}*/
375	int rc = VINF_SUCCESS;
376
377	switch (offPort)
378	{
379	case VPCI_HOST_FEATURES:
380	/* Tell the guest what features we support. */
381	ASSERT_GUEST_MSG(cb == 4, ("%d\n", cb));
382	*pu32 = vpciGetHostFeatures(pThis, pCallbacks) \| VPCI_F_BAD_FEATURE;
383	break;
384
385	case VPCI_GUEST_FEATURES:
386	ASSERT_GUEST_MSG(cb == 4, ("%d\n", cb));
387	*pu32 = pThis->uGuestFeatures;
388	break;
389
390	case VPCI_QUEUE_PFN:
391	ASSERT_GUEST_MSG(cb == 4, ("%d\n", cb));
392	*pu32 = pThis->Queues[pThis->uQueueSelector].uPageNumber;
393	break;
394
395	case VPCI_QUEUE_NUM:
396	ASSERT_GUEST_MSG(cb == 2, ("%d\n", cb));
397	*pu32 = pThis->Queues[pThis->uQueueSelector].VRing.uSize;
398	break;
399
400	case VPCI_QUEUE_SEL:
401	ASSERT_GUEST_MSG(cb == 2, ("%d\n", cb));
402	*pu32 = pThis->uQueueSelector;
403	break;
404
405	case VPCI_STATUS:
406	ASSERT_GUEST_MSG(cb == 1, ("%d\n", cb));
407	*pu32 = pThis->uStatus;
408	break;
409
410	case VPCI_ISR:
411	ASSERT_GUEST_MSG(cb == 1, ("%d\n", cb));
412	*pu32 = pThis->uISR;
413	pThis->uISR = 0; /* read clears all interrupts */
414	vpciLowerInterrupt(pDevIns, pThis);
415	break;
416
417	default:
418	if (offPort >= VPCI_CONFIG)
419	rc = pCallbacks->pfnGetConfig(pThis, offPort - VPCI_CONFIG, cb, pu32);
420	else
421	{
422	*pu32 = UINT32_MAX;
423	rc = PDMDevHlpDBGFStop(pDevIns, RT_SRC_POS, "%s vpciIOPortIn: no valid port at offset port=%RTiop cb=%08x\n",
424	INSTANCE(pThis), offPort, cb);
425	}
426	break;
427	}
428	Log3(("%s vpciIOPortIn: At %RTiop in %0x\n", INSTANCE(pThis), offPort, cb2, *pu32));
429
430	//vpciCsLeave(pThis);
431
432	STAM_PROFILE_ADV_STOP(&pThis->CTX_SUFF(StatIORead), a);
433	return rc;
434	}
435
436
437	/**
438	* Port I/O Handler for OUT operations.
439	*
440	* @returns VBox status code.
441	*
442	* @param pDevIns The device instance.
443	* @param pThis The shared virtio core instance data.
444	* @param offPort The offset into the I/O range of the port being written.
445	* @param u32 The value to output.
446	* @param cb The value size in bytes.
447	* @param pCallbacks Pointer to the callbacks.
448	* @thread EMT
449	*/
450	int vpciIOPortOut(PPDMDEVINS pDevIns,
451	PVPCISTATE pThis,
452	PVPCISTATECC pThisCC,
453	RTIOPORT offPort,
454	uint32_t u32,
455	unsigned cb,
456	PCVPCIIOCALLBACKS pCallbacks)
457	{
458	STAM_PROFILE_ADV_START(&pThis->CTX_SUFF(StatIOWrite), a);
459	int rc = VINF_SUCCESS;
460	bool fHasBecomeReady;
461	#ifndef IN_RING3
462	RT_NOREF_PV(pThisCC);
463	#endif
464
465	Log3(("%s virtioIOPortOut: At offPort=%RTiop out %0x\n", INSTANCE(pThis), offPort, cb2, u32));
466
467	switch (offPort)
468	{
469	case VPCI_GUEST_FEATURES:
470	{
471	const uint32_t fHostFeatures = vpciGetHostFeatures(pThis, pCallbacks);
472
473	if (RT_LIKELY((u32 & ~fHostFeatures) == 0))
474	pThis->uGuestFeatures = u32;
475	else
476	{
477	/*
478	* Guest requests features we don't advertise. Stick
479	* to the minimum if negotiation looks completely
480	* botched, otherwise restrict to advertised features.
481	*/
482	if (u32 & VPCI_F_BAD_FEATURE)
483	{
484	Log(("%s WARNING! Guest failed to negotiate properly (guest=%x)\n",
485	INSTANCE(pThis), u32));
486	pThis->uGuestFeatures = pCallbacks->pfnGetHostMinimalFeatures(pThis);
487	}
488	else
489	{
490	Log(("%s Guest asked for features host does not support! (host=%x guest=%x)\n",
491	INSTANCE(pThis), fHostFeatures, u32));
492	pThis->uGuestFeatures = u32 & fHostFeatures;
493	}
494	}
495	pCallbacks->pfnSetHostFeatures(pThis, pThis->uGuestFeatures);
496	break;
497	}
498
499	case VPCI_QUEUE_PFN:
500	/*
501	* The guest is responsible for allocating the pages for queues,
502	* here it provides us with the page number of descriptor table.
503	* Note that we provide the size of the queue to the guest via
504	* VIRTIO_PCI_QUEUE_NUM.
505	*/
506	pThis->Queues[pThis->uQueueSelector].uPageNumber = u32;
507	if (u32)
508	vqueueInit(&pThis->Queues[pThis->uQueueSelector], u32);
509	else
510	rc = pCallbacks->pfnReset(pDevIns);
511	break;
512
513	case VPCI_QUEUE_SEL:
514	ASSERT_GUEST_MSG(cb == 2, ("cb=%u\n", cb));
515	u32 &= 0xFFFF;
516	if (u32 < pThis->cQueues)
517	pThis->uQueueSelector = u32;
518	else
519	Log3(("%s vpciIOPortOut: Invalid queue selector %08x\n", INSTANCE(pThis), u32));
520	break;
521
522	case VPCI_QUEUE_NOTIFY:
523	#ifdef IN_RING3
524	ASSERT_GUEST_MSG(cb == 2, ("cb=%u\n", cb));
525	u32 &= 0xFFFF;
526	if (u32 < pThis->cQueues)
527	{
528	RT_UNTRUSTED_VALIDATED_FENCE();
529	if (pThis->Queues[u32].VRing.addrDescriptors)
530	{
531
532	// rc = vpciCsEnter(pThis, VERR_SEM_BUSY);
533	// if (RT_LIKELY(rc == VINF_SUCCESS))
534	// {
535	pThisCC->Queues[u32].pfnCallback(pDevIns, &pThis->Queues[u32]);
536	// vpciCsLeave(pThis);
537	// }
538	}
539	else
540	Log(("%s The queue (#%d) being notified has not been initialized.\n",
541	INSTANCE(pThis), u32));
542	}
543	else
544	Log(("%s Invalid queue number (%d)\n", INSTANCE(pThis), u32));
545	#else
546	rc = VINF_IOM_R3_IOPORT_WRITE;
547	#endif
548	break;
549
550	case VPCI_STATUS:
551	ASSERT_GUEST_MSG(cb == 1, ("cb=%u\n", cb));
552	u32 &= 0xFF;
553	fHasBecomeReady = !(pThis->uStatus & VPCI_STATUS_DRV_OK) && (u32 & VPCI_STATUS_DRV_OK);
554	pThis->uStatus = u32;
555	/* Writing 0 to the status port triggers device reset. */
556	if (u32 == 0)
557	rc = pCallbacks->pfnReset(pDevIns);
558	else if (fHasBecomeReady)
559	{
560	/* Older hypervisors were lax and did not enforce bus mastering. Older guests
561	* (Linux prior to 2.6.34, NetBSD 6.x) were lazy and did not enable bus mastering.
562	* We automagically enable bus mastering on driver initialization to make existing
563	* drivers work.
564	*/
565	PPDMPCIDEV pPciDev = pDevIns->apPciDevs[0];
566	PDMPciDevSetCommand(pPciDev, PDMPciDevGetCommand(pPciDev) \| PCI_COMMAND_BUSMASTER);
567
568	pCallbacks->pfnReady(pDevIns);
569	}
570	break;
571
572	default:
573	if (offPort >= VPCI_CONFIG)
574	rc = pCallbacks->pfnSetConfig(pThis, offPort - VPCI_CONFIG, cb, &u32);
575	else
576	rc = PDMDevHlpDBGFStop(pDevIns, RT_SRC_POS, "%s vpciIOPortOut: no valid port at offset offPort=%RTiop cb=%08x\n",
577	INSTANCE(pThis), offPort, cb);
578	break;
579	}
580
581	STAM_PROFILE_ADV_STOP(&pThis->CTX_SUFF(StatIOWrite), a);
582	return rc;
583	}
584
585	#ifdef IN_RING3
586
587	/**
588	* Handles common IBase.pfnQueryInterface requests.
589	*/
590	void vpciR3QueryInterface(PVPCISTATECC pThisCC, const char pszIID)
591	{
592	PDMIBASE_RETURN_INTERFACE(pszIID, PDMIBASE, &pThisCC->IBase);
593	PDMIBASE_RETURN_INTERFACE(pszIID, PDMILEDPORTS, &pThisCC->ILeds);
594	return NULL;
595	}
596
597	/**
598	* @interface_method_impl{PDMILEDPORTS,pfnQueryStatusLed}
599	*/
600	static DECLCALLBACK(int) vpciR3QueryStatusLed(PPDMILEDPORTS pInterface, unsigned iLUN, PPDMLED *ppLed)
601	{
602	PVPCISTATECC pThisCC = RT_FROM_MEMBER(pInterface, VPCISTATECC, ILeds);
603	if (iLUN == 0)
604	{
605	*ppLed = &pThisCC->pShared->led;
606	return VINF_SUCCESS;
607	}
608	return VERR_PDM_LUN_NOT_FOUND;
609	}
610
611	/**
612	* Turns on/off the write status LED.
613	*
614	* @returns VBox status code.
615	* @param pThis Pointer to the device state structure.
616	* @param fOn New LED state.
617	*/
618	void vpciR3SetWriteLed(PVPCISTATE pThis, bool fOn)
619	{
620	LogFlow(("%s vpciR3SetWriteLed: %s\n", INSTANCE(pThis), fOn?"on":"off"));
621	if (fOn)
622	pThis->led.Asserted.s.fWriting = pThis->led.Actual.s.fWriting = 1;
623	else
624	pThis->led.Actual.s.fWriting = fOn;
625	}
626
627	/**
628	* Turns on/off the read status LED.
629	*
630	* @returns VBox status code.
631	* @param pThis Pointer to the device state structure.
632	* @param fOn New LED state.
633	*/
634	void vpciR3SetReadLed(PVPCISTATE pThis, bool fOn)
635	{
636	LogFlow(("%s vpciR3SetReadLed: %s\n", INSTANCE(pThis), fOn?"on":"off"));
637	if (fOn)
638	pThis->led.Asserted.s.fReading = pThis->led.Actual.s.fReading = 1;
639	else
640	pThis->led.Actual.s.fReading = fOn;
641	}
642
643	# if 0 /* unused */
644	/**
645	* Sets 32-bit register in PCI configuration space.
646	* @param refPciDev The PCI device.
647	* @param uOffset The register offset.
648	* @param u32Value The value to store in the register.
649	* @thread EMT
650	*/
651	DECLINLINE(void) vpciCfgSetU32(PDMPCIDEV& refPciDev, uint32_t uOffset, uint32_t u32Value)
652	{
653	Assert(uOffset+sizeof(u32Value) <= sizeof(refPciDev.config));
654	(uint32_t)&refPciDev.config[uOffset] = u32Value;
655	}
656	# endif /* unused */
657
658	/**
659	* Dumps the state (useful for both logging and info items).
660	*/
661	void vpciR3DumpStateWorker(PVPCISTATE pThis, PCDBGFINFOHLP pHlp)
662	{
663
664	pHlp->pfnPrintf(pHlp,
665	" uGuestFeatures = 0x%08x\n"
666	" uQueueSelector = 0x%04x\n"
667	" uStatus = 0x%02x\n"
668	" uISR = 0x%02x\n",
669	pThis->uGuestFeatures,
670	pThis->uQueueSelector,
671	pThis->uStatus,
672	pThis->uISR);
673
674	for (unsigned i = 0; i < pThis->cQueues; i++)
675	pHlp->pfnPrintf(pHlp,
676	" %s queue:\n"
677	" VRing.uSize = %u\n"
678	" VRing.addrDescriptors = %p\n"
679	" VRing.addrAvail = %p\n"
680	" VRing.addrUsed = %p\n"
681	" uNextAvailIndex = %u\n"
682	" uNextUsedIndex = %u\n"
683	" uPageNumber = %x\n",
684	pThis->Queues[i].szName,
685	pThis->Queues[i].VRing.uSize,
686	pThis->Queues[i].VRing.addrDescriptors,
687	pThis->Queues[i].VRing.addrAvail,
688	pThis->Queues[i].VRing.addrUsed,
689	pThis->Queues[i].uNextAvailIndex,
690	pThis->Queues[i].uNextUsedIndex,
691	pThis->Queues[i].uPageNumber);
692	}
693
694	# ifdef LOG_ENABLED
695	void vpciR3DumpState(PVPCISTATE pThis, const char *pcszCaller)
696	{
697	if (LogIs2Enabled())
698	{
699	Log2(("vpciR3DumpState: (called from %s)\n", pcszCaller));
700	vpciR3DumpStateWorker(pThis, DBGFR3InfoLogHlp());
701	}
702	}
703	# else
704	# define vpciR3DumpState(x, s) do {} while (0)
705	# endif
706
707	/**
708	* Saved the core virtio state.
709	*
710	* @returns VBox status code.
711	* @param pHlp The device helpers.
712	* @param pThis The shared virtio core instance data.
713	* @param pSSM The handle to the saved state.
714	*/
715	int vpciR3SaveExec(PCPDMDEVHLPR3 pHlp, PVPCISTATE pThis, PSSMHANDLE pSSM)
716	{
717	vpciR3DumpState(pThis, "vpciR3SaveExec");
718
719	pHlp->pfnSSMPutU32(pSSM, pThis->uGuestFeatures);
720	pHlp->pfnSSMPutU16(pSSM, pThis->uQueueSelector);
721	pHlp->pfnSSMPutU8( pSSM, pThis->uStatus);
722	pHlp->pfnSSMPutU8( pSSM, pThis->uISR);
723
724	/* Save queue states */
725	int rc = pHlp->pfnSSMPutU32(pSSM, pThis->cQueues);
726	AssertRCReturn(rc, rc);
727	for (unsigned i = 0; i < pThis->cQueues; i++)
728	{
729	pHlp->pfnSSMPutU16(pSSM, pThis->Queues[i].VRing.uSize);
730	pHlp->pfnSSMPutU32(pSSM, pThis->Queues[i].uPageNumber);
731	pHlp->pfnSSMPutU16(pSSM, pThis->Queues[i].uNextAvailIndex);
732	rc = pHlp->pfnSSMPutU16(pSSM, pThis->Queues[i].uNextUsedIndex);
733	AssertRCReturn(rc, rc);
734	}
735
736	return VINF_SUCCESS;
737	}
738
739	/**
740	* Loads a saved device state.
741	*
742	* @returns VBox status code.
743	* @param pHlp The device helpers.
744	* @param pThis The shared virtio core instance data.
745	* @param pSSM The handle to the saved state.
746	* @param uVersion The data unit version number.
747	* @param uPass The data pass.
748	* @param cQueues The default queue count (for old states).
749	*/
750	int vpciR3LoadExec(PCPDMDEVHLPR3 pHlp, PVPCISTATE pThis, PSSMHANDLE pSSM, uint32_t uVersion, uint32_t uPass, uint32_t cQueues)
751	{
752	int rc;
753
754	if (uPass == SSM_PASS_FINAL)
755	{
756	/* Restore state data */
757	pHlp->pfnSSMGetU32(pSSM, &pThis->uGuestFeatures);
758	pHlp->pfnSSMGetU16(pSSM, &pThis->uQueueSelector);
759	pHlp->pfnSSMGetU8( pSSM, &pThis->uStatus);
760	pHlp->pfnSSMGetU8( pSSM, &pThis->uISR);
761
762	/* Restore queues */
763	if (uVersion > VIRTIO_SAVEDSTATE_VERSION_3_1_BETA1)
764	{
765	rc = pHlp->pfnSSMGetU32(pSSM, &pThis->cQueues);
766	AssertRCReturn(rc, rc);
767	}
768	else
769	pThis->cQueues = cQueues;
770	AssertLogRelMsgReturn(pThis->cQueues <= VIRTIO_MAX_NQUEUES, ("%#x\n", pThis->cQueues), VERR_SSM_LOAD_CONFIG_MISMATCH);
771	AssertLogRelMsgReturn(pThis->uQueueSelector < pThis->cQueues \|\| (pThis->cQueues == 0 && pThis->uQueueSelector),
772	("uQueueSelector=%u cQueues=%u\n", pThis->uQueueSelector, pThis->cQueues),
773	VERR_SSM_LOAD_CONFIG_MISMATCH);
774
775	for (unsigned i = 0; i < pThis->cQueues; i++)
776	{
777	rc = pHlp->pfnSSMGetU16(pSSM, &pThis->Queues[i].VRing.uSize);
778	AssertRCReturn(rc, rc);
779	rc = pHlp->pfnSSMGetU32(pSSM, &pThis->Queues[i].uPageNumber);
780	AssertRCReturn(rc, rc);
781
782	if (pThis->Queues[i].uPageNumber)
783	vqueueInit(&pThis->Queues[i], pThis->Queues[i].uPageNumber);
784
785	rc = pHlp->pfnSSMGetU16(pSSM, &pThis->Queues[i].uNextAvailIndex);
786	AssertRCReturn(rc, rc);
787	rc = pHlp->pfnSSMGetU16(pSSM, &pThis->Queues[i].uNextUsedIndex);
788	AssertRCReturn(rc, rc);
789	}
790	}
791
792	vpciR3DumpState(pThis, "vpciLoadExec");
793
794	return VINF_SUCCESS;
795	}
796
797	PVQUEUE vpciR3AddQueue(PVPCISTATE pThis, PVPCISTATECC pThisCC, unsigned uSize,
798	PFNVPCIQUEUECALLBACK pfnCallback, const char *pcszName)
799	{
800	/* Find an empty queue slot */
801	for (unsigned i = 0; i < pThis->cQueues; i++)
802	{
803	if (pThis->Queues[i].VRing.uSize == 0)
804	{
805	PVQUEUE pQueue = &pThis->Queues[i];
806	pQueue->VRing.uSize = uSize;
807	pQueue->VRing.addrDescriptors = 0;
808	pQueue->uPageNumber = 0;
809	int rc = RTStrCopy(pQueue->szName, sizeof(pQueue->szName), pcszName);
810	AssertRC(rc);
811	pThisCC->Queues[i].pfnCallback = pfnCallback;
812	return pQueue;
813	}
814	}
815	AssertMsgFailedReturn(("%s Too many queues being added, no empty slots available!\n", INSTANCE(pThis)), NULL);
816	}
817
818	/**
819	* Destruct PCI-related part of device.
820	*
821	* We need to free non-VM resources only.
822	*
823	* @returns VBox status code.
824	* @param pThis The shared virtio core instance data.
825	*/
826	int vpciR3Term(PPDMDEVINS pDevIns, PVPCISTATE pThis)
827	{
828	Log(("%s Destroying PCI instance\n", INSTANCE(pThis)));
829
830	if (PDMDevHlpCritSectIsInitialized(pDevIns, &pThis->cs))
831	PDMDevHlpCritSectDelete(pDevIns, &pThis->cs);
832
833	return VINF_SUCCESS;
834	}
835
836	/**
837	* Set PCI configuration space registers.
838	*
839	* @param pPciDev Pointer to the PCI device structure.
840	* @param uDeviceId VirtiO Device Id
841	* @param uClass Class of PCI device (network, etc)
842	* @thread EMT
843	*/
844	static void vpciConfigure(PPDMPCIDEV pPciDev, uint16_t uDeviceId, uint16_t uClass)
845	{
846	/* Configure PCI Device, assume 32-bit mode ******************************/
847	PDMPciDevSetVendorId(pPciDev, DEVICE_PCI_VENDOR_ID);
848	PDMPciDevSetDeviceId(pPciDev, DEVICE_PCI_BASE_ID + uDeviceId);
849	PDMPciDevSetWord(pPciDev, VBOX_PCI_SUBSYSTEM_VENDOR_ID, DEVICE_PCI_SUBSYSTEM_VENDOR_ID);
850	PDMPciDevSetWord(pPciDev, VBOX_PCI_SUBSYSTEM_ID, DEVICE_PCI_SUBSYSTEM_BASE_ID + uDeviceId);
851
852	/* ABI version, must be equal 0 as of 2.6.30 kernel. */
853	PDMPciDevSetByte(pPciDev, VBOX_PCI_REVISION_ID, 0x00);
854	/* Ethernet adapter */
855	PDMPciDevSetByte(pPciDev, VBOX_PCI_CLASS_PROG, 0x00);
856	PDMPciDevSetWord(pPciDev, VBOX_PCI_CLASS_DEVICE, uClass);
857	/* Interrupt Pin: INTA# */
858	PDMPciDevSetByte(pPciDev, VBOX_PCI_INTERRUPT_PIN, 0x01);
859
860	# ifdef VBOX_WITH_MSI_DEVICES
861	PDMPciDevSetCapabilityList(pPciDev, 0x80);
862	PDMPciDevSetStatus(pPciDev, VBOX_PCI_STATUS_CAP_LIST);
863	# endif
864	}
865
866	int vpciR3Init(PPDMDEVINS pDevIns, PVPCISTATE pThis, PVPCISTATECC pThisCC, uint16_t uDeviceId, uint16_t uClass, uint32_t cQueues)
867	{
868	/* Init data members. */
869	pThis->cQueues = cQueues;
870	pThis->led.u32Magic = PDMLED_MAGIC;
871	pThisCC->pShared = pThis;
872	pThisCC->ILeds.pfnQueryStatusLed = vpciR3QueryStatusLed;
873	AssertReturn(pThisCC->IBase.pfnQueryInterface, VERR_INVALID_POINTER);
874	AssertReturn(pThis->szInstance[0], VERR_INVALID_PARAMETER);
875	AssertReturn(strlen(pThis->szInstance) < sizeof(pThis->szInstance), VERR_INVALID_PARAMETER);
876
877	/* Initialize critical section. */
878	int rc = PDMDevHlpCritSectInit(pDevIns, &pThis->cs, RT_SRC_POS, "%s", pThis->szInstance);
879	AssertRCReturn(rc, rc);
880
881	/*
882	* Set up the PCI device.
883	*/
884	PPDMPCIDEV pPciDev = pDevIns->apPciDevs[0];
885	PDMPCIDEV_ASSERT_VALID(pDevIns, pPciDev);
886
887	/* Set PCI config registers */
888	vpciConfigure(pPciDev, uDeviceId, uClass);
889
890	/* Register PCI device */
891	rc = PDMDevHlpPCIRegister(pDevIns, pPciDev);
892	AssertRCReturn(rc, rc);
893
894	# ifdef VBOX_WITH_MSI_DEVICES
895	# if 0
896	{
897	PDMMSIREG aMsiReg;
898
899	RT_ZERO(aMsiReg);
900	aMsiReg.cMsixVectors = 1;
901	aMsiReg.iMsixCapOffset = 0x80;
902	aMsiReg.iMsixNextOffset = 0x0;
903	aMsiReg.iMsixBar = 0;
904	rc = PDMDevHlpPCIRegisterMsi(pDevIns, &aMsiReg);
905	if (RT_FAILURE (rc))
906	PCIDevSetCapabilityList(&pThis->pciDevice, 0x0);
907	}
908	# endif
909	# endif
910
911	/*
912	* Attach the status driver (optional).
913	*/
914	PPDMIBASE pBase;
915	rc = PDMDevHlpDriverAttach(pDevIns, PDM_STATUS_LUN, &pThisCC->IBase, &pBase, "Status Port");
916	if (RT_SUCCESS(rc))
917	pThisCC->pLedsConnector = PDMIBASE_QUERY_INTERFACE(pBase, PDMILEDCONNECTORS);
918	else if (rc != VERR_PDM_NO_ATTACHED_DRIVER)
919	return PDMDEV_SET_ERROR(pDevIns, rc, N_("Failed to attach the status LUN"));
920
921	/*
922	* Statistics.
923	*/
924	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatIntsRaised, STAMTYPE_COUNTER, "Interrupts/Raised", STAMUNIT_OCCURENCES, "Number of raised interrupts");
925	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatIntsSkipped, STAMTYPE_COUNTER, "Interrupts/Skipped", STAMUNIT_OCCURENCES, "Number of skipped interrupts");
926	# ifdef VBOX_WITH_STATISTICS
927	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatIOReadR3, STAMTYPE_PROFILE, "IO/ReadR3", STAMUNIT_TICKS_PER_CALL, "Profiling IO reads in R3");
928	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatIOReadR0, STAMTYPE_PROFILE, "IO/ReadR0", STAMUNIT_TICKS_PER_CALL, "Profiling IO reads in R0");
929	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatIOReadRC, STAMTYPE_PROFILE, "IO/ReadRC", STAMUNIT_TICKS_PER_CALL, "Profiling IO reads in RC");
930	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatIOWriteR3, STAMTYPE_PROFILE, "IO/WriteR3", STAMUNIT_TICKS_PER_CALL, "Profiling IO writes in R3");
931	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatIOWriteR0, STAMTYPE_PROFILE, "IO/WriteR0", STAMUNIT_TICKS_PER_CALL, "Profiling IO writes in R0");
932	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatIOWriteRC, STAMTYPE_PROFILE, "IO/WriteRC", STAMUNIT_TICKS_PER_CALL, "Profiling IO writes in RC");
933	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatCsR3, STAMTYPE_PROFILE, "Cs/CsR3", STAMUNIT_TICKS_PER_CALL, "Profiling CS wait in R3");
934	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatCsR0, STAMTYPE_PROFILE, "Cs/CsR0", STAMUNIT_TICKS_PER_CALL, "Profiling CS wait in R0");
935	PDMDevHlpSTAMRegister(pDevIns, &pThis->StatCsRC, STAMTYPE_PROFILE, "Cs/CsRC", STAMUNIT_TICKS_PER_CALL, "Profiling CS wait in RC");
936	# endif /* VBOX_WITH_STATISTICS */
937
938	return VINF_SUCCESS;
939	}
940
941	#else /* !IN_RING3 */
942
943	/**
944	* Does ring-0/raw-mode initialization.
945	*/
946	int vpciRZInit(PPDMDEVINS pDevIns, PVPCISTATE pThis, PVPCISTATECC pThisCC)
947	{
948	RT_NOREF(pDevIns, pThis, pThisCC);
949	return VINF_SUCCESS;
950	}
951
952	#endif /* !IN_RING3 */
953

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

source: vbox/trunk/src/VBox/Devices/VirtIO/Virtio.cpp@ 86997

Download in other formats: