VirtualBox

source: vbox/trunk/src/VBox/Devices/VirtIO/Virtio_1_0.cpp@ 82145

Last change on this file since 82145 was 82145, checked in by vboxsync, 5 years ago

Storage/DevVirtioSCSI.cpp: Fixed problem with scsi offline errors on Linux guest. Fixed bug in save/load exec.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 80.0 KB
Line 
1/* $Id: Virtio_1_0.cpp 82145 2019-11-24 05:44:12Z vboxsync $ */
2/** @file
3 * Virtio_1_0 - Virtio Common (PCI, feature & config mgt, queue mgt & proxy, notification mgt)
4 */
5
6/*
7 * Copyright (C) 2009-2019 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 <VBox/log.h>
25#include <VBox/msi.h>
26#include <VBox/AssertGuest.h>
27#include <iprt/param.h>
28#include <iprt/assert.h>
29#include <iprt/uuid.h>
30#include <iprt/mem.h>
31#include <iprt/assert.h>
32#include <iprt/sg.h>
33#include <iprt/string.h>
34#include <VBox/vmm/pdmdev.h>
35#include "Virtio_1_0.h"
36
37
38/*********************************************************************************************************************************
39* Defined Constants And Macros *
40*********************************************************************************************************************************/
41#define INSTANCE(a_pVirtio) ((a_pVirtio)->szInstance)
42#define QUEUE_NAME(a_pVirtio, a_idxQueue) ((a_pVirtio)->virtqState[(a_idxQueue)].szVirtqName)
43#define IS_DRIVER_OK(a_pVirtio) ((a_pVirtio)->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK)
44
45/**
46 * This macro returns true if the @a a_offAccess and access length (@a
47 * a_cbAccess) are within the range of the mapped capability struct described by
48 * @a a_LocCapData.
49 *
50 * @param[in] a_offAccess The offset into the MMIO bar of the access.
51 * @param[in] a_cbAccess The access size.
52 * @param[out] a_offIntraVar The variable to return the intra-capability
53 * offset into. ASSUMES this is uint32_t.
54 * @param[in] a_LocCapData The capability location info.
55 */
56#define MATCHES_VIRTIO_CAP_STRUCT(a_offAccess, a_cbAccess, a_offIntraVar, a_LocCapData) \
57 ( ((a_offIntraVar) = (uint32_t)((a_offAccess) - (a_LocCapData).offMmio)) < (uint32_t)(a_LocCapData).cbMmio \
58 && (a_offIntraVar) + (uint32_t)(a_cbAccess) <= (uint32_t)(a_LocCapData).cbMmio )
59
60
61/** Marks the start of the virtio saved state (just for sanity). */
62#define VIRTIO_SAVEDSTATE_MARKER UINT64_C(0x1133557799bbddff)
63/** The current saved state version for the virtio core. */
64#define VIRTIO_SAVEDSTATE_VERSION UINT32_C(1)
65
66
67/*********************************************************************************************************************************
68* Structures and Typedefs *
69*********************************************************************************************************************************/
70/**
71 * virtq related structs
72 * (struct names follow VirtIO 1.0 spec, typedef use VBox style)
73 */
74typedef struct virtq_desc
75{
76 uint64_t GCPhysBuf; /**< addr GC Phys. address of buffer */
77 uint32_t cb; /**< len Buffer length */
78 uint16_t fFlags; /**< flags Buffer specific flags */
79 uint16_t uDescIdxNext; /**< next Idx set if VIRTIO_DESC_F_NEXT */
80} VIRTQ_DESC_T, *PVIRTQ_DESC_T;
81
82typedef struct virtq_avail
83{
84 uint16_t fFlags; /**< flags avail ring drv to dev flags */
85 uint16_t uIdx; /**< idx Index of next free ring slot */
86 uint16_t auRing[RT_FLEXIBLE_ARRAY]; /**< ring Ring: avail drv to dev bufs */
87 /* uint16_t uUsedEventIdx; - used_event (if VIRTQ_USED_F_EVENT_IDX) */
88} VIRTQ_AVAIL_T, *PVIRTQ_AVAIL_T;
89
90typedef struct virtq_used_elem
91{
92 uint32_t uDescIdx; /**< idx Start of used desc chain */
93 uint32_t cbElem; /**< len Total len of used desc chain */
94} VIRTQ_USED_ELEM_T;
95
96typedef struct virt_used
97{
98 uint16_t fFlags; /**< flags used ring host-to-guest flags */
99 uint16_t uIdx; /**< idx Index of next ring slot */
100 VIRTQ_USED_ELEM_T aRing[RT_FLEXIBLE_ARRAY]; /**< ring Ring: used dev to drv bufs */
101 /* uint16_t uAvailEventIdx; - avail_event if (VIRTQ_USED_F_EVENT_IDX) */
102} VIRTQ_USED_T, *PVIRTQ_USED_T;
103
104
105const char *virtioCoreGetStateChangeText(VIRTIOVMSTATECHANGED enmState)
106{
107 switch (enmState)
108 {
109 case kvirtIoVmStateChangedReset: return "VM RESET";
110 case kvirtIoVmStateChangedSuspend: return "VM SUSPEND";
111 case kvirtIoVmStateChangedPowerOff: return "VM POWER OFF";
112 case kvirtIoVmStateChangedResume: return "VM RESUME";
113 default: return "<BAD ENUM>";
114 }
115}
116
117/* Internal Functions */
118
119static void virtioNotifyGuestDriver(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue, bool fForce);
120static int virtioKick(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint8_t uCause, uint16_t uVec, bool fForce);
121
122/** @name Internal queue operations
123 * @{ */
124
125#if 0 /* unused */
126DECLINLINE(int) virtqIsEventNeeded(uint16_t uEventIdx, uint16_t uDescIdxNew, uint16_t uDescIdxOld)
127{
128 return (uint16_t)(uDescIdxNew - uEventIdx - 1) < (uint16_t)(uDescIdxNew - uDescIdxOld);
129}
130#endif
131
132/**
133 * Accessor for virtq descriptor
134 */
135DECLINLINE(void) virtioReadDesc(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue,
136 uint32_t idxDesc, PVIRTQ_DESC_T pDesc)
137{
138 //Log(("%s virtioQueueReadDesc: ring=%p idx=%u\n", INSTANCE(pState), pVirtQ, idx));
139 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
140 uint16_t const cQueueItems = RT_MAX(pVirtio->uQueueSize[idxQueue], 1); /* Make sure to avoid div-by-zero. */
141 PDMDevHlpPCIPhysRead(pDevIns,
142 pVirtio->aGCPhysQueueDesc[idxQueue] + sizeof(VIRTQ_DESC_T) * (idxDesc % cQueueItems),
143 pDesc, sizeof(VIRTQ_DESC_T));
144}
145
146/**
147 * Accessors for virtq avail ring
148 */
149DECLINLINE(uint16_t) virtioReadAvailDescIdx(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue, uint32_t availIdx)
150{
151 uint16_t uDescIdx;
152 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
153 uint16_t const cQueueItems = RT_MAX(pVirtio->uQueueSize[idxQueue], 1); /* Make sure to avoid div-by-zero. */
154 PDMDevHlpPCIPhysRead(pDevIns,
155 pVirtio->aGCPhysQueueAvail[idxQueue]
156 + RT_UOFFSETOF_DYN(VIRTQ_AVAIL_T, auRing[availIdx % cQueueItems]),
157 &uDescIdx, sizeof(uDescIdx));
158 return uDescIdx;
159}
160
161DECLINLINE(uint16_t) virtioReadAvailRingIdx(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue)
162{
163 uint16_t uIdx = 0;
164 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
165 PDMDevHlpPCIPhysRead(pDevIns,
166 pVirtio->aGCPhysQueueAvail[idxQueue] + RT_UOFFSETOF(VIRTQ_AVAIL_T, uIdx),
167 &uIdx, sizeof(uIdx));
168 return uIdx;
169}
170
171DECLINLINE(bool) virtqIsEmpty(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue)
172{
173 return virtioReadAvailRingIdx(pDevIns, pVirtio, idxQueue) == pVirtio->virtqState[idxQueue].uAvailIdx;
174}
175
176#if 0 /* unused - Will be used when VIRTIO_F_EVENT_IDX optional feature is implemented, VirtIO 1.0, 2.4.7 */
177DECLINLINE(uint16_t) virtioReadAvailFlags(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue)
178{
179 uint16_t fFlags;
180 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
181 PDMDevHlpPCIPhysRead(pDevIns,
182 pVirtio->aGCPhysQueueAvail[idxQueue] + RT_UOFFSETOF(VIRTQ_AVAIL_T, fFlags),
183 &fFlags, sizeof(fFlags));
184 return fFlags;
185}
186#endif
187
188DECLINLINE(uint16_t) virtioReadAvailUsedEvent(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue)
189{
190 uint16_t uUsedEventIdx;
191 /* VirtIO 1.0 uUsedEventIdx (used_event) immediately follows ring */
192 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
193 PDMDevHlpPCIPhysRead(pDevIns,
194 pVirtio->aGCPhysQueueAvail[idxQueue] + RT_UOFFSETOF_DYN(VIRTQ_AVAIL_T, auRing[pVirtio->uQueueSize[idxQueue]]),
195 &uUsedEventIdx, sizeof(uUsedEventIdx));
196 return uUsedEventIdx;
197}
198/** @} */
199
200/** @name Accessors for virtq used ring
201 * @{
202 */
203DECLINLINE(void) virtioWriteUsedElem(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue,
204 uint32_t usedIdx, uint32_t uDescIdx, uint32_t uLen)
205{
206 VIRTQ_USED_ELEM_T elem = { uDescIdx, uLen };
207 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
208 uint16_t const cQueueItems = RT_MAX(pVirtio->uQueueSize[idxQueue], 1); /* Make sure to avoid div-by-zero. */
209 PDMDevHlpPCIPhysWrite(pDevIns,
210 pVirtio->aGCPhysQueueUsed[idxQueue] + RT_UOFFSETOF_DYN(VIRTQ_USED_T, aRing[usedIdx % cQueueItems]),
211 &elem, sizeof(elem));
212}
213
214DECLINLINE(void) virtioWriteUsedRingIdx(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue, uint16_t uIdx)
215{
216 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
217 PDMDevHlpPCIPhysWrite(pDevIns,
218 pVirtio->aGCPhysQueueUsed[idxQueue] + RT_UOFFSETOF(VIRTQ_USED_T, uIdx),
219 &uIdx, sizeof(uIdx));
220}
221
222#ifdef LOG_ENABLED
223DECLINLINE(uint16_t) virtioReadUsedRingIdx(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue)
224{
225 uint16_t uIdx = 0;
226 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
227 PDMDevHlpPCIPhysRead(pDevIns,
228 pVirtio->aGCPhysQueueUsed[idxQueue] + RT_UOFFSETOF(VIRTQ_USED_T, uIdx),
229 &uIdx, sizeof(uIdx));
230 return uIdx;
231}
232#endif
233
234DECLINLINE(uint16_t) virtioReadUsedFlags(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue)
235{
236 uint16_t fFlags = 0;
237 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
238 PDMDevHlpPCIPhysRead(pDevIns,
239 pVirtio->aGCPhysQueueUsed[idxQueue] + RT_UOFFSETOF(VIRTQ_USED_T, fFlags),
240 &fFlags, sizeof(fFlags));
241 return fFlags;
242}
243
244#if 0 /* unused - This may eventually be used to set no-notify for the ring as an optimization */
245DECLINLINE(void) virtioWriteUsedFlags(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue, uint32_t fFlags)
246{
247 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
248 RT_UNTRUSTED_VALIDATED_FENCE(); /* VirtIO 1.0, Section 3.2.1.4.1 */
249 PDMDevHlpPCIPhysWrite(pDevIns,
250 pVirtio->aGCPhysQueueUsed[idxQueue] + RT_UOFFSETOF(VIRTQ_USED_T, fFlags),
251 &fFlags, sizeof(fFlags));
252}
253#endif
254
255#if 0 /* unused - *May* be used when VIRTIO_F_EVENT_IDX optional feature is implemented VirtIO 1.0, 2.4.9.2*/
256DECLINLINE(void) virtioWriteUsedAvailEvent(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue, uint32_t uAvailEventIdx)
257{
258 /** VirtIO 1.0 uAvailEventIdx (avail_event) immediately follows ring */
259 AssertMsg(pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK, ("Called with guest driver not ready\n"));
260 PDMDevHlpPCIPhysWrite(pDevIns,
261 pVirtio->aGCPhysQueueUsed[idxQueue] + RT_UOFFSETOF_DYN(VIRTQ_USED_T, aRing[pVirtio->uQueueSize[idxQueue]]),
262 &uAvailEventIdx, sizeof(uAvailEventIdx));
263}
264#endif
265
266/** @} */
267
268void virtioCoreSgBufInit(PVIRTIOSGBUF pGcSgBuf, PVIRTIOSGSEG paSegs, size_t cSegs)
269{
270 AssertPtr(pGcSgBuf);
271 Assert( (cSegs > 0 && VALID_PTR(paSegs)) || (!cSegs && !paSegs));
272 Assert(cSegs < (~(unsigned)0 >> 1));
273
274 pGcSgBuf->paSegs = paSegs;
275 pGcSgBuf->cSegs = (unsigned)cSegs;
276 pGcSgBuf->idxSeg = 0;
277 if (cSegs && paSegs)
278 {
279 pGcSgBuf->pGcSegCur = paSegs[0].pGcSeg;
280 pGcSgBuf->cbSegLeft = paSegs[0].cbSeg;
281 }
282 else
283 {
284 pGcSgBuf->pGcSegCur = 0;
285 pGcSgBuf->cbSegLeft = 0;
286 }
287}
288
289static RTGCPHYS virtioCoreSgBufGet(PVIRTIOSGBUF pGcSgBuf, size_t *pcbData)
290{
291 size_t cbData;
292 RTGCPHYS pGcBuf;
293
294 /* Check that the S/G buffer has memory left. */
295 if (RT_LIKELY(pGcSgBuf->idxSeg < pGcSgBuf->cSegs && pGcSgBuf->cbSegLeft))
296 { /* likely */ }
297 else
298 {
299 *pcbData = 0;
300 return 0;
301 }
302
303 AssertMsg( pGcSgBuf->cbSegLeft <= 128 * _1M
304 && (RTGCPHYS)pGcSgBuf->pGcSegCur >= (RTGCPHYS)pGcSgBuf->paSegs[pGcSgBuf->idxSeg].pGcSeg
305 && (RTGCPHYS)pGcSgBuf->pGcSegCur + pGcSgBuf->cbSegLeft <=
306 (RTGCPHYS)pGcSgBuf->paSegs[pGcSgBuf->idxSeg].pGcSeg + pGcSgBuf->paSegs[pGcSgBuf->idxSeg].cbSeg,
307 ("pGcSgBuf->idxSeg=%d pGcSgBuf->cSegs=%d pGcSgBuf->pGcSegCur=%p pGcSgBuf->cbSegLeft=%zd "
308 "pGcSgBuf->paSegs[%d].pGcSeg=%p pGcSgBuf->paSegs[%d].cbSeg=%zd\n",
309 pGcSgBuf->idxSeg, pGcSgBuf->cSegs, pGcSgBuf->pGcSegCur, pGcSgBuf->cbSegLeft,
310 pGcSgBuf->idxSeg, pGcSgBuf->paSegs[pGcSgBuf->idxSeg].pGcSeg, pGcSgBuf->idxSeg,
311 pGcSgBuf->paSegs[pGcSgBuf->idxSeg].cbSeg));
312
313 cbData = RT_MIN(*pcbData, pGcSgBuf->cbSegLeft);
314 pGcBuf = pGcSgBuf->pGcSegCur;
315 pGcSgBuf->cbSegLeft -= cbData;
316 if (!pGcSgBuf->cbSegLeft)
317 {
318 pGcSgBuf->idxSeg++;
319
320 if (pGcSgBuf->idxSeg < pGcSgBuf->cSegs)
321 {
322 pGcSgBuf->pGcSegCur = pGcSgBuf->paSegs[pGcSgBuf->idxSeg].pGcSeg;
323 pGcSgBuf->cbSegLeft = pGcSgBuf->paSegs[pGcSgBuf->idxSeg].cbSeg;
324 }
325 *pcbData = cbData;
326 }
327 else
328 pGcSgBuf->pGcSegCur = pGcSgBuf->pGcSegCur + cbData;
329
330 return pGcBuf;
331}
332
333void virtioCoreSgBufReset(PVIRTIOSGBUF pGcSgBuf)
334{
335 AssertPtrReturnVoid(pGcSgBuf);
336
337 pGcSgBuf->idxSeg = 0;
338 if (pGcSgBuf->cSegs)
339 {
340 pGcSgBuf->pGcSegCur = pGcSgBuf->paSegs[0].pGcSeg;
341 pGcSgBuf->cbSegLeft = pGcSgBuf->paSegs[0].cbSeg;
342 }
343 else
344 {
345 pGcSgBuf->pGcSegCur = 0;
346 pGcSgBuf->cbSegLeft = 0;
347 }
348}
349
350RTGCPHYS virtioCoreSgBufAdvance(PVIRTIOSGBUF pGcSgBuf, size_t cbAdvance)
351{
352 AssertReturn(pGcSgBuf, 0);
353
354 size_t cbLeft = cbAdvance;
355 while (cbLeft)
356 {
357 size_t cbThisAdvance = cbLeft;
358 virtioCoreSgBufGet(pGcSgBuf, &cbThisAdvance);
359 if (!cbThisAdvance)
360 break;
361
362 cbLeft -= cbThisAdvance;
363 }
364 return cbAdvance - cbLeft;
365}
366
367RTGCPHYS virtioCoreSgBufGetNextSegment(PVIRTIOSGBUF pGcSgBuf, size_t *pcbSeg)
368{
369 AssertReturn(pGcSgBuf, 0);
370 AssertPtrReturn(pcbSeg, 0);
371
372 if (!*pcbSeg)
373 *pcbSeg = pGcSgBuf->cbSegLeft;
374
375 return virtioCoreSgBufGet(pGcSgBuf, pcbSeg);
376}
377
378#ifdef LOG_ENABLED
379
380/**
381 * Does a formatted hex dump using Log(()), recommend using VIRTIO_HEX_DUMP() macro to
382 * control enabling of logging efficiently.
383 *
384 * @param pv pointer to buffer to dump contents of
385 * @param cb count of characters to dump from buffer
386 * @param uBase base address of per-row address prefixing of hex output
387 * @param pszTitle Optional title. If present displays title that lists
388 * provided text with value of cb to indicate size next to it.
389 */
390void virtioCoreHexDump(uint8_t *pv, uint32_t cb, uint32_t uBase, const char *pszTitle)
391{
392 if (pszTitle)
393 Log(("%s [%d bytes]:\n", pszTitle, cb));
394 for (uint32_t row = 0; row < RT_MAX(1, (cb / 16) + 1) && row * 16 < cb; row++)
395 {
396 Log(("%04x: ", row * 16 + uBase)); /* line address */
397 for (uint8_t col = 0; col < 16; col++)
398 {
399 uint32_t idx = row * 16 + col;
400 if (idx >= cb)
401 Log(("-- %s", (col + 1) % 8 ? "" : " "));
402 else
403 Log(("%02x %s", pv[idx], (col + 1) % 8 ? "" : " "));
404 }
405 for (uint32_t idx = row * 16; idx < row * 16 + 16; idx++)
406 Log(("%c", (idx >= cb) ? ' ' : (pv[idx] >= 0x20 && pv[idx] <= 0x7e ? pv[idx] : '.')));
407 Log(("\n"));
408 }
409 Log(("\n"));
410 RT_NOREF2(uBase, pv);
411}
412
413#endif /* LOG_ENABLED */
414
415/**
416 * Log memory-mapped I/O input or output value.
417 *
418 * This is designed to be invoked by macros that can make contextual assumptions
419 * (e.g. implicitly derive MACRO parameters from the invoking function). It is exposed
420 * for the VirtIO client doing the device-specific implementation in order to log in a
421 * similar fashion accesses to the device-specific MMIO configuration structure. Macros
422 * that leverage this function are found in virtioCommonCfgAccessed() and can be
423 * used as an example of how to use this effectively for the device-specific
424 * code.
425 *
426 * @param pszFunc To avoid displaying this function's name via __FUNCTION__ or LogFunc()
427 * @param pszMember Name of struct member
428 * @param pv pointer to value
429 * @param cb size of value
430 * @param uOffset offset into member where value starts
431 * @param fWrite True if write I/O
432 * @param fHasIndex True if the member is indexed
433 * @param idx The index if fHasIndex
434 */
435void virtioCoreLogMappedIoValue(const char *pszFunc, const char *pszMember, uint32_t uMemberSize,
436 const void *pv, uint32_t cb, uint32_t uOffset, int fWrite,
437 int fHasIndex, uint32_t idx)
438{
439 if (!LogIs6Enabled())
440 return;
441
442 char szIdx[16];
443 if (fHasIndex)
444 RTStrPrintf(szIdx, sizeof(szIdx), "[%d]", idx);
445 else
446 szIdx[0] = '\0';
447
448 if (cb == 1 || cb == 2 || cb == 4 || cb == 8)
449 {
450 char szDepiction[64];
451 size_t cchDepiction;
452 if (uOffset != 0 || cb != uMemberSize) /* display bounds if partial member access */
453 cchDepiction = RTStrPrintf(szDepiction, sizeof(szDepiction), "%s%s[%d:%d]",
454 pszMember, szIdx, uOffset, uOffset + cb - 1);
455 else
456 cchDepiction = RTStrPrintf(szDepiction, sizeof(szDepiction), "%s%s", pszMember, szIdx);
457
458 /* padding */
459 if (cchDepiction < 30)
460 szDepiction[cchDepiction++] = ' ';
461 while (cchDepiction < 30)
462 szDepiction[cchDepiction++] = '.';
463 szDepiction[cchDepiction] = '\0';
464
465 RTUINT64U uValue;
466 uValue.u = 0;
467 memcpy(uValue.au8, pv, cb);
468 Log6(("%s: Guest %s %s %#0*RX64\n",
469 pszFunc, fWrite ? "wrote" : "read ", szDepiction, 2 + cb * 2, uValue.u));
470 }
471 else /* odd number or oversized access, ... log inline hex-dump style */
472 {
473 Log6(("%s: Guest %s %s%s[%d:%d]: %.*Rhxs\n",
474 pszFunc, fWrite ? "wrote" : "read ", pszMember,
475 szIdx, uOffset, uOffset + cb, cb, pv));
476 }
477 RT_NOREF2(fWrite, pszFunc);
478}
479
480
481/**
482 * Makes the MMIO-mapped Virtio uDeviceStatus registers non-cryptic
483 */
484DECLINLINE(void) virtioLogDeviceStatus(uint8_t bStatus)
485{
486 if (bStatus == 0)
487 Log6(("RESET"));
488 else
489 {
490 int primed = 0;
491 if (bStatus & VIRTIO_STATUS_ACKNOWLEDGE)
492 Log6(("%sACKNOWLEDGE", primed++ ? "" : ""));
493 if (bStatus & VIRTIO_STATUS_DRIVER)
494 Log6(("%sDRIVER", primed++ ? " | " : ""));
495 if (bStatus & VIRTIO_STATUS_FEATURES_OK)
496 Log6(("%sFEATURES_OK", primed++ ? " | " : ""));
497 if (bStatus & VIRTIO_STATUS_DRIVER_OK)
498 Log6(("%sDRIVER_OK", primed++ ? " | " : ""));
499 if (bStatus & VIRTIO_STATUS_FAILED)
500 Log6(("%sFAILED", primed++ ? " | " : ""));
501 if (bStatus & VIRTIO_STATUS_DEVICE_NEEDS_RESET)
502 Log6(("%sNEEDS_RESET", primed++ ? " | " : ""));
503 (void)primed;
504 }
505}
506
507#ifdef IN_RING3
508/**
509 * Allocate client context for client to work with VirtIO-provided with queue
510 *
511 * @param pVirtio Pointer to the shared virtio state.
512 * @param idxQueue Queue number
513 * @param pcszName Name to give queue
514 *
515 * @returns VBox status code.
516 */
517int virtioCoreR3QueueAttach(PVIRTIOCORE pVirtio, uint16_t idxQueue, const char *pcszName)
518{
519 LogFunc(("%s\n", pcszName));
520 PVIRTQSTATE pVirtq = &pVirtio->virtqState[idxQueue];
521 pVirtq->uAvailIdx = 0;
522 pVirtq->uUsedIdx = 0;
523 pVirtq->fEventThresholdReached = false;
524 RTStrCopy(pVirtq->szVirtqName, sizeof(pVirtq->szVirtqName), pcszName);
525 return VINF_SUCCESS;
526}
527#endif /* IN_RING3 */
528
529/**
530 * See API comments in header file for description
531 */
532int virtioQueueSkip(PVIRTIOCORE pVirtio, uint16_t idxQueue)
533{
534 Assert(idxQueue < RT_ELEMENTS(pVirtio->virtqState));
535 PVIRTQSTATE pVirtq = &pVirtio->virtqState[idxQueue];
536
537 AssertMsgReturn(IS_DRIVER_OK(pVirtio) && pVirtio->uQueueEnable[idxQueue],
538 ("Guest driver not in ready state.\n"), VERR_INVALID_STATE);
539
540 if (virtioCoreQueueIsEmpty(pVirtio->pDevIns, pVirtio, idxQueue))
541 return VERR_NOT_AVAILABLE;
542
543 Log2Func(("%s avail_idx=%u\n", pVirtq->szVirtqName, pVirtq->uAvailIdx));
544 pVirtq->uAvailIdx++;
545
546 return VINF_SUCCESS;
547}
548
549/**
550 * Check if the associated queue is empty
551 *
552 * @param pDevIns The device instance (for reading).
553 * @param pVirtio Pointer to the shared virtio state.
554 * @param idxQueue Queue number
555 *
556 * @retval true Queue is empty or unavailable.
557 * @retval false Queue is available and has entries
558 */
559bool virtioCoreQueueIsEmpty(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue)
560{
561 if (pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK)
562 return virtqIsEmpty(pDevIns, pVirtio, idxQueue);
563 return true;
564}
565
566#ifdef IN_RING3
567
568
569int virtioCoreR3DescChainGet(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue,
570 uint16_t uHeadIdx, PPVIRTIO_DESC_CHAIN_T ppDescChain)
571{
572 AssertReturn(ppDescChain, VERR_INVALID_PARAMETER);
573
574 Assert(idxQueue < RT_ELEMENTS(pVirtio->virtqState));
575
576 PVIRTQSTATE pVirtq = &pVirtio->virtqState[idxQueue];
577
578 PVIRTIOSGSEG paSegsIn = (PVIRTIOSGSEG)RTMemAlloc(VIRTQ_MAX_SIZE * sizeof(VIRTIOSGSEG));
579 AssertReturn(paSegsIn, VERR_NO_MEMORY);
580
581 PVIRTIOSGSEG paSegsOut = (PVIRTIOSGSEG)RTMemAlloc(VIRTQ_MAX_SIZE * sizeof(VIRTIOSGSEG));
582 AssertReturn(paSegsOut, VERR_NO_MEMORY);
583
584 AssertMsgReturn(IS_DRIVER_OK(pVirtio) && pVirtio->uQueueEnable[idxQueue],
585 ("Guest driver not in ready state.\n"), VERR_INVALID_STATE);
586
587 uint16_t uDescIdx = uHeadIdx;
588
589 Log3Func(("%s DESC CHAIN: (head) desc_idx=%u\n", pVirtq->szVirtqName, uHeadIdx));
590 RT_NOREF(pVirtq);
591
592 VIRTQ_DESC_T desc;
593
594 uint32_t cbIn = 0, cbOut = 0, cSegsIn = 0, cSegsOut = 0;
595
596 do
597 {
598 PVIRTIOSGSEG pSeg;
599
600 /*
601 * Malicious guests may go beyond paSegsIn or paSegsOut boundaries by linking
602 * several descriptors into a loop. Since there is no legitimate way to get a sequences of
603 * linked descriptors exceeding the total number of descriptors in the ring (see @bugref{8620}),
604 * the following aborts I/O if breach and employs a simple log throttling algorithm to notify.
605 */
606 if (cSegsIn + cSegsOut >= VIRTQ_MAX_SIZE)
607 {
608 static volatile uint32_t s_cMessages = 0;
609 static volatile uint32_t s_cThreshold = 1;
610 if (ASMAtomicIncU32(&s_cMessages) == ASMAtomicReadU32(&s_cThreshold))
611 {
612 LogRelMax(64, ("Too many linked descriptors; check if the guest arranges descriptors in a loop.\n"));
613 if (ASMAtomicReadU32(&s_cMessages) != 1)
614 LogRelMax(64, ("(the above error has occured %u times so far)\n", ASMAtomicReadU32(&s_cMessages)));
615 ASMAtomicWriteU32(&s_cThreshold, ASMAtomicReadU32(&s_cThreshold) * 10);
616 }
617 break;
618 }
619 RT_UNTRUSTED_VALIDATED_FENCE();
620
621 virtioReadDesc(pDevIns, pVirtio, idxQueue, uDescIdx, &desc);
622
623 if (desc.fFlags & VIRTQ_DESC_F_WRITE)
624 {
625 Log3Func(("%s IN desc_idx=%u seg=%u addr=%RGp cb=%u\n", QUEUE_NAME(pVirtio, idxQueue), uDescIdx, cSegsIn, desc.GCPhysBuf, desc.cb));
626 cbIn += desc.cb;
627 pSeg = &(paSegsIn[cSegsIn++]);
628 }
629 else
630 {
631 Log3Func(("%s OUT desc_idx=%u seg=%u addr=%RGp cb=%u\n", QUEUE_NAME(pVirtio, idxQueue), uDescIdx, cSegsOut, desc.GCPhysBuf, desc.cb));
632 cbOut += desc.cb;
633 pSeg = &(paSegsOut[cSegsOut++]);
634 }
635
636 pSeg->pGcSeg = desc.GCPhysBuf;
637 pSeg->cbSeg = desc.cb;
638
639 uDescIdx = desc.uDescIdxNext;
640 } while (desc.fFlags & VIRTQ_DESC_F_NEXT);
641
642 PVIRTIOSGBUF pSgPhysIn = (PVIRTIOSGBUF)RTMemAllocZ(sizeof(VIRTIOSGBUF));
643 AssertReturn(pSgPhysIn, VERR_NO_MEMORY);
644
645 virtioCoreSgBufInit(pSgPhysIn, paSegsIn, cSegsIn);
646
647 PVIRTIOSGBUF pSgPhysOut = (PVIRTIOSGBUF)RTMemAllocZ(sizeof(VIRTIOSGBUF));
648 AssertReturn(pSgPhysOut, VERR_NO_MEMORY);
649
650 virtioCoreSgBufInit(pSgPhysOut, paSegsOut, cSegsOut);
651
652 PVIRTIO_DESC_CHAIN_T pDescChain = (PVIRTIO_DESC_CHAIN_T)RTMemAllocZ(sizeof(VIRTIO_DESC_CHAIN_T));
653 AssertReturn(pDescChain, VERR_NO_MEMORY);
654
655 pDescChain->uHeadIdx = uHeadIdx;
656 pDescChain->cbPhysSend = cbOut;
657 pDescChain->pSgPhysSend = pSgPhysOut;
658 pDescChain->cbPhysReturn = cbIn;
659 pDescChain->pSgPhysReturn = pSgPhysIn;
660 *ppDescChain = pDescChain;
661
662 Log3Func(("%s -- segs OUT: %u (%u bytes) IN: %u (%u bytes) --\n", pVirtq->szVirtqName, cSegsOut, cbOut, cSegsIn, cbIn));
663
664 return VINF_SUCCESS;
665}
666
667/**
668 * Fetches descriptor chain using avail ring of indicated queue and converts the descriptor
669 * chain into its OUT (to device) and IN to guest components.
670 *
671 * Additionally it converts the OUT desc chain data to a contiguous virtual
672 * memory buffer for easy consumption by the caller. The caller must return the
673 * descriptor chain pointer via virtioCoreR3QueuePut() and then call virtioCoreQueueSync()
674 * at some point to return the data to the guest and complete the transaction.
675 *
676 * @param pDevIns The device instance.
677 * @param pVirtio Pointer to the shared virtio state.
678 * @param idxQueue Queue number
679 * @param fRemove flags whether to remove desc chain from queue (false = peek)
680 * @param ppDescChain Address to store pointer to descriptor chain that contains the
681 * pre-processed transaction information pulled from the virtq.
682 *
683 * @returns VBox status code:
684 * @retval VINF_SUCCESS Success
685 * @retval VERR_INVALID_STATE VirtIO not in ready state (asserted).
686 * @retval VERR_NOT_AVAILABLE If the queue is empty.
687 */
688int virtioCoreR3QueueGet(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue,
689 PPVIRTIO_DESC_CHAIN_T ppDescChain, bool fRemove)
690{
691 PVIRTQSTATE pVirtq = &pVirtio->virtqState[idxQueue];
692
693 if (virtqIsEmpty(pDevIns, pVirtio, idxQueue))
694 return VERR_NOT_AVAILABLE;
695
696 uint16_t uHeadIdx = virtioReadAvailDescIdx(pDevIns, pVirtio, idxQueue, pVirtq->uAvailIdx);
697
698 if (fRemove)
699 pVirtq->uAvailIdx++;
700
701 int rc = virtioCoreR3DescChainGet(pDevIns, pVirtio, idxQueue, uHeadIdx, ppDescChain);
702 return rc;
703}
704
705/**
706 * Returns data to the guest to complete a transaction initiated by virtQueueGet().
707 *
708 * The caller passes in a pointer to a scatter-gather buffer of virtual memory segments
709 * and a pointer to the descriptor chain context originally derived from the pulled
710 * queue entry, and this function will write the virtual memory s/g buffer into the
711 * guest's physical memory free the descriptor chain. The caller handles the freeing
712 * (as needed) of the virtual memory buffer.
713 *
714 * @note This does a write-ahead to the used ring of the guest's queue. The data
715 * written won't be seen by the guest until the next call to virtioCoreQueueSync()
716 *
717 *
718 * @param pDevIns The device instance (for reading).
719 * @param pVirtio Pointer to the shared virtio state.
720 * @param idxQueue Queue number
721 *
722 * @param pSgVirtReturn Points toscatter-gather buffer of virtual memory
723 * segments the caller is returning to the guest.
724 *
725 * @param pDescChain This contains the context of the scatter-gather
726 * buffer originally pulled from the queue.
727 *
728 * @param fFence If true, put up copy fence (memory barrier) after
729 * copying to guest phys. mem.
730 *
731 * @returns VBox status code.
732 * @retval VINF_SUCCESS Success
733 * @retval VERR_INVALID_STATE VirtIO not in ready state
734 * @retval VERR_NOT_AVAILABLE Queue is empty
735 */
736int virtioCoreR3QueuePut(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue, PRTSGBUF pSgVirtReturn,
737 PVIRTIO_DESC_CHAIN_T pDescChain, bool fFence)
738{
739 Assert(idxQueue < RT_ELEMENTS(pVirtio->virtqState));
740 PVIRTQSTATE pVirtq = &pVirtio->virtqState[idxQueue];
741 PVIRTIOSGBUF pSgPhysReturn = pDescChain->pSgPhysReturn;
742
743 AssertMsgReturn(IS_DRIVER_OK(pVirtio) /*&& pVirtio->uQueueEnable[idxQueue]*/,
744 ("Guest driver not in ready state.\n"), VERR_INVALID_STATE);
745
746 Log3Func(("Copying client data to %s, desc chain (head desc_idx %d)\n",
747 QUEUE_NAME(pVirtio, idxQueue), virtioReadUsedRingIdx(pDevIns, pVirtio, idxQueue)));
748
749 /*
750 * Copy s/g buf (virtual memory) to guest phys mem (IN direction). This virtual memory
751 * block will be small (fixed portion of response header + sense buffer area or
752 * control commands or error return values)... The bulk of req data xfers to phys mem
753 * is handled by client */
754
755 size_t cbCopy = 0;
756 size_t cbRemain = RTSgBufCalcTotalLength(pSgVirtReturn);
757 virtioCoreSgBufReset(pSgPhysReturn); /* Reset ptr because req data may have already been written */
758 while (cbRemain)
759 {
760 PVIRTIOSGSEG paSeg = &pSgPhysReturn->paSegs[pSgPhysReturn->idxSeg];
761 uint64_t dstSgStart = (uint64_t)paSeg->pGcSeg;
762 uint64_t dstSgLen = (uint64_t)paSeg->cbSeg;
763 uint64_t dstSgCur = (uint64_t)pSgPhysReturn->pGcSegCur;
764 cbCopy = RT_MIN((uint64_t)pSgVirtReturn->cbSegLeft, dstSgLen - (dstSgCur - dstSgStart));
765 PDMDevHlpPhysWrite(pDevIns, (RTGCPHYS)pSgPhysReturn->pGcSegCur, pSgVirtReturn->pvSegCur, cbCopy);
766 RTSgBufAdvance(pSgVirtReturn, cbCopy);
767 virtioCoreSgBufAdvance(pSgPhysReturn, cbCopy);
768 cbRemain -= cbCopy;
769 }
770
771 if (fFence)
772 RT_UNTRUSTED_NONVOLATILE_COPY_FENCE(); /* needed? */
773
774 /* If this write-ahead crosses threshold where the driver wants to get an event flag it */
775 if (pVirtio->uDriverFeatures & VIRTIO_F_EVENT_IDX)
776 if (pVirtq->uUsedIdx == virtioReadAvailUsedEvent(pDevIns, pVirtio, idxQueue))
777 pVirtq->fEventThresholdReached = true;
778
779 Assert(!(cbCopy >> 32));
780
781 /*
782 * Place used buffer's descriptor in used ring but don't update used ring's slot index.
783 * That will be done with a subsequent client call to virtioCoreQueueSync() */
784 virtioWriteUsedElem(pDevIns, pVirtio, idxQueue, pVirtq->uUsedIdx++, pDescChain->uHeadIdx, (uint32_t)cbCopy);
785
786 Log2Func((".... Copied %zu bytes to %u byte buffer, residual=%zu\n",
787 cbCopy, pDescChain->cbPhysReturn, pDescChain->cbPhysReturn - cbCopy));
788
789 Log6Func(("Write ahead used_idx=%u, %s used_idx=%u\n",
790 pVirtq->uUsedIdx, QUEUE_NAME(pVirtio, idxQueue), virtioReadUsedRingIdx(pDevIns, pVirtio, idxQueue)));
791
792 RTMemFree((void *)pDescChain->pSgPhysSend->paSegs);
793 RTMemFree(pDescChain->pSgPhysSend);
794 RTMemFree((void *)pSgPhysReturn->paSegs);
795 RTMemFree(pSgPhysReturn);
796 RTMemFree(pDescChain);
797
798 return VINF_SUCCESS;
799}
800
801#endif /* IN_RING3 */
802
803/**
804 * Updates the indicated virtq's "used ring" descriptor index to match the
805 * current write-head index, thus exposing the data added to the used ring by all
806 * virtioCoreR3QueuePut() calls since the last sync. This should be called after one or
807 * more virtioCoreR3QueuePut() calls to inform the guest driver there is data in the queue.
808 * Explicit notifications (e.g. interrupt or MSI-X) will be sent to the guest,
809 * depending on VirtIO features negotiated and conditions, otherwise the guest
810 * will detect the update by polling. (see VirtIO 1.0
811 * specification, Section 2.4 "Virtqueues").
812 *
813 * @param pDevIns The device instance.
814 * @param pVirtio Pointer to the shared virtio state.
815 * @param idxQueue Queue number
816 *
817 * @returns VBox status code.
818 * @retval VINF_SUCCESS Success
819 * @retval VERR_INVALID_STATE VirtIO not in ready state
820 */
821int virtioCoreQueueSync(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue)
822{
823 Assert(idxQueue < RT_ELEMENTS(pVirtio->virtqState));
824 PVIRTQSTATE pVirtq = &pVirtio->virtqState[idxQueue];
825
826 AssertMsgReturn(IS_DRIVER_OK(pVirtio) && pVirtio->uQueueEnable[idxQueue],
827 ("Guest driver not in ready state.\n"), VERR_INVALID_STATE);
828
829 Log6Func(("Updating %s used_idx from %u to %u\n",
830 QUEUE_NAME(pVirtio, idxQueue), virtioReadUsedRingIdx(pDevIns, pVirtio, idxQueue), pVirtq->uUsedIdx));
831
832 virtioWriteUsedRingIdx(pDevIns, pVirtio, idxQueue, pVirtq->uUsedIdx);
833 virtioNotifyGuestDriver(pDevIns, pVirtio, idxQueue, false);
834
835 return VINF_SUCCESS;
836}
837
838#ifdef IN_RING3
839/**
840 */
841static void virtioR3QueueNotified(PVIRTIOCORE pVirtio, PVIRTIOCORECC pVirtioCC, uint16_t idxQueue, uint16_t uNotifyIdx)
842{
843 /* See VirtIO 1.0, section 4.1.5.2 It implies that idxQueue and uNotifyIdx should match.
844 * Disregarding this notification may cause throughput to stop, however there's no way to know
845 * which was queue was intended for wake-up if the two parameters disagree. */
846
847 AssertMsg(uNotifyIdx == idxQueue,
848 ("Notification param disagreement. Guest kicked virtq %d's notify addr w/non-corresponding virtq idx %d\n",
849 idxQueue, uNotifyIdx));
850
851// AssertMsgReturn(uNotifyIdx == idxQueue,
852// ("Notification param disagreement. Guest kicked virtq %d's notify addr w/non-corresponding virtq idx %d\n",
853// idxQueue, uNotifyIdx));
854 RT_NOREF(uNotifyIdx);
855
856 AssertReturnVoid(idxQueue < RT_ELEMENTS(pVirtio->virtqState));
857 Log6Func(("%s\n", pVirtio->virtqState[idxQueue].szVirtqName));
858
859 /* Inform client */
860 pVirtioCC->pfnQueueNotified(pVirtio, pVirtioCC, idxQueue);
861}
862#endif /* IN_RING3 */
863
864/**
865 * Trigger MSI-X or INT# interrupt to notify guest of data added to used ring of
866 * the specified virtq, depending on the interrupt configuration of the device
867 * and depending on negotiated and realtime constraints flagged by the guest driver.
868 *
869 * See VirtIO 1.0 specification (section 2.4.7).
870 *
871 * @param pDevIns The device instance.
872 * @param pVirtio Pointer to the shared virtio state.
873 * @param idxQueue Queue to check for guest interrupt handling preference
874 * @param fForce Overrides idxQueue, forcing notification regardless of driver's
875 * notification preferences. This is a safeguard to prevent
876 * stalls upon resuming the VM. VirtIO 1.0 specification Section 4.1.5.5
877 * indicates spurious interrupts are harmless to guest driver's state,
878 * as they only cause the guest driver to [re]scan queues for work to do.
879 */
880static void virtioNotifyGuestDriver(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint16_t idxQueue, bool fForce)
881{
882
883 Assert(idxQueue < RT_ELEMENTS(pVirtio->virtqState));
884 PVIRTQSTATE pVirtq = &pVirtio->virtqState[idxQueue];
885
886 AssertMsgReturnVoid(IS_DRIVER_OK(pVirtio), ("Guest driver not in ready state.\n"));
887 if (pVirtio->uDriverFeatures & VIRTIO_F_EVENT_IDX)
888 {
889 if (pVirtq->fEventThresholdReached)
890 {
891 virtioKick(pDevIns, pVirtio, VIRTIO_ISR_VIRTQ_INTERRUPT, pVirtio->uQueueMsixVector[idxQueue], fForce);
892 pVirtq->fEventThresholdReached = false;
893 return;
894 }
895 Log6Func(("...skipping interrupt: VIRTIO_F_EVENT_IDX set but threshold not reached\n"));
896 }
897 else
898 {
899 /** If guest driver hasn't suppressed interrupts, interrupt */
900 if (fForce || !(virtioReadUsedFlags(pDevIns, pVirtio, idxQueue) & VIRTQ_AVAIL_F_NO_INTERRUPT))
901 {
902 virtioKick(pDevIns, pVirtio, VIRTIO_ISR_VIRTQ_INTERRUPT, pVirtio->uQueueMsixVector[idxQueue], fForce);
903 return;
904 }
905 Log6Func(("...skipping interrupt. Guest flagged VIRTQ_AVAIL_F_NO_INTERRUPT for queue\n"));
906 }
907}
908
909/**
910 * Raise interrupt or MSI-X
911 *
912 * @param pDevIns The device instance.
913 * @param pVirtio Pointer to the shared virtio state.
914 * @param uCause Interrupt cause bit mask to set in PCI ISR port.
915 * @param uVec MSI-X vector, if enabled
916 * @param uForce True of out-of-band
917 */
918static int virtioKick(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, uint8_t uCause, uint16_t uMsixVector, bool fForce)
919{
920 if (fForce)
921 Log6Func(("reason: resumed after suspend\n"));
922 else
923 if (uCause == VIRTIO_ISR_VIRTQ_INTERRUPT)
924 Log6Func(("reason: buffer added to 'used' ring.\n"));
925 else
926 if (uCause == VIRTIO_ISR_DEVICE_CONFIG)
927 Log6Func(("reason: device config change\n"));
928
929 if (!pVirtio->fMsiSupport)
930 {
931 pVirtio->uISR |= uCause;
932 PDMDevHlpPCISetIrq(pDevIns, 0, PDM_IRQ_LEVEL_HIGH);
933 }
934 else if (uMsixVector != VIRTIO_MSI_NO_VECTOR)
935 PDMDevHlpPCISetIrq(pDevIns, uMsixVector, 1);
936 return VINF_SUCCESS;
937}
938
939/**
940 * Lower interrupt. (Called when guest reads ISR)
941 *
942 * @param pDevIns The device instance.
943 */
944static void virtioLowerInterrupt(PPDMDEVINS pDevIns, uint16_t uMsixVector)
945{
946 PVIRTIOCORE pVirtio = PDMINS_2_DATA(pDevIns, PVIRTIOCORE);
947 if (!pVirtio->fMsiSupport)
948 PDMDevHlpPCISetIrq(pDevIns, 0, PDM_IRQ_LEVEL_LOW);
949 else if (uMsixVector != VIRTIO_MSI_NO_VECTOR)
950 PDMDevHlpPCISetIrq(pDevIns, pVirtio->uMsixConfig, PDM_IRQ_LEVEL_LOW);
951}
952
953static void virtioResetQueue(PVIRTIOCORE pVirtio, uint16_t idxQueue)
954{
955 Assert(idxQueue < RT_ELEMENTS(pVirtio->virtqState));
956 PVIRTQSTATE pVirtQ = &pVirtio->virtqState[idxQueue];
957 pVirtQ->uAvailIdx = 0;
958 pVirtQ->uUsedIdx = 0;
959 pVirtio->uQueueEnable[idxQueue] = false;
960 pVirtio->uQueueSize[idxQueue] = VIRTQ_MAX_SIZE;
961 pVirtio->uQueueNotifyOff[idxQueue] = idxQueue;
962
963 pVirtio->uQueueMsixVector[idxQueue] = idxQueue + 2;
964 if (!pVirtio->fMsiSupport) /* VirtIO 1.0, 4.1.4.3 and 4.1.5.1.2 */
965 pVirtio->uQueueMsixVector[idxQueue] = VIRTIO_MSI_NO_VECTOR;
966}
967
968static void virtioResetDevice(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio)
969{
970 Log2Func(("\n"));
971 pVirtio->uDeviceFeaturesSelect = 0;
972 pVirtio->uDriverFeaturesSelect = 0;
973 pVirtio->uConfigGeneration = 0;
974 pVirtio->uDeviceStatus = 0;
975 pVirtio->uISR = 0;
976
977 if (!pVirtio->fMsiSupport)
978 virtioLowerInterrupt(pDevIns, 0);
979 else
980 {
981 virtioLowerInterrupt(pDevIns, pVirtio->uMsixConfig);
982 for (int i = 0; i < VIRTQ_MAX_CNT; i++)
983 {
984 virtioLowerInterrupt(pDevIns, pVirtio->uQueueMsixVector[i]);
985 pVirtio->uQueueMsixVector[i];
986 }
987 }
988
989 if (!pVirtio->fMsiSupport) /* VirtIO 1.0, 4.1.4.3 and 4.1.5.1.2 */
990 pVirtio->uMsixConfig = VIRTIO_MSI_NO_VECTOR;
991
992 for (uint16_t idxQueue = 0; idxQueue < VIRTQ_MAX_CNT; idxQueue++)
993 virtioResetQueue(pVirtio, idxQueue);
994}
995
996/**
997 * Initiate orderly reset procedure. This is an exposed API for clients that might need it.
998 * Invoked by client to reset the device and driver (see VirtIO 1.0 section 2.1.1/2.1.2)
999 */
1000void virtioCoreResetAll(PVIRTIOCORE pVirtio)
1001{
1002 LogFunc(("\n"));
1003 pVirtio->uDeviceStatus |= VIRTIO_STATUS_DEVICE_NEEDS_RESET;
1004 if (pVirtio->uDeviceStatus & VIRTIO_STATUS_DRIVER_OK)
1005 {
1006 pVirtio->fGenUpdatePending = true;
1007 virtioKick(pVirtio->pDevIns, pVirtio, VIRTIO_ISR_DEVICE_CONFIG, pVirtio->uMsixConfig, false /* fForce */);
1008 }
1009}
1010
1011#ifdef IN_RING3
1012/**
1013 * Invoked by this implementation when guest driver resets the device.
1014 * The driver itself will not until the device has read the status change.
1015 */
1016static void virtioGuestR3WasReset(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, PVIRTIOCORECC pVirtioCC)
1017{
1018 LogFunc(("Guest reset the device\n"));
1019
1020 /* Let the client know */
1021 pVirtioCC->pfnStatusChanged(pVirtio, pVirtioCC, 0);
1022 virtioResetDevice(pDevIns, pVirtio);
1023}
1024#endif /* IN_RING3 */
1025
1026/**
1027 * Handle accesses to Common Configuration capability
1028 *
1029 * @returns VBox status code
1030 *
1031 * @param pDevIns The device instance.
1032 * @param pVirtio Pointer to the shared virtio state.
1033 * @param pVirtioCC Pointer to the current context virtio state.
1034 * @param fWrite Set if write access, clear if read access.
1035 * @param offCfg The common configuration capability offset.
1036 * @param cb Number of bytes to read or write
1037 * @param pv Pointer to location to write to or read from
1038 */
1039static int virtioCommonCfgAccessed(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, PVIRTIOCORECC pVirtioCC,
1040 int fWrite, uint32_t offCfg, unsigned cb, void *pv)
1041{
1042/**
1043 * This macro resolves to boolean true if the implied parameters, offCfg and cb,
1044 * match the field offset and size of a field in the Common Cfg struct, (or if
1045 * it is a 64-bit field, if it accesses either 32-bit part as a 32-bit access)
1046 * This is mandated by section 4.1.3.1 of the VirtIO 1.0 specification)
1047 *
1048 * @param member Member of VIRTIO_PCI_COMMON_CFG_T
1049 * @param offCfg Implied parameter: Offset into VIRTIO_PCI_COMMON_CFG_T
1050 * @param cb Implied parameter: Number of bytes to access
1051 * @result true or false
1052 */
1053#define MATCH_COMMON_CFG(member) \
1054 ( ( RT_SIZEOFMEMB(VIRTIO_PCI_COMMON_CFG_T, member) == 8 \
1055 && ( offCfg == RT_OFFSETOF(VIRTIO_PCI_COMMON_CFG_T, member) \
1056 || offCfg == RT_OFFSETOF(VIRTIO_PCI_COMMON_CFG_T, member) + sizeof(uint32_t)) \
1057 && cb == sizeof(uint32_t)) \
1058 || ( offCfg == RT_OFFSETOF(VIRTIO_PCI_COMMON_CFG_T, member) \
1059 && cb == RT_SIZEOFMEMB(VIRTIO_PCI_COMMON_CFG_T, member)) )
1060
1061#ifdef LOG_ENABLED
1062# define LOG_COMMON_CFG_ACCESS(member, a_offIntra) \
1063 virtioCoreLogMappedIoValue(__FUNCTION__, #member, RT_SIZEOFMEMB(VIRTIO_PCI_COMMON_CFG_T, member), \
1064 pv, cb, a_offIntra, fWrite, false, 0);
1065# define LOG_COMMON_CFG_ACCESS_INDEXED(member, idx, a_offIntra) \
1066 virtioCoreLogMappedIoValue(__FUNCTION__, #member, RT_SIZEOFMEMB(VIRTIO_PCI_COMMON_CFG_T, member), \
1067 pv, cb, a_offIntra, fWrite, true, idx);
1068#else
1069# define LOG_COMMON_CFG_ACCESS(member, a_offIntra) do { } while (0)
1070# define LOG_COMMON_CFG_ACCESS_INDEXED(member, idx, a_offIntra) do { } while (0)
1071#endif
1072
1073#define COMMON_CFG_ACCESSOR(member) \
1074 do \
1075 { \
1076 uint32_t offIntra = offCfg - RT_OFFSETOF(VIRTIO_PCI_COMMON_CFG_T, member); \
1077 if (fWrite) \
1078 memcpy((char *)&pVirtio->member + offIntra, (const char *)pv, cb); \
1079 else \
1080 memcpy(pv, (const char *)&pVirtio->member + offIntra, cb); \
1081 LOG_COMMON_CFG_ACCESS(member, offIntra); \
1082 } while(0)
1083
1084#define COMMON_CFG_ACCESSOR_INDEXED(member, idx) \
1085 do \
1086 { \
1087 uint32_t offIntra = offCfg - RT_OFFSETOF(VIRTIO_PCI_COMMON_CFG_T, member); \
1088 if (fWrite) \
1089 memcpy((char *)&pVirtio->member[idx] + offIntra, pv, cb); \
1090 else \
1091 memcpy(pv, (const char *)&pVirtio->member[idx] + offIntra, cb); \
1092 LOG_COMMON_CFG_ACCESS_INDEXED(member, idx, offIntra); \
1093 } while(0)
1094
1095#define COMMON_CFG_ACCESSOR_READONLY(member) \
1096 do \
1097 { \
1098 uint32_t offIntra = offCfg - RT_OFFSETOF(VIRTIO_PCI_COMMON_CFG_T, member); \
1099 if (fWrite) \
1100 LogFunc(("Guest attempted to write readonly virtio_pci_common_cfg.%s\n", #member)); \
1101 else \
1102 { \
1103 memcpy(pv, (const char *)&pVirtio->member + offIntra, cb); \
1104 LOG_COMMON_CFG_ACCESS(member, offIntra); \
1105 } \
1106 } while(0)
1107
1108#define COMMON_CFG_ACCESSOR_INDEXED_READONLY(member, idx) \
1109 do \
1110 { \
1111 uint32_t offIntra = offCfg - RT_OFFSETOF(VIRTIO_PCI_COMMON_CFG_T, member); \
1112 if (fWrite) \
1113 LogFunc(("Guest attempted to write readonly virtio_pci_common_cfg.%s[%d]\n", #member, idx)); \
1114 else \
1115 { \
1116 memcpy(pv, (char const *)&pVirtio->member[idx] + offIntra, cb); \
1117 LOG_COMMON_CFG_ACCESS_INDEXED(member, idx, offIntra); \
1118 } \
1119 } while(0)
1120
1121
1122 int rc = VINF_SUCCESS;
1123 uint64_t val;
1124 if (MATCH_COMMON_CFG(uDeviceFeatures))
1125 {
1126 if (fWrite) /* Guest WRITE pCommonCfg>uDeviceFeatures */
1127 {
1128 LogFunc(("Guest attempted to write readonly virtio_pci_common_cfg.device_feature\n"));
1129 return VINF_SUCCESS;
1130 }
1131 else /* Guest READ pCommonCfg->uDeviceFeatures */
1132 {
1133 switch (pVirtio->uDeviceFeaturesSelect)
1134 {
1135 case 0:
1136 val = pVirtio->uDeviceFeatures & UINT32_C(0xffffffff);
1137 memcpy(pv, &val, cb);
1138 LOG_COMMON_CFG_ACCESS(uDeviceFeatures, offCfg - RT_UOFFSETOF(VIRTIO_PCI_COMMON_CFG_T, uDeviceFeatures));
1139 break;
1140 case 1:
1141 val = pVirtio->uDeviceFeatures >> 32;
1142 memcpy(pv, &val, cb);
1143 LOG_COMMON_CFG_ACCESS(uDeviceFeatures, offCfg - RT_UOFFSETOF(VIRTIO_PCI_COMMON_CFG_T, uDeviceFeatures) + 4);
1144 break;
1145 default:
1146 LogFunc(("Guest read uDeviceFeatures with out of range selector (%#x), returning 0\n",
1147 pVirtio->uDeviceFeaturesSelect));
1148 return VINF_IOM_MMIO_UNUSED_00;
1149 }
1150 }
1151 }
1152 else if (MATCH_COMMON_CFG(uDriverFeatures))
1153 {
1154 if (fWrite) /* Guest WRITE pCommonCfg->udriverFeatures */
1155 {
1156 switch (pVirtio->uDriverFeaturesSelect)
1157 {
1158 case 0:
1159 memcpy(&pVirtio->uDriverFeatures, pv, cb);
1160 LOG_COMMON_CFG_ACCESS(uDriverFeatures, offCfg - RT_UOFFSETOF(VIRTIO_PCI_COMMON_CFG_T, uDriverFeatures));
1161 break;
1162 case 1:
1163 memcpy((char *)&pVirtio->uDriverFeatures + sizeof(uint32_t), pv, cb);
1164 LOG_COMMON_CFG_ACCESS(uDriverFeatures, offCfg - RT_UOFFSETOF(VIRTIO_PCI_COMMON_CFG_T, uDriverFeatures) + 4);
1165 break;
1166 default:
1167 LogFunc(("Guest wrote uDriverFeatures with out of range selector (%#x), returning 0\n",
1168 pVirtio->uDriverFeaturesSelect));
1169 return VINF_SUCCESS;
1170 }
1171 }
1172 else /* Guest READ pCommonCfg->udriverFeatures */
1173 {
1174 switch (pVirtio->uDriverFeaturesSelect)
1175 {
1176 case 0:
1177 val = pVirtio->uDriverFeatures & 0xffffffff;
1178 memcpy(pv, &val, cb);
1179 LOG_COMMON_CFG_ACCESS(uDriverFeatures, offCfg - RT_UOFFSETOF(VIRTIO_PCI_COMMON_CFG_T, uDriverFeatures));
1180 break;
1181 case 1:
1182 val = (pVirtio->uDriverFeatures >> 32) & 0xffffffff;
1183 memcpy(pv, &val, cb);
1184 LOG_COMMON_CFG_ACCESS(uDriverFeatures, offCfg - RT_UOFFSETOF(VIRTIO_PCI_COMMON_CFG_T, uDriverFeatures) + 4);
1185 break;
1186 default:
1187 LogFunc(("Guest read uDriverFeatures with out of range selector (%#x), returning 0\n",
1188 pVirtio->uDriverFeaturesSelect));
1189 return VINF_IOM_MMIO_UNUSED_00;
1190 }
1191 }
1192 }
1193 else if (MATCH_COMMON_CFG(uNumQueues))
1194 {
1195 if (fWrite)
1196 {
1197 Log2Func(("Guest attempted to write readonly virtio_pci_common_cfg.num_queues\n"));
1198 return VINF_SUCCESS;
1199 }
1200 else
1201 {
1202 *(uint16_t *)pv = VIRTQ_MAX_CNT;
1203 LOG_COMMON_CFG_ACCESS(uNumQueues, 0);
1204 }
1205 }
1206 else if (MATCH_COMMON_CFG(uDeviceStatus))
1207 {
1208 if (fWrite) /* Guest WRITE pCommonCfg->uDeviceStatus */
1209 {
1210 uint8_t const fNewStatus = *(uint8_t *)pv;
1211 Log6Func(("Guest wrote uDeviceStatus (%#x, was %#x, change #%x) ................ (",
1212 fNewStatus, pVirtio->uDeviceStatus, fNewStatus ^ pVirtio->uDeviceStatus));
1213 virtioLogDeviceStatus(fNewStatus);
1214 Log6((")\n"));
1215
1216 /* If the status changed or we were reset, we need to go to ring-3 as
1217 it requires notifying the parent device. */
1218 bool const fStatusChanged = (fNewStatus & VIRTIO_STATUS_DRIVER_OK)
1219 != (pVirtio->uPrevDeviceStatus & VIRTIO_STATUS_DRIVER_OK);
1220#ifndef IN_RING3
1221 if (fStatusChanged || fNewStatus == 0)
1222 {
1223 Log6Func(("=>ring3\n"));
1224 return VINF_IOM_R3_MMIO_WRITE;
1225 }
1226#endif
1227 pVirtio->uDeviceStatus = fNewStatus;
1228
1229#ifdef IN_RING3
1230 /*
1231 * Notify client only if status actually changed from last time and when we're reset.
1232 */
1233 if (pVirtio->uDeviceStatus == 0)
1234 virtioGuestR3WasReset(pDevIns, pVirtio, pVirtioCC);
1235 if (fStatusChanged)
1236 pVirtioCC->pfnStatusChanged(pVirtio, pVirtioCC, fNewStatus & VIRTIO_STATUS_DRIVER_OK);
1237#endif
1238 /*
1239 * Save the current status for the next write so we can see what changed.
1240 */
1241 pVirtio->uPrevDeviceStatus = pVirtio->uDeviceStatus;
1242 }
1243 else /* Guest READ pCommonCfg->uDeviceStatus */
1244 {
1245 Log6Func(("Guest read uDeviceStatus ................ ("));
1246 *(uint8_t *)pv = pVirtio->uDeviceStatus;
1247 virtioLogDeviceStatus(pVirtio->uDeviceStatus);
1248 Log6((")\n"));
1249 }
1250 }
1251 else
1252 if (MATCH_COMMON_CFG(uMsixConfig))
1253 COMMON_CFG_ACCESSOR(uMsixConfig);
1254 else
1255 if (MATCH_COMMON_CFG(uDeviceFeaturesSelect))
1256 COMMON_CFG_ACCESSOR(uDeviceFeaturesSelect);
1257 else
1258 if (MATCH_COMMON_CFG(uDriverFeaturesSelect))
1259 COMMON_CFG_ACCESSOR(uDriverFeaturesSelect);
1260 else
1261 if (MATCH_COMMON_CFG(uConfigGeneration))
1262 COMMON_CFG_ACCESSOR_READONLY(uConfigGeneration);
1263 else
1264 if (MATCH_COMMON_CFG(uQueueSelect))
1265 COMMON_CFG_ACCESSOR(uQueueSelect);
1266 else
1267 if (MATCH_COMMON_CFG(uQueueSize))
1268 COMMON_CFG_ACCESSOR_INDEXED(uQueueSize, pVirtio->uQueueSelect);
1269 else
1270 if (MATCH_COMMON_CFG(uQueueMsixVector))
1271 COMMON_CFG_ACCESSOR_INDEXED(uQueueMsixVector, pVirtio->uQueueSelect);
1272 else
1273 if (MATCH_COMMON_CFG(uQueueEnable))
1274 COMMON_CFG_ACCESSOR_INDEXED(uQueueEnable, pVirtio->uQueueSelect);
1275 else
1276 if (MATCH_COMMON_CFG(uQueueNotifyOff))
1277 COMMON_CFG_ACCESSOR_INDEXED_READONLY(uQueueNotifyOff, pVirtio->uQueueSelect);
1278 else
1279 if (MATCH_COMMON_CFG(aGCPhysQueueDesc))
1280 COMMON_CFG_ACCESSOR_INDEXED(aGCPhysQueueDesc, pVirtio->uQueueSelect);
1281 else
1282 if (MATCH_COMMON_CFG(aGCPhysQueueAvail))
1283 COMMON_CFG_ACCESSOR_INDEXED(aGCPhysQueueAvail, pVirtio->uQueueSelect);
1284 else
1285 if (MATCH_COMMON_CFG(aGCPhysQueueUsed))
1286 COMMON_CFG_ACCESSOR_INDEXED(aGCPhysQueueUsed, pVirtio->uQueueSelect);
1287 else
1288 {
1289 Log2Func(("Bad guest %s access to virtio_pci_common_cfg: offCfg=%#x (%d), cb=%d\n",
1290 fWrite ? "write" : "read ", offCfg, offCfg, cb));
1291 return fWrite ? VINF_SUCCESS : VINF_IOM_MMIO_UNUSED_00;
1292 }
1293
1294#undef COMMON_CFG_ACCESSOR_READONLY
1295#undef COMMON_CFG_ACCESSOR_INDEXED_READONLY
1296#undef COMMON_CFG_ACCESSOR_INDEXED
1297#undef COMMON_CFG_ACCESSOR
1298#undef LOG_COMMON_CFG_ACCESS_INDEXED
1299#undef LOG_COMMON_CFG_ACCESS
1300#undef MATCH_COMMON_CFG
1301#ifndef IN_RING3
1302 RT_NOREF(pDevIns, pVirtioCC);
1303#endif
1304 return rc;
1305}
1306
1307/**
1308 * @callback_method_impl{FNIOMMMIONEWREAD,
1309 * Memory mapped I/O Handler for PCI Capabilities read operations.}
1310 *
1311 * This MMIO handler specifically supports the VIRTIO_PCI_CAP_PCI_CFG capability defined
1312 * in the VirtIO 1.0 specification, section 4.1.4.7, and as such is limited to cb == 1, cb == 2, or cb==4 type reads.
1313 *
1314 */
1315static DECLCALLBACK(VBOXSTRICTRC) virtioMmioRead(PPDMDEVINS pDevIns, void *pvUser, RTGCPHYS off, void *pv, unsigned cb)
1316{
1317 PVIRTIOCORE pVirtio = PDMINS_2_DATA(pDevIns, PVIRTIOCORE);
1318 PVIRTIOCORECC pVirtioCC = PDMINS_2_DATA_CC(pDevIns, PVIRTIOCORECC);
1319 AssertReturn(cb == 1 || cb == 2 || cb == 4, VERR_INVALID_PARAMETER);
1320 Assert(pVirtio == (PVIRTIOCORE)pvUser); RT_NOREF(pvUser);
1321
1322 uint32_t offIntra;
1323 if (MATCHES_VIRTIO_CAP_STRUCT(off, cb, offIntra, pVirtio->LocDeviceCap))
1324 {
1325#ifdef IN_RING3
1326 /*
1327 * Callback to client to manage device-specific configuration.
1328 */
1329 VBOXSTRICTRC rcStrict = pVirtioCC->pfnDevCapRead(pDevIns, offIntra, pv, cb);
1330
1331 /*
1332 * Additionally, anytime any part of the device-specific configuration (which our client maintains)
1333 * is READ it needs to be checked to see if it changed since the last time any part was read, in
1334 * order to maintain the config generation (see VirtIO 1.0 spec, section 4.1.4.3.1)
1335 */
1336 bool fDevSpecificFieldChanged = !!memcmp(pVirtioCC->pbDevSpecificCfg + offIntra,
1337 pVirtioCC->pbPrevDevSpecificCfg + offIntra,
1338 RT_MIN(cb, pVirtioCC->cbDevSpecificCfg - offIntra));
1339
1340 memcpy(pVirtioCC->pbPrevDevSpecificCfg, pVirtioCC->pbDevSpecificCfg, pVirtioCC->cbDevSpecificCfg);
1341
1342 if (pVirtio->fGenUpdatePending || fDevSpecificFieldChanged)
1343 {
1344 ++pVirtio->uConfigGeneration;
1345 Log6Func(("Bumped cfg. generation to %d because %s%s\n",
1346 pVirtio->uConfigGeneration,
1347 fDevSpecificFieldChanged ? "<dev cfg changed> " : "",
1348 pVirtio->fGenUpdatePending ? "<update was pending>" : ""));
1349 pVirtio->fGenUpdatePending = false;
1350 }
1351
1352 if (pVirtio->fMsiSupport)
1353 PDMDevHlpPCISetIrq(pDevIns, pVirtio->uMsixConfig, PDM_IRQ_LEVEL_LOW);
1354
1355 return rcStrict;
1356#else
1357 return VINF_IOM_R3_MMIO_READ;
1358#endif
1359 }
1360
1361 if (MATCHES_VIRTIO_CAP_STRUCT(off, cb, offIntra, pVirtio->LocCommonCfgCap))
1362 return virtioCommonCfgAccessed(pDevIns, pVirtio, pVirtioCC, false /* fWrite */, offIntra, cb, pv);
1363
1364 if (MATCHES_VIRTIO_CAP_STRUCT(off, cb, offIntra, pVirtio->LocIsrCap) && cb == sizeof(uint8_t))
1365 {
1366 *(uint8_t *)pv = pVirtio->uISR;
1367 Log6Func(("Read and clear ISR\n"));
1368 pVirtio->uISR = 0; /* VirtIO specification requires reads of ISR to clear it */
1369 virtioLowerInterrupt(pDevIns, 0);
1370 return VINF_SUCCESS;
1371 }
1372
1373 ASSERT_GUEST_MSG_FAILED(("Bad read access to mapped capabilities region: off=%RGp cb=%u\n", off, cb));
1374 return VINF_IOM_MMIO_UNUSED_00;
1375}
1376
1377/**
1378 * @callback_method_impl{FNIOMMMIONEWREAD,
1379 * Memory mapped I/O Handler for PCI Capabilities write operations.}
1380 *
1381 * This MMIO handler specifically supports the VIRTIO_PCI_CAP_PCI_CFG capability defined
1382 * in the VirtIO 1.0 specification, section 4.1.4.7, and as such is limited to cb == 1, cb == 2, or cb==4 type writes.
1383 */
1384static DECLCALLBACK(VBOXSTRICTRC) virtioMmioWrite(PPDMDEVINS pDevIns, void *pvUser, RTGCPHYS off, void const *pv, unsigned cb)
1385{
1386 PVIRTIOCORE pVirtio = PDMINS_2_DATA(pDevIns, PVIRTIOCORE);
1387 PVIRTIOCORECC pVirtioCC = PDMINS_2_DATA_CC(pDevIns, PVIRTIOCORECC);
1388
1389 AssertReturn(cb == 1 || cb == 2 || cb == 4, VERR_INVALID_PARAMETER);
1390
1391 Assert(pVirtio == (PVIRTIOCORE)pvUser); RT_NOREF(pvUser);
1392
1393 uint32_t offIntra;
1394 if (MATCHES_VIRTIO_CAP_STRUCT(off, cb, offIntra, pVirtio->LocDeviceCap))
1395 {
1396#ifdef IN_RING3
1397 /*
1398 * Pass this MMIO write access back to the client to handle
1399 */
1400 return pVirtioCC->pfnDevCapWrite(pDevIns, offIntra, pv, cb);
1401#else
1402 return VINF_IOM_R3_MMIO_WRITE;
1403#endif
1404 }
1405
1406 if (MATCHES_VIRTIO_CAP_STRUCT(off, cb, offIntra, pVirtio->LocCommonCfgCap))
1407 return virtioCommonCfgAccessed(pDevIns, pVirtio, pVirtioCC, true /* fWrite */, offIntra, cb, (void *)pv);
1408
1409 if (MATCHES_VIRTIO_CAP_STRUCT(off, cb, offIntra, pVirtio->LocIsrCap) && cb == sizeof(uint8_t))
1410 {
1411 pVirtio->uISR = *(uint8_t *)pv;
1412 Log6Func(("Setting uISR = 0x%02x (virtq interrupt: %d, dev confg interrupt: %d)\n",
1413 pVirtio->uISR & 0xff,
1414 pVirtio->uISR & VIRTIO_ISR_VIRTQ_INTERRUPT,
1415 RT_BOOL(pVirtio->uISR & VIRTIO_ISR_DEVICE_CONFIG)));
1416 return VINF_SUCCESS;
1417 }
1418
1419 /* This *should* be guest driver dropping index of a new descriptor in avail ring */
1420 if (MATCHES_VIRTIO_CAP_STRUCT(off, cb, offIntra, pVirtio->LocNotifyCap) && cb == sizeof(uint16_t))
1421 {
1422#ifdef IN_RING3
1423 virtioR3QueueNotified(pVirtio, pVirtioCC, offIntra / VIRTIO_NOTIFY_OFFSET_MULTIPLIER, *(uint16_t *)pv);
1424 return VINF_SUCCESS;
1425#else
1426 return VINF_IOM_R3_MMIO_WRITE;
1427#endif
1428 }
1429
1430 ASSERT_GUEST_MSG_FAILED(("Bad write access to mapped capabilities region: off=%RGp pv=%#p{%.*Rhxs} cb=%u\n", off, pv, cb, pv, cb));
1431 return VINF_SUCCESS;
1432}
1433
1434#ifdef IN_RING3
1435
1436/**
1437 * @callback_method_impl{FNPCICONFIGREAD}
1438 */
1439static DECLCALLBACK(VBOXSTRICTRC) virtioR3PciConfigRead(PPDMDEVINS pDevIns, PPDMPCIDEV pPciDev,
1440 uint32_t uAddress, unsigned cb, uint32_t *pu32Value)
1441{
1442 PVIRTIOCORE pVirtio = PDMINS_2_DATA(pDevIns, PVIRTIOCORE);
1443 PVIRTIOCORECC pVirtioCC = PDMINS_2_DATA_CC(pDevIns, PVIRTIOCORECC);
1444 RT_NOREF(pPciDev);
1445
1446 LogFlowFunc(("pDevIns=%p pPciDev=%p uAddress=%#x cb=%u pu32Value=%p\n",
1447 pDevIns, pPciDev, uAddress, cb, pu32Value));
1448 if (uAddress == pVirtio->uPciCfgDataOff)
1449 {
1450 /*
1451 * VirtIO 1.0 spec section 4.1.4.7 describes a required alternative access capability
1452 * whereby the guest driver can specify a bar, offset, and length via the PCI configuration space
1453 * (the virtio_pci_cfg_cap capability), and access data items.
1454 */
1455 uint32_t uLength = pVirtioCC->pPciCfgCap->pciCap.uLength;
1456 uint32_t uOffset = pVirtioCC->pPciCfgCap->pciCap.uOffset;
1457 uint8_t uBar = pVirtioCC->pPciCfgCap->pciCap.uBar;
1458
1459 if ( (uLength != 1 && uLength != 2 && uLength != 4)
1460 || cb != uLength
1461 || uBar != VIRTIO_REGION_PCI_CAP)
1462 {
1463 ASSERT_GUEST_MSG_FAILED(("Guest read virtio_pci_cfg_cap.pci_cfg_data using mismatching config. Ignoring\n"));
1464 *pu32Value = UINT32_MAX;
1465 return VINF_SUCCESS;
1466 }
1467
1468 VBOXSTRICTRC rcStrict = virtioMmioRead(pDevIns, pVirtio, uOffset, pu32Value, cb);
1469 Log2Func(("virtio: Guest read virtio_pci_cfg_cap.pci_cfg_data, bar=%d, offset=%d, length=%d, result=%d -> %Rrc\n",
1470 uBar, uOffset, uLength, *pu32Value, VBOXSTRICTRC_VAL(rcStrict)));
1471 return rcStrict;
1472 }
1473 return VINF_PDM_PCI_DO_DEFAULT;
1474}
1475
1476/**
1477 * @callback_method_impl{FNPCICONFIGWRITE}
1478 */
1479static DECLCALLBACK(VBOXSTRICTRC) virtioR3PciConfigWrite(PPDMDEVINS pDevIns, PPDMPCIDEV pPciDev,
1480 uint32_t uAddress, unsigned cb, uint32_t u32Value)
1481{
1482 PVIRTIOCORE pVirtio = PDMINS_2_DATA(pDevIns, PVIRTIOCORE);
1483 PVIRTIOCORECC pVirtioCC = PDMINS_2_DATA_CC(pDevIns, PVIRTIOCORECC);
1484 RT_NOREF(pPciDev);
1485
1486 LogFlowFunc(("pDevIns=%p pPciDev=%p uAddress=%#x cb=%u u32Value=%#x\n", pDevIns, pPciDev, uAddress, cb, u32Value));
1487 if (uAddress == pVirtio->uPciCfgDataOff)
1488 {
1489 /* VirtIO 1.0 spec section 4.1.4.7 describes a required alternative access capability
1490 * whereby the guest driver can specify a bar, offset, and length via the PCI configuration space
1491 * (the virtio_pci_cfg_cap capability), and access data items. */
1492
1493 uint32_t uLength = pVirtioCC->pPciCfgCap->pciCap.uLength;
1494 uint32_t uOffset = pVirtioCC->pPciCfgCap->pciCap.uOffset;
1495 uint8_t uBar = pVirtioCC->pPciCfgCap->pciCap.uBar;
1496
1497 if ( (uLength != 1 && uLength != 2 && uLength != 4)
1498 || cb != uLength
1499 || uBar != VIRTIO_REGION_PCI_CAP)
1500 {
1501 ASSERT_GUEST_MSG_FAILED(("Guest write virtio_pci_cfg_cap.pci_cfg_data using mismatching config. Ignoring\n"));
1502 return VINF_SUCCESS;
1503 }
1504
1505 VBOXSTRICTRC rcStrict = virtioMmioWrite(pDevIns, pVirtio, uOffset, &u32Value, cb);
1506 Log2Func(("Guest wrote virtio_pci_cfg_cap.pci_cfg_data, bar=%d, offset=%x, length=%x, value=%d -> %Rrc\n",
1507 uBar, uOffset, uLength, u32Value, VBOXSTRICTRC_VAL(rcStrict)));
1508 return rcStrict;
1509 }
1510 return VINF_PDM_PCI_DO_DEFAULT;
1511}
1512
1513
1514/*********************************************************************************************************************************
1515* Saved state. *
1516*********************************************************************************************************************************/
1517
1518/**
1519 * Called from the FNSSMDEVSAVEEXEC function of the device.
1520 *
1521 * @param pVirtio Pointer to the shared virtio state.
1522 * @param pHlp The ring-3 device helpers.
1523 * @param pSSM The saved state handle.
1524 * @returns VBox status code.
1525 */
1526int virtioCoreR3SaveExec(PVIRTIOCORE pVirtio, PCPDMDEVHLPR3 pHlp, PSSMHANDLE pSSM)
1527{
1528 LogFunc(("\n"));
1529 pHlp->pfnSSMPutU64(pSSM, VIRTIO_SAVEDSTATE_MARKER);
1530 pHlp->pfnSSMPutU32(pSSM, VIRTIO_SAVEDSTATE_VERSION);
1531
1532 pHlp->pfnSSMPutBool(pSSM, pVirtio->fGenUpdatePending);
1533 pHlp->pfnSSMPutU8(pSSM, pVirtio->uDeviceStatus);
1534 pHlp->pfnSSMPutU8(pSSM, pVirtio->uConfigGeneration);
1535 pHlp->pfnSSMPutU8(pSSM, pVirtio->uPciCfgDataOff);
1536 pHlp->pfnSSMPutU8(pSSM, pVirtio->uISR);
1537 pHlp->pfnSSMPutU16(pSSM, pVirtio->uQueueSelect);
1538 pHlp->pfnSSMPutU32(pSSM, pVirtio->uDeviceFeaturesSelect);
1539 pHlp->pfnSSMPutU32(pSSM, pVirtio->uDriverFeaturesSelect);
1540 pHlp->pfnSSMPutU64(pSSM, pVirtio->uDriverFeatures);
1541
1542 for (uint32_t i = 0; i < VIRTQ_MAX_CNT; i++)
1543 {
1544 pHlp->pfnSSMPutGCPhys64(pSSM, pVirtio->aGCPhysQueueDesc[i]);
1545 pHlp->pfnSSMPutGCPhys64(pSSM, pVirtio->aGCPhysQueueAvail[i]);
1546 pHlp->pfnSSMPutGCPhys64(pSSM, pVirtio->aGCPhysQueueUsed[i]);
1547 pHlp->pfnSSMPutU16(pSSM, pVirtio->uQueueNotifyOff[i]);
1548 pHlp->pfnSSMPutU16(pSSM, pVirtio->uQueueMsixVector[i]);
1549 pHlp->pfnSSMPutU16(pSSM, pVirtio->uQueueEnable[i]);
1550 pHlp->pfnSSMPutU16(pSSM, pVirtio->uQueueSize[i]);
1551 pHlp->pfnSSMPutU16(pSSM, pVirtio->virtqState[i].uAvailIdx);
1552 pHlp->pfnSSMPutU16(pSSM, pVirtio->virtqState[i].uUsedIdx);
1553 int rc = pHlp->pfnSSMPutMem(pSSM, pVirtio->virtqState[i].szVirtqName, 32);
1554 AssertRCReturn(rc, rc);
1555 }
1556
1557 return VINF_SUCCESS;
1558}
1559
1560/**
1561 * Called from the FNSSMDEVLOADEXEC function of the device.
1562 *
1563 * @param pVirtio Pointer to the shared virtio state.
1564 * @param pHlp The ring-3 device helpers.
1565 * @param pSSM The saved state handle.
1566 * @returns VBox status code.
1567 */
1568int virtioCoreR3LoadExec(PVIRTIOCORE pVirtio, PCPDMDEVHLPR3 pHlp, PSSMHANDLE pSSM)
1569{
1570 LogFunc(("\n"));
1571 /*
1572 * Check the marker and (embedded) version number.
1573 */
1574 uint64_t uMarker = 0;
1575 int rc = pHlp->pfnSSMGetU64(pSSM, &uMarker);
1576 AssertRCReturn(rc, rc);
1577 if (uMarker != VIRTIO_SAVEDSTATE_MARKER)
1578 return pHlp->pfnSSMSetLoadError(pSSM, VERR_SSM_DATA_UNIT_FORMAT_CHANGED, RT_SRC_POS,
1579 N_("Expected marker value %#RX64 found %#RX64 instead"),
1580 VIRTIO_SAVEDSTATE_MARKER, uMarker);
1581 uint32_t uVersion = 0;
1582 rc = pHlp->pfnSSMGetU32(pSSM, &uVersion);
1583 AssertRCReturn(rc, rc);
1584 if (uVersion != VIRTIO_SAVEDSTATE_VERSION)
1585 return pHlp->pfnSSMSetLoadError(pSSM, VERR_SSM_DATA_UNIT_FORMAT_CHANGED, RT_SRC_POS,
1586 N_("Unsupported virtio version: %u"), uVersion);
1587 /*
1588 * Load the state.
1589 */
1590 pHlp->pfnSSMGetBool(pSSM, &pVirtio->fGenUpdatePending);
1591 pHlp->pfnSSMGetU8(pSSM, &pVirtio->uDeviceStatus);
1592 pHlp->pfnSSMGetU8(pSSM, &pVirtio->uConfigGeneration);
1593 pHlp->pfnSSMGetU8(pSSM, &pVirtio->uPciCfgDataOff);
1594 pHlp->pfnSSMGetU8(pSSM, &pVirtio->uISR);
1595 pHlp->pfnSSMGetU16(pSSM, &pVirtio->uQueueSelect);
1596 pHlp->pfnSSMGetU32(pSSM, &pVirtio->uDeviceFeaturesSelect);
1597 pHlp->pfnSSMGetU32(pSSM, &pVirtio->uDriverFeaturesSelect);
1598 pHlp->pfnSSMGetU64(pSSM, &pVirtio->uDriverFeatures);
1599
1600 for (uint32_t i = 0; i < VIRTQ_MAX_CNT; i++)
1601 {
1602 pHlp->pfnSSMGetGCPhys64(pSSM, &pVirtio->aGCPhysQueueDesc[i]);
1603 pHlp->pfnSSMGetGCPhys64(pSSM, &pVirtio->aGCPhysQueueAvail[i]);
1604 pHlp->pfnSSMGetGCPhys64(pSSM, &pVirtio->aGCPhysQueueUsed[i]);
1605 pHlp->pfnSSMGetU16(pSSM, &pVirtio->uQueueNotifyOff[i]);
1606 pHlp->pfnSSMGetU16(pSSM, &pVirtio->uQueueMsixVector[i]);
1607 pHlp->pfnSSMGetU16(pSSM, &pVirtio->uQueueEnable[i]);
1608 pHlp->pfnSSMGetU16(pSSM, &pVirtio->uQueueSize[i]);
1609 pHlp->pfnSSMGetU16(pSSM, &pVirtio->virtqState[i].uAvailIdx);
1610 pHlp->pfnSSMGetU16(pSSM, &pVirtio->virtqState[i].uUsedIdx);
1611 rc = pHlp->pfnSSMGetMem(pSSM, pVirtio->virtqState[i].szVirtqName,
1612 sizeof(pVirtio->virtqState[i].szVirtqName));
1613 AssertRCReturn(rc, rc);
1614 }
1615
1616 return VINF_SUCCESS;
1617}
1618
1619
1620/*********************************************************************************************************************************
1621* Device Level *
1622*********************************************************************************************************************************/
1623
1624/**
1625 * This must be called by the client to handle VM state changes
1626 * after the client takes care of its device-specific tasks for the state change.
1627 * (i.e. Reset, suspend, power-off, resume)
1628 *
1629 * @param pDevIns The device instance.
1630 * @param pVirtio Pointer to the shared virtio state.
1631 */
1632void virtioCoreR3VmStateChanged(PVIRTIOCORE pVirtio, VIRTIOVMSTATECHANGED enmState)
1633{
1634 LogFunc(("State changing to %s\n",
1635 virtioCoreGetStateChangeText(enmState)));
1636
1637 switch(enmState)
1638 {
1639 case kvirtIoVmStateChangedReset:
1640 virtioCoreResetAll(pVirtio);
1641 break;
1642 case kvirtIoVmStateChangedSuspend:
1643 break;
1644 case kvirtIoVmStateChangedPowerOff:
1645 break;
1646 case kvirtIoVmStateChangedResume:
1647 virtioNotifyGuestDriver(pVirtio->pDevIns, pVirtio, 0 /* idxQueue */, true /* fForce */);
1648 break;
1649 default:
1650 LogRelFunc(("Bad enum value"));
1651 return;
1652 }
1653}
1654
1655/**
1656 * This should be called from PDMDEVREGR3::pfnDestruct.
1657 *
1658 * @param pDevIns The device instance.
1659 * @param pVirtio Pointer to the shared virtio state.
1660 * @param pVirtioCC Pointer to the ring-3 virtio state.
1661 */
1662void virtioCoreR3Term(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, PVIRTIOCORECC pVirtioCC)
1663{
1664 if (pVirtioCC->pbPrevDevSpecificCfg)
1665 {
1666 RTMemFree(pVirtioCC->pbPrevDevSpecificCfg);
1667 pVirtioCC->pbPrevDevSpecificCfg = NULL;
1668 }
1669 RT_NOREF(pDevIns, pVirtio);
1670}
1671
1672
1673/**
1674 * Setup PCI device controller and Virtio state
1675 *
1676 * This should be called from PDMDEVREGR3::pfnConstruct.
1677 *
1678 * @param pDevIns The device instance.
1679 * @param pVirtio Pointer to the shared virtio state. This
1680 * must be the first member in the shared
1681 * device instance data!
1682 * @param pVirtioCC Pointer to the ring-3 virtio state. This
1683 * must be the first member in the ring-3
1684 * device instance data!
1685 * @param pPciParams Values to populate industry standard PCI Configuration Space data structure
1686 * @param pcszInstance Device instance name (format-specifier)
1687 * @param fDevSpecificFeatures VirtIO device-specific features offered by
1688 * client
1689 * @param cbDevSpecificCfg Size of virtio_pci_device_cap device-specific struct
1690 * @param pvDevSpecificCfg Address of client's dev-specific
1691 * configuration struct.
1692 */
1693int virtioCoreR3Init(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, PVIRTIOCORECC pVirtioCC, PVIRTIOPCIPARAMS pPciParams,
1694 const char *pcszInstance, uint64_t fDevSpecificFeatures, void *pvDevSpecificCfg, uint16_t cbDevSpecificCfg)
1695{
1696 /*
1697 * The pVirtio state must be the first member of the shared device instance
1698 * data, otherwise we cannot get our bearings in the PCI configuration callbacks.
1699 */
1700 AssertLogRelReturn(pVirtio == PDMINS_2_DATA(pDevIns, PVIRTIOCORE), VERR_STATE_CHANGED);
1701 AssertLogRelReturn(pVirtioCC == PDMINS_2_DATA_CC(pDevIns, PVIRTIOCORECC), VERR_STATE_CHANGED);
1702
1703 pVirtio->pDevIns = pDevIns;
1704
1705 /*
1706 * Caller must initialize these.
1707 */
1708 AssertReturn(pVirtioCC->pfnStatusChanged, VERR_INVALID_POINTER);
1709 AssertReturn(pVirtioCC->pfnQueueNotified, VERR_INVALID_POINTER);
1710 AssertReturn(pVirtioCC->pfnDevCapRead, VERR_INVALID_POINTER);
1711 AssertReturn(pVirtioCC->pfnDevCapWrite, VERR_INVALID_POINTER);
1712
1713#if 0 /* Until pdmR3DvHlp_PCISetIrq() impl is fixed and Assert that limits vec to 0 is removed */
1714# ifdef VBOX_WITH_MSI_DEVICES
1715 pVirtio->fMsiSupport = true;
1716# endif
1717#endif
1718
1719 /*
1720 * The host features offered include both device-specific features
1721 * and reserved feature bits (device independent)
1722 */
1723 pVirtio->uDeviceFeatures = VIRTIO_F_VERSION_1
1724 | VIRTIO_DEV_INDEPENDENT_FEATURES_OFFERED
1725 | fDevSpecificFeatures;
1726
1727 RTStrCopy(pVirtio->szInstance, sizeof(pVirtio->szInstance), pcszInstance);
1728
1729 pVirtio->uDeviceStatus = 0;
1730 pVirtioCC->cbDevSpecificCfg = cbDevSpecificCfg;
1731 pVirtioCC->pbDevSpecificCfg = (uint8_t *)pvDevSpecificCfg;
1732 pVirtioCC->pbPrevDevSpecificCfg = (uint8_t *)RTMemDup(pvDevSpecificCfg, cbDevSpecificCfg);
1733 AssertLogRelReturn(pVirtioCC->pbPrevDevSpecificCfg, VERR_NO_MEMORY);
1734
1735 /* Set PCI config registers (assume 32-bit mode) */
1736 PPDMPCIDEV pPciDev = pDevIns->apPciDevs[0];
1737 PDMPCIDEV_ASSERT_VALID(pDevIns, pPciDev);
1738
1739 PDMPciDevSetRevisionId(pPciDev, DEVICE_PCI_REVISION_ID_VIRTIO);
1740 PDMPciDevSetVendorId(pPciDev, DEVICE_PCI_VENDOR_ID_VIRTIO);
1741 PDMPciDevSetSubSystemVendorId(pPciDev, DEVICE_PCI_VENDOR_ID_VIRTIO);
1742 PDMPciDevSetDeviceId(pPciDev, pPciParams->uDeviceId);
1743 PDMPciDevSetClassBase(pPciDev, pPciParams->uClassBase);
1744 PDMPciDevSetClassSub(pPciDev, pPciParams->uClassSub);
1745 PDMPciDevSetClassProg(pPciDev, pPciParams->uClassProg);
1746 PDMPciDevSetSubSystemId(pPciDev, pPciParams->uSubsystemId);
1747 PDMPciDevSetInterruptLine(pPciDev, pPciParams->uInterruptLine);
1748 PDMPciDevSetInterruptPin(pPciDev, pPciParams->uInterruptPin);
1749
1750 /* Register PCI device */
1751 int rc = PDMDevHlpPCIRegister(pDevIns, pPciDev);
1752 if (RT_FAILURE(rc))
1753 return PDMDEV_SET_ERROR(pDevIns, rc, N_("virtio: cannot register PCI Device")); /* can we put params in this error? */
1754
1755 rc = PDMDevHlpPCIInterceptConfigAccesses(pDevIns, pPciDev, virtioR3PciConfigRead, virtioR3PciConfigWrite);
1756 AssertRCReturn(rc, rc);
1757
1758
1759 /* Construct & map PCI vendor-specific capabilities for virtio host negotiation with guest driver */
1760
1761 /* The following capability mapped via VirtIO 1.0: struct virtio_pci_cfg_cap (VIRTIO_PCI_CFG_CAP_T)
1762 * as a mandatory but suboptimal alternative interface to host device capabilities, facilitating
1763 * access the memory of any BAR. If the guest uses it (the VirtIO driver on Linux doesn't),
1764 * Unlike Common, Notify, ISR and Device capabilities, it is accessed directly via PCI Config region.
1765 * therefore does not contribute to the capabilities region (BAR) the other capabilities use.
1766 */
1767#define CFG_ADDR_2_IDX(addr) ((uint8_t)(((uintptr_t)(addr) - (uintptr_t)&pPciDev->abConfig[0])))
1768#define SET_PCI_CAP_LOC(a_pPciDev, a_pCfg, a_LocCap, a_uMmioLengthAlign) \
1769 do { \
1770 (a_LocCap).offMmio = (a_pCfg)->uOffset; \
1771 (a_LocCap).cbMmio = RT_ALIGN_T((a_pCfg)->uLength, a_uMmioLengthAlign, uint16_t); \
1772 (a_LocCap).offPci = (uint16_t)(uintptr_t)((uint8_t *)(a_pCfg) - &(a_pPciDev)->abConfig[0]); \
1773 (a_LocCap).cbPci = (a_pCfg)->uCapLen; \
1774 } while (0)
1775
1776 PVIRTIO_PCI_CAP_T pCfg;
1777 uint32_t cbRegion = 0;
1778
1779 /* Common capability (VirtIO 1.0 spec, section 4.1.4.3) */
1780 pCfg = (PVIRTIO_PCI_CAP_T)&pPciDev->abConfig[0x40];
1781 pCfg->uCfgType = VIRTIO_PCI_CAP_COMMON_CFG;
1782 pCfg->uCapVndr = VIRTIO_PCI_CAP_ID_VENDOR;
1783 pCfg->uCapLen = sizeof(VIRTIO_PCI_CAP_T);
1784 pCfg->uCapNext = CFG_ADDR_2_IDX(pCfg) + pCfg->uCapLen;
1785 pCfg->uBar = VIRTIO_REGION_PCI_CAP;
1786 pCfg->uOffset = RT_ALIGN_32(0, 4); /* reminder, in case someone changes offset */
1787 pCfg->uLength = sizeof(VIRTIO_PCI_COMMON_CFG_T);
1788 cbRegion += pCfg->uLength;
1789 SET_PCI_CAP_LOC(pPciDev, pCfg, pVirtio->LocCommonCfgCap, 2);
1790 pVirtioCC->pCommonCfgCap = pCfg;
1791
1792 /*
1793 * Notify capability (VirtIO 1.0 spec, section 4.1.4.4). Note: uLength is based the choice
1794 * of this implementation that each queue's uQueueNotifyOff is set equal to (QueueSelect) ordinal
1795 * value of the queue */
1796 pCfg = (PVIRTIO_PCI_CAP_T)&pPciDev->abConfig[pCfg->uCapNext];
1797 pCfg->uCfgType = VIRTIO_PCI_CAP_NOTIFY_CFG;
1798 pCfg->uCapVndr = VIRTIO_PCI_CAP_ID_VENDOR;
1799 pCfg->uCapLen = sizeof(VIRTIO_PCI_NOTIFY_CAP_T);
1800 pCfg->uCapNext = CFG_ADDR_2_IDX(pCfg) + pCfg->uCapLen;
1801 pCfg->uBar = VIRTIO_REGION_PCI_CAP;
1802 pCfg->uOffset = pVirtioCC->pCommonCfgCap->uOffset + pVirtioCC->pCommonCfgCap->uLength;
1803 pCfg->uOffset = RT_ALIGN_32(pCfg->uOffset, 4);
1804
1805
1806 pCfg->uLength = VIRTQ_MAX_CNT * VIRTIO_NOTIFY_OFFSET_MULTIPLIER + 2; /* will change in VirtIO 1.1 */
1807 cbRegion += pCfg->uLength;
1808 SET_PCI_CAP_LOC(pPciDev, pCfg, pVirtio->LocNotifyCap, 1);
1809 pVirtioCC->pNotifyCap = (PVIRTIO_PCI_NOTIFY_CAP_T)pCfg;
1810 pVirtioCC->pNotifyCap->uNotifyOffMultiplier = VIRTIO_NOTIFY_OFFSET_MULTIPLIER;
1811
1812 /* ISR capability (VirtIO 1.0 spec, section 4.1.4.5)
1813 *
1814 * VirtIO 1.0 spec says 8-bit, unaligned in MMIO space. Example/diagram
1815 * of spec shows it as a 32-bit field with upper bits 'reserved'
1816 * Will take spec words more literally than the diagram for now.
1817 */
1818 pCfg = (PVIRTIO_PCI_CAP_T)&pPciDev->abConfig[pCfg->uCapNext];
1819 pCfg->uCfgType = VIRTIO_PCI_CAP_ISR_CFG;
1820 pCfg->uCapVndr = VIRTIO_PCI_CAP_ID_VENDOR;
1821 pCfg->uCapLen = sizeof(VIRTIO_PCI_CAP_T);
1822 pCfg->uCapNext = CFG_ADDR_2_IDX(pCfg) + pCfg->uCapLen;
1823 pCfg->uBar = VIRTIO_REGION_PCI_CAP;
1824 pCfg->uOffset = pVirtioCC->pNotifyCap->pciCap.uOffset + pVirtioCC->pNotifyCap->pciCap.uLength;
1825 pCfg->uOffset = RT_ALIGN_32(pCfg->uOffset, 4);
1826 pCfg->uLength = sizeof(uint8_t);
1827 cbRegion += pCfg->uLength;
1828 SET_PCI_CAP_LOC(pPciDev, pCfg, pVirtio->LocIsrCap, 4);
1829 pVirtioCC->pIsrCap = pCfg;
1830
1831 /* PCI Cfg capability (VirtIO 1.0 spec, section 4.1.4.7)
1832 * This capability doesn't get page-MMIO mapped. Instead uBar, uOffset and uLength are intercepted
1833 * by trapping PCI configuration I/O and get modulated by consumers to locate fetch and read/write
1834 * values from any region. NOTE: The linux driver not only doesn't use this feature, it will not
1835 * even list it as present if uLength isn't non-zero and 4-byte-aligned as the linux driver is
1836 * initializing. */
1837
1838 pVirtio->uPciCfgDataOff = pCfg->uCapNext + RT_OFFSETOF(VIRTIO_PCI_CFG_CAP_T, uPciCfgData);
1839 pCfg = (PVIRTIO_PCI_CAP_T)&pPciDev->abConfig[pCfg->uCapNext];
1840 pCfg->uCfgType = VIRTIO_PCI_CAP_PCI_CFG;
1841 pCfg->uCapVndr = VIRTIO_PCI_CAP_ID_VENDOR;
1842 pCfg->uCapLen = sizeof(VIRTIO_PCI_CFG_CAP_T);
1843 pCfg->uCapNext = (pVirtio->fMsiSupport || pVirtioCC->pbDevSpecificCfg) ? CFG_ADDR_2_IDX(pCfg) + pCfg->uCapLen : 0;
1844 pCfg->uBar = 0;
1845 pCfg->uOffset = 0;
1846 pCfg->uLength = 0;
1847 cbRegion += pCfg->uLength;
1848 SET_PCI_CAP_LOC(pPciDev, pCfg, pVirtio->LocPciCfgCap, 1);
1849 pVirtioCC->pPciCfgCap = (PVIRTIO_PCI_CFG_CAP_T)pCfg;
1850
1851 if (pVirtioCC->pbDevSpecificCfg)
1852 {
1853 /* Following capability (via VirtIO 1.0, section 4.1.4.6). Client defines the
1854 * device-specific config fields struct and passes size to this constructor */
1855 pCfg = (PVIRTIO_PCI_CAP_T)&pPciDev->abConfig[pCfg->uCapNext];
1856 pCfg->uCfgType = VIRTIO_PCI_CAP_DEVICE_CFG;
1857 pCfg->uCapVndr = VIRTIO_PCI_CAP_ID_VENDOR;
1858 pCfg->uCapLen = sizeof(VIRTIO_PCI_CAP_T);
1859 pCfg->uCapNext = pVirtio->fMsiSupport ? CFG_ADDR_2_IDX(pCfg) + pCfg->uCapLen : 0;
1860 pCfg->uBar = VIRTIO_REGION_PCI_CAP;
1861 pCfg->uOffset = pVirtioCC->pIsrCap->uOffset + pVirtioCC->pIsrCap->uLength;
1862 pCfg->uOffset = RT_ALIGN_32(pCfg->uOffset, 4);
1863 pCfg->uLength = cbDevSpecificCfg;
1864 cbRegion += pCfg->uLength;
1865 SET_PCI_CAP_LOC(pPciDev, pCfg, pVirtio->LocDeviceCap, 4);
1866 pVirtioCC->pDeviceCap = pCfg;
1867 }
1868 else
1869 Assert(pVirtio->LocDeviceCap.cbMmio == 0 && pVirtio->LocDeviceCap.cbPci == 0);
1870
1871 if (pVirtio->fMsiSupport)
1872 {
1873 PDMMSIREG aMsiReg;
1874 RT_ZERO(aMsiReg);
1875 aMsiReg.iMsixCapOffset = pCfg->uCapNext;
1876 aMsiReg.iMsixNextOffset = 0;
1877 aMsiReg.iMsixBar = VIRTIO_REGION_MSIX_CAP;
1878 aMsiReg.cMsixVectors = VBOX_MSIX_MAX_ENTRIES;
1879 rc = PDMDevHlpPCIRegisterMsi(pDevIns, &aMsiReg); /* see MsixR3init() */
1880 if (RT_FAILURE(rc))
1881 {
1882 /* See PDMDevHlp.cpp:pdmR3DevHlp_PCIRegisterMsi */
1883 LogFunc(("Failed to configure MSI-X (%Rrc). Reverting to INTx\n", rc));
1884 pVirtio->fMsiSupport = false;
1885 }
1886 else
1887 Log2Func(("Using MSI-X for guest driver notification\n"));
1888 }
1889 else
1890 LogFunc(("MSI-X not available for VBox, using INTx notification\n"));
1891
1892
1893 /* Set offset to first capability and enable PCI dev capabilities */
1894 PDMPciDevSetCapabilityList(pPciDev, 0x40);
1895 PDMPciDevSetStatus(pPciDev, VBOX_PCI_STATUS_CAP_LIST);
1896
1897 /* Linux drivers/virtio/virtio_pci_modern.c tries to map at least a page for the
1898 * 'unknown' device-specific capability without querying the capability to figure
1899 * out size, so pad with an extra page */
1900
1901 rc = PDMDevHlpPCIIORegionCreateMmio(pDevIns, VIRTIO_REGION_PCI_CAP, RT_ALIGN_32(cbRegion + PAGE_SIZE, PAGE_SIZE),
1902 PCI_ADDRESS_SPACE_MEM, virtioMmioWrite, virtioMmioRead, pVirtio,
1903 IOMMMIO_FLAGS_READ_PASSTHRU | IOMMMIO_FLAGS_WRITE_PASSTHRU, "virtio-scsi MMIO",
1904 &pVirtio->hMmioPciCap);
1905 AssertLogRelRCReturn(rc, PDMDEV_SET_ERROR(pDevIns, rc, N_("virtio: cannot register PCI Capabilities address space")));
1906
1907 return rc;
1908}
1909
1910#else /* !IN_RING3 */
1911
1912/**
1913 * Sets up the core ring-0/raw-mode virtio bits.
1914 *
1915 * @returns VBox status code.
1916 * @param pDevIns The device instance.
1917 * @param pVirtio Pointer to the shared virtio state. This must be the first
1918 * member in the shared device instance data!
1919 * @param pVirtioCC Pointer to the current context virtio state. This must be the
1920 * first member in the currenct context's device instance data!
1921 */
1922int virtioCoreRZInit(PPDMDEVINS pDevIns, PVIRTIOCORE pVirtio, PVIRTIOCORECC pVirtioCC)
1923{
1924 AssertLogRelReturn(pVirtio == PDMINS_2_DATA(pDevIns, PVIRTIOCORE), VERR_STATE_CHANGED);
1925 AssertLogRelReturn(pVirtioCC == PDMINS_2_DATA_CC(pDevIns, PVIRTIOCORECC), VERR_STATE_CHANGED);
1926
1927 int rc = PDMDevHlpMmioSetUpContext(pDevIns, pVirtio->hMmioPciCap, virtioMmioWrite, virtioMmioRead, pVirtio);
1928 AssertRCReturn(rc, rc);
1929 return rc;
1930}
1931
1932#endif /* !IN_RING3 */
1933
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