2 * Copyright (c) 2009-2012,2016 Microsoft Corp.
3 * Copyright (c) 2012 NetApp Inc.
4 * Copyright (c) 2012 Citrix Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
34 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/mutex.h>
39 #include <sys/sysctl.h>
40 #include <sys/systm.h>
42 #include <machine/atomic.h>
43 #include <machine/stdarg.h>
45 #include <dev/hyperv/include/hyperv_busdma.h>
46 #include <dev/hyperv/include/vmbus_xact.h>
47 #include <dev/hyperv/vmbus/hyperv_var.h>
48 #include <dev/hyperv/vmbus/vmbus_reg.h>
49 #include <dev/hyperv/vmbus/vmbus_var.h>
50 #include <dev/hyperv/vmbus/vmbus_brvar.h>
51 #include <dev/hyperv/vmbus/vmbus_chanvar.h>
53 static void vmbus_chan_update_evtflagcnt(
55 const struct vmbus_channel *);
56 static int vmbus_chan_close_internal(
57 struct vmbus_channel *);
58 static int vmbus_chan_sysctl_mnf(SYSCTL_HANDLER_ARGS);
59 static void vmbus_chan_sysctl_create(
60 struct vmbus_channel *);
61 static struct vmbus_channel *vmbus_chan_alloc(struct vmbus_softc *);
62 static void vmbus_chan_free(struct vmbus_channel *);
63 static int vmbus_chan_add(struct vmbus_channel *);
64 static void vmbus_chan_cpu_default(struct vmbus_channel *);
65 static int vmbus_chan_release(struct vmbus_channel *);
66 static void vmbus_chan_set_chmap(struct vmbus_channel *);
67 static void vmbus_chan_clear_chmap(struct vmbus_channel *);
68 static void vmbus_chan_detach(struct vmbus_channel *);
69 static bool vmbus_chan_wait_revoke(
70 const struct vmbus_channel *);
72 static void vmbus_chan_ins_prilist(struct vmbus_softc *,
73 struct vmbus_channel *);
74 static void vmbus_chan_rem_prilist(struct vmbus_softc *,
75 struct vmbus_channel *);
76 static void vmbus_chan_ins_list(struct vmbus_softc *,
77 struct vmbus_channel *);
78 static void vmbus_chan_rem_list(struct vmbus_softc *,
79 struct vmbus_channel *);
80 static void vmbus_chan_ins_sublist(struct vmbus_channel *,
81 struct vmbus_channel *);
82 static void vmbus_chan_rem_sublist(struct vmbus_channel *,
83 struct vmbus_channel *);
85 static void vmbus_chan_task(void *, int);
86 static void vmbus_chan_task_nobatch(void *, int);
87 static void vmbus_chan_clrchmap_task(void *, int);
88 static void vmbus_prichan_attach_task(void *, int);
89 static void vmbus_subchan_attach_task(void *, int);
90 static void vmbus_prichan_detach_task(void *, int);
91 static void vmbus_subchan_detach_task(void *, int);
93 static void vmbus_chan_msgproc_choffer(struct vmbus_softc *,
94 const struct vmbus_message *);
95 static void vmbus_chan_msgproc_chrescind(
97 const struct vmbus_message *);
99 static int vmbus_chan_printf(const struct vmbus_channel *,
100 const char *, ...) __printflike(2, 3);
103 * Vmbus channel message processing.
105 static const vmbus_chanmsg_proc_t
106 vmbus_chan_msgprocs[VMBUS_CHANMSG_TYPE_MAX] = {
107 VMBUS_CHANMSG_PROC(CHOFFER, vmbus_chan_msgproc_choffer),
108 VMBUS_CHANMSG_PROC(CHRESCIND, vmbus_chan_msgproc_chrescind),
110 VMBUS_CHANMSG_PROC_WAKEUP(CHOPEN_RESP),
111 VMBUS_CHANMSG_PROC_WAKEUP(GPADL_CONNRESP),
112 VMBUS_CHANMSG_PROC_WAKEUP(GPADL_DISCONNRESP)
116 * Notify host that there are data pending on our TX bufring.
119 vmbus_chan_signal_tx(const struct vmbus_channel *chan)
121 atomic_set_long(chan->ch_evtflag, chan->ch_evtflag_mask);
122 if (chan->ch_txflags & VMBUS_CHAN_TXF_HASMNF)
123 atomic_set_int(chan->ch_montrig, chan->ch_montrig_mask);
125 hypercall_signal_event(chan->ch_monprm_dma.hv_paddr);
129 vmbus_chan_ins_prilist(struct vmbus_softc *sc, struct vmbus_channel *chan)
132 mtx_assert(&sc->vmbus_prichan_lock, MA_OWNED);
133 if (atomic_testandset_int(&chan->ch_stflags,
134 VMBUS_CHAN_ST_ONPRIL_SHIFT))
135 panic("channel is already on the prilist");
136 TAILQ_INSERT_TAIL(&sc->vmbus_prichans, chan, ch_prilink);
140 vmbus_chan_rem_prilist(struct vmbus_softc *sc, struct vmbus_channel *chan)
143 mtx_assert(&sc->vmbus_prichan_lock, MA_OWNED);
144 if (atomic_testandclear_int(&chan->ch_stflags,
145 VMBUS_CHAN_ST_ONPRIL_SHIFT) == 0)
146 panic("channel is not on the prilist");
147 TAILQ_REMOVE(&sc->vmbus_prichans, chan, ch_prilink);
151 vmbus_chan_ins_sublist(struct vmbus_channel *prichan,
152 struct vmbus_channel *chan)
155 mtx_assert(&prichan->ch_subchan_lock, MA_OWNED);
157 if (atomic_testandset_int(&chan->ch_stflags,
158 VMBUS_CHAN_ST_ONSUBL_SHIFT))
159 panic("channel is already on the sublist");
160 TAILQ_INSERT_TAIL(&prichan->ch_subchans, chan, ch_sublink);
162 /* Bump sub-channel count. */
163 prichan->ch_subchan_cnt++;
167 vmbus_chan_rem_sublist(struct vmbus_channel *prichan,
168 struct vmbus_channel *chan)
171 mtx_assert(&prichan->ch_subchan_lock, MA_OWNED);
173 KASSERT(prichan->ch_subchan_cnt > 0,
174 ("invalid subchan_cnt %d", prichan->ch_subchan_cnt));
175 prichan->ch_subchan_cnt--;
177 if (atomic_testandclear_int(&chan->ch_stflags,
178 VMBUS_CHAN_ST_ONSUBL_SHIFT) == 0)
179 panic("channel is not on the sublist");
180 TAILQ_REMOVE(&prichan->ch_subchans, chan, ch_sublink);
184 vmbus_chan_ins_list(struct vmbus_softc *sc, struct vmbus_channel *chan)
187 mtx_assert(&sc->vmbus_chan_lock, MA_OWNED);
188 if (atomic_testandset_int(&chan->ch_stflags,
189 VMBUS_CHAN_ST_ONLIST_SHIFT))
190 panic("channel is already on the list");
191 TAILQ_INSERT_TAIL(&sc->vmbus_chans, chan, ch_link);
195 vmbus_chan_rem_list(struct vmbus_softc *sc, struct vmbus_channel *chan)
198 mtx_assert(&sc->vmbus_chan_lock, MA_OWNED);
199 if (atomic_testandclear_int(&chan->ch_stflags,
200 VMBUS_CHAN_ST_ONLIST_SHIFT) == 0)
201 panic("channel is not on the list");
202 TAILQ_REMOVE(&sc->vmbus_chans, chan, ch_link);
206 vmbus_chan_sysctl_mnf(SYSCTL_HANDLER_ARGS)
208 struct vmbus_channel *chan = arg1;
211 if (chan->ch_txflags & VMBUS_CHAN_TXF_HASMNF)
213 return sysctl_handle_int(oidp, &mnf, 0, req);
217 vmbus_chan_sysctl_create(struct vmbus_channel *chan)
219 struct sysctl_oid *ch_tree, *chid_tree, *br_tree;
220 struct sysctl_ctx_list *ctx;
225 * Add sysctl nodes related to this channel to this
226 * channel's sysctl ctx, so that they can be destroyed
227 * independently upon close of this channel, which can
228 * happen even if the device is not detached.
230 ctx = &chan->ch_sysctl_ctx;
231 sysctl_ctx_init(ctx);
234 * Create dev.NAME.UNIT.channel tree.
236 ch_tree = SYSCTL_ADD_NODE(ctx,
237 SYSCTL_CHILDREN(device_get_sysctl_tree(chan->ch_dev)),
238 OID_AUTO, "channel", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
243 * Create dev.NAME.UNIT.channel.CHANID tree.
245 if (VMBUS_CHAN_ISPRIMARY(chan))
248 ch_id = chan->ch_prichan->ch_id;
249 snprintf(name, sizeof(name), "%d", ch_id);
250 chid_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(ch_tree),
251 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
252 if (chid_tree == NULL)
255 if (!VMBUS_CHAN_ISPRIMARY(chan)) {
257 * Create dev.NAME.UNIT.channel.CHANID.sub tree.
259 ch_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(chid_tree),
260 OID_AUTO, "sub", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
265 * Create dev.NAME.UNIT.channel.CHANID.sub.SUBIDX tree.
268 * chid_tree is changed to this new sysctl tree.
270 snprintf(name, sizeof(name), "%d", chan->ch_subidx);
271 chid_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(ch_tree),
272 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
273 if (chid_tree == NULL)
276 SYSCTL_ADD_UINT(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO,
277 "chanid", CTLFLAG_RD, &chan->ch_id, 0, "channel id");
280 SYSCTL_ADD_UINT(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO,
281 "cpu", CTLFLAG_RD, &chan->ch_cpuid, 0, "owner CPU id");
282 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO,
283 "mnf", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
284 chan, 0, vmbus_chan_sysctl_mnf, "I",
285 "has monitor notification facilities");
287 br_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO,
288 "br", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
289 if (br_tree != NULL) {
291 * Create sysctl tree for RX bufring.
293 vmbus_br_sysctl_create(ctx, br_tree, &chan->ch_rxbr.rxbr, "rx");
295 * Create sysctl tree for TX bufring.
297 vmbus_br_sysctl_create(ctx, br_tree, &chan->ch_txbr.txbr, "tx");
302 vmbus_chan_open(struct vmbus_channel *chan, int txbr_size, int rxbr_size,
303 const void *udata, int udlen, vmbus_chan_callback_t cb, void *cbarg)
305 struct vmbus_chan_br cbr;
309 * Allocate the TX+RX bufrings.
311 KASSERT(chan->ch_bufring == NULL, ("bufrings are allocated"));
312 chan->ch_bufring = hyperv_dmamem_alloc(bus_get_dma_tag(chan->ch_dev),
313 PAGE_SIZE, 0, txbr_size + rxbr_size, &chan->ch_bufring_dma,
315 if (chan->ch_bufring == NULL) {
316 vmbus_chan_printf(chan, "bufring allocation failed\n");
320 cbr.cbr = chan->ch_bufring;
321 cbr.cbr_paddr = chan->ch_bufring_dma.hv_paddr;
322 cbr.cbr_txsz = txbr_size;
323 cbr.cbr_rxsz = rxbr_size;
325 error = vmbus_chan_open_br(chan, &cbr, udata, udlen, cb, cbarg);
327 if (error == EISCONN) {
330 * The bufring GPADL is still connected; abandon
331 * this bufring, instead of having mysterious
332 * crash or trashed data later on.
334 vmbus_chan_printf(chan, "chan%u bufring GPADL "
335 "is still connected upon channel open error; "
336 "leak %d bytes memory\n", chan->ch_id,
337 txbr_size + rxbr_size);
339 hyperv_dmamem_free(&chan->ch_bufring_dma,
342 chan->ch_bufring = NULL;
348 vmbus_chan_open_br(struct vmbus_channel *chan, const struct vmbus_chan_br *cbr,
349 const void *udata, int udlen, vmbus_chan_callback_t cb, void *cbarg)
351 struct vmbus_softc *sc = chan->ch_vmbus;
352 const struct vmbus_chanmsg_chopen_resp *resp;
353 const struct vmbus_message *msg;
354 struct vmbus_chanmsg_chopen *req;
355 struct vmbus_msghc *mh;
357 int error, txbr_size, rxbr_size;
361 if (udlen > VMBUS_CHANMSG_CHOPEN_UDATA_SIZE) {
362 vmbus_chan_printf(chan,
363 "invalid udata len %d for chan%u\n", udlen, chan->ch_id);
368 txbr_size = cbr->cbr_txsz;
369 rxbr_size = cbr->cbr_rxsz;
370 KASSERT((txbr_size & PAGE_MASK) == 0,
371 ("send bufring size is not multiple page"));
372 KASSERT((rxbr_size & PAGE_MASK) == 0,
373 ("recv bufring size is not multiple page"));
374 KASSERT((cbr->cbr_paddr & PAGE_MASK) == 0,
375 ("bufring is not page aligned"));
378 * Zero out the TX/RX bufrings, in case that they were used before.
380 memset(br, 0, txbr_size + rxbr_size);
382 if (atomic_testandset_int(&chan->ch_stflags,
383 VMBUS_CHAN_ST_OPENED_SHIFT))
384 panic("double-open chan%u", chan->ch_id);
387 chan->ch_cbarg = cbarg;
389 vmbus_chan_update_evtflagcnt(sc, chan);
391 chan->ch_tq = VMBUS_PCPU_GET(chan->ch_vmbus, event_tq, chan->ch_cpuid);
392 if (chan->ch_flags & VMBUS_CHAN_FLAG_BATCHREAD)
393 task_fn = vmbus_chan_task;
395 task_fn = vmbus_chan_task_nobatch;
396 TASK_INIT(&chan->ch_task, 0, task_fn, chan);
398 /* TX bufring comes first */
399 vmbus_txbr_setup(&chan->ch_txbr, br, txbr_size);
400 /* RX bufring immediately follows TX bufring */
401 vmbus_rxbr_setup(&chan->ch_rxbr, br + txbr_size, rxbr_size);
403 /* Create sysctl tree for this channel */
404 vmbus_chan_sysctl_create(chan);
407 * Connect the bufrings, both RX and TX, to this channel.
409 KASSERT(chan->ch_bufring_gpadl == 0,
410 ("bufring GPADL is still connected"));
411 error = vmbus_chan_gpadl_connect(chan, cbr->cbr_paddr,
412 txbr_size + rxbr_size, &chan->ch_bufring_gpadl);
414 vmbus_chan_printf(chan,
415 "failed to connect bufring GPADL to chan%u\n", chan->ch_id);
420 * Install this channel, before it is opened, but after everything
421 * else has been setup.
423 vmbus_chan_set_chmap(chan);
426 * Open channel w/ the bufring GPADL on the target CPU.
428 mh = vmbus_msghc_get(sc, sizeof(*req));
430 vmbus_chan_printf(chan,
431 "can not get msg hypercall for chopen(chan%u)\n",
437 req = vmbus_msghc_dataptr(mh);
438 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHOPEN;
439 req->chm_chanid = chan->ch_id;
440 req->chm_openid = chan->ch_id;
441 req->chm_gpadl = chan->ch_bufring_gpadl;
442 req->chm_vcpuid = chan->ch_vcpuid;
443 req->chm_txbr_pgcnt = txbr_size >> PAGE_SHIFT;
445 memcpy(req->chm_udata, udata, udlen);
447 error = vmbus_msghc_exec(sc, mh);
449 vmbus_chan_printf(chan,
450 "chopen(chan%u) msg hypercall exec failed: %d\n",
452 vmbus_msghc_put(sc, mh);
456 msg = vmbus_msghc_wait_result(sc, mh);
457 resp = (const struct vmbus_chanmsg_chopen_resp *)msg->msg_data;
458 status = resp->chm_status;
460 vmbus_msghc_put(sc, mh);
464 vmbus_chan_printf(chan, "chan%u opened\n", chan->ch_id);
468 vmbus_chan_printf(chan, "failed to open chan%u\n", chan->ch_id);
472 sysctl_ctx_free(&chan->ch_sysctl_ctx);
473 vmbus_chan_clear_chmap(chan);
474 if (chan->ch_bufring_gpadl != 0) {
477 error1 = vmbus_chan_gpadl_disconnect(chan,
478 chan->ch_bufring_gpadl);
481 * Give caller a hint that the bufring GPADL is still
486 chan->ch_bufring_gpadl = 0;
488 atomic_clear_int(&chan->ch_stflags, VMBUS_CHAN_ST_OPENED);
493 vmbus_chan_gpadl_connect(struct vmbus_channel *chan, bus_addr_t paddr,
494 int size, uint32_t *gpadl0)
496 struct vmbus_softc *sc = chan->ch_vmbus;
497 struct vmbus_msghc *mh;
498 struct vmbus_chanmsg_gpadl_conn *req;
499 const struct vmbus_message *msg;
501 uint32_t gpadl, status;
502 int page_count, range_len, i, cnt, error;
505 KASSERT(*gpadl0 == 0, ("GPADL is not zero"));
508 * Preliminary checks.
511 KASSERT((size & PAGE_MASK) == 0,
512 ("invalid GPA size %d, not multiple page size", size));
513 page_count = size >> PAGE_SHIFT;
515 KASSERT((paddr & PAGE_MASK) == 0,
516 ("GPA is not page aligned %jx", (uintmax_t)paddr));
517 page_id = paddr >> PAGE_SHIFT;
519 range_len = __offsetof(struct vmbus_gpa_range, gpa_page[page_count]);
521 * We don't support multiple GPA ranges.
523 if (range_len > UINT16_MAX) {
524 vmbus_chan_printf(chan, "GPA too large, %d pages\n",
532 gpadl = vmbus_gpadl_alloc(sc);
535 * Connect this GPADL to the target channel.
538 * Since each message can only hold small set of page
539 * addresses, several messages may be required to
540 * complete the connection.
542 if (page_count > VMBUS_CHANMSG_GPADL_CONN_PGMAX)
543 cnt = VMBUS_CHANMSG_GPADL_CONN_PGMAX;
548 reqsz = __offsetof(struct vmbus_chanmsg_gpadl_conn,
549 chm_range.gpa_page[cnt]);
550 mh = vmbus_msghc_get(sc, reqsz);
552 vmbus_chan_printf(chan,
553 "can not get msg hypercall for gpadl_conn(chan%u)\n",
558 req = vmbus_msghc_dataptr(mh);
559 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_CONN;
560 req->chm_chanid = chan->ch_id;
561 req->chm_gpadl = gpadl;
562 req->chm_range_len = range_len;
563 req->chm_range_cnt = 1;
564 req->chm_range.gpa_len = size;
565 req->chm_range.gpa_ofs = 0;
566 for (i = 0; i < cnt; ++i)
567 req->chm_range.gpa_page[i] = page_id++;
569 error = vmbus_msghc_exec(sc, mh);
571 vmbus_chan_printf(chan,
572 "gpadl_conn(chan%u) msg hypercall exec failed: %d\n",
574 vmbus_msghc_put(sc, mh);
578 while (page_count > 0) {
579 struct vmbus_chanmsg_gpadl_subconn *subreq;
581 if (page_count > VMBUS_CHANMSG_GPADL_SUBCONN_PGMAX)
582 cnt = VMBUS_CHANMSG_GPADL_SUBCONN_PGMAX;
587 reqsz = __offsetof(struct vmbus_chanmsg_gpadl_subconn,
589 vmbus_msghc_reset(mh, reqsz);
591 subreq = vmbus_msghc_dataptr(mh);
592 subreq->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_SUBCONN;
593 subreq->chm_gpadl = gpadl;
594 for (i = 0; i < cnt; ++i)
595 subreq->chm_gpa_page[i] = page_id++;
597 vmbus_msghc_exec_noresult(mh);
599 KASSERT(page_count == 0, ("invalid page count %d", page_count));
601 msg = vmbus_msghc_wait_result(sc, mh);
602 status = ((const struct vmbus_chanmsg_gpadl_connresp *)
603 msg->msg_data)->chm_status;
605 vmbus_msghc_put(sc, mh);
608 vmbus_chan_printf(chan, "gpadl_conn(chan%u) failed: %u\n",
609 chan->ch_id, status);
613 /* Done; commit the GPADL id. */
616 vmbus_chan_printf(chan, "gpadl_conn(chan%u) succeeded\n",
623 vmbus_chan_wait_revoke(const struct vmbus_channel *chan)
625 #define WAIT_COUNT 200 /* 200ms */
629 for (i = 0; i < WAIT_COUNT; ++i) {
630 if (vmbus_chan_is_revoked(chan))
632 /* Not sure about the context; use busy-wait. */
641 * Disconnect the GPA from the target channel
644 vmbus_chan_gpadl_disconnect(struct vmbus_channel *chan, uint32_t gpadl)
646 struct vmbus_softc *sc = chan->ch_vmbus;
647 struct vmbus_msghc *mh;
648 struct vmbus_chanmsg_gpadl_disconn *req;
651 KASSERT(gpadl != 0, ("GPADL is zero"));
653 mh = vmbus_msghc_get(sc, sizeof(*req));
655 vmbus_chan_printf(chan,
656 "can not get msg hypercall for gpadl_disconn(chan%u)\n",
661 req = vmbus_msghc_dataptr(mh);
662 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_DISCONN;
663 req->chm_chanid = chan->ch_id;
664 req->chm_gpadl = gpadl;
666 error = vmbus_msghc_exec(sc, mh);
668 vmbus_msghc_put(sc, mh);
670 if (vmbus_chan_wait_revoke(chan)) {
672 * Error is benign; this channel is revoked,
673 * so this GPADL will not be touched anymore.
675 vmbus_chan_printf(chan,
676 "gpadl_disconn(revoked chan%u) msg hypercall "
677 "exec failed: %d\n", chan->ch_id, error);
680 vmbus_chan_printf(chan,
681 "gpadl_disconn(chan%u) msg hypercall exec failed: %d\n",
686 vmbus_msghc_wait_result(sc, mh);
687 /* Discard result; no useful information */
688 vmbus_msghc_put(sc, mh);
694 vmbus_chan_detach(struct vmbus_channel *chan)
698 KASSERT(chan->ch_refs > 0, ("chan%u: invalid refcnt %d",
699 chan->ch_id, chan->ch_refs));
700 refs = atomic_fetchadd_int(&chan->ch_refs, -1);
702 if (VMBUS_CHAN_ISPRIMARY(chan)) {
703 KASSERT(refs == 1, ("chan%u: invalid refcnt %d for prichan",
704 chan->ch_id, refs + 1));
709 * Detach the target channel.
712 vmbus_chan_printf(chan, "chan%u detached\n",
715 taskqueue_enqueue(chan->ch_mgmt_tq, &chan->ch_detach_task);
720 vmbus_chan_clrchmap_task(void *xchan, int pending __unused)
722 struct vmbus_channel *chan = xchan;
725 chan->ch_vmbus->vmbus_chmap[chan->ch_id] = NULL;
730 vmbus_chan_clear_chmap(struct vmbus_channel *chan)
732 struct task chmap_task;
734 TASK_INIT(&chmap_task, 0, vmbus_chan_clrchmap_task, chan);
735 taskqueue_enqueue(chan->ch_tq, &chmap_task);
736 taskqueue_drain(chan->ch_tq, &chmap_task);
740 vmbus_chan_set_chmap(struct vmbus_channel *chan)
743 chan->ch_vmbus->vmbus_chmap[chan->ch_id] = chan;
747 vmbus_chan_close_internal(struct vmbus_channel *chan)
749 struct vmbus_softc *sc = chan->ch_vmbus;
750 struct vmbus_msghc *mh;
751 struct vmbus_chanmsg_chclose *req;
752 uint32_t old_stflags;
757 * Sub-channels are closed upon their primary channel closing,
758 * so they can be closed even before they are opened.
761 old_stflags = chan->ch_stflags;
762 if (atomic_cmpset_int(&chan->ch_stflags, old_stflags,
763 old_stflags & ~VMBUS_CHAN_ST_OPENED))
766 if ((old_stflags & VMBUS_CHAN_ST_OPENED) == 0) {
767 /* Not opened yet; done */
769 vmbus_chan_printf(chan, "chan%u not opened\n",
776 * Free this channel's sysctl tree attached to its device's
779 sysctl_ctx_free(&chan->ch_sysctl_ctx);
783 * Order is critical. This channel _must_ be uninstalled first,
784 * else the channel task may be enqueued by the IDT after it has
787 vmbus_chan_clear_chmap(chan);
788 taskqueue_drain(chan->ch_tq, &chan->ch_task);
792 * Close this channel.
794 mh = vmbus_msghc_get(sc, sizeof(*req));
796 vmbus_chan_printf(chan,
797 "can not get msg hypercall for chclose(chan%u)\n",
803 req = vmbus_msghc_dataptr(mh);
804 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHCLOSE;
805 req->chm_chanid = chan->ch_id;
807 error = vmbus_msghc_exec_noresult(mh);
808 vmbus_msghc_put(sc, mh);
811 vmbus_chan_printf(chan,
812 "chclose(chan%u) msg hypercall exec failed: %d\n",
818 vmbus_chan_printf(chan, "chan%u closed\n", chan->ch_id);
822 * Disconnect the TX+RX bufrings from this channel.
824 if (chan->ch_bufring_gpadl != 0) {
827 error1 = vmbus_chan_gpadl_disconnect(chan,
828 chan->ch_bufring_gpadl);
832 * The bufring GPADL is still connected; abandon
833 * this bufring, instead of having mysterious
834 * crash or trashed data later on.
836 vmbus_chan_printf(chan, "chan%u bufring GPADL "
837 "is still connected after close\n", chan->ch_id);
838 chan->ch_bufring = NULL;
840 * Give caller a hint that the bufring GPADL is
845 chan->ch_bufring_gpadl = 0;
849 * Destroy the TX+RX bufrings.
851 if (chan->ch_bufring != NULL) {
852 hyperv_dmamem_free(&chan->ch_bufring_dma, chan->ch_bufring);
853 chan->ch_bufring = NULL;
859 vmbus_chan_close_direct(struct vmbus_channel *chan)
864 if (VMBUS_CHAN_ISPRIMARY(chan)) {
865 struct vmbus_channel *subchan;
868 * All sub-channels _must_ have been closed, or are _not_
871 mtx_lock(&chan->ch_subchan_lock);
872 TAILQ_FOREACH(subchan, &chan->ch_subchans, ch_sublink) {
874 (subchan->ch_stflags & VMBUS_CHAN_ST_OPENED) == 0,
875 ("chan%u: subchan%u is still opened",
876 chan->ch_id, subchan->ch_subidx));
878 mtx_unlock(&chan->ch_subchan_lock);
882 error = vmbus_chan_close_internal(chan);
883 if (!VMBUS_CHAN_ISPRIMARY(chan)) {
885 * This sub-channel is referenced, when it is linked to
886 * the primary channel; drop that reference now.
888 vmbus_chan_detach(chan);
894 * Caller should make sure that all sub-channels have
895 * been added to 'chan' and all to-be-closed channels
896 * are not being opened.
899 vmbus_chan_close(struct vmbus_channel *chan)
903 if (!VMBUS_CHAN_ISPRIMARY(chan)) {
905 * Sub-channel is closed when its primary channel
912 * Close all sub-channels, if any.
914 subchan_cnt = chan->ch_subchan_cnt;
915 if (subchan_cnt > 0) {
916 struct vmbus_channel **subchan;
919 subchan = vmbus_subchan_get(chan, subchan_cnt);
920 for (i = 0; i < subchan_cnt; ++i) {
921 vmbus_chan_close_internal(subchan[i]);
923 * This sub-channel is referenced, when it is
924 * linked to the primary channel; drop that
927 vmbus_chan_detach(subchan[i]);
929 vmbus_subchan_rel(subchan, subchan_cnt);
932 /* Then close the primary channel. */
933 vmbus_chan_close_internal(chan);
937 vmbus_chan_intr_drain(struct vmbus_channel *chan)
940 taskqueue_drain(chan->ch_tq, &chan->ch_task);
944 vmbus_chan_send(struct vmbus_channel *chan, uint16_t type, uint16_t flags,
945 void *data, int dlen, uint64_t xactid)
947 struct vmbus_chanpkt pkt;
948 int pktlen, pad_pktlen, hlen, error;
954 pktlen = hlen + dlen;
955 pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen);
956 KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr),
957 ("invalid packet size %d", pad_pktlen));
959 pkt.cp_hdr.cph_type = type;
960 pkt.cp_hdr.cph_flags = flags;
961 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen);
962 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen);
963 pkt.cp_hdr.cph_xactid = xactid;
965 iov[0].iov_base = &pkt;
966 iov[0].iov_len = hlen;
967 iov[1].iov_base = data;
968 iov[1].iov_len = dlen;
969 iov[2].iov_base = &pad;
970 iov[2].iov_len = pad_pktlen - pktlen;
972 error = vmbus_txbr_write(&chan->ch_txbr, iov, 3, &send_evt);
973 if (!error && send_evt)
974 vmbus_chan_signal_tx(chan);
979 vmbus_chan_send_sglist(struct vmbus_channel *chan,
980 struct vmbus_gpa sg[], int sglen, void *data, int dlen, uint64_t xactid)
982 struct vmbus_chanpkt_sglist pkt;
983 int pktlen, pad_pktlen, hlen, error;
988 hlen = __offsetof(struct vmbus_chanpkt_sglist, cp_gpa[sglen]);
989 pktlen = hlen + dlen;
990 pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen);
991 KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr),
992 ("invalid packet size %d", pad_pktlen));
994 pkt.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
995 pkt.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
996 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen);
997 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen);
998 pkt.cp_hdr.cph_xactid = xactid;
1000 pkt.cp_gpa_cnt = sglen;
1002 iov[0].iov_base = &pkt;
1003 iov[0].iov_len = sizeof(pkt);
1004 iov[1].iov_base = sg;
1005 iov[1].iov_len = sizeof(struct vmbus_gpa) * sglen;
1006 iov[2].iov_base = data;
1007 iov[2].iov_len = dlen;
1008 iov[3].iov_base = &pad;
1009 iov[3].iov_len = pad_pktlen - pktlen;
1011 error = vmbus_txbr_write(&chan->ch_txbr, iov, 4, &send_evt);
1012 if (!error && send_evt)
1013 vmbus_chan_signal_tx(chan);
1018 vmbus_chan_send_prplist(struct vmbus_channel *chan,
1019 struct vmbus_gpa_range *prp, int prp_cnt, void *data, int dlen,
1022 struct vmbus_chanpkt_prplist pkt;
1023 int pktlen, pad_pktlen, hlen, error;
1024 struct iovec iov[4];
1028 hlen = __offsetof(struct vmbus_chanpkt_prplist,
1029 cp_range[0].gpa_page[prp_cnt]);
1030 pktlen = hlen + dlen;
1031 pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen);
1032 KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr),
1033 ("invalid packet size %d", pad_pktlen));
1035 pkt.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
1036 pkt.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
1037 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen);
1038 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen);
1039 pkt.cp_hdr.cph_xactid = xactid;
1041 pkt.cp_range_cnt = 1;
1043 iov[0].iov_base = &pkt;
1044 iov[0].iov_len = sizeof(pkt);
1045 iov[1].iov_base = prp;
1046 iov[1].iov_len = __offsetof(struct vmbus_gpa_range, gpa_page[prp_cnt]);
1047 iov[2].iov_base = data;
1048 iov[2].iov_len = dlen;
1049 iov[3].iov_base = &pad;
1050 iov[3].iov_len = pad_pktlen - pktlen;
1052 error = vmbus_txbr_write(&chan->ch_txbr, iov, 4, &send_evt);
1053 if (!error && send_evt)
1054 vmbus_chan_signal_tx(chan);
1059 vmbus_chan_recv(struct vmbus_channel *chan, void *data, int *dlen0,
1062 struct vmbus_chanpkt_hdr pkt;
1063 int error, dlen, hlen;
1065 error = vmbus_rxbr_peek(&chan->ch_rxbr, &pkt, sizeof(pkt));
1069 if (__predict_false(pkt.cph_hlen < VMBUS_CHANPKT_HLEN_MIN)) {
1070 vmbus_chan_printf(chan, "invalid hlen %u\n", pkt.cph_hlen);
1071 /* XXX this channel is dead actually. */
1074 if (__predict_false(pkt.cph_hlen > pkt.cph_tlen)) {
1075 vmbus_chan_printf(chan, "invalid hlen %u and tlen %u\n",
1076 pkt.cph_hlen, pkt.cph_tlen);
1077 /* XXX this channel is dead actually. */
1081 hlen = VMBUS_CHANPKT_GETLEN(pkt.cph_hlen);
1082 dlen = VMBUS_CHANPKT_GETLEN(pkt.cph_tlen) - hlen;
1084 if (*dlen0 < dlen) {
1085 /* Return the size of this packet's data. */
1090 *xactid = pkt.cph_xactid;
1093 /* Skip packet header */
1094 error = vmbus_rxbr_read(&chan->ch_rxbr, data, dlen, hlen);
1095 KASSERT(!error, ("vmbus_rxbr_read failed"));
1101 vmbus_chan_recv_pkt(struct vmbus_channel *chan,
1102 struct vmbus_chanpkt_hdr *pkt, int *pktlen0)
1104 int error, pktlen, pkt_hlen;
1106 pkt_hlen = sizeof(*pkt);
1107 error = vmbus_rxbr_peek(&chan->ch_rxbr, pkt, pkt_hlen);
1111 if (__predict_false(pkt->cph_hlen < VMBUS_CHANPKT_HLEN_MIN)) {
1112 vmbus_chan_printf(chan, "invalid hlen %u\n", pkt->cph_hlen);
1113 /* XXX this channel is dead actually. */
1116 if (__predict_false(pkt->cph_hlen > pkt->cph_tlen)) {
1117 vmbus_chan_printf(chan, "invalid hlen %u and tlen %u\n",
1118 pkt->cph_hlen, pkt->cph_tlen);
1119 /* XXX this channel is dead actually. */
1123 pktlen = VMBUS_CHANPKT_GETLEN(pkt->cph_tlen);
1124 if (*pktlen0 < pktlen) {
1125 /* Return the size of this packet. */
1132 * Skip the fixed-size packet header, which has been filled
1133 * by the above vmbus_rxbr_peek().
1135 error = vmbus_rxbr_read(&chan->ch_rxbr, pkt + 1,
1136 pktlen - pkt_hlen, pkt_hlen);
1137 KASSERT(!error, ("vmbus_rxbr_read failed"));
1143 vmbus_chan_task(void *xchan, int pending __unused)
1145 struct vmbus_channel *chan = xchan;
1146 vmbus_chan_callback_t cb = chan->ch_cb;
1147 void *cbarg = chan->ch_cbarg;
1150 * Optimize host to guest signaling by ensuring:
1151 * 1. While reading the channel, we disable interrupts from
1153 * 2. Ensure that we process all posted messages from the host
1154 * before returning from this callback.
1155 * 3. Once we return, enable signaling from the host. Once this
1156 * state is set we check to see if additional packets are
1157 * available to read. In this case we repeat the process.
1159 * NOTE: Interrupt has been disabled in the ISR.
1166 left = vmbus_rxbr_intr_unmask(&chan->ch_rxbr);
1168 /* No more data in RX bufring; done */
1171 vmbus_rxbr_intr_mask(&chan->ch_rxbr);
1176 vmbus_chan_task_nobatch(void *xchan, int pending __unused)
1178 struct vmbus_channel *chan = xchan;
1180 chan->ch_cb(chan, chan->ch_cbarg);
1183 static __inline void
1184 vmbus_event_flags_proc(struct vmbus_softc *sc, volatile u_long *event_flags,
1189 for (f = 0; f < flag_cnt; ++f) {
1194 if (event_flags[f] == 0)
1197 flags = atomic_swap_long(&event_flags[f], 0);
1198 chid_base = f << VMBUS_EVTFLAG_SHIFT;
1200 while ((chid_ofs = ffsl(flags)) != 0) {
1201 struct vmbus_channel *chan;
1203 --chid_ofs; /* NOTE: ffsl is 1-based */
1204 flags &= ~(1UL << chid_ofs);
1206 chan = sc->vmbus_chmap[chid_base + chid_ofs];
1207 if (__predict_false(chan == NULL)) {
1208 /* Channel is closed. */
1211 __compiler_membar();
1213 if (chan->ch_flags & VMBUS_CHAN_FLAG_BATCHREAD)
1214 vmbus_rxbr_intr_mask(&chan->ch_rxbr);
1215 taskqueue_enqueue(chan->ch_tq, &chan->ch_task);
1221 vmbus_event_proc(struct vmbus_softc *sc, int cpu)
1223 struct vmbus_evtflags *eventf;
1226 * On Host with Win8 or above, the event page can be checked directly
1227 * to get the id of the channel that has the pending interrupt.
1229 eventf = VMBUS_PCPU_GET(sc, event_flags, cpu) + VMBUS_SINT_MESSAGE;
1230 vmbus_event_flags_proc(sc, eventf->evt_flags,
1231 VMBUS_PCPU_GET(sc, event_flags_cnt, cpu));
1235 vmbus_event_proc_compat(struct vmbus_softc *sc, int cpu)
1237 struct vmbus_evtflags *eventf;
1239 eventf = VMBUS_PCPU_GET(sc, event_flags, cpu) + VMBUS_SINT_MESSAGE;
1240 if (atomic_testandclear_long(&eventf->evt_flags[0], 0)) {
1241 vmbus_event_flags_proc(sc, sc->vmbus_rx_evtflags,
1242 VMBUS_CHAN_MAX_COMPAT >> VMBUS_EVTFLAG_SHIFT);
1247 vmbus_chan_update_evtflagcnt(struct vmbus_softc *sc,
1248 const struct vmbus_channel *chan)
1250 volatile int *flag_cnt_ptr;
1253 flag_cnt = (chan->ch_id / VMBUS_EVTFLAG_LEN) + 1;
1254 flag_cnt_ptr = VMBUS_PCPU_PTR(sc, event_flags_cnt, chan->ch_cpuid);
1259 old_flag_cnt = *flag_cnt_ptr;
1260 if (old_flag_cnt >= flag_cnt)
1262 if (atomic_cmpset_int(flag_cnt_ptr, old_flag_cnt, flag_cnt)) {
1264 vmbus_chan_printf(chan,
1265 "chan%u update cpu%d flag_cnt to %d\n",
1266 chan->ch_id, chan->ch_cpuid, flag_cnt);
1273 static struct vmbus_channel *
1274 vmbus_chan_alloc(struct vmbus_softc *sc)
1276 struct vmbus_channel *chan;
1278 chan = malloc(sizeof(*chan), M_DEVBUF, M_WAITOK | M_ZERO);
1280 chan->ch_monprm = hyperv_dmamem_alloc(bus_get_dma_tag(sc->vmbus_dev),
1281 HYPERCALL_PARAM_ALIGN, 0, sizeof(struct hyperv_mon_param),
1282 &chan->ch_monprm_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
1283 if (chan->ch_monprm == NULL) {
1284 device_printf(sc->vmbus_dev, "monprm alloc failed\n");
1285 free(chan, M_DEVBUF);
1290 chan->ch_vmbus = sc;
1291 mtx_init(&chan->ch_subchan_lock, "vmbus subchan", NULL, MTX_DEF);
1292 sx_init(&chan->ch_orphan_lock, "vmbus chorphan");
1293 TAILQ_INIT(&chan->ch_subchans);
1294 vmbus_rxbr_init(&chan->ch_rxbr);
1295 vmbus_txbr_init(&chan->ch_txbr);
1301 vmbus_chan_free(struct vmbus_channel *chan)
1304 KASSERT(TAILQ_EMPTY(&chan->ch_subchans) && chan->ch_subchan_cnt == 0,
1305 ("still owns sub-channels"));
1306 KASSERT((chan->ch_stflags &
1307 (VMBUS_CHAN_ST_OPENED |
1308 VMBUS_CHAN_ST_ONPRIL |
1309 VMBUS_CHAN_ST_ONSUBL |
1310 VMBUS_CHAN_ST_ONLIST)) == 0, ("free busy channel"));
1311 KASSERT(chan->ch_orphan_xact == NULL,
1312 ("still has orphan xact installed"));
1313 KASSERT(chan->ch_refs == 0, ("chan%u: invalid refcnt %d",
1314 chan->ch_id, chan->ch_refs));
1316 hyperv_dmamem_free(&chan->ch_monprm_dma, chan->ch_monprm);
1317 mtx_destroy(&chan->ch_subchan_lock);
1318 sx_destroy(&chan->ch_orphan_lock);
1319 vmbus_rxbr_deinit(&chan->ch_rxbr);
1320 vmbus_txbr_deinit(&chan->ch_txbr);
1321 free(chan, M_DEVBUF);
1325 vmbus_chan_add(struct vmbus_channel *newchan)
1327 struct vmbus_softc *sc = newchan->ch_vmbus;
1328 struct vmbus_channel *prichan;
1330 if (newchan->ch_id == 0) {
1333 * Chan0 will neither be processed nor should be offered;
1336 device_printf(sc->vmbus_dev, "got chan0 offer, discard\n");
1338 } else if (newchan->ch_id >= VMBUS_CHAN_MAX) {
1339 device_printf(sc->vmbus_dev, "invalid chan%u offer\n",
1344 mtx_lock(&sc->vmbus_prichan_lock);
1345 TAILQ_FOREACH(prichan, &sc->vmbus_prichans, ch_prilink) {
1347 * Sub-channel will have the same type GUID and instance
1348 * GUID as its primary channel.
1350 if (memcmp(&prichan->ch_guid_type, &newchan->ch_guid_type,
1351 sizeof(struct hyperv_guid)) == 0 &&
1352 memcmp(&prichan->ch_guid_inst, &newchan->ch_guid_inst,
1353 sizeof(struct hyperv_guid)) == 0)
1356 if (VMBUS_CHAN_ISPRIMARY(newchan)) {
1357 if (prichan == NULL) {
1358 /* Install the new primary channel */
1359 vmbus_chan_ins_prilist(sc, newchan);
1360 mtx_unlock(&sc->vmbus_prichan_lock);
1363 mtx_unlock(&sc->vmbus_prichan_lock);
1364 device_printf(sc->vmbus_dev,
1365 "duplicated primary chan%u\n", newchan->ch_id);
1368 } else { /* Sub-channel */
1369 if (prichan == NULL) {
1370 mtx_unlock(&sc->vmbus_prichan_lock);
1371 device_printf(sc->vmbus_dev,
1372 "no primary chan for chan%u\n", newchan->ch_id);
1376 * Found the primary channel for this sub-channel and
1379 * XXX refcnt prichan
1382 mtx_unlock(&sc->vmbus_prichan_lock);
1385 * This is a sub-channel; link it with the primary channel.
1387 KASSERT(!VMBUS_CHAN_ISPRIMARY(newchan),
1388 ("new channel is not sub-channel"));
1389 KASSERT(prichan != NULL, ("no primary channel"));
1392 * Reference count this sub-channel; it will be dereferenced
1393 * when this sub-channel is closed.
1395 KASSERT(newchan->ch_refs == 1, ("chan%u: invalid refcnt %d",
1396 newchan->ch_id, newchan->ch_refs));
1397 atomic_add_int(&newchan->ch_refs, 1);
1399 newchan->ch_prichan = prichan;
1400 newchan->ch_dev = prichan->ch_dev;
1402 mtx_lock(&prichan->ch_subchan_lock);
1403 vmbus_chan_ins_sublist(prichan, newchan);
1404 mtx_unlock(&prichan->ch_subchan_lock);
1406 * Notify anyone that is interested in this sub-channel,
1407 * after this sub-channel is setup.
1412 * Hook this channel up for later revocation.
1414 mtx_lock(&sc->vmbus_chan_lock);
1415 vmbus_chan_ins_list(sc, newchan);
1416 mtx_unlock(&sc->vmbus_chan_lock);
1419 vmbus_chan_printf(newchan, "chan%u subidx%u offer\n",
1420 newchan->ch_id, newchan->ch_subidx);
1423 /* Select default cpu for this channel. */
1424 vmbus_chan_cpu_default(newchan);
1430 vmbus_chan_cpu_set(struct vmbus_channel *chan, int cpu)
1432 KASSERT(cpu >= 0 && cpu < mp_ncpus, ("invalid cpu %d", cpu));
1434 if (chan->ch_vmbus->vmbus_version == VMBUS_VERSION_WS2008 ||
1435 chan->ch_vmbus->vmbus_version == VMBUS_VERSION_WIN7) {
1436 /* Only cpu0 is supported */
1440 chan->ch_cpuid = cpu;
1441 chan->ch_vcpuid = VMBUS_PCPU_GET(chan->ch_vmbus, vcpuid, cpu);
1444 vmbus_chan_printf(chan,
1445 "chan%u assigned to cpu%u [vcpu%u]\n",
1446 chan->ch_id, chan->ch_cpuid, chan->ch_vcpuid);
1451 vmbus_chan_cpu_rr(struct vmbus_channel *chan)
1453 static uint32_t vmbus_chan_nextcpu;
1456 cpu = atomic_fetchadd_int(&vmbus_chan_nextcpu, 1) % mp_ncpus;
1457 vmbus_chan_cpu_set(chan, cpu);
1461 vmbus_chan_cpu_default(struct vmbus_channel *chan)
1464 * By default, pin the channel to cpu0. Devices having
1465 * special channel-cpu mapping requirement should call
1466 * vmbus_chan_cpu_{set,rr}().
1468 vmbus_chan_cpu_set(chan, 0);
1472 vmbus_chan_msgproc_choffer(struct vmbus_softc *sc,
1473 const struct vmbus_message *msg)
1475 const struct vmbus_chanmsg_choffer *offer;
1476 struct vmbus_channel *chan;
1477 task_fn_t *detach_fn, *attach_fn;
1480 offer = (const struct vmbus_chanmsg_choffer *)msg->msg_data;
1482 chan = vmbus_chan_alloc(sc);
1484 device_printf(sc->vmbus_dev, "allocate chan%u failed\n",
1489 chan->ch_id = offer->chm_chanid;
1490 chan->ch_subidx = offer->chm_subidx;
1491 chan->ch_guid_type = offer->chm_chtype;
1492 chan->ch_guid_inst = offer->chm_chinst;
1494 /* Batch reading is on by default */
1495 chan->ch_flags |= VMBUS_CHAN_FLAG_BATCHREAD;
1497 chan->ch_monprm->mp_connid = VMBUS_CONNID_EVENT;
1498 if (sc->vmbus_version != VMBUS_VERSION_WS2008)
1499 chan->ch_monprm->mp_connid = offer->chm_connid;
1501 if (offer->chm_flags1 & VMBUS_CHOFFER_FLAG1_HASMNF) {
1507 chan->ch_txflags |= VMBUS_CHAN_TXF_HASMNF;
1509 trig_idx = offer->chm_montrig / VMBUS_MONTRIG_LEN;
1510 if (trig_idx >= VMBUS_MONTRIGS_MAX)
1511 panic("invalid monitor trigger %u", offer->chm_montrig);
1513 &sc->vmbus_mnf2->mnf_trigs[trig_idx].mt_pending;
1515 chan->ch_montrig_mask =
1516 1 << (offer->chm_montrig % VMBUS_MONTRIG_LEN);
1523 &sc->vmbus_tx_evtflags[chan->ch_id >> VMBUS_EVTFLAG_SHIFT];
1524 chan->ch_evtflag_mask = 1UL << (chan->ch_id & VMBUS_EVTFLAG_MASK);
1527 * Setup attach and detach tasks.
1529 if (VMBUS_CHAN_ISPRIMARY(chan)) {
1530 chan->ch_mgmt_tq = sc->vmbus_devtq;
1531 attach_fn = vmbus_prichan_attach_task;
1532 detach_fn = vmbus_prichan_detach_task;
1534 chan->ch_mgmt_tq = sc->vmbus_subchtq;
1535 attach_fn = vmbus_subchan_attach_task;
1536 detach_fn = vmbus_subchan_detach_task;
1538 TASK_INIT(&chan->ch_attach_task, 0, attach_fn, chan);
1539 TASK_INIT(&chan->ch_detach_task, 0, detach_fn, chan);
1541 error = vmbus_chan_add(chan);
1543 device_printf(sc->vmbus_dev, "add chan%u failed: %d\n",
1544 chan->ch_id, error);
1545 atomic_subtract_int(&chan->ch_refs, 1);
1546 vmbus_chan_free(chan);
1549 taskqueue_enqueue(chan->ch_mgmt_tq, &chan->ch_attach_task);
1553 vmbus_chan_msgproc_chrescind(struct vmbus_softc *sc,
1554 const struct vmbus_message *msg)
1556 const struct vmbus_chanmsg_chrescind *note;
1557 struct vmbus_channel *chan;
1559 note = (const struct vmbus_chanmsg_chrescind *)msg->msg_data;
1560 if (note->chm_chanid > VMBUS_CHAN_MAX) {
1561 device_printf(sc->vmbus_dev, "invalid revoked chan%u\n",
1567 * Find and remove the target channel from the channel list.
1569 mtx_lock(&sc->vmbus_chan_lock);
1570 TAILQ_FOREACH(chan, &sc->vmbus_chans, ch_link) {
1571 if (chan->ch_id == note->chm_chanid)
1575 mtx_unlock(&sc->vmbus_chan_lock);
1576 device_printf(sc->vmbus_dev, "chan%u is not offered\n",
1580 vmbus_chan_rem_list(sc, chan);
1581 mtx_unlock(&sc->vmbus_chan_lock);
1583 if (VMBUS_CHAN_ISPRIMARY(chan)) {
1585 * The target channel is a primary channel; remove the
1586 * target channel from the primary channel list now,
1587 * instead of later, so that it will not be found by
1588 * other sub-channel offers, which are processed in
1591 mtx_lock(&sc->vmbus_prichan_lock);
1592 vmbus_chan_rem_prilist(sc, chan);
1593 mtx_unlock(&sc->vmbus_prichan_lock);
1598 * The following processing order is critical:
1599 * Set the REVOKED state flag before orphaning the installed xact.
1602 if (atomic_testandset_int(&chan->ch_stflags,
1603 VMBUS_CHAN_ST_REVOKED_SHIFT))
1604 panic("channel has already been revoked");
1606 sx_xlock(&chan->ch_orphan_lock);
1607 if (chan->ch_orphan_xact != NULL)
1608 vmbus_xact_ctx_orphan(chan->ch_orphan_xact);
1609 sx_xunlock(&chan->ch_orphan_lock);
1612 vmbus_chan_printf(chan, "chan%u revoked\n", note->chm_chanid);
1613 vmbus_chan_detach(chan);
1617 vmbus_chan_release(struct vmbus_channel *chan)
1619 struct vmbus_softc *sc = chan->ch_vmbus;
1620 struct vmbus_chanmsg_chfree *req;
1621 struct vmbus_msghc *mh;
1624 mh = vmbus_msghc_get(sc, sizeof(*req));
1626 vmbus_chan_printf(chan,
1627 "can not get msg hypercall for chfree(chan%u)\n",
1632 req = vmbus_msghc_dataptr(mh);
1633 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHFREE;
1634 req->chm_chanid = chan->ch_id;
1636 error = vmbus_msghc_exec_noresult(mh);
1637 vmbus_msghc_put(sc, mh);
1640 vmbus_chan_printf(chan,
1641 "chfree(chan%u) msg hypercall exec failed: %d\n",
1642 chan->ch_id, error);
1645 vmbus_chan_printf(chan, "chan%u freed\n", chan->ch_id);
1651 vmbus_prichan_detach_task(void *xchan, int pending __unused)
1653 struct vmbus_channel *chan = xchan;
1655 KASSERT(VMBUS_CHAN_ISPRIMARY(chan),
1656 ("chan%u is not primary channel", chan->ch_id));
1658 /* Delete and detach the device associated with this channel. */
1659 vmbus_delete_child(chan);
1661 /* Release this channel (back to vmbus). */
1662 vmbus_chan_release(chan);
1664 /* Free this channel's resource. */
1665 vmbus_chan_free(chan);
1669 vmbus_subchan_detach_task(void *xchan, int pending __unused)
1671 struct vmbus_channel *chan = xchan;
1672 struct vmbus_channel *pri_chan = chan->ch_prichan;
1674 KASSERT(!VMBUS_CHAN_ISPRIMARY(chan),
1675 ("chan%u is primary channel", chan->ch_id));
1677 /* Release this channel (back to vmbus). */
1678 vmbus_chan_release(chan);
1680 /* Unlink from its primary channel's sub-channel list. */
1681 mtx_lock(&pri_chan->ch_subchan_lock);
1682 vmbus_chan_rem_sublist(pri_chan, chan);
1683 mtx_unlock(&pri_chan->ch_subchan_lock);
1684 /* Notify anyone that is waiting for this sub-channel to vanish. */
1687 /* Free this channel's resource. */
1688 vmbus_chan_free(chan);
1692 vmbus_prichan_attach_task(void *xchan, int pending __unused)
1696 * Add device for this primary channel.
1698 vmbus_add_child(xchan);
1702 vmbus_subchan_attach_task(void *xchan __unused, int pending __unused)
1709 vmbus_chan_destroy_all(struct vmbus_softc *sc)
1713 * Detach all devices and destroy the corresponding primary
1717 struct vmbus_channel *chan;
1719 mtx_lock(&sc->vmbus_chan_lock);
1720 TAILQ_FOREACH(chan, &sc->vmbus_chans, ch_link) {
1721 if (VMBUS_CHAN_ISPRIMARY(chan))
1725 /* No more primary channels; done. */
1726 mtx_unlock(&sc->vmbus_chan_lock);
1729 vmbus_chan_rem_list(sc, chan);
1730 mtx_unlock(&sc->vmbus_chan_lock);
1732 mtx_lock(&sc->vmbus_prichan_lock);
1733 vmbus_chan_rem_prilist(sc, chan);
1734 mtx_unlock(&sc->vmbus_prichan_lock);
1736 taskqueue_enqueue(chan->ch_mgmt_tq, &chan->ch_detach_task);
1740 struct vmbus_channel **
1741 vmbus_subchan_get(struct vmbus_channel *pri_chan, int subchan_cnt)
1743 struct vmbus_channel **ret, *chan;
1746 KASSERT(subchan_cnt > 0, ("invalid sub-channel count %d", subchan_cnt));
1748 ret = malloc(subchan_cnt * sizeof(struct vmbus_channel *), M_TEMP,
1751 mtx_lock(&pri_chan->ch_subchan_lock);
1753 while (pri_chan->ch_subchan_cnt < subchan_cnt)
1754 mtx_sleep(pri_chan, &pri_chan->ch_subchan_lock, 0, "subch", 0);
1757 TAILQ_FOREACH(chan, &pri_chan->ch_subchans, ch_sublink) {
1758 /* TODO: refcnt chan */
1762 if (i == subchan_cnt)
1765 KASSERT(i == subchan_cnt, ("invalid subchan count %d, should be %d",
1766 pri_chan->ch_subchan_cnt, subchan_cnt));
1768 mtx_unlock(&pri_chan->ch_subchan_lock);
1774 vmbus_subchan_rel(struct vmbus_channel **subchan, int subchan_cnt __unused)
1777 free(subchan, M_TEMP);
1781 vmbus_subchan_drain(struct vmbus_channel *pri_chan)
1783 mtx_lock(&pri_chan->ch_subchan_lock);
1784 while (pri_chan->ch_subchan_cnt > 0)
1785 mtx_sleep(pri_chan, &pri_chan->ch_subchan_lock, 0, "dsubch", 0);
1786 mtx_unlock(&pri_chan->ch_subchan_lock);
1790 vmbus_chan_msgproc(struct vmbus_softc *sc, const struct vmbus_message *msg)
1792 vmbus_chanmsg_proc_t msg_proc;
1795 msg_type = ((const struct vmbus_chanmsg_hdr *)msg->msg_data)->chm_type;
1796 KASSERT(msg_type < VMBUS_CHANMSG_TYPE_MAX,
1797 ("invalid message type %u", msg_type));
1799 msg_proc = vmbus_chan_msgprocs[msg_type];
1800 if (msg_proc != NULL)
1805 vmbus_chan_set_readbatch(struct vmbus_channel *chan, bool on)
1808 chan->ch_flags &= ~VMBUS_CHAN_FLAG_BATCHREAD;
1810 chan->ch_flags |= VMBUS_CHAN_FLAG_BATCHREAD;
1814 vmbus_chan_id(const struct vmbus_channel *chan)
1820 vmbus_chan_subidx(const struct vmbus_channel *chan)
1822 return chan->ch_subidx;
1826 vmbus_chan_is_primary(const struct vmbus_channel *chan)
1828 if (VMBUS_CHAN_ISPRIMARY(chan))
1834 const struct hyperv_guid *
1835 vmbus_chan_guid_inst(const struct vmbus_channel *chan)
1837 return &chan->ch_guid_inst;
1841 vmbus_chan_prplist_nelem(int br_size, int prpcnt_max, int dlen_max)
1845 elem_size = __offsetof(struct vmbus_chanpkt_prplist,
1846 cp_range[0].gpa_page[prpcnt_max]);
1847 elem_size += dlen_max;
1848 elem_size = VMBUS_CHANPKT_TOTLEN(elem_size);
1850 return (vmbus_br_nelem(br_size, elem_size));
1854 vmbus_chan_tx_empty(const struct vmbus_channel *chan)
1857 return (vmbus_txbr_empty(&chan->ch_txbr));
1861 vmbus_chan_rx_empty(const struct vmbus_channel *chan)
1864 return (vmbus_rxbr_empty(&chan->ch_rxbr));
1868 vmbus_chan_printf(const struct vmbus_channel *chan, const char *fmt, ...)
1874 if (chan->ch_dev == NULL || !device_is_alive(chan->ch_dev))
1875 dev = chan->ch_vmbus->vmbus_dev;
1879 retval = device_print_prettyname(dev);
1881 retval += vprintf(fmt, ap);
1888 vmbus_chan_run_task(struct vmbus_channel *chan, struct task *task)
1891 taskqueue_enqueue(chan->ch_tq, task);
1892 taskqueue_drain(chan->ch_tq, task);
1896 vmbus_chan_mgmt_tq(const struct vmbus_channel *chan)
1899 return (chan->ch_mgmt_tq);
1903 vmbus_chan_is_revoked(const struct vmbus_channel *chan)
1906 if (chan->ch_stflags & VMBUS_CHAN_ST_REVOKED)
1912 vmbus_chan_set_orphan(struct vmbus_channel *chan, struct vmbus_xact_ctx *xact)
1915 sx_xlock(&chan->ch_orphan_lock);
1916 chan->ch_orphan_xact = xact;
1917 sx_xunlock(&chan->ch_orphan_lock);
1921 vmbus_chan_unset_orphan(struct vmbus_channel *chan)
1924 sx_xlock(&chan->ch_orphan_lock);
1925 chan->ch_orphan_xact = NULL;
1926 sx_xunlock(&chan->ch_orphan_lock);
1930 vmbus_chan_xact_wait(const struct vmbus_channel *chan,
1931 struct vmbus_xact *xact, size_t *resp_len, bool can_sleep)
1936 ret = vmbus_xact_wait(xact, resp_len);
1938 ret = vmbus_xact_busywait(xact, resp_len);
1939 if (vmbus_chan_is_revoked(chan)) {
1941 * This xact probably is interrupted, and the
1942 * interruption can race the reply reception,
1943 * so we have to make sure that there are nothing
1944 * left on the RX bufring, i.e. this xact will
1945 * not be touched, once this function returns.
1947 * Since the hypervisor will not put more data
1948 * onto the RX bufring once the channel is revoked,
1949 * the following loop will be terminated, once all
1950 * data are drained by the driver's channel
1953 while (!vmbus_chan_rx_empty(chan)) {