2 * Copyright (c) 2009-2012 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.
30 * VM Bus Driver Implementation
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
37 #include <sys/kernel.h>
39 #include <sys/malloc.h>
40 #include <sys/module.h>
41 #include <sys/sysctl.h>
42 #include <sys/syslog.h>
43 #include <sys/systm.h>
44 #include <sys/rtprio.h>
45 #include <sys/interrupt.h>
47 #include <sys/taskqueue.h>
48 #include <sys/mutex.h>
51 #include <machine/resource.h>
54 #include <machine/stdarg.h>
55 #include <machine/intr_machdep.h>
56 #include <machine/md_var.h>
57 #include <machine/segments.h>
59 #include <x86/apicvar.h>
61 #include "hv_vmbus_priv.h"
66 static device_t vmbus_devp;
67 static int vmbus_inited;
68 static hv_setup_args setup_args; /* only CPU 0 supported at this time */
71 * @brief Software interrupt thread routine to handle channel messages from
75 vmbus_msg_swintr(void *arg)
79 hv_vmbus_channel_msg_header *hdr;
80 hv_vmbus_channel_msg_table_entry *entry;
81 hv_vmbus_channel_msg_type msg_type;
82 hv_vmbus_message* msg;
83 hv_vmbus_message* copied;
84 static bool warned = false;
87 KASSERT(cpu <= mp_maxid, ("VMBUS: vmbus_msg_swintr: "
88 "cpu out of range!"));
90 page_addr = hv_vmbus_g_context.syn_ic_msg_page[cpu];
91 msg = (hv_vmbus_message*) page_addr + HV_VMBUS_MESSAGE_SINT;
94 if (msg->header.message_type == HV_MESSAGE_TYPE_NONE)
95 break; /* no message */
97 hdr = (hv_vmbus_channel_msg_header *)msg->u.payload;
98 msg_type = hdr->message_type;
100 if (msg_type >= HV_CHANNEL_MESSAGE_COUNT && !warned) {
102 printf("VMBUS: unknown message type = %d\n", msg_type);
106 entry = &g_channel_message_table[msg_type];
108 if (entry->handler_no_sleep)
109 entry->messageHandler(hdr);
112 copied = malloc(sizeof(hv_vmbus_message),
114 KASSERT(copied != NULL,
115 ("Error VMBUS: malloc failed to allocate"
116 " hv_vmbus_message!"));
120 memcpy(copied, msg, sizeof(hv_vmbus_message));
121 hv_queue_work_item(hv_vmbus_g_connection.work_queue,
122 hv_vmbus_on_channel_message,
126 msg->header.message_type = HV_MESSAGE_TYPE_NONE;
129 * Make sure the write to message_type (ie set to
130 * HV_MESSAGE_TYPE_NONE) happens before we read the
131 * message_pending and EOMing. Otherwise, the EOMing will
132 * not deliver any more messages
133 * since there is no empty slot
137 if (msg->header.message_flags.u.message_pending) {
139 * This will cause message queue rescan to possibly
140 * deliver another msg from the hypervisor
142 wrmsr(HV_X64_MSR_EOM, 0);
148 * @brief Interrupt filter routine for VMBUS.
150 * The purpose of this routine is to determine the type of VMBUS protocol
151 * message to process - an event or a channel message.
154 hv_vmbus_isr(void *unused)
157 hv_vmbus_message* msg;
158 hv_vmbus_synic_event_flags* event;
161 cpu = PCPU_GET(cpuid);
164 * The Windows team has advised that we check for events
165 * before checking for messages. This is the way they do it
166 * in Windows when running as a guest in Hyper-V
169 page_addr = hv_vmbus_g_context.syn_ic_event_page[cpu];
170 event = (hv_vmbus_synic_event_flags*)
171 page_addr + HV_VMBUS_MESSAGE_SINT;
173 if ((hv_vmbus_protocal_version == HV_VMBUS_VERSION_WS2008) ||
174 (hv_vmbus_protocal_version == HV_VMBUS_VERSION_WIN7)) {
175 /* Since we are a child, we only need to check bit 0 */
176 if (synch_test_and_clear_bit(0, &event->flags32[0])) {
177 swi_sched(hv_vmbus_g_context.event_swintr[cpu], 0);
181 * On host with Win8 or above, we can directly look at
182 * the event page. If bit n is set, we have an interrupt
183 * on the channel with id n.
184 * Directly schedule the event software interrupt on
187 swi_sched(hv_vmbus_g_context.event_swintr[cpu], 0);
190 /* Check if there are actual msgs to be process */
191 page_addr = hv_vmbus_g_context.syn_ic_msg_page[cpu];
192 msg = (hv_vmbus_message*) page_addr + HV_VMBUS_MESSAGE_SINT;
194 if (msg->header.message_type != HV_MESSAGE_TYPE_NONE) {
195 swi_sched(hv_vmbus_g_context.msg_swintr[cpu], 0);
198 return FILTER_HANDLED;
202 uint32_t hv_intr_count = 0;
204 uint32_t hv_vmbus_swintr_event_cpu[MAXCPU];
205 uint32_t hv_vmbus_intr_cpu[MAXCPU];
208 hv_vector_handler(struct trapframe *trap_frame)
215 * Disable preemption.
221 * Do a little interrupt counting.
223 cpu = PCPU_GET(cpuid);
224 hv_vmbus_intr_cpu[cpu]++;
243 struct hv_device *child_dev_ctx = device_get_ivars(child);
247 case HV_VMBUS_IVAR_TYPE:
248 *result = (uintptr_t) &child_dev_ctx->class_id;
250 case HV_VMBUS_IVAR_INSTANCE:
251 *result = (uintptr_t) &child_dev_ctx->device_id;
253 case HV_VMBUS_IVAR_DEVCTX:
254 *result = (uintptr_t) child_dev_ctx;
256 case HV_VMBUS_IVAR_NODE:
257 *result = (uintptr_t) child_dev_ctx->device;
272 case HV_VMBUS_IVAR_TYPE:
273 case HV_VMBUS_IVAR_INSTANCE:
274 case HV_VMBUS_IVAR_DEVCTX:
275 case HV_VMBUS_IVAR_NODE:
283 hv_vmbus_child_device_create(
286 hv_vmbus_channel* channel)
288 hv_device* child_dev;
291 * Allocate the new child device
293 child_dev = malloc(sizeof(hv_device), M_DEVBUF,
295 KASSERT(child_dev != NULL,
296 ("Error VMBUS: malloc failed to allocate hv_device!"));
298 if (child_dev == NULL)
301 child_dev->channel = channel;
302 memcpy(&child_dev->class_id, &type, sizeof(hv_guid));
303 memcpy(&child_dev->device_id, &instance, sizeof(hv_guid));
309 print_dev_guid(struct hv_device *dev)
312 unsigned char guid_name[100];
313 for (i = 0; i < 32; i += 2)
314 sprintf(&guid_name[i], "%02x", dev->class_id.data[i / 2]);
316 printf("VMBUS: Class ID: %s\n", guid_name);
320 hv_vmbus_child_device_register(struct hv_device *child_dev)
325 print_dev_guid(child_dev);
328 child = device_add_child(vmbus_devp, NULL, -1);
329 child_dev->device = child;
330 device_set_ivars(child, child_dev);
333 ret = device_probe_and_attach(child);
340 hv_vmbus_child_device_unregister(struct hv_device *child_dev)
344 * XXXKYS: Ensure that this is the opposite of
348 ret = device_delete_child(vmbus_devp, child_dev->device);
354 vmbus_identify(driver_t *driver, device_t parent)
356 if (!hv_vmbus_query_hypervisor_presence())
359 vm_guest = VM_GUEST_HV;
361 BUS_ADD_CHILD(parent, 0, "vmbus", 0);
365 vmbus_probe(device_t dev) {
367 device_printf(dev, "VMBUS: probe\n");
369 device_set_desc(dev, "Vmbus Devices");
371 return (BUS_PROBE_NOWILDCARD);
375 extern inthand_t IDTVEC(rsvd), IDTVEC(hv_vmbus_callback);
378 * @brief Find a free IDT slot and setup the interrupt handler.
381 vmbus_vector_alloc(void)
385 struct gate_descriptor *ip;
388 * Search backwards form the highest IDT vector available for use
389 * as vmbus channel callback vector. We install 'hv_vmbus_callback'
390 * handler at that vector and use it to interrupt vcpus.
392 vector = APIC_SPURIOUS_INT;
393 while (--vector >= APIC_IPI_INTS) {
395 func = ((long)ip->gd_hioffset << 16 | ip->gd_looffset);
396 if (func == (uintptr_t)&IDTVEC(rsvd)) {
398 setidt(vector , IDTVEC(hv_vmbus_callback), SDT_SYS386IGT,
399 SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
401 setidt(vector , IDTVEC(hv_vmbus_callback), SDT_SYSIGT,
412 * @brief Restore the IDT slot to rsvd.
415 vmbus_vector_free(int vector)
418 struct gate_descriptor *ip;
423 KASSERT(vector >= APIC_IPI_INTS && vector < APIC_SPURIOUS_INT,
424 ("invalid vector %d", vector));
427 func = ((long)ip->gd_hioffset << 16 | ip->gd_looffset);
428 KASSERT(func == (uintptr_t)&IDTVEC(hv_vmbus_callback),
429 ("invalid vector %d", vector));
431 setidt(vector, IDTVEC(rsvd), SDT_SYSIGT, SEL_KPL, 0);
437 vmbus_vector_alloc(void)
443 vmbus_vector_free(int vector)
450 * @brief Main vmbus driver initialization routine.
453 * - initialize the vmbus driver context
454 * - setup various driver entry points
455 * - invoke the vmbus hv main init routine
456 * - get the irq resource
457 * - invoke the vmbus to add the vmbus root device
458 * - setup the vmbus root device
459 * - retrieve the channel offers
471 ret = hv_vmbus_init();
475 printf("Error VMBUS: Hypervisor Initialization Failed!\n");
480 * Find a free IDT slot for vmbus callback.
482 hv_vmbus_g_context.hv_cb_vector = vmbus_vector_alloc();
484 if (hv_vmbus_g_context.hv_cb_vector == 0) {
486 printf("Error VMBUS: Cannot find free IDT slot for "
487 "vmbus callback!\n");
492 printf("VMBUS: vmbus callback vector %d\n",
493 hv_vmbus_g_context.hv_cb_vector);
496 * Notify the hypervisor of our vector.
498 setup_args.vector = hv_vmbus_g_context.hv_cb_vector;
501 hv_vmbus_intr_cpu[j] = 0;
502 hv_vmbus_swintr_event_cpu[j] = 0;
503 hv_vmbus_g_context.hv_event_intr_event[j] = NULL;
504 hv_vmbus_g_context.hv_msg_intr_event[j] = NULL;
505 hv_vmbus_g_context.event_swintr[j] = NULL;
506 hv_vmbus_g_context.msg_swintr[j] = NULL;
508 for (i = 0; i < 2; i++)
509 setup_args.page_buffers[2 * j + i] = NULL;
517 * Setup software interrupt thread and handler for msg handling.
519 ret = swi_add(&hv_vmbus_g_context.hv_msg_intr_event[j],
520 "hv_msg", vmbus_msg_swintr, (void *)(long)j, SWI_CLOCK, 0,
521 &hv_vmbus_g_context.msg_swintr[j]);
524 printf("VMBUS: failed to setup msg swi for "
530 * Bind the swi thread to the cpu.
532 ret = intr_event_bind(hv_vmbus_g_context.hv_msg_intr_event[j],
536 printf("VMBUS: failed to bind msg swi thread "
542 * Setup software interrupt thread and handler for
545 ret = swi_add(&hv_vmbus_g_context.hv_event_intr_event[j],
546 "hv_event", hv_vmbus_on_events, (void *)(long)j,
547 SWI_CLOCK, 0, &hv_vmbus_g_context.event_swintr[j]);
550 printf("VMBUS: failed to setup event swi for "
556 * Prepare the per cpu msg and event pages to be called on each cpu.
558 for(i = 0; i < 2; i++) {
559 setup_args.page_buffers[2 * j + i] =
560 malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT | M_ZERO);
561 if (setup_args.page_buffers[2 * j + i] == NULL) {
562 KASSERT(setup_args.page_buffers[2 * j + i] != NULL,
563 ("Error VMBUS: malloc failed!"));
570 printf("VMBUS: Calling smp_rendezvous, smp_started = %d\n",
573 smp_rendezvous(NULL, hv_vmbus_synic_init, NULL, &setup_args);
576 * Connect to VMBus in the root partition
578 ret = hv_vmbus_connect();
583 hv_vmbus_request_channel_offers();
588 * Free pages alloc'ed
590 for (n = 0; n < 2 * MAXCPU; n++)
591 if (setup_args.page_buffers[n] != NULL)
592 free(setup_args.page_buffers[n], M_DEVBUF);
595 * remove swi and vmbus callback vector;
598 if (hv_vmbus_g_context.msg_swintr[j] != NULL)
599 swi_remove(hv_vmbus_g_context.msg_swintr[j]);
600 if (hv_vmbus_g_context.event_swintr[j] != NULL)
601 swi_remove(hv_vmbus_g_context.event_swintr[j]);
602 hv_vmbus_g_context.hv_msg_intr_event[j] = NULL;
603 hv_vmbus_g_context.hv_event_intr_event[j] = NULL;
606 vmbus_vector_free(hv_vmbus_g_context.hv_cb_vector);
615 vmbus_attach(device_t dev)
618 device_printf(dev, "VMBUS: attach dev: %p\n", dev);
622 * If the system has already booted and thread
623 * scheduling is possible indicated by the global
624 * cold set to zero, we just call the driver
625 * initialization directly.
636 if (vm_guest != VM_GUEST_HV)
640 * If the system has already booted and thread
641 * scheduling is possible, as indicated by the
642 * global cold set to zero, we just call the driver
643 * initialization directly.
654 hv_vmbus_release_unattached_channels();
655 hv_vmbus_disconnect();
657 smp_rendezvous(NULL, hv_vmbus_synic_cleanup, NULL, NULL);
659 for(i = 0; i < 2 * MAXCPU; i++) {
660 if (setup_args.page_buffers[i] != 0)
661 free(setup_args.page_buffers[i], M_DEVBUF);
668 if (hv_vmbus_g_context.msg_swintr[i] != NULL)
669 swi_remove(hv_vmbus_g_context.msg_swintr[i]);
670 if (hv_vmbus_g_context.event_swintr[i] != NULL)
671 swi_remove(hv_vmbus_g_context.event_swintr[i]);
672 hv_vmbus_g_context.hv_msg_intr_event[i] = NULL;
673 hv_vmbus_g_context.hv_event_intr_event[i] = NULL;
676 vmbus_vector_free(hv_vmbus_g_context.hv_cb_vector);
688 vmbus_detach(device_t dev)
698 printf("VMBUS: load\n");
702 vmbus_mod_unload(void)
705 printf("VMBUS: unload\n");
709 vmbus_modevent(module_t mod, int what, void *arg)
724 static device_method_t vmbus_methods[] = {
725 /** Device interface */
726 DEVMETHOD(device_identify, vmbus_identify),
727 DEVMETHOD(device_probe, vmbus_probe),
728 DEVMETHOD(device_attach, vmbus_attach),
729 DEVMETHOD(device_detach, vmbus_detach),
730 DEVMETHOD(device_shutdown, bus_generic_shutdown),
731 DEVMETHOD(device_suspend, bus_generic_suspend),
732 DEVMETHOD(device_resume, bus_generic_resume),
735 DEVMETHOD(bus_add_child, bus_generic_add_child),
736 DEVMETHOD(bus_print_child, bus_generic_print_child),
737 DEVMETHOD(bus_read_ivar, vmbus_read_ivar),
738 DEVMETHOD(bus_write_ivar, vmbus_write_ivar),
742 static char driver_name[] = "vmbus";
743 static driver_t vmbus_driver = { driver_name, vmbus_methods,0, };
746 devclass_t vmbus_devclass;
748 DRIVER_MODULE(vmbus, nexus, vmbus_driver, vmbus_devclass, vmbus_modevent, 0);
749 MODULE_VERSION(vmbus,1);
751 /* We want to be started after SMP is initialized */
752 SYSINIT(vmb_init, SI_SUB_SMP + 1, SI_ORDER_FIRST, vmbus_init, NULL);