2 * Copyright (c) 2008, 2013 Citrix Systems, Inc.
3 * Copyright (c) 2012 Spectra Logic Corporation
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include <sys/param.h>
33 #include <sys/kernel.h>
34 #include <sys/malloc.h>
37 #include <sys/systm.h>
42 #include <dev/pci/pcivar.h>
44 #include <machine/cpufunc.h>
45 #include <machine/cpu.h>
46 #include <machine/smp.h>
48 #include <x86/apicreg.h>
50 #include <xen/xen-os.h>
51 #include <xen/features.h>
52 #include <xen/gnttab.h>
53 #include <xen/hypervisor.h>
55 #include <xen/xen_intr.h>
57 #include <xen/interface/hvm/params.h>
58 #include <xen/interface/vcpu.h>
60 /*--------------------------- Forward Declarations ---------------------------*/
62 static driver_filter_t xen_smp_rendezvous_action;
63 static driver_filter_t xen_invltlb;
64 static driver_filter_t xen_invlpg;
65 static driver_filter_t xen_invlrng;
66 static driver_filter_t xen_invlcache;
68 static driver_filter_t xen_lazypmap;
70 static driver_filter_t xen_ipi_bitmap_handler;
71 static driver_filter_t xen_cpustop_handler;
72 static driver_filter_t xen_cpususpend_handler;
73 static driver_filter_t xen_cpustophard_handler;
74 static void xen_ipi_vectored(u_int vector, int dest);
75 static void xen_hvm_cpu_resume(void);
77 static void xen_hvm_cpu_init(void);
79 /*---------------------------- Extern Declarations ---------------------------*/
81 extern void pmap_lazyfix_action(void);
84 extern int pmap_pcid_enabled;
87 /* Variables used by mp_machdep to perform the bitmap IPI */
88 extern volatile u_int cpu_ipi_pending[MAXCPU];
90 /*---------------------------------- Macros ----------------------------------*/
91 #define IPI_TO_IDX(ipi) ((ipi) - APIC_IPI_INTS)
93 /*-------------------------------- Local Types -------------------------------*/
94 enum xen_hvm_init_type {
96 XEN_HVM_INIT_CANCELLED_SUSPEND,
100 struct xen_ipi_handler
102 driver_filter_t *filter;
103 const char *description;
106 /*-------------------------------- Global Data -------------------------------*/
107 enum xen_domain_type xen_domain_type = XEN_NATIVE;
110 struct cpu_ops xen_hvm_cpu_ops = {
111 .ipi_vectored = lapic_ipi_vectored,
112 .cpu_init = xen_hvm_cpu_init,
113 .cpu_resume = xen_hvm_cpu_resume
117 static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support");
120 static struct xen_ipi_handler xen_ipis[] =
122 [IPI_TO_IDX(IPI_RENDEZVOUS)] = { xen_smp_rendezvous_action, "r" },
123 [IPI_TO_IDX(IPI_INVLTLB)] = { xen_invltlb, "itlb"},
124 [IPI_TO_IDX(IPI_INVLPG)] = { xen_invlpg, "ipg" },
125 [IPI_TO_IDX(IPI_INVLRNG)] = { xen_invlrng, "irg" },
126 [IPI_TO_IDX(IPI_INVLCACHE)] = { xen_invlcache, "ic" },
128 [IPI_TO_IDX(IPI_LAZYPMAP)] = { xen_lazypmap, "lp" },
130 [IPI_TO_IDX(IPI_BITMAP_VECTOR)] = { xen_ipi_bitmap_handler, "b" },
131 [IPI_TO_IDX(IPI_STOP)] = { xen_cpustop_handler, "st" },
132 [IPI_TO_IDX(IPI_SUSPEND)] = { xen_cpususpend_handler, "sp" },
133 [IPI_TO_IDX(IPI_STOP_HARD)] = { xen_cpustophard_handler, "sth" },
138 * If non-zero, the hypervisor has been configured to use a direct
139 * IDT event callback for interrupt injection.
141 int xen_vector_callback_enabled;
143 /*------------------------------- Per-CPU Data -------------------------------*/
144 DPCPU_DEFINE(struct vcpu_info, vcpu_local_info);
145 DPCPU_DEFINE(struct vcpu_info *, vcpu_info);
147 DPCPU_DEFINE(xen_intr_handle_t, ipi_handle[nitems(xen_ipis)]);
150 /*------------------ Hypervisor Access Shared Memory Regions -----------------*/
151 /** Hypercall table accessed via HYPERVISOR_*_op() methods. */
152 extern char *hypercall_page;
153 shared_info_t *HYPERVISOR_shared_info;
154 start_info_t *HYPERVISOR_start_info;
157 /*---------------------------- XEN PV IPI Handlers ---------------------------*/
159 * This are C clones of the ASM functions found in apic_vector.s
162 xen_ipi_bitmap_handler(void *arg)
164 struct trapframe *frame;
167 ipi_bitmap_handler(*frame);
168 return (FILTER_HANDLED);
172 xen_smp_rendezvous_action(void *arg)
175 (*ipi_rendezvous_counts[PCPU_GET(cpuid)])++;
176 #endif /* COUNT_IPIS */
178 smp_rendezvous_action();
179 return (FILTER_HANDLED);
183 xen_invltlb(void *arg)
187 return (FILTER_HANDLED);
192 xen_invltlb_pcid(void *arg)
195 invltlb_pcid_handler();
196 return (FILTER_HANDLED);
201 xen_invlpg(void *arg)
205 return (FILTER_HANDLED);
210 xen_invlpg_pcid(void *arg)
213 invlpg_pcid_handler();
214 return (FILTER_HANDLED);
219 xen_invlrng(void *arg)
223 return (FILTER_HANDLED);
227 xen_invlcache(void *arg)
231 return (FILTER_HANDLED);
236 xen_lazypmap(void *arg)
239 pmap_lazyfix_action();
240 return (FILTER_HANDLED);
245 xen_cpustop_handler(void *arg)
249 return (FILTER_HANDLED);
253 xen_cpususpend_handler(void *arg)
256 cpususpend_handler();
257 return (FILTER_HANDLED);
261 xen_cpustophard_handler(void *arg)
265 return (FILTER_HANDLED);
268 /* Xen PV IPI sender */
270 xen_ipi_vectored(u_int vector, int dest)
272 xen_intr_handle_t *ipi_handle;
273 int ipi_idx, to_cpu, self;
275 ipi_idx = IPI_TO_IDX(vector);
276 if (ipi_idx > nitems(xen_ipis))
277 panic("IPI out of range");
280 case APIC_IPI_DEST_SELF:
281 ipi_handle = DPCPU_GET(ipi_handle);
282 xen_intr_signal(ipi_handle[ipi_idx]);
284 case APIC_IPI_DEST_ALL:
285 CPU_FOREACH(to_cpu) {
286 ipi_handle = DPCPU_ID_GET(to_cpu, ipi_handle);
287 xen_intr_signal(ipi_handle[ipi_idx]);
290 case APIC_IPI_DEST_OTHERS:
291 self = PCPU_GET(cpuid);
292 CPU_FOREACH(to_cpu) {
293 if (to_cpu != self) {
294 ipi_handle = DPCPU_ID_GET(to_cpu, ipi_handle);
295 xen_intr_signal(ipi_handle[ipi_idx]);
300 to_cpu = apic_cpuid(dest);
301 ipi_handle = DPCPU_ID_GET(to_cpu, ipi_handle);
302 xen_intr_signal(ipi_handle[ipi_idx]);
307 /* XEN diverged cpu operations */
309 xen_hvm_cpu_resume(void)
311 u_int cpuid = PCPU_GET(cpuid);
314 * Reset pending bitmap IPIs, because Xen doesn't preserve pending
315 * event channels on migration.
317 cpu_ipi_pending[cpuid] = 0;
319 /* register vcpu_info area */
324 xen_cpu_ipi_init(int cpu)
326 xen_intr_handle_t *ipi_handle;
327 const struct xen_ipi_handler *ipi;
331 ipi_handle = DPCPU_ID_GET(cpu, ipi_handle);
332 dev = pcpu_find(cpu)->pc_device;
333 KASSERT((dev != NULL), ("NULL pcpu device_t"));
335 for (ipi = xen_ipis, idx = 0; idx < nitems(xen_ipis); ipi++, idx++) {
337 if (ipi->filter == NULL) {
338 ipi_handle[idx] = NULL;
342 rc = xen_intr_alloc_and_bind_ipi(dev, cpu, ipi->filter,
343 INTR_TYPE_TTY, &ipi_handle[idx]);
345 panic("Unable to allocate a XEN IPI port");
346 xen_intr_describe(ipi_handle[idx], "%s", ipi->description);
355 if (!xen_vector_callback_enabled)
359 if (pmap_pcid_enabled) {
360 xen_ipis[IPI_TO_IDX(IPI_INVLTLB)].filter = xen_invltlb_pcid;
361 xen_ipis[IPI_TO_IDX(IPI_INVLPG)].filter = xen_invlpg_pcid;
367 /* Set the xen pv ipi ops to replace the native ones */
368 cpu_ops.ipi_vectored = xen_ipi_vectored;
372 /*---------------------- XEN Hypervisor Probe and Setup ----------------------*/
374 xen_hvm_cpuid_base(void)
376 uint32_t base, regs[4];
378 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
379 do_cpuid(base, regs);
380 if (!memcmp("XenVMMXenVMM", ®s[1], 12)
381 && (regs[0] - base) >= 2)
388 * Allocate and fill in the hypcall page.
391 xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type)
393 uint32_t base, regs[4];
396 if (xen_pv_domain()) {
397 /* hypercall page is already set in the PV case */
401 base = xen_hvm_cpuid_base();
405 if (init_type == XEN_HVM_INIT_COLD) {
406 do_cpuid(base + 1, regs);
407 printf("XEN: Hypervisor version %d.%d detected.\n",
408 regs[0] >> 16, regs[0] & 0xffff);
412 * Find the hypercall pages.
414 do_cpuid(base + 2, regs);
416 for (i = 0; i < regs[0]; i++)
417 wrmsr(regs[1], vtophys(&hypercall_page + i * PAGE_SIZE) + i);
423 xen_hvm_init_shared_info_page(void)
425 struct xen_add_to_physmap xatp;
427 if (xen_pv_domain()) {
429 * Already setup in the PV case, shared_info is passed inside
430 * of the start_info struct at start of day.
435 if (HYPERVISOR_shared_info == NULL) {
436 HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT);
437 if (HYPERVISOR_shared_info == NULL)
438 panic("Unable to allocate Xen shared info page");
441 xatp.domid = DOMID_SELF;
443 xatp.space = XENMAPSPACE_shared_info;
444 xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT;
445 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
446 panic("HYPERVISOR_memory_op failed");
450 * Tell the hypervisor how to contact us for event channel callbacks.
453 xen_hvm_set_callback(device_t dev)
455 struct xen_hvm_param xhp;
458 if (xen_vector_callback_enabled)
461 xhp.domid = DOMID_SELF;
462 xhp.index = HVM_PARAM_CALLBACK_IRQ;
463 if (xen_feature(XENFEAT_hvm_callback_vector) != 0) {
466 xhp.value = HVM_CALLBACK_VECTOR(IDT_EVTCHN);
467 error = HYPERVISOR_hvm_op(HVMOP_set_param, &xhp);
469 xen_vector_callback_enabled = 1;
472 printf("Xen HVM callback vector registration failed (%d). "
473 "Falling back to emulated device interrupt\n", error);
475 xen_vector_callback_enabled = 0;
478 * Called from early boot or resume.
479 * xenpci will invoke us again later.
484 irq = pci_get_irq(dev);
486 xhp.value = HVM_CALLBACK_GSI(irq);
491 slot = pci_get_slot(dev);
492 pin = pci_get_intpin(dev) - 1;
493 xhp.value = HVM_CALLBACK_PCI_INTX(slot, pin);
496 if (HYPERVISOR_hvm_op(HVMOP_set_param, &xhp) != 0)
497 panic("Can't set evtchn callback");
500 #define XEN_MAGIC_IOPORT 0x10
503 XMI_UNPLUG_IDE_DISKS = 0x01,
504 XMI_UNPLUG_NICS = 0x02,
505 XMI_UNPLUG_IDE_EXCEPT_PRI_MASTER = 0x04
509 xen_hvm_disable_emulated_devices(void)
512 if (xen_pv_domain()) {
514 * No emulated devices in the PV case, so no need to unplug
520 if (inw(XEN_MAGIC_IOPORT) != XMI_MAGIC)
524 printf("XEN: Disabling emulated block and network devices\n");
525 outw(XEN_MAGIC_IOPORT, XMI_UNPLUG_IDE_DISKS|XMI_UNPLUG_NICS);
529 xen_hvm_init(enum xen_hvm_init_type init_type)
534 if (init_type == XEN_HVM_INIT_CANCELLED_SUSPEND)
537 error = xen_hvm_init_hypercall_stubs(init_type);
540 case XEN_HVM_INIT_COLD:
545 * If xen_domain_type is not set at this point
546 * it means we are inside a (PV)HVM guest, because
547 * for PVH the guest type is set much earlier
548 * (see hammer_time_xen).
551 xen_domain_type = XEN_HVM_DOMAIN;
552 vm_guest = VM_GUEST_XEN;
555 setup_xen_features();
557 cpu_ops = xen_hvm_cpu_ops;
560 case XEN_HVM_INIT_RESUME:
562 panic("Unable to init Xen hypercall stubs on resume");
564 /* Clear stale vcpu_info. */
566 DPCPU_ID_SET(i, vcpu_info, NULL);
569 panic("Unsupported HVM initialization type");
572 xen_vector_callback_enabled = 0;
573 xen_hvm_set_callback(NULL);
576 * On (PV)HVM domains we need to request the hypervisor to
577 * fill the shared info page, for PVH guest the shared_info page
578 * is passed inside the start_info struct and is already set, so this
579 * functions are no-ops.
581 xen_hvm_init_shared_info_page();
582 xen_hvm_disable_emulated_devices();
586 xen_hvm_suspend(void)
591 xen_hvm_resume(bool suspend_cancelled)
594 xen_hvm_init(suspend_cancelled ?
595 XEN_HVM_INIT_CANCELLED_SUSPEND : XEN_HVM_INIT_RESUME);
597 /* Register vcpu_info area for CPU#0. */
602 xen_hvm_sysinit(void *arg __unused)
604 xen_hvm_init(XEN_HVM_INIT_COLD);
608 xen_set_vcpu_id(void)
613 if (!xen_hvm_domain())
616 /* Set vcpu_id to acpi_id */
619 pc->pc_vcpu_id = pc->pc_acpi_id;
621 printf("XEN: CPU %u has VCPU ID %u\n",
627 xen_hvm_cpu_init(void)
629 struct vcpu_register_vcpu_info info;
630 struct vcpu_info *vcpu_info;
636 if (DPCPU_GET(vcpu_info) != NULL) {
638 * vcpu_info is already set. We're resuming
639 * from a failed migration and our pre-suspend
640 * configuration is still valid.
645 vcpu_info = DPCPU_PTR(vcpu_local_info);
646 cpu = PCPU_GET(vcpu_id);
647 info.mfn = vtophys(vcpu_info) >> PAGE_SHIFT;
648 info.offset = vtophys(vcpu_info) - trunc_page(vtophys(vcpu_info));
650 rc = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
652 DPCPU_SET(vcpu_info, &HYPERVISOR_shared_info->vcpu_info[cpu]);
654 DPCPU_SET(vcpu_info, vcpu_info);
657 SYSINIT(xen_hvm_init, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, xen_hvm_sysinit, NULL);
659 /* We need to setup IPIs before APs are started */
660 SYSINIT(xen_setup_cpus, SI_SUB_SMP-1, SI_ORDER_FIRST, xen_setup_cpus, NULL);
662 SYSINIT(xen_hvm_cpu_init, SI_SUB_INTR, SI_ORDER_FIRST, xen_hvm_cpu_init, NULL);
663 SYSINIT(xen_set_vcpu_id, SI_SUB_CPU, SI_ORDER_ANY, xen_set_vcpu_id, NULL);