3 * Copyright (c) 2004 Christian Limpach.
4 * Copyright (c) 2004-2006,2008 Kip Macy
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, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. All advertising materials mentioning features or use of this software
16 * must display the following acknowledgement:
17 * This product includes software developed by Christian Limpach.
18 * 4. The name of the author may not be used to endorse or promote products
19 * derived from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
22 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
23 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
24 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
26 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
30 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include <sys/param.h>
37 #include <sys/systm.h>
41 #include <sys/mount.h>
42 #include <sys/malloc.h>
43 #include <sys/mutex.h>
44 #include <sys/kernel.h>
46 #include <sys/reboot.h>
47 #include <sys/rwlock.h>
48 #include <sys/sysproto.h>
51 #include <xen/xen-os.h>
55 #include <machine/segments.h>
56 #include <machine/pcb.h>
57 #include <machine/stdarg.h>
58 #include <machine/vmparam.h>
59 #include <machine/cpu.h>
60 #include <machine/intr_machdep.h>
61 #include <machine/md_var.h>
62 #include <machine/asmacros.h>
66 #include <xen/hypervisor.h>
67 #include <machine/xen/xenvar.h>
68 #include <machine/xen/xenfunc.h>
69 #include <machine/xen/xenpmap.h>
70 #include <machine/xen/xenfunc.h>
71 #include <xen/interface/memory.h>
72 #include <machine/xen/features.h>
74 #include <machine/privatespace.h>
78 #include <vm/vm_page.h>
81 #define IDTVEC(name) __CONCAT(X,name)
84 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
85 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
86 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
87 IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align),
88 IDTVEC(xmm), IDTVEC(lcall_syscall), IDTVEC(int0x80_syscall);
92 start_info_t *xen_start_info;
93 start_info_t *HYPERVISOR_start_info;
94 shared_info_t *HYPERVISOR_shared_info;
95 xen_pfn_t *xen_machine_phys = machine_to_phys_mapping;
96 xen_pfn_t *xen_phys_machine;
97 xen_pfn_t *xen_pfn_to_mfn_frame_list[16];
98 xen_pfn_t *xen_pfn_to_mfn_frame_list_list;
99 int preemptable, init_first;
100 extern unsigned int avail_space;
101 int xen_vector_callback_enabled = 0;
102 enum xen_domain_type xen_domain_type = XEN_PV_DOMAIN;
111 CTR0(KTR_SPARE2, "ni_cli disabling interrupts");
112 __asm__("pushl %edx;"
125 __asm__("pushl %edx;"
137 force_evtchn_callback(void)
139 (void)HYPERVISOR_xen_version(0, NULL);
143 * Modify the cmd_line by converting ',' to NULLs so that it is in a format
144 * suitable for the static env vars.
147 xen_setbootenv(char *cmd_line)
151 /* Skip leading spaces */
152 for (; *cmd_line == ' '; cmd_line++);
154 xc_printf("xen_setbootenv(): cmd_line='%s'\n", cmd_line);
156 for (cmd_line_next = cmd_line; strsep(&cmd_line_next, ",") != NULL;);
161 xen_boothowto(char *envp)
165 /* get equivalents from the environment */
166 for (i = 0; howto_names[i].ev != NULL; i++)
167 if (getenv(howto_names[i].ev) != NULL)
168 howto |= howto_names[i].mask;
173 #define XPQUEUE_SIZE 128
181 /* per-cpu queues and indices */
183 static struct mmu_log xpq_queue_log[XEN_LEGACY_MAX_VCPUS][XPQUEUE_SIZE];
186 static int xpq_idx[XEN_LEGACY_MAX_VCPUS];
187 static mmu_update_t xpq_queue[XEN_LEGACY_MAX_VCPUS][XPQUEUE_SIZE];
189 #define XPQ_QUEUE_LOG xpq_queue_log[vcpu]
190 #define XPQ_QUEUE xpq_queue[vcpu]
191 #define XPQ_IDX xpq_idx[vcpu]
192 #define SET_VCPU() int vcpu = smp_processor_id()
195 static mmu_update_t xpq_queue[XPQUEUE_SIZE];
197 static struct mmu_log xpq_queue_log[XPQUEUE_SIZE];
199 static int xpq_idx = 0;
201 #define XPQ_QUEUE_LOG xpq_queue_log
202 #define XPQ_QUEUE xpq_queue
203 #define XPQ_IDX xpq_idx
207 #define XPQ_IDX_INC atomic_add_int(&XPQ_IDX, 1);
213 int _xpq_idx = XPQ_IDX;
219 xc_printf("xen_dump_queue(): %u entries\n", _xpq_idx);
220 for (i = 0; i < _xpq_idx; i++) {
221 xc_printf(" val: %llx ptr: %llx\n", XPQ_QUEUE[i].val,
229 _xen_flush_queue(void)
232 int _xpq_idx = XPQ_IDX;
236 if (__predict_true(gdtset))
237 CRITICAL_ASSERT(curthread);
241 /* Make sure index is cleared first to avoid double updates. */
242 error = HYPERVISOR_mmu_update((mmu_update_t *)&XPQ_QUEUE,
243 _xpq_idx, NULL, DOMID_SELF);
246 if (__predict_true(gdtset))
247 for (i = _xpq_idx; i > 0;) {
249 CTR6(KTR_PMAP, "mmu:val: %lx ptr: %lx val: %lx "
250 "ptr: %lx val: %lx ptr: %lx",
251 (XPQ_QUEUE[i-1].val & 0xffffffff),
252 (XPQ_QUEUE[i-1].ptr & 0xffffffff),
253 (XPQ_QUEUE[i-2].val & 0xffffffff),
254 (XPQ_QUEUE[i-2].ptr & 0xffffffff),
255 (XPQ_QUEUE[i-3].val & 0xffffffff),
256 (XPQ_QUEUE[i-3].ptr & 0xffffffff));
259 CTR4(KTR_PMAP, "mmu: val: %lx ptr: %lx val: %lx ptr: %lx",
260 (XPQ_QUEUE[i-1].val & 0xffffffff),
261 (XPQ_QUEUE[i-1].ptr & 0xffffffff),
262 (XPQ_QUEUE[i-2].val & 0xffffffff),
263 (XPQ_QUEUE[i-2].ptr & 0xffffffff));
266 CTR2(KTR_PMAP, "mmu: val: %lx ptr: %lx",
267 (XPQ_QUEUE[i-1].val & 0xffffffff),
268 (XPQ_QUEUE[i-1].ptr & 0xffffffff));
273 if (__predict_false(error < 0)) {
274 for (i = 0; i < _xpq_idx; i++)
275 printf("val: %llx ptr: %llx\n",
276 XPQ_QUEUE[i].val, XPQ_QUEUE[i].ptr);
277 panic("Failed to execute MMU updates: %d", error);
283 xen_flush_queue(void)
287 if (__predict_true(gdtset))
289 if (XPQ_IDX != 0) _xen_flush_queue();
290 if (__predict_true(gdtset))
295 xen_increment_idx(void)
300 if (__predict_false(XPQ_IDX == XPQUEUE_SIZE))
305 xen_check_queue(void)
310 KASSERT(XPQ_IDX == 0, ("pending operations XPQ_IDX=%d", XPQ_IDX));
315 xen_invlpg(vm_offset_t va)
318 op.cmd = MMUEXT_INVLPG_ALL;
319 op.arg1.linear_addr = va & ~PAGE_MASK;
320 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
324 xen_load_cr3(u_int val)
330 KASSERT(XPQ_IDX == 0, ("pending operations XPQ_IDX=%d", XPQ_IDX));
332 op.cmd = MMUEXT_NEW_BASEPTR;
333 op.arg1.mfn = xpmap_ptom(val) >> PAGE_SHIFT;
334 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
338 static __inline u_int
343 __asm __volatile("movl 4(%%ebp),%0" : "=r" (data));
354 eflags = _read_eflags();
355 _vcpu = &HYPERVISOR_shared_info->vcpu_info[smp_processor_id()];
356 if (_vcpu->evtchn_upcall_mask)
363 write_eflags(u_int eflags)
367 CTR2(KTR_SPARE2, "%x xen_restore_flags eflags %x", rebp(), eflags);
368 intr = ((eflags & PSL_I) == 0);
369 __restore_flags(intr);
370 _write_eflags(eflags);
376 CTR1(KTR_SPARE2, "%x xen_cli disabling interrupts", rebp());
383 CTR1(KTR_SPARE2, "%x xen_sti enabling interrupts", rebp());
391 return (HYPERVISOR_shared_info->vcpu_info[curcpu].arch.cr2);
395 _xen_machphys_update(vm_paddr_t mfn, vm_paddr_t pfn, char *file, int line)
399 if (__predict_true(gdtset))
401 XPQ_QUEUE[XPQ_IDX].ptr = (mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE;
402 XPQ_QUEUE[XPQ_IDX].val = pfn;
404 XPQ_QUEUE_LOG[XPQ_IDX].file = file;
405 XPQ_QUEUE_LOG[XPQ_IDX].line = line;
408 if (__predict_true(gdtset))
412 extern struct rwlock pvh_global_lock;
415 _xen_queue_pt_update(vm_paddr_t ptr, vm_paddr_t val, char *file, int line)
419 if (__predict_true(gdtset))
420 rw_assert(&pvh_global_lock, RA_WLOCKED);
422 KASSERT((ptr & 7) == 0, ("misaligned update"));
424 if (__predict_true(gdtset))
427 XPQ_QUEUE[XPQ_IDX].ptr = ((uint64_t)ptr) | MMU_NORMAL_PT_UPDATE;
428 XPQ_QUEUE[XPQ_IDX].val = (uint64_t)val;
430 XPQ_QUEUE_LOG[XPQ_IDX].file = file;
431 XPQ_QUEUE_LOG[XPQ_IDX].line = line;
434 if (__predict_true(gdtset))
439 xen_pgdpt_pin(vm_paddr_t ma)
442 op.cmd = MMUEXT_PIN_L3_TABLE;
443 op.arg1.mfn = ma >> PAGE_SHIFT;
445 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
449 xen_pgd_pin(vm_paddr_t ma)
452 op.cmd = MMUEXT_PIN_L2_TABLE;
453 op.arg1.mfn = ma >> PAGE_SHIFT;
455 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
459 xen_pgd_unpin(vm_paddr_t ma)
462 op.cmd = MMUEXT_UNPIN_TABLE;
463 op.arg1.mfn = ma >> PAGE_SHIFT;
465 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
469 xen_pt_pin(vm_paddr_t ma)
472 op.cmd = MMUEXT_PIN_L1_TABLE;
473 op.arg1.mfn = ma >> PAGE_SHIFT;
475 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
479 xen_pt_unpin(vm_paddr_t ma)
482 op.cmd = MMUEXT_UNPIN_TABLE;
483 op.arg1.mfn = ma >> PAGE_SHIFT;
485 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
489 xen_set_ldt(vm_paddr_t ptr, unsigned long len)
492 op.cmd = MMUEXT_SET_LDT;
493 op.arg1.linear_addr = ptr;
494 op.arg2.nr_ents = len;
496 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
499 void xen_tlb_flush(void)
502 op.cmd = MMUEXT_TLB_FLUSH_LOCAL;
504 PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
508 xen_update_descriptor(union descriptor *table, union descriptor *entry)
513 ptp = vtopte((vm_offset_t)table);
514 pa = (*ptp & PG_FRAME) | ((vm_offset_t)table & PAGE_MASK);
515 if (HYPERVISOR_update_descriptor(pa, *(uint64_t *)entry))
516 panic("HYPERVISOR_update_descriptor failed\n");
522 * Bitmap is indexed by page number. If bit is set, the page is part of a
523 * xen_create_contiguous_region() area of memory.
525 unsigned long *contiguous_bitmap;
528 contiguous_bitmap_set(unsigned long first_page, unsigned long nr_pages)
530 unsigned long start_off, end_off, curr_idx, end_idx;
532 curr_idx = first_page / BITS_PER_LONG;
533 start_off = first_page & (BITS_PER_LONG-1);
534 end_idx = (first_page + nr_pages) / BITS_PER_LONG;
535 end_off = (first_page + nr_pages) & (BITS_PER_LONG-1);
537 if (curr_idx == end_idx) {
538 contiguous_bitmap[curr_idx] |=
539 ((1UL<<end_off)-1) & -(1UL<<start_off);
541 contiguous_bitmap[curr_idx] |= -(1UL<<start_off);
542 while ( ++curr_idx < end_idx )
543 contiguous_bitmap[curr_idx] = ~0UL;
544 contiguous_bitmap[curr_idx] |= (1UL<<end_off)-1;
549 contiguous_bitmap_clear(unsigned long first_page, unsigned long nr_pages)
551 unsigned long start_off, end_off, curr_idx, end_idx;
553 curr_idx = first_page / BITS_PER_LONG;
554 start_off = first_page & (BITS_PER_LONG-1);
555 end_idx = (first_page + nr_pages) / BITS_PER_LONG;
556 end_off = (first_page + nr_pages) & (BITS_PER_LONG-1);
558 if (curr_idx == end_idx) {
559 contiguous_bitmap[curr_idx] &=
560 -(1UL<<end_off) | ((1UL<<start_off)-1);
562 contiguous_bitmap[curr_idx] &= (1UL<<start_off)-1;
563 while ( ++curr_idx != end_idx )
564 contiguous_bitmap[curr_idx] = 0;
565 contiguous_bitmap[curr_idx] &= -(1UL<<end_off);
570 /* Ensure multi-page extents are contiguous in machine memory. */
572 xen_create_contiguous_region(vm_page_t pages, int npages)
574 unsigned long mfn, i, flags;
576 struct xen_memory_reservation reservation = {
581 set_xen_guest_handle(reservation.extent_start, &mfn);
585 /* can currently only handle power of two allocation */
586 PANIC_IF(ffs(npages) != fls(npages));
588 /* 0. determine order */
589 order = (ffs(npages) == fls(npages)) ? fls(npages) - 1 : fls(npages);
591 /* 1. give away machine pages. */
592 for (i = 0; i < (1 << order); i++) {
594 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
596 PFNTOMFN(pfn) = INVALID_P2M_ENTRY;
597 PANIC_IF(HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation) != 1);
601 /* 2. Get a new contiguous memory extent. */
602 reservation.extent_order = order;
603 /* xenlinux hardcodes this because of aacraid - maybe set to 0 if we're not
604 * running with a broxen driver XXXEN
606 reservation.address_bits = 31;
607 if (HYPERVISOR_memory_op(XENMEM_increase_reservation, &reservation) != 1)
610 /* 3. Map the new extent in place of old pages. */
611 for (i = 0; i < (1 << order); i++) {
613 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
614 xen_machphys_update(mfn+i, pfn);
615 PFNTOMFN(pfn) = mfn+i;
621 contiguous_bitmap_set(VM_PAGE_TO_PHYS(&pages[0]) >> PAGE_SHIFT, 1UL << order);
624 balloon_unlock(flags);
629 reservation.extent_order = 0;
630 reservation.address_bits = 0;
632 for (i = 0; i < (1 << order); i++) {
634 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
635 PANIC_IF(HYPERVISOR_memory_op(
636 XENMEM_increase_reservation, &reservation) != 1);
637 xen_machphys_update(mfn, pfn);
643 balloon_unlock(flags);
649 xen_destroy_contiguous_region(void *addr, int npages)
651 unsigned long mfn, i, flags, order, pfn0;
652 struct xen_memory_reservation reservation = {
657 set_xen_guest_handle(reservation.extent_start, &mfn);
659 pfn0 = vtophys(addr) >> PAGE_SHIFT;
661 scrub_pages(vstart, 1 << order);
663 /* can currently only handle power of two allocation */
664 PANIC_IF(ffs(npages) != fls(npages));
666 /* 0. determine order */
667 order = (ffs(npages) == fls(npages)) ? fls(npages) - 1 : fls(npages);
672 contiguous_bitmap_clear(vtophys(addr) >> PAGE_SHIFT, 1UL << order);
675 /* 1. Zap current PTEs, giving away the underlying pages. */
676 for (i = 0; i < (1 << order); i++) {
678 uint64_t new_val = 0;
679 pfn = vtomach((char *)addr + i*PAGE_SIZE) >> PAGE_SHIFT;
681 PANIC_IF(HYPERVISOR_update_va_mapping((vm_offset_t)((char *)addr + (i * PAGE_SIZE)), new_val, 0));
682 PFNTOMFN(pfn) = INVALID_P2M_ENTRY;
683 PANIC_IF(HYPERVISOR_memory_op(
684 XENMEM_decrease_reservation, &reservation) != 1);
687 /* 2. Map new pages in place of old pages. */
688 for (i = 0; i < (1 << order); i++) {
692 PANIC_IF(HYPERVISOR_memory_op(XENMEM_increase_reservation, &reservation) != 1);
694 new_val = mfn << PAGE_SHIFT;
695 PANIC_IF(HYPERVISOR_update_va_mapping((vm_offset_t)addr + (i * PAGE_SIZE),
696 new_val, PG_KERNEL));
697 xen_machphys_update(mfn, pfn);
703 balloon_unlock(flags);
706 extern vm_offset_t proc0kstack;
707 extern int vm86paddr, vm86phystk;
708 char *bootmem_start, *bootmem_current, *bootmem_end;
710 pteinfo_t *pteinfo_list;
711 void initvalues(start_info_t *startinfo);
713 struct xenstore_domain_interface;
714 extern struct xenstore_domain_interface *xen_store;
717 bootmem_alloc(unsigned int size)
721 retptr = bootmem_current;
722 PANIC_IF(retptr + size > bootmem_end);
723 bootmem_current += size;
729 bootmem_free(void *ptr, unsigned int size)
734 PANIC_IF(tptr != bootmem_current - size ||
735 bootmem_current - size < bootmem_start);
737 bootmem_current -= size;
742 xpmap_mtop2(vm_paddr_t mpa)
744 return ((machine_to_phys_mapping[mpa >> PAGE_SHIFT] << PAGE_SHIFT)
745 ) | (mpa & ~PG_FRAME);
749 xpmap_get_bootpde(vm_paddr_t va)
752 return ((pd_entry_t *)xen_start_info->pt_base)[va >> 22];
756 xpmap_get_vbootpde(vm_paddr_t va)
760 pde = xpmap_get_bootpde(va);
761 if ((pde & PG_V) == 0)
762 return (pde & ~PG_FRAME);
763 return (pde & ~PG_FRAME) |
764 (xpmap_mtop2(pde & PG_FRAME) + KERNBASE);
768 xpmap_get_bootptep(vm_paddr_t va)
772 pde = xpmap_get_vbootpde(va);
773 if ((pde & PG_V) == 0)
775 #define PT_MASK 0x003ff000 /* page table address bits */
776 return &(((pt_entry_t *)(pde & PG_FRAME))[(va & PT_MASK) >> PAGE_SHIFT]);
780 xpmap_get_bootpte(vm_paddr_t va)
783 return xpmap_get_bootptep(va)[0];
790 shift_phys_machine(unsigned long *phys_machine, int nr_pages)
793 unsigned long *tmp_page, *current_page, *next_page;
796 tmp_page = bootmem_alloc(PAGE_SIZE);
797 current_page = phys_machine + nr_pages - (PAGE_SIZE/sizeof(unsigned long));
798 next_page = current_page - (PAGE_SIZE/sizeof(unsigned long));
799 bcopy(phys_machine, tmp_page, PAGE_SIZE);
801 while (current_page > phys_machine) {
803 bcopy(next_page, tmp_page, PAGE_SIZE);
804 /* shift down page */
805 bcopy(current_page, next_page, PAGE_SIZE);
807 bcopy(tmp_page, current_page, PAGE_SIZE);
809 current_page -= (PAGE_SIZE/sizeof(unsigned long));
810 next_page -= (PAGE_SIZE/sizeof(unsigned long));
812 bootmem_free(tmp_page, PAGE_SIZE);
814 for (i = 0; i < nr_pages; i++) {
815 xen_machphys_update(phys_machine[i], i);
817 memset(phys_machine, INVALID_P2M_ENTRY, PAGE_SIZE);
820 #endif /* ADD_ISA_HOLE */
823 * Build a directory of the pages that make up our Physical to Machine
824 * mapping table. The Xen suspend/restore code uses this to find our
828 init_frame_list_list(void *arg)
830 unsigned long nr_pages = xen_start_info->nr_pages;
831 #define FPP (PAGE_SIZE/sizeof(xen_pfn_t))
834 xen_pfn_to_mfn_frame_list_list = malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK);
835 for (i = 0, j = 0, k = -1; i < nr_pages;
837 if ((j & (FPP - 1)) == 0) {
839 xen_pfn_to_mfn_frame_list[k] =
840 malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK);
841 xen_pfn_to_mfn_frame_list_list[k] =
842 VTOMFN(xen_pfn_to_mfn_frame_list[k]);
845 xen_pfn_to_mfn_frame_list[k][j] =
846 VTOMFN(&xen_phys_machine[i]);
849 HYPERVISOR_shared_info->arch.max_pfn = nr_pages;
850 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list
851 = VTOMFN(xen_pfn_to_mfn_frame_list_list);
853 SYSINIT(init_fll, SI_SUB_DEVFS, SI_ORDER_ANY, init_frame_list_list, NULL);
855 extern unsigned long physfree;
860 extern uint32_t kernbase;
863 initvalues(start_info_t *startinfo)
865 vm_offset_t cur_space, cur_space_pt;
866 struct physdev_set_iopl set_iopl;
868 int l3_pages, l2_pages, l1_pages, offset;
869 vm_paddr_t console_page_ma, xen_store_ma;
873 vm_paddr_t IdlePDPTma, IdlePDPTnewma;
874 vm_paddr_t IdlePTDnewma[4];
875 pd_entry_t *IdlePDPTnew, *IdlePTDnew;
876 vm_paddr_t IdlePTDma[4];
878 vm_paddr_t IdlePTDma[1];
885 max((startinfo->nr_pages >> NPGPTD_SHIFT), nkpt),
886 NPGPTD*NPDEPG - KPTDI),
887 (HYPERVISOR_VIRT_START - KERNBASE) >> PDRSHIFT);
889 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
892 * need to install handler
894 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments_notify);
896 xen_start_info = startinfo;
897 HYPERVISOR_start_info = startinfo;
898 xen_phys_machine = (xen_pfn_t *)startinfo->mfn_list;
900 IdlePTD = (pd_entry_t *)((uint8_t *)startinfo->pt_base + PAGE_SIZE);
906 IdlePDPT = (pd_entry_t *)startinfo->pt_base;
907 IdlePDPTma = VTOM(startinfo->pt_base);
908 for (i = (KERNBASE >> 30);
909 (i < 4) && (IdlePDPT[i] != 0); i++)
912 * Note that only one page directory has been allocated at this point.
915 for (i = 0; i < l2_pages; i++)
916 IdlePTDma[i] = VTOM(IdlePTD + i*PAGE_SIZE);
918 l2_pages = (l2_pages == 0) ? 1 : l2_pages;
923 for (i = (((KERNBASE>>18) & PAGE_MASK)>>PAGE_SHIFT);
924 (i<l2_pages*NPDEPG) && (i<(VM_MAX_KERNEL_ADDRESS>>PDRSHIFT)); i++) {
931 /* number of pages allocated after the pts + 1*/;
932 cur_space = xen_start_info->pt_base +
933 (l3_pages + l2_pages + l1_pages + 1)*PAGE_SIZE;
935 xc_printf("initvalues(): wooh - availmem=%x,%x\n", avail_space,
938 xc_printf("KERNBASE=%x,pt_base=%lx, VTOPFN(base)=%x, nr_pt_frames=%lx\n",
939 KERNBASE,xen_start_info->pt_base, VTOPFN(xen_start_info->pt_base),
940 xen_start_info->nr_pt_frames);
941 xendebug_flags = 0; /* 0xffffffff; */
944 shift_phys_machine(xen_phys_machine, xen_start_info->nr_pages);
946 XENPRINTF("IdlePTD %p\n", IdlePTD);
947 XENPRINTF("nr_pages: %ld shared_info: 0x%lx flags: 0x%x pt_base: 0x%lx "
948 "mod_start: 0x%lx mod_len: 0x%lx\n",
949 xen_start_info->nr_pages, xen_start_info->shared_info,
950 xen_start_info->flags, xen_start_info->pt_base,
951 xen_start_info->mod_start, xen_start_info->mod_len);
954 IdlePDPTnew = (pd_entry_t *)cur_space; cur_space += PAGE_SIZE;
955 bzero(IdlePDPTnew, PAGE_SIZE);
957 IdlePDPTnewma = VTOM(IdlePDPTnew);
958 IdlePTDnew = (pd_entry_t *)cur_space; cur_space += 4*PAGE_SIZE;
959 bzero(IdlePTDnew, 4*PAGE_SIZE);
961 for (i = 0; i < 4; i++)
962 IdlePTDnewma[i] = VTOM((uint8_t *)IdlePTDnew + i*PAGE_SIZE);
966 * Copy the 4 machine addresses of the new PTDs in to the PDPT
969 for (i = 0; i < 4; i++)
970 IdlePDPTnew[i] = IdlePTDnewma[i] | PG_V;
975 * re-map the new PDPT read-only
977 PT_SET_MA(IdlePDPTnew, IdlePDPTnewma | PG_V);
980 * Unpin the current PDPT
982 xen_pt_unpin(IdlePDPTma);
986 /* Map proc0's KSTACK */
987 proc0kstack = cur_space; cur_space += (KSTACK_PAGES * PAGE_SIZE);
988 xc_printf("proc0kstack=%u\n", proc0kstack);
990 /* vm86/bios stack */
991 cur_space += PAGE_SIZE;
993 /* Map space for the vm86 region */
994 vm86paddr = (vm_offset_t)cur_space;
995 cur_space += (PAGE_SIZE * 3);
997 /* allocate 4 pages for bootmem allocator */
998 bootmem_start = bootmem_current = (char *)cur_space;
999 cur_space += (4 * PAGE_SIZE);
1000 bootmem_end = (char *)cur_space;
1002 /* allocate pages for gdt */
1003 gdt = (union descriptor *)cur_space;
1004 cur_space += PAGE_SIZE*ncpus;
1006 /* allocate page for ldt */
1007 ldt = (union descriptor *)cur_space; cur_space += PAGE_SIZE;
1008 cur_space += PAGE_SIZE;
1010 /* unmap remaining pages from initial chunk
1013 for (tmpva = cur_space; tmpva < (((uint32_t)&kernbase) + (l1_pages<<PDRSHIFT));
1014 tmpva += PAGE_SIZE) {
1015 bzero((char *)tmpva, PAGE_SIZE);
1016 PT_SET_MA(tmpva, (vm_paddr_t)0);
1021 memcpy(((uint8_t *)IdlePTDnew) + ((unsigned int)(KERNBASE >> 18)),
1022 ((uint8_t *)IdlePTD) + ((KERNBASE >> 18) & PAGE_MASK),
1023 l1_pages*sizeof(pt_entry_t));
1025 for (i = 0; i < 4; i++) {
1026 PT_SET_MA((uint8_t *)IdlePTDnew + i*PAGE_SIZE,
1027 IdlePTDnewma[i] | PG_V);
1029 xen_load_cr3(VTOP(IdlePDPTnew));
1030 xen_pgdpt_pin(VTOM(IdlePDPTnew));
1032 /* allocate remainder of nkpt pages */
1033 cur_space_pt = cur_space;
1034 for (offset = (KERNBASE >> PDRSHIFT), i = l1_pages; i < nkpt;
1035 i++, cur_space += PAGE_SIZE) {
1036 pdir = (offset + i) / NPDEPG;
1037 curoffset = ((offset + i) % NPDEPG);
1038 if (((offset + i) << PDRSHIFT) == VM_MAX_KERNEL_ADDRESS)
1042 * make sure that all the initial page table pages
1045 PT_SET_MA(cur_space, VTOM(cur_space) | PG_V | PG_RW);
1046 bzero((char *)cur_space, PAGE_SIZE);
1047 PT_SET_MA(cur_space, (vm_paddr_t)0);
1048 xen_pt_pin(VTOM(cur_space));
1049 xen_queue_pt_update((vm_paddr_t)(IdlePTDnewma[pdir] +
1050 curoffset*sizeof(vm_paddr_t)),
1051 VTOM(cur_space) | PG_KERNEL);
1055 for (i = 0; i < 4; i++) {
1056 pdir = (PTDPTDI + i) / NPDEPG;
1057 curoffset = (PTDPTDI + i) % NPDEPG;
1059 xen_queue_pt_update((vm_paddr_t)(IdlePTDnewma[pdir] +
1060 curoffset*sizeof(vm_paddr_t)),
1061 IdlePTDnewma[i] | PG_V);
1066 IdlePTD = IdlePTDnew;
1067 IdlePDPT = IdlePDPTnew;
1068 IdlePDPTma = IdlePDPTnewma;
1070 HYPERVISOR_shared_info = (shared_info_t *)cur_space;
1071 cur_space += PAGE_SIZE;
1073 xen_store = (struct xenstore_domain_interface *)cur_space;
1074 cur_space += PAGE_SIZE;
1076 console_page = (char *)cur_space;
1077 cur_space += PAGE_SIZE;
1080 * shared_info is an unsigned long so this will randomly break if
1081 * it is allocated above 4GB - I guess people are used to that
1082 * sort of thing with Xen ... sigh
1084 shinfo = xen_start_info->shared_info;
1085 PT_SET_MA(HYPERVISOR_shared_info, shinfo | PG_KERNEL);
1089 xen_store_ma = (((vm_paddr_t)xen_start_info->store_mfn) << PAGE_SHIFT);
1090 PT_SET_MA(xen_store, xen_store_ma | PG_KERNEL);
1091 console_page_ma = (((vm_paddr_t)xen_start_info->console.domU.mfn) << PAGE_SHIFT);
1092 PT_SET_MA(console_page, console_page_ma | PG_KERNEL);
1097 PANIC_IF(HYPERVISOR_physdev_op(PHYSDEVOP_SET_IOPL, &set_iopl));
1100 /* add page table for KERNBASE */
1101 xen_queue_pt_update(IdlePTDma + KPTDI*sizeof(vm_paddr_t),
1102 VTOM(cur_space) | PG_KERNEL);
1105 xen_queue_pt_update(pdir_shadow_ma[3] + KPTDI*sizeof(vm_paddr_t),
1106 VTOM(cur_space) | PG_V | PG_A);
1108 xen_queue_pt_update(pdir_shadow_ma + KPTDI*sizeof(vm_paddr_t),
1109 VTOM(cur_space) | PG_V | PG_A);
1112 cur_space += PAGE_SIZE;
1116 if (xen_start_info->flags & SIF_INITDOMAIN) {
1117 /* Map first megabyte */
1118 for (i = 0; i < (256 << PAGE_SHIFT); i += PAGE_SIZE)
1119 PT_SET_MA(KERNBASE + i, i | PG_KERNEL | PG_NC_PCD);
1124 * re-map kernel text read-only
1127 for (i = (((vm_offset_t)&btext) & ~PAGE_MASK);
1128 i < (((vm_offset_t)&etext) & ~PAGE_MASK); i += PAGE_SIZE)
1129 PT_SET_MA(i, VTOM(i) | PG_V | PG_A);
1132 physfree = VTOP(cur_space);
1133 init_first = physfree >> PAGE_SHIFT;
1134 IdlePTD = (pd_entry_t *)VTOP(IdlePTD);
1135 IdlePDPT = (pd_entry_t *)VTOP(IdlePDPT);
1136 setup_xen_features();
1137 xc_printf("#8, proc0kstack=%u\n", proc0kstack);
1141 trap_info_t trap_table[] = {
1142 { 0, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(div)},
1143 { 1, 0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(dbg)},
1144 { 3, 3|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(bpt)},
1145 { 4, 3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(ofl)},
1146 /* This is UPL on Linux and KPL on BSD */
1147 { 5, 3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(bnd)},
1148 { 6, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(ill)},
1149 { 7, 0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(dna)},
1151 * { 8, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(XXX)},
1152 * no handler for double fault
1154 { 9, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(fpusegm)},
1155 {10, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(tss)},
1156 {11, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(missing)},
1157 {12, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(stk)},
1158 {13, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(prot)},
1159 {14, 0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(page)},
1160 {15, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(rsvd)},
1161 {16, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(fpu)},
1162 {17, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(align)},
1163 {18, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(mchk)},
1164 {19, 0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(xmm)},
1165 {0x80, 3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(int0x80_syscall)},
1169 /* Perform a multicall and check that individual calls succeeded. */
1171 HYPERVISOR_multicall(struct multicall_entry * call_list, int nr_calls)
1176 /* Perform the multicall. */
1177 PANIC_IF(_HYPERVISOR_multicall(call_list, nr_calls));
1179 /* Check the results of individual hypercalls. */
1180 for (i = 0; i < nr_calls; i++)
1181 if (__predict_false(call_list[i].result < 0))
1183 if (__predict_false(ret > 0))
1184 panic("%d multicall(s) failed: cpu %d\n",
1185 ret, smp_processor_id());
1187 /* If we didn't panic already, everything succeeded. */
1191 /********** CODE WORTH KEEPING ABOVE HERE *****************/
1193 void xen_failsafe_handler(void);
1196 xen_failsafe_handler(void)
1199 panic("xen_failsafe_handler called!\n");
1202 void xen_handle_thread_switch(struct pcb *pcb);
1204 /* This is called by cpu_switch() when switching threads. */
1205 /* The pcb arg refers to the process control block of the */
1206 /* next thread which is to run */
1208 xen_handle_thread_switch(struct pcb *pcb)
1210 uint32_t *a = (uint32_t *)&PCPU_GET(fsgs_gdt)[0];
1211 uint32_t *b = (uint32_t *)&pcb->pcb_fsd;
1212 multicall_entry_t mcl[3];
1215 /* Notify Xen of task switch */
1216 mcl[i].op = __HYPERVISOR_stack_switch;
1217 mcl[i].args[0] = GSEL(GDATA_SEL, SEL_KPL);
1218 mcl[i++].args[1] = (unsigned long)pcb;
1220 /* Check for update of fsd */
1221 if (*a != *b || *(a+1) != *(b+1)) {
1222 mcl[i].op = __HYPERVISOR_update_descriptor;
1223 *(uint64_t *)&mcl[i].args[0] = vtomach((vm_offset_t)a);
1224 *(uint64_t *)&mcl[i++].args[2] = *(uint64_t *)b;
1230 /* Check for update of gsd */
1231 if (*a != *b || *(a+1) != *(b+1)) {
1232 mcl[i].op = __HYPERVISOR_update_descriptor;
1233 *(uint64_t *)&mcl[i].args[0] = vtomach((vm_offset_t)a);
1234 *(uint64_t *)&mcl[i++].args[2] = *(uint64_t *)b;
1237 (void)HYPERVISOR_multicall(mcl, i);