2 * Copyright (c) 2006 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 #include <sys/types.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
41 #include <sys/random.h>
42 #include <sys/vkernel.h>
44 #include <sys/reboot.h>
46 #include <sys/msgbuf.h>
47 #include <sys/vmspace.h>
48 #include <sys/socket.h>
49 #include <sys/sockio.h>
50 #include <sys/sysctl.h>
52 #include <vm/vm_page.h>
53 #include <vm/vm_map.h>
54 #include <sys/mplock2.h>
58 #include <machine/cpu.h>
59 #include <machine/globaldata.h>
60 #include <machine/tls.h>
61 #include <machine/md_var.h>
62 #include <machine/vmparam.h>
63 #include <cpu/specialreg.h>
66 #include <net/if_arp.h>
67 #include <net/ethernet.h>
68 #include <net/bridge/if_bridgevar.h>
69 #include <netinet/in.h>
70 #include <arpa/inet.h>
71 #include <net/if_var.h>
85 #define EX_VKERNEL_REBOOT 32
87 vm_phystable_t phys_avail[16];
89 vm_paddr_t Maxmem_bytes;
92 struct vkdisk_info DiskInfo[VKDISK_MAX];
94 struct vknetif_info NetifInfo[VKNETIF_MAX];
100 vm_offset_t virtual_start;
101 vm_offset_t virtual_end;
102 vm_offset_t virtual2_start;
103 vm_offset_t virtual2_end;
104 vm_offset_t kernel_vm_end;
105 vm_offset_t crashdumpmap;
106 vm_offset_t clean_sva;
107 vm_offset_t clean_eva;
108 struct msgbuf *msgbufp;
111 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */
112 void *dmap_min_address;
114 u_int cpu_feature; /* XXX */
118 int64_t tsc_frequency;
119 int optcpus; /* number of cpus - see mp_start() */
121 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */
122 int real_ncpus; /* number of real CPUs */
123 int next_cpu; /* next real CPU to lock a virtual CPU to */
124 int vkernel_b_arg; /* no of logical CPU bits - only SMP */
125 int vkernel_B_arg; /* no of core bits - only SMP */
126 int vmm_enabled; /* VMM HW assisted enable */
127 int use_precise_timer = 0; /* use a precise timer (more expensive) */
128 struct privatespace *CPU_prvspace;
130 extern uint64_t KPML4phys; /* phys addr of kernel level 4 */
132 static struct trapframe proc0_tf;
133 static void *proc0paddr;
135 static void init_sys_memory(char *imageFile);
136 static void init_kern_memory(void);
137 static void init_kern_memory_vmm(void);
138 static void init_globaldata(void);
139 static void init_vkernel(void);
140 static void init_disk(char **diskExp, int *diskFlags, int diskFileNum, enum vkdisk_type type);
141 static void init_netif(char *netifExp[], int netifFileNum);
142 static void writepid(void);
143 static void cleanpid(void);
144 static int unix_connect(const char *path);
145 static void usage_err(const char *ctl, ...);
146 static void usage_help(_Bool);
147 static void init_locks(void);
148 static void handle_term(int);
153 static int prezeromem;
154 static char **save_av;
157 * Kernel startup for virtual kernels - standard main()
160 main(int ac, char **av)
162 char *memImageFile = NULL;
163 char *netifFile[VKNETIF_MAX];
164 char *diskFile[VKDISK_MAX];
165 char *cdFile[VKDISK_MAX];
170 int diskFlags[VKDISK_MAX];
171 int netifFileNum = 0;
174 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
182 int dflag = 0; /* disable vmm */
183 int real_vkernel_enable;
194 * Currently a bad hack but rtld-elf needs LD_SHAREDLIB_BASE to
195 * be set to force it to mmap() shared libraries into low memory,
196 * so our module loader can link against the related symbols.
198 if (getenv("LD_SHAREDLIB_BASE") == NULL) {
199 setenv("LD_SHAREDLIB_BASE", "0x10000000", 1);
201 fprintf(stderr, "Must run %s with full path\n", av[0]);
205 while ((childpid = fork()) != 0) {
207 bzero(&sa, sizeof(sa));
208 sigemptyset(&sa.sa_mask);
209 sa.sa_handler = SIG_IGN;
210 sigaction(SIGINT, &sa, NULL);
211 sigaction(SIGQUIT, &sa, NULL);
212 sigaction(SIGHUP, &sa, NULL);
215 * Forward SIGTERM to the child so that
216 * the shutdown process initiates correctly.
218 sa.sa_handler = handle_term;
219 sigaction(SIGTERM, &sa, NULL);
222 * Wait for child to terminate, exit if
223 * someone stole our child.
225 while (waitpid(childpid, &status, 0) != childpid) {
229 if (WEXITSTATUS(status) != EX_VKERNEL_REBOOT)
245 kernel_mem_readonly = 1;
250 lwp_cpu_lock = LCL_NONE;
252 real_vkernel_enable = 0;
253 vsize = sizeof(real_vkernel_enable);
254 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
256 if (real_vkernel_enable == 0) {
257 errx(1, "vm.vkernel_enable is 0, must be set "
258 "to 1 to execute a vkernel!");
262 vsize = sizeof(real_ncpus);
263 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
268 while ((c = getopt(ac, av, "c:hsvztl:m:n:r:R:e:i:p:I:Ud")) != -1) {
275 * name=value:name=value:name=value...
278 * Allow values to be quoted but note that shells
279 * may remove the quotes, so using this feature
280 * to embed colons may require a backslash.
287 kern_envp = malloc(kenv_size);
288 if (kern_envp == NULL)
289 errx(1, "Couldn't allocate %zd bytes for kern_envp", kenv_size);
291 kenv_size2 = kenv_size + n + 1;
293 if ((tmp = realloc(kern_envp, kenv_size2)) == NULL)
294 errx(1, "Couldn't reallocate %zd bytes for kern_envp", kenv_size2);
296 kenv_size = kenv_size2;
299 for (i = 0, j = pos; i < n; ++i) {
300 if (optarg[i] == '"')
302 else if (optarg[i] == '\'')
304 else if (isq == 0 && optarg[i] == ':')
307 kern_envp[j++] = optarg[i];
314 boothowto |= RB_SINGLE;
317 use_precise_timer = 1;
323 memImageFile = optarg;
326 if (netifFileNum < VKNETIF_MAX)
327 netifFile[netifFileNum++] = strdup(optarg);
333 if (diskFileNum + cdFileNum < VKDISK_MAX) {
334 diskFile[diskFileNum] = strdup(optarg);
335 diskFlags[diskFileNum] = (c == 'R');
342 if (diskFileNum + cdFileNum < VKDISK_MAX)
343 cdFile[cdFileNum++] = strdup(optarg);
346 Maxmem_bytes = strtoull(optarg, &suffix, 0);
363 usage_err("Bad maxmem option");
371 if (strncmp("map", optarg, 3) == 0) {
372 lwp_cpu_lock = LCL_PER_CPU;
373 if (optarg[3] == ',') {
374 next_cpu = strtol(optarg+4, &endp, 0);
376 usage_err("Bad target CPU number at '%s'", endp);
380 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
381 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
382 } else if (strncmp("any", optarg, 3) == 0) {
383 lwp_cpu_lock = LCL_NONE;
385 lwp_cpu_lock = LCL_SINGLE_CPU;
386 next_cpu = strtol(optarg, &endp, 0);
388 usage_err("Bad target CPU number at '%s'", endp);
389 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
390 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
395 * This value is set up by mp_start(), don't just
398 tok = strtok(optarg, ":");
399 optcpus = strtol(tok, NULL, 0);
400 if (optcpus < 1 || optcpus > MAXCPU)
401 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU);
403 while ((1 << cpu_bits) < optcpus)
407 * By default assume simple hyper-threading
410 vkernel_B_arg = cpu_bits - vkernel_b_arg;
413 * [:lbits[:cbits]] override # of cpu bits
414 * for logical and core extraction, supplying
415 * defaults for any omission.
417 tok = strtok(NULL, ":");
419 vkernel_b_arg = strtol(tok, NULL, 0);
420 vkernel_B_arg = cpu_bits - vkernel_b_arg;
422 /* :cbits argument */
423 tok = strtok(NULL, ":");
425 vkernel_B_arg = strtol(tok, NULL, 0);
433 kernel_mem_readonly = 0;
449 vsize = sizeof(vmm_enabled);
450 sysctlbyname("hw.vmm.enable", &vmm_enabled, &vsize, NULL, 0);
451 vmm_enabled = (vmm_enabled && !dflag);
456 /* use a MAP_ANON directly */
457 printf("VMM is available\n");
458 init_kern_memory_vmm();
460 printf("VMM is not available\n");
461 init_sys_memory(memImageFile);
469 vmm_guest = VMM_GUEST_VKERNEL;
474 vsize = sizeof(tsc_present);
475 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
476 vsize = sizeof(tsc_invariant);
477 sysctlbyname("hw.tsc_invariant", &tsc_invariant, &vsize, NULL, 0);
478 vsize = sizeof(tsc_mpsync);
479 sysctlbyname("hw.tsc_mpsync", &tsc_mpsync, &vsize, NULL, 0);
480 vsize = sizeof(tsc_frequency);
481 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
483 cpu_feature |= CPUID_TSC;
488 vsize = sizeof(supports_sse);
490 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
491 sysctlbyname("hw.mxcsr_mask", &mxcsr_mask, &msize, NULL, 0);
492 init_fpu(supports_sse);
494 cpu_feature |= CPUID_SSE | CPUID_FXSR;
497 * We boot from the first installed disk.
499 if (bootOnDisk == 1) {
500 init_disk(diskFile, diskFlags, diskFileNum, VKD_DISK);
501 init_disk(cdFile, NULL, cdFileNum, VKD_CD);
503 init_disk(cdFile, NULL, cdFileNum, VKD_CD);
504 init_disk(diskFile, diskFlags, diskFileNum, VKD_DISK);
507 init_netif(netifFile, netifFileNum);
514 /* SIGTERM handler */
523 * Initialize system memory. This is the virtual kernel's 'RAM'.
527 init_sys_memory(char *imageFile)
534 * Figure out the system memory image size. If an image file was
535 * specified and -m was not specified, use the image file's size.
537 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
538 Maxmem_bytes = (vm_paddr_t)st.st_size;
539 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
541 errx(1, "Cannot create new memory file %s unless "
542 "system memory size is specified with -m",
548 * Maxmem must be known at this time
550 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
551 errx(1, "Bad maxmem specification: 64MB minimum, "
552 "multiples of %dMB only",
553 SEG_SIZE / 1024 / 1024);
558 * Generate an image file name if necessary, then open/create the
559 * file exclusively locked. Do not allow multiple virtual kernels
560 * to use the same image file.
562 * Don't iterate through a million files if we do not have write
563 * access to the directory, stop if our open() failed on a
564 * non-existant file. Otherwise opens can fail for any number
566 if (imageFile == NULL) {
567 for (i = 0; i < 1000000; ++i) {
568 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
570 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
571 if (fd < 0 && stat(imageFile, &st) == 0) {
578 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
580 fprintf(stderr, "Using memory file: %s\n", imageFile);
581 if (fd < 0 || fstat(fd, &st) < 0) {
582 err(1, "Unable to open/create %s", imageFile);
587 * Truncate or extend the file as necessary. Clean out the contents
588 * of the file, we want it to be full of holes so we don't waste
589 * time reading in data from an old file that we no longer care
593 ftruncate(fd, Maxmem_bytes);
596 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
601 * Initialize kernel memory. This reserves kernel virtual memory by using
607 init_kern_memory(void)
614 * Memory map our kernel virtual memory space. Note that the
615 * kernel image itself is not made part of this memory for the
618 * The memory map must be segment-aligned so we can properly
621 * If the system kernel has a different MAXDSIZ, it might not
622 * be possible to map kernel memory in its prefered location.
623 * Try a number of different locations.
626 base = mmap((void*)KERNEL_KVA_START, KERNEL_KVA_SIZE,
627 PROT_READ|PROT_WRITE|PROT_EXEC,
628 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED|MAP_TRYFIXED,
629 MemImageFd, (off_t)KERNEL_KVA_START);
631 if (base == MAP_FAILED) {
632 err(1, "Unable to mmap() kernel virtual memory!");
635 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
636 KvaStart = (vm_offset_t)base;
637 KvaSize = KERNEL_KVA_SIZE;
638 KvaEnd = KvaStart + KvaSize;
640 /* cannot use kprintf yet */
641 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
644 dmap_min_address = mmap(0, DMAP_SIZE, PROT_READ|PROT_WRITE,
645 MAP_NOCORE|MAP_NOSYNC|MAP_SHARED,
647 if (dmap_min_address == MAP_FAILED) {
648 err(1, "Unable to mmap() kernel DMAP region!");
653 * Prefault the memory. The vkernel is going to fault it all in
654 * anyway, and faults on the backing store itself are very expensive
655 * once we go SMP (contend a lot). So do it now.
658 bzero(dmap_min_address, Maxmem_bytes);
661 * Bootstrap the kernel_pmap
664 pmap_bootstrap((vm_paddr_t *)&firstfree, (int64_t)base);
666 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
667 0 | VPTE_RW | VPTE_V);
670 * phys_avail[] represents unallocated physical memory. MI code
671 * will use phys_avail[] to create the vm_page array.
673 phys_avail[0].phys_beg = (vm_paddr_t)firstfree;
674 phys_avail[0].phys_beg = (phys_avail[0].phys_beg + PAGE_MASK) &
675 ~(vm_paddr_t)PAGE_MASK;
676 phys_avail[0].phys_end = Maxmem_bytes;
680 * (virtual_start, virtual_end) represent unallocated kernel virtual
681 * memory. MI code will create kernel_map using these parameters.
683 virtual_start = KvaStart + (long)firstfree;
684 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
685 virtual_end = KvaStart + KERNEL_KVA_SIZE;
689 * pmap_growkernel() will set the correct value.
694 * Allocate space for process 0's UAREA.
696 proc0paddr = (void *)virtual_start;
697 for (i = 0; i < UPAGES; ++i) {
698 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg);
699 virtual_start += PAGE_SIZE;
700 phys_avail[0].phys_beg += PAGE_SIZE;
706 crashdumpmap = virtual_start;
707 virtual_start += MAXDUMPPGS * PAGE_SIZE;
710 * msgbufp maps the system message buffer
712 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
713 msgbufp = (void *)virtual_start;
714 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
715 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg);
716 virtual_start += PAGE_SIZE;
717 phys_avail[0].phys_beg += PAGE_SIZE;
719 msgbufinit(msgbufp, MSGBUF_SIZE);
722 * used by kern_memio for /dev/mem access
724 ptvmmap = (caddr_t)virtual_start;
725 virtual_start += PAGE_SIZE;
730 init_kern_memory_vmm(void)
734 struct vmm_guest_options options;
737 KvaStart = (vm_offset_t)KERNEL_KVA_START;
738 KvaSize = KERNEL_KVA_SIZE;
739 KvaEnd = KvaStart + KvaSize;
741 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
744 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
745 errx(1, "Bad maxmem specification: 64MB minimum, "
746 "multiples of %dMB only",
747 SEG_SIZE / 1024 / 1024);
751 /* Call the vmspace_create to allocate the internal
752 * vkernel structures. Won't do anything else (no new
755 if (vmspace_create(NULL, 0, NULL) < 0)
756 panic("vmspace_create() failed");
760 * MAP_ANON the region of the VKERNEL phyisical memory
761 * (known as GPA - Guest Physical Address
763 dmap_address = mmap(NULL, Maxmem_bytes,
764 PROT_READ|PROT_WRITE|PROT_EXEC,
765 MAP_ANON|MAP_SHARED, -1, 0);
766 if (dmap_address == MAP_FAILED) {
767 err(1, "Unable to mmap() RAM region!");
771 bzero(dmap_address, Maxmem_bytes);
773 /* Alloc a new stack in the lowmem */
774 vkernel_stack = mmap(NULL, KERNEL_STACK_SIZE,
775 PROT_READ|PROT_WRITE|PROT_EXEC|PROT_EXEC,
777 if (vkernel_stack == MAP_FAILED) {
778 err(1, "Unable to allocate stack\n");
782 * Bootstrap the kernel_pmap
784 firstfree = dmap_address;
785 dmap_min_address = NULL; /* VIRT == PHYS in the first 512G */
786 pmap_bootstrap((vm_paddr_t *)&firstfree, (uint64_t)KvaStart);
791 bzero(&options, sizeof(options));
792 options.guest_cr3 = (register_t) KPML4phys;
793 options.new_stack = (uint64_t) vkernel_stack + KERNEL_STACK_SIZE;
795 if (vmm_guest_ctl(VMM_GUEST_RUN, &options)) {
796 err(1, "Unable to enter VMM mode.");
800 * phys_avail[] represents unallocated physical memory. MI code
801 * will use phys_avail[] to create the vm_page array.
803 phys_avail[0].phys_beg = (vm_paddr_t)firstfree;
804 phys_avail[0].phys_beg = (phys_avail[0].phys_beg + PAGE_MASK) &
805 ~(vm_paddr_t)PAGE_MASK;
806 phys_avail[0].phys_end = (vm_paddr_t)dmap_address + Maxmem_bytes;
809 * pmap_growkernel() will set the correct value.
814 * Allocate space for process 0's UAREA.
816 proc0paddr = (void *)virtual_start;
817 for (i = 0; i < UPAGES; ++i) {
818 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg);
819 virtual_start += PAGE_SIZE;
820 phys_avail[0].phys_beg += PAGE_SIZE;
826 crashdumpmap = virtual_start;
827 virtual_start += MAXDUMPPGS * PAGE_SIZE;
830 * msgbufp maps the system message buffer
832 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
833 msgbufp = (void *)virtual_start;
834 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
836 pmap_kenter_quick(virtual_start, phys_avail[0].phys_beg);
837 virtual_start += PAGE_SIZE;
838 phys_avail[0].phys_beg += PAGE_SIZE;
841 msgbufinit(msgbufp, MSGBUF_SIZE);
844 * used by kern_memio for /dev/mem access
846 ptvmmap = (caddr_t)virtual_start;
847 virtual_start += PAGE_SIZE;
849 printf("vmm: Hardware pagetable enabled for guest\n");
854 * Map the per-cpu globaldata for cpu #0. Allocate the space using
855 * virtual_start and phys_avail[0]
859 init_globaldata(void)
866 * Reserve enough KVA to cover possible cpus. This is a considerable
867 * amount of KVA since the privatespace structure includes two
868 * whole page table mappings.
870 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
871 CPU_prvspace = (void *)virtual_start;
872 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
875 * Allocate enough physical memory to cover the mdglobaldata
876 * portion of the space and the idle stack and map the pages
877 * into KVA. For cpu #0 only.
879 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
880 pa = phys_avail[0].phys_beg;
881 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
882 pmap_kenter_quick(va, pa);
883 phys_avail[0].phys_beg += PAGE_SIZE;
885 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
886 pa = phys_avail[0].phys_beg;
887 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
888 pmap_kenter_quick(va, pa);
889 phys_avail[0].phys_beg += PAGE_SIZE;
893 * Setup the %gs for cpu #0. The mycpu macro works after this
894 * point. Note that %fs is used by pthreads.
896 tls_set_gs(&CPU_prvspace[0], sizeof(struct privatespace));
901 * Initialize pool tokens and other necessary locks
908 * Get the initial mplock with a count of 1 for the BSP.
909 * This uses a LOGICAL cpu ID, ie BSP == 0.
911 cpu_get_initial_mplock();
913 /* our token pool needs to work early */
914 lwkt_token_pool_init();
920 * Initialize very low level systems including thread0, proc0, etc.
926 struct mdglobaldata *gd;
928 gd = &CPU_prvspace[0].mdglobaldata;
929 bzero(gd, sizeof(*gd));
931 gd->mi.gd_curthread = &thread0;
932 thread0.td_gd = &gd->mi;
934 ncpus2 = 1; /* rounded down power of 2 */
935 ncpus_fit = 1; /* rounded up power of 2 */
936 /* ncpus2_mask and ncpus_fit_mask are 0 */
938 gd->mi.gd_prvspace = &CPU_prvspace[0];
939 mi_gdinit(&gd->mi, 0);
941 mi_proc0init(&gd->mi, proc0paddr);
942 lwp0.lwp_md.md_regs = &proc0_tf;
947 #if 0 /* #ifdef DDB */
949 if (boothowto & RB_KDB)
950 Debugger("Boot flags requested debugger");
954 initializecpu(); /* Initialize CPU registers */
956 init_param2((phys_avail[0].phys_end -
957 phys_avail[0].phys_beg) / PAGE_SIZE);
961 * Map the message buffer
963 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
964 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
965 msgbufinit(msgbufp, MSGBUF_SIZE);
968 thread0.td_pcb_cr3 ... MMU
969 lwp0.lwp_md.md_regs = &proc0_tf;
974 * Filesystem image paths for the virtual kernel are optional.
975 * If specified they each should point to a disk image,
976 * the first of which will become the root disk.
978 * The virtual kernel caches data from our 'disk' just like a normal kernel,
979 * so we do not really want the real kernel to cache the data too. Use
980 * O_DIRECT to remove the duplication.
984 init_disk(char **diskExp, int *diskFlags, int diskFileNum, enum vkdisk_type type)
989 if (diskFileNum == 0)
992 for (i=0; i < diskFileNum; i++){
997 warnx("Invalid argument to '-r'");
1001 * Check for a serial number for the virtual disk
1002 * passed from the command line.
1005 strsep(&serno, ":");
1007 if (DiskNum < VKDISK_MAX) {
1009 struct vkdisk_info *info = NULL;
1013 if (type == VKD_DISK)
1014 fd = open(fname, O_RDWR|O_DIRECT, 0644);
1016 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
1017 if (fd < 0 || fstat(fd, &st) < 0) {
1018 err(1, "Unable to open/create %s", fname);
1021 if (S_ISREG(st.st_mode) && (diskFlags[i] & 1) == 0) {
1022 if (flock(fd, LOCK_EX|LOCK_NB) < 0) {
1023 errx(1, "Disk image %s is already "
1029 info = &DiskInfo[DiskNum];
1035 info->flags = diskFlags[i];
1036 memcpy(info->fname, fname, l);
1039 if ((info->serno = malloc(SERNOLEN)) != NULL)
1040 strlcpy(info->serno, serno, SERNOLEN);
1042 warnx("Couldn't allocate memory for the operation");
1046 if (type == VKD_CD) {
1047 rootdevnames[0] = "cd9660:vcd0";
1048 } else if (type == VKD_DISK) {
1049 rootdevnames[0] = "ufs:vkd0s0a";
1050 rootdevnames[1] = "ufs:vkd0s1a";
1056 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
1064 netif_set_tapflags(int tap_unit, int f, int s)
1069 bzero(&ifr, sizeof(ifr));
1071 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
1072 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
1073 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
1080 * If the flags are already set/cleared, then we return
1081 * immediately to avoid extra syscalls
1083 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
1085 /* Turn off flags */
1087 if ((flags & f) == 0)
1098 * Fix up ifreq.ifr_name, since it may be trashed
1099 * in previous ioctl(SIOCGIFFLAGS)
1101 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
1103 ifr.ifr_flags = flags & 0xffff;
1104 ifr.ifr_flagshigh = flags >> 16;
1105 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
1106 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
1114 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
1116 struct ifaliasreq ifra;
1117 struct sockaddr_in *in;
1119 bzero(&ifra, sizeof(ifra));
1120 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
1123 in = (struct sockaddr_in *)&ifra.ifra_addr;
1124 in->sin_family = AF_INET;
1125 in->sin_len = sizeof(*in);
1126 in->sin_addr.s_addr = addr;
1130 in = (struct sockaddr_in *)&ifra.ifra_mask;
1131 in->sin_len = sizeof(*in);
1132 in->sin_addr.s_addr = mask;
1135 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
1136 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
1144 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
1149 bzero(&ifbr, sizeof(ifbr));
1150 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
1153 bzero(&ifd, sizeof(ifd));
1154 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
1155 ifd.ifd_cmd = BRDGADD;
1156 ifd.ifd_len = sizeof(ifbr);
1157 ifd.ifd_data = &ifbr;
1159 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
1161 * 'errno == EEXIST' means that the tap(4) is already
1162 * a member of the bridge(4)
1164 if (errno != EEXIST) {
1165 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
1172 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
1175 * Locate the first unused tap(4) device file if auto mode is requested,
1176 * or open the user supplied device file, and bring up the corresponding
1179 * NOTE: Only tap(4) device file is supported currently
1183 netif_open_tap(const char *netif, int *tap_unit, int s)
1185 char tap_dev[MAXPATHLEN];
1192 if (strcmp(netif, "auto") == 0) {
1194 * Find first unused tap(4) device file
1196 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
1198 warnc(errno, "Unable to find a free tap(4)");
1203 * User supplied tap(4) device file or unix socket.
1205 if (netif[0] == '/') /* Absolute path */
1206 strlcpy(tap_dev, netif, sizeof(tap_dev));
1208 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
1210 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
1213 * If we cannot open normally try to connect to it.
1216 tap_fd = unix_connect(tap_dev);
1219 warn("Unable to open %s", tap_dev);
1225 * Check whether the device file is a tap(4)
1227 if (fstat(tap_fd, &st) < 0) {
1229 } else if (S_ISCHR(st.st_mode)) {
1230 dname = fdevname(tap_fd);
1232 dname = strstr(dname, "tap");
1235 * Bring up the corresponding tap(4) interface
1237 *tap_unit = strtol(dname + 3, NULL, 10);
1238 printf("TAP UNIT %d\n", *tap_unit);
1239 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
1246 } else if (S_ISSOCK(st.st_mode)) {
1248 * Special socket connection (typically to vknet). We
1249 * do not have to do anything.
1257 warnx("%s is not a tap(4) device or socket", tap_dev);
1266 unix_connect(const char *path)
1268 struct sockaddr_un sunx;
1271 int sndbuf = 262144;
1274 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
1275 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
1276 ++len; /* include nul */
1277 sunx.sun_family = AF_UNIX;
1280 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1283 if (connect(net_fd, (void *)&sunx, len) < 0) {
1287 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1288 if (fstat(net_fd, &st) == 0)
1289 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1290 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1296 #undef TAPDEV_OFLAGS
1299 * Following syntax is supported,
1300 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1302 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1303 * tap(4)'s netmask len is z
1305 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1306 * pseudo netif's address is y.y.y.y
1308 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1309 * pseudo netif's address is y.y.y.y
1310 * tap(4) and pseudo netif's netmask len are z
1312 * 5) bridgeX tap(4) will be added to bridgeX
1314 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1315 * pseudo netif's address is y.y.y.y
1317 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1318 * pseudo netif's address is y.y.y.y
1319 * pseudo netif's netmask len is z
1323 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1325 in_addr_t tap_addr, netmask, netif_addr;
1326 int next_netif_addr;
1327 char *tok, *masklen_str, *ifbridge;
1332 tok = strtok(NULL, ":/");
1335 * Nothing special, simply use tap(4) as backend
1340 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1342 * tap(4)'s address is supplied
1347 * If there is next token, then it may be pseudo
1348 * netif's address or netmask len for tap(4)
1350 next_netif_addr = 0;
1353 * Not tap(4)'s address, assume it as a bridge(4)
1360 * If there is next token, then it must be pseudo
1363 next_netif_addr = 1;
1366 netmask = netif_addr = 0;
1368 tok = strtok(NULL, ":/");
1372 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1373 if (next_netif_addr) {
1374 warnx("Invalid pseudo netif address: %s", tok);
1380 * Current token is not address, then it must be netmask len
1385 * Current token is pseudo netif address, if there is next token
1386 * it must be netmask len
1388 masklen_str = strtok(NULL, "/");
1391 /* Calculate netmask */
1392 if (masklen_str != NULL) {
1395 masklen = strtoul(masklen_str, NULL, 10);
1396 if (masklen < 32 && masklen > 0) {
1397 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1400 warnx("Invalid netmask len: %lu", masklen);
1405 /* Make sure there is no more token left */
1406 if (strtok(NULL, ":/") != NULL) {
1407 warnx("Invalid argument to '-I'");
1414 } else if (ifbridge == NULL) {
1415 /* Set tap(4) address/netmask */
1416 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1419 /* Tie tap(4) to bridge(4) */
1420 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1430 * NetifInfo[] will be filled for pseudo netif initialization.
1431 * NetifNum will be bumped to reflect the number of valid entries
1436 init_netif(char *netifExp[], int netifExpNum)
1441 if (netifExpNum == 0)
1444 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1448 for (i = 0; i < netifExpNum; ++i) {
1449 struct vknetif_info *info;
1450 in_addr_t netif_addr, netif_mask;
1451 int tap_fd, tap_unit;
1454 /* Extract MAC address if there is one */
1458 netif = strtok(netifExp[i], ":");
1459 if (netif == NULL) {
1460 warnx("Invalid argument to '-I'");
1465 * Open tap(4) device file and bring up the
1466 * corresponding interface
1468 tap_fd = netif_open_tap(netif, &tap_unit, s);
1473 * Initialize tap(4) and get address/netmask
1476 * NB: Rest part of netifExp[i] is passed
1477 * to netif_init_tap() implicitly.
1479 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1481 * NB: Closing tap(4) device file will bring
1482 * down the corresponding interface
1488 info = &NetifInfo[NetifNum];
1489 bzero(info, sizeof(*info));
1490 info->tap_fd = tap_fd;
1491 info->tap_unit = tap_unit;
1492 info->netif_addr = netif_addr;
1493 info->netif_mask = netif_mask;
1495 * If tmp isn't NULL it means a MAC could have been
1496 * specified so attempt to convert it.
1497 * Setting enaddr to NULL will tell vke_attach() we
1498 * need a pseudo-random MAC address.
1501 if ((info->enaddr = malloc(ETHER_ADDR_LEN)) == NULL)
1502 warnx("Couldn't allocate memory for the operation");
1504 if ((kether_aton(tmp, info->enaddr)) == NULL) {
1506 info->enaddr = NULL;
1512 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1519 * Create the pid file and leave it open and locked while the vkernel is
1520 * running. This allows a script to use /usr/bin/lockf to probe whether
1521 * a vkernel is still running (so as not to accidently kill an unrelated
1522 * process from a stale pid file).
1531 if (pid_file != NULL) {
1532 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid());
1533 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666);
1535 if (errno == EWOULDBLOCK) {
1536 perror("Failed to lock pidfile, "
1537 "vkernel already running");
1539 perror("Failed to create pidfile");
1544 write(fd, buf, strlen(buf));
1545 /* leave the file open to maintain the lock */
1553 if (pid_file != NULL) {
1554 if (unlink(pid_file) < 0)
1555 perror("Warning: couldn't remove pidfile");
1561 usage_err(const char *ctl, ...)
1566 vfprintf(stderr, ctl, va);
1568 fprintf(stderr, "\n");
1574 usage_help(_Bool help)
1576 fprintf(stderr, "Usage: %s [-hsUvd] [-c file] [-e name=value:name=value:...]\n"
1577 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n"
1578 "\t[-m size] [-n numcpus[:lbits[:cbits]]]\n"
1579 "\t[-p file] [-r file]\n", save_av[0]);
1582 fprintf(stderr, "\nArguments:\n"
1583 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n"
1584 "\t-e\tSpecify an environment to be used by the kernel.\n"
1585 "\t-h\tThis list of options.\n"
1586 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n"
1587 "\t-I\tCreate a virtual network device.\n"
1588 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n"
1589 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n"
1590 "\t-n\tSpecify the number of CPUs and the topology you wish to emulate:\n"
1591 "\t\t\tnumcpus - number of cpus\n"
1592 "\t\t\tlbits - specify the number of bits within APICID(=CPUID)\n"
1593 "\t\t\t needed for representing the logical ID.\n"
1594 "\t\t\t Controls the number of threads/core:\n"
1595 "\t\t\t (0 bits - 1 thread, 1 bit - 2 threads).\n"
1596 "\t\t\tcbits - specify the number of bits within APICID(=CPUID)\n"
1597 "\t\t\t needed for representing the core ID.\n"
1598 "\t\t\t Controls the number of cores/package:\n"
1599 "\t\t\t (0 bits - 1 core, 1 bit - 2 cores).\n"
1600 "\t-p\tSpecify a file in which to store the process ID.\n"
1601 "\t-r\tSpecify a R/W disk image file, iterates vkd0..n\n"
1602 "\t-R\tSpecify a COW disk image file, iterates vkd0..n\n"
1603 "\t-s\tBoot into single-user mode.\n"
1604 "\t-U\tEnable writing to kernel memory and module loading.\n"
1605 "\t-v\tTurn on verbose booting.\n");
1611 cpu_smp_stopped(void)
1618 kprintf("cpu reset, rebooting vkernel\n");
1621 exit(EX_VKERNEL_REBOOT);
1627 kprintf("cpu halt, exiting vkernel\n");
1635 switch(lwp_cpu_lock) {
1638 kprintf("Locking CPU%d to real cpu %d\n",
1640 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1642 if (next_cpu >= real_ncpus)
1645 case LCL_SINGLE_CPU:
1647 kprintf("Locking CPU%d to real cpu %d\n",
1649 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1652 /* do not map virtual cpus to real cpus */
1658 * Allocate and free memory for module loading. The loaded module
1659 * has to be placed somewhere near the current kernel binary load
1660 * point or the relocations will not work.
1662 * I'm not sure why this isn't working.
1665 vkernel_module_memory_alloc(vm_offset_t *basep, size_t bytes)
1669 xtra = (PAGE_SIZE - (vm_offset_t)sbrk(0)) & PAGE_MASK;
1670 *basep = (vm_offset_t)sbrk(xtra + bytes) + xtra;
1671 bzero((void *)*basep, bytes);
1673 *basep = (vm_offset_t)mmap((void *)0x000000000, bytes,
1674 PROT_READ|PROT_WRITE|PROT_EXEC|PROT_EXEC,
1675 MAP_ANON|MAP_SHARED, -1, 0);
1676 if ((void *)*basep == MAP_FAILED)
1683 vkernel_module_memory_free(vm_offset_t base, size_t bytes)
1687 munmap((void *)base, bytes);