2 * Copyright (c) 2000,2001 Doug Rabson
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include "opt_compat.h"
32 #include "opt_msgbuf.h"
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/eventhandler.h>
37 #include <sys/sysproto.h>
38 #include <sys/signalvar.h>
39 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/reboot.h>
48 #include <sys/vmmeter.h>
49 #include <sys/msgbuf.h>
51 #include <sys/sysctl.h>
53 #include <sys/linker.h>
54 #include <sys/random.h>
56 #include <net/netisr.h>
58 #include <vm/vm_kern.h>
59 #include <vm/vm_page.h>
60 #include <vm/vm_map.h>
61 #include <vm/vm_extern.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_pager.h>
65 #include <sys/ptrace.h>
66 #include <machine/clock.h>
67 #include <machine/md_var.h>
68 #include <machine/reg.h>
69 #include <machine/fpu.h>
70 #include <machine/pal.h>
71 #include <machine/sal.h>
72 #include <machine/bootinfo.h>
73 #include <machine/mutex.h>
74 #include <machine/vmparam.h>
75 #include <machine/elf.h>
77 #include <sys/vnode.h>
78 #include <sys/ucontext.h>
79 #include <machine/sigframe.h>
80 #include <machine/efi.h>
81 #include <machine/inst.h>
82 #include <machine/rse.h>
83 #include <machine/unwind.h>
85 void ia64_probe_sapics(void);
86 void map_pal_code(void);
89 extern void ia64_ski_init(void);
92 u_int64_t processor_frequency;
93 u_int64_t bus_frequency;
94 u_int64_t itc_frequency;
96 struct bootinfo bootinfo;
98 struct mtx sched_lock;
101 extern char kstack[];
102 struct user *proc0uarea;
103 vm_offset_t proc0kstack;
105 extern u_int64_t kernel_text[], _end[];
106 extern u_int64_t _ia64_unwind_start[];
107 extern u_int64_t _ia64_unwind_end[];
109 FPSWA_INTERFACE *fpswa_interface;
111 u_int64_t ia64_pal_base;
112 u_int64_t ia64_port_base;
114 char machine[] = "ia64";
115 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
117 static char cpu_model[128];
118 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0, "");
121 /* start and end of kernel symbol table */
122 void *ksym_start, *ksym_end;
125 int ia64_unaligned_print = 1; /* warn about unaligned accesses */
126 int ia64_unaligned_fix = 1; /* fix up unaligned accesses */
127 int ia64_unaligned_sigbus = 0; /* don't SIGBUS on fixed-up accesses */
129 SYSCTL_INT(_machdep, CPU_UNALIGNED_PRINT, unaligned_print,
130 CTLFLAG_RW, &ia64_unaligned_print, 0, "");
132 SYSCTL_INT(_machdep, CPU_UNALIGNED_FIX, unaligned_fix,
133 CTLFLAG_RW, &ia64_unaligned_fix, 0, "");
135 SYSCTL_INT(_machdep, CPU_UNALIGNED_SIGBUS, unaligned_sigbus,
136 CTLFLAG_RW, &ia64_unaligned_sigbus, 0, "");
138 static void cpu_startup __P((void *));
139 SYSINIT(cpu, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL)
141 struct msgbuf *msgbufp=0;
146 int totalphysmem; /* total amount of physical memory in system */
147 int physmem; /* physical memory used by NetBSD + some rsvd */
148 int resvmem; /* amount of memory reserved for PROM */
150 vm_offset_t phys_avail[20];
153 sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
155 int error = sysctl_handle_int(oidp, 0, ia64_ptob(physmem), req);
159 SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_INT|CTLFLAG_RD,
160 0, 0, sysctl_hw_physmem, "I", "");
163 sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
165 int error = sysctl_handle_int(oidp, 0,
166 ia64_ptob(physmem - cnt.v_wire_count), req);
170 SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
171 0, 0, sysctl_hw_usermem, "I", "");
173 SYSCTL_INT(_hw, OID_AUTO, availpages, CTLFLAG_RD, &physmem, 0, "");
175 /* must be 2 less so 0 0 can signal end of chunks */
176 #define PHYS_AVAIL_ARRAY_END ((sizeof(phys_avail) / sizeof(vm_offset_t)) - 2)
178 static void identifycpu __P((void));
180 struct kva_md_info kmi;
188 * Good {morning,afternoon,evening,night}.
192 /* startrtclock(); */
196 printf("real memory = %ld (%ldK bytes)\n", ia64_ptob(Maxmem), ia64_ptob(Maxmem) / 1024);
199 * Display any holes after the first chunk of extended memory.
204 printf("Physical memory chunk(s):\n");
205 for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
206 int size1 = phys_avail[indx + 1] - phys_avail[indx];
208 printf("0x%08lx - 0x%08lx, %d bytes (%d pages)\n", phys_avail[indx],
209 phys_avail[indx + 1] - 1, size1, size1 / PAGE_SIZE);
213 vm_ksubmap_init(&kmi);
215 #if defined(USERCONFIG)
216 #if defined(USERCONFIG_BOOT)
219 if (boothowto & RB_CONFIG)
223 cninit(); /* the preferred console may have changed */
227 printf("avail memory = %ld (%ldK bytes)\n", ptoa(cnt.v_free_count),
228 ptoa(cnt.v_free_count) / 1024);
230 if (fpswa_interface == NULL)
231 printf("Warning: no FPSWA package supplied\n");
233 printf("FPSWA Revision = 0x%lx, Entry = %p\n",
234 (long)fpswa_interface->Revision,
235 (void *)fpswa_interface->Fpswa);
238 * Set up buffers, so they can be used to read disk labels.
241 vm_pager_bufferinit();
245 * Traverse the MADT to discover IOSAPIC and Local SAPIC
253 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
255 KASSERT(size >= sizeof(struct pcpu) + sizeof(struct pcb),
256 ("%s: too small an allocation for pcpu", __func__));
257 pcpu->pc_pcb = (void*)(pcpu+1);
265 int number, revision, model, family, archrev;
269 * Assumes little-endian.
271 *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
272 *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
275 t = ia64_get_cpuid(3);
276 number = (t >> 0) & 0xff;
277 revision = (t >> 8) & 0xff;
278 model = (t >> 16) & 0xff;
279 family = (t >> 24) & 0xff;
280 archrev = (t >> 32) & 0xff;
283 strcpy(cpu_model, "Itanium");
284 else if (family == 0x1f)
285 strcpy(cpu_model, "McKinley");
287 snprintf(cpu_model, sizeof(cpu_model), "Family=%d", family);
289 features = ia64_get_cpuid(4);
291 printf("CPU: %s", cpu_model);
292 if (processor_frequency)
293 printf(" (%ld.%02ld-Mhz)\n",
294 (processor_frequency + 4999) / 1000000,
295 ((processor_frequency + 4999) / 10000) % 100);
298 printf(" Origin = \"%s\" Model = %d Revision = %d\n",
299 vendor, model, revision);
300 printf(" Features = 0x%b\n", (u_int32_t) features,
306 add_kernel_unwind_tables(void *arg)
309 * Register the kernel's unwind table.
311 ia64_add_unwind_table(kernel_text,
315 SYSINIT(unwind, SI_SUB_KMEM, SI_ORDER_ANY, add_kernel_unwind_tables, 0);
323 if (ia64_pal_base == 0)
326 bzero(&pte, sizeof(pte));
328 pte.pte_ma = PTE_MA_WB;
331 pte.pte_pl = PTE_PL_KERN;
332 pte.pte_ar = PTE_AR_RWX;
333 pte.pte_ppn = ia64_pal_base >> 12;
335 __asm __volatile("mov %0=psr;;" : "=r" (psr));
336 __asm __volatile("rsm psr.ic|psr.i;; srlz.i;;");
337 __asm __volatile("mov cr.ifa=%0" ::
338 "r"(IA64_PHYS_TO_RR7(ia64_pal_base)));
339 __asm __volatile("mov cr.itir=%0" :: "r"(28 << 2));
340 __asm __volatile("srlz.i;;");
341 __asm __volatile("itr.i itr[%0]=%1;;" ::
342 "r"(2), "r"(*(u_int64_t*)&pte));
343 __asm __volatile("srlz.i;;");
344 __asm __volatile("mov psr.l=%0;; srlz.i;;" :: "r" (psr));
348 calculate_frequencies(void)
350 struct ia64_sal_result sal;
351 struct ia64_pal_result pal;
353 sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
354 pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
356 if (sal.sal_status == 0 && pal.pal_status == 0) {
358 printf("Platform clock frequency %ld Hz\n",
360 printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
361 "ITC ratio %ld/%ld\n",
362 pal.pal_result[0] >> 32,
363 pal.pal_result[0] & ((1L << 32) - 1),
364 pal.pal_result[1] >> 32,
365 pal.pal_result[1] & ((1L << 32) - 1),
366 pal.pal_result[2] >> 32,
367 pal.pal_result[2] & ((1L << 32) - 1));
369 processor_frequency =
370 sal.sal_result[0] * (pal.pal_result[0] >> 32)
371 / (pal.pal_result[0] & ((1L << 32) - 1));
373 sal.sal_result[0] * (pal.pal_result[1] >> 32)
374 / (pal.pal_result[1] & ((1L << 32) - 1));
376 sal.sal_result[0] * (pal.pal_result[2] >> 32)
377 / (pal.pal_result[2] & ((1L << 32) - 1));
382 ia64_init(u_int64_t arg1, u_int64_t arg2)
385 vm_offset_t kernstart, kernend;
386 vm_offset_t kernstartpfn, kernendpfn, pfn0, pfn1;
388 EFI_MEMORY_DESCRIPTOR *md, *mdp;
391 /* NO OUTPUT ALLOWED UNTIL FURTHER NOTICE */
394 * TODO: Disable interrupts, floating point etc.
395 * Maybe flush cache and tlb
397 ia64_set_fpsr(IA64_FPSR_DEFAULT);
400 * TODO: Get critical system information (if possible, from the
401 * information provided by the boot program).
405 * Gross and disgusting hack. The bootinfo is written into
406 * memory at a fixed address.
408 bootinfo = *(struct bootinfo *) 0xe000000000508000;
409 if (bootinfo.bi_magic != BOOTINFO_MAGIC
410 || bootinfo.bi_version != 1) {
411 bzero(&bootinfo, sizeof(bootinfo));
412 bootinfo.bi_kernend = (vm_offset_t) round_page(_end);
416 * Look for the I/O ports first - we need them for console
419 mdcount = bootinfo.bi_memmap_size / bootinfo.bi_memdesc_size;
420 md = (EFI_MEMORY_DESCRIPTOR *) IA64_PHYS_TO_RR7(bootinfo.bi_memmap);
421 if (md == NULL || mdcount == 0) {
423 static EFI_MEMORY_DESCRIPTOR ski_md[2];
425 * XXX hack for ski. In reality, the loader will probably ask
426 * EFI and pass the results to us. Possibly, we will call EFI
429 ski_md[0].Type = EfiConventionalMemory;
430 ski_md[0].PhysicalStart = 2L*1024*1024;
431 ski_md[0].VirtualStart = 0;
432 ski_md[0].NumberOfPages = (64L*1024*1024)>>12;
433 ski_md[0].Attribute = EFI_MEMORY_WB;
435 ski_md[1].Type = EfiMemoryMappedIOPortSpace;
436 ski_md[1].PhysicalStart = 0xffffc000000;
437 ski_md[1].VirtualStart = 0;
438 ski_md[1].NumberOfPages = (64L*1024*1024)>>12;
439 ski_md[1].Attribute = EFI_MEMORY_UC;
446 for (i = 0, mdp = md; i < mdcount; i++,
447 mdp = NextMemoryDescriptor(mdp, bootinfo.bi_memdesc_size)) {
448 if (mdp->Type == EfiMemoryMappedIOPortSpace)
449 ia64_port_base = IA64_PHYS_TO_RR6(mdp->PhysicalStart);
450 else if (mdp->Type == EfiPalCode)
451 ia64_pal_base = mdp->PhysicalStart;
454 KASSERT(ia64_port_base != 0,
455 ("%s: no I/O memory region", __func__));
458 * Look at arguments passed to us and compute boothowto.
460 boothowto = bootinfo.bi_boothowto;
466 * Catch case of boot_verbose set in environment.
468 if ((p = getenv("boot_verbose")) != NULL) {
469 if (strcmp(p, "yes") == 0 || strcmp(p, "YES") == 0) {
470 boothowto |= RB_VERBOSE;
474 if (boothowto & RB_VERBOSE)
478 * Initialize the console before we print anything out.
482 /* OUTPUT NOW ALLOWED */
484 if (ia64_pal_base != 0) {
485 ia64_pal_base &= ~((1 << 28) - 1);
487 * We use a TR to map the first 256M of memory - this might
488 * cover the palcode too.
490 if (ia64_pal_base == 0)
491 printf("PAL code mapped by the kernel's TR\n");
493 printf("PAL code not found\n");
496 * Wire things up so we can call the firmware.
503 calculate_frequencies();
506 * Find the beginning and end of the kernel.
508 kernstart = trunc_page(kernel_text);
509 ksym_start = (void *)bootinfo.bi_symtab;
510 ksym_end = (void *)bootinfo.bi_esymtab;
511 kernend = (vm_offset_t)round_page(ksym_end);
512 /* But if the bootstrap tells us otherwise, believe it! */
513 if (bootinfo.bi_kernend)
514 kernend = round_page(bootinfo.bi_kernend);
515 preload_metadata = (caddr_t)bootinfo.bi_modulep;
517 kern_envp = static_env;
519 kern_envp = (caddr_t)bootinfo.bi_envp;
521 /* get fpswa interface */
522 fpswa_interface = (FPSWA_INTERFACE*)IA64_PHYS_TO_RR7(bootinfo.bi_fpswa);
524 /* Init basic tunables, including hz */
527 p = getenv("kernelname");
529 strncpy(kernelname, p, sizeof(kernelname) - 1);
531 kernstartpfn = atop(IA64_RR_MASK(kernstart));
532 kernendpfn = atop(IA64_RR_MASK(kernend));
535 * Size the memory regions and load phys_avail[] with the results.
539 * Find out how much memory is available, by looking at
540 * the memory descriptors.
544 printf("Memory descriptor count: %d\n", mdcount);
548 for (i = 0, mdp = md; i < mdcount; i++,
549 mdp = NextMemoryDescriptor(mdp, bootinfo.bi_memdesc_size)) {
551 printf("MD %d: type %d pa 0x%lx cnt 0x%lx\n", i,
557 pfn0 = ia64_btop(round_page(mdp->PhysicalStart));
558 pfn1 = ia64_btop(trunc_page(mdp->PhysicalStart
559 + mdp->NumberOfPages * 4096));
563 if (mdp->Type != EfiConventionalMemory) {
564 resvmem += (pfn1 - pfn0);
568 totalphysmem += (pfn1 - pfn0);
571 * We have a memory descriptors available for system
572 * software use. We must determine if this cluster
575 physmem += (pfn1 - pfn0);
576 if (pfn0 <= kernendpfn && kernstartpfn <= pfn1) {
578 * Must compute the location of the kernel
579 * within the segment.
582 printf("Descriptor %d contains kernel\n", i);
584 if (pfn0 < kernstartpfn) {
586 * There is a chunk before the kernel.
589 printf("Loading chunk before kernel: "
590 "0x%lx / 0x%lx\n", pfn0, kernstartpfn);
592 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
593 phys_avail[phys_avail_cnt+1] = ia64_ptob(kernstartpfn);
596 if (kernendpfn < pfn1) {
598 * There is a chunk after the kernel.
601 printf("Loading chunk after kernel: "
602 "0x%lx / 0x%lx\n", kernendpfn, pfn1);
604 phys_avail[phys_avail_cnt] = ia64_ptob(kernendpfn);
605 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
610 * Just load this cluster as one chunk.
613 printf("Loading descriptor %d: 0x%lx / 0x%lx\n", i,
616 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
617 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
622 phys_avail[phys_avail_cnt] = 0;
625 init_param2(physmem);
628 * Initialize error message buffer (at end of core).
631 size_t sz = round_page(MSGBUF_SIZE);
632 int i = phys_avail_cnt - 2;
634 /* shrink so that it'll fit in the last segment */
635 if (phys_avail[i+1] - phys_avail[i] < sz)
636 sz = phys_avail[i+1] - phys_avail[i];
638 phys_avail[i+1] -= sz;
639 msgbufp = (struct msgbuf*) IA64_PHYS_TO_RR7(phys_avail[i+1]);
641 msgbufinit(msgbufp, sz);
643 /* Remove the last segment if it now has no pages. */
644 if (phys_avail[i] == phys_avail[i+1]) {
649 /* warn if the message buffer had to be shrunk */
650 if (sz != round_page(MSGBUF_SIZE))
651 printf("WARNING: %ld bytes not available for msgbuf in last cluster (%ld used)\n",
652 round_page(MSGBUF_SIZE), sz);
656 proc_linkup(&proc0, &proc0.p_ksegrp, &proc0.p_kse, &thread0);
658 * Init mapping for u page(s) for proc 0
660 proc0uarea = (struct user *)pmap_steal_memory(UAREA_PAGES * PAGE_SIZE);
661 proc0kstack = (vm_offset_t)kstack;
662 proc0.p_uarea = proc0uarea;
663 thread0.td_kstack = proc0kstack;
664 thread0.td_pcb = (struct pcb *)
665 (thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
667 * Setup the global data for the bootstrap cpu.
669 pcpup = (struct pcpu *) pmap_steal_memory(PAGE_SIZE);
670 pcpu_init(pcpup, 0, PAGE_SIZE);
671 ia64_set_k4((u_int64_t) pcpup);
674 * Initialize the virtual memory system.
679 * Initialize the rest of proc 0's PCB.
681 * Set the kernel sp, reserving space for an (empty) trapframe,
682 * and make proc0's trapframe pointer point to it for sanity.
683 * Initialise proc0's backing store to start after u area.
685 * XXX what is all this +/- 16 stuff?
687 thread0.td_frame = (struct trapframe *)thread0.td_pcb - 1;
688 thread0.td_pcb->pcb_sp = (u_int64_t)thread0.td_frame - 16;
689 thread0.td_pcb->pcb_bspstore = (u_int64_t)proc0kstack;
691 /* Setup curproc so that mutexes work */
692 PCPU_SET(curthread, &thread0);
694 LIST_INIT(&thread0.td_contested);
697 * Initialise mutexes.
699 mtx_init(&Giant, "Giant", MTX_DEF | MTX_RECURSE);
700 mtx_init(&sched_lock, "sched lock", MTX_SPIN | MTX_RECURSE);
701 mtx_init(&proc0.p_mtx, "process lock", MTX_DEF);
705 * Initialize debuggers, and break into them if appropriate.
709 if (boothowto & RB_KDB) {
710 printf("Boot flags requested debugger\n");
717 ia64_running_in_simulator()
719 return bootinfo.bi_systab == 0;
723 bzero(void *buf, size_t len)
727 while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
731 while (len >= sizeof(u_long) * 8) {
733 *((u_long*) p + 1) = 0;
734 *((u_long*) p + 2) = 0;
735 *((u_long*) p + 3) = 0;
736 len -= sizeof(u_long) * 8;
737 *((u_long*) p + 4) = 0;
738 *((u_long*) p + 5) = 0;
739 *((u_long*) p + 6) = 0;
740 *((u_long*) p + 7) = 0;
741 p += sizeof(u_long) * 8;
743 while (len >= sizeof(u_long)) {
745 len -= sizeof(u_long);
757 u_int64_t start, end, now;
759 start = ia64_get_itc();
760 end = start + (itc_frequency * n) / 1000000;
761 /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
763 now = ia64_get_itc();
764 } while (now < end || (now > start && end < start));
768 * Send an interrupt to process.
770 * Stack is set up to allow sigcode stored
771 * at top to call routine, followed by kcall
772 * to sigreturn routine below. After sigreturn
773 * resets the signal mask, the stack, and the
774 * frame pointer, it returns to the user
778 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
782 struct trapframe *frame;
784 struct sigframe sf, *sfp;
786 int oonstack, rndfsize;
790 PROC_LOCK_ASSERT(p, MA_OWNED);
792 frame = td->td_frame;
793 oonstack = sigonstack(frame->tf_r[FRAME_SP]);
794 rndfsize = ((sizeof(sf) + 15) / 16) * 16;
797 * Make sure that we restore the entire trapframe after a
800 frame->tf_flags &= ~FRAME_SYSCALL;
802 /* save user context */
803 bzero(&sf, sizeof(struct sigframe));
804 sf.sf_uc.uc_sigmask = *mask;
805 sf.sf_uc.uc_stack = p->p_sigstk;
806 sf.sf_uc.uc_stack.ss_flags = (p->p_flag & P_ALTSTACK)
807 ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
808 sf.sf_uc.uc_mcontext.mc_flags = IA64_MC_FLAG_ONSTACK;
809 sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
811 sf.sf_uc.uc_mcontext.mc_nat = 0; /* XXX */
812 sf.sf_uc.uc_mcontext.mc_sp = frame->tf_r[FRAME_SP];
813 sf.sf_uc.uc_mcontext.mc_ip = (frame->tf_cr_iip
814 | ((frame->tf_cr_ipsr >> 41) & 3));
815 sf.sf_uc.uc_mcontext.mc_cfm = frame->tf_cr_ifs & ~(1<<31);
816 sf.sf_uc.uc_mcontext.mc_um = frame->tf_cr_ipsr & 0x1fff;
817 sf.sf_uc.uc_mcontext.mc_ar_rsc = frame->tf_ar_rsc;
818 sf.sf_uc.uc_mcontext.mc_ar_bsp = frame->tf_ar_bspstore;
819 sf.sf_uc.uc_mcontext.mc_ar_rnat = frame->tf_ar_rnat;
820 sf.sf_uc.uc_mcontext.mc_ar_ccv = frame->tf_ar_ccv;
821 sf.sf_uc.uc_mcontext.mc_ar_unat = frame->tf_ar_unat;
822 sf.sf_uc.uc_mcontext.mc_ar_fpsr = frame->tf_ar_fpsr;
823 sf.sf_uc.uc_mcontext.mc_ar_pfs = frame->tf_ar_pfs;
824 sf.sf_uc.uc_mcontext.mc_pr = frame->tf_pr;
826 bcopy(&frame->tf_b[0],
827 &sf.sf_uc.uc_mcontext.mc_br[0],
828 8 * sizeof(unsigned long));
829 sf.sf_uc.uc_mcontext.mc_gr[0] = 0;
830 bcopy(&frame->tf_r[0],
831 &sf.sf_uc.uc_mcontext.mc_gr[1],
832 31 * sizeof(unsigned long));
837 * Allocate and validate space for the signal handler
838 * context. Note that if the stack is in P0 space, the
839 * call to grow() is a nop, and the useracc() check
840 * will fail if the process has not already allocated
841 * the space with a `brk'.
843 if ((p->p_flag & P_ALTSTACK) != 0 && !oonstack &&
844 SIGISMEMBER(psp->ps_sigonstack, sig)) {
845 sbs = (u_int64_t) p->p_sigstk.ss_sp;
846 sfp = (struct sigframe *)((caddr_t)p->p_sigstk.ss_sp +
847 p->p_sigstk.ss_size - rndfsize);
851 sbs = (sbs + 15) & ~15;
852 sfp = (struct sigframe *)((u_int64_t)sfp & ~15);
853 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
854 p->p_sigstk.ss_flags |= SS_ONSTACK;
857 sfp = (struct sigframe *)(frame->tf_r[FRAME_SP] - rndfsize);
860 (void)grow_stack(p, (u_long)sfp);
862 if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid)
863 printf("sendsig(%d): sig %d ssp %p usp %p\n", p->p_pid,
866 if (!useracc((caddr_t)sfp, sizeof(sf), VM_PROT_WRITE)) {
868 if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid)
869 printf("sendsig(%d): useracc failed on sig %d\n",
873 * Process has trashed its stack; give it an illegal
874 * instruction to halt it in its tracks.
877 SIGACTION(p, SIGILL) = SIG_DFL;
878 SIGDELSET(p->p_sigignore, SIGILL);
879 SIGDELSET(p->p_sigcatch, SIGILL);
880 SIGDELSET(p->p_sigmask, SIGILL);
886 /* save the floating-point state, if necessary, then copy it. */
887 ia64_fpstate_save(td, 1);
888 sf.sf_uc.uc_mcontext.mc_ownedfp = td->td_md.md_flags & MDP_FPUSED;
889 bcopy(&td->td_pcb->pcb_fp,
890 (struct fpreg *)sf.sf_uc.uc_mcontext.mc_fpregs,
891 sizeof(struct fpreg));
892 sf.sf_uc.uc_mcontext.mc_fp_control = td->td_pcb.pcb_fp_control;
896 * copy the frame out to userland.
898 (void) copyout((caddr_t)&sf, (caddr_t)sfp, sizeof(sf));
900 if (sigdebug & SDB_FOLLOW)
901 printf("sendsig(%d): sig %d sfp %p code %lx\n", p->p_pid, sig,
906 * Set up the registers to return to sigcode.
908 frame->tf_cr_ipsr &= ~IA64_PSR_RI;
909 frame->tf_cr_iip = PS_STRINGS - (esigcode - sigcode);
910 frame->tf_r[FRAME_R1] = sig;
912 if (SIGISMEMBER(p->p_sigacts->ps_siginfo, sig)) {
913 frame->tf_r[FRAME_R15] = (u_int64_t)&(sfp->sf_si);
915 /* Fill in POSIX parts */
916 sf.sf_si.si_signo = sig;
917 sf.sf_si.si_code = code;
918 sf.sf_si.si_addr = (void*)frame->tf_cr_ifa;
921 frame->tf_r[FRAME_R15] = code;
923 frame->tf_r[FRAME_SP] = (u_int64_t)sfp - 16;
924 frame->tf_r[FRAME_R14] = sig;
925 frame->tf_r[FRAME_R15] = (u_int64_t) &sfp->sf_si;
926 frame->tf_r[FRAME_R16] = (u_int64_t) &sfp->sf_uc;
927 frame->tf_r[FRAME_R17] = (u_int64_t)catcher;
928 frame->tf_r[FRAME_R18] = sbs;
931 if (sigdebug & SDB_FOLLOW)
932 printf("sendsig(%d): pc %lx, catcher %lx\n", p->p_pid,
933 frame->tf_cr_iip, frame->tf_regs[FRAME_R4]);
934 if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid)
935 printf("sendsig(%d): sig %d returns\n",
941 * Stub to satisfy the reference to osigreturn in the syscall table. This
942 * is needed even for newer arches that don't support old signals because
943 * the syscall table is machine-independent.
946 osigreturn(struct thread *td, struct osigreturn_args *uap)
949 return (nosys(td, (struct nosys_args *)uap));
953 * System call to cleanup state after a signal
954 * has been taken. Reset signal mask and
955 * stack state from context left by sendsig (above).
956 * Return to previous pc and psl as specified by
957 * context left by sendsig. Check carefully to
958 * make sure that the user has not modified the
959 * state to gain improper privileges.
962 sigreturn(struct thread *td,
963 struct sigreturn_args /* {
964 ucontext_t *sigcntxp;
969 struct trapframe *frame = td->td_frame;
970 struct __mcontext *mcp;
978 if (sigdebug & SDB_FOLLOW)
979 printf("sigreturn: pid %d, scp %p\n", p->p_pid, ucp);
983 * Fetch the entire context structure at once for speed.
984 * We don't use a normal argument to simplify RSE handling.
986 if (copyin((caddr_t)frame->tf_r[FRAME_R4],
987 (caddr_t)&uc, sizeof(ucontext_t)))
990 if (frame->tf_ndirty != 0) {
991 printf("sigreturn: dirty user stacked registers\n");
995 * Restore the user-supplied information
997 mcp = &uc.uc_mcontext;
998 bcopy(&mcp->mc_br[0], &frame->tf_b[0], 8*sizeof(u_int64_t));
999 bcopy(&mcp->mc_gr[1], &frame->tf_r[0], 31*sizeof(u_int64_t));
1002 frame->tf_flags &= ~FRAME_SYSCALL;
1003 frame->tf_cr_iip = mcp->mc_ip & ~15;
1004 frame->tf_cr_ipsr &= ~IA64_PSR_RI;
1005 switch (mcp->mc_ip & 15) {
1007 frame->tf_cr_ipsr |= IA64_PSR_RI_1;
1010 frame->tf_cr_ipsr |= IA64_PSR_RI_2;
1013 frame->tf_cr_ipsr = ((frame->tf_cr_ipsr & ~0x1fff)
1014 | (mcp->mc_um & 0x1fff));
1015 frame->tf_pr = mcp->mc_pr;
1016 frame->tf_ar_rsc = (mcp->mc_ar_rsc & 3) | 12; /* user, loadrs=0 */
1017 frame->tf_ar_pfs = mcp->mc_ar_pfs;
1018 frame->tf_cr_ifs = mcp->mc_cfm | (1UL<<63);
1019 frame->tf_ar_bspstore = mcp->mc_ar_bsp;
1020 frame->tf_ar_rnat = mcp->mc_ar_rnat;
1021 frame->tf_ndirty = 0; /* assumes flushrs in sigcode */
1022 frame->tf_ar_unat = mcp->mc_ar_unat;
1023 frame->tf_ar_ccv = mcp->mc_ar_ccv;
1024 frame->tf_ar_fpsr = mcp->mc_ar_fpsr;
1026 frame->tf_r[FRAME_SP] = mcp->mc_sp;
1029 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
1030 if (uc.uc_mcontext.mc_onstack & 1)
1031 p->p_sigstk.ss_flags |= SS_ONSTACK;
1033 p->p_sigstk.ss_flags &= ~SS_ONSTACK;
1036 p->p_sigmask = uc.uc_sigmask;
1037 SIG_CANTMASK(p->p_sigmask);
1040 /* XXX ksc.sc_ownedfp ? */
1041 ia64_fpstate_drop(td);
1043 bcopy((struct fpreg *)uc.uc_mcontext.mc_fpregs,
1044 &td->td_pcb->pcb_fp, sizeof(struct fpreg));
1045 td->td_pcb->pcb_fp_control = uc.uc_mcontext.mc_fp_control;
1049 if (sigdebug & SDB_FOLLOW)
1050 printf("sigreturn(%d): returns\n", p->p_pid);
1052 return (EJUSTRETURN);
1056 * Machine dependent boot() routine
1062 ia64_efi_runtime->ResetSystem(EfiResetWarm, EFI_SUCCESS, 0, 0);
1066 * Shutdown the CPU as much as possible
1072 ia64_efi_runtime->ResetSystem(EfiResetWarm, EFI_SUCCESS, 0, 0);
1076 * Clear registers on exec
1079 setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
1081 struct trapframe *frame;
1083 frame = td->td_frame;
1086 * Make sure that we restore the entire trapframe after an
1089 frame->tf_flags &= ~FRAME_SYSCALL;
1091 bzero(frame->tf_r, sizeof(frame->tf_r));
1092 bzero(frame->tf_f, sizeof(frame->tf_f));
1093 frame->tf_cr_iip = entry;
1094 frame->tf_cr_ipsr = (IA64_PSR_IC
1101 | IA64_PSR_CPL_USER);
1103 * Make sure that sp is aligned to a 16 byte boundary and
1104 * reserve 16 bytes of scratch space for _start.
1106 frame->tf_r[FRAME_SP] = (stack & ~15) - 16;
1109 * Write values for out0, out1 and out2 to the user's backing
1110 * store and arrange for them to be restored into the user's
1111 * initial register frame. Assumes that (bspstore & 0x1f8) <
1114 frame->tf_ar_bspstore = td->td_md.md_bspstore + 24;
1115 suword((caddr_t) frame->tf_ar_bspstore - 24, stack);
1116 suword((caddr_t) frame->tf_ar_bspstore - 16, ps_strings);
1117 suword((caddr_t) frame->tf_ar_bspstore - 8, 0);
1118 frame->tf_ndirty = 0;
1119 frame->tf_cr_ifs = (1L<<63) | 3; /* sof=3, v=1 */
1121 frame->tf_ar_rsc = 0xf; /* user mode rsc */
1122 frame->tf_ar_fpsr = IA64_FPSR_DEFAULT;
1124 td->td_md.md_flags &= ~MDP_FPUSED;
1125 ia64_fpstate_drop(td);
1129 ptrace_set_pc(struct thread *td, unsigned long addr)
1131 /* TODO set pc in trapframe */
1136 ptrace_single_step(struct thread *td)
1138 /* TODO arrange for user process to single step */
1143 ia64_pa_access(vm_offset_t pa)
1145 return VM_PROT_READ|VM_PROT_WRITE;
1153 /* TODO copy trapframe to regs */
1162 /* TODO copy regs to trapframe */
1167 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1174 set_dbregs(struct thread *td, struct dbreg *dbregs)
1181 fill_fpregs(td, fpregs)
1183 struct fpreg *fpregs;
1185 /* TODO copy fpu state to fpregs */
1186 ia64_fpstate_save(td, 0);
1189 bcopy(&td->td_pcb->pcb_fp, fpregs, sizeof *fpregs);
1195 set_fpregs(td, fpregs)
1197 struct fpreg *fpregs;
1199 /* TODO copy fpregs fpu state */
1200 ia64_fpstate_drop(td);
1203 bcopy(fpregs, &td->td_pcb->pcb_fp, sizeof *fpregs);
1210 Debugger(const char *msg)
1212 printf("Debugger(\"%s\") called.\n", msg);
1216 #include <sys/disklabel.h>
1219 * Determine the size of the transfer, and make sure it is
1220 * within the boundaries of the partition. Adjust transfer
1221 * if needed, and signal errors or early completion.
1224 bounds_check_with_label(struct bio *bp, struct disklabel *lp, int wlabel)
1227 struct partition *p = lp->d_partitions + dkpart(bp->bio_dev);
1228 int labelsect = lp->d_partitions[0].p_offset;
1229 int maxsz = p->p_size,
1230 sz = (bp->bio_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT;
1232 /* overwriting disk label ? */
1233 /* XXX should also protect bootstrap in first 8K */
1234 if (bp->bio_blkno + p->p_offset <= LABELSECTOR + labelsect &&
1235 #if LABELSECTOR != 0
1236 bp->bio_blkno + p->p_offset + sz > LABELSECTOR + labelsect &&
1238 (bp->bio_cmd == BIO_WRITE) && wlabel == 0) {
1239 bp->bio_error = EROFS;
1243 #if defined(DOSBBSECTOR) && defined(notyet)
1244 /* overwriting master boot record? */
1245 if (bp->bio_blkno + p->p_offset <= DOSBBSECTOR &&
1246 (bp->bio_cmd == BIO_WRITE) && wlabel == 0) {
1247 bp->bio_error = EROFS;
1252 /* beyond partition? */
1253 if (bp->bio_blkno < 0 || bp->bio_blkno + sz > maxsz) {
1254 /* if exactly at end of disk, return an EOF */
1255 if (bp->bio_blkno == maxsz) {
1256 bp->bio_resid = bp->bio_bcount;
1259 /* or truncate if part of it fits */
1260 sz = maxsz - bp->bio_blkno;
1262 bp->bio_error = EINVAL;
1265 bp->bio_bcount = sz << DEV_BSHIFT;
1268 bp->bio_pblkno = bp->bio_blkno + p->p_offset;
1273 bp->bio_flags |= BIO_ERROR;
1279 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
1282 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
1284 if (!error && req->newptr)
1289 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
1290 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
1292 SYSCTL_INT(_machdep, CPU_DISRTCSET, disable_rtc_set,
1293 CTLFLAG_RW, &disable_rtc_set, 0, "");
1295 SYSCTL_INT(_machdep, CPU_WALLCLOCK, wall_cmos_clock,
1296 CTLFLAG_RW, &wall_cmos_clock, 0, "");
1299 ia64_fpstate_check(struct thread *td)
1301 if ((td->td_frame->tf_cr_ipsr & IA64_PSR_DFH) == 0)
1302 if (td != PCPU_GET(fpcurthread))
1303 panic("ia64_check_fpcurthread: bogus");
1307 * Save the high floating point state in the pcb. Use this to get
1308 * read-only access to the floating point state. If write is true, the
1309 * current fp process is cleared so that fp state can safely be
1310 * modified. The process will automatically reload the changed state
1311 * by generating a disabled fp trap.
1314 ia64_fpstate_save(struct thread *td, int write)
1316 if (td == PCPU_GET(fpcurthread)) {
1318 * Save the state in the pcb.
1320 savehighfp(td->td_pcb->pcb_highfp);
1323 td->td_frame->tf_cr_ipsr |= IA64_PSR_DFH;
1324 PCPU_SET(fpcurthread, NULL);
1330 * Relinquish ownership of the FP state. This is called instead of
1331 * ia64_save_fpstate() if the entire FP state is being changed
1332 * (e.g. on sigreturn).
1335 ia64_fpstate_drop(struct thread *td)
1337 if (td == PCPU_GET(fpcurthread)) {
1338 td->td_frame->tf_cr_ipsr |= IA64_PSR_DFH;
1339 PCPU_SET(fpcurthread, NULL);
1344 * Switch the current owner of the fp state to p, reloading the state
1348 ia64_fpstate_switch(struct thread *td)
1350 if (PCPU_GET(fpcurthread)) {
1352 * Dump the old fp state if its valid.
1354 savehighfp(PCPU_GET(fpcurthread)->td_pcb->pcb_highfp);
1355 PCPU_GET(fpcurthread)->td_frame->tf_cr_ipsr |= IA64_PSR_DFH;
1359 * Remember the new FP owner and reload its state.
1361 PCPU_SET(fpcurthread, td);
1362 restorehighfp(td->td_pcb->pcb_highfp);
1363 td->td_frame->tf_cr_ipsr &= ~IA64_PSR_DFH;
1365 td->td_md.md_flags |= MDP_FPUSED;
1369 * Utility functions for manipulating instruction bundles.
1372 ia64_unpack_bundle(u_int64_t low, u_int64_t high, struct ia64_bundle *bp)
1374 bp->template = low & 0x1f;
1375 bp->slot[0] = (low >> 5) & ((1L<<41) - 1);
1376 bp->slot[1] = (low >> 46) | ((high & ((1L<<23) - 1)) << 18);
1377 bp->slot[2] = (high >> 23);
1381 ia64_pack_bundle(u_int64_t *lowp, u_int64_t *highp,
1382 const struct ia64_bundle *bp)
1384 u_int64_t low, high;
1386 low = bp->template | (bp->slot[0] << 5) | (bp->slot[1] << 46);
1387 high = (bp->slot[1] >> 18) | (bp->slot[2] << 23);
1393 rse_slot(u_int64_t *bsp)
1395 return ((u_int64_t) bsp >> 3) & 0x3f;
1399 * Return the address of register regno (regno >= 32) given that bsp
1400 * points at the base of the register stack frame.
1403 ia64_rse_register_address(u_int64_t *bsp, int regno)
1405 int off = regno - 32;
1406 u_int64_t rnats = (rse_slot(bsp) + off) / 63;
1407 return bsp + off + rnats;
1411 * Calculate the base address of the previous frame given that the
1412 * current frame's locals area is 'size'.
1415 ia64_rse_previous_frame(u_int64_t *bsp, int size)
1417 int slot = rse_slot(bsp);
1421 while (count > slot) {
1426 return bsp - size - rnats;