2 * Copyright (c) 2003,2004 Marcel Moolenaar
3 * Copyright (c) 2000,2001 Doug Rabson
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include "opt_compat.h"
33 #include "opt_kstack_pages.h"
34 #include "opt_msgbuf.h"
35 #include "opt_sched.h"
37 #include <sys/param.h>
39 #include <sys/systm.h>
45 #include <sys/eventhandler.h>
47 #include <sys/imgact.h>
49 #include <sys/kernel.h>
50 #include <sys/linker.h>
52 #include <sys/malloc.h>
54 #include <sys/msgbuf.h>
56 #include <sys/ptrace.h>
57 #include <sys/random.h>
58 #include <sys/reboot.h>
59 #include <sys/sched.h>
60 #include <sys/signalvar.h>
61 #include <sys/syscall.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysproto.h>
64 #include <sys/ucontext.h>
67 #include <sys/vmmeter.h>
68 #include <sys/vnode.h>
72 #include <net/netisr.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_pager.h>
82 #include <machine/bootinfo.h>
83 #include <machine/clock.h>
84 #include <machine/cpu.h>
85 #include <machine/efi.h>
86 #include <machine/elf.h>
87 #include <machine/fpu.h>
88 #include <machine/mca.h>
89 #include <machine/md_var.h>
90 #include <machine/mutex.h>
91 #include <machine/pal.h>
92 #include <machine/pcb.h>
93 #include <machine/reg.h>
94 #include <machine/sal.h>
95 #include <machine/sigframe.h>
97 #include <machine/smp.h>
99 #include <machine/unwind.h>
100 #include <machine/vmparam.h>
102 #include <i386/include/specialreg.h>
104 u_int64_t processor_frequency;
105 u_int64_t bus_frequency;
106 u_int64_t itc_frequency;
109 u_int64_t pa_bootinfo;
110 struct bootinfo bootinfo;
114 extern u_int64_t kernel_text[], _end[];
116 extern u_int64_t ia64_gateway_page[];
117 extern u_int64_t break_sigtramp[];
118 extern u_int64_t epc_sigtramp[];
120 struct fpswa_iface *fpswa_iface;
122 u_int64_t ia64_pal_base;
123 u_int64_t ia64_port_base;
125 static int ia64_sync_icache_needed;
127 char machine[] = MACHINE;
128 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
130 static char cpu_model[64];
131 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0,
132 "The CPU model name");
134 static char cpu_family[64];
135 SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0,
136 "The CPU family name");
139 extern vm_offset_t ksym_start, ksym_end;
143 struct msgbuf *msgbufp = NULL;
145 /* Other subsystems (e.g., ACPI) can hook this later. */
146 void (*cpu_idle_hook)(void) = NULL;
151 #define PHYSMAP_SIZE (2 * VM_PHYSSEG_MAX)
153 vm_paddr_t phys_avail[PHYSMAP_SIZE + 2];
155 /* must be 2 less so 0 0 can signal end of chunks */
156 #define PHYS_AVAIL_ARRAY_END ((sizeof(phys_avail) / sizeof(vm_offset_t)) - 2)
158 struct kva_md_info kmi;
161 #define Ghz (1000L*Mhz)
167 char *family_name, *model_name;
168 u_int64_t features, tmp;
169 int number, revision, model, family, archrev;
172 * Assumes little-endian.
174 *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
175 *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
178 tmp = ia64_get_cpuid(3);
179 number = (tmp >> 0) & 0xff;
180 revision = (tmp >> 8) & 0xff;
181 model = (tmp >> 16) & 0xff;
182 family = (tmp >> 24) & 0xff;
183 archrev = (tmp >> 32) & 0xff;
185 family_name = model_name = "unknown";
188 family_name = "Itanium";
189 model_name = "Merced";
192 family_name = "Itanium 2";
195 model_name = "McKinley";
199 * Deerfield is a low-voltage variant based on the
200 * Madison core. We need circumstantial evidence
201 * (i.e. the clock frequency) to identify those.
202 * Allow for roughly 1% error margin.
204 tmp = processor_frequency >> 7;
205 if ((processor_frequency - tmp) < 1*Ghz &&
206 (processor_frequency + tmp) >= 1*Ghz)
207 model_name = "Deerfield";
209 model_name = "Madison";
212 model_name = "Madison II";
217 ia64_sync_icache_needed = 1;
219 family_name = "Itanium 2";
222 model_name = "Montecito";
227 snprintf(cpu_family, sizeof(cpu_family), "%s", family_name);
228 snprintf(cpu_model, sizeof(cpu_model), "%s", model_name);
230 features = ia64_get_cpuid(4);
232 printf("CPU: %s (", model_name);
233 if (processor_frequency) {
234 printf("%ld.%02ld-Mhz ",
235 (processor_frequency + 4999) / Mhz,
236 ((processor_frequency + 4999) / (Mhz/100)) % 100);
238 printf("%s)\n", family_name);
239 printf(" Origin = \"%s\" Revision = %d\n", vendor, revision);
240 printf(" Features = 0x%b\n", (u_int32_t) features,
242 "\001LB" /* long branch (brl) instruction. */
243 "\002SD" /* Spontaneous deferral. */
244 "\003AO" /* 16-byte atomic operations (ld, st, cmpxchg). */ );
248 cpu_startup(void *dummy)
252 struct pcpu_stats *pcs;
255 * Good {morning,afternoon,evening,night}.
262 printf("real memory = %ld (%ld MB)\n", ia64_ptob(Maxmem),
263 ia64_ptob(Maxmem) / 1048576);
267 * Display any holes after the first chunk of extended memory.
272 printf("Physical memory chunk(s):\n");
273 for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
274 long size1 = phys_avail[indx + 1] - phys_avail[indx];
276 printf("0x%08lx - 0x%08lx, %ld bytes (%ld pages)\n",
277 phys_avail[indx], phys_avail[indx + 1] - 1, size1,
278 size1 >> PAGE_SHIFT);
282 vm_ksubmap_init(&kmi);
284 printf("avail memory = %ld (%ld MB)\n", ptoa(cnt.v_free_count),
285 ptoa(cnt.v_free_count) / 1048576);
287 if (fpswa_iface == NULL)
288 printf("Warning: no FPSWA package supplied\n");
290 printf("FPSWA Revision = 0x%lx, Entry = %p\n",
291 (long)fpswa_iface->if_rev, (void *)fpswa_iface->if_fpswa);
294 * Set up buffers, so they can be used to read disk labels.
297 vm_pager_bufferinit();
300 * Traverse the MADT to discover IOSAPIC and Local SAPIC
307 * Create sysctl tree for per-CPU information.
309 SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
311 snprintf(nodename, sizeof(nodename), "cpu%u", pc->pc_cpuid);
312 sysctl_ctx_init(&pcs->pcs_sysctl_ctx);
313 pcs->pcs_sysctl_tree = SYSCTL_ADD_NODE(&pcs->pcs_sysctl_ctx,
314 SYSCTL_STATIC_CHILDREN(_machdep), OID_AUTO, nodename,
315 CTLFLAG_RD, NULL, "");
316 if (pcs->pcs_sysctl_tree == NULL)
319 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
320 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
321 "nasts", CTLFLAG_RD, &pcs->pcs_nasts,
322 "Number of IPI_AST interrupts");
324 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
325 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
326 "nclks", CTLFLAG_RD, &pcs->pcs_nclks,
327 "Number of clock interrupts");
329 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
330 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
331 "nextints", CTLFLAG_RD, &pcs->pcs_nextints,
332 "Number of ExtINT interrupts");
334 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
335 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
336 "nhighfps", CTLFLAG_RD, &pcs->pcs_nhighfps,
337 "Number of IPI_HIGH_FP interrupts");
339 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
340 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
341 "nhwints", CTLFLAG_RD, &pcs->pcs_nhwints,
342 "Number of hardware (device) interrupts");
344 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
345 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
346 "npreempts", CTLFLAG_RD, &pcs->pcs_npreempts,
347 "Number of IPI_PREEMPT interrupts");
349 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
350 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
351 "nrdvs", CTLFLAG_RD, &pcs->pcs_nrdvs,
352 "Number of IPI_RENDEZVOUS interrupts");
354 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
355 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
356 "nstops", CTLFLAG_RD, &pcs->pcs_nstops,
357 "Number of IPI_STOP interrupts");
359 SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx,
360 SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO,
361 "nstrays", CTLFLAG_RD, &pcs->pcs_nstrays,
362 "Number of stray vectors");
365 SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
375 cpu_flush_dcache(void *ptr, size_t len)
379 va = (uintptr_t)ptr & ~31;
380 lim = (uintptr_t)ptr + len;
389 /* Get current clock frequency for the given cpu id. */
391 cpu_est_clockrate(int cpu_id, uint64_t *rate)
394 if (pcpu_find(cpu_id) == NULL || rate == NULL)
396 *rate = processor_frequency;
410 struct ia64_pal_result res;
412 if (cpu_idle_hook != NULL)
415 res = ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
419 cpu_idle_wakeup(int cpu)
433 cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx)
435 struct pcb *oldpcb, *newpcb;
437 oldpcb = old->td_pcb;
439 ia32_savectx(oldpcb);
441 if (PCPU_GET(fpcurthread) == old)
442 old->td_frame->tf_special.psr |= IA64_PSR_DFH;
443 if (!savectx(oldpcb)) {
445 #if defined(SCHED_ULE) && defined(SMP)
446 /* td_lock is volatile */
447 while (new->td_lock == &blocked_lock)
450 newpcb = new->td_pcb;
451 oldpcb->pcb_current_pmap =
452 pmap_switch(newpcb->pcb_current_pmap);
453 PCPU_SET(curthread, new);
455 ia32_restorectx(newpcb);
457 if (PCPU_GET(fpcurthread) == new)
458 new->td_frame->tf_special.psr &= ~IA64_PSR_DFH;
460 /* We should not get here. */
461 panic("cpu_switch: restorectx() returned");
467 cpu_throw(struct thread *old __unused, struct thread *new)
471 newpcb = new->td_pcb;
472 (void)pmap_switch(newpcb->pcb_current_pmap);
473 PCPU_SET(curthread, new);
475 ia32_restorectx(newpcb);
478 /* We should not get here. */
479 panic("cpu_throw: restorectx() returned");
484 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
488 * Set pc_acpi_id to "uninitialized".
489 * See sys/dev/acpica/acpi_cpu.c
491 pcpu->pc_acpi_id = 0xffffffff;
500 if (td->td_md.md_spinlock_count == 0)
501 td->td_md.md_saved_intr = intr_disable();
502 td->td_md.md_spinlock_count++;
513 td->td_md.md_spinlock_count--;
514 if (td->td_md.md_spinlock_count == 0)
515 intr_restore(td->td_md.md_saved_intr);
519 map_vhpt(uintptr_t vhpt)
524 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
525 PTE_PL_KERN | PTE_AR_RW;
526 pte |= vhpt & PTE_PPN_MASK;
528 __asm __volatile("ptr.d %0,%1" :: "r"(vhpt),
529 "r"(IA64_ID_PAGE_SHIFT<<2));
531 __asm __volatile("mov %0=psr" : "=r"(psr));
532 __asm __volatile("rsm psr.ic|psr.i");
535 ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
537 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(2), "r"(pte));
538 __asm __volatile("mov psr.l=%0" :: "r" (psr));
548 if (ia64_pal_base == 0)
551 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
552 PTE_PL_KERN | PTE_AR_RWX;
553 pte |= ia64_pal_base & PTE_PPN_MASK;
555 __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
556 "r"(IA64_PHYS_TO_RR7(ia64_pal_base)), "r"(IA64_ID_PAGE_SHIFT<<2));
558 __asm __volatile("mov %0=psr" : "=r"(psr));
559 __asm __volatile("rsm psr.ic|psr.i");
561 ia64_set_ifa(IA64_PHYS_TO_RR7(ia64_pal_base));
562 ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
564 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(1), "r"(pte));
566 __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte));
567 __asm __volatile("mov psr.l=%0" :: "r" (psr));
572 map_gateway_page(void)
577 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
578 PTE_PL_KERN | PTE_AR_X_RX;
579 pte |= (uint64_t)ia64_gateway_page & PTE_PPN_MASK;
581 __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
582 "r"(VM_MAX_ADDRESS), "r"(PAGE_SHIFT << 2));
584 __asm __volatile("mov %0=psr" : "=r"(psr));
585 __asm __volatile("rsm psr.ic|psr.i");
587 ia64_set_ifa(VM_MAX_ADDRESS);
588 ia64_set_itir(PAGE_SHIFT << 2);
590 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte));
592 __asm __volatile("itr.i itr[%0]=%1" :: "r"(3), "r"(pte));
593 __asm __volatile("mov psr.l=%0" :: "r" (psr));
596 /* Expose the mapping to userland in ar.k5 */
597 ia64_set_k5(VM_MAX_ADDRESS);
601 calculate_frequencies(void)
603 struct ia64_sal_result sal;
604 struct ia64_pal_result pal;
606 sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
607 pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
609 if (sal.sal_status == 0 && pal.pal_status == 0) {
611 printf("Platform clock frequency %ld Hz\n",
613 printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
614 "ITC ratio %ld/%ld\n",
615 pal.pal_result[0] >> 32,
616 pal.pal_result[0] & ((1L << 32) - 1),
617 pal.pal_result[1] >> 32,
618 pal.pal_result[1] & ((1L << 32) - 1),
619 pal.pal_result[2] >> 32,
620 pal.pal_result[2] & ((1L << 32) - 1));
622 processor_frequency =
623 sal.sal_result[0] * (pal.pal_result[0] >> 32)
624 / (pal.pal_result[0] & ((1L << 32) - 1));
626 sal.sal_result[0] * (pal.pal_result[1] >> 32)
627 / (pal.pal_result[1] & ((1L << 32) - 1));
629 sal.sal_result[0] * (pal.pal_result[2] >> 32)
630 / (pal.pal_result[2] & ((1L << 32) - 1));
634 struct ia64_init_return
637 struct ia64_init_return ret;
639 vm_offset_t kernstart, kernend;
640 vm_offset_t kernstartpfn, kernendpfn, pfn0, pfn1;
643 int metadata_missing;
645 /* NO OUTPUT ALLOWED UNTIL FURTHER NOTICE */
648 * TODO: Disable interrupts, floating point etc.
649 * Maybe flush cache and tlb
651 ia64_set_fpsr(IA64_FPSR_DEFAULT);
654 * TODO: Get critical system information (if possible, from the
655 * information provided by the boot program).
659 * pa_bootinfo is the physical address of the bootinfo block as
660 * passed to us by the loader and set in locore.s.
662 bootinfo = *(struct bootinfo *)(IA64_PHYS_TO_RR7(pa_bootinfo));
664 if (bootinfo.bi_magic != BOOTINFO_MAGIC || bootinfo.bi_version != 1) {
665 bzero(&bootinfo, sizeof(bootinfo));
666 bootinfo.bi_kernend = (vm_offset_t) round_page(_end);
670 * Look for the I/O ports first - we need them for console
673 for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
674 switch (md->md_type) {
675 case EFI_MD_TYPE_IOPORT:
676 ia64_port_base = IA64_PHYS_TO_RR6(md->md_phys);
678 case EFI_MD_TYPE_PALCODE:
679 ia64_pal_base = md->md_phys;
684 metadata_missing = 0;
685 if (bootinfo.bi_modulep)
686 preload_metadata = (caddr_t)bootinfo.bi_modulep;
688 metadata_missing = 1;
690 if (envmode == 0 && bootinfo.bi_envp)
691 kern_envp = (caddr_t)bootinfo.bi_envp;
693 kern_envp = static_env;
696 * Look at arguments passed to us and compute boothowto.
698 boothowto = bootinfo.bi_boothowto;
701 * Catch case of boot_verbose set in environment.
703 if ((p = getenv("boot_verbose")) != NULL) {
704 if (strcmp(p, "yes") == 0 || strcmp(p, "YES") == 0) {
705 boothowto |= RB_VERBOSE;
710 if (boothowto & RB_VERBOSE)
714 * Find the beginning and end of the kernel.
716 kernstart = trunc_page(kernel_text);
718 ksym_start = bootinfo.bi_symtab;
719 ksym_end = bootinfo.bi_esymtab;
720 kernend = (vm_offset_t)round_page(ksym_end);
722 kernend = (vm_offset_t)round_page(_end);
724 /* But if the bootstrap tells us otherwise, believe it! */
725 if (bootinfo.bi_kernend)
726 kernend = round_page(bootinfo.bi_kernend);
729 * Setup the PCPU data for the bootstrap processor. It is needed
730 * by printf(). Also, since printf() has critical sections, we
731 * need to initialize at least pc_curthread.
734 ia64_set_k4((u_int64_t)pcpup);
735 pcpu_init(pcpup, 0, sizeof(pcpu0));
736 dpcpu_init((void *)kernend, 0);
737 kernend += DPCPU_SIZE;
738 PCPU_SET(curthread, &thread0);
741 * Initialize the console before we print anything out.
745 /* OUTPUT NOW ALLOWED */
747 if (ia64_pal_base != 0) {
748 ia64_pal_base &= ~IA64_ID_PAGE_MASK;
750 * We use a TR to map the first 256M of memory - this might
751 * cover the palcode too.
753 if (ia64_pal_base == 0)
754 printf("PAL code mapped by the kernel's TR\n");
756 printf("PAL code not found\n");
759 * Wire things up so we can call the firmware.
762 efi_boot_minimal(bootinfo.bi_systab);
764 calculate_frequencies();
766 if (metadata_missing)
767 printf("WARNING: loader(8) metadata is missing!\n");
769 /* Get FPSWA interface */
770 fpswa_iface = (bootinfo.bi_fpswa == 0) ? NULL :
771 (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo.bi_fpswa);
773 /* Init basic tunables, including hz */
776 p = getenv("kernelname");
778 strncpy(kernelname, p, sizeof(kernelname) - 1);
782 kernstartpfn = atop(IA64_RR_MASK(kernstart));
783 kernendpfn = atop(IA64_RR_MASK(kernend));
786 * Size the memory regions and load phys_avail[] with the results.
790 * Find out how much memory is available, by looking at
791 * the memory descriptors.
795 printf("Memory descriptor count: %d\n", mdcount);
799 for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
801 printf("MD %p: type %d pa 0x%lx cnt 0x%lx\n", md,
802 md->md_type, md->md_phys, md->md_pages);
805 pfn0 = ia64_btop(round_page(md->md_phys));
806 pfn1 = ia64_btop(trunc_page(md->md_phys + md->md_pages * 4096));
810 if (md->md_type != EFI_MD_TYPE_FREE)
814 * We have a memory descriptor that describes conventional
815 * memory that is for general use. We must determine if the
816 * loader has put the kernel in this region.
818 physmem += (pfn1 - pfn0);
819 if (pfn0 <= kernendpfn && kernstartpfn <= pfn1) {
821 * Must compute the location of the kernel
822 * within the segment.
825 printf("Descriptor %p contains kernel\n", mp);
827 if (pfn0 < kernstartpfn) {
829 * There is a chunk before the kernel.
832 printf("Loading chunk before kernel: "
833 "0x%lx / 0x%lx\n", pfn0, kernstartpfn);
835 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
836 phys_avail[phys_avail_cnt+1] = ia64_ptob(kernstartpfn);
839 if (kernendpfn < pfn1) {
841 * There is a chunk after the kernel.
844 printf("Loading chunk after kernel: "
845 "0x%lx / 0x%lx\n", kernendpfn, pfn1);
847 phys_avail[phys_avail_cnt] = ia64_ptob(kernendpfn);
848 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
853 * Just load this cluster as one chunk.
856 printf("Loading descriptor %d: 0x%lx / 0x%lx\n", i,
859 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
860 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
865 phys_avail[phys_avail_cnt] = 0;
868 init_param2(physmem);
871 * Initialize error message buffer (at end of core).
873 msgbufp = (struct msgbuf *)pmap_steal_memory(MSGBUF_SIZE);
874 msgbufinit(msgbufp, MSGBUF_SIZE);
876 proc_linkup0(&proc0, &thread0);
878 * Init mapping for kernel stack for proc 0
880 thread0.td_kstack = pmap_steal_memory(KSTACK_PAGES * PAGE_SIZE);
881 thread0.td_kstack_pages = KSTACK_PAGES;
886 * Initialize the rest of proc 0's PCB.
888 * Set the kernel sp, reserving space for an (empty) trapframe,
889 * and make proc0's trapframe pointer point to it for sanity.
890 * Initialise proc0's backing store to start after u area.
892 cpu_thread_alloc(&thread0);
893 thread0.td_frame->tf_flags = FRAME_SYSCALL;
894 thread0.td_pcb->pcb_special.sp =
895 (u_int64_t)thread0.td_frame - 16;
896 thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack;
899 * Initialize the virtual memory system.
904 * Initialize debuggers, and break into them if appropriate.
909 if (boothowto & RB_KDB)
910 kdb_enter(KDB_WHY_BOOTFLAGS,
911 "Boot flags requested debugger\n");
917 ret.bspstore = thread0.td_pcb->pcb_special.bspstore;
918 ret.sp = thread0.td_pcb->pcb_special.sp;
923 ia64_ioport_address(u_int port)
927 addr = (port > 0xffff) ? IA64_PHYS_TO_RR6((uint64_t)port) :
928 ia64_port_base | ((port & 0xfffc) << 10) | (port & 0xFFF);
929 return ((void *)addr);
936 return (bootinfo.bi_hcdp);
940 bzero(void *buf, size_t len)
944 while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
948 while (len >= sizeof(u_long) * 8) {
950 *((u_long*) p + 1) = 0;
951 *((u_long*) p + 2) = 0;
952 *((u_long*) p + 3) = 0;
953 len -= sizeof(u_long) * 8;
954 *((u_long*) p + 4) = 0;
955 *((u_long*) p + 5) = 0;
956 *((u_long*) p + 6) = 0;
957 *((u_long*) p + 7) = 0;
958 p += sizeof(u_long) * 8;
960 while (len >= sizeof(u_long)) {
962 len -= sizeof(u_long);
974 u_int64_t start, end, now;
978 start = ia64_get_itc();
979 end = start + (itc_frequency * n) / 1000000;
980 /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
982 now = ia64_get_itc();
983 } while (now < end || (now > start && end < start));
989 * Send an interrupt (signal) to a process.
992 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
996 struct trapframe *tf;
998 struct sigframe sf, *sfp;
1006 PROC_LOCK_ASSERT(p, MA_OWNED);
1007 sig = ksi->ksi_signo;
1008 code = ksi->ksi_code;
1010 mtx_assert(&psp->ps_mtx, MA_OWNED);
1012 sp = tf->tf_special.sp;
1013 oonstack = sigonstack(sp);
1016 /* save user context */
1017 bzero(&sf, sizeof(struct sigframe));
1018 sf.sf_uc.uc_sigmask = *mask;
1019 sf.sf_uc.uc_stack = td->td_sigstk;
1020 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
1021 ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1024 * Allocate and validate space for the signal handler
1025 * context. Note that if the stack is in P0 space, the
1026 * call to grow() is a nop, and the useracc() check
1027 * will fail if the process has not already allocated
1028 * the space with a `brk'.
1030 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
1031 SIGISMEMBER(psp->ps_sigonstack, sig)) {
1032 sbs = (u_int64_t)td->td_sigstk.ss_sp;
1033 sbs = (sbs + 15) & ~15;
1034 sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size);
1035 #if defined(COMPAT_43)
1036 td->td_sigstk.ss_flags |= SS_ONSTACK;
1039 sfp = (struct sigframe *)sp;
1040 sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15);
1042 /* Fill in the siginfo structure for POSIX handlers. */
1043 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
1044 sf.sf_si = ksi->ksi_info;
1045 sf.sf_si.si_signo = sig;
1047 * XXX this shouldn't be here after code in trap.c
1050 sf.sf_si.si_addr = (void*)tf->tf_special.ifa;
1051 code = (u_int64_t)&sfp->sf_si;
1054 mtx_unlock(&psp->ps_mtx);
1057 get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
1059 /* Copy the frame out to userland. */
1060 if (copyout(&sf, sfp, sizeof(sf)) != 0) {
1062 * Process has trashed its stack; give it an illegal
1063 * instruction to halt it in its tracks.
1066 sigexit(td, SIGILL);
1070 if ((tf->tf_flags & FRAME_SYSCALL) == 0) {
1071 tf->tf_special.psr &= ~IA64_PSR_RI;
1072 tf->tf_special.iip = ia64_get_k5() +
1073 ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
1075 tf->tf_special.iip = ia64_get_k5() +
1076 ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
1079 * Setup the trapframe to return to the signal trampoline. We pass
1080 * information to the trampoline in the following registers:
1082 * gp new backing store or NULL
1084 * r9 signal code or siginfo pointer
1085 * r10 signal handler (function descriptor)
1087 tf->tf_special.sp = (u_int64_t)sfp - 16;
1088 tf->tf_special.gp = sbs;
1089 tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore;
1090 tf->tf_special.ndirty = 0;
1091 tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat;
1092 tf->tf_scratch.gr8 = sig;
1093 tf->tf_scratch.gr9 = code;
1094 tf->tf_scratch.gr10 = (u_int64_t)catcher;
1097 mtx_lock(&psp->ps_mtx);
1101 * System call to cleanup state after a signal
1102 * has been taken. Reset signal mask and
1103 * stack state from context left by sendsig (above).
1104 * Return to previous pc and psl as specified by
1105 * context left by sendsig. Check carefully to
1106 * make sure that the user has not modified the
1107 * state to gain improper privileges.
1112 sigreturn(struct thread *td,
1113 struct sigreturn_args /* {
1114 ucontext_t *sigcntxp;
1118 struct trapframe *tf;
1125 * Fetch the entire context structure at once for speed.
1126 * We don't use a normal argument to simplify RSE handling.
1128 if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc)))
1131 set_mcontext(td, &uc.uc_mcontext);
1133 #if defined(COMPAT_43)
1134 if (sigonstack(tf->tf_special.sp))
1135 td->td_sigstk.ss_flags |= SS_ONSTACK;
1137 td->td_sigstk.ss_flags &= ~SS_ONSTACK;
1139 kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
1141 return (EJUSTRETURN);
1144 #ifdef COMPAT_FREEBSD4
1146 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
1149 return sigreturn(td, (struct sigreturn_args *)uap);
1154 * Construct a PCB from a trapframe. This is called from kdb_trap() where
1155 * we want to start a backtrace from the function that caused us to enter
1156 * the debugger. We have the context in the trapframe, but base the trace
1157 * on the PCB. The PCB doesn't have to be perfect, as long as it contains
1158 * enough for a backtrace.
1161 makectx(struct trapframe *tf, struct pcb *pcb)
1164 pcb->pcb_special = tf->tf_special;
1165 pcb->pcb_special.__spare = ~0UL; /* XXX see unwind.c */
1166 save_callee_saved(&pcb->pcb_preserved);
1167 save_callee_saved_fp(&pcb->pcb_preserved_fp);
1171 ia64_flush_dirty(struct thread *td, struct _special *r)
1175 uint64_t bspst, kstk, rnat;
1181 kstk = td->td_kstack + (r->bspstore & 0x1ffUL);
1182 if (td == curthread) {
1183 __asm __volatile("mov ar.rsc=0;;");
1184 __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst));
1185 /* Make sure we have all the user registers written out. */
1186 if (bspst - kstk < r->ndirty) {
1187 __asm __volatile("flushrs;;");
1188 __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst));
1190 __asm __volatile("mov %0=ar.rnat;;" : "=r"(rnat));
1191 __asm __volatile("mov ar.rsc=3");
1192 error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty);
1194 r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL))
1195 ? *(uint64_t*)(kstk | 0x1f8L) : rnat;
1197 locked = PROC_LOCKED(td->td_proc);
1200 iov.iov_base = (void*)(uintptr_t)kstk;
1201 iov.iov_len = r->ndirty;
1204 uio.uio_offset = r->bspstore;
1205 uio.uio_resid = r->ndirty;
1206 uio.uio_segflg = UIO_SYSSPACE;
1207 uio.uio_rw = UIO_WRITE;
1209 error = proc_rwmem(td->td_proc, &uio);
1211 * XXX proc_rwmem() doesn't currently return ENOSPC,
1212 * so I think it can bogusly return 0. Neither do
1213 * we allow short writes.
1215 if (uio.uio_resid != 0 && error == 0)
1221 r->bspstore += r->ndirty;
1227 get_mcontext(struct thread *td, mcontext_t *mc, int flags)
1229 struct trapframe *tf;
1233 bzero(mc, sizeof(*mc));
1234 mc->mc_special = tf->tf_special;
1235 error = ia64_flush_dirty(td, &mc->mc_special);
1236 if (tf->tf_flags & FRAME_SYSCALL) {
1237 mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT;
1238 mc->mc_scratch = tf->tf_scratch;
1239 if (flags & GET_MC_CLEAR_RET) {
1240 mc->mc_scratch.gr8 = 0;
1241 mc->mc_scratch.gr9 = 0;
1242 mc->mc_scratch.gr10 = 0;
1243 mc->mc_scratch.gr11 = 0;
1246 mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT;
1247 mc->mc_scratch = tf->tf_scratch;
1248 mc->mc_scratch_fp = tf->tf_scratch_fp;
1250 * XXX If the thread never used the high FP registers, we
1251 * probably shouldn't waste time saving them.
1253 ia64_highfp_save(td);
1254 mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID;
1255 mc->mc_high_fp = td->td_pcb->pcb_high_fp;
1257 save_callee_saved(&mc->mc_preserved);
1258 save_callee_saved_fp(&mc->mc_preserved_fp);
1263 set_mcontext(struct thread *td, const mcontext_t *mc)
1266 struct trapframe *tf;
1271 KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1272 ("Whoa there! We have more than 8KB of dirty registers!"));
1276 * Only copy the user mask and the restart instruction bit from
1279 psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL |
1280 IA64_PSR_MFH | IA64_PSR_RI;
1281 s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask);
1282 /* We don't have any dirty registers of the new context. */
1284 if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) {
1286 * We can get an async context passed to us while we
1287 * entered the kernel through a syscall: sigreturn(2)
1288 * takes contexts that could previously be the result of
1289 * a trap or interrupt.
1290 * Hence, we cannot assert that the trapframe is not
1291 * a syscall frame, but we can assert that it's at
1292 * least an expected syscall.
1294 if (tf->tf_flags & FRAME_SYSCALL) {
1295 KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo"));
1296 tf->tf_flags &= ~FRAME_SYSCALL;
1298 tf->tf_scratch = mc->mc_scratch;
1299 tf->tf_scratch_fp = mc->mc_scratch_fp;
1300 if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID)
1301 td->td_pcb->pcb_high_fp = mc->mc_high_fp;
1303 KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
1304 if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) {
1305 s.cfm = tf->tf_special.cfm;
1306 s.iip = tf->tf_special.iip;
1307 tf->tf_scratch.gr15 = 0; /* Clear syscall nr. */
1309 tf->tf_scratch = mc->mc_scratch;
1312 restore_callee_saved(&mc->mc_preserved);
1313 restore_callee_saved_fp(&mc->mc_preserved_fp);
1319 * Clear registers on exec.
1322 exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
1324 struct trapframe *tf;
1325 uint64_t *ksttop, *kst;
1328 ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty +
1329 (tf->tf_special.bspstore & 0x1ffUL));
1332 * We can ignore up to 8KB of dirty registers by masking off the
1333 * lower 13 bits in exception_restore() or epc_syscall(). This
1334 * should be enough for a couple of years, but if there are more
1335 * than 8KB of dirty registers, we lose track of the bottom of
1336 * the kernel stack. The solution is to copy the active part of
1337 * the kernel stack down 1 page (or 2, but not more than that)
1338 * so that we always have less than 8KB of dirty registers.
1340 KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1341 ("Whoa there! We have more than 8KB of dirty registers!"));
1343 bzero(&tf->tf_special, sizeof(tf->tf_special));
1344 if ((tf->tf_flags & FRAME_SYSCALL) == 0) { /* break syscalls. */
1345 bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
1346 bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
1347 tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
1348 tf->tf_special.bspstore = IA64_BACKINGSTORE;
1350 * Copy the arguments onto the kernel register stack so that
1351 * they get loaded by the loadrs instruction. Skip over the
1352 * NaT collection points.
1355 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1358 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1360 *kst-- = ps_strings;
1361 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1364 tf->tf_special.ndirty = (ksttop - kst) << 3;
1365 } else { /* epc syscalls (default). */
1366 tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
1367 tf->tf_special.bspstore = IA64_BACKINGSTORE + 24;
1369 * Write values for out0, out1 and out2 to the user's backing
1370 * store and arrange for them to be restored into the user's
1371 * initial register frame.
1372 * Assumes that (bspstore & 0x1f8) < 0x1e0.
1374 suword((caddr_t)tf->tf_special.bspstore - 24, stack);
1375 suword((caddr_t)tf->tf_special.bspstore - 16, ps_strings);
1376 suword((caddr_t)tf->tf_special.bspstore - 8, 0);
1379 tf->tf_special.iip = entry;
1380 tf->tf_special.sp = (stack & ~15) - 16;
1381 tf->tf_special.rsc = 0xf;
1382 tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
1383 tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT |
1384 IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN |
1389 ptrace_set_pc(struct thread *td, unsigned long addr)
1393 switch (addr & 0xFUL) {
1395 slot = IA64_PSR_RI_0;
1398 /* XXX we need to deal with MLX bundles here */
1399 slot = IA64_PSR_RI_1;
1402 slot = IA64_PSR_RI_2;
1408 td->td_frame->tf_special.iip = addr & ~0x0FULL;
1409 td->td_frame->tf_special.psr =
1410 (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot;
1415 ptrace_single_step(struct thread *td)
1417 struct trapframe *tf;
1420 * There's no way to set single stepping when we're leaving the
1421 * kernel through the EPC syscall path. The way we solve this is
1422 * by enabling the lower-privilege trap so that we re-enter the
1423 * kernel as soon as the privilege level changes. See trap.c for
1424 * how we proceed from there.
1427 if (tf->tf_flags & FRAME_SYSCALL)
1428 tf->tf_special.psr |= IA64_PSR_LP;
1430 tf->tf_special.psr |= IA64_PSR_SS;
1435 ptrace_clear_single_step(struct thread *td)
1437 struct trapframe *tf;
1440 * Clear any and all status bits we may use to implement single
1444 tf->tf_special.psr &= ~IA64_PSR_SS;
1445 tf->tf_special.psr &= ~IA64_PSR_LP;
1446 tf->tf_special.psr &= ~IA64_PSR_TB;
1451 fill_regs(struct thread *td, struct reg *regs)
1453 struct trapframe *tf;
1456 regs->r_special = tf->tf_special;
1457 regs->r_scratch = tf->tf_scratch;
1458 save_callee_saved(®s->r_preserved);
1463 set_regs(struct thread *td, struct reg *regs)
1465 struct trapframe *tf;
1469 error = ia64_flush_dirty(td, &tf->tf_special);
1471 tf->tf_special = regs->r_special;
1472 tf->tf_special.bspstore += tf->tf_special.ndirty;
1473 tf->tf_special.ndirty = 0;
1474 tf->tf_scratch = regs->r_scratch;
1475 restore_callee_saved(®s->r_preserved);
1481 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1488 set_dbregs(struct thread *td, struct dbreg *dbregs)
1495 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1497 struct trapframe *frame = td->td_frame;
1498 struct pcb *pcb = td->td_pcb;
1500 /* Save the high FP registers. */
1501 ia64_highfp_save(td);
1503 fpregs->fpr_scratch = frame->tf_scratch_fp;
1504 save_callee_saved_fp(&fpregs->fpr_preserved);
1505 fpregs->fpr_high = pcb->pcb_high_fp;
1510 set_fpregs(struct thread *td, struct fpreg *fpregs)
1512 struct trapframe *frame = td->td_frame;
1513 struct pcb *pcb = td->td_pcb;
1515 /* Throw away the high FP registers (should be redundant). */
1516 ia64_highfp_drop(td);
1518 frame->tf_scratch_fp = fpregs->fpr_scratch;
1519 restore_callee_saved_fp(&fpregs->fpr_preserved);
1520 pcb->pcb_high_fp = fpregs->fpr_high;
1525 ia64_sync_icache(vm_offset_t va, vm_offset_t sz)
1529 if (!ia64_sync_icache_needed)