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 SYSCTL_NODE(_hw, OID_AUTO, freq, CTLFLAG_RD, 0, "");
105 SYSCTL_NODE(_machdep, OID_AUTO, cpu, CTLFLAG_RD, 0, "");
107 static u_int bus_freq;
108 SYSCTL_UINT(_hw_freq, OID_AUTO, bus, CTLFLAG_RD, &bus_freq, 0,
109 "Bus clock frequency");
111 static u_int cpu_freq;
112 SYSCTL_UINT(_hw_freq, OID_AUTO, cpu, CTLFLAG_RD, &cpu_freq, 0,
113 "CPU clock frequency");
115 static u_int itc_freq;
116 SYSCTL_UINT(_hw_freq, OID_AUTO, itc, CTLFLAG_RD, &itc_freq, 0,
121 u_int64_t pa_bootinfo;
122 struct bootinfo bootinfo;
126 extern u_int64_t kernel_text[], _end[];
128 extern u_int64_t ia64_gateway_page[];
129 extern u_int64_t break_sigtramp[];
130 extern u_int64_t epc_sigtramp[];
132 struct fpswa_iface *fpswa_iface;
134 u_int64_t ia64_pal_base;
135 u_int64_t ia64_port_base;
137 static int ia64_sync_icache_needed;
139 char machine[] = MACHINE;
140 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
142 static char cpu_model[64];
143 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0,
144 "The CPU model name");
146 static char cpu_family[64];
147 SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0,
148 "The CPU family name");
151 extern vm_offset_t ksym_start, ksym_end;
155 struct msgbuf *msgbufp = NULL;
157 /* Other subsystems (e.g., ACPI) can hook this later. */
158 void (*cpu_idle_hook)(void) = NULL;
163 #define PHYSMAP_SIZE (2 * VM_PHYSSEG_MAX)
165 vm_paddr_t phys_avail[PHYSMAP_SIZE + 2];
167 /* must be 2 less so 0 0 can signal end of chunks */
168 #define PHYS_AVAIL_ARRAY_END ((sizeof(phys_avail) / sizeof(vm_offset_t)) - 2)
170 struct kva_md_info kmi;
173 #define Ghz (1000L*Mhz)
179 char *family_name, *model_name;
180 u_int64_t features, tmp;
181 int number, revision, model, family, archrev;
184 * Assumes little-endian.
186 *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
187 *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
190 tmp = ia64_get_cpuid(3);
191 number = (tmp >> 0) & 0xff;
192 revision = (tmp >> 8) & 0xff;
193 model = (tmp >> 16) & 0xff;
194 family = (tmp >> 24) & 0xff;
195 archrev = (tmp >> 32) & 0xff;
197 family_name = model_name = "unknown";
200 family_name = "Itanium";
201 model_name = "Merced";
204 family_name = "Itanium 2";
207 model_name = "McKinley";
211 * Deerfield is a low-voltage variant based on the
212 * Madison core. We need circumstantial evidence
213 * (i.e. the clock frequency) to identify those.
214 * Allow for roughly 1% error margin.
216 if (cpu_freq > 990 && cpu_freq < 1010)
217 model_name = "Deerfield";
219 model_name = "Madison";
222 model_name = "Madison II";
227 ia64_sync_icache_needed = 1;
229 family_name = "Itanium 2";
232 model_name = "Montecito";
237 snprintf(cpu_family, sizeof(cpu_family), "%s", family_name);
238 snprintf(cpu_model, sizeof(cpu_model), "%s", model_name);
240 features = ia64_get_cpuid(4);
242 printf("CPU: %s (", model_name);
244 printf("%u Mhz ", cpu_freq);
245 printf("%s)\n", family_name);
246 printf(" Origin = \"%s\" Revision = %d\n", vendor, revision);
247 printf(" Features = 0x%b\n", (u_int32_t) features,
249 "\001LB" /* long branch (brl) instruction. */
250 "\002SD" /* Spontaneous deferral. */
251 "\003AO" /* 16-byte atomic operations (ld, st, cmpxchg). */ );
255 cpu_startup(void *dummy)
259 struct pcpu_stats *pcs;
262 * Good {morning,afternoon,evening,night}.
269 printf("real memory = %ld (%ld MB)\n", ia64_ptob(Maxmem),
270 ia64_ptob(Maxmem) / 1048576);
274 * Display any holes after the first chunk of extended memory.
279 printf("Physical memory chunk(s):\n");
280 for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
281 long size1 = phys_avail[indx + 1] - phys_avail[indx];
283 printf("0x%08lx - 0x%08lx, %ld bytes (%ld pages)\n",
284 phys_avail[indx], phys_avail[indx + 1] - 1, size1,
285 size1 >> PAGE_SHIFT);
289 vm_ksubmap_init(&kmi);
291 printf("avail memory = %ld (%ld MB)\n", ptoa(cnt.v_free_count),
292 ptoa(cnt.v_free_count) / 1048576);
294 if (fpswa_iface == NULL)
295 printf("Warning: no FPSWA package supplied\n");
297 printf("FPSWA Revision = 0x%lx, Entry = %p\n",
298 (long)fpswa_iface->if_rev, (void *)fpswa_iface->if_fpswa);
301 * Set up buffers, so they can be used to read disk labels.
304 vm_pager_bufferinit();
307 * Traverse the MADT to discover IOSAPIC and Local SAPIC
314 * Create sysctl tree for per-CPU information.
316 SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
317 snprintf(nodename, sizeof(nodename), "%u", pc->pc_cpuid);
318 sysctl_ctx_init(&pc->pc_md.sysctl_ctx);
319 pc->pc_md.sysctl_tree = SYSCTL_ADD_NODE(&pc->pc_md.sysctl_ctx,
320 SYSCTL_STATIC_CHILDREN(_machdep_cpu), OID_AUTO, nodename,
321 CTLFLAG_RD, NULL, "");
322 if (pc->pc_md.sysctl_tree == NULL)
325 pcs = &pc->pc_md.stats;
327 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
328 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
329 "nasts", CTLFLAG_RD, &pcs->pcs_nasts,
330 "Number of IPI_AST interrupts");
332 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
333 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
334 "nclks", CTLFLAG_RD, &pcs->pcs_nclks,
335 "Number of clock interrupts");
337 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
338 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
339 "nextints", CTLFLAG_RD, &pcs->pcs_nextints,
340 "Number of ExtINT interrupts");
342 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
343 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
344 "nhighfps", CTLFLAG_RD, &pcs->pcs_nhighfps,
345 "Number of IPI_HIGH_FP interrupts");
347 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
348 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
349 "nhwints", CTLFLAG_RD, &pcs->pcs_nhwints,
350 "Number of hardware (device) interrupts");
352 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
353 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
354 "npreempts", CTLFLAG_RD, &pcs->pcs_npreempts,
355 "Number of IPI_PREEMPT interrupts");
357 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
358 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
359 "nrdvs", CTLFLAG_RD, &pcs->pcs_nrdvs,
360 "Number of IPI_RENDEZVOUS interrupts");
362 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
363 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
364 "nstops", CTLFLAG_RD, &pcs->pcs_nstops,
365 "Number of IPI_STOP interrupts");
367 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
368 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
369 "nstrays", CTLFLAG_RD, &pcs->pcs_nstrays,
370 "Number of stray vectors");
373 SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
383 cpu_flush_dcache(void *ptr, size_t len)
387 va = (uintptr_t)ptr & ~31;
388 lim = (uintptr_t)ptr + len;
397 /* Get current clock frequency for the given cpu id. */
399 cpu_est_clockrate(int cpu_id, uint64_t *rate)
402 if (pcpu_find(cpu_id) == NULL || rate == NULL)
404 *rate = (u_long)cpu_freq * 1000000ul;
418 struct ia64_pal_result res;
420 if (cpu_idle_hook != NULL)
423 res = ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
427 cpu_idle_wakeup(int cpu)
441 cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx)
443 struct pcb *oldpcb, *newpcb;
445 oldpcb = old->td_pcb;
447 ia32_savectx(oldpcb);
449 if (PCPU_GET(fpcurthread) == old)
450 old->td_frame->tf_special.psr |= IA64_PSR_DFH;
451 if (!savectx(oldpcb)) {
453 #if defined(SCHED_ULE) && defined(SMP)
454 /* td_lock is volatile */
455 while (new->td_lock == &blocked_lock)
458 newpcb = new->td_pcb;
459 oldpcb->pcb_current_pmap =
460 pmap_switch(newpcb->pcb_current_pmap);
461 PCPU_SET(curthread, new);
463 ia32_restorectx(newpcb);
465 if (PCPU_GET(fpcurthread) == new)
466 new->td_frame->tf_special.psr &= ~IA64_PSR_DFH;
468 /* We should not get here. */
469 panic("cpu_switch: restorectx() returned");
475 cpu_throw(struct thread *old __unused, struct thread *new)
479 newpcb = new->td_pcb;
480 (void)pmap_switch(newpcb->pcb_current_pmap);
481 PCPU_SET(curthread, new);
483 ia32_restorectx(newpcb);
486 /* We should not get here. */
487 panic("cpu_throw: restorectx() returned");
492 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
496 * Set pc_acpi_id to "uninitialized".
497 * See sys/dev/acpica/acpi_cpu.c
499 pcpu->pc_acpi_id = 0xffffffff;
508 if (td->td_md.md_spinlock_count == 0)
509 td->td_md.md_saved_intr = intr_disable();
510 td->td_md.md_spinlock_count++;
521 td->td_md.md_spinlock_count--;
522 if (td->td_md.md_spinlock_count == 0)
523 intr_restore(td->td_md.md_saved_intr);
527 map_vhpt(uintptr_t vhpt)
532 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
533 PTE_PL_KERN | PTE_AR_RW;
534 pte |= vhpt & PTE_PPN_MASK;
536 __asm __volatile("ptr.d %0,%1" :: "r"(vhpt),
537 "r"(IA64_ID_PAGE_SHIFT<<2));
539 __asm __volatile("mov %0=psr" : "=r"(psr));
540 __asm __volatile("rsm psr.ic|psr.i");
543 ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
545 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(2), "r"(pte));
546 __asm __volatile("mov psr.l=%0" :: "r" (psr));
556 if (ia64_pal_base == 0)
559 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
560 PTE_PL_KERN | PTE_AR_RWX;
561 pte |= ia64_pal_base & PTE_PPN_MASK;
563 __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
564 "r"(IA64_PHYS_TO_RR7(ia64_pal_base)), "r"(IA64_ID_PAGE_SHIFT<<2));
566 __asm __volatile("mov %0=psr" : "=r"(psr));
567 __asm __volatile("rsm psr.ic|psr.i");
569 ia64_set_ifa(IA64_PHYS_TO_RR7(ia64_pal_base));
570 ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
572 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(1), "r"(pte));
574 __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte));
575 __asm __volatile("mov psr.l=%0" :: "r" (psr));
580 map_gateway_page(void)
585 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
586 PTE_PL_KERN | PTE_AR_X_RX;
587 pte |= (uint64_t)ia64_gateway_page & PTE_PPN_MASK;
589 __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
590 "r"(VM_MAX_ADDRESS), "r"(PAGE_SHIFT << 2));
592 __asm __volatile("mov %0=psr" : "=r"(psr));
593 __asm __volatile("rsm psr.ic|psr.i");
595 ia64_set_ifa(VM_MAX_ADDRESS);
596 ia64_set_itir(PAGE_SHIFT << 2);
598 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte));
600 __asm __volatile("itr.i itr[%0]=%1" :: "r"(3), "r"(pte));
601 __asm __volatile("mov psr.l=%0" :: "r" (psr));
604 /* Expose the mapping to userland in ar.k5 */
605 ia64_set_k5(VM_MAX_ADDRESS);
609 freq_ratio(u_long base, u_long ratio)
613 f = (base * (ratio >> 32)) / (ratio & 0xfffffffful);
614 return ((f + 500000) / 1000000);
618 calculate_frequencies(void)
620 struct ia64_sal_result sal;
621 struct ia64_pal_result pal;
623 sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
624 pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
626 if (sal.sal_status == 0 && pal.pal_status == 0) {
628 printf("Platform clock frequency %ld Hz\n",
630 printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
631 "ITC ratio %ld/%ld\n",
632 pal.pal_result[0] >> 32,
633 pal.pal_result[0] & ((1L << 32) - 1),
634 pal.pal_result[1] >> 32,
635 pal.pal_result[1] & ((1L << 32) - 1),
636 pal.pal_result[2] >> 32,
637 pal.pal_result[2] & ((1L << 32) - 1));
639 cpu_freq = freq_ratio(sal.sal_result[0], pal.pal_result[0]);
640 bus_freq = freq_ratio(sal.sal_result[0], pal.pal_result[1]);
641 itc_freq = freq_ratio(sal.sal_result[0], pal.pal_result[2]);
645 struct ia64_init_return
648 struct ia64_init_return ret;
650 vm_offset_t kernstart, kernend;
651 vm_offset_t kernstartpfn, kernendpfn, pfn0, pfn1;
654 int metadata_missing;
656 /* NO OUTPUT ALLOWED UNTIL FURTHER NOTICE */
659 * TODO: Disable interrupts, floating point etc.
660 * Maybe flush cache and tlb
662 ia64_set_fpsr(IA64_FPSR_DEFAULT);
665 * TODO: Get critical system information (if possible, from the
666 * information provided by the boot program).
670 * pa_bootinfo is the physical address of the bootinfo block as
671 * passed to us by the loader and set in locore.s.
673 bootinfo = *(struct bootinfo *)(IA64_PHYS_TO_RR7(pa_bootinfo));
675 if (bootinfo.bi_magic != BOOTINFO_MAGIC || bootinfo.bi_version != 1) {
676 bzero(&bootinfo, sizeof(bootinfo));
677 bootinfo.bi_kernend = (vm_offset_t) round_page(_end);
681 * Look for the I/O ports first - we need them for console
684 for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
685 switch (md->md_type) {
686 case EFI_MD_TYPE_IOPORT:
687 ia64_port_base = IA64_PHYS_TO_RR6(md->md_phys);
689 case EFI_MD_TYPE_PALCODE:
690 ia64_pal_base = md->md_phys;
695 metadata_missing = 0;
696 if (bootinfo.bi_modulep)
697 preload_metadata = (caddr_t)bootinfo.bi_modulep;
699 metadata_missing = 1;
701 if (envmode == 0 && bootinfo.bi_envp)
702 kern_envp = (caddr_t)bootinfo.bi_envp;
704 kern_envp = static_env;
707 * Look at arguments passed to us and compute boothowto.
709 boothowto = bootinfo.bi_boothowto;
712 * Catch case of boot_verbose set in environment.
714 if ((p = getenv("boot_verbose")) != NULL) {
715 if (strcmp(p, "yes") == 0 || strcmp(p, "YES") == 0) {
716 boothowto |= RB_VERBOSE;
721 if (boothowto & RB_VERBOSE)
725 * Find the beginning and end of the kernel.
727 kernstart = trunc_page(kernel_text);
729 ksym_start = bootinfo.bi_symtab;
730 ksym_end = bootinfo.bi_esymtab;
731 kernend = (vm_offset_t)round_page(ksym_end);
733 kernend = (vm_offset_t)round_page(_end);
735 /* But if the bootstrap tells us otherwise, believe it! */
736 if (bootinfo.bi_kernend)
737 kernend = round_page(bootinfo.bi_kernend);
740 * Setup the PCPU data for the bootstrap processor. It is needed
741 * by printf(). Also, since printf() has critical sections, we
742 * need to initialize at least pc_curthread.
745 ia64_set_k4((u_int64_t)pcpup);
746 pcpu_init(pcpup, 0, sizeof(pcpu0));
747 dpcpu_init((void *)kernend, 0);
748 kernend += DPCPU_SIZE;
749 PCPU_SET(curthread, &thread0);
752 * Initialize the console before we print anything out.
756 /* OUTPUT NOW ALLOWED */
758 if (ia64_pal_base != 0) {
759 ia64_pal_base &= ~IA64_ID_PAGE_MASK;
761 * We use a TR to map the first 256M of memory - this might
762 * cover the palcode too.
764 if (ia64_pal_base == 0)
765 printf("PAL code mapped by the kernel's TR\n");
767 printf("PAL code not found\n");
770 * Wire things up so we can call the firmware.
773 efi_boot_minimal(bootinfo.bi_systab);
775 calculate_frequencies();
777 if (metadata_missing)
778 printf("WARNING: loader(8) metadata is missing!\n");
780 /* Get FPSWA interface */
781 fpswa_iface = (bootinfo.bi_fpswa == 0) ? NULL :
782 (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo.bi_fpswa);
784 /* Init basic tunables, including hz */
787 p = getenv("kernelname");
789 strncpy(kernelname, p, sizeof(kernelname) - 1);
793 kernstartpfn = atop(IA64_RR_MASK(kernstart));
794 kernendpfn = atop(IA64_RR_MASK(kernend));
797 * Size the memory regions and load phys_avail[] with the results.
801 * Find out how much memory is available, by looking at
802 * the memory descriptors.
806 printf("Memory descriptor count: %d\n", mdcount);
810 for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
812 printf("MD %p: type %d pa 0x%lx cnt 0x%lx\n", md,
813 md->md_type, md->md_phys, md->md_pages);
816 pfn0 = ia64_btop(round_page(md->md_phys));
817 pfn1 = ia64_btop(trunc_page(md->md_phys + md->md_pages * 4096));
821 if (md->md_type != EFI_MD_TYPE_FREE)
825 * We have a memory descriptor that describes conventional
826 * memory that is for general use. We must determine if the
827 * loader has put the kernel in this region.
829 physmem += (pfn1 - pfn0);
830 if (pfn0 <= kernendpfn && kernstartpfn <= pfn1) {
832 * Must compute the location of the kernel
833 * within the segment.
836 printf("Descriptor %p contains kernel\n", mp);
838 if (pfn0 < kernstartpfn) {
840 * There is a chunk before the kernel.
843 printf("Loading chunk before kernel: "
844 "0x%lx / 0x%lx\n", pfn0, kernstartpfn);
846 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
847 phys_avail[phys_avail_cnt+1] = ia64_ptob(kernstartpfn);
850 if (kernendpfn < pfn1) {
852 * There is a chunk after the kernel.
855 printf("Loading chunk after kernel: "
856 "0x%lx / 0x%lx\n", kernendpfn, pfn1);
858 phys_avail[phys_avail_cnt] = ia64_ptob(kernendpfn);
859 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
864 * Just load this cluster as one chunk.
867 printf("Loading descriptor %d: 0x%lx / 0x%lx\n", i,
870 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
871 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
876 phys_avail[phys_avail_cnt] = 0;
879 init_param2(physmem);
882 * Initialize error message buffer (at end of core).
884 msgbufp = (struct msgbuf *)pmap_steal_memory(MSGBUF_SIZE);
885 msgbufinit(msgbufp, MSGBUF_SIZE);
887 proc_linkup0(&proc0, &thread0);
889 * Init mapping for kernel stack for proc 0
891 thread0.td_kstack = pmap_steal_memory(KSTACK_PAGES * PAGE_SIZE);
892 thread0.td_kstack_pages = KSTACK_PAGES;
897 * Initialize the rest of proc 0's PCB.
899 * Set the kernel sp, reserving space for an (empty) trapframe,
900 * and make proc0's trapframe pointer point to it for sanity.
901 * Initialise proc0's backing store to start after u area.
903 cpu_thread_alloc(&thread0);
904 thread0.td_frame->tf_flags = FRAME_SYSCALL;
905 thread0.td_pcb->pcb_special.sp =
906 (u_int64_t)thread0.td_frame - 16;
907 thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack;
910 * Initialize the virtual memory system.
915 * Initialize debuggers, and break into them if appropriate.
920 if (boothowto & RB_KDB)
921 kdb_enter(KDB_WHY_BOOTFLAGS,
922 "Boot flags requested debugger\n");
928 ret.bspstore = thread0.td_pcb->pcb_special.bspstore;
929 ret.sp = thread0.td_pcb->pcb_special.sp;
934 ia64_ioport_address(u_int port)
938 addr = (port > 0xffff) ? IA64_PHYS_TO_RR6((uint64_t)port) :
939 ia64_port_base | ((port & 0xfffc) << 10) | (port & 0xFFF);
940 return ((void *)addr);
947 return (bootinfo.bi_hcdp);
951 bzero(void *buf, size_t len)
955 while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
959 while (len >= sizeof(u_long) * 8) {
961 *((u_long*) p + 1) = 0;
962 *((u_long*) p + 2) = 0;
963 *((u_long*) p + 3) = 0;
964 len -= sizeof(u_long) * 8;
965 *((u_long*) p + 4) = 0;
966 *((u_long*) p + 5) = 0;
967 *((u_long*) p + 6) = 0;
968 *((u_long*) p + 7) = 0;
969 p += sizeof(u_long) * 8;
971 while (len >= sizeof(u_long)) {
973 len -= sizeof(u_long);
992 u_int64_t start, end, now;
996 start = ia64_get_itc();
997 end = start + itc_freq * n;
998 /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
1000 now = ia64_get_itc();
1001 } while (now < end || (now > start && end < start));
1007 * Send an interrupt (signal) to a process.
1010 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
1014 struct trapframe *tf;
1015 struct sigacts *psp;
1016 struct sigframe sf, *sfp;
1024 PROC_LOCK_ASSERT(p, MA_OWNED);
1025 sig = ksi->ksi_signo;
1026 code = ksi->ksi_code;
1028 mtx_assert(&psp->ps_mtx, MA_OWNED);
1030 sp = tf->tf_special.sp;
1031 oonstack = sigonstack(sp);
1034 /* save user context */
1035 bzero(&sf, sizeof(struct sigframe));
1036 sf.sf_uc.uc_sigmask = *mask;
1037 sf.sf_uc.uc_stack = td->td_sigstk;
1038 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
1039 ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1042 * Allocate and validate space for the signal handler
1043 * context. Note that if the stack is in P0 space, the
1044 * call to grow() is a nop, and the useracc() check
1045 * will fail if the process has not already allocated
1046 * the space with a `brk'.
1048 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
1049 SIGISMEMBER(psp->ps_sigonstack, sig)) {
1050 sbs = (u_int64_t)td->td_sigstk.ss_sp;
1051 sbs = (sbs + 15) & ~15;
1052 sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size);
1053 #if defined(COMPAT_43)
1054 td->td_sigstk.ss_flags |= SS_ONSTACK;
1057 sfp = (struct sigframe *)sp;
1058 sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15);
1060 /* Fill in the siginfo structure for POSIX handlers. */
1061 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
1062 sf.sf_si = ksi->ksi_info;
1063 sf.sf_si.si_signo = sig;
1065 * XXX this shouldn't be here after code in trap.c
1068 sf.sf_si.si_addr = (void*)tf->tf_special.ifa;
1069 code = (u_int64_t)&sfp->sf_si;
1072 mtx_unlock(&psp->ps_mtx);
1075 get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
1077 /* Copy the frame out to userland. */
1078 if (copyout(&sf, sfp, sizeof(sf)) != 0) {
1080 * Process has trashed its stack; give it an illegal
1081 * instruction to halt it in its tracks.
1084 sigexit(td, SIGILL);
1088 if ((tf->tf_flags & FRAME_SYSCALL) == 0) {
1089 tf->tf_special.psr &= ~IA64_PSR_RI;
1090 tf->tf_special.iip = ia64_get_k5() +
1091 ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
1093 tf->tf_special.iip = ia64_get_k5() +
1094 ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
1097 * Setup the trapframe to return to the signal trampoline. We pass
1098 * information to the trampoline in the following registers:
1100 * gp new backing store or NULL
1102 * r9 signal code or siginfo pointer
1103 * r10 signal handler (function descriptor)
1105 tf->tf_special.sp = (u_int64_t)sfp - 16;
1106 tf->tf_special.gp = sbs;
1107 tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore;
1108 tf->tf_special.ndirty = 0;
1109 tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat;
1110 tf->tf_scratch.gr8 = sig;
1111 tf->tf_scratch.gr9 = code;
1112 tf->tf_scratch.gr10 = (u_int64_t)catcher;
1115 mtx_lock(&psp->ps_mtx);
1119 * System call to cleanup state after a signal
1120 * has been taken. Reset signal mask and
1121 * stack state from context left by sendsig (above).
1122 * Return to previous pc and psl as specified by
1123 * context left by sendsig. Check carefully to
1124 * make sure that the user has not modified the
1125 * state to gain improper privileges.
1130 sigreturn(struct thread *td,
1131 struct sigreturn_args /* {
1132 ucontext_t *sigcntxp;
1136 struct trapframe *tf;
1143 * Fetch the entire context structure at once for speed.
1144 * We don't use a normal argument to simplify RSE handling.
1146 if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc)))
1149 set_mcontext(td, &uc.uc_mcontext);
1151 #if defined(COMPAT_43)
1152 if (sigonstack(tf->tf_special.sp))
1153 td->td_sigstk.ss_flags |= SS_ONSTACK;
1155 td->td_sigstk.ss_flags &= ~SS_ONSTACK;
1157 kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
1159 return (EJUSTRETURN);
1162 #ifdef COMPAT_FREEBSD4
1164 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
1167 return sigreturn(td, (struct sigreturn_args *)uap);
1172 * Construct a PCB from a trapframe. This is called from kdb_trap() where
1173 * we want to start a backtrace from the function that caused us to enter
1174 * the debugger. We have the context in the trapframe, but base the trace
1175 * on the PCB. The PCB doesn't have to be perfect, as long as it contains
1176 * enough for a backtrace.
1179 makectx(struct trapframe *tf, struct pcb *pcb)
1182 pcb->pcb_special = tf->tf_special;
1183 pcb->pcb_special.__spare = ~0UL; /* XXX see unwind.c */
1184 save_callee_saved(&pcb->pcb_preserved);
1185 save_callee_saved_fp(&pcb->pcb_preserved_fp);
1189 ia64_flush_dirty(struct thread *td, struct _special *r)
1193 uint64_t bspst, kstk, rnat;
1199 kstk = td->td_kstack + (r->bspstore & 0x1ffUL);
1200 if (td == curthread) {
1201 __asm __volatile("mov ar.rsc=0;;");
1202 __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst));
1203 /* Make sure we have all the user registers written out. */
1204 if (bspst - kstk < r->ndirty) {
1205 __asm __volatile("flushrs;;");
1206 __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst));
1208 __asm __volatile("mov %0=ar.rnat;;" : "=r"(rnat));
1209 __asm __volatile("mov ar.rsc=3");
1210 error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty);
1212 r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL))
1213 ? *(uint64_t*)(kstk | 0x1f8L) : rnat;
1215 locked = PROC_LOCKED(td->td_proc);
1218 iov.iov_base = (void*)(uintptr_t)kstk;
1219 iov.iov_len = r->ndirty;
1222 uio.uio_offset = r->bspstore;
1223 uio.uio_resid = r->ndirty;
1224 uio.uio_segflg = UIO_SYSSPACE;
1225 uio.uio_rw = UIO_WRITE;
1227 error = proc_rwmem(td->td_proc, &uio);
1229 * XXX proc_rwmem() doesn't currently return ENOSPC,
1230 * so I think it can bogusly return 0. Neither do
1231 * we allow short writes.
1233 if (uio.uio_resid != 0 && error == 0)
1239 r->bspstore += r->ndirty;
1245 get_mcontext(struct thread *td, mcontext_t *mc, int flags)
1247 struct trapframe *tf;
1251 bzero(mc, sizeof(*mc));
1252 mc->mc_special = tf->tf_special;
1253 error = ia64_flush_dirty(td, &mc->mc_special);
1254 if (tf->tf_flags & FRAME_SYSCALL) {
1255 mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT;
1256 mc->mc_scratch = tf->tf_scratch;
1257 if (flags & GET_MC_CLEAR_RET) {
1258 mc->mc_scratch.gr8 = 0;
1259 mc->mc_scratch.gr9 = 0;
1260 mc->mc_scratch.gr10 = 0;
1261 mc->mc_scratch.gr11 = 0;
1264 mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT;
1265 mc->mc_scratch = tf->tf_scratch;
1266 mc->mc_scratch_fp = tf->tf_scratch_fp;
1268 * XXX If the thread never used the high FP registers, we
1269 * probably shouldn't waste time saving them.
1271 ia64_highfp_save(td);
1272 mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID;
1273 mc->mc_high_fp = td->td_pcb->pcb_high_fp;
1275 save_callee_saved(&mc->mc_preserved);
1276 save_callee_saved_fp(&mc->mc_preserved_fp);
1281 set_mcontext(struct thread *td, const mcontext_t *mc)
1284 struct trapframe *tf;
1289 KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1290 ("Whoa there! We have more than 8KB of dirty registers!"));
1294 * Only copy the user mask and the restart instruction bit from
1297 psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL |
1298 IA64_PSR_MFH | IA64_PSR_RI;
1299 s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask);
1300 /* We don't have any dirty registers of the new context. */
1302 if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) {
1304 * We can get an async context passed to us while we
1305 * entered the kernel through a syscall: sigreturn(2)
1306 * takes contexts that could previously be the result of
1307 * a trap or interrupt.
1308 * Hence, we cannot assert that the trapframe is not
1309 * a syscall frame, but we can assert that it's at
1310 * least an expected syscall.
1312 if (tf->tf_flags & FRAME_SYSCALL) {
1313 KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo"));
1314 tf->tf_flags &= ~FRAME_SYSCALL;
1316 tf->tf_scratch = mc->mc_scratch;
1317 tf->tf_scratch_fp = mc->mc_scratch_fp;
1318 if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID)
1319 td->td_pcb->pcb_high_fp = mc->mc_high_fp;
1321 KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
1322 if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) {
1323 s.cfm = tf->tf_special.cfm;
1324 s.iip = tf->tf_special.iip;
1325 tf->tf_scratch.gr15 = 0; /* Clear syscall nr. */
1327 tf->tf_scratch = mc->mc_scratch;
1330 restore_callee_saved(&mc->mc_preserved);
1331 restore_callee_saved_fp(&mc->mc_preserved_fp);
1337 * Clear registers on exec.
1340 exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
1342 struct trapframe *tf;
1343 uint64_t *ksttop, *kst;
1346 ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty +
1347 (tf->tf_special.bspstore & 0x1ffUL));
1350 * We can ignore up to 8KB of dirty registers by masking off the
1351 * lower 13 bits in exception_restore() or epc_syscall(). This
1352 * should be enough for a couple of years, but if there are more
1353 * than 8KB of dirty registers, we lose track of the bottom of
1354 * the kernel stack. The solution is to copy the active part of
1355 * the kernel stack down 1 page (or 2, but not more than that)
1356 * so that we always have less than 8KB of dirty registers.
1358 KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1359 ("Whoa there! We have more than 8KB of dirty registers!"));
1361 bzero(&tf->tf_special, sizeof(tf->tf_special));
1362 if ((tf->tf_flags & FRAME_SYSCALL) == 0) { /* break syscalls. */
1363 bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
1364 bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
1365 tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
1366 tf->tf_special.bspstore = IA64_BACKINGSTORE;
1368 * Copy the arguments onto the kernel register stack so that
1369 * they get loaded by the loadrs instruction. Skip over the
1370 * NaT collection points.
1373 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1376 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1378 *kst-- = ps_strings;
1379 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1382 tf->tf_special.ndirty = (ksttop - kst) << 3;
1383 } else { /* epc syscalls (default). */
1384 tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
1385 tf->tf_special.bspstore = IA64_BACKINGSTORE + 24;
1387 * Write values for out0, out1 and out2 to the user's backing
1388 * store and arrange for them to be restored into the user's
1389 * initial register frame.
1390 * Assumes that (bspstore & 0x1f8) < 0x1e0.
1392 suword((caddr_t)tf->tf_special.bspstore - 24, stack);
1393 suword((caddr_t)tf->tf_special.bspstore - 16, ps_strings);
1394 suword((caddr_t)tf->tf_special.bspstore - 8, 0);
1397 tf->tf_special.iip = entry;
1398 tf->tf_special.sp = (stack & ~15) - 16;
1399 tf->tf_special.rsc = 0xf;
1400 tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
1401 tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT |
1402 IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN |
1407 ptrace_set_pc(struct thread *td, unsigned long addr)
1411 switch (addr & 0xFUL) {
1413 slot = IA64_PSR_RI_0;
1416 /* XXX we need to deal with MLX bundles here */
1417 slot = IA64_PSR_RI_1;
1420 slot = IA64_PSR_RI_2;
1426 td->td_frame->tf_special.iip = addr & ~0x0FULL;
1427 td->td_frame->tf_special.psr =
1428 (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot;
1433 ptrace_single_step(struct thread *td)
1435 struct trapframe *tf;
1438 * There's no way to set single stepping when we're leaving the
1439 * kernel through the EPC syscall path. The way we solve this is
1440 * by enabling the lower-privilege trap so that we re-enter the
1441 * kernel as soon as the privilege level changes. See trap.c for
1442 * how we proceed from there.
1445 if (tf->tf_flags & FRAME_SYSCALL)
1446 tf->tf_special.psr |= IA64_PSR_LP;
1448 tf->tf_special.psr |= IA64_PSR_SS;
1453 ptrace_clear_single_step(struct thread *td)
1455 struct trapframe *tf;
1458 * Clear any and all status bits we may use to implement single
1462 tf->tf_special.psr &= ~IA64_PSR_SS;
1463 tf->tf_special.psr &= ~IA64_PSR_LP;
1464 tf->tf_special.psr &= ~IA64_PSR_TB;
1469 fill_regs(struct thread *td, struct reg *regs)
1471 struct trapframe *tf;
1474 regs->r_special = tf->tf_special;
1475 regs->r_scratch = tf->tf_scratch;
1476 save_callee_saved(®s->r_preserved);
1481 set_regs(struct thread *td, struct reg *regs)
1483 struct trapframe *tf;
1487 error = ia64_flush_dirty(td, &tf->tf_special);
1489 tf->tf_special = regs->r_special;
1490 tf->tf_special.bspstore += tf->tf_special.ndirty;
1491 tf->tf_special.ndirty = 0;
1492 tf->tf_scratch = regs->r_scratch;
1493 restore_callee_saved(®s->r_preserved);
1499 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1506 set_dbregs(struct thread *td, struct dbreg *dbregs)
1513 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1515 struct trapframe *frame = td->td_frame;
1516 struct pcb *pcb = td->td_pcb;
1518 /* Save the high FP registers. */
1519 ia64_highfp_save(td);
1521 fpregs->fpr_scratch = frame->tf_scratch_fp;
1522 save_callee_saved_fp(&fpregs->fpr_preserved);
1523 fpregs->fpr_high = pcb->pcb_high_fp;
1528 set_fpregs(struct thread *td, struct fpreg *fpregs)
1530 struct trapframe *frame = td->td_frame;
1531 struct pcb *pcb = td->td_pcb;
1533 /* Throw away the high FP registers (should be redundant). */
1534 ia64_highfp_drop(td);
1536 frame->tf_scratch_fp = fpregs->fpr_scratch;
1537 restore_callee_saved_fp(&fpregs->fpr_preserved);
1538 pcb->pcb_high_fp = fpregs->fpr_high;
1543 ia64_sync_icache(vm_offset_t va, vm_offset_t sz)
1547 if (!ia64_sync_icache_needed)