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/intr.h>
89 #include <machine/mca.h>
90 #include <machine/md_var.h>
91 #include <machine/mutex.h>
92 #include <machine/pal.h>
93 #include <machine/pcb.h>
94 #include <machine/reg.h>
95 #include <machine/sal.h>
96 #include <machine/sigframe.h>
98 #include <machine/smp.h>
100 #include <machine/unwind.h>
101 #include <machine/vmparam.h>
103 #include <i386/include/specialreg.h>
105 SYSCTL_NODE(_hw, OID_AUTO, freq, CTLFLAG_RD, 0, "");
106 SYSCTL_NODE(_machdep, OID_AUTO, cpu, CTLFLAG_RD, 0, "");
108 static u_int bus_freq;
109 SYSCTL_UINT(_hw_freq, OID_AUTO, bus, CTLFLAG_RD, &bus_freq, 0,
110 "Bus clock frequency");
112 static u_int cpu_freq;
113 SYSCTL_UINT(_hw_freq, OID_AUTO, cpu, CTLFLAG_RD, &cpu_freq, 0,
114 "CPU clock frequency");
116 static u_int itc_freq;
117 SYSCTL_UINT(_hw_freq, OID_AUTO, itc, CTLFLAG_RD, &itc_freq, 0,
122 u_int64_t pa_bootinfo;
123 struct bootinfo bootinfo;
127 extern u_int64_t kernel_text[], _end[];
129 extern u_int64_t ia64_gateway_page[];
130 extern u_int64_t break_sigtramp[];
131 extern u_int64_t epc_sigtramp[];
133 struct fpswa_iface *fpswa_iface;
135 u_int64_t ia64_pal_base;
136 u_int64_t ia64_port_base;
138 u_int64_t ia64_lapic_addr = PAL_PIB_DEFAULT_ADDR;
140 struct ia64_pib *ia64_pib;
142 static int ia64_sync_icache_needed;
144 char machine[] = MACHINE;
145 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
147 static char cpu_model[64];
148 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0,
149 "The CPU model name");
151 static char cpu_family[64];
152 SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0,
153 "The CPU family name");
156 extern vm_offset_t ksym_start, ksym_end;
160 struct msgbuf *msgbufp = NULL;
162 /* Other subsystems (e.g., ACPI) can hook this later. */
163 void (*cpu_idle_hook)(void) = NULL;
168 #define PHYSMAP_SIZE (2 * VM_PHYSSEG_MAX)
170 vm_paddr_t phys_avail[PHYSMAP_SIZE + 2];
172 /* must be 2 less so 0 0 can signal end of chunks */
173 #define PHYS_AVAIL_ARRAY_END ((sizeof(phys_avail) / sizeof(vm_offset_t)) - 2)
175 struct kva_md_info kmi;
178 #define Ghz (1000L*Mhz)
184 char *family_name, *model_name;
185 u_int64_t features, tmp;
186 int number, revision, model, family, archrev;
189 * Assumes little-endian.
191 *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
192 *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
195 tmp = ia64_get_cpuid(3);
196 number = (tmp >> 0) & 0xff;
197 revision = (tmp >> 8) & 0xff;
198 model = (tmp >> 16) & 0xff;
199 family = (tmp >> 24) & 0xff;
200 archrev = (tmp >> 32) & 0xff;
202 family_name = model_name = "unknown";
205 family_name = "Itanium";
206 model_name = "Merced";
209 family_name = "Itanium 2";
212 model_name = "McKinley";
216 * Deerfield is a low-voltage variant based on the
217 * Madison core. We need circumstantial evidence
218 * (i.e. the clock frequency) to identify those.
219 * Allow for roughly 1% error margin.
221 if (cpu_freq > 990 && cpu_freq < 1010)
222 model_name = "Deerfield";
224 model_name = "Madison";
227 model_name = "Madison II";
232 ia64_sync_icache_needed = 1;
234 family_name = "Itanium 2";
237 model_name = "Montecito";
242 snprintf(cpu_family, sizeof(cpu_family), "%s", family_name);
243 snprintf(cpu_model, sizeof(cpu_model), "%s", model_name);
245 features = ia64_get_cpuid(4);
247 printf("CPU: %s (", model_name);
249 printf("%u Mhz ", cpu_freq);
250 printf("%s)\n", family_name);
251 printf(" Origin = \"%s\" Revision = %d\n", vendor, revision);
252 printf(" Features = 0x%b\n", (u_int32_t) features,
254 "\001LB" /* long branch (brl) instruction. */
255 "\002SD" /* Spontaneous deferral. */
256 "\003AO" /* 16-byte atomic operations (ld, st, cmpxchg). */ );
260 cpu_startup(void *dummy)
264 struct pcpu_stats *pcs;
267 * Good {morning,afternoon,evening,night}.
274 printf("real memory = %ld (%ld MB)\n", ia64_ptob(Maxmem),
275 ia64_ptob(Maxmem) / 1048576);
279 * Display any holes after the first chunk of extended memory.
284 printf("Physical memory chunk(s):\n");
285 for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
286 long size1 = phys_avail[indx + 1] - phys_avail[indx];
288 printf("0x%08lx - 0x%08lx, %ld bytes (%ld pages)\n",
289 phys_avail[indx], phys_avail[indx + 1] - 1, size1,
290 size1 >> PAGE_SHIFT);
294 vm_ksubmap_init(&kmi);
296 printf("avail memory = %ld (%ld MB)\n", ptoa(cnt.v_free_count),
297 ptoa(cnt.v_free_count) / 1048576);
299 if (fpswa_iface == NULL)
300 printf("Warning: no FPSWA package supplied\n");
302 printf("FPSWA Revision = 0x%lx, Entry = %p\n",
303 (long)fpswa_iface->if_rev, (void *)fpswa_iface->if_fpswa);
306 * Set up buffers, so they can be used to read disk labels.
309 vm_pager_bufferinit();
312 * Traverse the MADT to discover IOSAPIC and Local SAPIC
316 ia64_pib = pmap_mapdev(ia64_lapic_addr, sizeof(*ia64_pib));
321 * Create sysctl tree for per-CPU information.
323 SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
324 snprintf(nodename, sizeof(nodename), "%u", pc->pc_cpuid);
325 sysctl_ctx_init(&pc->pc_md.sysctl_ctx);
326 pc->pc_md.sysctl_tree = SYSCTL_ADD_NODE(&pc->pc_md.sysctl_ctx,
327 SYSCTL_STATIC_CHILDREN(_machdep_cpu), OID_AUTO, nodename,
328 CTLFLAG_RD, NULL, "");
329 if (pc->pc_md.sysctl_tree == NULL)
332 pcs = &pc->pc_md.stats;
334 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
335 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
336 "nasts", CTLFLAG_RD, &pcs->pcs_nasts,
337 "Number of IPI_AST interrupts");
339 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
340 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
341 "nclks", CTLFLAG_RD, &pcs->pcs_nclks,
342 "Number of clock interrupts");
344 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
345 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
346 "nextints", CTLFLAG_RD, &pcs->pcs_nextints,
347 "Number of ExtINT interrupts");
349 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
350 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
351 "nhighfps", CTLFLAG_RD, &pcs->pcs_nhighfps,
352 "Number of IPI_HIGH_FP interrupts");
354 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
355 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
356 "nhwints", CTLFLAG_RD, &pcs->pcs_nhwints,
357 "Number of hardware (device) interrupts");
359 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
360 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
361 "npreempts", CTLFLAG_RD, &pcs->pcs_npreempts,
362 "Number of IPI_PREEMPT interrupts");
364 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
365 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
366 "nrdvs", CTLFLAG_RD, &pcs->pcs_nrdvs,
367 "Number of IPI_RENDEZVOUS interrupts");
369 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
370 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
371 "nstops", CTLFLAG_RD, &pcs->pcs_nstops,
372 "Number of IPI_STOP interrupts");
374 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
375 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
376 "nstrays", CTLFLAG_RD, &pcs->pcs_nstrays,
377 "Number of stray vectors");
380 SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
390 cpu_flush_dcache(void *ptr, size_t len)
394 va = (uintptr_t)ptr & ~31;
395 lim = (uintptr_t)ptr + len;
404 /* Get current clock frequency for the given cpu id. */
406 cpu_est_clockrate(int cpu_id, uint64_t *rate)
409 if (pcpu_find(cpu_id) == NULL || rate == NULL)
411 *rate = (u_long)cpu_freq * 1000000ul;
425 struct ia64_pal_result res;
427 if (cpu_idle_hook != NULL)
430 res = ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
434 cpu_idle_wakeup(int cpu)
448 cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx)
450 struct pcb *oldpcb, *newpcb;
452 oldpcb = old->td_pcb;
454 ia32_savectx(oldpcb);
456 if (PCPU_GET(fpcurthread) == old)
457 old->td_frame->tf_special.psr |= IA64_PSR_DFH;
458 if (!savectx(oldpcb)) {
460 #if defined(SCHED_ULE) && defined(SMP)
461 /* td_lock is volatile */
462 while (new->td_lock == &blocked_lock)
465 newpcb = new->td_pcb;
466 oldpcb->pcb_current_pmap =
467 pmap_switch(newpcb->pcb_current_pmap);
468 PCPU_SET(curthread, new);
470 ia32_restorectx(newpcb);
472 if (PCPU_GET(fpcurthread) == new)
473 new->td_frame->tf_special.psr &= ~IA64_PSR_DFH;
475 /* We should not get here. */
476 panic("cpu_switch: restorectx() returned");
482 cpu_throw(struct thread *old __unused, struct thread *new)
486 newpcb = new->td_pcb;
487 (void)pmap_switch(newpcb->pcb_current_pmap);
488 PCPU_SET(curthread, new);
490 ia32_restorectx(newpcb);
493 /* We should not get here. */
494 panic("cpu_throw: restorectx() returned");
499 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
503 * Set pc_acpi_id to "uninitialized".
504 * See sys/dev/acpica/acpi_cpu.c
506 pcpu->pc_acpi_id = 0xffffffff;
515 if (td->td_md.md_spinlock_count == 0)
516 td->td_md.md_saved_intr = intr_disable();
517 td->td_md.md_spinlock_count++;
528 td->td_md.md_spinlock_count--;
529 if (td->td_md.md_spinlock_count == 0)
530 intr_restore(td->td_md.md_saved_intr);
534 map_vhpt(uintptr_t vhpt)
539 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
540 PTE_PL_KERN | PTE_AR_RW;
541 pte |= vhpt & PTE_PPN_MASK;
543 __asm __volatile("ptr.d %0,%1" :: "r"(vhpt),
544 "r"(IA64_ID_PAGE_SHIFT<<2));
546 __asm __volatile("mov %0=psr" : "=r"(psr));
547 __asm __volatile("rsm psr.ic|psr.i");
550 ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
552 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(2), "r"(pte));
553 __asm __volatile("mov psr.l=%0" :: "r" (psr));
563 if (ia64_pal_base == 0)
566 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
567 PTE_PL_KERN | PTE_AR_RWX;
568 pte |= ia64_pal_base & PTE_PPN_MASK;
570 __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
571 "r"(IA64_PHYS_TO_RR7(ia64_pal_base)), "r"(IA64_ID_PAGE_SHIFT<<2));
573 __asm __volatile("mov %0=psr" : "=r"(psr));
574 __asm __volatile("rsm psr.ic|psr.i");
576 ia64_set_ifa(IA64_PHYS_TO_RR7(ia64_pal_base));
577 ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
579 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(1), "r"(pte));
581 __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte));
582 __asm __volatile("mov psr.l=%0" :: "r" (psr));
587 map_gateway_page(void)
592 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
593 PTE_PL_KERN | PTE_AR_X_RX;
594 pte |= (uint64_t)ia64_gateway_page & PTE_PPN_MASK;
596 __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
597 "r"(VM_MAX_ADDRESS), "r"(PAGE_SHIFT << 2));
599 __asm __volatile("mov %0=psr" : "=r"(psr));
600 __asm __volatile("rsm psr.ic|psr.i");
602 ia64_set_ifa(VM_MAX_ADDRESS);
603 ia64_set_itir(PAGE_SHIFT << 2);
605 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte));
607 __asm __volatile("itr.i itr[%0]=%1" :: "r"(3), "r"(pte));
608 __asm __volatile("mov psr.l=%0" :: "r" (psr));
611 /* Expose the mapping to userland in ar.k5 */
612 ia64_set_k5(VM_MAX_ADDRESS);
616 freq_ratio(u_long base, u_long ratio)
620 f = (base * (ratio >> 32)) / (ratio & 0xfffffffful);
621 return ((f + 500000) / 1000000);
625 calculate_frequencies(void)
627 struct ia64_sal_result sal;
628 struct ia64_pal_result pal;
630 sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
631 pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
633 if (sal.sal_status == 0 && pal.pal_status == 0) {
635 printf("Platform clock frequency %ld Hz\n",
637 printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
638 "ITC ratio %ld/%ld\n",
639 pal.pal_result[0] >> 32,
640 pal.pal_result[0] & ((1L << 32) - 1),
641 pal.pal_result[1] >> 32,
642 pal.pal_result[1] & ((1L << 32) - 1),
643 pal.pal_result[2] >> 32,
644 pal.pal_result[2] & ((1L << 32) - 1));
646 cpu_freq = freq_ratio(sal.sal_result[0], pal.pal_result[0]);
647 bus_freq = freq_ratio(sal.sal_result[0], pal.pal_result[1]);
648 itc_freq = freq_ratio(sal.sal_result[0], pal.pal_result[2]);
652 struct ia64_init_return
655 struct ia64_init_return ret;
657 vm_offset_t kernstart, kernend;
658 vm_offset_t kernstartpfn, kernendpfn, pfn0, pfn1;
661 int metadata_missing;
663 /* NO OUTPUT ALLOWED UNTIL FURTHER NOTICE */
666 * TODO: Disable interrupts, floating point etc.
667 * Maybe flush cache and tlb
669 ia64_set_fpsr(IA64_FPSR_DEFAULT);
672 * TODO: Get critical system information (if possible, from the
673 * information provided by the boot program).
677 * pa_bootinfo is the physical address of the bootinfo block as
678 * passed to us by the loader and set in locore.s.
680 bootinfo = *(struct bootinfo *)(IA64_PHYS_TO_RR7(pa_bootinfo));
682 if (bootinfo.bi_magic != BOOTINFO_MAGIC || bootinfo.bi_version != 1) {
683 bzero(&bootinfo, sizeof(bootinfo));
684 bootinfo.bi_kernend = (vm_offset_t) round_page(_end);
688 * Look for the I/O ports first - we need them for console
691 for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
692 switch (md->md_type) {
693 case EFI_MD_TYPE_IOPORT:
694 ia64_port_base = (uintptr_t)pmap_mapdev(md->md_phys,
695 md->md_pages * EFI_PAGE_SIZE);
697 case EFI_MD_TYPE_PALCODE:
698 ia64_pal_base = md->md_phys;
703 metadata_missing = 0;
704 if (bootinfo.bi_modulep)
705 preload_metadata = (caddr_t)bootinfo.bi_modulep;
707 metadata_missing = 1;
709 if (envmode == 0 && bootinfo.bi_envp)
710 kern_envp = (caddr_t)bootinfo.bi_envp;
712 kern_envp = static_env;
715 * Look at arguments passed to us and compute boothowto.
717 boothowto = bootinfo.bi_boothowto;
720 * Catch case of boot_verbose set in environment.
722 if ((p = getenv("boot_verbose")) != NULL) {
723 if (strcmp(p, "yes") == 0 || strcmp(p, "YES") == 0) {
724 boothowto |= RB_VERBOSE;
729 if (boothowto & RB_VERBOSE)
733 * Find the beginning and end of the kernel.
735 kernstart = trunc_page(kernel_text);
737 ksym_start = bootinfo.bi_symtab;
738 ksym_end = bootinfo.bi_esymtab;
739 kernend = (vm_offset_t)round_page(ksym_end);
741 kernend = (vm_offset_t)round_page(_end);
743 /* But if the bootstrap tells us otherwise, believe it! */
744 if (bootinfo.bi_kernend)
745 kernend = round_page(bootinfo.bi_kernend);
748 * Setup the PCPU data for the bootstrap processor. It is needed
749 * by printf(). Also, since printf() has critical sections, we
750 * need to initialize at least pc_curthread.
753 ia64_set_k4((u_int64_t)pcpup);
754 pcpu_init(pcpup, 0, sizeof(pcpu0));
755 dpcpu_init((void *)kernend, 0);
756 kernend += DPCPU_SIZE;
757 PCPU_SET(curthread, &thread0);
760 * Initialize the console before we print anything out.
764 /* OUTPUT NOW ALLOWED */
766 if (ia64_pal_base != 0) {
767 ia64_pal_base &= ~IA64_ID_PAGE_MASK;
769 * We use a TR to map the first 256M of memory - this might
770 * cover the palcode too.
772 if (ia64_pal_base == 0)
773 printf("PAL code mapped by the kernel's TR\n");
775 printf("PAL code not found\n");
778 * Wire things up so we can call the firmware.
781 efi_boot_minimal(bootinfo.bi_systab);
783 calculate_frequencies();
785 if (metadata_missing)
786 printf("WARNING: loader(8) metadata is missing!\n");
788 /* Get FPSWA interface */
789 fpswa_iface = (bootinfo.bi_fpswa == 0) ? NULL :
790 (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo.bi_fpswa);
792 /* Init basic tunables, including hz */
795 p = getenv("kernelname");
797 strncpy(kernelname, p, sizeof(kernelname) - 1);
801 kernstartpfn = atop(IA64_RR_MASK(kernstart));
802 kernendpfn = atop(IA64_RR_MASK(kernend));
805 * Size the memory regions and load phys_avail[] with the results.
809 * Find out how much memory is available, by looking at
810 * the memory descriptors.
814 printf("Memory descriptor count: %d\n", mdcount);
818 for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
820 printf("MD %p: type %d pa 0x%lx cnt 0x%lx\n", md,
821 md->md_type, md->md_phys, md->md_pages);
824 pfn0 = ia64_btop(round_page(md->md_phys));
825 pfn1 = ia64_btop(trunc_page(md->md_phys + md->md_pages * 4096));
829 if (md->md_type != EFI_MD_TYPE_FREE)
833 * We have a memory descriptor that describes conventional
834 * memory that is for general use. We must determine if the
835 * loader has put the kernel in this region.
837 physmem += (pfn1 - pfn0);
838 if (pfn0 <= kernendpfn && kernstartpfn <= pfn1) {
840 * Must compute the location of the kernel
841 * within the segment.
844 printf("Descriptor %p contains kernel\n", mp);
846 if (pfn0 < kernstartpfn) {
848 * There is a chunk before the kernel.
851 printf("Loading chunk before kernel: "
852 "0x%lx / 0x%lx\n", pfn0, kernstartpfn);
854 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
855 phys_avail[phys_avail_cnt+1] = ia64_ptob(kernstartpfn);
858 if (kernendpfn < pfn1) {
860 * There is a chunk after the kernel.
863 printf("Loading chunk after kernel: "
864 "0x%lx / 0x%lx\n", kernendpfn, pfn1);
866 phys_avail[phys_avail_cnt] = ia64_ptob(kernendpfn);
867 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
872 * Just load this cluster as one chunk.
875 printf("Loading descriptor %d: 0x%lx / 0x%lx\n", i,
878 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
879 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
884 phys_avail[phys_avail_cnt] = 0;
887 init_param2(physmem);
890 * Initialize error message buffer (at end of core).
892 msgbufp = (struct msgbuf *)pmap_steal_memory(MSGBUF_SIZE);
893 msgbufinit(msgbufp, MSGBUF_SIZE);
895 proc_linkup0(&proc0, &thread0);
897 * Init mapping for kernel stack for proc 0
899 thread0.td_kstack = pmap_steal_memory(KSTACK_PAGES * PAGE_SIZE);
900 thread0.td_kstack_pages = KSTACK_PAGES;
905 * Initialize the rest of proc 0's PCB.
907 * Set the kernel sp, reserving space for an (empty) trapframe,
908 * and make proc0's trapframe pointer point to it for sanity.
909 * Initialise proc0's backing store to start after u area.
911 cpu_thread_alloc(&thread0);
912 thread0.td_frame->tf_flags = FRAME_SYSCALL;
913 thread0.td_pcb->pcb_special.sp =
914 (u_int64_t)thread0.td_frame - 16;
915 thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack;
918 * Initialize the virtual memory system.
923 * Initialize debuggers, and break into them if appropriate.
928 if (boothowto & RB_KDB)
929 kdb_enter(KDB_WHY_BOOTFLAGS,
930 "Boot flags requested debugger\n");
936 ret.bspstore = thread0.td_pcb->pcb_special.bspstore;
937 ret.sp = thread0.td_pcb->pcb_special.sp;
945 return (bootinfo.bi_hcdp);
949 bzero(void *buf, size_t len)
953 while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
957 while (len >= sizeof(u_long) * 8) {
959 *((u_long*) p + 1) = 0;
960 *((u_long*) p + 2) = 0;
961 *((u_long*) p + 3) = 0;
962 len -= sizeof(u_long) * 8;
963 *((u_long*) p + 4) = 0;
964 *((u_long*) p + 5) = 0;
965 *((u_long*) p + 6) = 0;
966 *((u_long*) p + 7) = 0;
967 p += sizeof(u_long) * 8;
969 while (len >= sizeof(u_long)) {
971 len -= sizeof(u_long);
990 u_int64_t start, end, now;
994 start = ia64_get_itc();
995 end = start + itc_freq * n;
996 /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
998 now = ia64_get_itc();
999 } while (now < end || (now > start && end < start));
1005 * Send an interrupt (signal) to a process.
1008 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
1012 struct trapframe *tf;
1013 struct sigacts *psp;
1014 struct sigframe sf, *sfp;
1022 PROC_LOCK_ASSERT(p, MA_OWNED);
1023 sig = ksi->ksi_signo;
1024 code = ksi->ksi_code;
1026 mtx_assert(&psp->ps_mtx, MA_OWNED);
1028 sp = tf->tf_special.sp;
1029 oonstack = sigonstack(sp);
1032 /* save user context */
1033 bzero(&sf, sizeof(struct sigframe));
1034 sf.sf_uc.uc_sigmask = *mask;
1035 sf.sf_uc.uc_stack = td->td_sigstk;
1036 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
1037 ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1040 * Allocate and validate space for the signal handler
1041 * context. Note that if the stack is in P0 space, the
1042 * call to grow() is a nop, and the useracc() check
1043 * will fail if the process has not already allocated
1044 * the space with a `brk'.
1046 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
1047 SIGISMEMBER(psp->ps_sigonstack, sig)) {
1048 sbs = (u_int64_t)td->td_sigstk.ss_sp;
1049 sbs = (sbs + 15) & ~15;
1050 sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size);
1051 #if defined(COMPAT_43)
1052 td->td_sigstk.ss_flags |= SS_ONSTACK;
1055 sfp = (struct sigframe *)sp;
1056 sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15);
1058 /* Fill in the siginfo structure for POSIX handlers. */
1059 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
1060 sf.sf_si = ksi->ksi_info;
1061 sf.sf_si.si_signo = sig;
1063 * XXX this shouldn't be here after code in trap.c
1066 sf.sf_si.si_addr = (void*)tf->tf_special.ifa;
1067 code = (u_int64_t)&sfp->sf_si;
1070 mtx_unlock(&psp->ps_mtx);
1073 get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
1075 /* Copy the frame out to userland. */
1076 if (copyout(&sf, sfp, sizeof(sf)) != 0) {
1078 * Process has trashed its stack; give it an illegal
1079 * instruction to halt it in its tracks.
1082 sigexit(td, SIGILL);
1086 if ((tf->tf_flags & FRAME_SYSCALL) == 0) {
1087 tf->tf_special.psr &= ~IA64_PSR_RI;
1088 tf->tf_special.iip = ia64_get_k5() +
1089 ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
1091 tf->tf_special.iip = ia64_get_k5() +
1092 ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
1095 * Setup the trapframe to return to the signal trampoline. We pass
1096 * information to the trampoline in the following registers:
1098 * gp new backing store or NULL
1100 * r9 signal code or siginfo pointer
1101 * r10 signal handler (function descriptor)
1103 tf->tf_special.sp = (u_int64_t)sfp - 16;
1104 tf->tf_special.gp = sbs;
1105 tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore;
1106 tf->tf_special.ndirty = 0;
1107 tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat;
1108 tf->tf_scratch.gr8 = sig;
1109 tf->tf_scratch.gr9 = code;
1110 tf->tf_scratch.gr10 = (u_int64_t)catcher;
1113 mtx_lock(&psp->ps_mtx);
1117 * System call to cleanup state after a signal
1118 * has been taken. Reset signal mask and
1119 * stack state from context left by sendsig (above).
1120 * Return to previous pc and psl as specified by
1121 * context left by sendsig. Check carefully to
1122 * make sure that the user has not modified the
1123 * state to gain improper privileges.
1128 sigreturn(struct thread *td,
1129 struct sigreturn_args /* {
1130 ucontext_t *sigcntxp;
1134 struct trapframe *tf;
1141 * Fetch the entire context structure at once for speed.
1142 * We don't use a normal argument to simplify RSE handling.
1144 if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc)))
1147 set_mcontext(td, &uc.uc_mcontext);
1149 #if defined(COMPAT_43)
1150 if (sigonstack(tf->tf_special.sp))
1151 td->td_sigstk.ss_flags |= SS_ONSTACK;
1153 td->td_sigstk.ss_flags &= ~SS_ONSTACK;
1155 kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
1157 return (EJUSTRETURN);
1160 #ifdef COMPAT_FREEBSD4
1162 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
1165 return sigreturn(td, (struct sigreturn_args *)uap);
1170 * Construct a PCB from a trapframe. This is called from kdb_trap() where
1171 * we want to start a backtrace from the function that caused us to enter
1172 * the debugger. We have the context in the trapframe, but base the trace
1173 * on the PCB. The PCB doesn't have to be perfect, as long as it contains
1174 * enough for a backtrace.
1177 makectx(struct trapframe *tf, struct pcb *pcb)
1180 pcb->pcb_special = tf->tf_special;
1181 pcb->pcb_special.__spare = ~0UL; /* XXX see unwind.c */
1182 save_callee_saved(&pcb->pcb_preserved);
1183 save_callee_saved_fp(&pcb->pcb_preserved_fp);
1187 ia64_flush_dirty(struct thread *td, struct _special *r)
1191 uint64_t bspst, kstk, rnat;
1197 kstk = td->td_kstack + (r->bspstore & 0x1ffUL);
1198 if (td == curthread) {
1199 __asm __volatile("mov ar.rsc=0;;");
1200 __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst));
1201 /* Make sure we have all the user registers written out. */
1202 if (bspst - kstk < r->ndirty) {
1203 __asm __volatile("flushrs;;");
1204 __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst));
1206 __asm __volatile("mov %0=ar.rnat;;" : "=r"(rnat));
1207 __asm __volatile("mov ar.rsc=3");
1208 error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty);
1210 r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL))
1211 ? *(uint64_t*)(kstk | 0x1f8L) : rnat;
1213 locked = PROC_LOCKED(td->td_proc);
1216 iov.iov_base = (void*)(uintptr_t)kstk;
1217 iov.iov_len = r->ndirty;
1220 uio.uio_offset = r->bspstore;
1221 uio.uio_resid = r->ndirty;
1222 uio.uio_segflg = UIO_SYSSPACE;
1223 uio.uio_rw = UIO_WRITE;
1225 error = proc_rwmem(td->td_proc, &uio);
1227 * XXX proc_rwmem() doesn't currently return ENOSPC,
1228 * so I think it can bogusly return 0. Neither do
1229 * we allow short writes.
1231 if (uio.uio_resid != 0 && error == 0)
1237 r->bspstore += r->ndirty;
1243 get_mcontext(struct thread *td, mcontext_t *mc, int flags)
1245 struct trapframe *tf;
1249 bzero(mc, sizeof(*mc));
1250 mc->mc_special = tf->tf_special;
1251 error = ia64_flush_dirty(td, &mc->mc_special);
1252 if (tf->tf_flags & FRAME_SYSCALL) {
1253 mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT;
1254 mc->mc_scratch = tf->tf_scratch;
1255 if (flags & GET_MC_CLEAR_RET) {
1256 mc->mc_scratch.gr8 = 0;
1257 mc->mc_scratch.gr9 = 0;
1258 mc->mc_scratch.gr10 = 0;
1259 mc->mc_scratch.gr11 = 0;
1262 mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT;
1263 mc->mc_scratch = tf->tf_scratch;
1264 mc->mc_scratch_fp = tf->tf_scratch_fp;
1266 * XXX If the thread never used the high FP registers, we
1267 * probably shouldn't waste time saving them.
1269 ia64_highfp_save(td);
1270 mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID;
1271 mc->mc_high_fp = td->td_pcb->pcb_high_fp;
1273 save_callee_saved(&mc->mc_preserved);
1274 save_callee_saved_fp(&mc->mc_preserved_fp);
1279 set_mcontext(struct thread *td, const mcontext_t *mc)
1282 struct trapframe *tf;
1287 KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1288 ("Whoa there! We have more than 8KB of dirty registers!"));
1292 * Only copy the user mask and the restart instruction bit from
1295 psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL |
1296 IA64_PSR_MFH | IA64_PSR_RI;
1297 s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask);
1298 /* We don't have any dirty registers of the new context. */
1300 if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) {
1302 * We can get an async context passed to us while we
1303 * entered the kernel through a syscall: sigreturn(2)
1304 * takes contexts that could previously be the result of
1305 * a trap or interrupt.
1306 * Hence, we cannot assert that the trapframe is not
1307 * a syscall frame, but we can assert that it's at
1308 * least an expected syscall.
1310 if (tf->tf_flags & FRAME_SYSCALL) {
1311 KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo"));
1312 tf->tf_flags &= ~FRAME_SYSCALL;
1314 tf->tf_scratch = mc->mc_scratch;
1315 tf->tf_scratch_fp = mc->mc_scratch_fp;
1316 if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID)
1317 td->td_pcb->pcb_high_fp = mc->mc_high_fp;
1319 KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
1320 if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) {
1321 s.cfm = tf->tf_special.cfm;
1322 s.iip = tf->tf_special.iip;
1323 tf->tf_scratch.gr15 = 0; /* Clear syscall nr. */
1325 tf->tf_scratch = mc->mc_scratch;
1328 restore_callee_saved(&mc->mc_preserved);
1329 restore_callee_saved_fp(&mc->mc_preserved_fp);
1335 * Clear registers on exec.
1338 exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
1340 struct trapframe *tf;
1341 uint64_t *ksttop, *kst;
1344 ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty +
1345 (tf->tf_special.bspstore & 0x1ffUL));
1348 * We can ignore up to 8KB of dirty registers by masking off the
1349 * lower 13 bits in exception_restore() or epc_syscall(). This
1350 * should be enough for a couple of years, but if there are more
1351 * than 8KB of dirty registers, we lose track of the bottom of
1352 * the kernel stack. The solution is to copy the active part of
1353 * the kernel stack down 1 page (or 2, but not more than that)
1354 * so that we always have less than 8KB of dirty registers.
1356 KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1357 ("Whoa there! We have more than 8KB of dirty registers!"));
1359 bzero(&tf->tf_special, sizeof(tf->tf_special));
1360 if ((tf->tf_flags & FRAME_SYSCALL) == 0) { /* break syscalls. */
1361 bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
1362 bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
1363 tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
1364 tf->tf_special.bspstore = IA64_BACKINGSTORE;
1366 * Copy the arguments onto the kernel register stack so that
1367 * they get loaded by the loadrs instruction. Skip over the
1368 * NaT collection points.
1371 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1374 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1376 *kst-- = ps_strings;
1377 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1380 tf->tf_special.ndirty = (ksttop - kst) << 3;
1381 } else { /* epc syscalls (default). */
1382 tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
1383 tf->tf_special.bspstore = IA64_BACKINGSTORE + 24;
1385 * Write values for out0, out1 and out2 to the user's backing
1386 * store and arrange for them to be restored into the user's
1387 * initial register frame.
1388 * Assumes that (bspstore & 0x1f8) < 0x1e0.
1390 suword((caddr_t)tf->tf_special.bspstore - 24, stack);
1391 suword((caddr_t)tf->tf_special.bspstore - 16, ps_strings);
1392 suword((caddr_t)tf->tf_special.bspstore - 8, 0);
1395 tf->tf_special.iip = entry;
1396 tf->tf_special.sp = (stack & ~15) - 16;
1397 tf->tf_special.rsc = 0xf;
1398 tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
1399 tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT |
1400 IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN |
1405 ptrace_set_pc(struct thread *td, unsigned long addr)
1409 switch (addr & 0xFUL) {
1411 slot = IA64_PSR_RI_0;
1414 /* XXX we need to deal with MLX bundles here */
1415 slot = IA64_PSR_RI_1;
1418 slot = IA64_PSR_RI_2;
1424 td->td_frame->tf_special.iip = addr & ~0x0FULL;
1425 td->td_frame->tf_special.psr =
1426 (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot;
1431 ptrace_single_step(struct thread *td)
1433 struct trapframe *tf;
1436 * There's no way to set single stepping when we're leaving the
1437 * kernel through the EPC syscall path. The way we solve this is
1438 * by enabling the lower-privilege trap so that we re-enter the
1439 * kernel as soon as the privilege level changes. See trap.c for
1440 * how we proceed from there.
1443 if (tf->tf_flags & FRAME_SYSCALL)
1444 tf->tf_special.psr |= IA64_PSR_LP;
1446 tf->tf_special.psr |= IA64_PSR_SS;
1451 ptrace_clear_single_step(struct thread *td)
1453 struct trapframe *tf;
1456 * Clear any and all status bits we may use to implement single
1460 tf->tf_special.psr &= ~IA64_PSR_SS;
1461 tf->tf_special.psr &= ~IA64_PSR_LP;
1462 tf->tf_special.psr &= ~IA64_PSR_TB;
1467 fill_regs(struct thread *td, struct reg *regs)
1469 struct trapframe *tf;
1472 regs->r_special = tf->tf_special;
1473 regs->r_scratch = tf->tf_scratch;
1474 save_callee_saved(®s->r_preserved);
1479 set_regs(struct thread *td, struct reg *regs)
1481 struct trapframe *tf;
1485 error = ia64_flush_dirty(td, &tf->tf_special);
1487 tf->tf_special = regs->r_special;
1488 tf->tf_special.bspstore += tf->tf_special.ndirty;
1489 tf->tf_special.ndirty = 0;
1490 tf->tf_scratch = regs->r_scratch;
1491 restore_callee_saved(®s->r_preserved);
1497 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1504 set_dbregs(struct thread *td, struct dbreg *dbregs)
1511 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1513 struct trapframe *frame = td->td_frame;
1514 struct pcb *pcb = td->td_pcb;
1516 /* Save the high FP registers. */
1517 ia64_highfp_save(td);
1519 fpregs->fpr_scratch = frame->tf_scratch_fp;
1520 save_callee_saved_fp(&fpregs->fpr_preserved);
1521 fpregs->fpr_high = pcb->pcb_high_fp;
1526 set_fpregs(struct thread *td, struct fpreg *fpregs)
1528 struct trapframe *frame = td->td_frame;
1529 struct pcb *pcb = td->td_pcb;
1531 /* Throw away the high FP registers (should be redundant). */
1532 ia64_highfp_drop(td);
1534 frame->tf_scratch_fp = fpregs->fpr_scratch;
1535 restore_callee_saved_fp(&fpregs->fpr_preserved);
1536 pcb->pcb_high_fp = fpregs->fpr_high;
1541 ia64_sync_icache(vm_offset_t va, vm_offset_t sz)
1545 if (!ia64_sync_icache_needed)