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_sched.h"
35 #include "opt_xtrace.h"
37 #include <sys/param.h>
39 #include <sys/systm.h>
46 #include <sys/eventhandler.h>
48 #include <sys/imgact.h>
50 #include <sys/kernel.h>
51 #include <sys/linker.h>
53 #include <sys/malloc.h>
55 #include <sys/msgbuf.h>
57 #include <sys/ptrace.h>
58 #include <sys/random.h>
59 #include <sys/reboot.h>
60 #include <sys/rwlock.h>
61 #include <sys/sched.h>
62 #include <sys/signalvar.h>
63 #include <sys/syscall.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysproto.h>
67 #include <sys/ucontext.h>
70 #include <sys/vmmeter.h>
71 #include <sys/vnode.h>
75 #include <net/netisr.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_pager.h>
85 #include <machine/bootinfo.h>
86 #include <machine/cpu.h>
87 #include <machine/elf.h>
88 #include <machine/fpu.h>
89 #include <machine/intr.h>
90 #include <machine/kdb.h>
91 #include <machine/mca.h>
92 #include <machine/md_var.h>
93 #include <machine/pal.h>
94 #include <machine/pcb.h>
95 #include <machine/reg.h>
96 #include <machine/sal.h>
97 #include <machine/sigframe.h>
99 #include <machine/smp.h>
101 #include <machine/unwind.h>
102 #include <machine/vmparam.h>
105 * For atomicity reasons, we demand that pc_curthread is the first
106 * field in the struct pcpu. It allows us to read the pointer with
107 * a single atomic instruction:
108 * ld8 %curthread = [r13]
109 * Otherwise we would first have to calculate the load address and
110 * store the result in a temporary register and that for the load:
111 * add %temp = %offsetof(struct pcpu), r13
112 * ld8 %curthread = [%temp]
113 * A context switch inbetween the add and the ld8 could have the
114 * thread migrate to a different core. In that case, %curthread
115 * would be the thread running on the original core and not actually
116 * the current thread.
118 CTASSERT(offsetof(struct pcpu, pc_curthread) == 0);
120 static SYSCTL_NODE(_hw, OID_AUTO, freq, CTLFLAG_RD, 0, "");
121 static SYSCTL_NODE(_machdep, OID_AUTO, cpu, CTLFLAG_RD, 0, "");
123 static u_int bus_freq;
124 SYSCTL_UINT(_hw_freq, OID_AUTO, bus, CTLFLAG_RD, &bus_freq, 0,
125 "Bus clock frequency");
127 static u_int cpu_freq;
128 SYSCTL_UINT(_hw_freq, OID_AUTO, cpu, CTLFLAG_RD, &cpu_freq, 0,
129 "CPU clock frequency");
131 static u_int itc_freq;
132 SYSCTL_UINT(_hw_freq, OID_AUTO, itc, CTLFLAG_RD, &itc_freq, 0,
136 int unmapped_buf_allowed = 0;
138 struct bootinfo *bootinfo;
142 extern u_int64_t kernel_text[], _end[];
144 extern u_int64_t ia64_gateway_page[];
145 extern u_int64_t break_sigtramp[];
146 extern u_int64_t epc_sigtramp[];
148 struct fpswa_iface *fpswa_iface;
150 vm_size_t ia64_pal_size;
151 vm_paddr_t ia64_pal_base;
152 vm_offset_t ia64_port_base;
154 u_int64_t ia64_lapic_addr = PAL_PIB_DEFAULT_ADDR;
156 struct ia64_pib *ia64_pib;
158 static int ia64_sync_icache_needed;
160 char machine[] = MACHINE;
161 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
163 static char cpu_model[64];
164 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0,
165 "The CPU model name");
167 static char cpu_family[64];
168 SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0,
169 "The CPU family name");
172 extern vm_offset_t ksym_start, ksym_end;
175 struct msgbuf *msgbufp = NULL;
177 /* Other subsystems (e.g., ACPI) can hook this later. */
178 void (*cpu_idle_hook)(sbintime_t) = NULL;
180 struct kva_md_info kmi;
186 char *family_name, *model_name;
187 u_int64_t features, tmp;
188 int number, revision, model, family, archrev;
191 * Assumes little-endian.
193 *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
194 *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
197 tmp = ia64_get_cpuid(3);
198 number = (tmp >> 0) & 0xff;
199 revision = (tmp >> 8) & 0xff;
200 model = (tmp >> 16) & 0xff;
201 family = (tmp >> 24) & 0xff;
202 archrev = (tmp >> 32) & 0xff;
204 family_name = model_name = "unknown";
207 family_name = "Itanium";
208 model_name = "Merced";
211 family_name = "Itanium 2";
214 model_name = "McKinley";
218 * Deerfield is a low-voltage variant based on the
219 * Madison core. We need circumstantial evidence
220 * (i.e. the clock frequency) to identify those.
221 * Allow for roughly 1% error margin.
223 if (cpu_freq > 990 && cpu_freq < 1010)
224 model_name = "Deerfield";
226 model_name = "Madison";
229 model_name = "Madison II";
234 ia64_sync_icache_needed = 1;
236 family_name = "Itanium 2";
239 model_name = "Montecito";
242 model_name = "Montvale";
247 snprintf(cpu_family, sizeof(cpu_family), "%s", family_name);
248 snprintf(cpu_model, sizeof(cpu_model), "%s", model_name);
250 features = ia64_get_cpuid(4);
252 printf("CPU: %s (", model_name);
254 printf("%u MHz ", cpu_freq);
255 printf("%s)\n", family_name);
256 printf(" Origin = \"%s\" Revision = %d\n", vendor, revision);
257 printf(" Features = 0x%b\n", (u_int32_t) features,
259 "\001LB" /* long branch (brl) instruction. */
260 "\002SD" /* Spontaneous deferral. */
261 "\003AO" /* 16-byte atomic operations (ld, st, cmpxchg). */ );
265 cpu_startup(void *dummy)
269 struct pcpu_stats *pcs;
272 * Good {morning,afternoon,evening,night}.
279 printf("real memory = %ld (%ld MB)\n", ptoa(realmem),
280 ptoa(realmem) / 1048576);
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
304 ia64_pib = pmap_mapdev(ia64_lapic_addr, sizeof(*ia64_pib));
309 * Create sysctl tree for per-CPU information.
311 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
312 snprintf(nodename, sizeof(nodename), "%u", pc->pc_cpuid);
313 sysctl_ctx_init(&pc->pc_md.sysctl_ctx);
314 pc->pc_md.sysctl_tree = SYSCTL_ADD_NODE(&pc->pc_md.sysctl_ctx,
315 SYSCTL_STATIC_CHILDREN(_machdep_cpu), OID_AUTO, nodename,
316 CTLFLAG_RD, NULL, "");
317 if (pc->pc_md.sysctl_tree == NULL)
320 pcs = &pc->pc_md.stats;
322 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
323 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
324 "nasts", CTLFLAG_RD, &pcs->pcs_nasts,
325 "Number of IPI_AST interrupts");
327 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
328 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
329 "nclks", CTLFLAG_RD, &pcs->pcs_nclks,
330 "Number of clock interrupts");
332 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
333 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
334 "nextints", CTLFLAG_RD, &pcs->pcs_nextints,
335 "Number of ExtINT interrupts");
337 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
338 SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
339 "nhardclocks", CTLFLAG_RD, &pcs->pcs_nhardclocks,
340 "Number of IPI_HARDCLOCK 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 interrupts");
373 SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
376 cpu_flush_dcache(void *ptr, size_t len)
380 va = (uintptr_t)ptr & ~31;
381 lim = (uintptr_t)ptr + len;
390 /* Get current clock frequency for the given cpu id. */
392 cpu_est_clockrate(int cpu_id, uint64_t *rate)
395 if (pcpu_find(cpu_id) == NULL || rate == NULL)
397 *rate = (u_long)cpu_freq * 1000000ul;
416 sbt = cpu_idleclock();
420 KASSERT(ie != 0, ("%s called with interrupts disabled\n", __func__));
422 if (sched_runnable())
424 else if (cpu_idle_hook != NULL) {
425 (*cpu_idle_hook)(sbt);
426 /* The hook must enable interrupts! */
428 ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
439 cpu_idle_wakeup(int cpu)
453 cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx)
455 struct pcb *oldpcb, *newpcb;
457 oldpcb = old->td_pcb;
458 #ifdef COMPAT_FREEBSD32
459 ia32_savectx(oldpcb);
461 if (pcpup->pc_fpcurthread == old)
462 old->td_frame->tf_special.psr |= IA64_PSR_DFH;
463 if (!savectx(oldpcb)) {
464 newpcb = new->td_pcb;
465 oldpcb->pcb_current_pmap =
466 pmap_switch(newpcb->pcb_current_pmap);
470 atomic_store_rel_ptr(&old->td_lock, mtx);
472 #if defined(SCHED_ULE) && defined(SMP)
473 while (atomic_load_acq_ptr(&new->td_lock) == &blocked_lock)
477 pcpup->pc_curthread = new;
479 #ifdef COMPAT_FREEBSD32
480 ia32_restorectx(newpcb);
483 if (pcpup->pc_fpcurthread == new)
484 new->td_frame->tf_special.psr &= ~IA64_PSR_DFH;
486 /* We should not get here. */
487 panic("cpu_switch: restorectx() returned");
493 cpu_throw(struct thread *old __unused, struct thread *new)
497 newpcb = new->td_pcb;
498 (void)pmap_switch(newpcb->pcb_current_pmap);
500 #if defined(SCHED_ULE) && defined(SMP)
501 while (atomic_load_acq_ptr(&new->td_lock) == &blocked_lock)
505 pcpup->pc_curthread = new;
507 #ifdef COMPAT_FREEBSD32
508 ia32_restorectx(newpcb);
512 /* We should not get here. */
513 panic("cpu_throw: restorectx() returned");
518 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
522 * Set pc_acpi_id to "uninitialized".
523 * See sys/dev/acpica/acpi_cpu.c
525 pcpu->pc_acpi_id = 0xffffffff;
529 cpu_pcpu_setup(struct pcpu *pc, u_int acpi_id, u_int sapic_id)
532 pc->pc_acpi_id = acpi_id;
533 pc->pc_md.lid = IA64_LID_SET_SAPIC_ID(sapic_id);
543 if (td->td_md.md_spinlock_count == 0) {
544 intr = intr_disable();
545 td->td_md.md_spinlock_count = 1;
546 td->td_md.md_saved_intr = intr;
548 td->td_md.md_spinlock_count++;
560 intr = td->td_md.md_saved_intr;
561 td->td_md.md_spinlock_count--;
562 if (td->td_md.md_spinlock_count == 0)
567 kdb_cpu_trap(int vector, int code __unused)
573 __asm __volatile("flushrs;;");
575 /* Restart after the break instruction. */
576 if (vector == IA64_VEC_BREAK &&
577 kdb_frame->tf_special.ifa == IA64_FIXED_BREAK)
578 kdb_frame->tf_special.psr += IA64_PSR_RI_1;
582 map_vhpt(uintptr_t vhpt)
587 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
588 PTE_PL_KERN | PTE_AR_RW;
589 pte |= vhpt & PTE_PPN_MASK;
591 __asm __volatile("ptr.d %0,%1" :: "r"(vhpt),
592 "r"(pmap_vhpt_log2size << 2));
594 __asm __volatile("mov %0=psr" : "=r"(psr));
595 __asm __volatile("rsm psr.ic|psr.i");
598 ia64_set_itir(pmap_vhpt_log2size << 2);
600 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte));
601 __asm __volatile("mov psr.l=%0" :: "r" (psr));
614 if (ia64_pal_size == 0)
617 va = IA64_PHYS_TO_RR7(ia64_pal_base);
626 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
627 PTE_PL_KERN | PTE_AR_RWX;
628 pte |= ia64_pal_base & PTE_PPN_MASK;
630 __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" :: "r"(va), "r"(shft<<2));
632 __asm __volatile("mov %0=psr" : "=r"(psr));
633 __asm __volatile("rsm psr.ic|psr.i");
636 ia64_set_itir(shft << 2);
638 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(4), "r"(pte));
640 __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte));
641 __asm __volatile("mov psr.l=%0" :: "r" (psr));
646 map_gateway_page(void)
651 pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
652 PTE_PL_KERN | PTE_AR_X_RX;
653 pte |= ia64_tpa((uint64_t)ia64_gateway_page) & PTE_PPN_MASK;
655 __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
656 "r"(VM_MAXUSER_ADDRESS), "r"(PAGE_SHIFT << 2));
658 __asm __volatile("mov %0=psr" : "=r"(psr));
659 __asm __volatile("rsm psr.ic|psr.i");
661 ia64_set_ifa(VM_MAXUSER_ADDRESS);
662 ia64_set_itir(PAGE_SHIFT << 2);
664 __asm __volatile("itr.d dtr[%0]=%1" :: "r"(5), "r"(pte));
666 __asm __volatile("itr.i itr[%0]=%1" :: "r"(2), "r"(pte));
667 __asm __volatile("mov psr.l=%0" :: "r" (psr));
670 /* Expose the mapping to userland in ar.k5 */
671 ia64_set_k5(VM_MAXUSER_ADDRESS);
675 freq_ratio(u_long base, u_long ratio)
679 f = (base * (ratio >> 32)) / (ratio & 0xfffffffful);
680 return ((f + 500000) / 1000000);
684 calculate_frequencies(void)
686 struct ia64_sal_result sal;
687 struct ia64_pal_result pal;
691 sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
692 pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
695 if (sal.sal_status == 0 && pal.pal_status == 0) {
697 printf("Platform clock frequency %ld Hz\n",
699 printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
700 "ITC ratio %ld/%ld\n",
701 pal.pal_result[0] >> 32,
702 pal.pal_result[0] & ((1L << 32) - 1),
703 pal.pal_result[1] >> 32,
704 pal.pal_result[1] & ((1L << 32) - 1),
705 pal.pal_result[2] >> 32,
706 pal.pal_result[2] & ((1L << 32) - 1));
708 cpu_freq = freq_ratio(sal.sal_result[0], pal.pal_result[0]);
709 bus_freq = freq_ratio(sal.sal_result[0], pal.pal_result[1]);
710 itc_freq = freq_ratio(sal.sal_result[0], pal.pal_result[2]);
714 struct ia64_init_return
717 struct ia64_init_return ret;
719 pt_entry_t *pbvm_pgtbl_ent, *pbvm_pgtbl_lim;
722 int metadata_missing;
725 * NO OUTPUT ALLOWED UNTIL FURTHER NOTICE.
728 ia64_set_fpsr(IA64_FPSR_DEFAULT);
731 * Region 6 is direct mapped UC and region 7 is direct mapped
732 * WC. The details of this is controlled by the Alt {I,D}TLB
733 * handlers. Here we just make sure that they have the largest
734 * possible page size to minimise TLB usage.
736 ia64_set_rr(IA64_RR_BASE(6), (6 << 8) | (LOG2_ID_PAGE_SIZE << 2));
737 ia64_set_rr(IA64_RR_BASE(7), (7 << 8) | (LOG2_ID_PAGE_SIZE << 2));
740 /* Initialize/setup physical memory datastructures */
744 * Process the memory map. This gives us the PAL locations,
745 * the I/O port base address, the available memory regions
746 * for initializing the physical memory map.
748 for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
749 mdlen = md->md_pages * EFI_PAGE_SIZE;
750 switch (md->md_type) {
751 case EFI_MD_TYPE_IOPORT:
752 ia64_port_base = pmap_mapdev_priv(md->md_phys,
753 mdlen, VM_MEMATTR_UNCACHEABLE);
755 case EFI_MD_TYPE_PALCODE:
756 ia64_pal_base = md->md_phys;
757 ia64_pal_size = mdlen;
759 case EFI_MD_TYPE_BAD:
760 case EFI_MD_TYPE_FIRMWARE:
761 case EFI_MD_TYPE_RECLAIM:
762 case EFI_MD_TYPE_RT_CODE:
763 case EFI_MD_TYPE_RT_DATA:
764 /* Don't use these memory regions. */
765 ia64_physmem_track(md->md_phys, mdlen);
767 case EFI_MD_TYPE_BS_CODE:
768 case EFI_MD_TYPE_BS_DATA:
769 case EFI_MD_TYPE_CODE:
770 case EFI_MD_TYPE_DATA:
771 case EFI_MD_TYPE_FREE:
772 /* These are ok to use. */
773 ia64_physmem_add(md->md_phys, mdlen);
779 * Remove the PBVM and its page table from phys_avail. The loader
780 * passes the physical address of the page table to us. The virtual
781 * address of the page table is fixed.
782 * Track and the PBVM limit for later use.
784 ia64_physmem_delete(bootinfo->bi_pbvm_pgtbl, bootinfo->bi_pbvm_pgtblsz);
785 pbvm_pgtbl_ent = (void *)IA64_PBVM_PGTBL;
786 pbvm_pgtbl_lim = (void *)(IA64_PBVM_PGTBL + bootinfo->bi_pbvm_pgtblsz);
787 while (pbvm_pgtbl_ent < pbvm_pgtbl_lim) {
788 if ((*pbvm_pgtbl_ent & PTE_PRESENT) == 0)
790 ia64_physmem_delete(*pbvm_pgtbl_ent & PTE_PPN_MASK,
791 IA64_PBVM_PAGE_SIZE);
795 /* Finalize physical memory datastructures */
798 metadata_missing = 0;
799 if (bootinfo->bi_modulep)
800 preload_metadata = (caddr_t)bootinfo->bi_modulep;
802 metadata_missing = 1;
804 if (envmode == 0 && bootinfo->bi_envp)
805 kern_envp = (caddr_t)bootinfo->bi_envp;
807 kern_envp = static_env;
810 * Look at arguments passed to us and compute boothowto.
812 boothowto = bootinfo->bi_boothowto;
814 if (boothowto & RB_VERBOSE)
818 * Wire things up so we can call the firmware.
821 efi_boot_minimal(bootinfo->bi_systab);
824 calculate_frequencies();
826 set_cputicker(ia64_get_itc, (u_long)itc_freq * 1000000, 0);
829 * Setup the PCPU data for the bootstrap processor. It is needed
830 * by printf(). Also, since printf() has critical sections, we
831 * need to initialize at least pc_curthread.
834 ia64_set_k4((u_int64_t)pcpup);
835 pcpu_init(pcpup, 0, sizeof(pcpu0));
836 dpcpu_init(ia64_physmem_alloc(DPCPU_SIZE, PAGE_SIZE), 0);
837 cpu_pcpu_setup(pcpup, ~0U, ia64_get_lid());
838 pcpup->pc_curthread = &thread0;
841 * Initialize the console before we print anything out.
845 /* OUTPUT NOW ALLOWED */
847 if (metadata_missing)
848 printf("WARNING: loader(8) metadata is missing!\n");
850 /* Get FPSWA interface */
851 fpswa_iface = (bootinfo->bi_fpswa == 0) ? NULL :
852 (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo->bi_fpswa);
854 /* Init basic tunables, including hz */
857 p = getenv("kernelname");
859 strlcpy(kernelname, p, sizeof(kernelname));
863 init_param2(physmem);
866 * Initialize error message buffer (at end of core).
868 msgbufp = ia64_physmem_alloc(msgbufsize, PAGE_SIZE);
869 msgbufinit(msgbufp, msgbufsize);
871 proc_linkup0(&proc0, &thread0);
873 * Init mapping for kernel stack for proc 0
875 p = ia64_physmem_alloc(KSTACK_PAGES * PAGE_SIZE, PAGE_SIZE);
876 thread0.td_kstack = (uintptr_t)p;
877 thread0.td_kstack_pages = KSTACK_PAGES;
882 * Initialize the rest of proc 0's PCB.
884 * Set the kernel sp, reserving space for an (empty) trapframe,
885 * and make proc0's trapframe pointer point to it for sanity.
886 * Initialise proc0's backing store to start after u area.
888 cpu_thread_alloc(&thread0);
889 thread0.td_frame->tf_flags = FRAME_SYSCALL;
890 thread0.td_pcb->pcb_special.sp =
891 (u_int64_t)thread0.td_frame - 16;
892 thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack;
895 * Initialize the virtual memory system.
900 ia64_xtrace_init_bsp();
904 * Initialize debuggers, and break into them if appropriate.
907 ksym_start = bootinfo->bi_symtab;
908 ksym_end = bootinfo->bi_esymtab;
914 if (boothowto & RB_KDB)
915 kdb_enter(KDB_WHY_BOOTFLAGS,
916 "Boot flags requested debugger\n");
922 ret.bspstore = thread0.td_pcb->pcb_special.bspstore;
923 ret.sp = thread0.td_pcb->pcb_special.sp;
931 return (bootinfo->bi_hcdp);
935 bzero(void *buf, size_t len)
939 while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
943 while (len >= sizeof(u_long) * 8) {
945 *((u_long*) p + 1) = 0;
946 *((u_long*) p + 2) = 0;
947 *((u_long*) p + 3) = 0;
948 len -= sizeof(u_long) * 8;
949 *((u_long*) p + 4) = 0;
950 *((u_long*) p + 5) = 0;
951 *((u_long*) p + 6) = 0;
952 *((u_long*) p + 7) = 0;
953 p += sizeof(u_long) * 8;
955 while (len >= sizeof(u_long)) {
957 len -= sizeof(u_long);
976 u_int64_t start, end, now;
980 start = ia64_get_itc();
981 end = start + itc_freq * n;
982 /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
984 now = ia64_get_itc();
985 } while (now < end || (now > start && end < start));
991 * Send an interrupt (signal) to a process.
994 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
998 struct trapframe *tf;
1000 struct sigframe sf, *sfp;
1008 PROC_LOCK_ASSERT(p, MA_OWNED);
1009 sig = ksi->ksi_signo;
1010 code = ksi->ksi_code;
1012 mtx_assert(&psp->ps_mtx, MA_OWNED);
1014 sp = tf->tf_special.sp;
1015 oonstack = sigonstack(sp);
1018 /* save user context */
1019 bzero(&sf, sizeof(struct sigframe));
1020 sf.sf_uc.uc_sigmask = *mask;
1021 sf.sf_uc.uc_stack = td->td_sigstk;
1022 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
1023 ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1026 * Allocate and validate space for the signal handler
1027 * context. Note that if the stack is in P0 space, the
1028 * call to grow() is a nop, and the useracc() check
1029 * will fail if the process has not already allocated
1030 * the space with a `brk'.
1032 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
1033 SIGISMEMBER(psp->ps_sigonstack, sig)) {
1034 sbs = (u_int64_t)td->td_sigstk.ss_sp;
1035 sbs = (sbs + 15) & ~15;
1036 sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size);
1037 #if defined(COMPAT_43)
1038 td->td_sigstk.ss_flags |= SS_ONSTACK;
1041 sfp = (struct sigframe *)sp;
1042 sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15);
1044 /* Fill in the siginfo structure for POSIX handlers. */
1045 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
1046 sf.sf_si = ksi->ksi_info;
1047 sf.sf_si.si_signo = sig;
1049 * XXX this shouldn't be here after code in trap.c
1052 sf.sf_si.si_addr = (void*)tf->tf_special.ifa;
1053 code = (u_int64_t)&sfp->sf_si;
1056 mtx_unlock(&psp->ps_mtx);
1059 get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
1061 /* Copy the frame out to userland. */
1062 if (copyout(&sf, sfp, sizeof(sf)) != 0) {
1064 * Process has trashed its stack; give it an illegal
1065 * instruction to halt it in its tracks.
1068 sigexit(td, SIGILL);
1072 if ((tf->tf_flags & FRAME_SYSCALL) == 0) {
1073 tf->tf_special.psr &= ~IA64_PSR_RI;
1074 tf->tf_special.iip = ia64_get_k5() +
1075 ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
1077 tf->tf_special.iip = ia64_get_k5() +
1078 ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
1081 * Setup the trapframe to return to the signal trampoline. We pass
1082 * information to the trampoline in the following registers:
1084 * gp new backing store or NULL
1086 * r9 signal code or siginfo pointer
1087 * r10 signal handler (function descriptor)
1089 tf->tf_special.sp = (u_int64_t)sfp - 16;
1090 tf->tf_special.gp = sbs;
1091 tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore;
1092 tf->tf_special.ndirty = 0;
1093 tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat;
1094 tf->tf_scratch.gr8 = sig;
1095 tf->tf_scratch.gr9 = code;
1096 tf->tf_scratch.gr10 = (u_int64_t)catcher;
1099 mtx_lock(&psp->ps_mtx);
1103 * System call to cleanup state after a signal
1104 * has been taken. Reset signal mask and
1105 * stack state from context left by sendsig (above).
1106 * Return to previous pc and psl as specified by
1107 * context left by sendsig. Check carefully to
1108 * make sure that the user has not modified the
1109 * state to gain improper privileges.
1114 sys_sigreturn(struct thread *td,
1115 struct sigreturn_args /* {
1116 ucontext_t *sigcntxp;
1120 struct trapframe *tf;
1127 * Fetch the entire context structure at once for speed.
1128 * We don't use a normal argument to simplify RSE handling.
1130 if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc)))
1133 set_mcontext(td, &uc.uc_mcontext);
1135 #if defined(COMPAT_43)
1136 if (sigonstack(tf->tf_special.sp))
1137 td->td_sigstk.ss_flags |= SS_ONSTACK;
1139 td->td_sigstk.ss_flags &= ~SS_ONSTACK;
1141 kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
1143 return (EJUSTRETURN);
1146 #ifdef COMPAT_FREEBSD4
1148 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
1151 return sys_sigreturn(td, (struct sigreturn_args *)uap);
1156 * Construct a PCB from a trapframe. This is called from kdb_trap() where
1157 * we want to start a backtrace from the function that caused us to enter
1158 * the debugger. We have the context in the trapframe, but base the trace
1159 * on the PCB. The PCB doesn't have to be perfect, as long as it contains
1160 * enough for a backtrace.
1163 makectx(struct trapframe *tf, struct pcb *pcb)
1166 pcb->pcb_special = tf->tf_special;
1167 pcb->pcb_special.__spare = ~0UL; /* XXX see unwind.c */
1168 save_callee_saved(&pcb->pcb_preserved);
1169 save_callee_saved_fp(&pcb->pcb_preserved_fp);
1173 ia64_flush_dirty(struct thread *td, struct _special *r)
1177 uint64_t bspst, kstk, rnat;
1183 kstk = td->td_kstack + (r->bspstore & 0x1ffUL);
1184 if (td == curthread) {
1185 __asm __volatile("mov ar.rsc=0;;");
1186 __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst));
1187 /* Make sure we have all the user registers written out. */
1188 if (bspst - kstk < r->ndirty) {
1189 __asm __volatile("flushrs;;");
1190 __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst));
1192 __asm __volatile("mov %0=ar.rnat;;" : "=r"(rnat));
1193 __asm __volatile("mov ar.rsc=3");
1194 error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty);
1196 r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL))
1197 ? *(uint64_t*)(kstk | 0x1f8L) : rnat;
1199 locked = PROC_LOCKED(td->td_proc);
1202 iov.iov_base = (void*)(uintptr_t)kstk;
1203 iov.iov_len = r->ndirty;
1206 uio.uio_offset = r->bspstore;
1207 uio.uio_resid = r->ndirty;
1208 uio.uio_segflg = UIO_SYSSPACE;
1209 uio.uio_rw = UIO_WRITE;
1211 error = proc_rwmem(td->td_proc, &uio);
1213 * XXX proc_rwmem() doesn't currently return ENOSPC,
1214 * so I think it can bogusly return 0. Neither do
1215 * we allow short writes.
1217 if (uio.uio_resid != 0 && error == 0)
1223 r->bspstore += r->ndirty;
1229 get_mcontext(struct thread *td, mcontext_t *mc, int flags)
1231 struct trapframe *tf;
1235 bzero(mc, sizeof(*mc));
1236 mc->mc_special = tf->tf_special;
1237 error = ia64_flush_dirty(td, &mc->mc_special);
1238 if (tf->tf_flags & FRAME_SYSCALL) {
1239 mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT;
1240 mc->mc_scratch = tf->tf_scratch;
1241 if (flags & GET_MC_CLEAR_RET) {
1242 mc->mc_scratch.gr8 = 0;
1243 mc->mc_scratch.gr9 = 0;
1244 mc->mc_scratch.gr10 = 0;
1245 mc->mc_scratch.gr11 = 0;
1248 mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT;
1249 mc->mc_scratch = tf->tf_scratch;
1250 mc->mc_scratch_fp = tf->tf_scratch_fp;
1252 * XXX If the thread never used the high FP registers, we
1253 * probably shouldn't waste time saving them.
1255 ia64_highfp_save(td);
1256 mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID;
1257 mc->mc_high_fp = td->td_pcb->pcb_high_fp;
1259 save_callee_saved(&mc->mc_preserved);
1260 save_callee_saved_fp(&mc->mc_preserved_fp);
1265 set_mcontext(struct thread *td, const mcontext_t *mc)
1268 struct trapframe *tf;
1273 KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1274 ("Whoa there! We have more than 8KB of dirty registers!"));
1278 * Only copy the user mask and the restart instruction bit from
1281 psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL |
1282 IA64_PSR_MFH | IA64_PSR_RI;
1283 s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask);
1284 /* We don't have any dirty registers of the new context. */
1286 if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) {
1288 * We can get an async context passed to us while we
1289 * entered the kernel through a syscall: sigreturn(2)
1290 * takes contexts that could previously be the result of
1291 * a trap or interrupt.
1292 * Hence, we cannot assert that the trapframe is not
1293 * a syscall frame, but we can assert that it's at
1294 * least an expected syscall.
1296 if (tf->tf_flags & FRAME_SYSCALL) {
1297 KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo"));
1298 tf->tf_flags &= ~FRAME_SYSCALL;
1300 tf->tf_scratch = mc->mc_scratch;
1301 tf->tf_scratch_fp = mc->mc_scratch_fp;
1302 if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID)
1303 td->td_pcb->pcb_high_fp = mc->mc_high_fp;
1305 KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
1306 if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) {
1307 s.cfm = tf->tf_special.cfm;
1308 s.iip = tf->tf_special.iip;
1309 tf->tf_scratch.gr15 = 0; /* Clear syscall nr. */
1311 tf->tf_scratch = mc->mc_scratch;
1314 restore_callee_saved(&mc->mc_preserved);
1315 restore_callee_saved_fp(&mc->mc_preserved_fp);
1321 * Clear registers on exec.
1324 exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
1326 struct trapframe *tf;
1327 uint64_t *ksttop, *kst;
1330 ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty +
1331 (tf->tf_special.bspstore & 0x1ffUL));
1334 * We can ignore up to 8KB of dirty registers by masking off the
1335 * lower 13 bits in exception_restore() or epc_syscall(). This
1336 * should be enough for a couple of years, but if there are more
1337 * than 8KB of dirty registers, we lose track of the bottom of
1338 * the kernel stack. The solution is to copy the active part of
1339 * the kernel stack down 1 page (or 2, but not more than that)
1340 * so that we always have less than 8KB of dirty registers.
1342 KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1343 ("Whoa there! We have more than 8KB of dirty registers!"));
1345 bzero(&tf->tf_special, sizeof(tf->tf_special));
1346 if ((tf->tf_flags & FRAME_SYSCALL) == 0) { /* break syscalls. */
1347 bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
1348 bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
1349 tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
1350 tf->tf_special.bspstore = IA64_BACKINGSTORE;
1352 * Copy the arguments onto the kernel register stack so that
1353 * they get loaded by the loadrs instruction. Skip over the
1354 * NaT collection points.
1357 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1360 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1362 *kst-- = imgp->ps_strings;
1363 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1366 tf->tf_special.ndirty = (ksttop - kst) << 3;
1367 } else { /* epc syscalls (default). */
1368 tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
1369 tf->tf_special.bspstore = IA64_BACKINGSTORE + 24;
1371 * Write values for out0, out1 and out2 to the user's backing
1372 * store and arrange for them to be restored into the user's
1373 * initial register frame.
1374 * Assumes that (bspstore & 0x1f8) < 0x1e0.
1376 suword((caddr_t)tf->tf_special.bspstore - 24, stack);
1377 suword((caddr_t)tf->tf_special.bspstore - 16, imgp->ps_strings);
1378 suword((caddr_t)tf->tf_special.bspstore - 8, 0);
1381 tf->tf_special.iip = imgp->entry_addr;
1382 tf->tf_special.sp = (stack & ~15) - 16;
1383 tf->tf_special.rsc = 0xf;
1384 tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
1385 tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT |
1386 IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN |
1391 ptrace_set_pc(struct thread *td, unsigned long addr)
1395 switch (addr & 0xFUL) {
1397 slot = IA64_PSR_RI_0;
1400 /* XXX we need to deal with MLX bundles here */
1401 slot = IA64_PSR_RI_1;
1404 slot = IA64_PSR_RI_2;
1410 td->td_frame->tf_special.iip = addr & ~0x0FULL;
1411 td->td_frame->tf_special.psr =
1412 (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot;
1417 ptrace_single_step(struct thread *td)
1419 struct trapframe *tf;
1422 * There's no way to set single stepping when we're leaving the
1423 * kernel through the EPC syscall path. The way we solve this is
1424 * by enabling the lower-privilege trap so that we re-enter the
1425 * kernel as soon as the privilege level changes. See trap.c for
1426 * how we proceed from there.
1429 if (tf->tf_flags & FRAME_SYSCALL)
1430 tf->tf_special.psr |= IA64_PSR_LP;
1432 tf->tf_special.psr |= IA64_PSR_SS;
1437 ptrace_clear_single_step(struct thread *td)
1439 struct trapframe *tf;
1442 * Clear any and all status bits we may use to implement single
1446 tf->tf_special.psr &= ~IA64_PSR_SS;
1447 tf->tf_special.psr &= ~IA64_PSR_LP;
1448 tf->tf_special.psr &= ~IA64_PSR_TB;
1453 fill_regs(struct thread *td, struct reg *regs)
1455 struct trapframe *tf;
1458 regs->r_special = tf->tf_special;
1459 regs->r_scratch = tf->tf_scratch;
1460 save_callee_saved(®s->r_preserved);
1465 set_regs(struct thread *td, struct reg *regs)
1467 struct trapframe *tf;
1471 error = ia64_flush_dirty(td, &tf->tf_special);
1473 tf->tf_special = regs->r_special;
1474 tf->tf_special.bspstore += tf->tf_special.ndirty;
1475 tf->tf_special.ndirty = 0;
1476 tf->tf_scratch = regs->r_scratch;
1477 restore_callee_saved(®s->r_preserved);
1483 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1490 set_dbregs(struct thread *td, struct dbreg *dbregs)
1497 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1499 struct trapframe *frame = td->td_frame;
1500 struct pcb *pcb = td->td_pcb;
1502 /* Save the high FP registers. */
1503 ia64_highfp_save(td);
1505 fpregs->fpr_scratch = frame->tf_scratch_fp;
1506 save_callee_saved_fp(&fpregs->fpr_preserved);
1507 fpregs->fpr_high = pcb->pcb_high_fp;
1512 set_fpregs(struct thread *td, struct fpreg *fpregs)
1514 struct trapframe *frame = td->td_frame;
1515 struct pcb *pcb = td->td_pcb;
1517 /* Throw away the high FP registers (should be redundant). */
1518 ia64_highfp_drop(td);
1520 frame->tf_scratch_fp = fpregs->fpr_scratch;
1521 restore_callee_saved_fp(&fpregs->fpr_preserved);
1522 pcb->pcb_high_fp = fpregs->fpr_high;
1527 ia64_sync_icache(vm_offset_t va, vm_offset_t sz)
1531 if (!ia64_sync_icache_needed)