machdep.c revision 225617
1/*-
2 * Copyright (c) 2001 Jake Burkholder.
3 * Copyright (c) 1992 Terrence R. Lambert.
4 * Copyright (c) 1982, 1987, 1990 The Regents of the University of California.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * William Jolitz.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 *	from: @(#)machdep.c	7.4 (Berkeley) 6/3/91
35 *	from: FreeBSD: src/sys/i386/i386/machdep.c,v 1.477 2001/08/27
36 */
37
38#include <sys/cdefs.h>
39__FBSDID("$FreeBSD: head/sys/sparc64/sparc64/machdep.c 225617 2011-09-16 13:58:51Z kmacy $");
40
41#include "opt_compat.h"
42#include "opt_ddb.h"
43#include "opt_kstack_pages.h"
44
45#include <sys/param.h>
46#include <sys/malloc.h>
47#include <sys/proc.h>
48#include <sys/systm.h>
49#include <sys/bio.h>
50#include <sys/buf.h>
51#include <sys/bus.h>
52#include <sys/cpu.h>
53#include <sys/cons.h>
54#include <sys/eventhandler.h>
55#include <sys/exec.h>
56#include <sys/imgact.h>
57#include <sys/interrupt.h>
58#include <sys/kdb.h>
59#include <sys/kernel.h>
60#include <sys/ktr.h>
61#include <sys/linker.h>
62#include <sys/lock.h>
63#include <sys/msgbuf.h>
64#include <sys/mutex.h>
65#include <sys/pcpu.h>
66#include <sys/ptrace.h>
67#include <sys/reboot.h>
68#include <sys/signalvar.h>
69#include <sys/smp.h>
70#include <sys/syscallsubr.h>
71#include <sys/sysent.h>
72#include <sys/sysproto.h>
73#include <sys/timetc.h>
74#include <sys/ucontext.h>
75
76#include <dev/ofw/openfirm.h>
77
78#include <vm/vm.h>
79#include <vm/vm_extern.h>
80#include <vm/vm_kern.h>
81#include <vm/vm_page.h>
82#include <vm/vm_map.h>
83#include <vm/vm_object.h>
84#include <vm/vm_pager.h>
85#include <vm/vm_param.h>
86
87#include <ddb/ddb.h>
88
89#include <machine/bus.h>
90#include <machine/cache.h>
91#include <machine/clock.h>
92#include <machine/cmt.h>
93#include <machine/cpu.h>
94#include <machine/fireplane.h>
95#include <machine/fp.h>
96#include <machine/fsr.h>
97#include <machine/intr_machdep.h>
98#include <machine/jbus.h>
99#include <machine/md_var.h>
100#include <machine/metadata.h>
101#include <machine/ofw_machdep.h>
102#include <machine/ofw_mem.h>
103#include <machine/pcb.h>
104#include <machine/pmap.h>
105#include <machine/pstate.h>
106#include <machine/reg.h>
107#include <machine/sigframe.h>
108#include <machine/smp.h>
109#include <machine/tick.h>
110#include <machine/tlb.h>
111#include <machine/tstate.h>
112#include <machine/upa.h>
113#include <machine/ver.h>
114
115typedef int ofw_vec_t(void *);
116
117#ifdef DDB
118extern vm_offset_t ksym_start, ksym_end;
119#endif
120
121int dtlb_slots;
122int itlb_slots;
123struct tlb_entry *kernel_tlbs;
124int kernel_tlb_slots;
125
126int cold = 1;
127long Maxmem;
128long realmem;
129
130void *dpcpu0;
131char pcpu0[PCPU_PAGES * PAGE_SIZE];
132struct trapframe frame0;
133
134vm_offset_t kstack0;
135vm_paddr_t kstack0_phys;
136
137struct kva_md_info kmi;
138
139u_long ofw_vec;
140u_long ofw_tba;
141u_int tba_taken_over;
142
143char sparc64_model[32];
144
145static int cpu_use_vis = 1;
146
147cpu_block_copy_t *cpu_block_copy;
148cpu_block_zero_t *cpu_block_zero;
149
150static phandle_t find_bsp(phandle_t node, uint32_t bspid, u_int cpu_impl);
151void sparc64_init(caddr_t mdp, u_long o1, u_long o2, u_long o3,
152    ofw_vec_t *vec);
153static void sparc64_shutdown_final(void *dummy, int howto);
154
155static void cpu_startup(void *arg);
156SYSINIT(cpu, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
157
158CTASSERT((1 << INT_SHIFT) == sizeof(int));
159CTASSERT((1 << PTR_SHIFT) == sizeof(char *));
160
161CTASSERT(sizeof(struct reg) == 256);
162CTASSERT(sizeof(struct fpreg) == 272);
163CTASSERT(sizeof(struct __mcontext) == 512);
164
165CTASSERT((sizeof(struct pcb) & (64 - 1)) == 0);
166CTASSERT((offsetof(struct pcb, pcb_kfp) & (64 - 1)) == 0);
167CTASSERT((offsetof(struct pcb, pcb_ufp) & (64 - 1)) == 0);
168CTASSERT(sizeof(struct pcb) <= ((KSTACK_PAGES * PAGE_SIZE) / 8));
169
170CTASSERT(sizeof(struct pcpu) <= ((PCPU_PAGES * PAGE_SIZE) / 2));
171
172static void
173cpu_startup(void *arg)
174{
175	vm_paddr_t physsz;
176	int i;
177
178	physsz = 0;
179	for (i = 0; i < sparc64_nmemreg; i++)
180		physsz += sparc64_memreg[i].mr_size;
181	printf("real memory  = %lu (%lu MB)\n", physsz,
182	    physsz / (1024 * 1024));
183	realmem = (long)physsz / PAGE_SIZE;
184
185	vm_ksubmap_init(&kmi);
186
187	bufinit();
188	vm_pager_bufferinit();
189
190	EVENTHANDLER_REGISTER(shutdown_final, sparc64_shutdown_final, NULL,
191	    SHUTDOWN_PRI_LAST);
192
193	printf("avail memory = %lu (%lu MB)\n", cnt.v_free_count * PAGE_SIZE,
194	    cnt.v_free_count / ((1024 * 1024) / PAGE_SIZE));
195
196	if (bootverbose)
197		printf("machine: %s\n", sparc64_model);
198
199	cpu_identify(rdpr(ver), PCPU_GET(clock), curcpu);
200}
201
202void
203cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
204{
205	struct intr_request *ir;
206	int i;
207
208	pcpu->pc_irtail = &pcpu->pc_irhead;
209	for (i = 0; i < IR_FREE; i++) {
210		ir = &pcpu->pc_irpool[i];
211		ir->ir_next = pcpu->pc_irfree;
212		pcpu->pc_irfree = ir;
213	}
214}
215
216void
217spinlock_enter(void)
218{
219	struct thread *td;
220	register_t pil;
221
222	td = curthread;
223	if (td->td_md.md_spinlock_count == 0) {
224		pil = rdpr(pil);
225		wrpr(pil, 0, PIL_TICK);
226		td->td_md.md_spinlock_count = 1;
227		td->td_md.md_saved_pil = pil;
228	} else
229		td->td_md.md_spinlock_count++;
230	critical_enter();
231}
232
233void
234spinlock_exit(void)
235{
236	struct thread *td;
237	register_t pil;
238
239	td = curthread;
240	critical_exit();
241	pil = td->td_md.md_saved_pil;
242	td->td_md.md_spinlock_count--;
243	if (td->td_md.md_spinlock_count == 0)
244		wrpr(pil, pil, 0);
245}
246
247static phandle_t
248find_bsp(phandle_t node, uint32_t bspid, u_int cpu_impl)
249{
250	char type[sizeof("cpu")];
251	phandle_t child;
252	uint32_t cpuid;
253
254	for (; node != 0; node = OF_peer(node)) {
255		child = OF_child(node);
256		if (child > 0) {
257			child = find_bsp(child, bspid, cpu_impl);
258			if (child > 0)
259				return (child);
260		} else {
261			if (OF_getprop(node, "device_type", type,
262			    sizeof(type)) <= 0)
263				continue;
264			if (strcmp(type, "cpu") != 0)
265				continue;
266			if (OF_getprop(node, cpu_cpuid_prop(cpu_impl), &cpuid,
267			    sizeof(cpuid)) <= 0)
268				continue;
269			if (cpuid == bspid)
270				return (node);
271		}
272	}
273	return (0);
274}
275
276const char *
277cpu_cpuid_prop(u_int cpu_impl)
278{
279
280	switch (cpu_impl) {
281	case CPU_IMPL_SPARC64:
282	case CPU_IMPL_SPARC64V:
283	case CPU_IMPL_ULTRASPARCI:
284	case CPU_IMPL_ULTRASPARCII:
285	case CPU_IMPL_ULTRASPARCIIi:
286	case CPU_IMPL_ULTRASPARCIIe:
287		return ("upa-portid");
288	case CPU_IMPL_ULTRASPARCIII:
289	case CPU_IMPL_ULTRASPARCIIIp:
290	case CPU_IMPL_ULTRASPARCIIIi:
291	case CPU_IMPL_ULTRASPARCIIIip:
292		return ("portid");
293	case CPU_IMPL_ULTRASPARCIV:
294	case CPU_IMPL_ULTRASPARCIVp:
295		return ("cpuid");
296	default:
297		return ("");
298	}
299}
300
301uint32_t
302cpu_get_mid(u_int cpu_impl)
303{
304
305	switch (cpu_impl) {
306	case CPU_IMPL_SPARC64:
307	case CPU_IMPL_SPARC64V:
308	case CPU_IMPL_ULTRASPARCI:
309	case CPU_IMPL_ULTRASPARCII:
310	case CPU_IMPL_ULTRASPARCIIi:
311	case CPU_IMPL_ULTRASPARCIIe:
312		return (UPA_CR_GET_MID(ldxa(0, ASI_UPA_CONFIG_REG)));
313	case CPU_IMPL_ULTRASPARCIII:
314	case CPU_IMPL_ULTRASPARCIIIp:
315		return (FIREPLANE_CR_GET_AID(ldxa(AA_FIREPLANE_CONFIG,
316		    ASI_FIREPLANE_CONFIG_REG)));
317	case CPU_IMPL_ULTRASPARCIIIi:
318	case CPU_IMPL_ULTRASPARCIIIip:
319		return (JBUS_CR_GET_JID(ldxa(0, ASI_JBUS_CONFIG_REG)));
320	case CPU_IMPL_ULTRASPARCIV:
321	case CPU_IMPL_ULTRASPARCIVp:
322		return (INTR_ID_GET_ID(ldxa(AA_INTR_ID, ASI_INTR_ID)));
323	default:
324		return (0);
325	}
326}
327
328void
329sparc64_init(caddr_t mdp, u_long o1, u_long o2, u_long o3, ofw_vec_t *vec)
330{
331	char *env;
332	struct pcpu *pc;
333	vm_offset_t end;
334	vm_offset_t va;
335	caddr_t kmdp;
336	phandle_t root;
337	u_int cpu_impl;
338
339	end = 0;
340	kmdp = NULL;
341
342	/*
343	 * Find out what kind of CPU we have first, for anything that changes
344	 * behaviour.
345	 */
346	cpu_impl = VER_IMPL(rdpr(ver));
347
348	/*
349	 * Do CPU-specific initialization.
350	 */
351	if (cpu_impl >= CPU_IMPL_ULTRASPARCIII)
352		cheetah_init(cpu_impl);
353	else if (cpu_impl == CPU_IMPL_SPARC64V)
354		zeus_init(cpu_impl);
355
356	/*
357	 * Clear (S)TICK timer (including NPT).
358	 */
359	tick_clear(cpu_impl);
360
361	/*
362	 * UltraSparc II[e,i] based systems come up with the tick interrupt
363	 * enabled and a handler that resets the tick counter, causing DELAY()
364	 * to not work properly when used early in boot.
365	 * UltraSPARC III based systems come up with the system tick interrupt
366	 * enabled, causing an interrupt storm on startup since they are not
367	 * handled.
368	 */
369	tick_stop(cpu_impl);
370
371	/*
372	 * Set up Open Firmware entry points.
373	 */
374	ofw_tba = rdpr(tba);
375	ofw_vec = (u_long)vec;
376
377	/*
378	 * Parse metadata if present and fetch parameters.  Must be before the
379	 * console is inited so cninit gets the right value of boothowto.
380	 */
381	if (mdp != NULL) {
382		preload_metadata = mdp;
383		kmdp = preload_search_by_type("elf kernel");
384		if (kmdp != NULL) {
385			boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
386			kern_envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *);
387			end = MD_FETCH(kmdp, MODINFOMD_KERNEND, vm_offset_t);
388			kernel_tlb_slots = MD_FETCH(kmdp, MODINFOMD_DTLB_SLOTS,
389			    int);
390			kernel_tlbs = (void *)preload_search_info(kmdp,
391			    MODINFO_METADATA | MODINFOMD_DTLB);
392		}
393	}
394
395	init_param1();
396
397	/*
398	 * Initialize Open Firmware (needed for console).
399	 */
400	OF_install(OFW_STD_DIRECT, 0);
401	OF_init(ofw_entry);
402
403	/*
404	 * Prime our per-CPU data page for use.  Note, we are using it for
405	 * our stack, so don't pass the real size (PAGE_SIZE) to pcpu_init
406	 * or it'll zero it out from under us.
407	 */
408	pc = (struct pcpu *)(pcpu0 + (PCPU_PAGES * PAGE_SIZE)) - 1;
409	pcpu_init(pc, 0, sizeof(struct pcpu));
410	pc->pc_addr = (vm_offset_t)pcpu0;
411	pc->pc_impl = cpu_impl;
412	pc->pc_mid = cpu_get_mid(cpu_impl);
413	pc->pc_tlb_ctx = TLB_CTX_USER_MIN;
414	pc->pc_tlb_ctx_min = TLB_CTX_USER_MIN;
415	pc->pc_tlb_ctx_max = TLB_CTX_USER_MAX;
416
417	/*
418	 * Determine the OFW node and frequency of the BSP (and ensure the
419	 * BSP is in the device tree in the first place).
420	 */
421	root = OF_peer(0);
422	pc->pc_node = find_bsp(root, pc->pc_mid, cpu_impl);
423	if (pc->pc_node == 0)
424		OF_exit();
425	if (OF_getprop(pc->pc_node, "clock-frequency", &pc->pc_clock,
426	    sizeof(pc->pc_clock)) <= 0)
427		OF_exit();
428
429	/*
430	 * Provide a DELAY() that works before PCPU_REG is set.  We can't
431	 * set PCPU_REG without also taking over the trap table or the
432	 * firmware will overwrite it.  Unfortunately, it's way to early
433	 * to also take over the trap table at this point.
434	 */
435	clock_boot = pc->pc_clock;
436	delay_func = delay_boot;
437
438	/*
439	 * Initialize the console before printing anything.
440	 * NB: the low-level console drivers require a working DELAY() at
441	 * this point.
442	 */
443	cninit();
444
445	/*
446	 * Panic if there is no metadata.  Most likely the kernel was booted
447	 * directly, instead of through loader(8).
448	 */
449	if (mdp == NULL || kmdp == NULL || end == 0 ||
450	    kernel_tlb_slots == 0 || kernel_tlbs == NULL) {
451		printf("sparc64_init: missing loader metadata.\n"
452		    "This probably means you are not using loader(8).\n");
453		panic("sparc64_init");
454	}
455
456	/*
457	 * Work around the broken loader behavior of not demapping no
458	 * longer used kernel TLB slots when unloading the kernel or
459	 * modules.
460	 */
461	for (va = KERNBASE + (kernel_tlb_slots - 1) * PAGE_SIZE_4M;
462	    va >= roundup2(end, PAGE_SIZE_4M); va -= PAGE_SIZE_4M) {
463		if (bootverbose)
464			printf("demapping unused kernel TLB slot "
465			    "(va %#lx - %#lx)\n", va, va + PAGE_SIZE_4M - 1);
466		stxa(TLB_DEMAP_VA(va) | TLB_DEMAP_PRIMARY | TLB_DEMAP_PAGE,
467		    ASI_DMMU_DEMAP, 0);
468		stxa(TLB_DEMAP_VA(va) | TLB_DEMAP_PRIMARY | TLB_DEMAP_PAGE,
469		    ASI_IMMU_DEMAP, 0);
470		flush(KERNBASE);
471		kernel_tlb_slots--;
472	}
473
474	/*
475	 * Determine the TLB slot maxima, which are expected to be
476	 * equal across all CPUs.
477	 * NB: for cheetah-class CPUs, these properties only refer
478	 * to the t16s.
479	 */
480	if (OF_getprop(pc->pc_node, "#dtlb-entries", &dtlb_slots,
481	    sizeof(dtlb_slots)) == -1)
482		panic("sparc64_init: cannot determine number of dTLB slots");
483	if (OF_getprop(pc->pc_node, "#itlb-entries", &itlb_slots,
484	    sizeof(itlb_slots)) == -1)
485		panic("sparc64_init: cannot determine number of iTLB slots");
486
487	/*
488	 * Initialize and enable the caches.  Note that his may include
489	 * applying workarounds.
490	 */
491	cache_init(pc);
492	cache_enable(cpu_impl);
493	uma_set_align(pc->pc_cache.dc_linesize - 1);
494
495	cpu_block_copy = bcopy;
496	cpu_block_zero = bzero;
497	getenv_int("machdep.use_vis", &cpu_use_vis);
498	if (cpu_use_vis) {
499		switch (cpu_impl) {
500		case CPU_IMPL_SPARC64:
501		case CPU_IMPL_ULTRASPARCI:
502		case CPU_IMPL_ULTRASPARCII:
503		case CPU_IMPL_ULTRASPARCIIi:
504		case CPU_IMPL_ULTRASPARCIIe:
505		case CPU_IMPL_ULTRASPARCIII:	/* NB: we've disabled P$. */
506		case CPU_IMPL_ULTRASPARCIIIp:
507		case CPU_IMPL_ULTRASPARCIIIi:
508		case CPU_IMPL_ULTRASPARCIV:
509		case CPU_IMPL_ULTRASPARCIVp:
510		case CPU_IMPL_ULTRASPARCIIIip:
511			cpu_block_copy = spitfire_block_copy;
512			cpu_block_zero = spitfire_block_zero;
513			break;
514		case CPU_IMPL_SPARC64V:
515			cpu_block_copy = zeus_block_copy;
516			cpu_block_zero = zeus_block_zero;
517			break;
518		}
519	}
520
521#ifdef SMP
522	mp_init(cpu_impl);
523#endif
524
525	/*
526	 * Initialize virtual memory and calculate physmem.
527	 */
528	pmap_bootstrap(cpu_impl);
529
530	/*
531	 * Initialize tunables.
532	 */
533	init_param2(physmem);
534	env = getenv("kernelname");
535	if (env != NULL) {
536		strlcpy(kernelname, env, sizeof(kernelname));
537		freeenv(env);
538	}
539
540	/*
541	 * Initialize the interrupt tables.
542	 */
543	intr_init1();
544
545	/*
546	 * Initialize proc0, set kstack0, frame0, curthread and curpcb.
547	 */
548	proc_linkup0(&proc0, &thread0);
549	proc0.p_md.md_sigtramp = NULL;
550	proc0.p_md.md_utrap = NULL;
551	thread0.td_kstack = kstack0;
552	thread0.td_kstack_pages = KSTACK_PAGES;
553	thread0.td_pcb = (struct pcb *)
554	    (thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
555	frame0.tf_tstate = TSTATE_IE | TSTATE_PEF | TSTATE_PRIV;
556	thread0.td_frame = &frame0;
557	pc->pc_curthread = &thread0;
558	pc->pc_curpcb = thread0.td_pcb;
559
560	/*
561	 * Initialize global registers.
562	 */
563	cpu_setregs(pc);
564
565	/*
566	 * Take over the trap table via the PROM.  Using the PROM for this
567	 * is necessary in order to set obp-control-relinquished to true
568	 * within the PROM so obtaining /virtual-memory/translations doesn't
569	 * trigger a fatal reset error or worse things further down the road.
570	 * XXX it should be possible to use this solely instead of writing
571	 * %tba in cpu_setregs().  Doing so causes a hang however.
572	 */
573	sun4u_set_traptable(tl0_base);
574
575	/*
576	 * It's now safe to use the real DELAY().
577	 */
578	delay_func = delay_tick;
579
580	/*
581	 * Initialize the dynamic per-CPU area for the BSP and the message
582	 * buffer (after setting the trap table).
583	 */
584	dpcpu_init(dpcpu0, 0);
585	msgbufinit(msgbufp, msgbufsize);
586
587	/*
588	 * Initialize mutexes.
589	 */
590	mutex_init();
591
592	/*
593	 * Finish the interrupt initialization now that mutexes work and
594	 * enable them.
595	 */
596	intr_init2();
597	wrpr(pil, 0, 0);
598	wrpr(pstate, 0, PSTATE_KERNEL);
599
600	/*
601	 * Finish pmap initialization now that we're ready for mutexes.
602	 */
603	PMAP_LOCK_INIT(kernel_pmap);
604
605	OF_getprop(root, "name", sparc64_model, sizeof(sparc64_model) - 1);
606
607	kdb_init();
608
609#ifdef KDB
610	if (boothowto & RB_KDB)
611		kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger");
612#endif
613}
614
615void
616sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
617{
618	struct trapframe *tf;
619	struct sigframe *sfp;
620	struct sigacts *psp;
621	struct sigframe sf;
622	struct thread *td;
623	struct frame *fp;
624	struct proc *p;
625	u_long sp;
626	int oonstack;
627	int sig;
628
629	oonstack = 0;
630	td = curthread;
631	p = td->td_proc;
632	PROC_LOCK_ASSERT(p, MA_OWNED);
633	sig = ksi->ksi_signo;
634	psp = p->p_sigacts;
635	mtx_assert(&psp->ps_mtx, MA_OWNED);
636	tf = td->td_frame;
637	sp = tf->tf_sp + SPOFF;
638	oonstack = sigonstack(sp);
639
640	CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm,
641	    catcher, sig);
642
643	/* Make sure we have a signal trampoline to return to. */
644	if (p->p_md.md_sigtramp == NULL) {
645		/*
646		 * No signal trampoline... kill the process.
647		 */
648		CTR0(KTR_SIG, "sendsig: no sigtramp");
649		printf("sendsig: %s is too old, rebuild it\n", p->p_comm);
650		sigexit(td, sig);
651		/* NOTREACHED */
652	}
653
654	/* Save user context. */
655	bzero(&sf, sizeof(sf));
656	get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
657	sf.sf_uc.uc_sigmask = *mask;
658	sf.sf_uc.uc_stack = td->td_sigstk;
659	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ?
660	    ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
661
662	/* Allocate and validate space for the signal handler context. */
663	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
664	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
665		sfp = (struct sigframe *)(td->td_sigstk.ss_sp +
666		    td->td_sigstk.ss_size - sizeof(struct sigframe));
667	} else
668		sfp = (struct sigframe *)sp - 1;
669	mtx_unlock(&psp->ps_mtx);
670	PROC_UNLOCK(p);
671
672	fp = (struct frame *)sfp - 1;
673
674	/* Translate the signal if appropriate. */
675	if (p->p_sysent->sv_sigtbl && sig <= p->p_sysent->sv_sigsize)
676		sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
677
678	/* Build the argument list for the signal handler. */
679	tf->tf_out[0] = sig;
680	tf->tf_out[2] = (register_t)&sfp->sf_uc;
681	tf->tf_out[4] = (register_t)catcher;
682	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
683		/* Signal handler installed with SA_SIGINFO. */
684		tf->tf_out[1] = (register_t)&sfp->sf_si;
685
686		/* Fill in POSIX parts. */
687		sf.sf_si = ksi->ksi_info;
688		sf.sf_si.si_signo = sig; /* maybe a translated signal */
689	} else {
690		/* Old FreeBSD-style arguments. */
691		tf->tf_out[1] = ksi->ksi_code;
692		tf->tf_out[3] = (register_t)ksi->ksi_addr;
693	}
694
695	/* Copy the sigframe out to the user's stack. */
696	if (rwindow_save(td) != 0 || copyout(&sf, sfp, sizeof(*sfp)) != 0 ||
697	    suword(&fp->fr_in[6], tf->tf_out[6]) != 0) {
698		/*
699		 * Something is wrong with the stack pointer.
700		 * ...Kill the process.
701		 */
702		CTR2(KTR_SIG, "sendsig: sigexit td=%p sfp=%p", td, sfp);
703		PROC_LOCK(p);
704		sigexit(td, SIGILL);
705		/* NOTREACHED */
706	}
707
708	tf->tf_tpc = (u_long)p->p_md.md_sigtramp;
709	tf->tf_tnpc = tf->tf_tpc + 4;
710	tf->tf_sp = (u_long)fp - SPOFF;
711
712	CTR3(KTR_SIG, "sendsig: return td=%p pc=%#lx sp=%#lx", td, tf->tf_tpc,
713	    tf->tf_sp);
714
715	PROC_LOCK(p);
716	mtx_lock(&psp->ps_mtx);
717}
718
719#ifndef	_SYS_SYSPROTO_H_
720struct sigreturn_args {
721	ucontext_t *ucp;
722};
723#endif
724
725/*
726 * MPSAFE
727 */
728int
729sys_sigreturn(struct thread *td, struct sigreturn_args *uap)
730{
731	struct proc *p;
732	mcontext_t *mc;
733	ucontext_t uc;
734	int error;
735
736	p = td->td_proc;
737	if (rwindow_save(td)) {
738		PROC_LOCK(p);
739		sigexit(td, SIGILL);
740	}
741
742	CTR2(KTR_SIG, "sigreturn: td=%p ucp=%p", td, uap->sigcntxp);
743	if (copyin(uap->sigcntxp, &uc, sizeof(uc)) != 0) {
744		CTR1(KTR_SIG, "sigreturn: efault td=%p", td);
745		return (EFAULT);
746	}
747
748	mc = &uc.uc_mcontext;
749	error = set_mcontext(td, mc);
750	if (error != 0)
751		return (error);
752
753	kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
754
755	CTR4(KTR_SIG, "sigreturn: return td=%p pc=%#lx sp=%#lx tstate=%#lx",
756	    td, mc->mc_tpc, mc->mc_sp, mc->mc_tstate);
757	return (EJUSTRETURN);
758}
759
760/*
761 * Construct a PCB from a trapframe. This is called from kdb_trap() where
762 * we want to start a backtrace from the function that caused us to enter
763 * the debugger. We have the context in the trapframe, but base the trace
764 * on the PCB. The PCB doesn't have to be perfect, as long as it contains
765 * enough for a backtrace.
766 */
767void
768makectx(struct trapframe *tf, struct pcb *pcb)
769{
770
771	pcb->pcb_pc = tf->tf_tpc;
772	pcb->pcb_sp = tf->tf_sp;
773}
774
775int
776get_mcontext(struct thread *td, mcontext_t *mc, int flags)
777{
778	struct trapframe *tf;
779	struct pcb *pcb;
780
781	tf = td->td_frame;
782	pcb = td->td_pcb;
783	/*
784	 * Copy the registers which will be restored by tl0_ret() from the
785	 * trapframe.
786	 * Note that we skip %g7 which is used as the userland TLS register
787	 * and %wstate.
788	 */
789	mc->mc_flags = _MC_VERSION;
790	mc->mc_global[1] = tf->tf_global[1];
791	mc->mc_global[2] = tf->tf_global[2];
792	mc->mc_global[3] = tf->tf_global[3];
793	mc->mc_global[4] = tf->tf_global[4];
794	mc->mc_global[5] = tf->tf_global[5];
795	mc->mc_global[6] = tf->tf_global[6];
796	if (flags & GET_MC_CLEAR_RET) {
797		mc->mc_out[0] = 0;
798		mc->mc_out[1] = 0;
799	} else {
800		mc->mc_out[0] = tf->tf_out[0];
801		mc->mc_out[1] = tf->tf_out[1];
802	}
803	mc->mc_out[2] = tf->tf_out[2];
804	mc->mc_out[3] = tf->tf_out[3];
805	mc->mc_out[4] = tf->tf_out[4];
806	mc->mc_out[5] = tf->tf_out[5];
807	mc->mc_out[6] = tf->tf_out[6];
808	mc->mc_out[7] = tf->tf_out[7];
809	mc->mc_fprs = tf->tf_fprs;
810	mc->mc_fsr = tf->tf_fsr;
811	mc->mc_gsr = tf->tf_gsr;
812	mc->mc_tnpc = tf->tf_tnpc;
813	mc->mc_tpc = tf->tf_tpc;
814	mc->mc_tstate = tf->tf_tstate;
815	mc->mc_y = tf->tf_y;
816	critical_enter();
817	if ((tf->tf_fprs & FPRS_FEF) != 0) {
818		savefpctx(pcb->pcb_ufp);
819		tf->tf_fprs &= ~FPRS_FEF;
820		pcb->pcb_flags |= PCB_FEF;
821	}
822	if ((pcb->pcb_flags & PCB_FEF) != 0) {
823		bcopy(pcb->pcb_ufp, mc->mc_fp, sizeof(mc->mc_fp));
824		mc->mc_fprs |= FPRS_FEF;
825	}
826	critical_exit();
827	return (0);
828}
829
830int
831set_mcontext(struct thread *td, const mcontext_t *mc)
832{
833	struct trapframe *tf;
834	struct pcb *pcb;
835
836	if (!TSTATE_SECURE(mc->mc_tstate) ||
837	    (mc->mc_flags & ((1L << _MC_VERSION_BITS) - 1)) != _MC_VERSION)
838		return (EINVAL);
839	tf = td->td_frame;
840	pcb = td->td_pcb;
841	/* Make sure the windows are spilled first. */
842	flushw();
843	/*
844	 * Copy the registers which will be restored by tl0_ret() to the
845	 * trapframe.
846	 * Note that we skip %g7 which is used as the userland TLS register
847	 * and %wstate.
848	 */
849	tf->tf_global[1] = mc->mc_global[1];
850	tf->tf_global[2] = mc->mc_global[2];
851	tf->tf_global[3] = mc->mc_global[3];
852	tf->tf_global[4] = mc->mc_global[4];
853	tf->tf_global[5] = mc->mc_global[5];
854	tf->tf_global[6] = mc->mc_global[6];
855	tf->tf_out[0] = mc->mc_out[0];
856	tf->tf_out[1] = mc->mc_out[1];
857	tf->tf_out[2] = mc->mc_out[2];
858	tf->tf_out[3] = mc->mc_out[3];
859	tf->tf_out[4] = mc->mc_out[4];
860	tf->tf_out[5] = mc->mc_out[5];
861	tf->tf_out[6] = mc->mc_out[6];
862	tf->tf_out[7] = mc->mc_out[7];
863	tf->tf_fprs = mc->mc_fprs;
864	tf->tf_fsr = mc->mc_fsr;
865	tf->tf_gsr = mc->mc_gsr;
866	tf->tf_tnpc = mc->mc_tnpc;
867	tf->tf_tpc = mc->mc_tpc;
868	tf->tf_tstate = mc->mc_tstate;
869	tf->tf_y = mc->mc_y;
870	if ((mc->mc_fprs & FPRS_FEF) != 0) {
871		tf->tf_fprs = 0;
872		bcopy(mc->mc_fp, pcb->pcb_ufp, sizeof(pcb->pcb_ufp));
873		pcb->pcb_flags |= PCB_FEF;
874	}
875	return (0);
876}
877
878/*
879 * Exit the kernel and execute a firmware call that will not return, as
880 * specified by the arguments.
881 */
882void
883cpu_shutdown(void *args)
884{
885
886#ifdef SMP
887	cpu_mp_shutdown();
888#endif
889	ofw_exit(args);
890}
891
892/*
893 * Flush the D-cache for non-DMA I/O so that the I-cache can
894 * be made coherent later.
895 */
896void
897cpu_flush_dcache(void *ptr, size_t len)
898{
899
900	/* TBD */
901}
902
903/* Get current clock frequency for the given CPU ID. */
904int
905cpu_est_clockrate(int cpu_id, uint64_t *rate)
906{
907	struct pcpu *pc;
908
909	pc = pcpu_find(cpu_id);
910	if (pc == NULL || rate == NULL)
911		return (EINVAL);
912	*rate = pc->pc_clock;
913	return (0);
914}
915
916/*
917 * Duplicate OF_exit() with a different firmware call function that restores
918 * the trap table, otherwise a RED state exception is triggered in at least
919 * some firmware versions.
920 */
921void
922cpu_halt(void)
923{
924	static struct {
925		cell_t name;
926		cell_t nargs;
927		cell_t nreturns;
928	} args = {
929		(cell_t)"exit",
930		0,
931		0
932	};
933
934	cpu_shutdown(&args);
935}
936
937static void
938sparc64_shutdown_final(void *dummy, int howto)
939{
940	static struct {
941		cell_t name;
942		cell_t nargs;
943		cell_t nreturns;
944	} args = {
945		(cell_t)"SUNW,power-off",
946		0,
947		0
948	};
949
950	/* Turn the power off? */
951	if ((howto & RB_POWEROFF) != 0)
952		cpu_shutdown(&args);
953	/* In case of halt, return to the firmware. */
954	if ((howto & RB_HALT) != 0)
955		cpu_halt();
956}
957
958void
959cpu_idle(int busy)
960{
961
962	/* Insert code to halt (until next interrupt) for the idle loop. */
963}
964
965int
966cpu_idle_wakeup(int cpu)
967{
968
969	return (1);
970}
971
972int
973ptrace_set_pc(struct thread *td, u_long addr)
974{
975
976	td->td_frame->tf_tpc = addr;
977	td->td_frame->tf_tnpc = addr + 4;
978	return (0);
979}
980
981int
982ptrace_single_step(struct thread *td)
983{
984
985	/* TODO; */
986	return (0);
987}
988
989int
990ptrace_clear_single_step(struct thread *td)
991{
992
993	/* TODO; */
994	return (0);
995}
996
997void
998exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
999{
1000	struct trapframe *tf;
1001	struct pcb *pcb;
1002	struct proc *p;
1003	u_long sp;
1004
1005	/* XXX no cpu_exec */
1006	p = td->td_proc;
1007	p->p_md.md_sigtramp = NULL;
1008	if (p->p_md.md_utrap != NULL) {
1009		utrap_free(p->p_md.md_utrap);
1010		p->p_md.md_utrap = NULL;
1011	}
1012
1013	pcb = td->td_pcb;
1014	tf = td->td_frame;
1015	sp = rounddown(stack, 16);
1016	bzero(pcb, sizeof(*pcb));
1017	bzero(tf, sizeof(*tf));
1018	tf->tf_out[0] = stack;
1019	tf->tf_out[3] = p->p_sysent->sv_psstrings;
1020	tf->tf_out[6] = sp - SPOFF - sizeof(struct frame);
1021	tf->tf_tnpc = imgp->entry_addr + 4;
1022	tf->tf_tpc = imgp->entry_addr;
1023	tf->tf_tstate = TSTATE_IE | TSTATE_PEF | TSTATE_MM_TSO;
1024
1025	td->td_retval[0] = tf->tf_out[0];
1026	td->td_retval[1] = tf->tf_out[1];
1027}
1028
1029int
1030fill_regs(struct thread *td, struct reg *regs)
1031{
1032
1033	bcopy(td->td_frame, regs, sizeof(*regs));
1034	return (0);
1035}
1036
1037int
1038set_regs(struct thread *td, struct reg *regs)
1039{
1040	struct trapframe *tf;
1041
1042	if (!TSTATE_SECURE(regs->r_tstate))
1043		return (EINVAL);
1044	tf = td->td_frame;
1045	regs->r_wstate = tf->tf_wstate;
1046	bcopy(regs, tf, sizeof(*regs));
1047	return (0);
1048}
1049
1050int
1051fill_dbregs(struct thread *td, struct dbreg *dbregs)
1052{
1053
1054	return (ENOSYS);
1055}
1056
1057int
1058set_dbregs(struct thread *td, struct dbreg *dbregs)
1059{
1060
1061	return (ENOSYS);
1062}
1063
1064int
1065fill_fpregs(struct thread *td, struct fpreg *fpregs)
1066{
1067	struct trapframe *tf;
1068	struct pcb *pcb;
1069
1070	pcb = td->td_pcb;
1071	tf = td->td_frame;
1072	bcopy(pcb->pcb_ufp, fpregs->fr_regs, sizeof(fpregs->fr_regs));
1073	fpregs->fr_fsr = tf->tf_fsr;
1074	fpregs->fr_gsr = tf->tf_gsr;
1075	return (0);
1076}
1077
1078int
1079set_fpregs(struct thread *td, struct fpreg *fpregs)
1080{
1081	struct trapframe *tf;
1082	struct pcb *pcb;
1083
1084	pcb = td->td_pcb;
1085	tf = td->td_frame;
1086	tf->tf_fprs &= ~FPRS_FEF;
1087	bcopy(fpregs->fr_regs, pcb->pcb_ufp, sizeof(pcb->pcb_ufp));
1088	tf->tf_fsr = fpregs->fr_fsr;
1089	tf->tf_gsr = fpregs->fr_gsr;
1090	return (0);
1091}
1092
1093struct md_utrap *
1094utrap_alloc(void)
1095{
1096	struct md_utrap *ut;
1097
1098	ut = malloc(sizeof(struct md_utrap), M_SUBPROC, M_WAITOK | M_ZERO);
1099	ut->ut_refcnt = 1;
1100	return (ut);
1101}
1102
1103void
1104utrap_free(struct md_utrap *ut)
1105{
1106	int refcnt;
1107
1108	if (ut == NULL)
1109		return;
1110	mtx_pool_lock(mtxpool_sleep, ut);
1111	ut->ut_refcnt--;
1112	refcnt = ut->ut_refcnt;
1113	mtx_pool_unlock(mtxpool_sleep, ut);
1114	if (refcnt == 0)
1115		free(ut, M_SUBPROC);
1116}
1117
1118struct md_utrap *
1119utrap_hold(struct md_utrap *ut)
1120{
1121
1122	if (ut == NULL)
1123		return (NULL);
1124	mtx_pool_lock(mtxpool_sleep, ut);
1125	ut->ut_refcnt++;
1126	mtx_pool_unlock(mtxpool_sleep, ut);
1127	return (ut);
1128}
1129