machdep.c revision 294683
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: stable/10/sys/sparc64/sparc64/machdep.c 294683 2016-01-24 21:04:06Z ian $");
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/rwlock.h>
69#include <sys/signalvar.h>
70#include <sys/smp.h>
71#include <sys/syscallsubr.h>
72#include <sys/sysent.h>
73#include <sys/sysproto.h>
74#include <sys/timetc.h>
75#include <sys/ucontext.h>
76
77#include <dev/ofw/openfirm.h>
78
79#include <vm/vm.h>
80#include <vm/vm_extern.h>
81#include <vm/vm_kern.h>
82#include <vm/vm_page.h>
83#include <vm/vm_map.h>
84#include <vm/vm_object.h>
85#include <vm/vm_pager.h>
86#include <vm/vm_param.h>
87
88#include <ddb/ddb.h>
89
90#include <machine/bus.h>
91#include <machine/cache.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 portid;
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_portid_prop(cpu_impl),
267			    &portid, sizeof(portid)) <= 0)
268				continue;
269			if (portid == bspid)
270				return (node);
271		}
272	}
273	return (0);
274}
275
276const char *
277cpu_portid_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			init_static_kenv(MD_FETCH(kmdp, MODINFOMD_ENVP, char *),
387			    0);
388			end = MD_FETCH(kmdp, MODINFOMD_KERNEND, vm_offset_t);
389			kernel_tlb_slots = MD_FETCH(kmdp, MODINFOMD_DTLB_SLOTS,
390			    int);
391			kernel_tlbs = (void *)preload_search_info(kmdp,
392			    MODINFO_METADATA | MODINFOMD_DTLB);
393		}
394	}
395
396	init_param1();
397
398	/*
399	 * Initialize Open Firmware (needed for console).
400	 */
401	OF_install(OFW_STD_DIRECT, 0);
402	OF_init(ofw_entry);
403
404	/*
405	 * Prime our per-CPU data page for use.  Note, we are using it for
406	 * our stack, so don't pass the real size (PAGE_SIZE) to pcpu_init
407	 * or it'll zero it out from under us.
408	 */
409	pc = (struct pcpu *)(pcpu0 + (PCPU_PAGES * PAGE_SIZE)) - 1;
410	pcpu_init(pc, 0, sizeof(struct pcpu));
411	pc->pc_addr = (vm_offset_t)pcpu0;
412	pc->pc_impl = cpu_impl;
413	pc->pc_mid = cpu_get_mid(cpu_impl);
414	pc->pc_tlb_ctx = TLB_CTX_USER_MIN;
415	pc->pc_tlb_ctx_min = TLB_CTX_USER_MIN;
416	pc->pc_tlb_ctx_max = TLB_CTX_USER_MAX;
417
418	/*
419	 * Determine the OFW node and frequency of the BSP (and ensure the
420	 * BSP is in the device tree in the first place).
421	 */
422	root = OF_peer(0);
423	pc->pc_node = find_bsp(root, pc->pc_mid, cpu_impl);
424	if (pc->pc_node == 0)
425		OF_panic("%s: cannot find boot CPU node", __func__);
426	if (OF_getprop(pc->pc_node, "clock-frequency", &pc->pc_clock,
427	    sizeof(pc->pc_clock)) <= 0)
428		OF_panic("%s: cannot determine boot CPU clock", __func__);
429
430	/*
431	 * Panic if there is no metadata.  Most likely the kernel was booted
432	 * directly, instead of through loader(8).
433	 */
434	if (mdp == NULL || kmdp == NULL || end == 0 ||
435	    kernel_tlb_slots == 0 || kernel_tlbs == NULL)
436		OF_panic("%s: missing loader metadata.\nThis probably means "
437		    "you are not using loader(8).", __func__);
438
439	/*
440	 * Work around the broken loader behavior of not demapping no
441	 * longer used kernel TLB slots when unloading the kernel or
442	 * modules.
443	 */
444	for (va = KERNBASE + (kernel_tlb_slots - 1) * PAGE_SIZE_4M;
445	    va >= roundup2(end, PAGE_SIZE_4M); va -= PAGE_SIZE_4M) {
446		if (bootverbose)
447			OF_printf("demapping unused kernel TLB slot "
448			    "(va %#lx - %#lx)\n", va, va + PAGE_SIZE_4M - 1);
449		stxa(TLB_DEMAP_VA(va) | TLB_DEMAP_PRIMARY | TLB_DEMAP_PAGE,
450		    ASI_DMMU_DEMAP, 0);
451		stxa(TLB_DEMAP_VA(va) | TLB_DEMAP_PRIMARY | TLB_DEMAP_PAGE,
452		    ASI_IMMU_DEMAP, 0);
453		flush(KERNBASE);
454		kernel_tlb_slots--;
455	}
456
457	/*
458	 * Determine the TLB slot maxima, which are expected to be
459	 * equal across all CPUs.
460	 * NB: for cheetah-class CPUs, these properties only refer
461	 * to the t16s.
462	 */
463	if (OF_getprop(pc->pc_node, "#dtlb-entries", &dtlb_slots,
464	    sizeof(dtlb_slots)) == -1)
465		OF_panic("%s: cannot determine number of dTLB slots",
466		    __func__);
467	if (OF_getprop(pc->pc_node, "#itlb-entries", &itlb_slots,
468	    sizeof(itlb_slots)) == -1)
469		OF_panic("%s: cannot determine number of iTLB slots",
470		    __func__);
471
472	/*
473	 * Initialize and enable the caches.  Note that this may include
474	 * applying workarounds.
475	 */
476	cache_init(pc);
477	cache_enable(cpu_impl);
478	uma_set_align(pc->pc_cache.dc_linesize - 1);
479
480	cpu_block_copy = bcopy;
481	cpu_block_zero = bzero;
482	getenv_int("machdep.use_vis", &cpu_use_vis);
483	if (cpu_use_vis) {
484		switch (cpu_impl) {
485		case CPU_IMPL_SPARC64:
486		case CPU_IMPL_ULTRASPARCI:
487		case CPU_IMPL_ULTRASPARCII:
488		case CPU_IMPL_ULTRASPARCIIi:
489		case CPU_IMPL_ULTRASPARCIIe:
490		case CPU_IMPL_ULTRASPARCIII:	/* NB: we've disabled P$. */
491		case CPU_IMPL_ULTRASPARCIIIp:
492		case CPU_IMPL_ULTRASPARCIIIi:
493		case CPU_IMPL_ULTRASPARCIV:
494		case CPU_IMPL_ULTRASPARCIVp:
495		case CPU_IMPL_ULTRASPARCIIIip:
496			cpu_block_copy = spitfire_block_copy;
497			cpu_block_zero = spitfire_block_zero;
498			break;
499		case CPU_IMPL_SPARC64V:
500			cpu_block_copy = zeus_block_copy;
501			cpu_block_zero = zeus_block_zero;
502			break;
503		}
504	}
505
506#ifdef SMP
507	mp_init();
508#endif
509
510	/*
511	 * Initialize virtual memory and calculate physmem.
512	 */
513	pmap_bootstrap(cpu_impl);
514
515	/*
516	 * Initialize tunables.
517	 */
518	init_param2(physmem);
519	env = getenv("kernelname");
520	if (env != NULL) {
521		strlcpy(kernelname, env, sizeof(kernelname));
522		freeenv(env);
523	}
524
525	/*
526	 * Initialize the interrupt tables.
527	 */
528	intr_init1();
529
530	/*
531	 * Initialize proc0, set kstack0, frame0, curthread and curpcb.
532	 */
533	proc_linkup0(&proc0, &thread0);
534	proc0.p_md.md_sigtramp = NULL;
535	proc0.p_md.md_utrap = NULL;
536	thread0.td_kstack = kstack0;
537	thread0.td_kstack_pages = KSTACK_PAGES;
538	thread0.td_pcb = (struct pcb *)
539	    (thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
540	frame0.tf_tstate = TSTATE_IE | TSTATE_PEF | TSTATE_PRIV;
541	thread0.td_frame = &frame0;
542	pc->pc_curthread = &thread0;
543	pc->pc_curpcb = thread0.td_pcb;
544
545	/*
546	 * Initialize global registers.
547	 */
548	cpu_setregs(pc);
549
550	/*
551	 * Take over the trap table via the PROM.  Using the PROM for this
552	 * is necessary in order to set obp-control-relinquished to true
553	 * within the PROM so obtaining /virtual-memory/translations doesn't
554	 * trigger a fatal reset error or worse things further down the road.
555	 * XXX it should be possible to use this solely instead of writing
556	 * %tba in cpu_setregs().  Doing so causes a hang however.
557	 *
558	 * NB: the low-level console drivers require a working DELAY() and
559	 * some compiler optimizations may cause the curthread accesses of
560	 * mutex(9) to be factored out even if the latter aren't actually
561	 * called.  Both of these require PCPU_REG to be set.  However, we
562	 * can't set PCPU_REG without also taking over the trap table or the
563	 * firmware will overwrite it.
564	 */
565	sun4u_set_traptable(tl0_base);
566
567	/*
568	 * Initialize the dynamic per-CPU area for the BSP and the message
569	 * buffer (after setting the trap table).
570	 */
571	dpcpu_init(dpcpu0, 0);
572	msgbufinit(msgbufp, msgbufsize);
573
574	/*
575	 * Initialize mutexes.
576	 */
577	mutex_init();
578
579	/*
580	 * Initialize console now that we have a reasonable set of system
581	 * services.
582	 */
583	cninit();
584
585	/*
586	 * Finish the interrupt initialization now that mutexes work and
587	 * enable them.
588	 */
589	intr_init2();
590	wrpr(pil, 0, 0);
591	wrpr(pstate, 0, PSTATE_KERNEL);
592
593	OF_getprop(root, "name", sparc64_model, sizeof(sparc64_model) - 1);
594
595	kdb_init();
596
597#ifdef KDB
598	if (boothowto & RB_KDB)
599		kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger");
600#endif
601}
602
603void
604sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
605{
606	struct trapframe *tf;
607	struct sigframe *sfp;
608	struct sigacts *psp;
609	struct sigframe sf;
610	struct thread *td;
611	struct frame *fp;
612	struct proc *p;
613	u_long sp;
614	int oonstack;
615	int sig;
616
617	oonstack = 0;
618	td = curthread;
619	p = td->td_proc;
620	PROC_LOCK_ASSERT(p, MA_OWNED);
621	sig = ksi->ksi_signo;
622	psp = p->p_sigacts;
623	mtx_assert(&psp->ps_mtx, MA_OWNED);
624	tf = td->td_frame;
625	sp = tf->tf_sp + SPOFF;
626	oonstack = sigonstack(sp);
627
628	CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm,
629	    catcher, sig);
630
631	/* Make sure we have a signal trampoline to return to. */
632	if (p->p_md.md_sigtramp == NULL) {
633		/*
634		 * No signal trampoline... kill the process.
635		 */
636		CTR0(KTR_SIG, "sendsig: no sigtramp");
637		printf("sendsig: %s is too old, rebuild it\n", p->p_comm);
638		sigexit(td, sig);
639		/* NOTREACHED */
640	}
641
642	/* Save user context. */
643	bzero(&sf, sizeof(sf));
644	get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
645	sf.sf_uc.uc_sigmask = *mask;
646	sf.sf_uc.uc_stack = td->td_sigstk;
647	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ?
648	    ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
649
650	/* Allocate and validate space for the signal handler context. */
651	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
652	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
653		sfp = (struct sigframe *)(td->td_sigstk.ss_sp +
654		    td->td_sigstk.ss_size - sizeof(struct sigframe));
655	} else
656		sfp = (struct sigframe *)sp - 1;
657	mtx_unlock(&psp->ps_mtx);
658	PROC_UNLOCK(p);
659
660	fp = (struct frame *)sfp - 1;
661
662	/* Build the argument list for the signal handler. */
663	tf->tf_out[0] = sig;
664	tf->tf_out[2] = (register_t)&sfp->sf_uc;
665	tf->tf_out[4] = (register_t)catcher;
666	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
667		/* Signal handler installed with SA_SIGINFO. */
668		tf->tf_out[1] = (register_t)&sfp->sf_si;
669
670		/* Fill in POSIX parts. */
671		sf.sf_si = ksi->ksi_info;
672		sf.sf_si.si_signo = sig; /* maybe a translated signal */
673	} else {
674		/* Old FreeBSD-style arguments. */
675		tf->tf_out[1] = ksi->ksi_code;
676		tf->tf_out[3] = (register_t)ksi->ksi_addr;
677	}
678
679	/* Copy the sigframe out to the user's stack. */
680	if (rwindow_save(td) != 0 || copyout(&sf, sfp, sizeof(*sfp)) != 0 ||
681	    suword(&fp->fr_in[6], tf->tf_out[6]) != 0) {
682		/*
683		 * Something is wrong with the stack pointer.
684		 * ...Kill the process.
685		 */
686		CTR2(KTR_SIG, "sendsig: sigexit td=%p sfp=%p", td, sfp);
687		PROC_LOCK(p);
688		sigexit(td, SIGILL);
689		/* NOTREACHED */
690	}
691
692	tf->tf_tpc = (u_long)p->p_md.md_sigtramp;
693	tf->tf_tnpc = tf->tf_tpc + 4;
694	tf->tf_sp = (u_long)fp - SPOFF;
695
696	CTR3(KTR_SIG, "sendsig: return td=%p pc=%#lx sp=%#lx", td, tf->tf_tpc,
697	    tf->tf_sp);
698
699	PROC_LOCK(p);
700	mtx_lock(&psp->ps_mtx);
701}
702
703#ifndef	_SYS_SYSPROTO_H_
704struct sigreturn_args {
705	ucontext_t *ucp;
706};
707#endif
708
709/*
710 * MPSAFE
711 */
712int
713sys_sigreturn(struct thread *td, struct sigreturn_args *uap)
714{
715	struct proc *p;
716	mcontext_t *mc;
717	ucontext_t uc;
718	int error;
719
720	p = td->td_proc;
721	if (rwindow_save(td)) {
722		PROC_LOCK(p);
723		sigexit(td, SIGILL);
724	}
725
726	CTR2(KTR_SIG, "sigreturn: td=%p ucp=%p", td, uap->sigcntxp);
727	if (copyin(uap->sigcntxp, &uc, sizeof(uc)) != 0) {
728		CTR1(KTR_SIG, "sigreturn: efault td=%p", td);
729		return (EFAULT);
730	}
731
732	mc = &uc.uc_mcontext;
733	error = set_mcontext(td, mc);
734	if (error != 0)
735		return (error);
736
737	kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
738
739	CTR4(KTR_SIG, "sigreturn: return td=%p pc=%#lx sp=%#lx tstate=%#lx",
740	    td, mc->_mc_tpc, mc->_mc_sp, mc->_mc_tstate);
741	return (EJUSTRETURN);
742}
743
744/*
745 * Construct a PCB from a trapframe. This is called from kdb_trap() where
746 * we want to start a backtrace from the function that caused us to enter
747 * the debugger. We have the context in the trapframe, but base the trace
748 * on the PCB. The PCB doesn't have to be perfect, as long as it contains
749 * enough for a backtrace.
750 */
751void
752makectx(struct trapframe *tf, struct pcb *pcb)
753{
754
755	pcb->pcb_pc = tf->tf_tpc;
756	pcb->pcb_sp = tf->tf_sp;
757}
758
759int
760get_mcontext(struct thread *td, mcontext_t *mc, int flags)
761{
762	struct trapframe *tf;
763	struct pcb *pcb;
764
765	tf = td->td_frame;
766	pcb = td->td_pcb;
767	/*
768	 * Copy the registers which will be restored by tl0_ret() from the
769	 * trapframe.
770	 * Note that we skip %g7 which is used as the userland TLS register
771	 * and %wstate.
772	 */
773	mc->_mc_flags = _MC_VERSION;
774	mc->mc_global[1] = tf->tf_global[1];
775	mc->mc_global[2] = tf->tf_global[2];
776	mc->mc_global[3] = tf->tf_global[3];
777	mc->mc_global[4] = tf->tf_global[4];
778	mc->mc_global[5] = tf->tf_global[5];
779	mc->mc_global[6] = tf->tf_global[6];
780	if (flags & GET_MC_CLEAR_RET) {
781		mc->mc_out[0] = 0;
782		mc->mc_out[1] = 0;
783	} else {
784		mc->mc_out[0] = tf->tf_out[0];
785		mc->mc_out[1] = tf->tf_out[1];
786	}
787	mc->mc_out[2] = tf->tf_out[2];
788	mc->mc_out[3] = tf->tf_out[3];
789	mc->mc_out[4] = tf->tf_out[4];
790	mc->mc_out[5] = tf->tf_out[5];
791	mc->mc_out[6] = tf->tf_out[6];
792	mc->mc_out[7] = tf->tf_out[7];
793	mc->_mc_fprs = tf->tf_fprs;
794	mc->_mc_fsr = tf->tf_fsr;
795	mc->_mc_gsr = tf->tf_gsr;
796	mc->_mc_tnpc = tf->tf_tnpc;
797	mc->_mc_tpc = tf->tf_tpc;
798	mc->_mc_tstate = tf->tf_tstate;
799	mc->_mc_y = tf->tf_y;
800	critical_enter();
801	if ((tf->tf_fprs & FPRS_FEF) != 0) {
802		savefpctx(pcb->pcb_ufp);
803		tf->tf_fprs &= ~FPRS_FEF;
804		pcb->pcb_flags |= PCB_FEF;
805	}
806	if ((pcb->pcb_flags & PCB_FEF) != 0) {
807		bcopy(pcb->pcb_ufp, mc->mc_fp, sizeof(mc->mc_fp));
808		mc->_mc_fprs |= FPRS_FEF;
809	}
810	critical_exit();
811	return (0);
812}
813
814int
815set_mcontext(struct thread *td, mcontext_t *mc)
816{
817	struct trapframe *tf;
818	struct pcb *pcb;
819
820	if (!TSTATE_SECURE(mc->_mc_tstate) ||
821	    (mc->_mc_flags & ((1L << _MC_VERSION_BITS) - 1)) != _MC_VERSION)
822		return (EINVAL);
823	tf = td->td_frame;
824	pcb = td->td_pcb;
825	/* Make sure the windows are spilled first. */
826	flushw();
827	/*
828	 * Copy the registers which will be restored by tl0_ret() to the
829	 * trapframe.
830	 * Note that we skip %g7 which is used as the userland TLS register
831	 * and %wstate.
832	 */
833	tf->tf_global[1] = mc->mc_global[1];
834	tf->tf_global[2] = mc->mc_global[2];
835	tf->tf_global[3] = mc->mc_global[3];
836	tf->tf_global[4] = mc->mc_global[4];
837	tf->tf_global[5] = mc->mc_global[5];
838	tf->tf_global[6] = mc->mc_global[6];
839	tf->tf_out[0] = mc->mc_out[0];
840	tf->tf_out[1] = mc->mc_out[1];
841	tf->tf_out[2] = mc->mc_out[2];
842	tf->tf_out[3] = mc->mc_out[3];
843	tf->tf_out[4] = mc->mc_out[4];
844	tf->tf_out[5] = mc->mc_out[5];
845	tf->tf_out[6] = mc->mc_out[6];
846	tf->tf_out[7] = mc->mc_out[7];
847	tf->tf_fprs = mc->_mc_fprs;
848	tf->tf_fsr = mc->_mc_fsr;
849	tf->tf_gsr = mc->_mc_gsr;
850	tf->tf_tnpc = mc->_mc_tnpc;
851	tf->tf_tpc = mc->_mc_tpc;
852	tf->tf_tstate = mc->_mc_tstate;
853	tf->tf_y = mc->_mc_y;
854	if ((mc->_mc_fprs & FPRS_FEF) != 0) {
855		tf->tf_fprs = 0;
856		bcopy(mc->mc_fp, pcb->pcb_ufp, sizeof(pcb->pcb_ufp));
857		pcb->pcb_flags |= PCB_FEF;
858	}
859	return (0);
860}
861
862/*
863 * Exit the kernel and execute a firmware call that will not return, as
864 * specified by the arguments.
865 */
866void
867cpu_shutdown(void *args)
868{
869
870#ifdef SMP
871	cpu_mp_shutdown();
872#endif
873	ofw_exit(args);
874}
875
876/*
877 * Flush the D-cache for non-DMA I/O so that the I-cache can
878 * be made coherent later.
879 */
880void
881cpu_flush_dcache(void *ptr, size_t len)
882{
883
884	/* TBD */
885}
886
887/* Get current clock frequency for the given CPU ID. */
888int
889cpu_est_clockrate(int cpu_id, uint64_t *rate)
890{
891	struct pcpu *pc;
892
893	pc = pcpu_find(cpu_id);
894	if (pc == NULL || rate == NULL)
895		return (EINVAL);
896	*rate = pc->pc_clock;
897	return (0);
898}
899
900/*
901 * Duplicate OF_exit() with a different firmware call function that restores
902 * the trap table, otherwise a RED state exception is triggered in at least
903 * some firmware versions.
904 */
905void
906cpu_halt(void)
907{
908	static struct {
909		cell_t name;
910		cell_t nargs;
911		cell_t nreturns;
912	} args = {
913		(cell_t)"exit",
914		0,
915		0
916	};
917
918	cpu_shutdown(&args);
919}
920
921static void
922sparc64_shutdown_final(void *dummy, int howto)
923{
924	static struct {
925		cell_t name;
926		cell_t nargs;
927		cell_t nreturns;
928	} args = {
929		(cell_t)"SUNW,power-off",
930		0,
931		0
932	};
933
934	/* Turn the power off? */
935	if ((howto & RB_POWEROFF) != 0)
936		cpu_shutdown(&args);
937	/* In case of halt, return to the firmware. */
938	if ((howto & RB_HALT) != 0)
939		cpu_halt();
940}
941
942void
943cpu_idle(int busy)
944{
945
946	/* Insert code to halt (until next interrupt) for the idle loop. */
947}
948
949int
950cpu_idle_wakeup(int cpu)
951{
952
953	return (1);
954}
955
956int
957ptrace_set_pc(struct thread *td, u_long addr)
958{
959
960	td->td_frame->tf_tpc = addr;
961	td->td_frame->tf_tnpc = addr + 4;
962	return (0);
963}
964
965int
966ptrace_single_step(struct thread *td)
967{
968
969	/* TODO; */
970	return (0);
971}
972
973int
974ptrace_clear_single_step(struct thread *td)
975{
976
977	/* TODO; */
978	return (0);
979}
980
981void
982exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
983{
984	struct trapframe *tf;
985	struct pcb *pcb;
986	struct proc *p;
987	u_long sp;
988
989	/* XXX no cpu_exec */
990	p = td->td_proc;
991	p->p_md.md_sigtramp = NULL;
992	if (p->p_md.md_utrap != NULL) {
993		utrap_free(p->p_md.md_utrap);
994		p->p_md.md_utrap = NULL;
995	}
996
997	pcb = td->td_pcb;
998	tf = td->td_frame;
999	sp = rounddown(stack, 16);
1000	bzero(pcb, sizeof(*pcb));
1001	bzero(tf, sizeof(*tf));
1002	tf->tf_out[0] = stack;
1003	tf->tf_out[3] = p->p_sysent->sv_psstrings;
1004	tf->tf_out[6] = sp - SPOFF - sizeof(struct frame);
1005	tf->tf_tnpc = imgp->entry_addr + 4;
1006	tf->tf_tpc = imgp->entry_addr;
1007	/*
1008	 * While we could adhere to the memory model indicated in the ELF
1009	 * header, it turns out that just always using TSO performs best.
1010	 */
1011	tf->tf_tstate = TSTATE_IE | TSTATE_PEF | TSTATE_MM_TSO;
1012
1013	td->td_retval[0] = tf->tf_out[0];
1014	td->td_retval[1] = tf->tf_out[1];
1015}
1016
1017int
1018fill_regs(struct thread *td, struct reg *regs)
1019{
1020
1021	bcopy(td->td_frame, regs, sizeof(*regs));
1022	return (0);
1023}
1024
1025int
1026set_regs(struct thread *td, struct reg *regs)
1027{
1028	struct trapframe *tf;
1029
1030	if (!TSTATE_SECURE(regs->r_tstate))
1031		return (EINVAL);
1032	tf = td->td_frame;
1033	regs->r_wstate = tf->tf_wstate;
1034	bcopy(regs, tf, sizeof(*regs));
1035	return (0);
1036}
1037
1038int
1039fill_dbregs(struct thread *td, struct dbreg *dbregs)
1040{
1041
1042	return (ENOSYS);
1043}
1044
1045int
1046set_dbregs(struct thread *td, struct dbreg *dbregs)
1047{
1048
1049	return (ENOSYS);
1050}
1051
1052int
1053fill_fpregs(struct thread *td, struct fpreg *fpregs)
1054{
1055	struct trapframe *tf;
1056	struct pcb *pcb;
1057
1058	pcb = td->td_pcb;
1059	tf = td->td_frame;
1060	bcopy(pcb->pcb_ufp, fpregs->fr_regs, sizeof(fpregs->fr_regs));
1061	fpregs->fr_fsr = tf->tf_fsr;
1062	fpregs->fr_gsr = tf->tf_gsr;
1063	return (0);
1064}
1065
1066int
1067set_fpregs(struct thread *td, struct fpreg *fpregs)
1068{
1069	struct trapframe *tf;
1070	struct pcb *pcb;
1071
1072	pcb = td->td_pcb;
1073	tf = td->td_frame;
1074	tf->tf_fprs &= ~FPRS_FEF;
1075	bcopy(fpregs->fr_regs, pcb->pcb_ufp, sizeof(pcb->pcb_ufp));
1076	tf->tf_fsr = fpregs->fr_fsr;
1077	tf->tf_gsr = fpregs->fr_gsr;
1078	return (0);
1079}
1080
1081struct md_utrap *
1082utrap_alloc(void)
1083{
1084	struct md_utrap *ut;
1085
1086	ut = malloc(sizeof(struct md_utrap), M_SUBPROC, M_WAITOK | M_ZERO);
1087	ut->ut_refcnt = 1;
1088	return (ut);
1089}
1090
1091void
1092utrap_free(struct md_utrap *ut)
1093{
1094	int refcnt;
1095
1096	if (ut == NULL)
1097		return;
1098	mtx_pool_lock(mtxpool_sleep, ut);
1099	ut->ut_refcnt--;
1100	refcnt = ut->ut_refcnt;
1101	mtx_pool_unlock(mtxpool_sleep, ut);
1102	if (refcnt == 0)
1103		free(ut, M_SUBPROC);
1104}
1105
1106struct md_utrap *
1107utrap_hold(struct md_utrap *ut)
1108{
1109
1110	if (ut == NULL)
1111		return (NULL);
1112	mtx_pool_lock(mtxpool_sleep, ut);
1113	ut->ut_refcnt++;
1114	mtx_pool_unlock(mtxpool_sleep, ut);
1115	return (ut);
1116}
1117