machdep.c revision 325307
1/*	$NetBSD: arm32_machdep.c,v 1.44 2004/03/24 15:34:47 atatat Exp $	*/
2
3/*-
4 * Copyright (c) 2004 Olivier Houchard
5 * Copyright (c) 1994-1998 Mark Brinicombe.
6 * Copyright (c) 1994 Brini.
7 * All rights reserved.
8 *
9 * This code is derived from software written for Brini by Mark Brinicombe
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 *    notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 *    notice, this list of conditions and the following disclaimer in the
18 *    documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 *    must display the following acknowledgement:
21 *	This product includes software developed by Mark Brinicombe
22 *	for the NetBSD Project.
23 * 4. The name of the company nor the name of the author may be used to
24 *    endorse or promote products derived from this software without specific
25 *    prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
28 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
29 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
30 * IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
31 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
32 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
33 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 *
39 * Machine dependent functions for kernel setup
40 *
41 * Created      : 17/09/94
42 * Updated	: 18/04/01 updated for new wscons
43 */
44
45#include "opt_compat.h"
46#include "opt_ddb.h"
47#include "opt_kstack_pages.h"
48#include "opt_platform.h"
49#include "opt_sched.h"
50#include "opt_timer.h"
51
52#include <sys/cdefs.h>
53__FBSDID("$FreeBSD: stable/11/sys/arm/arm/machdep.c 325307 2017-11-02 07:08:13Z mmel $");
54
55#include <sys/param.h>
56#include <sys/buf.h>
57#include <sys/bus.h>
58#include <sys/cons.h>
59#include <sys/cpu.h>
60#include <sys/devmap.h>
61#include <sys/efi.h>
62#include <sys/imgact.h>
63#include <sys/kdb.h>
64#include <sys/kernel.h>
65#include <sys/linker.h>
66#include <sys/msgbuf.h>
67#include <sys/rwlock.h>
68#include <sys/sched.h>
69#include <sys/syscallsubr.h>
70#include <sys/sysent.h>
71#include <sys/sysproto.h>
72
73#include <vm/vm_object.h>
74#include <vm/vm_page.h>
75#include <vm/vm_pager.h>
76
77#include <machine/debug_monitor.h>
78#include <machine/machdep.h>
79#include <machine/metadata.h>
80#include <machine/pcb.h>
81#include <machine/physmem.h>
82#include <machine/platform.h>
83#include <machine/sysarch.h>
84#include <machine/undefined.h>
85#include <machine/vfp.h>
86#include <machine/vmparam.h>
87
88#ifdef FDT
89#include <dev/fdt/fdt_common.h>
90#include <machine/ofw_machdep.h>
91#endif
92
93#ifdef DEBUG
94#define	debugf(fmt, args...) printf(fmt, ##args)
95#else
96#define	debugf(fmt, args...)
97#endif
98
99#if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
100    defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) || \
101    defined(COMPAT_FREEBSD9)
102#error FreeBSD/arm doesn't provide compatibility with releases prior to 10
103#endif
104
105struct pcpu __pcpu[MAXCPU];
106struct pcpu *pcpup = &__pcpu[0];
107
108static struct trapframe proc0_tf;
109uint32_t cpu_reset_address = 0;
110int cold = 1;
111vm_offset_t vector_page;
112
113int (*_arm_memcpy)(void *, void *, int, int) = NULL;
114int (*_arm_bzero)(void *, int, int) = NULL;
115int _min_memcpy_size = 0;
116int _min_bzero_size = 0;
117
118extern int *end;
119
120#ifdef FDT
121vm_paddr_t pmap_pa;
122#if __ARM_ARCH >= 6
123vm_offset_t systempage;
124vm_offset_t irqstack;
125vm_offset_t undstack;
126vm_offset_t abtstack;
127#else
128/*
129 * This is the number of L2 page tables required for covering max
130 * (hypothetical) memsize of 4GB and all kernel mappings (vectors, msgbuf,
131 * stacks etc.), uprounded to be divisible by 4.
132 */
133#define KERNEL_PT_MAX	78
134static struct pv_addr kernel_pt_table[KERNEL_PT_MAX];
135struct pv_addr systempage;
136static struct pv_addr msgbufpv;
137struct pv_addr irqstack;
138struct pv_addr undstack;
139struct pv_addr abtstack;
140static struct pv_addr kernelstack;
141#endif /* __ARM_ARCH >= 6 */
142#endif /* FDT */
143
144#ifdef MULTIDELAY
145static delay_func *delay_impl;
146static void *delay_arg;
147#endif
148
149struct kva_md_info kmi;
150
151/*
152 * arm32_vector_init:
153 *
154 *	Initialize the vector page, and select whether or not to
155 *	relocate the vectors.
156 *
157 *	NOTE: We expect the vector page to be mapped at its expected
158 *	destination.
159 */
160
161extern unsigned int page0[], page0_data[];
162void
163arm_vector_init(vm_offset_t va, int which)
164{
165	unsigned int *vectors = (int *) va;
166	unsigned int *vectors_data = vectors + (page0_data - page0);
167	int vec;
168
169	/*
170	 * Loop through the vectors we're taking over, and copy the
171	 * vector's insn and data word.
172	 */
173	for (vec = 0; vec < ARM_NVEC; vec++) {
174		if ((which & (1 << vec)) == 0) {
175			/* Don't want to take over this vector. */
176			continue;
177		}
178		vectors[vec] = page0[vec];
179		vectors_data[vec] = page0_data[vec];
180	}
181
182	/* Now sync the vectors. */
183	icache_sync(va, (ARM_NVEC * 2) * sizeof(u_int));
184
185	vector_page = va;
186#if __ARM_ARCH < 6
187	if (va == ARM_VECTORS_HIGH) {
188		/*
189		 * Enable high vectors in the system control reg (SCTLR).
190		 *
191		 * Assume the MD caller knows what it's doing here, and really
192		 * does want the vector page relocated.
193		 *
194		 * Note: This has to be done here (and not just in
195		 * cpu_setup()) because the vector page needs to be
196		 * accessible *before* cpu_startup() is called.
197		 * Think ddb(9) ...
198		 */
199		cpu_control(CPU_CONTROL_VECRELOC, CPU_CONTROL_VECRELOC);
200	}
201#endif
202}
203
204static void
205cpu_startup(void *dummy)
206{
207	struct pcb *pcb = thread0.td_pcb;
208	const unsigned int mbyte = 1024 * 1024;
209#if __ARM_ARCH < 6 && !defined(ARM_CACHE_LOCK_ENABLE)
210	vm_page_t m;
211#endif
212
213	identify_arm_cpu();
214
215	vm_ksubmap_init(&kmi);
216
217	/*
218	 * Display the RAM layout.
219	 */
220	printf("real memory  = %ju (%ju MB)\n",
221	    (uintmax_t)arm32_ptob(realmem),
222	    (uintmax_t)arm32_ptob(realmem) / mbyte);
223	printf("avail memory = %ju (%ju MB)\n",
224	    (uintmax_t)arm32_ptob(vm_cnt.v_free_count),
225	    (uintmax_t)arm32_ptob(vm_cnt.v_free_count) / mbyte);
226	if (bootverbose) {
227		arm_physmem_print_tables();
228		devmap_print_table();
229	}
230
231	bufinit();
232	vm_pager_bufferinit();
233	pcb->pcb_regs.sf_sp = (u_int)thread0.td_kstack +
234	    USPACE_SVC_STACK_TOP;
235	pmap_set_pcb_pagedir(kernel_pmap, pcb);
236#if __ARM_ARCH < 6
237	vector_page_setprot(VM_PROT_READ);
238	pmap_postinit();
239#ifdef ARM_CACHE_LOCK_ENABLE
240	pmap_kenter_user(ARM_TP_ADDRESS, ARM_TP_ADDRESS);
241	arm_lock_cache_line(ARM_TP_ADDRESS);
242#else
243	m = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ | VM_ALLOC_ZERO);
244	pmap_kenter_user(ARM_TP_ADDRESS, VM_PAGE_TO_PHYS(m));
245#endif
246	*(uint32_t *)ARM_RAS_START = 0;
247	*(uint32_t *)ARM_RAS_END = 0xffffffff;
248#endif
249}
250
251SYSINIT(cpu, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
252
253/*
254 * Flush the D-cache for non-DMA I/O so that the I-cache can
255 * be made coherent later.
256 */
257void
258cpu_flush_dcache(void *ptr, size_t len)
259{
260
261	dcache_wb_poc((vm_offset_t)ptr, (vm_paddr_t)vtophys(ptr), len);
262}
263
264/* Get current clock frequency for the given cpu id. */
265int
266cpu_est_clockrate(int cpu_id, uint64_t *rate)
267{
268
269	return (ENXIO);
270}
271
272void
273cpu_idle(int busy)
274{
275
276	CTR2(KTR_SPARE2, "cpu_idle(%d) at %d", busy, curcpu);
277	spinlock_enter();
278#ifndef NO_EVENTTIMERS
279	if (!busy)
280		cpu_idleclock();
281#endif
282	if (!sched_runnable())
283		cpu_sleep(0);
284#ifndef NO_EVENTTIMERS
285	if (!busy)
286		cpu_activeclock();
287#endif
288	spinlock_exit();
289	CTR2(KTR_SPARE2, "cpu_idle(%d) at %d done", busy, curcpu);
290}
291
292int
293cpu_idle_wakeup(int cpu)
294{
295
296	return (0);
297}
298
299/*
300 * Most ARM platforms don't need to do anything special to init their clocks
301 * (they get intialized during normal device attachment), and by not defining a
302 * cpu_initclocks() function they get this generic one.  Any platform that needs
303 * to do something special can just provide their own implementation, which will
304 * override this one due to the weak linkage.
305 */
306void
307arm_generic_initclocks(void)
308{
309
310#ifndef NO_EVENTTIMERS
311#ifdef SMP
312	if (PCPU_GET(cpuid) == 0)
313		cpu_initclocks_bsp();
314	else
315		cpu_initclocks_ap();
316#else
317	cpu_initclocks_bsp();
318#endif
319#endif
320}
321__weak_reference(arm_generic_initclocks, cpu_initclocks);
322
323#ifdef MULTIDELAY
324void
325arm_set_delay(delay_func *impl, void *arg)
326{
327
328	KASSERT(impl != NULL, ("No DELAY implementation"));
329	delay_impl = impl;
330	delay_arg = arg;
331}
332
333void
334DELAY(int usec)
335{
336
337	delay_impl(usec, delay_arg);
338}
339#endif
340
341void
342cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
343{
344}
345
346void
347spinlock_enter(void)
348{
349	struct thread *td;
350	register_t cspr;
351
352	td = curthread;
353	if (td->td_md.md_spinlock_count == 0) {
354		cspr = disable_interrupts(PSR_I | PSR_F);
355		td->td_md.md_spinlock_count = 1;
356		td->td_md.md_saved_cspr = cspr;
357	} else
358		td->td_md.md_spinlock_count++;
359	critical_enter();
360}
361
362void
363spinlock_exit(void)
364{
365	struct thread *td;
366	register_t cspr;
367
368	td = curthread;
369	critical_exit();
370	cspr = td->td_md.md_saved_cspr;
371	td->td_md.md_spinlock_count--;
372	if (td->td_md.md_spinlock_count == 0)
373		restore_interrupts(cspr);
374}
375
376/*
377 * Clear registers on exec
378 */
379void
380exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
381{
382	struct trapframe *tf = td->td_frame;
383
384	memset(tf, 0, sizeof(*tf));
385	tf->tf_usr_sp = stack;
386	tf->tf_usr_lr = imgp->entry_addr;
387	tf->tf_svc_lr = 0x77777777;
388	tf->tf_pc = imgp->entry_addr;
389	tf->tf_spsr = PSR_USR32_MODE;
390}
391
392
393#ifdef VFP
394/*
395 * Get machine VFP context.
396 */
397static void
398get_vfpcontext(struct thread *td, mcontext_vfp_t *vfp)
399{
400	struct pcb *curpcb;
401
402	curpcb = curthread->td_pcb;
403	critical_enter();
404
405	vfp_store(&curpcb->pcb_vfpstate, false);
406	memcpy(vfp->mcv_reg, curpcb->pcb_vfpstate.reg,
407	    sizeof(vfp->mcv_reg));
408	vfp->mcv_fpscr = curpcb->pcb_vfpstate.fpscr;
409
410	critical_exit();
411}
412
413/*
414 * Set machine VFP context.
415 */
416static void
417set_vfpcontext(struct thread *td, mcontext_vfp_t *vfp)
418{
419	struct pcb *curpcb;
420
421	curpcb = curthread->td_pcb;
422	critical_enter();
423
424	vfp_discard(td);
425	memcpy(curpcb->pcb_vfpstate.reg, vfp->mcv_reg,
426	    sizeof(curpcb->pcb_vfpstate.reg));
427	curpcb->pcb_vfpstate.fpscr = vfp->mcv_fpscr;
428
429	critical_exit();
430}
431#endif
432
433int
434arm_get_vfpstate(struct thread *td, void *args)
435{
436	int rv;
437	struct arm_get_vfpstate_args ua;
438	mcontext_vfp_t	mcontext_vfp;
439
440	rv = copyin(args, &ua, sizeof(ua));
441	if (rv != 0)
442		return (rv);
443	if (ua.mc_vfp_size != sizeof(mcontext_vfp_t))
444		return (EINVAL);
445#ifdef VFP
446	get_vfpcontext(td, &mcontext_vfp);
447#else
448	bzero(&mcontext_vfp, sizeof(mcontext_vfp));
449#endif
450
451	rv = copyout(&mcontext_vfp, ua.mc_vfp,  sizeof(mcontext_vfp));
452	if (rv != 0)
453		return (rv);
454	return (0);
455}
456
457/*
458 * Get machine context.
459 */
460int
461get_mcontext(struct thread *td, mcontext_t *mcp, int clear_ret)
462{
463	struct trapframe *tf = td->td_frame;
464	__greg_t *gr = mcp->__gregs;
465
466	if (clear_ret & GET_MC_CLEAR_RET) {
467		gr[_REG_R0] = 0;
468		gr[_REG_CPSR] = tf->tf_spsr & ~PSR_C;
469	} else {
470		gr[_REG_R0]   = tf->tf_r0;
471		gr[_REG_CPSR] = tf->tf_spsr;
472	}
473	gr[_REG_R1]   = tf->tf_r1;
474	gr[_REG_R2]   = tf->tf_r2;
475	gr[_REG_R3]   = tf->tf_r3;
476	gr[_REG_R4]   = tf->tf_r4;
477	gr[_REG_R5]   = tf->tf_r5;
478	gr[_REG_R6]   = tf->tf_r6;
479	gr[_REG_R7]   = tf->tf_r7;
480	gr[_REG_R8]   = tf->tf_r8;
481	gr[_REG_R9]   = tf->tf_r9;
482	gr[_REG_R10]  = tf->tf_r10;
483	gr[_REG_R11]  = tf->tf_r11;
484	gr[_REG_R12]  = tf->tf_r12;
485	gr[_REG_SP]   = tf->tf_usr_sp;
486	gr[_REG_LR]   = tf->tf_usr_lr;
487	gr[_REG_PC]   = tf->tf_pc;
488
489	mcp->mc_vfp_size = 0;
490	mcp->mc_vfp_ptr = NULL;
491	memset(&mcp->mc_spare, 0, sizeof(mcp->mc_spare));
492
493	return (0);
494}
495
496/*
497 * Set machine context.
498 *
499 * However, we don't set any but the user modifiable flags, and we won't
500 * touch the cs selector.
501 */
502int
503set_mcontext(struct thread *td, mcontext_t *mcp)
504{
505	mcontext_vfp_t mc_vfp, *vfp;
506	struct trapframe *tf = td->td_frame;
507	const __greg_t *gr = mcp->__gregs;
508
509#ifdef WITNESS
510	if (mcp->mc_vfp_size != 0 && mcp->mc_vfp_size != sizeof(mc_vfp)) {
511		printf("%s: %s: Malformed mc_vfp_size: %d (0x%08X)\n",
512		    td->td_proc->p_comm, __func__,
513		    mcp->mc_vfp_size, mcp->mc_vfp_size);
514	} else if (mcp->mc_vfp_size != 0 && mcp->mc_vfp_ptr == NULL) {
515		printf("%s: %s: c_vfp_size != 0 but mc_vfp_ptr == NULL\n",
516		    td->td_proc->p_comm, __func__);
517	}
518#endif
519
520	if (mcp->mc_vfp_size == sizeof(mc_vfp) && mcp->mc_vfp_ptr != NULL) {
521		if (copyin(mcp->mc_vfp_ptr, &mc_vfp, sizeof(mc_vfp)) != 0)
522			return (EFAULT);
523		vfp = &mc_vfp;
524	} else {
525		vfp = NULL;
526	}
527
528	tf->tf_r0 = gr[_REG_R0];
529	tf->tf_r1 = gr[_REG_R1];
530	tf->tf_r2 = gr[_REG_R2];
531	tf->tf_r3 = gr[_REG_R3];
532	tf->tf_r4 = gr[_REG_R4];
533	tf->tf_r5 = gr[_REG_R5];
534	tf->tf_r6 = gr[_REG_R6];
535	tf->tf_r7 = gr[_REG_R7];
536	tf->tf_r8 = gr[_REG_R8];
537	tf->tf_r9 = gr[_REG_R9];
538	tf->tf_r10 = gr[_REG_R10];
539	tf->tf_r11 = gr[_REG_R11];
540	tf->tf_r12 = gr[_REG_R12];
541	tf->tf_usr_sp = gr[_REG_SP];
542	tf->tf_usr_lr = gr[_REG_LR];
543	tf->tf_pc = gr[_REG_PC];
544	tf->tf_spsr = gr[_REG_CPSR];
545#ifdef VFP
546	if (vfp != NULL)
547		set_vfpcontext(td, vfp);
548#endif
549	return (0);
550}
551
552void
553sendsig(catcher, ksi, mask)
554	sig_t catcher;
555	ksiginfo_t *ksi;
556	sigset_t *mask;
557{
558	struct thread *td;
559	struct proc *p;
560	struct trapframe *tf;
561	struct sigframe *fp, frame;
562	struct sigacts *psp;
563	struct sysentvec *sysent;
564	int onstack;
565	int sig;
566	int code;
567
568	td = curthread;
569	p = td->td_proc;
570	PROC_LOCK_ASSERT(p, MA_OWNED);
571	sig = ksi->ksi_signo;
572	code = ksi->ksi_code;
573	psp = p->p_sigacts;
574	mtx_assert(&psp->ps_mtx, MA_OWNED);
575	tf = td->td_frame;
576	onstack = sigonstack(tf->tf_usr_sp);
577
578	CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm,
579	    catcher, sig);
580
581	/* Allocate and validate space for the signal handler context. */
582	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !(onstack) &&
583	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
584		fp = (struct sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
585		    td->td_sigstk.ss_size);
586#if defined(COMPAT_43)
587		td->td_sigstk.ss_flags |= SS_ONSTACK;
588#endif
589	} else
590		fp = (struct sigframe *)td->td_frame->tf_usr_sp;
591
592	/* make room on the stack */
593	fp--;
594
595	/* make the stack aligned */
596	fp = (struct sigframe *)STACKALIGN(fp);
597	/* Populate the siginfo frame. */
598	get_mcontext(td, &frame.sf_uc.uc_mcontext, 0);
599#ifdef VFP
600	get_vfpcontext(td, &frame.sf_vfp);
601	frame.sf_uc.uc_mcontext.mc_vfp_size = sizeof(fp->sf_vfp);
602	frame.sf_uc.uc_mcontext.mc_vfp_ptr = &fp->sf_vfp;
603#else
604	frame.sf_uc.uc_mcontext.mc_vfp_size = 0;
605	frame.sf_uc.uc_mcontext.mc_vfp_ptr = NULL;
606#endif
607	frame.sf_si = ksi->ksi_info;
608	frame.sf_uc.uc_sigmask = *mask;
609	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK )
610	    ? ((onstack) ? SS_ONSTACK : 0) : SS_DISABLE;
611	frame.sf_uc.uc_stack = td->td_sigstk;
612	mtx_unlock(&psp->ps_mtx);
613	PROC_UNLOCK(td->td_proc);
614
615	/* Copy the sigframe out to the user's stack. */
616	if (copyout(&frame, fp, sizeof(*fp)) != 0) {
617		/* Process has trashed its stack. Kill it. */
618		CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp);
619		PROC_LOCK(p);
620		sigexit(td, SIGILL);
621	}
622
623	/*
624	 * Build context to run handler in.  We invoke the handler
625	 * directly, only returning via the trampoline.  Note the
626	 * trampoline version numbers are coordinated with machine-
627	 * dependent code in libc.
628	 */
629
630	tf->tf_r0 = sig;
631	tf->tf_r1 = (register_t)&fp->sf_si;
632	tf->tf_r2 = (register_t)&fp->sf_uc;
633
634	/* the trampoline uses r5 as the uc address */
635	tf->tf_r5 = (register_t)&fp->sf_uc;
636	tf->tf_pc = (register_t)catcher;
637	tf->tf_usr_sp = (register_t)fp;
638	sysent = p->p_sysent;
639	if (sysent->sv_sigcode_base != 0)
640		tf->tf_usr_lr = (register_t)sysent->sv_sigcode_base;
641	else
642		tf->tf_usr_lr = (register_t)(sysent->sv_psstrings -
643		    *(sysent->sv_szsigcode));
644	/* Set the mode to enter in the signal handler */
645#if __ARM_ARCH >= 7
646	if ((register_t)catcher & 1)
647		tf->tf_spsr |= PSR_T;
648	else
649		tf->tf_spsr &= ~PSR_T;
650#endif
651
652	CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_usr_lr,
653	    tf->tf_usr_sp);
654
655	PROC_LOCK(p);
656	mtx_lock(&psp->ps_mtx);
657}
658
659int
660sys_sigreturn(td, uap)
661	struct thread *td;
662	struct sigreturn_args /* {
663		const struct __ucontext *sigcntxp;
664	} */ *uap;
665{
666	ucontext_t uc;
667	int spsr;
668
669	if (uap == NULL)
670		return (EFAULT);
671	if (copyin(uap->sigcntxp, &uc, sizeof(uc)))
672		return (EFAULT);
673	/*
674	 * Make sure the processor mode has not been tampered with and
675	 * interrupts have not been disabled.
676	 */
677	spsr = uc.uc_mcontext.__gregs[_REG_CPSR];
678	if ((spsr & PSR_MODE) != PSR_USR32_MODE ||
679	    (spsr & (PSR_I | PSR_F)) != 0)
680		return (EINVAL);
681	/* Restore register context. */
682	set_mcontext(td, &uc.uc_mcontext);
683
684	/* Restore signal mask. */
685	kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
686
687	return (EJUSTRETURN);
688}
689
690/*
691 * Construct a PCB from a trapframe. This is called from kdb_trap() where
692 * we want to start a backtrace from the function that caused us to enter
693 * the debugger. We have the context in the trapframe, but base the trace
694 * on the PCB. The PCB doesn't have to be perfect, as long as it contains
695 * enough for a backtrace.
696 */
697void
698makectx(struct trapframe *tf, struct pcb *pcb)
699{
700	pcb->pcb_regs.sf_r4 = tf->tf_r4;
701	pcb->pcb_regs.sf_r5 = tf->tf_r5;
702	pcb->pcb_regs.sf_r6 = tf->tf_r6;
703	pcb->pcb_regs.sf_r7 = tf->tf_r7;
704	pcb->pcb_regs.sf_r8 = tf->tf_r8;
705	pcb->pcb_regs.sf_r9 = tf->tf_r9;
706	pcb->pcb_regs.sf_r10 = tf->tf_r10;
707	pcb->pcb_regs.sf_r11 = tf->tf_r11;
708	pcb->pcb_regs.sf_r12 = tf->tf_r12;
709	pcb->pcb_regs.sf_pc = tf->tf_pc;
710	pcb->pcb_regs.sf_lr = tf->tf_usr_lr;
711	pcb->pcb_regs.sf_sp = tf->tf_usr_sp;
712}
713
714void
715pcpu0_init(void)
716{
717#if __ARM_ARCH >= 6
718	set_curthread(&thread0);
719#endif
720	pcpu_init(pcpup, 0, sizeof(struct pcpu));
721	PCPU_SET(curthread, &thread0);
722}
723
724/*
725 * Initialize proc0
726 */
727void
728init_proc0(vm_offset_t kstack)
729{
730	proc_linkup0(&proc0, &thread0);
731	thread0.td_kstack = kstack;
732	thread0.td_pcb = (struct pcb *)
733		(thread0.td_kstack + kstack_pages * PAGE_SIZE) - 1;
734	thread0.td_pcb->pcb_flags = 0;
735	thread0.td_pcb->pcb_vfpcpu = -1;
736	thread0.td_pcb->pcb_vfpstate.fpscr = VFPSCR_DN;
737	thread0.td_frame = &proc0_tf;
738	pcpup->pc_curpcb = thread0.td_pcb;
739}
740
741#if __ARM_ARCH >= 6
742void
743set_stackptrs(int cpu)
744{
745
746	set_stackptr(PSR_IRQ32_MODE,
747	    irqstack + ((IRQ_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
748	set_stackptr(PSR_ABT32_MODE,
749	    abtstack + ((ABT_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
750	set_stackptr(PSR_UND32_MODE,
751	    undstack + ((UND_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
752}
753#else
754void
755set_stackptrs(int cpu)
756{
757
758	set_stackptr(PSR_IRQ32_MODE,
759	    irqstack.pv_va + ((IRQ_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
760	set_stackptr(PSR_ABT32_MODE,
761	    abtstack.pv_va + ((ABT_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
762	set_stackptr(PSR_UND32_MODE,
763	    undstack.pv_va + ((UND_STACK_SIZE * PAGE_SIZE) * (cpu + 1)));
764}
765#endif
766
767
768#ifdef FDT
769#if __ARM_ARCH < 6
770void *
771initarm(struct arm_boot_params *abp)
772{
773	struct mem_region mem_regions[FDT_MEM_REGIONS];
774	struct pv_addr kernel_l1pt;
775	struct pv_addr dpcpu;
776	vm_offset_t dtbp, freemempos, l2_start, lastaddr;
777	uint64_t memsize;
778	uint32_t l2size;
779	char *env;
780	void *kmdp;
781	u_int l1pagetable;
782	int i, j, err_devmap, mem_regions_sz;
783
784	lastaddr = parse_boot_param(abp);
785	arm_physmem_kernaddr = abp->abp_physaddr;
786
787	memsize = 0;
788
789	cpuinfo_init();
790	set_cpufuncs();
791
792	/*
793	 * Find the dtb passed in by the boot loader.
794	 */
795	kmdp = preload_search_by_type("elf kernel");
796	if (kmdp != NULL)
797		dtbp = MD_FETCH(kmdp, MODINFOMD_DTBP, vm_offset_t);
798	else
799		dtbp = (vm_offset_t)NULL;
800
801#if defined(FDT_DTB_STATIC)
802	/*
803	 * In case the device tree blob was not retrieved (from metadata) try
804	 * to use the statically embedded one.
805	 */
806	if (dtbp == (vm_offset_t)NULL)
807		dtbp = (vm_offset_t)&fdt_static_dtb;
808#endif
809
810	if (OF_install(OFW_FDT, 0) == FALSE)
811		panic("Cannot install FDT");
812
813	if (OF_init((void *)dtbp) != 0)
814		panic("OF_init failed with the found device tree");
815
816	/* Grab physical memory regions information from device tree. */
817	if (fdt_get_mem_regions(mem_regions, &mem_regions_sz, &memsize) != 0)
818		panic("Cannot get physical memory regions");
819	arm_physmem_hardware_regions(mem_regions, mem_regions_sz);
820
821	/* Grab reserved memory regions information from device tree. */
822	if (fdt_get_reserved_regions(mem_regions, &mem_regions_sz) == 0)
823		arm_physmem_exclude_regions(mem_regions, mem_regions_sz,
824		    EXFLAG_NODUMP | EXFLAG_NOALLOC);
825
826	/* Platform-specific initialisation */
827	platform_probe_and_attach();
828
829	pcpu0_init();
830
831	/* Do basic tuning, hz etc */
832	init_param1();
833
834	/* Calculate number of L2 tables needed for mapping vm_page_array */
835	l2size = (memsize / PAGE_SIZE) * sizeof(struct vm_page);
836	l2size = (l2size >> L1_S_SHIFT) + 1;
837
838	/*
839	 * Add one table for end of kernel map, one for stacks, msgbuf and
840	 * L1 and L2 tables map and one for vectors map.
841	 */
842	l2size += 3;
843
844	/* Make it divisible by 4 */
845	l2size = (l2size + 3) & ~3;
846
847	freemempos = (lastaddr + PAGE_MASK) & ~PAGE_MASK;
848
849	/* Define a macro to simplify memory allocation */
850#define valloc_pages(var, np)						\
851	alloc_pages((var).pv_va, (np));					\
852	(var).pv_pa = (var).pv_va + (abp->abp_physaddr - KERNVIRTADDR);
853
854#define alloc_pages(var, np)						\
855	(var) = freemempos;						\
856	freemempos += (np * PAGE_SIZE);					\
857	memset((char *)(var), 0, ((np) * PAGE_SIZE));
858
859	while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0)
860		freemempos += PAGE_SIZE;
861	valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE);
862
863	for (i = 0, j = 0; i < l2size; ++i) {
864		if (!(i % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) {
865			valloc_pages(kernel_pt_table[i],
866			    L2_TABLE_SIZE / PAGE_SIZE);
867			j = i;
868		} else {
869			kernel_pt_table[i].pv_va = kernel_pt_table[j].pv_va +
870			    L2_TABLE_SIZE_REAL * (i - j);
871			kernel_pt_table[i].pv_pa =
872			    kernel_pt_table[i].pv_va - KERNVIRTADDR +
873			    abp->abp_physaddr;
874
875		}
876	}
877	/*
878	 * Allocate a page for the system page mapped to 0x00000000
879	 * or 0xffff0000. This page will just contain the system vectors
880	 * and can be shared by all processes.
881	 */
882	valloc_pages(systempage, 1);
883
884	/* Allocate dynamic per-cpu area. */
885	valloc_pages(dpcpu, DPCPU_SIZE / PAGE_SIZE);
886	dpcpu_init((void *)dpcpu.pv_va, 0);
887
888	/* Allocate stacks for all modes */
889	valloc_pages(irqstack, IRQ_STACK_SIZE * MAXCPU);
890	valloc_pages(abtstack, ABT_STACK_SIZE * MAXCPU);
891	valloc_pages(undstack, UND_STACK_SIZE * MAXCPU);
892	valloc_pages(kernelstack, kstack_pages * MAXCPU);
893	valloc_pages(msgbufpv, round_page(msgbufsize) / PAGE_SIZE);
894
895	/*
896	 * Now we start construction of the L1 page table
897	 * We start by mapping the L2 page tables into the L1.
898	 * This means that we can replace L1 mappings later on if necessary
899	 */
900	l1pagetable = kernel_l1pt.pv_va;
901
902	/*
903	 * Try to map as much as possible of kernel text and data using
904	 * 1MB section mapping and for the rest of initial kernel address
905	 * space use L2 coarse tables.
906	 *
907	 * Link L2 tables for mapping remainder of kernel (modulo 1MB)
908	 * and kernel structures
909	 */
910	l2_start = lastaddr & ~(L1_S_OFFSET);
911	for (i = 0 ; i < l2size - 1; i++)
912		pmap_link_l2pt(l1pagetable, l2_start + i * L1_S_SIZE,
913		    &kernel_pt_table[i]);
914
915	pmap_curmaxkvaddr = l2_start + (l2size - 1) * L1_S_SIZE;
916
917	/* Map kernel code and data */
918	pmap_map_chunk(l1pagetable, KERNVIRTADDR, abp->abp_physaddr,
919	   (((uint32_t)(lastaddr) - KERNVIRTADDR) + PAGE_MASK) & ~PAGE_MASK,
920	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
921
922	/* Map L1 directory and allocated L2 page tables */
923	pmap_map_chunk(l1pagetable, kernel_l1pt.pv_va, kernel_l1pt.pv_pa,
924	    L1_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
925
926	pmap_map_chunk(l1pagetable, kernel_pt_table[0].pv_va,
927	    kernel_pt_table[0].pv_pa,
928	    L2_TABLE_SIZE_REAL * l2size,
929	    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
930
931	/* Map allocated DPCPU, stacks and msgbuf */
932	pmap_map_chunk(l1pagetable, dpcpu.pv_va, dpcpu.pv_pa,
933	    freemempos - dpcpu.pv_va,
934	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
935
936	/* Link and map the vector page */
937	pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH,
938	    &kernel_pt_table[l2size - 1]);
939	pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa,
940	    VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE, PTE_CACHE);
941
942	/* Establish static device mappings. */
943	err_devmap = platform_devmap_init();
944	devmap_bootstrap(l1pagetable, NULL);
945	vm_max_kernel_address = platform_lastaddr();
946
947	cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL * 2)) | DOMAIN_CLIENT);
948	pmap_pa = kernel_l1pt.pv_pa;
949	cpu_setttb(kernel_l1pt.pv_pa);
950	cpu_tlb_flushID();
951	cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL * 2));
952
953	/*
954	 * Now that proper page tables are installed, call cpu_setup() to enable
955	 * instruction and data caches and other chip-specific features.
956	 */
957	cpu_setup();
958
959	/*
960	 * Only after the SOC registers block is mapped we can perform device
961	 * tree fixups, as they may attempt to read parameters from hardware.
962	 */
963	OF_interpret("perform-fixup", 0);
964
965	platform_gpio_init();
966
967	cninit();
968
969	debugf("initarm: console initialized\n");
970	debugf(" arg1 kmdp = 0x%08x\n", (uint32_t)kmdp);
971	debugf(" boothowto = 0x%08x\n", boothowto);
972	debugf(" dtbp = 0x%08x\n", (uint32_t)dtbp);
973	arm_print_kenv();
974
975	env = kern_getenv("kernelname");
976	if (env != NULL) {
977		strlcpy(kernelname, env, sizeof(kernelname));
978		freeenv(env);
979	}
980
981	if (err_devmap != 0)
982		printf("WARNING: could not fully configure devmap, error=%d\n",
983		    err_devmap);
984
985	platform_late_init();
986
987	/*
988	 * Pages were allocated during the secondary bootstrap for the
989	 * stacks for different CPU modes.
990	 * We must now set the r13 registers in the different CPU modes to
991	 * point to these stacks.
992	 * Since the ARM stacks use STMFD etc. we must set r13 to the top end
993	 * of the stack memory.
994	 */
995	cpu_control(CPU_CONTROL_MMU_ENABLE, CPU_CONTROL_MMU_ENABLE);
996
997	set_stackptrs(0);
998
999	/*
1000	 * We must now clean the cache again....
1001	 * Cleaning may be done by reading new data to displace any
1002	 * dirty data in the cache. This will have happened in cpu_setttb()
1003	 * but since we are boot strapping the addresses used for the read
1004	 * may have just been remapped and thus the cache could be out
1005	 * of sync. A re-clean after the switch will cure this.
1006	 * After booting there are no gross relocations of the kernel thus
1007	 * this problem will not occur after initarm().
1008	 */
1009	cpu_idcache_wbinv_all();
1010
1011	undefined_init();
1012
1013	init_proc0(kernelstack.pv_va);
1014
1015	arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL);
1016	pmap_bootstrap(freemempos, &kernel_l1pt);
1017	msgbufp = (void *)msgbufpv.pv_va;
1018	msgbufinit(msgbufp, msgbufsize);
1019	mutex_init();
1020
1021	/*
1022	 * Exclude the kernel (and all the things we allocated which immediately
1023	 * follow the kernel) from the VM allocation pool but not from crash
1024	 * dumps.  virtual_avail is a global variable which tracks the kva we've
1025	 * "allocated" while setting up pmaps.
1026	 *
1027	 * Prepare the list of physical memory available to the vm subsystem.
1028	 */
1029	arm_physmem_exclude_region(abp->abp_physaddr,
1030	    (virtual_avail - KERNVIRTADDR), EXFLAG_NOALLOC);
1031	arm_physmem_init_kernel_globals();
1032
1033	init_param2(physmem);
1034	dbg_monitor_init();
1035	kdb_init();
1036
1037	return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP -
1038	    sizeof(struct pcb)));
1039}
1040#else /* __ARM_ARCH < 6 */
1041void *
1042initarm(struct arm_boot_params *abp)
1043{
1044	struct mem_region mem_regions[FDT_MEM_REGIONS];
1045	vm_paddr_t lastaddr;
1046	vm_offset_t dtbp, kernelstack, dpcpu;
1047	char *env;
1048	void *kmdp;
1049	int err_devmap, mem_regions_sz;
1050#ifdef EFI
1051	struct efi_map_header *efihdr;
1052#endif
1053
1054	/* get last allocated physical address */
1055	arm_physmem_kernaddr = abp->abp_physaddr;
1056	lastaddr = parse_boot_param(abp) - KERNVIRTADDR + arm_physmem_kernaddr;
1057
1058	set_cpufuncs();
1059	cpuinfo_init();
1060
1061	/*
1062	 * Find the dtb passed in by the boot loader.
1063	 */
1064	kmdp = preload_search_by_type("elf kernel");
1065	dtbp = MD_FETCH(kmdp, MODINFOMD_DTBP, vm_offset_t);
1066#if defined(FDT_DTB_STATIC)
1067	/*
1068	 * In case the device tree blob was not retrieved (from metadata) try
1069	 * to use the statically embedded one.
1070	 */
1071	if (dtbp == (vm_offset_t)NULL)
1072		dtbp = (vm_offset_t)&fdt_static_dtb;
1073#endif
1074
1075	if (OF_install(OFW_FDT, 0) == FALSE)
1076		panic("Cannot install FDT");
1077
1078	if (OF_init((void *)dtbp) != 0)
1079		panic("OF_init failed with the found device tree");
1080
1081#if defined(LINUX_BOOT_ABI)
1082	arm_parse_fdt_bootargs();
1083#endif
1084
1085#ifdef EFI
1086	efihdr = (struct efi_map_header *)preload_search_info(kmdp,
1087	    MODINFO_METADATA | MODINFOMD_EFI_MAP);
1088	if (efihdr != NULL) {
1089		arm_add_efi_map_entries(efihdr, mem_regions, &mem_regions_sz);
1090	} else
1091#endif
1092	{
1093		/* Grab physical memory regions information from device tree. */
1094		if (fdt_get_mem_regions(mem_regions, &mem_regions_sz,NULL) != 0)
1095			panic("Cannot get physical memory regions");
1096	}
1097	arm_physmem_hardware_regions(mem_regions, mem_regions_sz);
1098
1099	/* Grab reserved memory regions information from device tree. */
1100	if (fdt_get_reserved_regions(mem_regions, &mem_regions_sz) == 0)
1101		arm_physmem_exclude_regions(mem_regions, mem_regions_sz,
1102		    EXFLAG_NODUMP | EXFLAG_NOALLOC);
1103
1104	/*
1105	 * Set TEX remapping registers.
1106	 * Setup kernel page tables and switch to kernel L1 page table.
1107	 */
1108	pmap_set_tex();
1109	pmap_bootstrap_prepare(lastaddr);
1110
1111	/*
1112	 * Now that proper page tables are installed, call cpu_setup() to enable
1113	 * instruction and data caches and other chip-specific features.
1114	 */
1115	cpu_setup();
1116
1117	/* Platform-specific initialisation */
1118	platform_probe_and_attach();
1119	pcpu0_init();
1120
1121	/* Do basic tuning, hz etc */
1122	init_param1();
1123
1124	/*
1125	 * Allocate a page for the system page mapped to 0xffff0000
1126	 * This page will just contain the system vectors and can be
1127	 * shared by all processes.
1128	 */
1129	systempage = pmap_preboot_get_pages(1);
1130
1131	/* Map the vector page. */
1132	pmap_preboot_map_pages(systempage, ARM_VECTORS_HIGH,  1);
1133	if (virtual_end >= ARM_VECTORS_HIGH)
1134		virtual_end = ARM_VECTORS_HIGH - 1;
1135
1136	/* Allocate dynamic per-cpu area. */
1137	dpcpu = pmap_preboot_get_vpages(DPCPU_SIZE / PAGE_SIZE);
1138	dpcpu_init((void *)dpcpu, 0);
1139
1140	/* Allocate stacks for all modes */
1141	irqstack    = pmap_preboot_get_vpages(IRQ_STACK_SIZE * MAXCPU);
1142	abtstack    = pmap_preboot_get_vpages(ABT_STACK_SIZE * MAXCPU);
1143	undstack    = pmap_preboot_get_vpages(UND_STACK_SIZE * MAXCPU );
1144	kernelstack = pmap_preboot_get_vpages(kstack_pages * MAXCPU);
1145
1146	/* Allocate message buffer. */
1147	msgbufp = (void *)pmap_preboot_get_vpages(
1148	    round_page(msgbufsize) / PAGE_SIZE);
1149
1150	/*
1151	 * Pages were allocated during the secondary bootstrap for the
1152	 * stacks for different CPU modes.
1153	 * We must now set the r13 registers in the different CPU modes to
1154	 * point to these stacks.
1155	 * Since the ARM stacks use STMFD etc. we must set r13 to the top end
1156	 * of the stack memory.
1157	 */
1158	set_stackptrs(0);
1159	mutex_init();
1160
1161	/* Establish static device mappings. */
1162	err_devmap = platform_devmap_init();
1163	devmap_bootstrap(0, NULL);
1164	vm_max_kernel_address = platform_lastaddr();
1165
1166	/*
1167	 * Only after the SOC registers block is mapped we can perform device
1168	 * tree fixups, as they may attempt to read parameters from hardware.
1169	 */
1170	OF_interpret("perform-fixup", 0);
1171	platform_gpio_init();
1172	cninit();
1173
1174	debugf("initarm: console initialized\n");
1175	debugf(" arg1 kmdp = 0x%08x\n", (uint32_t)kmdp);
1176	debugf(" boothowto = 0x%08x\n", boothowto);
1177	debugf(" dtbp = 0x%08x\n", (uint32_t)dtbp);
1178	debugf(" lastaddr1: 0x%08x\n", lastaddr);
1179	arm_print_kenv();
1180
1181	env = kern_getenv("kernelname");
1182	if (env != NULL)
1183		strlcpy(kernelname, env, sizeof(kernelname));
1184
1185	if (err_devmap != 0)
1186		printf("WARNING: could not fully configure devmap, error=%d\n",
1187		    err_devmap);
1188
1189	platform_late_init();
1190
1191	/*
1192	 * We must now clean the cache again....
1193	 * Cleaning may be done by reading new data to displace any
1194	 * dirty data in the cache. This will have happened in cpu_setttb()
1195	 * but since we are boot strapping the addresses used for the read
1196	 * may have just been remapped and thus the cache could be out
1197	 * of sync. A re-clean after the switch will cure this.
1198	 * After booting there are no gross relocations of the kernel thus
1199	 * this problem will not occur after initarm().
1200	 */
1201	/* Set stack for exception handlers */
1202	undefined_init();
1203	init_proc0(kernelstack);
1204	arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL);
1205	enable_interrupts(PSR_A);
1206	pmap_bootstrap(0);
1207
1208	/* Exclude the kernel (and all the things we allocated which immediately
1209	 * follow the kernel) from the VM allocation pool but not from crash
1210	 * dumps.  virtual_avail is a global variable which tracks the kva we've
1211	 * "allocated" while setting up pmaps.
1212	 *
1213	 * Prepare the list of physical memory available to the vm subsystem.
1214	 */
1215	arm_physmem_exclude_region(abp->abp_physaddr,
1216		pmap_preboot_get_pages(0) - abp->abp_physaddr, EXFLAG_NOALLOC);
1217	arm_physmem_init_kernel_globals();
1218
1219	init_param2(physmem);
1220	/* Init message buffer. */
1221	msgbufinit(msgbufp, msgbufsize);
1222	dbg_monitor_init();
1223	kdb_init();
1224	return ((void *)STACKALIGN(thread0.td_pcb));
1225
1226}
1227
1228#endif /* __ARM_ARCH < 6 */
1229#endif /* FDT */
1230