Home | History | Annotate | Line # | Download | only in hppa
trap.c revision 1.80
      1 /*	$NetBSD: trap.c,v 1.80 2010/03/20 23:31:27 chs Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2001, 2002 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Matthew Fredette.
      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  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*	$OpenBSD: trap.c,v 1.30 2001/09/19 20:50:56 mickey Exp $	*/
     33 
     34 /*
     35  * Copyright (c) 1998-2004 Michael Shalayeff
     36  * All rights reserved.
     37  *
     38  * Redistribution and use in source and binary forms, with or without
     39  * modification, are permitted provided that the following conditions
     40  * are met:
     41  * 1. Redistributions of source code must retain the above copyright
     42  *    notice, this list of conditions and the following disclaimer.
     43  * 2. Redistributions in binary form must reproduce the above copyright
     44  *    notice, this list of conditions and the following disclaimer in the
     45  *    documentation and/or other materials provided with the distribution.
     46  *
     47  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     48  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     49  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     50  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
     51  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
     52  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     53  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
     55  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
     56  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
     57  * THE POSSIBILITY OF SUCH DAMAGE.
     58  */
     59 
     60 #include <sys/cdefs.h>
     61 __KERNEL_RCSID(0, "$NetBSD: trap.c,v 1.80 2010/03/20 23:31:27 chs Exp $");
     62 
     63 /* #define INTRDEBUG */
     64 /* #define TRAPDEBUG */
     65 /* #define USERTRACE */
     66 
     67 #include "opt_kgdb.h"
     68 #include "opt_ptrace.h"
     69 #include "opt_sa.h"
     70 
     71 #include <sys/param.h>
     72 #include <sys/systm.h>
     73 #include <sys/kernel.h>
     74 #include <sys/syscall.h>
     75 #include <sys/syscallvar.h>
     76 #include <sys/sa.h>
     77 #include <sys/savar.h>
     78 #include <sys/mutex.h>
     79 #include <sys/ktrace.h>
     80 #include <sys/proc.h>
     81 #include <sys/signalvar.h>
     82 #include <sys/acct.h>
     83 #include <sys/signal.h>
     84 #include <sys/device.h>
     85 #include <sys/pool.h>
     86 #include <sys/userret.h>
     87 
     88 #include <net/netisr.h>
     89 
     90 #ifdef KGDB
     91 #include <sys/kgdb.h>
     92 #endif
     93 
     94 #include <uvm/uvm.h>
     95 
     96 #include <machine/iomod.h>
     97 #include <machine/cpufunc.h>
     98 #include <machine/reg.h>
     99 #include <machine/autoconf.h>
    100 
    101 #include <machine/db_machdep.h>
    102 
    103 #include <hppa/hppa/machdep.h>
    104 
    105 #include <ddb/db_output.h>
    106 #include <ddb/db_interface.h>
    107 
    108 #ifdef PTRACE
    109 void ss_clear_breakpoints(struct lwp *l);
    110 int ss_put_value(struct lwp *, vaddr_t, u_int);
    111 int ss_get_value(struct lwp *, vaddr_t, u_int *);
    112 #endif
    113 
    114 /* single-step breakpoint */
    115 #define SSBREAKPOINT   (HPPA_BREAK_KERNEL | (HPPA_BREAK_SS << 13))
    116 
    117 #if defined(DEBUG) || defined(DIAGNOSTIC)
    118 /*
    119  * 0x6fc1000 is a stwm r1, d(sr0, sp), which is the last
    120  * instruction in the function prologue that gcc -O0 uses.
    121  * When we have this instruction we know the relationship
    122  * between the stack pointer and the gcc -O0 frame pointer
    123  * (in r3, loaded with the initial sp) for the body of a
    124  * function.
    125  *
    126  * If the given instruction is a stwm r1, d(sr0, sp) where
    127  * d > 0, we evaluate to d, else we evaluate to zero.
    128  */
    129 #define STWM_R1_D_SR0_SP(inst) \
    130 	(((inst) & 0xffffc001) == 0x6fc10000 ? (((inst) & 0x00003ff) >> 1) : 0)
    131 #endif /* DEBUG || DIAGNOSTIC */
    132 
    133 const char *trap_type[] = {
    134 	"invalid",
    135 	"HPMC",
    136 	"power failure",
    137 	"recovery counter",
    138 	"external interrupt",
    139 	"LPMC",
    140 	"ITLB miss fault",
    141 	"instruction protection",
    142 	"Illegal instruction",
    143 	"break instruction",
    144 	"privileged operation",
    145 	"privileged register",
    146 	"overflow",
    147 	"conditional",
    148 	"assist exception",
    149 	"DTLB miss",
    150 	"ITLB non-access miss",
    151 	"DTLB non-access miss",
    152 	"data protection/rights/alignment",
    153 	"data break",
    154 	"TLB dirty",
    155 	"page reference",
    156 	"assist emulation",
    157 	"higher-priv transfer",
    158 	"lower-priv transfer",
    159 	"taken branch",
    160 	"data access rights",
    161 	"data protection",
    162 	"unaligned data ref",
    163 };
    164 int trap_types = __arraycount(trap_type);
    165 
    166 uint8_t fpopmap[] = {
    167 	0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x00,
    168 	0x00, 0x0c, 0x00, 0x0e, 0x00, 0x00, 0x00, 0x00,
    169 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    170 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    171 	0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0x00,
    172 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    173 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    174 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    175 };
    176 
    177 volatile int astpending;
    178 
    179 void pmap_hptdump(void);
    180 void syscall(struct trapframe *, int *);
    181 
    182 #if defined(DEBUG)
    183 struct trapframe *sanity_frame;
    184 struct lwp *sanity_lwp;
    185 int sanity_checked = 0;
    186 void frame_sanity_check(int, int, struct trapframe *, struct lwp *);
    187 #endif
    188 
    189 
    190 #ifdef USERTRACE
    191 /*
    192  * USERTRACE is a crude facility that traces the PC of
    193  * a single user process.  This tracing is normally
    194  * activated by the dispatching of a certain syscall
    195  * with certain arguments - see the activation code in
    196  * syscall().
    197  */
    198 static void user_backtrace(struct trapframe *, struct lwp *, int);
    199 static void user_backtrace_raw(u_int, u_int);
    200 
    201 u_int rctr_next_iioq;
    202 #endif
    203 
    204 static inline void
    205 userret(struct lwp *l, register_t pc, u_quad_t oticks)
    206 {
    207 	struct proc *p = l->l_proc;
    208 
    209 	if (astpending) {
    210 		astpending = 0;
    211 		if (curcpu()->ci_want_resched) {
    212 			preempt();
    213 		}
    214 	}
    215 
    216 	mi_userret(l);
    217 
    218 	/*
    219 	 * If profiling, charge recent system time to the trapped pc.
    220 	 */
    221 	if (p->p_stflag & PST_PROFIL) {
    222 		extern int psratio;
    223 
    224 		addupc_task(l, pc, (int)(p->p_sticks - oticks) * psratio);
    225 	}
    226 }
    227 
    228 /*
    229  * This handles some messy kernel debugger details.
    230  * It dispatches into either kgdb or DDB, and knows
    231  * about some special things to do, like skipping over
    232  * break instructions and how to really set up for
    233  * a single-step.
    234  */
    235 #if defined(KGDB) || defined(DDB)
    236 static int
    237 trap_kdebug(int type, int code, struct trapframe *frame)
    238 {
    239 	int handled;
    240 	u_int tf_iioq_head_old;
    241 	u_int tf_iioq_tail_old;
    242 
    243 	for(;;) {
    244 
    245 		/* This trap has not been handled. */
    246 		handled = 0;
    247 
    248 		/* Remember the instruction offset queue. */
    249 		tf_iioq_head_old = frame->tf_iioq_head;
    250 		tf_iioq_tail_old = frame->tf_iioq_tail;
    251 
    252 #ifdef	KGDB
    253 		/* Let KGDB handle it (if connected) */
    254 		if (!handled)
    255 			handled = kgdb_trap(type, frame);
    256 #endif
    257 #ifdef	DDB
    258 		/* Let DDB handle it. */
    259 		if (!handled)
    260 			handled = kdb_trap(type, code, frame);
    261 #endif
    262 
    263 		/* If this trap wasn't handled, return now. */
    264 		if (!handled)
    265 			return(0);
    266 
    267 		/*
    268 		 * If the instruction offset queue head changed, but the offset
    269 		 * queue tail didn't, assume that the user wants to jump to the
    270 		 * head offset, and adjust the tail accordingly.  This should
    271 		 * fix the kgdb `jump' command, and can help DDB users who `set'
    272 		 * the offset head but forget the tail.
    273 		 */
    274 		if (frame->tf_iioq_head != tf_iioq_head_old &&
    275 		    frame->tf_iioq_tail == tf_iioq_tail_old)
    276 			frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    277 
    278 		/*
    279 		 * This is some single-stepping support.  If we're trying to
    280 		 * step through a nullified instruction, just advance by hand
    281 		 * and trap again.  Otherwise, load the recovery counter with
    282 		 * zero.
    283 		 */
    284 		if (frame->tf_ipsw & PSW_R) {
    285 #ifdef TRAPDEBUG
    286 			printf("(single stepping at head 0x%x tail 0x%x)\n",
    287 			    frame->tf_iioq_head, frame->tf_iioq_tail);
    288 #endif
    289 			if (frame->tf_ipsw & PSW_N) {
    290 #ifdef TRAPDEBUG
    291 				printf("(single stepping past nullified)\n");
    292 #endif
    293 
    294 				/* Advance the program counter. */
    295 				frame->tf_iioq_head = frame->tf_iioq_tail;
    296 				frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    297 
    298 				/* Clear flags. */
    299 				frame->tf_ipsw &= ~(PSW_N|PSW_X|PSW_Y|PSW_Z|PSW_B|PSW_T|PSW_H|PSW_L);
    300 
    301 				/* Simulate another trap. */
    302 				type = T_RECOVERY;
    303 				continue;
    304 			}
    305 			frame->tf_rctr = 0;
    306 		}
    307 
    308 		/* We handled this trap. */
    309 		return (1);
    310 	}
    311 	/* NOTREACHED */
    312 }
    313 #else	/* !KGDB && !DDB */
    314 #define trap_kdebug(t, c, f)	(0)
    315 #endif	/* !KGDB && !DDB */
    316 
    317 #if defined(DEBUG) || defined(USERTRACE)
    318 /*
    319  * These functions give a crude usermode backtrace.  They really only work when
    320  * code has been compiled without optimization, as they assume a certain func-
    321  * tion prologue sets up a frame pointer and stores the return pointer and arg-
    322  * uments in it.
    323  */
    324 static void
    325 user_backtrace_raw(u_int pc, u_int fp)
    326 {
    327 	int frame_number;
    328 	int arg_number;
    329 
    330 	for (frame_number = 0;
    331 	     frame_number < 100 && pc > HPPA_PC_PRIV_MASK && fp;
    332 	     frame_number++) {
    333 
    334 		printf("%3d: pc=%08x%s fp=0x%08x", frame_number,
    335 		    pc & ~HPPA_PC_PRIV_MASK, USERMODE(pc) ? "  " : "**", fp);
    336 		for(arg_number = 0; arg_number < 4; arg_number++)
    337 			printf(" arg%d=0x%08x", arg_number,
    338 			    (int) fuword(HPPA_FRAME_CARG(arg_number, fp)));
    339 		printf("\n");
    340                 pc = fuword(((register_t *) fp) - 5);	/* fetch rp */
    341 		if (pc == -1) {
    342 			printf("  fuword for pc failed\n");
    343 			break;
    344 		}
    345                 fp = fuword(((register_t *) fp) + 0);	/* fetch previous fp */
    346 		if (fp == -1) {
    347 			printf("  fuword for fp failed\n");
    348 			break;
    349 		}
    350 	}
    351 	printf("  backtrace stopped with pc %08x fp 0x%08x\n", pc, fp);
    352 }
    353 
    354 static void
    355 user_backtrace(struct trapframe *tf, struct lwp *l, int type)
    356 {
    357 	struct proc *p = l->l_proc;
    358 	u_int pc, fp, inst;
    359 
    360 	/*
    361 	 * Display any trap type that we have.
    362 	 */
    363 	if (type >= 0)
    364 		printf("pid %d (%s) trap #%d\n",
    365 		    p->p_pid, p->p_comm, type & ~T_USER);
    366 
    367 	/*
    368 	 * Assuming that the frame pointer in r3 is valid,
    369 	 * dump out a stack trace.
    370 	 */
    371 	fp = tf->tf_r3;
    372 	printf("pid %d (%s) backtrace, starting with fp 0x%08x\n",
    373 		p->p_pid, p->p_comm, fp);
    374 	user_backtrace_raw(tf->tf_iioq_head, fp);
    375 
    376 	/*
    377 	 * In case the frame pointer in r3 is not valid,
    378 	 * assuming the stack pointer is valid and the
    379 	 * faulting function is a non-leaf, if we can
    380 	 * find its prologue we can recover its frame
    381 	 * pointer.
    382 	 */
    383 	pc = tf->tf_iioq_head;
    384 	fp = tf->tf_sp - HPPA_FRAME_SIZE;
    385 	printf("pid %d (%s) backtrace, starting with sp 0x%08x pc 0x%08x\n",
    386 		p->p_pid, p->p_comm, tf->tf_sp, pc);
    387 	for (pc &= ~HPPA_PC_PRIV_MASK; pc > 0; pc -= sizeof(inst)) {
    388 		inst = fuword((register_t *) pc);
    389 		if (inst == -1) {
    390 			printf("  fuword for inst at pc %08x failed\n", pc);
    391 			break;
    392 		}
    393 		/* Check for the prologue instruction that sets sp. */
    394 		if (STWM_R1_D_SR0_SP(inst)) {
    395 			fp = tf->tf_sp - STWM_R1_D_SR0_SP(inst);
    396 			printf("  sp from fp at pc %08x: %08x\n", pc, inst);
    397 			break;
    398 		}
    399 	}
    400 	user_backtrace_raw(tf->tf_iioq_head, fp);
    401 }
    402 #endif /* DEBUG || USERTRACE */
    403 
    404 #ifdef DEBUG
    405 /*
    406  * This sanity-checks a trapframe.  It is full of various
    407  * assumptions about what a healthy CPU state should be,
    408  * with some documented elsewhere, some not.
    409  */
    410 void
    411 frame_sanity_check(int where, int type, struct trapframe *tf, struct lwp *l)
    412 {
    413 	extern int kernel_text;
    414 	extern int etext;
    415 	extern register_t kpsw;
    416 
    417 #define SANITY(e)					\
    418 do {							\
    419 	if (sanity_frame == NULL && !(e)) {		\
    420 		sanity_frame = tf;			\
    421 		sanity_lwp = l;				\
    422 		sanity_checked = __LINE__;		\
    423 	}						\
    424 } while (/* CONSTCOND */ 0)
    425 
    426 	KASSERT(l != NULL);
    427 	SANITY((tf->tf_ipsw & kpsw) == kpsw);
    428 	SANITY((kpsw & PSW_I) == 0 || tf->tf_eiem != 0);
    429 	if (tf->tf_iisq_head == HPPA_SID_KERNEL) {
    430 		vaddr_t minsp, maxsp, uv;
    431 
    432 		uv = uvm_lwp_getuarea(l);
    433 
    434 		/*
    435 		 * If the trap happened in the gateway
    436 		 * page, we take the easy way out and
    437 		 * assume that the trapframe is okay.
    438 		 */
    439 		if ((tf->tf_iioq_head & ~PAGE_MASK) == SYSCALLGATE)
    440 			goto out;
    441 
    442 		SANITY(!USERMODE(tf->tf_iioq_head));
    443 		SANITY(!USERMODE(tf->tf_iioq_tail));
    444 
    445 		/*
    446 		 * Don't check the instruction queues or stack on interrupts
    447 		 * as we could be be in the sti code (outside normal kernel
    448 		 * text) or switching LWPs (curlwp and sp are not in sync)
    449 		 */
    450 		if ((type & ~T_USER) == T_INTERRUPT)
    451 			goto out;
    452 
    453 		SANITY(tf->tf_iioq_head >= (u_int) &kernel_text);
    454 		SANITY(tf->tf_iioq_head < (u_int) &etext);
    455 		SANITY(tf->tf_iioq_tail >= (u_int) &kernel_text);
    456 		SANITY(tf->tf_iioq_tail < (u_int) &etext);
    457 
    458 		maxsp = uv + USPACE + PAGE_SIZE;
    459 		minsp = uv + PAGE_SIZE;
    460 
    461 		SANITY(tf->tf_sp >= minsp && tf->tf_sp < maxsp);
    462 	} else {
    463 		struct pcb *pcb = lwp_getpcb(l);
    464 
    465 		SANITY(USERMODE(tf->tf_iioq_head));
    466 		SANITY(USERMODE(tf->tf_iioq_tail));
    467 		SANITY(tf->tf_cr30 == (u_int)pcb->pcb_fpregs);
    468 	}
    469 #undef SANITY
    470 out:
    471 	if (sanity_frame == tf) {
    472 		printf("insanity: where 0x%x type 0x%x tf %p lwp %p line %d "
    473 		       "sp 0x%x pc 0x%x\n",
    474 		       where, type, sanity_frame, sanity_lwp, sanity_checked,
    475 		       tf->tf_sp, tf->tf_iioq_head);
    476 		(void) trap_kdebug(T_IBREAK, 0, tf);
    477 		sanity_frame = NULL;
    478 		sanity_lwp = NULL;
    479 		sanity_checked = 0;
    480 	}
    481 }
    482 #endif /* DEBUG */
    483 
    484 void
    485 trap(int type, struct trapframe *frame)
    486 {
    487 	struct lwp *l;
    488 	struct proc *p;
    489 	struct pcb *pcb;
    490 	vaddr_t va;
    491 	struct vm_map *map;
    492 	struct vmspace *vm;
    493 	vm_prot_t vftype;
    494 	pa_space_t space;
    495 	ksiginfo_t ksi;
    496 	u_int opcode, onfault;
    497 	int ret;
    498 	const char *tts = "reserved";
    499 	int trapnum;
    500 #ifdef DIAGNOSTIC
    501 	extern int emergency_stack_start, emergency_stack_end;
    502 	int oldcpl = cpl;
    503 #endif
    504 
    505 	trapnum = type & ~T_USER;
    506 	opcode = frame->tf_iir;
    507 	if (trapnum == T_ITLBMISS || trapnum == T_ITLBMISSNA ||
    508 	    trapnum == T_IBREAK || trapnum == T_TAKENBR) {
    509 		va = frame->tf_iioq_head;
    510 		space = frame->tf_iisq_head;
    511 		vftype = VM_PROT_EXECUTE;
    512 	} else {
    513 		va = frame->tf_ior;
    514 		space = frame->tf_isr;
    515 		vftype = inst_store(opcode) ? VM_PROT_WRITE : VM_PROT_READ;
    516 	}
    517 
    518 	KASSERT(curlwp != NULL);
    519 	l = curlwp;
    520 	p = l->l_proc;
    521 	if ((type & T_USER) != 0)
    522 		LWP_CACHE_CREDS(l, p);
    523 
    524 #ifdef DIAGNOSTIC
    525 	/*
    526 	 * If we are on the emergency stack, then we either got
    527 	 * a fault on the kernel stack, or we're just handling
    528 	 * a trap for the machine check handler (which also
    529 	 * runs on the emergency stack).
    530 	 *
    531 	 * We *very crudely* differentiate between the two cases
    532 	 * by checking the faulting instruction: if it is the
    533 	 * function prologue instruction that stores the old
    534 	 * frame pointer and updates the stack pointer, we assume
    535 	 * that we faulted on the kernel stack.
    536 	 *
    537 	 * In this case, not completing that instruction will
    538 	 * probably confuse backtraces in kgdb/ddb.  Completing
    539 	 * it would be difficult, because we already faulted on
    540 	 * that part of the stack, so instead we fix up the
    541 	 * frame as if the function called has just returned.
    542 	 * This has peculiar knowledge about what values are in
    543 	 * what registers during the "normal gcc -g" prologue.
    544 	 */
    545 	if (&type >= &emergency_stack_start &&
    546 	    &type < &emergency_stack_end &&
    547 	    type != T_IBREAK && STWM_R1_D_SR0_SP(opcode)) {
    548 		/* Restore the caller's frame pointer. */
    549 		frame->tf_r3 = frame->tf_r1;
    550 		/* Restore the caller's instruction offsets. */
    551 		frame->tf_iioq_head = frame->tf_rp;
    552 		frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    553 		goto dead_end;
    554 	}
    555 #endif /* DIAGNOSTIC */
    556 
    557 #ifdef DEBUG
    558 	frame_sanity_check(0xdead01, type, frame, l);
    559 #endif /* DEBUG */
    560 
    561 	if (frame->tf_flags & TFF_LAST)
    562 		l->l_md.md_regs = frame;
    563 
    564 	if (trapnum <= trap_types)
    565 		tts = trap_type[trapnum];
    566 
    567 #ifdef TRAPDEBUG
    568 	if (trapnum != T_INTERRUPT && trapnum != T_IBREAK)
    569 		printf("trap: %d, %s for %x:%lx at %x:%x, fp=%p, rp=%x\n",
    570 		    type, tts, space, va, frame->tf_iisq_head,
    571 		    frame->tf_iioq_head, frame, frame->tf_rp);
    572 	else if (trapnum == T_IBREAK)
    573 		printf("trap: break instruction %x:%x at %x:%x, fp=%p\n",
    574 		    break5(opcode), break13(opcode),
    575 		    frame->tf_iisq_head, frame->tf_iioq_head, frame);
    576 
    577 	{
    578 		extern int etext;
    579 		if (frame < (struct trapframe *)&etext) {
    580 			printf("trap: bogus frame ptr %p\n", frame);
    581 			goto dead_end;
    582 		}
    583 	}
    584 #endif
    585 
    586 	pcb = lwp_getpcb(l);
    587 
    588 	/* If this is a trap, not an interrupt, reenable interrupts. */
    589 	if (trapnum != T_INTERRUPT) {
    590 		uvmexp.traps++;
    591 		mtctl(frame->tf_eiem, CR_EIEM);
    592 	}
    593 
    594 	switch (type) {
    595 	case T_NONEXIST:
    596 	case T_NONEXIST|T_USER:
    597 #if !defined(DDB) && !defined(KGDB)
    598 		/* we've got screwed up by the central scrutinizer */
    599 		panic ("trap: elvis has just left the building!");
    600 		break;
    601 #else
    602 		goto dead_end;
    603 #endif
    604 	case T_RECOVERY|T_USER:
    605 #ifdef USERTRACE
    606 		for(;;) {
    607 			if (frame->tf_iioq_head != rctr_next_iioq)
    608 				printf("-%08x\nr %08x",
    609 					rctr_next_iioq - 4,
    610 					frame->tf_iioq_head);
    611 			rctr_next_iioq = frame->tf_iioq_head + 4;
    612 			if (frame->tf_ipsw & PSW_N) {
    613 				/* Advance the program counter. */
    614 				frame->tf_iioq_head = frame->tf_iioq_tail;
    615 				frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    616 				/* Clear flags. */
    617 				frame->tf_ipsw &= ~(PSW_N|PSW_X|PSW_Y|PSW_Z|PSW_B|PSW_T|PSW_H|PSW_L);
    618 				/* Simulate another trap. */
    619 				continue;
    620 			}
    621 			break;
    622 		}
    623 		frame->tf_rctr = 0;
    624 		break;
    625 #endif /* USERTRACE */
    626 	case T_RECOVERY:
    627 #if !defined(DDB) && !defined(KGDB)
    628 		/* XXX will implement later */
    629 		printf ("trap: handicapped");
    630 		break;
    631 #else
    632 		goto dead_end;
    633 #endif
    634 
    635 	case T_EMULATION | T_USER:
    636 #ifdef FPEMUL
    637 		hppa_fpu_emulate(frame, l, opcode);
    638 #else  /* !FPEMUL */
    639 		/*
    640 		 * We don't have FPU emulation, so signal the
    641 		 * process with a SIGFPE.
    642 		 */
    643 
    644 		KSI_INIT_TRAP(&ksi);
    645 		ksi.ksi_signo = SIGFPE;
    646 		ksi.ksi_code = SI_NOINFO;
    647 		ksi.ksi_trap = type;
    648 		ksi.ksi_addr = (void *)frame->tf_iioq_head;
    649 		trapsignal(l, &ksi);
    650 #endif /* !FPEMUL */
    651 		break;
    652 
    653 	case T_DATALIGN:
    654 		onfault = pcb->pcb_onfault;
    655 		if (onfault) {
    656 			ret = EFAULT;
    657 do_onfault:
    658 			frame->tf_iioq_head = onfault;
    659 			frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    660 			frame->tf_ret0 = ret;
    661 			break;
    662 		}
    663 		/*FALLTHROUGH*/
    664 
    665 #ifdef DIAGNOSTIC
    666 		/* these just can't happen ever */
    667 	case T_PRIV_OP:
    668 	case T_PRIV_REG:
    669 		/* these just can't make it to the trap() ever */
    670 	case T_HPMC:
    671 	case T_HPMC | T_USER:
    672 	case T_EMULATION:
    673 	case T_EXCEPTION:
    674 #endif
    675 	case T_IBREAK:
    676 	case T_DBREAK:
    677 	dead_end:
    678 		if (type & T_USER) {
    679 #ifdef DEBUG
    680 			user_backtrace(frame, l, type);
    681 #endif
    682 			KSI_INIT_TRAP(&ksi);
    683 			ksi.ksi_signo = SIGILL;
    684 			ksi.ksi_code = ILL_ILLTRP;
    685 			ksi.ksi_trap = type;
    686 			ksi.ksi_addr = (void *)frame->tf_iioq_head;
    687 			trapsignal(l, &ksi);
    688 			break;
    689 		}
    690 		if (trap_kdebug(type, va, frame))
    691 			return;
    692 		else if (type == T_DATALIGN)
    693 			panic ("trap: %s at 0x%x", tts, (u_int) va);
    694 		else
    695 			panic ("trap: no debugger for \"%s\" (%d)", tts, type);
    696 		break;
    697 
    698 	case T_IBREAK | T_USER:
    699 	case T_DBREAK | T_USER:
    700 		KSI_INIT_TRAP(&ksi);
    701 		ksi.ksi_signo = SIGTRAP;
    702 		ksi.ksi_code = TRAP_TRACE;
    703 		ksi.ksi_trap = trapnum;
    704 		ksi.ksi_addr = (void *)frame->tf_iioq_head;
    705 #ifdef PTRACE
    706 		ss_clear_breakpoints(l);
    707 		if (opcode == SSBREAKPOINT)
    708 			ksi.ksi_code = TRAP_BRKPT;
    709 #endif
    710 		/* pass to user debugger */
    711 		trapsignal(l, &ksi);
    712 
    713 		break;
    714 
    715 #ifdef PTRACE
    716 	case T_TAKENBR | T_USER:
    717 		ss_clear_breakpoints(l);
    718 
    719 		KSI_INIT_TRAP(&ksi);
    720 		ksi.ksi_signo = SIGTRAP;
    721 		ksi.ksi_code = TRAP_TRACE;
    722 		ksi.ksi_trap = trapnum;
    723 		ksi.ksi_addr = (void *)frame->tf_iioq_head;
    724 
    725                 /* pass to user debugger */
    726 		trapsignal(l, &ksi);
    727 		break;
    728 #endif
    729 
    730 	case T_EXCEPTION | T_USER: {	/* co-proc assist trap */
    731 		uint64_t *fpp;
    732 		uint32_t *pex, ex, inst;
    733 		int i;
    734 
    735 		hppa_fpu_flush(l);
    736 		fpp = (uint64_t *)pcb->pcb_fpregs;
    737 
    738 		/* skip the status register */
    739 		pex = (uint32_t *)&fpp[0];
    740 		pex++;
    741 
    742 		/* loop through the exception registers */
    743 		for (i = 1; i < 8 && !*pex; i++, pex++)
    744 			;
    745 		KASSERT(i < 8);
    746 		ex = *pex;
    747 		*pex = 0;
    748 
    749 		/* reset the trap flag, as if there was none */
    750 		fpp[0] &= ~(((uint64_t)HPPA_FPU_T) << 32);
    751 
    752 		/* emulate the instruction */
    753 		inst = ((uint32_t)fpopmap[ex >> 26] << 26) | (ex & 0x03ffffff);
    754 		hppa_fpu_emulate(frame, l, inst);
    755 		}
    756 		break;
    757 
    758 	case T_OVERFLOW | T_USER:
    759 		KSI_INIT_TRAP(&ksi);
    760 		ksi.ksi_signo = SIGFPE;
    761 		ksi.ksi_code = SI_NOINFO;
    762 		ksi.ksi_trap = type;
    763 		ksi.ksi_addr = (void *)va;
    764 		trapsignal(l, &ksi);
    765 		break;
    766 
    767 	case T_CONDITION | T_USER:
    768 		KSI_INIT_TRAP(&ksi);
    769 		ksi.ksi_signo = SIGFPE;
    770 		ksi.ksi_code = FPE_INTDIV;
    771 		ksi.ksi_trap = type;
    772 		ksi.ksi_addr = (void *)va;
    773 		trapsignal(l, &ksi);
    774 		break;
    775 
    776 	case T_ILLEGAL | T_USER:
    777 #ifdef DEBUG
    778 		user_backtrace(frame, l, type);
    779 #endif
    780 		KSI_INIT_TRAP(&ksi);
    781 		ksi.ksi_signo = SIGILL;
    782 		ksi.ksi_code = ILL_ILLOPC;
    783 		ksi.ksi_trap = type;
    784 		ksi.ksi_addr = (void *)va;
    785 		trapsignal(l, &ksi);
    786 		break;
    787 
    788 	case T_PRIV_OP | T_USER:
    789 #ifdef DEBUG
    790 		user_backtrace(frame, l, type);
    791 #endif
    792 		KSI_INIT_TRAP(&ksi);
    793 		ksi.ksi_signo = SIGILL;
    794 		ksi.ksi_code = ILL_PRVOPC;
    795 		ksi.ksi_trap = type;
    796 		ksi.ksi_addr = (void *)va;
    797 		trapsignal(l, &ksi);
    798 		break;
    799 
    800 	case T_PRIV_REG | T_USER:
    801 #ifdef DEBUG
    802 		user_backtrace(frame, l, type);
    803 #endif
    804 		KSI_INIT_TRAP(&ksi);
    805 		ksi.ksi_signo = SIGILL;
    806 		ksi.ksi_code = ILL_PRVREG;
    807 		ksi.ksi_trap = type;
    808 		ksi.ksi_addr = (void *)va;
    809 		trapsignal(l, &ksi);
    810 		break;
    811 
    812 		/* these should never got here */
    813 	case T_HIGHERPL | T_USER:
    814 	case T_LOWERPL | T_USER:
    815 		KSI_INIT_TRAP(&ksi);
    816 		ksi.ksi_signo = SIGSEGV;
    817 		ksi.ksi_code = SEGV_ACCERR;
    818 		ksi.ksi_trap = type;
    819 		ksi.ksi_addr = (void *)va;
    820 		trapsignal(l, &ksi);
    821 		break;
    822 
    823 	case T_IPROT | T_USER:
    824 	case T_DPROT | T_USER:
    825 		KSI_INIT_TRAP(&ksi);
    826 		ksi.ksi_signo = SIGSEGV;
    827 		ksi.ksi_code = SEGV_ACCERR;
    828 		ksi.ksi_trap = type;
    829 		ksi.ksi_addr = (void *)va;
    830 		trapsignal(l, &ksi);
    831 		break;
    832 
    833 	case T_DATACC:   	case T_USER | T_DATACC:
    834 	case T_ITLBMISS:	case T_USER | T_ITLBMISS:
    835 	case T_DTLBMISS:	case T_USER | T_DTLBMISS:
    836 	case T_ITLBMISSNA:	case T_USER | T_ITLBMISSNA:
    837 	case T_DTLBMISSNA:	case T_USER | T_DTLBMISSNA:
    838 	case T_TLB_DIRTY:	case T_USER | T_TLB_DIRTY:
    839 		vm = p->p_vmspace;
    840 
    841 		if (!vm) {
    842 #ifdef TRAPDEBUG
    843 			printf("trap: no vm, p=%p\n", p);
    844 #endif
    845 			goto dead_end;
    846 		}
    847 
    848 		/*
    849 		 * it could be a kernel map for exec_map faults
    850 		 */
    851 		if (!(type & T_USER) && space == HPPA_SID_KERNEL)
    852 			map = kernel_map;
    853 		else {
    854 			map = &vm->vm_map;
    855 			if ((l->l_flag & LW_SA)
    856 			    && (~l->l_pflag & LP_SA_NOBLOCK)) {
    857 				l->l_savp->savp_faultaddr = va;
    858 				l->l_pflag |= LP_SA_PAGEFAULT;
    859 			}
    860 		}
    861 
    862 		va = trunc_page(va);
    863 
    864 		if (map->pmap->pm_space != space) {
    865 #ifdef TRAPDEBUG
    866 			printf("trap: space mismatch %d != %d\n",
    867 			    space, map->pmap->pm_space);
    868 #endif
    869 			/* actually dump the user, crap the kernel */
    870 			goto dead_end;
    871 		}
    872 
    873 		/* Never call uvm_fault in interrupt context. */
    874 		KASSERT(hppa_intr_depth == 0);
    875 
    876 		onfault = pcb->pcb_onfault;
    877 		pcb->pcb_onfault = 0;
    878 		ret = uvm_fault(map, va, vftype);
    879 		pcb->pcb_onfault = onfault;
    880 
    881 #ifdef TRAPDEBUG
    882 		printf("uvm_fault(%p, %x, %d)=%d\n",
    883 		    map, (u_int)va, vftype, ret);
    884 #endif
    885 
    886 		if (map != kernel_map)
    887 			l->l_pflag &= ~LP_SA_PAGEFAULT;
    888 
    889 		/*
    890 		 * If this was a stack access we keep track of the maximum
    891 		 * accessed stack size.  Also, if uvm_fault gets a protection
    892 		 * failure it is due to accessing the stack region outside
    893 		 * the current limit and we need to reflect that as an access
    894 		 * error.
    895 		 */
    896 		if (map != kernel_map && va >= (vaddr_t)vm->vm_minsaddr) {
    897 			if (ret == 0)
    898 				uvm_grow(l->l_proc, va);
    899 			else if (ret == EACCES)
    900 				ret = EFAULT;
    901 		}
    902 
    903 		if (ret != 0) {
    904 			if (type & T_USER) {
    905 #ifdef DEBUG
    906 				user_backtrace(frame, l, type);
    907 #endif
    908 				KSI_INIT_TRAP(&ksi);
    909 				ksi.ksi_signo = SIGSEGV;
    910 				ksi.ksi_code = (ret == EACCES ?
    911 						SEGV_ACCERR : SEGV_MAPERR);
    912 				ksi.ksi_trap = type;
    913 				ksi.ksi_addr = (void *)va;
    914 				trapsignal(l, &ksi);
    915 			} else {
    916 				if (onfault) {
    917 					goto do_onfault;
    918 				}
    919 				panic("trap: uvm_fault(%p, %lx, %d): %d",
    920 				    map, va, vftype, ret);
    921 			}
    922 		}
    923 		break;
    924 
    925 	case T_DATALIGN | T_USER:
    926 #ifdef DEBUG
    927 		user_backtrace(frame, l, type);
    928 #endif
    929 		KSI_INIT_TRAP(&ksi);
    930 		ksi.ksi_signo = SIGBUS;
    931 		ksi.ksi_code = BUS_ADRALN;
    932 		ksi.ksi_trap = type;
    933 		ksi.ksi_addr = (void *)va;
    934 		trapsignal(l, &ksi);
    935 		break;
    936 
    937 	case T_INTERRUPT:
    938 	case T_INTERRUPT|T_USER:
    939 		hppa_intr(frame);
    940 		mtctl(frame->tf_eiem, CR_EIEM);
    941 		break;
    942 
    943 	case T_LOWERPL:
    944 	case T_DPROT:
    945 	case T_IPROT:
    946 	case T_OVERFLOW:
    947 	case T_CONDITION:
    948 	case T_ILLEGAL:
    949 	case T_HIGHERPL:
    950 	case T_TAKENBR:
    951 	case T_POWERFAIL:
    952 	case T_LPMC:
    953 	case T_PAGEREF:
    954 	case T_DATAPID:  	case T_DATAPID  | T_USER:
    955 		if (0 /* T-chip */) {
    956 			break;
    957 		}
    958 		/* FALLTHROUGH to unimplemented */
    959 	default:
    960 		panic ("trap: unimplemented \'%s\' (%d)", tts, type);
    961 	}
    962 
    963 #ifdef DIAGNOSTIC
    964 	if (cpl != oldcpl)
    965 		printf("WARNING: SPL (%d) NOT LOWERED ON TRAP (%d) EXIT\n",
    966 		    cpl, trapnum);
    967 #endif
    968 
    969 	if (type & T_USER)
    970 		userret(l, l->l_md.md_regs->tf_iioq_head, 0);
    971 
    972 #ifdef DEBUG
    973 	frame_sanity_check(0xdead02, type, frame, l);
    974 	if (frame->tf_flags & TFF_LAST && (curlwp->l_flag & LW_IDLE) == 0)
    975 		frame_sanity_check(0xdead03, type, curlwp->l_md.md_regs,
    976 				   curlwp);
    977 #endif /* DEBUG */
    978 }
    979 
    980 void
    981 child_return(void *arg)
    982 {
    983 	struct lwp *l = arg;
    984 
    985 	/*
    986 	 * Return values in the frame set by cpu_lwp_fork().
    987 	 */
    988 
    989 	userret(l, l->l_md.md_regs->tf_iioq_head, 0);
    990 	ktrsysret(SYS_fork, 0, 0);
    991 #ifdef DEBUG
    992 	frame_sanity_check(0xdead04, 0, l->l_md.md_regs, l);
    993 #endif /* DEBUG */
    994 }
    995 
    996 #ifdef PTRACE
    997 
    998 #include <sys/ptrace.h>
    999 
   1000 int
   1001 ss_get_value(struct lwp *l, vaddr_t addr, u_int *value)
   1002 {
   1003 	struct uio uio;
   1004 	struct iovec iov;
   1005 
   1006 	iov.iov_base = (void *)value;
   1007 	iov.iov_len = sizeof(u_int);
   1008 	uio.uio_iov = &iov;
   1009 	uio.uio_iovcnt = 1;
   1010 	uio.uio_offset = (off_t)addr;
   1011 	uio.uio_resid = sizeof(u_int);
   1012 	uio.uio_rw = UIO_READ;
   1013 	UIO_SETUP_SYSSPACE(&uio);
   1014 
   1015 	return (process_domem(curlwp, l, &uio));
   1016 }
   1017 
   1018 int
   1019 ss_put_value(struct lwp *l, vaddr_t addr, u_int value)
   1020 {
   1021 	struct uio uio;
   1022 	struct iovec iov;
   1023 
   1024 	iov.iov_base = (void *)&value;
   1025 	iov.iov_len = sizeof(u_int);
   1026 	uio.uio_iov = &iov;
   1027 	uio.uio_iovcnt = 1;
   1028 	uio.uio_offset = (off_t)addr;
   1029 	uio.uio_resid = sizeof(u_int);
   1030 	uio.uio_rw = UIO_WRITE;
   1031 	UIO_SETUP_SYSSPACE(&uio);
   1032 
   1033 	return (process_domem(curlwp, l, &uio));
   1034 }
   1035 
   1036 void
   1037 ss_clear_breakpoints(struct lwp *l)
   1038 {
   1039 	/* Restore origional instructions. */
   1040 	if (l->l_md.md_bpva != 0) {
   1041 		ss_put_value(l, l->l_md.md_bpva, l->l_md.md_bpsave[0]);
   1042 		ss_put_value(l, l->l_md.md_bpva + 4, l->l_md.md_bpsave[1]);
   1043 		l->l_md.md_bpva = 0;
   1044 	}
   1045 }
   1046 
   1047 
   1048 int
   1049 process_sstep(struct lwp *l, int sstep)
   1050 {
   1051 	struct trapframe *tf = l->l_md.md_regs;
   1052 	int error;
   1053 
   1054 	ss_clear_breakpoints(l);
   1055 
   1056 	/* We're continuing... */
   1057 	/* Don't touch the syscall gateway page. */
   1058 	/* XXX head */
   1059 	if (sstep == 0 ||
   1060 	    (tf->tf_iioq_tail & ~PAGE_MASK) == SYSCALLGATE) {
   1061 		tf->tf_ipsw &= ~PSW_T;
   1062 		return 0;
   1063 	}
   1064 
   1065 	l->l_md.md_bpva = tf->tf_iioq_tail & ~HPPA_PC_PRIV_MASK;
   1066 
   1067 	/*
   1068 	 * Insert two breakpoint instructions; the first one might be
   1069 	 * nullified.  Of course we need to save two instruction
   1070 	 * first.
   1071 	 */
   1072 
   1073 	error = ss_get_value(l, l->l_md.md_bpva, &l->l_md.md_bpsave[0]);
   1074 	if (error)
   1075 		return (error);
   1076 	error = ss_get_value(l, l->l_md.md_bpva + 4, &l->l_md.md_bpsave[1]);
   1077 	if (error)
   1078 		return (error);
   1079 
   1080 	error = ss_put_value(l, l->l_md.md_bpva, SSBREAKPOINT);
   1081 	if (error)
   1082 		return error;
   1083 	error = ss_put_value(l, l->l_md.md_bpva + 4, SSBREAKPOINT);
   1084 	if (error)
   1085 		return error;
   1086 
   1087 	tf->tf_ipsw |= PSW_T;
   1088 
   1089 	return 0;
   1090 }
   1091 #endif
   1092 
   1093 
   1094 /*
   1095  * call actual syscall routine
   1096  * from the low-level syscall handler:
   1097  * - all HPPA_FRAME_NARGS syscall's arguments supposed to be copied onto
   1098  *   our stack, this wins compared to copyin just needed amount anyway
   1099  * - register args are copied onto stack too
   1100  */
   1101 void
   1102 syscall(struct trapframe *frame, int *args)
   1103 {
   1104 	struct lwp *l;
   1105 	struct proc *p;
   1106 	const struct sysent *callp;
   1107 	size_t nargs64;
   1108 	int nsys, code, error;
   1109 	int tmp;
   1110 	int rval[2];
   1111 #ifdef DIAGNOSTIC
   1112 	int oldcpl = cpl;
   1113 #endif
   1114 
   1115 	uvmexp.syscalls++;
   1116 
   1117 #ifdef DEBUG
   1118 	frame_sanity_check(0xdead04, 0, frame, curlwp);
   1119 #endif /* DEBUG */
   1120 
   1121 	if (!USERMODE(frame->tf_iioq_head))
   1122 		panic("syscall");
   1123 
   1124 	KASSERT(curlwp != NULL);
   1125 	l = curlwp;
   1126 	p = l->l_proc;
   1127 	l->l_md.md_regs = frame;
   1128 	nsys = p->p_emul->e_nsysent;
   1129 	callp = p->p_emul->e_sysent;
   1130 	code = frame->tf_t1;
   1131 	LWP_CACHE_CREDS(l, p);
   1132 
   1133 #ifdef KERN_SA
   1134 	if (__predict_false((l->l_savp)
   1135             && (l->l_savp->savp_pflags & SAVP_FLAG_DELIVERING)))
   1136 		l->l_savp->savp_pflags &= ~SAVP_FLAG_DELIVERING;
   1137 #endif
   1138 
   1139 	/*
   1140 	 * Restarting a system call is touchy on the HPPA, because syscall
   1141 	 * arguments are passed in registers and the program counter of the
   1142 	 * syscall "point" isn't easily divined.
   1143 	 *
   1144 	 * We handle the first problem by assuming that we will have to restart
   1145 	 * this system call, so we stuff the first four words of the original
   1146 	 * arguments back into the frame as arg0...arg3, which is where we
   1147 	 * found them in the first place.  Any further arguments are (still) on
   1148 	 * the user's stack and the  syscall code will fetch them from there
   1149 	 * (again).
   1150 	 *
   1151 	 * The program counter problem is addressed below.
   1152 	 */
   1153 	frame->tf_arg0 = args[0];
   1154 	frame->tf_arg1 = args[1];
   1155 	frame->tf_arg2 = args[2];
   1156 	frame->tf_arg3 = args[3];
   1157 
   1158 	/*
   1159 	 * Some special handling for the syscall(2) and
   1160 	 * __syscall(2) system calls.
   1161 	 */
   1162 	switch (code) {
   1163 	case SYS_syscall:
   1164 		code = *args;
   1165 		args += 1;
   1166 		break;
   1167 	case SYS___syscall:
   1168 		if (callp != sysent)
   1169 			break;
   1170 		/*
   1171 		 * NB: even though __syscall(2) takes a quad_t containing the
   1172 		 * system call number, because our argument copying word-swaps
   1173 		 * 64-bit arguments, the least significant word of that quad_t
   1174 		 * is the first word in the argument array.
   1175 		 */
   1176 		code = *args;
   1177 		args += 2;
   1178 	}
   1179 
   1180 	/*
   1181 	 * Stacks growing from lower addresses to higher addresses are not
   1182 	 * really such a good idea, because it makes it impossible to overlay a
   1183 	 * struct on top of C stack arguments (the arguments appear in
   1184 	 * reversed order).
   1185 	 *
   1186 	 * You can do the obvious thing (as locore.S does) and copy argument
   1187 	 * words one by one, laying them out in the "right" order in the dest-
   1188 	 * ination buffer, but this ends up word-swapping multi-word arguments
   1189 	 * (like off_t).
   1190 	 *
   1191 	 * FIXME - this works only on native binaries and
   1192 	 * will probably screw up any and all emulation.
   1193 	 *
   1194 	 */
   1195 
   1196 	if (code < 0 || code >= nsys)
   1197 		callp += p->p_emul->e_nosys;	/* bad syscall # */
   1198 	else
   1199 		callp += code;
   1200 
   1201 	nargs64 = SYCALL_NARGS64(callp);
   1202 	if (nargs64 != 0) {
   1203 		size_t nargs = callp->sy_narg;
   1204 
   1205 		for (size_t i = 0; i < nargs + nargs64;) {
   1206 			if (SYCALL_ARG_64_P(callp, i)) {
   1207 				tmp = args[i];
   1208 				args[i] = args[i + 1];
   1209 				args[i + 1] = tmp;
   1210 				i += 2;
   1211 			} else
   1212 				i++;
   1213 		}
   1214 	}
   1215 
   1216 #ifdef USERTRACE
   1217 	if (0) {
   1218 		user_backtrace(frame, l, -1);
   1219 		frame->tf_ipsw |= PSW_R;
   1220 		frame->tf_rctr = 0;
   1221 		printf("r %08x", frame->tf_iioq_head);
   1222 		rctr_next_iioq = frame->tf_iioq_head + 4;
   1223 	}
   1224 #endif
   1225 
   1226 	error = 0;
   1227 	if (__predict_false(p->p_trace_enabled)) {
   1228 		error = trace_enter(code, args, callp->sy_narg);
   1229 		if (error)
   1230 			goto out;
   1231 	}
   1232 
   1233 	rval[0] = 0;
   1234 	rval[1] = 0;
   1235 	error = sy_call(callp, l, args, rval);
   1236 out:
   1237 	switch (error) {
   1238 	case 0:
   1239 		l = curlwp;			/* changes on exec() */
   1240 		frame = l->l_md.md_regs;
   1241 		frame->tf_ret0 = rval[0];
   1242 		frame->tf_ret1 = rval[1];
   1243 		frame->tf_t1 = 0;
   1244 		break;
   1245 	case ERESTART:
   1246 		/*
   1247 		 * Now we have to wind back the instruction offset queue to the
   1248 		 * point where the system call will be made again.  This is
   1249 		 * inherently tied to the SYSCALL macro.
   1250 		 *
   1251 		 * Currently, the part of the SYSCALL macro that we want to re-
   1252 		 * run reads as:
   1253 		 *
   1254 		 *	ldil	L%SYSCALLGATE, r1
   1255 		 *	ble	4(sr7, r1)
   1256 		 *	ldi	__CONCAT(SYS_,x), t1
   1257 		 *	comb,<>	%r0, %t1, __cerror
   1258 		 *
   1259 		 * And our offset queue head points to the comb instruction.
   1260 		 * So we need to subtract twelve to reach the ldil.
   1261 		 */
   1262 		frame->tf_iioq_head -= 12;
   1263 		frame->tf_iioq_tail = frame->tf_iioq_head + 4;
   1264 		break;
   1265 	case EJUSTRETURN:
   1266 		p = curproc;
   1267 		break;
   1268 	default:
   1269 		if (p->p_emul->e_errno)
   1270 			error = p->p_emul->e_errno[error];
   1271 		frame->tf_t1 = error;
   1272 		break;
   1273 	}
   1274 
   1275 	if (__predict_false(p->p_trace_enabled))
   1276 		trace_exit(code, rval, error);
   1277 
   1278 	userret(l, frame->tf_iioq_head, 0);
   1279 
   1280 #ifdef DIAGNOSTIC
   1281 	if (cpl != oldcpl) {
   1282 		printf("WARNING: SPL (0x%x) NOT LOWERED ON "
   1283 		    "syscall(0x%x, 0x%x, 0x%x, 0x%x...) EXIT, PID %d\n",
   1284 		    cpl, code, args[0], args[1], args[2], p->p_pid);
   1285 		cpl = oldcpl;
   1286 	}
   1287 #endif
   1288 
   1289 #ifdef DEBUG
   1290 	frame_sanity_check(0xdead05, 0, frame, l);
   1291 #endif /* DEBUG */
   1292 }
   1293 
   1294 /*
   1295  * Start a new LWP
   1296  */
   1297 void
   1298 startlwp(void *arg)
   1299 {
   1300 	int err;
   1301 	ucontext_t *uc = arg;
   1302 	struct lwp *l = curlwp;
   1303 
   1304 	err = cpu_setmcontext(l, &uc->uc_mcontext, uc->uc_flags);
   1305 #if DIAGNOSTIC
   1306 	if (err) {
   1307 		printf("Error %d from cpu_setmcontext.", err);
   1308 	}
   1309 #endif
   1310 	pool_put(&lwp_uc_pool, uc);
   1311 
   1312 	userret(l, l->l_md.md_regs->tf_iioq_head, 0);
   1313 }
   1314 
   1315 /*
   1316  * XXX This is a terrible name.
   1317  */
   1318 void
   1319 upcallret(struct lwp *l)
   1320 {
   1321 	userret(l, l->l_md.md_regs->tf_iioq_head, 0);
   1322 }
   1323