Home | History | Annotate | Line # | Download | only in hppa
trap.c revision 1.69
      1 /*	$NetBSD: trap.c,v 1.69 2010/01/16 07:56:16 skrll 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 1.69 2010/01/16 07:56:16 skrll 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 = sizeof(trap_type)/sizeof(trap_type[0]);
    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 (curcpu()->ci_want_resched) {
    210 		preempt();
    211 	}
    212 
    213 	mi_userret(l);
    214 
    215 	/*
    216 	 * If profiling, charge recent system time to the trapped pc.
    217 	 */
    218 	if (p->p_stflag & PST_PROFIL) {
    219 		extern int psratio;
    220 
    221 		addupc_task(l, pc, (int)(p->p_sticks - oticks) * psratio);
    222 	}
    223 }
    224 
    225 /*
    226  * This handles some messy kernel debugger details.
    227  * It dispatches into either kgdb or DDB, and knows
    228  * about some special things to do, like skipping over
    229  * break instructions and how to really set up for
    230  * a single-step.
    231  */
    232 #if defined(KGDB) || defined(DDB)
    233 static int
    234 trap_kdebug(int type, int code, struct trapframe *frame)
    235 {
    236 	int handled;
    237 	u_int tf_iioq_head_old;
    238 	u_int tf_iioq_tail_old;
    239 
    240 	for(;;) {
    241 
    242 		/* This trap has not been handled. */
    243 		handled = 0;
    244 
    245 		/* Remember the instruction offset queue. */
    246 		tf_iioq_head_old = frame->tf_iioq_head;
    247 		tf_iioq_tail_old = frame->tf_iioq_tail;
    248 
    249 #ifdef	KGDB
    250 		/* Let KGDB handle it (if connected) */
    251 		if (!handled)
    252 			handled = kgdb_trap(type, frame);
    253 #endif
    254 #ifdef	DDB
    255 		/* Let DDB handle it. */
    256 		if (!handled)
    257 			handled = kdb_trap(type, code, frame);
    258 #endif
    259 
    260 		/* If this trap wasn't handled, return now. */
    261 		if (!handled)
    262 			return(0);
    263 
    264 		/*
    265 		 * If the instruction offset queue head changed,
    266 		 * but the offset queue tail didn't, assume that
    267 		 * the user wants to jump to the head offset, and
    268 		 * adjust the tail accordingly.  This should fix
    269 		 * the kgdb `jump' command, and can help DDB users
    270 		 * who `set' the offset head but forget the tail.
    271 		 */
    272 		if (frame->tf_iioq_head != tf_iioq_head_old &&
    273 		    frame->tf_iioq_tail == tf_iioq_tail_old)
    274 			frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    275 
    276 		/*
    277 		 * This is some single-stepping support.
    278 		 * If we're trying to step through a nullified
    279 		 * instruction, just advance by hand and trap
    280 		 * again.  Otherwise, load the recovery counter
    281 		 * with zero.
    282 		 */
    283 		if (frame->tf_ipsw & PSW_R) {
    284 #ifdef TRAPDEBUG
    285 			printf("(single stepping at head 0x%x tail 0x%x)\n",
    286 			    frame->tf_iioq_head, frame->tf_iioq_tail);
    287 #endif
    288 			if (frame->tf_ipsw & PSW_N) {
    289 #ifdef TRAPDEBUG
    290 				printf("(single stepping past nullified)\n");
    291 #endif
    292 
    293 				/* Advance the program counter. */
    294 				frame->tf_iioq_head = frame->tf_iioq_tail;
    295 				frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    296 
    297 				/* Clear flags. */
    298 				frame->tf_ipsw &= ~(PSW_N|PSW_X|PSW_Y|PSW_Z|PSW_B|PSW_T|PSW_H|PSW_L);
    299 
    300 				/* Simulate another trap. */
    301 				type = T_RECOVERY;
    302 				continue;
    303 			}
    304 			frame->tf_rctr = 0;
    305 		}
    306 
    307 		/* We handled this trap. */
    308 		return (1);
    309 	}
    310 	/* NOTREACHED */
    311 }
    312 #else	/* !KGDB && !DDB */
    313 #define trap_kdebug(t, c, f)	(0)
    314 #endif	/* !KGDB && !DDB */
    315 
    316 #if defined(DEBUG) || defined(USERTRACE)
    317 /*
    318  * These functions give a crude usermode backtrace.  They
    319  * really only work when code has been compiled without
    320  * optimization, as they assume a certain function prologue
    321  * sets up a frame pointer and stores the return pointer
    322  * and arguments 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 	SANITY((tf->tf_ipsw & kpsw) == kpsw);
    427 	SANITY((kpsw & PSW_I) == 0 || tf->tf_eiem != 0);
    428 	if (tf->tf_iisq_head == HPPA_SID_KERNEL) {
    429 		vaddr_t minsp, maxsp, uv;
    430 
    431 		uv = uvm_lwp_getuarea(l);
    432 
    433 		/*
    434 		 * If the trap happened in the gateway
    435 		 * page, we take the easy way out and
    436 		 * assume that the trapframe is okay.
    437 		 */
    438 		if ((tf->tf_iioq_head & ~PAGE_MASK) == SYSCALLGATE)
    439 			goto out;
    440 
    441 		SANITY(!USERMODE(tf->tf_iioq_head));
    442 		SANITY(!USERMODE(tf->tf_iioq_tail));
    443 
    444 		/*
    445 		 * Don't check the instruction queues or stack on interrupts
    446 		 * as we could be be in the sti code (outside normal kernel
    447 		 * text) or switching LWPs (curlwp and sp are not in sync)
    448 		 */
    449 		if ((type & ~T_USER) == T_INTERRUPT)
    450 			goto out;
    451 
    452 		SANITY(tf->tf_iioq_head >= (u_int) &kernel_text);
    453 		SANITY(tf->tf_iioq_head < (u_int) &etext);
    454 		SANITY(tf->tf_iioq_tail >= (u_int) &kernel_text);
    455 		SANITY(tf->tf_iioq_tail < (u_int) &etext);
    456 
    457 		maxsp = uv + USPACE + PAGE_SIZE;
    458 		minsp = uv + PAGE_SIZE;
    459 
    460 		SANITY(l != NULL || (tf->tf_sp >= minsp && tf->tf_sp < maxsp));
    461 	} else {
    462 		SANITY(USERMODE(tf->tf_iioq_head));
    463 		SANITY(USERMODE(tf->tf_iioq_tail));
    464 		SANITY(l != NULL &&
    465 		    tf->tf_cr30 == kvtop((void *)uvm_lwp_getuarea(l)));
    466 	}
    467 #undef SANITY
    468 out:
    469 	if (sanity_frame == tf) {
    470 		printf("insanity: where 0x%x type 0x%x tf %p lwp %p line %d "
    471 		       "sp 0x%x pc 0x%x\n",
    472 		       where, type, sanity_frame, sanity_lwp, sanity_checked,
    473 		       tf->tf_sp, tf->tf_iioq_head);
    474 		(void) trap_kdebug(T_IBREAK, 0, tf);
    475 		sanity_frame = NULL;
    476 		sanity_lwp = NULL;
    477 		sanity_checked = 0;
    478 	}
    479 }
    480 #endif /* DEBUG */
    481 
    482 void
    483 trap(int type, struct trapframe *frame)
    484 {
    485 	struct lwp *l;
    486 	struct proc *p;
    487 	struct pcb *pcb;
    488 	vaddr_t va;
    489 	struct vm_map *map;
    490 	struct vmspace *vm;
    491 	vm_prot_t vftype;
    492 	pa_space_t space;
    493 	ksiginfo_t ksi;
    494 	u_int opcode, onfault;
    495 	int ret;
    496 	const char *tts;
    497 	int trapnum;
    498 #ifdef DIAGNOSTIC
    499 	extern int emergency_stack_start, emergency_stack_end;
    500 	int oldcpl = cpl;
    501 #endif
    502 
    503 	trapnum = type & ~T_USER;
    504 	opcode = frame->tf_iir;
    505 	if (trapnum == T_ITLBMISS || trapnum == T_ITLBMISSNA ||
    506 	    trapnum == T_IBREAK || trapnum == T_TAKENBR) {
    507 		va = frame->tf_iioq_head;
    508 		space = frame->tf_iisq_head;
    509 		vftype = VM_PROT_EXECUTE;
    510 	} else {
    511 		va = frame->tf_ior;
    512 		space = frame->tf_isr;
    513 		vftype = inst_store(opcode) ? VM_PROT_WRITE : VM_PROT_READ;
    514 	}
    515 
    516 	KASSERT(curlwp != NULL);
    517 	l = curlwp;
    518 	p = l->l_proc;
    519 	if ((type & T_USER) != 0)
    520 		LWP_CACHE_CREDS(l, p);
    521 
    522 	tts = (trapnum > trap_types) ? "reserved" : trap_type[trapnum];
    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 #ifdef TRAPDEBUG
    565 	if (trapnum != T_INTERRUPT && trapnum != T_IBREAK)
    566 		printf("trap: %d, %s for %x:%x at %x:%x, fp=%p, rp=%x\n",
    567 		    type, tts, space, (u_int)va, frame->tf_iisq_head,
    568 		    frame->tf_iioq_head, frame, frame->tf_rp);
    569 	else if (trapnum == T_IBREAK)
    570 		printf("trap: break instruction %x:%x at %x:%x, fp=%p\n",
    571 		    break5(opcode), break13(opcode),
    572 		    frame->tf_iisq_head, frame->tf_iioq_head, frame);
    573 
    574 	{
    575 		extern int etext;
    576 		if (frame < (struct trapframe *)&etext) {
    577 			printf("trap: bogus frame ptr %p\n", frame);
    578 			goto dead_end;
    579 		}
    580 	}
    581 #endif
    582 	pcb = lwp_getpcb(l);
    583 
    584 	/* If this is a trap, not an interrupt, reenable interrupts. */
    585 	if (trapnum != T_INTERRUPT) {
    586 		uvmexp.traps++;
    587 		mtctl(frame->tf_eiem, CR_EIEM);
    588 	}
    589 
    590 	switch (type) {
    591 	case T_NONEXIST:
    592 	case T_NONEXIST|T_USER:
    593 #if !defined(DDB) && !defined(KGDB)
    594 		/* we've got screwed up by the central scrutinizer */
    595 		panic ("trap: elvis has just left the building!");
    596 		break;
    597 #else
    598 		goto dead_end;
    599 #endif
    600 	case T_RECOVERY|T_USER:
    601 #ifdef USERTRACE
    602 		for(;;) {
    603 			if (frame->tf_iioq_head != rctr_next_iioq)
    604 				printf("-%08x\nr %08x",
    605 					rctr_next_iioq - 4,
    606 					frame->tf_iioq_head);
    607 			rctr_next_iioq = frame->tf_iioq_head + 4;
    608 			if (frame->tf_ipsw & PSW_N) {
    609 				/* Advance the program counter. */
    610 				frame->tf_iioq_head = frame->tf_iioq_tail;
    611 				frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    612 				/* Clear flags. */
    613 				frame->tf_ipsw &= ~(PSW_N|PSW_X|PSW_Y|PSW_Z|PSW_B|PSW_T|PSW_H|PSW_L);
    614 				/* Simulate another trap. */
    615 				continue;
    616 			}
    617 			break;
    618 		}
    619 		frame->tf_rctr = 0;
    620 		break;
    621 #endif /* USERTRACE */
    622 	case T_RECOVERY:
    623 #if !defined(DDB) && !defined(KGDB)
    624 		/* XXX will implement later */
    625 		printf ("trap: handicapped");
    626 		break;
    627 #else
    628 		goto dead_end;
    629 #endif
    630 
    631 	case T_EMULATION | T_USER:
    632 #ifdef FPEMUL
    633 		hppa_fpu_emulate(frame, l, opcode);
    634 #else  /* !FPEMUL */
    635 		/*
    636 		 * We don't have FPU emulation, so signal the
    637 		 * process with a SIGFPE.
    638 		 */
    639 
    640 		KSI_INIT_TRAP(&ksi);
    641 		ksi.ksi_signo = SIGFPE;
    642 		ksi.ksi_code = SI_NOINFO;
    643 		ksi.ksi_trap = type;
    644 		ksi.ksi_addr = (void *)frame->tf_iioq_head;
    645 		trapsignal(l, &ksi);
    646 #endif /* !FPEMUL */
    647 		break;
    648 
    649 	case T_DATALIGN:
    650 		if (pcb->pcb_onfault) {
    651 do_onfault:
    652 			frame->tf_iioq_head = pcb->pcb_onfault;
    653 			frame->tf_iioq_tail = 4 + frame->tf_iioq_head;
    654 			pcb->pcb_onfault = 0;
    655 			break;
    656 		}
    657 		/*FALLTHROUGH*/
    658 
    659 #ifdef DIAGNOSTIC
    660 		/* these just can't happen ever */
    661 	case T_PRIV_OP:
    662 	case T_PRIV_REG:
    663 		/* these just can't make it to the trap() ever */
    664 	case T_HPMC:
    665 	case T_HPMC | T_USER:
    666 	case T_EMULATION:
    667 	case T_EXCEPTION:
    668 #endif
    669 	case T_IBREAK:
    670 	case T_DBREAK:
    671 	dead_end:
    672 		if (type & T_USER) {
    673 #ifdef DEBUG
    674 			user_backtrace(frame, l, type);
    675 #endif
    676 			KSI_INIT_TRAP(&ksi);
    677 			ksi.ksi_signo = SIGILL;
    678 			ksi.ksi_code = ILL_ILLTRP;
    679 			ksi.ksi_trap = type;
    680 			ksi.ksi_addr = (void *)frame->tf_iioq_head;
    681 			trapsignal(l, &ksi);
    682 			break;
    683 		}
    684 		if (trap_kdebug(type, va, frame))
    685 			return;
    686 		else if (type == T_DATALIGN)
    687 			panic ("trap: %s at 0x%x", tts, (u_int) va);
    688 		else
    689 			panic ("trap: no debugger for \"%s\" (%d)", tts, type);
    690 		break;
    691 
    692 	case T_IBREAK | T_USER:
    693 	case T_DBREAK | T_USER:
    694 		KSI_INIT_TRAP(&ksi);
    695 		ksi.ksi_signo = SIGTRAP;
    696 		ksi.ksi_code = TRAP_TRACE;
    697 		ksi.ksi_trap = trapnum;
    698 		ksi.ksi_addr = (void *)frame->tf_iioq_head;
    699 #ifdef PTRACE
    700 		ss_clear_breakpoints(l);
    701 		if (opcode == SSBREAKPOINT)
    702 			ksi.ksi_code = TRAP_BRKPT;
    703 #endif
    704 		/* pass to user debugger */
    705 		trapsignal(l, &ksi);
    706 
    707 		break;
    708 
    709 #ifdef PTRACE
    710 	case T_TAKENBR | T_USER:
    711 		ss_clear_breakpoints(l);
    712 
    713 		KSI_INIT_TRAP(&ksi);
    714 		ksi.ksi_signo = SIGTRAP;
    715 		ksi.ksi_code = TRAP_TRACE;
    716 		ksi.ksi_trap = trapnum;
    717 		ksi.ksi_addr = (void *)frame->tf_iioq_head;
    718 
    719                 /* pass to user debugger */
    720 		trapsignal(l, &ksi);
    721 		break;
    722 #endif
    723 
    724 	case T_EXCEPTION | T_USER: {	/* co-proc assist trap */
    725 		uint64_t *fpp;
    726 		uint32_t *pex, ex, inst;
    727 		int i;
    728 
    729 		hppa_fpu_flush(l);
    730 		fpp = pcb->pcb_fpregs;
    731 		pex = (uint32_t *)&fpp[1];
    732 		for (i = 1; i < 8 && !*pex; i++, pex++)
    733 			;
    734 		KASSERT(i < 8);
    735 		ex = *pex;
    736 		*pex = 0;
    737 
    738 		/* reset the trap flag, as if there was none */
    739 		fpp[0] &= ~(((uint64_t)HPPA_FPU_T) << 32);
    740 
    741 		/* emulate the instruction */
    742 		inst = ((uint32_t)fpopmap[ex >> 26] << 26) | (ex & 0x03ffffff);
    743 		hppa_fpu_emulate(frame, l, inst);
    744 		}
    745 		break;
    746 
    747 	case T_OVERFLOW | T_USER:
    748 		KSI_INIT_TRAP(&ksi);
    749 		ksi.ksi_signo = SIGFPE;
    750 		ksi.ksi_code = SI_NOINFO;
    751 		ksi.ksi_trap = type;
    752 		ksi.ksi_addr = (void *)va;
    753 		trapsignal(l, &ksi);
    754 		break;
    755 
    756 	case T_CONDITION | T_USER:
    757 		KSI_INIT_TRAP(&ksi);
    758 		ksi.ksi_signo = SIGFPE;
    759 		ksi.ksi_code = FPE_INTDIV;
    760 		ksi.ksi_trap = type;
    761 		ksi.ksi_addr = (void *)va;
    762 		trapsignal(l, &ksi);
    763 		break;
    764 
    765 	case T_ILLEGAL | T_USER:
    766 #ifdef DEBUG
    767 		user_backtrace(frame, l, type);
    768 #endif
    769 		KSI_INIT_TRAP(&ksi);
    770 		ksi.ksi_signo = SIGILL;
    771 		ksi.ksi_code = ILL_ILLOPC;
    772 		ksi.ksi_trap = type;
    773 		ksi.ksi_addr = (void *)va;
    774 		trapsignal(l, &ksi);
    775 		break;
    776 
    777 	case T_PRIV_OP | T_USER:
    778 #ifdef DEBUG
    779 		user_backtrace(frame, l, type);
    780 #endif
    781 		KSI_INIT_TRAP(&ksi);
    782 		ksi.ksi_signo = SIGILL;
    783 		ksi.ksi_code = ILL_PRVOPC;
    784 		ksi.ksi_trap = type;
    785 		ksi.ksi_addr = (void *)va;
    786 		trapsignal(l, &ksi);
    787 		break;
    788 
    789 	case T_PRIV_REG | T_USER:
    790 #ifdef DEBUG
    791 		user_backtrace(frame, l, type);
    792 #endif
    793 		KSI_INIT_TRAP(&ksi);
    794 		ksi.ksi_signo = SIGILL;
    795 		ksi.ksi_code = ILL_PRVREG;
    796 		ksi.ksi_trap = type;
    797 		ksi.ksi_addr = (void *)va;
    798 		trapsignal(l, &ksi);
    799 		break;
    800 
    801 		/* these should never got here */
    802 	case T_HIGHERPL | T_USER:
    803 	case T_LOWERPL | T_USER:
    804 		KSI_INIT_TRAP(&ksi);
    805 		ksi.ksi_signo = SIGSEGV;
    806 		ksi.ksi_code = SEGV_ACCERR;
    807 		ksi.ksi_trap = type;
    808 		ksi.ksi_addr = (void *)va;
    809 		trapsignal(l, &ksi);
    810 		break;
    811 
    812 	case T_IPROT | T_USER:
    813 	case T_DPROT | T_USER:
    814 		KSI_INIT_TRAP(&ksi);
    815 		ksi.ksi_signo = SIGSEGV;
    816 		ksi.ksi_code = SEGV_ACCERR;
    817 		ksi.ksi_trap = type;
    818 		ksi.ksi_addr = (void *)va;
    819 		trapsignal(l, &ksi);
    820 		break;
    821 
    822 	case T_DATACC:   	case T_USER | T_DATACC:
    823 	case T_ITLBMISS:	case T_USER | T_ITLBMISS:
    824 	case T_DTLBMISS:	case T_USER | T_DTLBMISS:
    825 	case T_ITLBMISSNA:	case T_USER | T_ITLBMISSNA:
    826 	case T_DTLBMISSNA:	case T_USER | T_DTLBMISSNA:
    827 	case T_TLB_DIRTY:	case T_USER | T_TLB_DIRTY:
    828 		vm = p->p_vmspace;
    829 
    830 		if (!vm) {
    831 #ifdef TRAPDEBUG
    832 			printf("trap: no vm, p=%p\n", p);
    833 #endif
    834 			goto dead_end;
    835 		}
    836 
    837 		/*
    838 		 * it could be a kernel map for exec_map faults
    839 		 */
    840 		if (!(type & T_USER) && space == HPPA_SID_KERNEL)
    841 			map = kernel_map;
    842 		else {
    843 			map = &vm->vm_map;
    844 			if ((l->l_flag & LW_SA)
    845 			    && (~l->l_pflag & LP_SA_NOBLOCK)) {
    846 				l->l_savp->savp_faultaddr = va;
    847 				l->l_pflag |= LP_SA_PAGEFAULT;
    848 			}
    849 		}
    850 
    851 		va = trunc_page(va);
    852 
    853 		if (map->pmap->pm_space != space) {
    854 #ifdef TRAPDEBUG
    855 			printf("trap: space mismatch %d != %d\n",
    856 			    space, map->pmap->pm_space);
    857 #endif
    858 			/* actually dump the user, crap the kernel */
    859 			goto dead_end;
    860 		}
    861 
    862 		/* Never call uvm_fault in interrupt context. */
    863 		KASSERT(hppa_intr_depth == 0);
    864 
    865 		onfault = pcb->pcb_onfault;
    866 		pcb->pcb_onfault = 0;
    867 		ret = uvm_fault(map, va, vftype);
    868 		pcb->pcb_onfault = onfault;
    869 
    870 #ifdef TRAPDEBUG
    871 		printf("uvm_fault(%p, %x, %d)=%d\n",
    872 		    map, (u_int)va, vftype, ret);
    873 #endif
    874 
    875 		if (map != kernel_map)
    876 			l->l_pflag &= ~LP_SA_PAGEFAULT;
    877 
    878 		/*
    879 		 * If this was a stack access we keep track of the maximum
    880 		 * accessed stack size.  Also, if uvm_fault gets a protection
    881 		 * failure it is due to accessing the stack region outside
    882 		 * the current limit and we need to reflect that as an access
    883 		 * error.
    884 		 */
    885 		if (map != kernel_map && va >= (vaddr_t)vm->vm_minsaddr) {
    886 			if (ret == 0)
    887 				uvm_grow(l->l_proc, va);
    888 			else if (ret == EACCES)
    889 				ret = EFAULT;
    890 		}
    891 
    892 		if (ret != 0) {
    893 			if (type & T_USER) {
    894 #ifdef DEBUG
    895 				user_backtrace(frame, l, type);
    896 #endif
    897 				KSI_INIT_TRAP(&ksi);
    898 				ksi.ksi_signo = SIGSEGV;
    899 				ksi.ksi_code = (ret == EACCES ?
    900 						SEGV_ACCERR : SEGV_MAPERR);
    901 				ksi.ksi_trap = type;
    902 				ksi.ksi_addr = (void *)va;
    903 				trapsignal(l, &ksi);
    904 			} else {
    905 				if (pcb->pcb_onfault) {
    906 					goto do_onfault;
    907 				}
    908 				panic("trap: uvm_fault(%p, %lx, %d): %d",
    909 				    map, va, vftype, ret);
    910 			}
    911 		}
    912 		break;
    913 
    914 	case T_DATALIGN | T_USER:
    915 #ifdef DEBUG
    916 		user_backtrace(frame, l, type);
    917 #endif
    918 		KSI_INIT_TRAP(&ksi);
    919 		ksi.ksi_signo = SIGBUS;
    920 		ksi.ksi_code = BUS_ADRALN;
    921 		ksi.ksi_trap = type;
    922 		ksi.ksi_addr = (void *)va;
    923 		trapsignal(l, &ksi);
    924 		break;
    925 
    926 	case T_INTERRUPT:
    927 	case T_INTERRUPT|T_USER:
    928 		hppa_intr(frame);
    929 		mtctl(frame->tf_eiem, CR_EIEM);
    930 		break;
    931 
    932 	case T_LOWERPL:
    933 	case T_DPROT:
    934 	case T_IPROT:
    935 	case T_OVERFLOW:
    936 	case T_CONDITION:
    937 	case T_ILLEGAL:
    938 	case T_HIGHERPL:
    939 	case T_TAKENBR:
    940 	case T_POWERFAIL:
    941 	case T_LPMC:
    942 	case T_PAGEREF:
    943 	case T_DATAPID:  	case T_DATAPID  | T_USER:
    944 		if (0 /* T-chip */) {
    945 			break;
    946 		}
    947 		/* FALLTHROUGH to unimplemented */
    948 	default:
    949 		panic ("trap: unimplemented \'%s\' (%d)", tts, type);
    950 	}
    951 
    952 #ifdef DIAGNOSTIC
    953 	if (cpl != oldcpl)
    954 		printf("WARNING: SPL (%d) NOT LOWERED ON TRAP (%d) EXIT\n",
    955 		    cpl, trapnum);
    956 #endif
    957 
    958 	if (type & T_USER)
    959 		userret(l, l->l_md.md_regs->tf_iioq_head, 0);
    960 
    961 #ifdef DEBUG
    962 	frame_sanity_check(0xdead02, type, frame, l);
    963 	if (frame->tf_flags & TFF_LAST && (curlwp->l_flag & LW_IDLE) == 0)
    964 		frame_sanity_check(0xdead03, type, curlwp->l_md.md_regs,
    965 				   curlwp);
    966 #endif /* DEBUG */
    967 }
    968 
    969 void
    970 child_return(void *arg)
    971 {
    972 	struct lwp *l = arg;
    973 
    974 	/*
    975 	 * Return values in the frame set by cpu_lwp_fork().
    976 	 */
    977 
    978 	userret(l, l->l_md.md_regs->tf_iioq_head, 0);
    979 	ktrsysret(SYS_fork, 0, 0);
    980 #ifdef DEBUG
    981 	frame_sanity_check(0xdead04, 0, l->l_md.md_regs, l);
    982 #endif /* DEBUG */
    983 }
    984 
    985 #ifdef PTRACE
    986 
    987 #include <sys/ptrace.h>
    988 
    989 int
    990 ss_get_value(struct lwp *l, vaddr_t addr, u_int *value)
    991 {
    992 	struct uio uio;
    993 	struct iovec iov;
    994 
    995 	iov.iov_base = (void *)value;
    996 	iov.iov_len = sizeof(u_int);
    997 	uio.uio_iov = &iov;
    998 	uio.uio_iovcnt = 1;
    999 	uio.uio_offset = (off_t)addr;
   1000 	uio.uio_resid = sizeof(u_int);
   1001 	uio.uio_rw = UIO_READ;
   1002 	UIO_SETUP_SYSSPACE(&uio);
   1003 
   1004 	return (process_domem(curlwp, l, &uio));
   1005 }
   1006 
   1007 int
   1008 ss_put_value(struct lwp *l, vaddr_t addr, u_int value)
   1009 {
   1010 	struct uio uio;
   1011 	struct iovec iov;
   1012 
   1013 	iov.iov_base = (void *)&value;
   1014 	iov.iov_len = sizeof(u_int);
   1015 	uio.uio_iov = &iov;
   1016 	uio.uio_iovcnt = 1;
   1017 	uio.uio_offset = (off_t)addr;
   1018 	uio.uio_resid = sizeof(u_int);
   1019 	uio.uio_rw = UIO_WRITE;
   1020 	UIO_SETUP_SYSSPACE(&uio);
   1021 
   1022 	return (process_domem(curlwp, l, &uio));
   1023 }
   1024 
   1025 void
   1026 ss_clear_breakpoints(struct lwp *l)
   1027 {
   1028 	/* Restore origional instructions. */
   1029 	if (l->l_md.md_bpva != 0) {
   1030 		ss_put_value(l, l->l_md.md_bpva, l->l_md.md_bpsave[0]);
   1031 		ss_put_value(l, l->l_md.md_bpva + 4, l->l_md.md_bpsave[1]);
   1032 		l->l_md.md_bpva = 0;
   1033 	}
   1034 }
   1035 
   1036 
   1037 int
   1038 process_sstep(struct lwp *l, int sstep)
   1039 {
   1040 	struct trapframe *tf = l->l_md.md_regs;
   1041 	int error;
   1042 
   1043 	ss_clear_breakpoints(l);
   1044 
   1045 	/* We're continuing... */
   1046 	/* Don't touch the syscall gateway page. */
   1047 	/* XXX head */
   1048 	if (sstep == 0 ||
   1049 	    (tf->tf_iioq_tail & ~PAGE_MASK) == SYSCALLGATE) {
   1050 		tf->tf_ipsw &= ~PSW_T;
   1051 		return 0;
   1052 	}
   1053 
   1054 	l->l_md.md_bpva = tf->tf_iioq_tail & ~HPPA_PC_PRIV_MASK;
   1055 
   1056 	/*
   1057 	 * Insert two breakpoint instructions; the first one might be
   1058 	 * nullified.  Of course we need to save two instruction
   1059 	 * first.
   1060 	 */
   1061 
   1062 	error = ss_get_value(l, l->l_md.md_bpva, &l->l_md.md_bpsave[0]);
   1063 	if (error)
   1064 		return (error);
   1065 	error = ss_get_value(l, l->l_md.md_bpva + 4, &l->l_md.md_bpsave[1]);
   1066 	if (error)
   1067 		return (error);
   1068 
   1069 	error = ss_put_value(l, l->l_md.md_bpva, SSBREAKPOINT);
   1070 	if (error)
   1071 		return error;
   1072 	error = ss_put_value(l, l->l_md.md_bpva + 4, SSBREAKPOINT);
   1073 	if (error)
   1074 		return error;
   1075 
   1076 	tf->tf_ipsw |= PSW_T;
   1077 
   1078 	return 0;
   1079 }
   1080 #endif
   1081 
   1082 
   1083 /*
   1084  * call actual syscall routine
   1085  * from the low-level syscall handler:
   1086  * - all HPPA_FRAME_NARGS syscall's arguments supposed to be copied onto
   1087  *   our stack, this wins compared to copyin just needed amount anyway
   1088  * - register args are copied onto stack too
   1089  */
   1090 void
   1091 syscall(struct trapframe *frame, int *args)
   1092 {
   1093 	struct lwp *l;
   1094 	struct proc *p;
   1095 	const struct sysent *callp;
   1096 	size_t nargs64;
   1097 	int nsys, code, error;
   1098 	int tmp;
   1099 	int rval[2];
   1100 #ifdef DIAGNOSTIC
   1101 	int oldcpl = cpl;
   1102 #endif
   1103 
   1104 	uvmexp.syscalls++;
   1105 
   1106 #ifdef DEBUG
   1107 	frame_sanity_check(0xdead04, 0, frame, curlwp);
   1108 #endif /* DEBUG */
   1109 
   1110 	if (!USERMODE(frame->tf_iioq_head))
   1111 		panic("syscall");
   1112 
   1113 	KASSERT(curlwp != NULL);
   1114 	l = curlwp;
   1115 	p = l->l_proc;
   1116 	l->l_md.md_regs = frame;
   1117 	nsys = p->p_emul->e_nsysent;
   1118 	callp = p->p_emul->e_sysent;
   1119 	code = frame->tf_t1;
   1120 	LWP_CACHE_CREDS(l, p);
   1121 
   1122 #ifdef KERN_SA
   1123 	if (__predict_false((l->l_savp)
   1124             && (l->l_savp->savp_pflags & SAVP_FLAG_DELIVERING)))
   1125 		l->l_savp->savp_pflags &= ~SAVP_FLAG_DELIVERING;
   1126 #endif
   1127 
   1128 	/*
   1129 	 * Restarting a system call is touchy on the HPPA, because syscall
   1130 	 * arguments are passed in registers and the program counter of the
   1131 	 * syscall "point" isn't easily divined.
   1132 	 *
   1133 	 * We handle the first problem by assuming that we will have to restart
   1134 	 * this system call, so we stuff the first four words of the original
   1135 	 * arguments back into the frame as arg0...arg3, which is where we
   1136 	 * found them in the first place.  Any further arguments are (still) on
   1137 	 * the user's stack and the  syscall code will fetch them from there
   1138 	 * (again).
   1139 	 *
   1140 	 * The program counter problem is addressed below.
   1141 	 */
   1142 	frame->tf_arg0 = args[0];
   1143 	frame->tf_arg1 = args[1];
   1144 	frame->tf_arg2 = args[2];
   1145 	frame->tf_arg3 = args[3];
   1146 
   1147 	/*
   1148 	 * Some special handling for the syscall(2) and
   1149 	 * __syscall(2) system calls.
   1150 	 */
   1151 	switch (code) {
   1152 	case SYS_syscall:
   1153 		code = *args;
   1154 		args += 1;
   1155 		break;
   1156 	case SYS___syscall:
   1157 		if (callp != sysent)
   1158 			break;
   1159 		/*
   1160 		 * NB: even though __syscall(2) takes a quad_t containing the
   1161 		 * system call number, because our argument copying word-swaps
   1162 		 * 64-bit arguments, the least significant word of that quad_t
   1163 		 * is the first word in the argument array.
   1164 		 */
   1165 		code = *args;
   1166 		args += 2;
   1167 	}
   1168 
   1169 	/*
   1170 	 * Stacks growing from lower addresses to higher addresses are not
   1171 	 * really such a good idea, because it makes it impossible to overlay a
   1172 	 * struct on top of C stack arguments (the arguments appear in
   1173 	 * reversed order).
   1174 	 *
   1175 	 * You can do the obvious thing (as locore.S does) and copy argument
   1176 	 * words one by one, laying them out in the "right" order in the dest-
   1177 	 * ination buffer, but this ends up word-swapping multi-word arguments
   1178 	 * (like off_t).
   1179 	 *
   1180 	 * FIXME - this works only on native binaries and
   1181 	 * will probably screw up any and all emulation.
   1182 	 *
   1183 	 */
   1184 
   1185 	if (code < 0 || code >= nsys)
   1186 		callp += p->p_emul->e_nosys;	/* bad syscall # */
   1187 	else
   1188 		callp += code;
   1189 
   1190 	nargs64 = SYCALL_NARGS64(callp);
   1191 	if (nargs64 != 0) {
   1192 		size_t nargs = callp->sy_narg;
   1193 
   1194 		for (size_t i = 0; i < nargs + nargs64;) {
   1195 			if (SYCALL_ARG_64_P(callp, i)) {
   1196 				tmp = args[i];
   1197 				args[i] = args[i + 1];
   1198 				args[i + 1] = tmp;
   1199 				i += 2;
   1200 			} else
   1201 				i++;
   1202 		}
   1203 	}
   1204 
   1205 #ifdef USERTRACE
   1206 	if (0) {
   1207 		user_backtrace(frame, l, -1);
   1208 		frame->tf_ipsw |= PSW_R;
   1209 		frame->tf_rctr = 0;
   1210 		printf("r %08x", frame->tf_iioq_head);
   1211 		rctr_next_iioq = frame->tf_iioq_head + 4;
   1212 	}
   1213 #endif
   1214 
   1215 	error = 0;
   1216 	if (__predict_false(p->p_trace_enabled)) {
   1217 		error = trace_enter(code, args, callp->sy_narg);
   1218 		if (error)
   1219 			goto out;
   1220 	}
   1221 
   1222 	rval[0] = 0;
   1223 	rval[1] = 0;
   1224 	error = sy_call(callp, l, args, rval);
   1225 out:
   1226 	switch (error) {
   1227 	case 0:
   1228 		l = curlwp;			/* changes on exec() */
   1229 		frame = l->l_md.md_regs;
   1230 		frame->tf_ret0 = rval[0];
   1231 		frame->tf_ret1 = rval[1];
   1232 		frame->tf_t1 = 0;
   1233 		break;
   1234 	case ERESTART:
   1235 		/*
   1236 		 * Now we have to wind back the instruction
   1237 		 * offset queue to the point where the system
   1238 		 * call will be made again.  This is inherently
   1239 		 * tied to the SYSCALL macro.
   1240 		 *
   1241 		 * Currently, the part of the SYSCALL macro
   1242 		 * that we want to rerun reads as:
   1243 		 *
   1244 		 *	ldil	L%SYSCALLGATE, r1
   1245 		 *	ble	4(sr7, r1)
   1246 		 *	ldi	__CONCAT(SYS_,x), t1
   1247 		 *	comb,<>	%r0, %t1, __cerror
   1248 		 *
   1249 		 * And our offset queue head points to the
   1250 		 * comb instruction.  So we need to
   1251 		 * subtract twelve to reach the ldil.
   1252 		 */
   1253 		frame->tf_iioq_head -= 12;
   1254 		frame->tf_iioq_tail = frame->tf_iioq_head + 4;
   1255 		break;
   1256 	case EJUSTRETURN:
   1257 		p = curproc;
   1258 		break;
   1259 	default:
   1260 		if (p->p_emul->e_errno)
   1261 			error = p->p_emul->e_errno[error];
   1262 		frame->tf_t1 = error;
   1263 		break;
   1264 	}
   1265 
   1266 	if (__predict_false(p->p_trace_enabled))
   1267 		trace_exit(code, rval, error);
   1268 
   1269 	userret(l, frame->tf_iioq_head, 0);
   1270 
   1271 #ifdef DIAGNOSTIC
   1272 	if (cpl != oldcpl) {
   1273 		printf("WARNING: SPL (0x%x) NOT LOWERED ON "
   1274 		    "syscall(0x%x, 0x%x, 0x%x, 0x%x...) EXIT, PID %d\n",
   1275 		    cpl, code, args[0], args[1], args[2], p->p_pid);
   1276 		cpl = oldcpl;
   1277 	}
   1278 #endif
   1279 
   1280 #ifdef DEBUG
   1281 	frame_sanity_check(0xdead05, 0, frame, l);
   1282 #endif /* DEBUG */
   1283 }
   1284 
   1285 /*
   1286  * Start a new LWP
   1287  */
   1288 void
   1289 startlwp(void *arg)
   1290 {
   1291 	int err;
   1292 	ucontext_t *uc = arg;
   1293 	struct lwp *l = curlwp;
   1294 
   1295 	err = cpu_setmcontext(l, &uc->uc_mcontext, uc->uc_flags);
   1296 #if DIAGNOSTIC
   1297 	if (err) {
   1298 		printf("Error %d from cpu_setmcontext.", err);
   1299 	}
   1300 #endif
   1301 	pool_put(&lwp_uc_pool, uc);
   1302 
   1303 	userret(l, l->l_md.md_regs->tf_iioq_head, 0);
   1304 }
   1305 
   1306 /*
   1307  * XXX This is a terrible name.
   1308  */
   1309 void
   1310 upcallret(struct lwp *l)
   1311 {
   1312 	userret(l, l->l_md.md_regs->tf_iioq_head, 0);
   1313 }
   1314