Home | History | Annotate | Line # | Download | only in hppa
      1 /*	$NetBSD: trap.c,v 1.128 2025/09/19 18:06:37 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 1.128 2025/09/19 18:06:37 skrll Exp $");
     62 
     63 /* #define INTRDEBUG */
     64 /* #define TRAPDEBUG */
     65 /* #define USERTRACE */
     66 
     67 #include "opt_kgdb.h"
     68 #include "opt_ptrace.h"
     69 
     70 #include <sys/param.h>
     71 #include <sys/systm.h>
     72 #include <sys/kernel.h>
     73 #include <sys/syscall.h>
     74 #include <sys/syscallvar.h>
     75 #include <sys/mutex.h>
     76 #include <sys/ktrace.h>
     77 #include <sys/proc.h>
     78 #include <sys/signalvar.h>
     79 #include <sys/acct.h>
     80 #include <sys/signal.h>
     81 #include <sys/device.h>
     82 #include <sys/kauth.h>
     83 #include <sys/kmem.h>
     84 #include <sys/userret.h>
     85 
     86 #ifdef KGDB
     87 #include <sys/kgdb.h>
     88 #endif
     89 
     90 #include <uvm/uvm.h>
     91 
     92 #include <machine/iomod.h>
     93 #include <machine/cpufunc.h>
     94 #include <machine/reg.h>
     95 #include <machine/autoconf.h>
     96 
     97 #include <machine/db_machdep.h>
     98 
     99 #include <hppa/hppa/machdep.h>
    100 
    101 #include <ddb/db_output.h>
    102 #include <ddb/db_interface.h>
    103 
    104 #ifdef PTRACE
    105 void ss_clear_breakpoints(struct lwp *l);
    106 int ss_put_value(struct lwp *, vaddr_t, u_int);
    107 int ss_get_value(struct lwp *, vaddr_t, u_int *);
    108 
    109 /* single-step breakpoint */
    110 #define SSBREAKPOINT   (HPPA_BREAK_KERNEL | (HPPA_BREAK_SS << 13))
    111 
    112 #endif
    113 
    114 #if defined(DEBUG) || defined(DIAGNOSTIC)
    115 /*
    116  * 0x6fc1000 is a stwm r1, d(sr0, sp), which is the last
    117  * instruction in the function prologue that gcc -O0 uses.
    118  * When we have this instruction we know the relationship
    119  * between the stack pointer and the gcc -O0 frame pointer
    120  * (in r3, loaded with the initial sp) for the body of a
    121  * function.
    122  *
    123  * If the given instruction is a stwm r1, d(sr0, sp) where
    124  * d > 0, we evaluate to d, else we evaluate to zero.
    125  */
    126 #define STWM_R1_D_SR0_SP(inst) \
    127 	(((inst) & 0xffffc001) == 0x6fc10000 ? (((inst) & 0x00003ff) >> 1) : 0)
    128 #endif /* DEBUG || DIAGNOSTIC */
    129 
    130 const char *trap_type[] = {
    131 	"invalid",
    132 	"HPMC",
    133 	"power failure",
    134 	"recovery counter",
    135 	"external interrupt",
    136 	"LPMC",
    137 	"ITLB miss fault",
    138 	"instruction protection",
    139 	"Illegal instruction",
    140 	"break instruction",
    141 	"privileged operation",
    142 	"privileged register",
    143 	"overflow",
    144 	"conditional",
    145 	"assist exception",
    146 	"DTLB miss",
    147 	"ITLB non-access miss",
    148 	"DTLB non-access miss",
    149 	"data protection/rights/alignment",
    150 	"data break",
    151 	"TLB dirty",
    152 	"page reference",
    153 	"assist emulation",
    154 	"higher-priv transfer",
    155 	"lower-priv transfer",
    156 	"taken branch",
    157 	"data access rights",
    158 	"data protection",
    159 	"unaligned data ref",
    160 };
    161 int trap_types = __arraycount(trap_type);
    162 
    163 uint8_t fpopmap[] = {
    164 	0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x00,
    165 	0x00, 0x0c, 0x00, 0x0e, 0x00, 0x00, 0x00, 0x00,
    166 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    167 	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    168 	0x00, 0x00, 0x00, 0x26, 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, 0x00, 0x00, 0x00, 0x00, 0x00,
    172 };
    173 
    174 void pmap_hptdump(void);
    175 void syscall(struct trapframe *, int *);
    176 
    177 #if defined(DEBUG)
    178 struct trapframe *sanity_frame;
    179 struct lwp *sanity_lwp;
    180 const char *sanity_string;
    181 void frame_sanity_check(const char *, int, int, struct trapframe *,
    182     struct lwp *);
    183 #endif
    184 
    185 
    186 #ifdef USERTRACE
    187 /*
    188  * USERTRACE is a crude facility that traces the PC of a single user process.
    189  * This tracing is normally activated by the dispatching of a certain syscall
    190  * with certain arguments - see the activation code in syscall().
    191  */
    192 static void user_backtrace(struct trapframe *, struct lwp *, int);
    193 static void user_backtrace_raw(u_int, u_int);
    194 
    195 u_int rctr_next_iioq;
    196 #endif
    197 
    198 static inline void
    199 userret(struct lwp *l, struct trapframe *tf)
    200 {
    201 	struct proc *p = l->l_proc;
    202 	int oticks = 0; /* XXX why zero? */
    203 
    204 	do {
    205 		l->l_md.md_astpending = 0;
    206 		//curcpu()->ci_data.cpu_nast++;
    207 		mi_userret(l);
    208 	} while (l->l_md.md_astpending);
    209 
    210 	/*
    211 	 * If profiling, charge recent system time to the trapped pc.
    212 	 */
    213 	if (p->p_stflag & PST_PROFIL) {
    214 		extern int psratio;
    215 
    216 		addupc_task(l, tf->tf_iioq_head,
    217 		    (int)(p->p_sticks - oticks) * psratio);
    218 	}
    219 }
    220 
    221 /*
    222  * This handles some messy kernel debugger details.
    223  * It dispatches into either kgdb or DDB, and knows
    224  * about some special things to do, like skipping over
    225  * break instructions and how to really set up for
    226  * a single-step.
    227  */
    228 #if defined(KGDB) || defined(DDB)
    229 static int
    230 trap_kdebug(int type, int code, struct trapframe *frame)
    231 {
    232 	int handled;
    233 	u_int tf_iioq_head_old;
    234 	u_int tf_iioq_tail_old;
    235 
    236 	for (;;) {
    237 
    238 		/* This trap has not been handled. */
    239 		handled = 0;
    240 
    241 		/* Remember the instruction offset queue. */
    242 		tf_iioq_head_old = frame->tf_iioq_head;
    243 		tf_iioq_tail_old = frame->tf_iioq_tail;
    244 
    245 #ifdef	KGDB
    246 		/* Let KGDB handle it (if connected) */
    247 		if (!handled)
    248 			handled = kgdb_trap(type, frame);
    249 #endif
    250 #ifdef	DDB
    251 		/* Let DDB handle it. */
    252 		if (!handled)
    253 			handled = kdb_trap(type, code, frame);
    254 #endif
    255 
    256 		/* If this trap wasn't handled, return now. */
    257 		if (!handled)
    258 			return(0);
    259 
    260 		/*
    261 		 * If the instruction offset queue head changed, but the offset
    262 		 * queue tail didn't, assume that the user wants to jump to the
    263 		 * head offset, and adjust the tail accordingly.  This should
    264 		 * fix the kgdb `jump' command, and can help DDB users who `set'
    265 		 * the offset head but forget the tail.
    266 		 */
    267 		if (frame->tf_iioq_head != tf_iioq_head_old &&
    268 		    frame->tf_iioq_tail == tf_iioq_tail_old)
    269 			frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    270 
    271 		/*
    272 		 * This is some single-stepping support.  If we're trying to
    273 		 * step through a nullified instruction, just advance by hand
    274 		 * and trap again.  Otherwise, load the recovery counter with
    275 		 * zero.
    276 		 */
    277 		if (frame->tf_ipsw & PSW_R) {
    278 #ifdef TRAPDEBUG
    279 			printf("(single stepping at head 0x%x tail 0x%x)\n",
    280 			    frame->tf_iioq_head, frame->tf_iioq_tail);
    281 #endif
    282 			if (frame->tf_ipsw & PSW_N) {
    283 #ifdef TRAPDEBUG
    284 				printf("(single stepping past nullified)\n");
    285 #endif
    286 
    287 				/* Advance the program counter. */
    288 				frame->tf_iioq_head = frame->tf_iioq_tail;
    289 				frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    290 
    291 				/* Clear flags. */
    292 				frame->tf_ipsw &= ~(PSW_N|PSW_X|PSW_Y|PSW_Z|PSW_B|PSW_T|PSW_H|PSW_L);
    293 
    294 				/* Simulate another trap. */
    295 				type = T_RECOVERY;
    296 				continue;
    297 			}
    298 			frame->tf_rctr = 0;
    299 		}
    300 
    301 		/* We handled this trap. */
    302 		return (1);
    303 	}
    304 	/* NOTREACHED */
    305 }
    306 #else	/* !KGDB && !DDB */
    307 #define trap_kdebug(t, c, f)	(0)
    308 #endif	/* !KGDB && !DDB */
    309 
    310 #if defined(DEBUG) || defined(USERTRACE)
    311 /*
    312  * These functions give a crude usermode backtrace.  They really only work when
    313  * code has been compiled without optimization, as they assume a certain func-
    314  * tion prologue sets up a frame pointer and stores the return pointer and arg-
    315  * uments in it.
    316  */
    317 static void
    318 user_backtrace_raw(u_int pc, u_int fp)
    319 {
    320 	int frame_number;
    321 	int arg_number;
    322 	uint32_t val;
    323 
    324 	for (frame_number = 0;
    325 	     frame_number < 100 && pc > HPPA_PC_PRIV_MASK && fp;
    326 	     frame_number++) {
    327 
    328 		printf("%3d: pc=%08x%s fp=0x%08x", frame_number,
    329 		    pc & ~HPPA_PC_PRIV_MASK, USERMODE(pc) ? "  " : "**", fp);
    330 		for (arg_number = 0; arg_number < 4; arg_number++) {
    331 			if (ufetch_32(HPPA_FRAME_CARG(arg_number, fp),
    332 				      &val) == 0) {
    333 				printf(" arg%d=0x%08x", arg_number, val);
    334 			} else {
    335 				printf(" arg%d=<bad address>", arg_number);
    336 			}
    337 		}
    338 		printf("\n");
    339 		if (ufetch_int((((uint32_t *) fp) - 5), &pc) != 0) {
    340 			printf("  ufetch for pc failed\n");
    341 			break;
    342 		}
    343 		if (ufetch_int((((uint32_t *) fp) + 0), &fp) != 0) {
    344 			printf("  ufetch for fp failed\n");
    345 			break;
    346 		}
    347 	}
    348 	printf("  backtrace stopped with pc %08x fp 0x%08x\n", pc, fp);
    349 }
    350 
    351 static void
    352 user_backtrace(struct trapframe *tf, struct lwp *l, int type)
    353 {
    354 	struct proc *p = l->l_proc;
    355 	u_int pc, fp, inst;
    356 
    357 	/*
    358 	 * Display any trap type that we have.
    359 	 */
    360 	if (type >= 0)
    361 		printf("pid %d (%s) trap #%d\n",
    362 		    p->p_pid, p->p_comm, type & ~T_USER);
    363 
    364 	/*
    365 	 * Assuming that the frame pointer in r3 is valid,
    366 	 * dump out a stack trace.
    367 	 */
    368 	fp = tf->tf_r3;
    369 	printf("pid %d (%s) backtrace, starting with fp 0x%08x\n",
    370 		p->p_pid, p->p_comm, fp);
    371 	user_backtrace_raw(tf->tf_iioq_head, fp);
    372 
    373 	/*
    374 	 * In case the frame pointer in r3 is not valid, assuming the stack
    375 	 * pointer is valid and the faulting function is a non-leaf, if we can
    376 	 * find its prologue we can recover its frame pointer.
    377 	 */
    378 	pc = tf->tf_iioq_head;
    379 	fp = tf->tf_sp - HPPA_FRAME_SIZE;
    380 	printf("pid %d (%s) backtrace, starting with sp 0x%08x pc 0x%08x\n",
    381 	    p->p_pid, p->p_comm, tf->tf_sp, pc);
    382 	for (pc &= ~HPPA_PC_PRIV_MASK; pc > 0; pc -= sizeof(inst)) {
    383 		if (ufetch_int((u_int *) pc, &inst) != 0) {
    384 			printf("  ufetch for inst at pc %08x failed\n", pc);
    385 			break;
    386 		}
    387 		/* Check for the prologue instruction that sets sp. */
    388 		if (STWM_R1_D_SR0_SP(inst)) {
    389 			fp = tf->tf_sp - STWM_R1_D_SR0_SP(inst);
    390 			printf("  sp from fp at pc %08x: %08x\n", pc, inst);
    391 			break;
    392 		}
    393 	}
    394 	user_backtrace_raw(tf->tf_iioq_head, fp);
    395 }
    396 #endif /* DEBUG || USERTRACE */
    397 
    398 #ifdef DEBUG
    399 /*
    400  * This sanity-checks a trapframe.  It is full of various assumptions about
    401  * what a healthy CPU state should be, with some documented elsewhere, some not.
    402  */
    403 void
    404 frame_sanity_check(const char *func, int line, int type, struct trapframe *tf,
    405     struct lwp *l)
    406 {
    407 #if 0
    408 	extern int kernel_text;
    409 	extern int etext;
    410 #endif
    411 	struct cpu_info *ci = curcpu();
    412 
    413 #define SANITY(e)					\
    414 do {							\
    415 	if (sanity_frame == NULL && !(e)) {		\
    416 		sanity_frame = tf;			\
    417 		sanity_lwp = l;				\
    418 		sanity_string = #e;			\
    419 	}						\
    420 } while (/* CONSTCOND */ 0)
    421 
    422 	KASSERT(l != NULL);
    423 	SANITY((tf->tf_ipsw & ci->ci_psw) == ci->ci_psw);
    424 	SANITY((ci->ci_psw & PSW_I) == 0 || tf->tf_eiem != 0);
    425 	if (tf->tf_iisq_head == HPPA_SID_KERNEL) {
    426 		vaddr_t minsp, maxsp, uv;
    427 
    428 		uv = uvm_lwp_getuarea(l);
    429 
    430 		/*
    431 		 * If the trap happened in the gateway page, we take the easy
    432 		 * way out and assume that the trapframe is okay.
    433 		 */
    434 		if ((tf->tf_iioq_head & ~PAGE_MASK) == SYSCALLGATE)
    435 			goto out;
    436 
    437 		SANITY(!USERMODE(tf->tf_iioq_head));
    438 		SANITY(!USERMODE(tf->tf_iioq_tail));
    439 
    440 		/*
    441 		 * Don't check the instruction queues or stack on interrupts
    442 		 * as we could be in the sti code (outside normal kernel
    443 		 * text) or switching LWPs (curlwp and sp are not in sync)
    444 		 */
    445 		if ((type & ~T_USER) == T_INTERRUPT)
    446 			goto out;
    447 #if 0
    448 		SANITY(tf->tf_iioq_head >= (u_int) &kernel_text);
    449 		SANITY(tf->tf_iioq_head < (u_int) &etext);
    450 		SANITY(tf->tf_iioq_tail >= (u_int) &kernel_text);
    451 		SANITY(tf->tf_iioq_tail < (u_int) &etext);
    452 #endif
    453 
    454 		maxsp = uv + USPACE + PAGE_SIZE;
    455 		minsp = uv + PAGE_SIZE;
    456 
    457 		SANITY(tf->tf_sp >= minsp && tf->tf_sp < maxsp);
    458 	} else {
    459 		struct pcb *pcb = lwp_getpcb(l);
    460 
    461 		SANITY(USERMODE(tf->tf_iioq_head));
    462 		SANITY(USERMODE(tf->tf_iioq_tail));
    463 		SANITY(tf->tf_cr30 == (u_int)pcb->pcb_fpregs);
    464 	}
    465 #undef SANITY
    466 out:
    467 	if (sanity_frame == tf) {
    468 		printf("insanity: '%s' at %s:%d type 0x%x tf %p lwp %p "
    469 		    "sp 0x%x pc 0x%x\n",
    470 		    sanity_string, func, line, type, sanity_frame, sanity_lwp,
    471 		    tf->tf_sp, tf->tf_iioq_head);
    472 		(void) trap_kdebug(T_IBREAK, 0, tf);
    473 		sanity_frame = NULL;
    474 		sanity_lwp = NULL;
    475 	}
    476 }
    477 #endif /* DEBUG */
    478 
    479 
    480 #define __PABITS(x, y)		__BITS(31 - (x), 31 - (y))
    481 #define __PABIT(x)		__BIT(31 - (x))
    482 
    483 #define LPA_MASK					  \
    484      (				__PABITS(0, 5)		| \
    485 				__PABITS(18, 25))
    486 #define LPA						  \
    487      (__SHIFTIN(1,		__PABITS(0, 5))		| \
    488       __SHIFTIN(0x4d,		__PABITS(18, 25)))
    489 
    490 
    491 #define PROBE_ENCS		(0x46 | 0xc6 | 0x47 | 0xc7)
    492 #define PROBE_PL		__PABITS(11, 15)
    493 #define PROBE_IMMED		__PABIT(18)
    494 #define PROBE_RW		__PABIT(25)
    495 
    496 #define PROBE_MASK					  \
    497     ((				__PABITS(0, 5)		| \
    498 				__PABITS(18, 25)	| \
    499 				__PABIT(26))		^ \
    500      (PROBE_IMMED | PROBE_RW))
    501 
    502 #define PROBE						  \
    503     ((__SHIFTIN(1,		__PABITS(0, 5))		| \
    504       __SHIFTIN(PROBE_ENCS,	__PABITS(18, 25))	| \
    505       __SHIFTIN(0,		__PABIT(26)))		^ \
    506      (PROBE_IMMED | PROBE_RW))
    507 
    508 #define PLMASK			__BITS(1, 0)
    509 
    510 
    511 /* for hppa64 */
    512 CTASSERT(sizeof(register_t) == sizeof(u_int));
    513 size_t hppa_regmap[] = {
    514 	0,	/* r0 is special case */
    515 	offsetof(struct trapframe, tf_r1  ) / sizeof(register_t),
    516 	offsetof(struct trapframe, tf_rp  ) / sizeof(register_t),
    517 	offsetof(struct trapframe, tf_r3  ) / sizeof(register_t),
    518 	offsetof(struct trapframe, tf_r4  ) / sizeof(register_t),
    519 	offsetof(struct trapframe, tf_r5  ) / sizeof(register_t),
    520 	offsetof(struct trapframe, tf_r6  ) / sizeof(register_t),
    521 	offsetof(struct trapframe, tf_r7  ) / sizeof(register_t),
    522 	offsetof(struct trapframe, tf_r8  ) / sizeof(register_t),
    523 	offsetof(struct trapframe, tf_r9  ) / sizeof(register_t),
    524 	offsetof(struct trapframe, tf_r10 ) / sizeof(register_t),
    525 	offsetof(struct trapframe, tf_r11 ) / sizeof(register_t),
    526 	offsetof(struct trapframe, tf_r12 ) / sizeof(register_t),
    527 	offsetof(struct trapframe, tf_r13 ) / sizeof(register_t),
    528 	offsetof(struct trapframe, tf_r14 ) / sizeof(register_t),
    529 	offsetof(struct trapframe, tf_r15 ) / sizeof(register_t),
    530 	offsetof(struct trapframe, tf_r16 ) / sizeof(register_t),
    531 	offsetof(struct trapframe, tf_r17 ) / sizeof(register_t),
    532 	offsetof(struct trapframe, tf_r18 ) / sizeof(register_t),
    533 	offsetof(struct trapframe, tf_t4  ) / sizeof(register_t),
    534 	offsetof(struct trapframe, tf_t3  ) / sizeof(register_t),
    535 	offsetof(struct trapframe, tf_t2  ) / sizeof(register_t),
    536 	offsetof(struct trapframe, tf_t1  ) / sizeof(register_t),
    537 	offsetof(struct trapframe, tf_arg3) / sizeof(register_t),
    538 	offsetof(struct trapframe, tf_arg2) / sizeof(register_t),
    539 	offsetof(struct trapframe, tf_arg1) / sizeof(register_t),
    540 	offsetof(struct trapframe, tf_arg0) / sizeof(register_t),
    541 	offsetof(struct trapframe, tf_dp  ) / sizeof(register_t),
    542 	offsetof(struct trapframe, tf_ret0) / sizeof(register_t),
    543 	offsetof(struct trapframe, tf_ret1) / sizeof(register_t),
    544 	offsetof(struct trapframe, tf_sp  ) / sizeof(register_t),
    545 	offsetof(struct trapframe, tf_r31 ) / sizeof(register_t),
    546 };
    547 
    548 
    549 static inline register_t
    550 tf_getregno(struct trapframe *tf, u_int regno)
    551 {
    552 	register_t *tf_reg = (register_t *)tf;
    553 	if (regno == 0)
    554 		return 0;
    555 	else
    556 		return tf_reg[hppa_regmap[regno]];
    557 }
    558 
    559 static inline void
    560 tf_setregno(struct trapframe *tf, u_int regno, register_t val)
    561 {
    562 	register_t *tf_reg = (register_t *)tf;
    563 	if (regno == 0)
    564 		return;
    565 	else
    566 		tf_reg[hppa_regmap[regno]] = val;
    567 }
    568 
    569 void
    570 trap(int type, struct trapframe *frame)
    571 {
    572 	struct lwp *l;
    573 	struct proc *p;
    574 	struct pcb *pcb;
    575 	vaddr_t va;
    576 	struct vm_map *map;
    577 	struct vmspace *vm;
    578 	vm_prot_t vftype;
    579 	pa_space_t space;
    580 	ksiginfo_t ksi;
    581 	u_int opcode, onfault;
    582 	int ret;
    583 	const char *tts = "reserved";
    584 	int trapnum;
    585 #ifdef DIAGNOSTIC
    586 	extern int emergency_stack_start, emergency_stack_end;
    587 	struct cpu_info *ci = curcpu();
    588 	int oldcpl = ci->ci_cpl;
    589 #endif
    590 
    591 	trapnum = type & ~T_USER;
    592 	opcode = frame->tf_iir;
    593 
    594 	if (trapnum <= T_EXCEPTION || trapnum == T_HIGHERPL ||
    595 	    trapnum == T_LOWERPL || trapnum == T_TAKENBR ||
    596 	    trapnum == T_IDEBUG || trapnum == T_PERFMON) {
    597 		va = frame->tf_iioq_head;
    598 		space = frame->tf_iisq_head;
    599 		vftype = VM_PROT_EXECUTE;
    600 	} else {
    601 		va = frame->tf_ior;
    602 		space = frame->tf_isr;
    603 		vftype = inst_store(opcode) ? VM_PROT_WRITE : VM_PROT_READ;
    604 	}
    605 
    606 	KASSERT(curlwp != NULL);
    607 	l = curlwp;
    608 	p = l->l_proc;
    609 
    610 #ifdef DIAGNOSTIC
    611 	/*
    612 	 * If we are on the emergency stack, then we either got
    613 	 * a fault on the kernel stack, or we're just handling
    614 	 * a trap for the machine check handler (which also
    615 	 * runs on the emergency stack).
    616 	 *
    617 	 * We *very crudely* differentiate between the two cases
    618 	 * by checking the faulting instruction: if it is the
    619 	 * function prologue instruction that stores the old
    620 	 * frame pointer and updates the stack pointer, we assume
    621 	 * that we faulted on the kernel stack.
    622 	 *
    623 	 * In this case, not completing that instruction will
    624 	 * probably confuse backtraces in kgdb/ddb.  Completing
    625 	 * it would be difficult, because we already faulted on
    626 	 * that part of the stack, so instead we fix up the
    627 	 * frame as if the function called has just returned.
    628 	 * This has peculiar knowledge about what values are in
    629 	 * what registers during the "normal gcc -g" prologue.
    630 	 */
    631 	if (&type >= &emergency_stack_start &&
    632 	    &type < &emergency_stack_end &&
    633 	    type != T_IBREAK && STWM_R1_D_SR0_SP(opcode)) {
    634 		/* Restore the caller's frame pointer. */
    635 		frame->tf_r3 = frame->tf_r1;
    636 		/* Restore the caller's instruction offsets. */
    637 		frame->tf_iioq_head = frame->tf_rp;
    638 		frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    639 		goto dead_end;
    640 	}
    641 #endif /* DIAGNOSTIC */
    642 
    643 #ifdef DEBUG
    644 	frame_sanity_check(__func__, __LINE__, type, frame, l);
    645 #endif /* DEBUG */
    646 
    647 	if (frame->tf_flags & TFF_LAST)
    648 		l->l_md.md_regs = frame;
    649 
    650 	if (trapnum <= trap_types)
    651 		tts = trap_type[trapnum];
    652 
    653 #ifdef TRAPDEBUG
    654 	if (trapnum != T_INTERRUPT && trapnum != T_IBREAK)
    655 		printf("trap: %d, %s for %x:%lx at %x:%x, fp=%p, rp=%x\n",
    656 		    type, tts, space, va, frame->tf_iisq_head,
    657 		    frame->tf_iioq_head, frame, frame->tf_rp);
    658 	else if (trapnum == T_IBREAK)
    659 		printf("trap: break instruction %x:%x at %x:%x, fp=%p\n",
    660 		    break5(opcode), break13(opcode),
    661 		    frame->tf_iisq_head, frame->tf_iioq_head, frame);
    662 
    663 	{
    664 		extern int etext;
    665 		if (frame < (struct trapframe *)&etext) {
    666 			printf("trap: bogus frame ptr %p\n", frame);
    667 			goto dead_end;
    668 		}
    669 	}
    670 #endif
    671 
    672 	pcb = lwp_getpcb(l);
    673 
    674 	/* If this is a trap, not an interrupt, reenable interrupts. */
    675 	if (trapnum != T_INTERRUPT) {
    676 		curcpu()->ci_data.cpu_ntrap++;
    677 		mtctl(frame->tf_eiem, CR_EIEM);
    678 	}
    679 
    680 	const bool user = (type & T_USER) != 0;
    681 	switch (type) {
    682 	case T_NONEXIST:
    683 	case T_NONEXIST | T_USER:
    684 #if !defined(DDB) && !defined(KGDB)
    685 		/* we've got screwed up by the central scrutinizer */
    686 		panic ("trap: elvis has just left the building!");
    687 		break;
    688 #else
    689 		goto dead_end;
    690 #endif
    691 	case T_RECOVERY | T_USER:
    692 #ifdef USERTRACE
    693 		for (;;) {
    694 			if (frame->tf_iioq_head != rctr_next_iioq)
    695 				printf("-%08x\nr %08x",
    696 					rctr_next_iioq - 4,
    697 					frame->tf_iioq_head);
    698 			rctr_next_iioq = frame->tf_iioq_head + 4;
    699 			if (frame->tf_ipsw & PSW_N) {
    700 				/* Advance the program counter. */
    701 				frame->tf_iioq_head = frame->tf_iioq_tail;
    702 				frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    703 				/* Clear flags. */
    704 				frame->tf_ipsw &= ~(PSW_N|PSW_X|PSW_Y|PSW_Z|PSW_B|PSW_T|PSW_H|PSW_L);
    705 				/* Simulate another trap. */
    706 				continue;
    707 			}
    708 			break;
    709 		}
    710 		frame->tf_rctr = 0;
    711 		break;
    712 #endif /* USERTRACE */
    713 	case T_RECOVERY:
    714 #if !defined(DDB) && !defined(KGDB)
    715 		/* XXX will implement later */
    716 		printf ("trap: handicapped");
    717 		break;
    718 #else
    719 		goto dead_end;
    720 #endif
    721 
    722 	case T_EMULATION | T_USER:
    723 		hppa_fpu_emulate(frame, l, opcode);
    724 		break;
    725 
    726 	case T_DATALIGN:
    727 		onfault = pcb->pcb_onfault;
    728 		if (onfault) {
    729 			ret = EFAULT;
    730 do_onfault:
    731 			frame->tf_iioq_head = onfault;
    732 			frame->tf_iioq_tail = frame->tf_iioq_head + 4;
    733 			frame->tf_ret0 = ret;
    734 			break;
    735 		}
    736 		/*FALLTHROUGH*/
    737 
    738 #ifdef DIAGNOSTIC
    739 		/* these just can't happen ever */
    740 	case T_PRIV_OP:
    741 	case T_PRIV_REG:
    742 		/* these just can't make it to the trap() ever */
    743 	case T_HPMC:
    744 	case T_HPMC | T_USER:
    745 	case T_EMULATION:
    746 	case T_EXCEPTION:
    747 #endif
    748 	case T_IBREAK:
    749 	case T_DBREAK:
    750 	dead_end:
    751 		if (type & T_USER) {
    752 #ifdef DEBUG
    753 			user_backtrace(frame, l, type);
    754 #endif
    755 			KSI_INIT_TRAP(&ksi);
    756 			ksi.ksi_signo = SIGILL;
    757 			ksi.ksi_code = ILL_ILLTRP;
    758 			ksi.ksi_trap = type;
    759 			ksi.ksi_addr = (void *)frame->tf_iioq_head;
    760 			trapsignal(l, &ksi);
    761 			break;
    762 		}
    763 		if (trap_kdebug(type, va, frame))
    764 			return;
    765 		else if (type == T_DATALIGN)
    766 			panic ("trap: %s at 0x%x", tts, (u_int) va);
    767 		else
    768 			panic ("trap: no debugger for \"%s\" (%d)", tts, type);
    769 		break;
    770 
    771 	case T_IBREAK | T_USER:
    772 	case T_DBREAK | T_USER:
    773 		KSI_INIT_TRAP(&ksi);
    774 		ksi.ksi_signo = SIGTRAP;
    775 		ksi.ksi_code = TRAP_BRKPT;
    776 		ksi.ksi_trap = trapnum;
    777 		ksi.ksi_addr = (void *)(frame->tf_iioq_head & ~HPPA_PC_PRIV_MASK);
    778 #ifdef PTRACE
    779 		ss_clear_breakpoints(l);
    780 		if (opcode == SSBREAKPOINT)
    781 			ksi.ksi_code = TRAP_TRACE;
    782 #endif
    783 		/* pass to user debugger */
    784 		trapsignal(l, &ksi);
    785 		break;
    786 
    787 #ifdef PTRACE
    788 	case T_TAKENBR | T_USER:
    789 		ss_clear_breakpoints(l);
    790 
    791 		KSI_INIT_TRAP(&ksi);
    792 		ksi.ksi_signo = SIGTRAP;
    793 		ksi.ksi_code = TRAP_TRACE;
    794 		ksi.ksi_trap = trapnum;
    795 		ksi.ksi_addr = (void *)(frame->tf_iioq_head & ~HPPA_PC_PRIV_MASK);
    796 
    797                 /* pass to user debugger */
    798 		trapsignal(l, &ksi);
    799 		break;
    800 #endif
    801 
    802 	case T_EXCEPTION | T_USER: {	/* co-proc assist trap */
    803 		uint64_t *fpp;
    804 		uint32_t *pex, ex, inst;
    805 		int i;
    806 
    807 		hppa_fpu_flush(l);
    808 		fpp = (uint64_t *)pcb->pcb_fpregs;
    809 
    810 		/* skip the status register */
    811 		pex = (uint32_t *)&fpp[0];
    812 		pex++;
    813 
    814 		/* loop through the exception registers */
    815 		for (i = 1; i < 8 && !*pex; i++, pex++)
    816 			;
    817 		KASSERT(i < 8);
    818 		ex = *pex;
    819 		*pex = 0;
    820 
    821 		/* reset the trap flag, as if there was none */
    822 		fpp[0] &= ~(((uint64_t)HPPA_FPU_T) << 32);
    823 
    824 		/* emulate the instruction */
    825 		inst = ((uint32_t)fpopmap[ex >> 26] << 26) | (ex & 0x03ffffff);
    826 		hppa_fpu_emulate(frame, l, inst);
    827 		}
    828 		break;
    829 
    830 	case T_OVERFLOW | T_USER:
    831 		KSI_INIT_TRAP(&ksi);
    832 		ksi.ksi_signo = SIGFPE;
    833 		ksi.ksi_code = SI_NOINFO;
    834 		ksi.ksi_trap = type;
    835 		ksi.ksi_addr = (void *)va;
    836 		trapsignal(l, &ksi);
    837 		break;
    838 
    839 	case T_CONDITION | T_USER:
    840 		KSI_INIT_TRAP(&ksi);
    841 		ksi.ksi_signo = SIGFPE;
    842 		ksi.ksi_code = FPE_INTDIV;
    843 		ksi.ksi_trap = type;
    844 		ksi.ksi_addr = (void *)va;
    845 		trapsignal(l, &ksi);
    846 		break;
    847 
    848 	case T_ILLEGAL | T_USER:
    849 #ifdef DEBUG
    850 		user_backtrace(frame, l, type);
    851 #endif
    852 		KSI_INIT_TRAP(&ksi);
    853 		ksi.ksi_signo = SIGILL;
    854 		ksi.ksi_code = ILL_ILLOPC;
    855 		ksi.ksi_trap = type;
    856 		ksi.ksi_addr = (void *)va;
    857 		trapsignal(l, &ksi);
    858 		break;
    859 
    860 	case T_PRIV_OP | T_USER:
    861 #ifdef DEBUG
    862 		user_backtrace(frame, l, type);
    863 #endif
    864 		KSI_INIT_TRAP(&ksi);
    865 		ksi.ksi_signo = SIGILL;
    866 		ksi.ksi_code = ILL_PRVOPC;
    867 		ksi.ksi_trap = type;
    868 		ksi.ksi_addr = (void *)va;
    869 		trapsignal(l, &ksi);
    870 		break;
    871 
    872 	case T_PRIV_REG | T_USER:
    873 #ifdef DEBUG
    874 		user_backtrace(frame, l, type);
    875 #endif
    876 		KSI_INIT_TRAP(&ksi);
    877 		ksi.ksi_signo = SIGILL;
    878 		ksi.ksi_code = ILL_PRVREG;
    879 		ksi.ksi_trap = type;
    880 		ksi.ksi_addr = (void *)va;
    881 		trapsignal(l, &ksi);
    882 		break;
    883 
    884 		/* these should never got here */
    885 	case T_HIGHERPL | T_USER:
    886 	case T_LOWERPL | T_USER:
    887 		KSI_INIT_TRAP(&ksi);
    888 		ksi.ksi_signo = SIGSEGV;
    889 		ksi.ksi_code = SEGV_ACCERR;
    890 		ksi.ksi_trap = type;
    891 		ksi.ksi_addr = (void *)va;
    892 		trapsignal(l, &ksi);
    893 		break;
    894 
    895 	case T_IPROT | T_USER:
    896 	case T_DPROT | T_USER:
    897 		KSI_INIT_TRAP(&ksi);
    898 		ksi.ksi_signo = SIGSEGV;
    899 		ksi.ksi_code = SEGV_ACCERR;
    900 		ksi.ksi_trap = type;
    901 		ksi.ksi_addr = (void *)va;
    902 		trapsignal(l, &ksi);
    903 		break;
    904 
    905 	case T_ITLBMISSNA:	case T_USER | T_ITLBMISSNA:
    906 	case T_DTLBMISSNA:	case T_USER | T_DTLBMISSNA:
    907 		vm = p->p_vmspace;
    908 
    909 		if (!vm) {
    910 #ifdef TRAPDEBUG
    911 			printf("trap: no vm, p=%p\n", p);
    912 #endif
    913 			goto dead_end;
    914 		}
    915 
    916 		/*
    917 		 * it could be a kernel map for exec_map faults
    918 		 */
    919 		if (!user && space == HPPA_SID_KERNEL)
    920 			map = kernel_map;
    921 		else {
    922 			map = &vm->vm_map;
    923 		}
    924 
    925 		va = trunc_page(va);
    926 
    927 		if ((opcode & LPA_MASK) == LPA) {
    928 			/* lpa failure case */
    929 			const u_int regno =
    930 			    __SHIFTOUT(opcode, __PABITS(27, 31));
    931 			tf_setregno(frame, regno, 0);
    932 			frame->tf_ipsw |= PSW_N;
    933 		} else if ((opcode & PROBE_MASK) == PROBE) {
    934 			u_int pl;
    935 			if ((opcode & PROBE_IMMED) != 0) {
    936 				pl = __SHIFTOUT(opcode, PROBE_PL) & PLMASK;
    937 			} else {
    938 				const u_int plreg =
    939 				    __SHIFTOUT(opcode, PROBE_PL);
    940 				pl = tf_getregno(frame, plreg) & PLMASK;
    941 			}
    942 
    943 			bool ok = true;
    944 			if ((user && space == HPPA_SID_KERNEL) ||
    945 			    (frame->tf_iioq_head & HPPA_PC_PRIV_MASK) != pl ||
    946 			    (user && va >= VM_MAXUSER_ADDRESS)) {
    947 				ok = false;
    948 			} else {
    949 				/* Never call uvm_fault in interrupt context. */
    950 				KASSERT(curcpu()->ci_intr_depth == 0);
    951 
    952 				const bool read =
    953 				    __SHIFTOUT(opcode, PROBE_RW) == 0;
    954 				onfault = pcb->pcb_onfault;
    955 				pcb->pcb_onfault = 0;
    956 				ret = uvm_fault(map, va, read ?
    957 				    VM_PROT_READ : VM_PROT_WRITE);
    958 				pcb->pcb_onfault = onfault;
    959 
    960 				if (ret)
    961 					ok = false;
    962 			}
    963 			if (!ok) {
    964 				const u_int regno =
    965 				    __SHIFTOUT(opcode, __PABITS(27, 31));
    966 				tf_setregno(frame, regno, 0);
    967 				frame->tf_ipsw |= PSW_N;
    968 			}
    969 		}
    970 		break;
    971 
    972 	case T_DATACC:   	case T_USER | T_DATACC:
    973 	case T_ITLBMISS:	case T_USER | T_ITLBMISS:
    974 	case T_DTLBMISS:	case T_USER | T_DTLBMISS:
    975 	case T_TLB_DIRTY:	case T_USER | T_TLB_DIRTY:
    976 		vm = p->p_vmspace;
    977 
    978 		if (!vm) {
    979 #ifdef TRAPDEBUG
    980 			printf("trap: no vm, p=%p\n", p);
    981 #endif
    982 			goto dead_end;
    983 		}
    984 
    985 		/*
    986 		 * it could be a kernel map for exec_map faults
    987 		 */
    988 		if (!(type & T_USER) && space == HPPA_SID_KERNEL)
    989 			map = kernel_map;
    990 		else {
    991 			map = &vm->vm_map;
    992 		}
    993 
    994 		va = trunc_page(va);
    995 
    996 		if (map->pmap->pm_space != space) {
    997 #ifdef TRAPDEBUG
    998 			printf("trap: space mismatch %d != %d\n",
    999 			    space, map->pmap->pm_space);
   1000 #endif
   1001 			/* actually dump the user, crap the kernel */
   1002 			goto dead_end;
   1003 		}
   1004 
   1005 		/* Never call uvm_fault in interrupt context. */
   1006 		KASSERT(curcpu()->ci_intr_depth == 0);
   1007 
   1008 		onfault = pcb->pcb_onfault;
   1009 		pcb->pcb_onfault = 0;
   1010 		ret = uvm_fault(map, va, vftype);
   1011 		pcb->pcb_onfault = onfault;
   1012 
   1013 #ifdef TRAPDEBUG
   1014 		printf("uvm_fault(%p, %x, %d)=%d\n",
   1015 		    map, (u_int)va, vftype, ret);
   1016 #endif
   1017 
   1018 		/*
   1019 		 * If this was a stack access we keep track of the maximum
   1020 		 * accessed stack size.  Also, if uvm_fault gets a protection
   1021 		 * failure it is due to accessing the stack region outside
   1022 		 * the current limit and we need to reflect that as an access
   1023 		 * error.
   1024 		 */
   1025 		if (map != kernel_map && va >= (vaddr_t)vm->vm_minsaddr) {
   1026 			if (ret == 0)
   1027 				uvm_grow(l->l_proc, va);
   1028 			else if (ret == EACCES)
   1029 				ret = EFAULT;
   1030 		}
   1031 
   1032 		if (ret != 0) {
   1033 			if (type & T_USER) {
   1034 #ifdef DEBUG
   1035 				user_backtrace(frame, l, type);
   1036 #endif
   1037 				KSI_INIT_TRAP(&ksi);
   1038 				switch (ret) {
   1039 				case EACCES:
   1040 					ksi.ksi_signo = SIGSEGV;
   1041 					ksi.ksi_code = SEGV_ACCERR;
   1042 					break;
   1043 				case ENOMEM:
   1044 					ksi.ksi_signo = SIGKILL;
   1045 					printf("UVM: pid %d (%s), uid %d "
   1046 					    "killed: out of swap\n",
   1047 					    p->p_pid, p->p_comm,
   1048 					    l->l_cred ?
   1049 						kauth_cred_geteuid(l->l_cred)
   1050 						: -1);
   1051 					break;
   1052 				case EINVAL:
   1053 					ksi.ksi_signo = SIGBUS;
   1054 					ksi.ksi_code = BUS_ADRERR;
   1055 					break;
   1056 				default:
   1057 					ksi.ksi_signo = SIGSEGV;
   1058 					ksi.ksi_code = SEGV_MAPERR;
   1059 					break;
   1060 				}
   1061 				ksi.ksi_trap = type;
   1062 				ksi.ksi_addr = (void *)va;
   1063 				trapsignal(l, &ksi);
   1064 			} else {
   1065 				if (onfault) {
   1066 					goto do_onfault;
   1067 				}
   1068 				panic("trap: uvm_fault(%p, %lx, %d): %d",
   1069 				    map, va, vftype, ret);
   1070 			}
   1071 		}
   1072 		break;
   1073 
   1074 	case T_DATALIGN | T_USER:
   1075 #ifdef DEBUG
   1076 		user_backtrace(frame, l, type);
   1077 #endif
   1078 		KSI_INIT_TRAP(&ksi);
   1079 		ksi.ksi_signo = SIGBUS;
   1080 		ksi.ksi_code = BUS_ADRALN;
   1081 		ksi.ksi_trap = type;
   1082 		ksi.ksi_addr = (void *)va;
   1083 		trapsignal(l, &ksi);
   1084 		break;
   1085 
   1086 	case T_INTERRUPT:
   1087 	case T_INTERRUPT | T_USER:
   1088 		hppa_intr(frame);
   1089 		mtctl(frame->tf_eiem, CR_EIEM);
   1090 		break;
   1091 
   1092 	case T_LOWERPL:
   1093 	case T_DPROT:
   1094 	case T_IPROT:
   1095 	case T_OVERFLOW:
   1096 	case T_CONDITION:
   1097 	case T_ILLEGAL:
   1098 	case T_HIGHERPL:
   1099 	case T_TAKENBR:
   1100 	case T_POWERFAIL:
   1101 	case T_LPMC:
   1102 	case T_PAGEREF:
   1103 	case T_DATAPID:  	case T_DATAPID  | T_USER:
   1104 		if (0 /* T-chip */) {
   1105 			break;
   1106 		}
   1107 		/* FALLTHROUGH to unimplemented */
   1108 	default:
   1109 		panic ("trap: unimplemented \'%s\' (%d)", tts, type);
   1110 	}
   1111 
   1112 #ifdef DIAGNOSTIC
   1113 	if (ci->ci_cpl != oldcpl)
   1114 		printf("WARNING: SPL (%d) NOT LOWERED ON TRAP (%d) EXIT\n",
   1115 		    ci->ci_cpl, trapnum);
   1116 #endif
   1117 
   1118 	if (type & T_USER)
   1119 		userret(l, l->l_md.md_regs);
   1120 
   1121 #ifdef DEBUG
   1122 	frame_sanity_check(__func__, __LINE__, type, frame, l);
   1123 	if (frame->tf_flags & TFF_LAST && (curlwp->l_flag & LW_IDLE) == 0)
   1124 		frame_sanity_check(__func__, __LINE__, type,
   1125 		    curlwp->l_md.md_regs, curlwp);
   1126 #endif /* DEBUG */
   1127 }
   1128 
   1129 void
   1130 md_child_return(struct lwp *l)
   1131 {
   1132 	/*
   1133 	 * Return values in the frame set by cpu_lwp_fork().
   1134 	 */
   1135 
   1136 	userret(l, l->l_md.md_regs);
   1137 #ifdef DEBUG
   1138 	frame_sanity_check(__func__, __LINE__, 0, l->l_md.md_regs, l);
   1139 #endif /* DEBUG */
   1140 }
   1141 
   1142 /*
   1143  * Process the tail end of a posix_spawn() for the child.
   1144  */
   1145 void
   1146 cpu_spawn_return(struct lwp *l)
   1147 {
   1148 
   1149 	userret(l, l->l_md.md_regs);
   1150 #ifdef DEBUG
   1151 	frame_sanity_check(__func__, __LINE__, 0, l->l_md.md_regs, l);
   1152 #endif /* DEBUG */
   1153 }
   1154 
   1155 #ifdef PTRACE
   1156 
   1157 #include <sys/ptrace.h>
   1158 
   1159 int
   1160 ss_get_value(struct lwp *l, vaddr_t addr, u_int *value)
   1161 {
   1162 	struct uio uio;
   1163 	struct iovec iov;
   1164 
   1165 	iov.iov_base = (void *)value;
   1166 	iov.iov_len = sizeof(u_int);
   1167 	uio.uio_iov = &iov;
   1168 	uio.uio_iovcnt = 1;
   1169 	uio.uio_offset = (off_t)addr;
   1170 	uio.uio_resid = sizeof(u_int);
   1171 	uio.uio_rw = UIO_READ;
   1172 	UIO_SETUP_SYSSPACE(&uio);
   1173 
   1174 	return (process_domem(curlwp, l, &uio));
   1175 }
   1176 
   1177 int
   1178 ss_put_value(struct lwp *l, vaddr_t addr, u_int value)
   1179 {
   1180 	struct uio uio;
   1181 	struct iovec iov;
   1182 
   1183 	iov.iov_base = (void *)&value;
   1184 	iov.iov_len = sizeof(u_int);
   1185 	uio.uio_iov = &iov;
   1186 	uio.uio_iovcnt = 1;
   1187 	uio.uio_offset = (off_t)addr;
   1188 	uio.uio_resid = sizeof(u_int);
   1189 	uio.uio_rw = UIO_WRITE;
   1190 	UIO_SETUP_SYSSPACE(&uio);
   1191 
   1192 	return (process_domem(curlwp, l, &uio));
   1193 }
   1194 
   1195 void
   1196 ss_clear_breakpoints(struct lwp *l)
   1197 {
   1198 	/* Restore original instructions. */
   1199 	if (l->l_md.md_bpva != 0) {
   1200 		ss_put_value(l, l->l_md.md_bpva, l->l_md.md_bpsave[0]);
   1201 		ss_put_value(l, l->l_md.md_bpva + 4, l->l_md.md_bpsave[1]);
   1202 		l->l_md.md_bpva = 0;
   1203 	}
   1204 }
   1205 
   1206 
   1207 int
   1208 process_sstep(struct lwp *l, int sstep)
   1209 {
   1210 	struct trapframe *tf = l->l_md.md_regs;
   1211 	int error;
   1212 
   1213 	ss_clear_breakpoints(l);
   1214 
   1215 	/* We're continuing... */
   1216 	if (sstep == 0) {
   1217 		tf->tf_ipsw &= ~PSW_T;
   1218 		return 0;
   1219 	}
   1220 
   1221 	/*
   1222 	 * Don't touch the syscall gateway page.  Instead, insert a
   1223 	 * breakpoint where we're supposed to return.
   1224 	 */
   1225 	if ((tf->tf_iioq_tail & ~PAGE_MASK) == SYSCALLGATE)
   1226 		l->l_md.md_bpva = tf->tf_r31 & ~HPPA_PC_PRIV_MASK;
   1227 	else
   1228 		l->l_md.md_bpva = tf->tf_iioq_tail & ~HPPA_PC_PRIV_MASK;
   1229 
   1230 	error = ss_get_value(l, l->l_md.md_bpva, &l->l_md.md_bpsave[0]);
   1231 	if (error)
   1232 		return error;
   1233 	error = ss_get_value(l, l->l_md.md_bpva + 4, &l->l_md.md_bpsave[1]);
   1234 	if (error)
   1235 		return error;
   1236 
   1237 	error = ss_put_value(l, l->l_md.md_bpva, SSBREAKPOINT);
   1238 	if (error)
   1239 		return error;
   1240 	error = ss_put_value(l, l->l_md.md_bpva + 4, SSBREAKPOINT);
   1241 	if (error)
   1242 		return error;
   1243 
   1244 	if ((tf->tf_iioq_tail & ~PAGE_MASK) == SYSCALLGATE)
   1245 		tf->tf_ipsw &= ~PSW_T;
   1246 	else
   1247 		tf->tf_ipsw |= PSW_T;
   1248 
   1249 	return 0;
   1250 }
   1251 #endif
   1252 
   1253 
   1254 void
   1255 syscall_intern(struct proc *p)
   1256 {
   1257 	p->p_md.md_syscall = syscall;
   1258 }
   1259 
   1260 /*
   1261  * call actual syscall routine
   1262  * from the low-level syscall handler:
   1263  * - all HPPA_FRAME_NARGS syscall's arguments supposed to be copied onto
   1264  *   our stack, this wins compared to copyin just needed amount anyway
   1265  * - register args are copied onto stack too
   1266  */
   1267 void
   1268 syscall(struct trapframe *frame, int *args)
   1269 {
   1270 	struct lwp *l;
   1271 	struct proc *p;
   1272 	const struct sysent *callp;
   1273 	size_t nargs64;
   1274 	int nsys, code, error;
   1275 	int tmp;
   1276 	int rval[2];
   1277 #ifdef DIAGNOSTIC
   1278 	struct cpu_info *ci = curcpu();
   1279 	int oldcpl = ci->ci_cpl;
   1280 #endif
   1281 
   1282 	curcpu()->ci_data.cpu_nsyscall++;
   1283 
   1284 #ifdef DEBUG
   1285 	frame_sanity_check(__func__, __LINE__, 0, frame, curlwp);
   1286 #endif /* DEBUG */
   1287 
   1288 	if (!USERMODE(frame->tf_iioq_head))
   1289 		panic("syscall");
   1290 
   1291 	KASSERT(curlwp != NULL);
   1292 	l = curlwp;
   1293 	p = l->l_proc;
   1294 	l->l_md.md_regs = frame;
   1295 	nsys = p->p_emul->e_nsysent;
   1296 	callp = p->p_emul->e_sysent;
   1297 	code = frame->tf_t1;
   1298 
   1299 	/*
   1300 	 * Restarting a system call is touchy on the HPPA, because syscall
   1301 	 * arguments are passed in registers and the program counter of the
   1302 	 * syscall "point" isn't easily divined.
   1303 	 *
   1304 	 * We handle the first problem by assuming that we will have to restart
   1305 	 * this system call, so we stuff the first four words of the original
   1306 	 * arguments back into the frame as arg0...arg3, which is where we
   1307 	 * found them in the first place.  Any further arguments are (still) on
   1308 	 * the user's stack and the  syscall code will fetch them from there
   1309 	 * (again).
   1310 	 *
   1311 	 * The program counter problem is addressed below.
   1312 	 */
   1313 	frame->tf_arg0 = args[0];
   1314 	frame->tf_arg1 = args[1];
   1315 	frame->tf_arg2 = args[2];
   1316 	frame->tf_arg3 = args[3];
   1317 
   1318 	/*
   1319 	 * Some special handling for the syscall(2) and
   1320 	 * __syscall(2) system calls.
   1321 	 */
   1322 	switch (code) {
   1323 	case SYS_syscall:
   1324 		code = *args;
   1325 		args += 1;
   1326 		break;
   1327 	case SYS___syscall:
   1328 		if (callp != sysent)
   1329 			break;
   1330 		/*
   1331 		 * NB: even though __syscall(2) takes a quad_t containing the
   1332 		 * system call number, because our argument copying word-swaps
   1333 		 * 64-bit arguments, the least significant word of that quad_t
   1334 		 * is the first word in the argument array.
   1335 		 */
   1336 		code = *args;
   1337 		args += 2;
   1338 	}
   1339 
   1340 	/*
   1341 	 * Stacks growing from lower addresses to higher addresses are not
   1342 	 * really such a good idea, because it makes it impossible to overlay a
   1343 	 * struct on top of C stack arguments (the arguments appear in
   1344 	 * reversed order).
   1345 	 *
   1346 	 * You can do the obvious thing (as locore.S does) and copy argument
   1347 	 * words one by one, laying them out in the "right" order in the dest-
   1348 	 * ination buffer, but this ends up word-swapping multi-word arguments
   1349 	 * (like off_t).
   1350 	 *
   1351 	 * FIXME - this works only on native binaries and
   1352 	 * will probably screw up any and all emulation.
   1353 	 *
   1354 	 */
   1355 
   1356 	if (code < 0 || code >= nsys)
   1357 		callp += p->p_emul->e_nosys;	/* bad syscall # */
   1358 	else
   1359 		callp += code;
   1360 
   1361 	nargs64 = SYCALL_NARGS64(callp);
   1362 	if (nargs64 != 0) {
   1363 		size_t nargs = callp->sy_narg;
   1364 
   1365 		for (size_t i = 0; i < nargs + nargs64;) {
   1366 			if (SYCALL_ARG_64_P(callp, i)) {
   1367 				tmp = args[i];
   1368 				args[i] = args[i + 1];
   1369 				args[i + 1] = tmp;
   1370 				i += 2;
   1371 			} else
   1372 				i++;
   1373 		}
   1374 	}
   1375 
   1376 #ifdef USERTRACE
   1377 	if (0) {
   1378 		user_backtrace(frame, l, -1);
   1379 		frame->tf_ipsw |= PSW_R;
   1380 		frame->tf_rctr = 0;
   1381 		printf("r %08x", frame->tf_iioq_head);
   1382 		rctr_next_iioq = frame->tf_iioq_head + 4;
   1383 	}
   1384 #endif
   1385 
   1386 	error = sy_invoke(callp, l, args, rval, code);
   1387 
   1388 	switch (error) {
   1389 	case 0:
   1390 		l = curlwp;			/* changes on exec() */
   1391 		frame = l->l_md.md_regs;
   1392 		frame->tf_ret0 = rval[0];
   1393 		frame->tf_ret1 = rval[1];
   1394 		frame->tf_t1 = 0;
   1395 		break;
   1396 	case ERESTART:
   1397 		/*
   1398 		 * Now we have to wind back the instruction offset queue to the
   1399 		 * point where the system call will be made again.  This is
   1400 		 * inherently tied to the SYSCALL macro.
   1401 		 *
   1402 		 * Currently, the part of the SYSCALL macro that we want to re-
   1403 		 * run reads as:
   1404 		 *
   1405 		 *	ldil	L%SYSCALLGATE, r1
   1406 		 *	ble	4(srX, r1)
   1407 		 *	ldi	__CONCAT(SYS_,x), t1
   1408 		 *	comb,<>	%r0, %t1, __cerror
   1409 		 *
   1410 		 * And our offset queue head points to the comb instruction.
   1411 		 * So we need to subtract twelve to reach the ldil.
   1412 		 */
   1413 		frame->tf_iioq_head -= 12;
   1414 		frame->tf_iioq_tail = frame->tf_iioq_head + 4;
   1415 		break;
   1416 	case EJUSTRETURN:
   1417 		p = curproc;
   1418 		break;
   1419 	default:
   1420 		if (p->p_emul->e_errno)
   1421 			error = p->p_emul->e_errno[error];
   1422 		frame->tf_t1 = error;
   1423 		break;
   1424 	}
   1425 
   1426 	userret(l, frame);
   1427 
   1428 #ifdef DIAGNOSTIC
   1429 	if (ci->ci_cpl != oldcpl) {
   1430 		printf("WARNING: SPL (0x%x) NOT LOWERED ON "
   1431 		    "syscall(0x%x, 0x%x, 0x%x, 0x%x...) EXIT, PID %d\n",
   1432 		    ci->ci_cpl, code, args[0], args[1], args[2], p->p_pid);
   1433 		ci->ci_cpl = oldcpl;
   1434 	}
   1435 #endif
   1436 
   1437 #ifdef DEBUG
   1438 	frame_sanity_check(__func__, __LINE__, 0, frame, l);
   1439 #endif /* DEBUG */
   1440 }
   1441 
   1442 /*
   1443  * Start a new LWP
   1444  */
   1445 void
   1446 startlwp(void *arg)
   1447 {
   1448 	ucontext_t *uc = arg;
   1449 	lwp_t *l = curlwp;
   1450 	int error __diagused;
   1451 
   1452 	error = cpu_setmcontext(l, &uc->uc_mcontext, uc->uc_flags);
   1453 	KASSERT(error == 0);
   1454 
   1455 	kmem_free(uc, sizeof(ucontext_t));
   1456 	userret(l, l->l_md.md_regs);
   1457 }
   1458