hp300/hp300/machdep.c

/*
 * Copyright (c) 1988 University of Utah.
 * Copyright (c) 1982, 1986, 1990 The Regents of the University of California.
 * All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * the Systems Programming Group of the University of Utah Computer
 * Science Department.
 *
 * %sccs.include.redist.c%
 *
 * from: Utah $Hdr: machdep.c 1.51 89/11/28$
 *
 *	@(#)machdep.c	7.3 (Berkeley) 05/25/90
 */

#include "param.h"
#include "systm.h"
#include "user.h"
#include "kernel.h"
#include "map.h"
#include "vm.h"
#include "proc.h"
#include "buf.h"
#include "reboot.h"
#include "conf.h"
#include "file.h"
#include "text.h"
#include "clist.h"
#include "callout.h"
#include "cmap.h"
#include "malloc.h"
#include "mbuf.h"
#include "msgbuf.h"
#ifdef SYSVSHM
#include "shm.h"
#endif
#ifdef HPUXCOMPAT
#include "../hpux/hpux.h"
#endif

#include "cpu.h"
#include "reg.h"
#include "pte.h"
#include "psl.h"
#include "isr.h"
#include "../net/netisr.h"

#define RETURN(value)   { u.u_error = (value); return; }

/*
 * Declare these as initialized data so we can patch them.
 */
int	nswbuf = 0;
#ifdef	NBUF
int	nbuf = NBUF;
#else
int	nbuf = 0;
#endif
#ifdef	BUFPAGES
int	bufpages = BUFPAGES;
#else
int	bufpages = 0;
#endif
int	msgbufmapped;		/* set when safe to use msgbuf */
int	physmem = MAXMEM;	/* max supported memory, changes to actual */

extern	u_int lowram;

/*
 * Machine-dependent startup code
 */
startup(firstaddr)
	int firstaddr;
{
	register int unixsize;
	register unsigned i;
	register struct pte *pte;
	int mapaddr, j, n;
	register caddr_t v;
	int maxbufs, base, residual;
	extern long Usrptsize;
	extern struct map *useriomap;

	/*
	 * Set cpuspeed immediately since cninit() called routines
	 * might use delay.
	 */
	switch (machineid) {
	case HP_320:
	case HP_330:
	case HP_340:
		cpuspeed = MHZ_16;
		break;
	case HP_350:
	case HP_360:
		cpuspeed = MHZ_25;
		break;
	case HP_370:
		cpuspeed = MHZ_33;
		break;
	case HP_375:
		cpuspeed = MHZ_50;
		break;
	}
	/*
         * Find what hardware is attached to this machine.
         */
	find_devs();
	/*
	 * Initialize the console before we print anything out.
	 */
	cninit();
	/*
	 * Initialize error message buffer (at end of core).
	 */
	maxmem -= btoc(sizeof (struct msgbuf));
	pte = msgbufmap;
	for (i = 0; i < btoc(sizeof (struct msgbuf)); i++)
		*(int *)pte++ = PG_CI | PG_V | PG_KW | (ctob(maxmem + i));
	TBIAS();
	msgbufmapped = 1;

	/*
	 * Good {morning,afternoon,evening,night}.
	 */
	printf(version);
	identifycpu();
	printf("real mem = %d\n", ctob(physmem));

	/*
	 * Allocate space for system data structures.
	 * The first available real memory address is in "firstaddr".
	 * The first available kernel virtual address is in "v".
	 * As pages of kernel virtual memory are allocated, "v" is incremented.
	 * As pages of memory are allocated and cleared,
	 * "firstaddr" is incremented.
	 * An index into the kernel page table corresponding to the
	 * virtual memory address maintained in "v" is kept in "mapaddr".
	 */
	v = (caddr_t)((firstaddr * NBPG) - lowram);
	mapaddr = (int)v;
#define	valloc(name, type, num) \
	    (name) = (type *)v; v = (caddr_t)((name)+(num))
#define	valloclim(name, type, num, lim) \
	    (name) = (type *)v; v = (caddr_t)((lim) = ((name)+(num)))
	valloclim(file, struct file, nfile, fileNFILE);
	valloclim(proc, struct proc, nproc, procNPROC);
	valloclim(text, struct text, ntext, textNTEXT);
	valloc(cfree, struct cblock, nclist);
	valloc(callout, struct callout, ncallout);
	valloc(swapmap, struct map, nswapmap = nproc * 2);
	valloc(argmap, struct map, ARGMAPSIZE);
	valloc(kernelmap, struct map, nproc);
	valloc(mbmap, struct map, nmbclusters/4);
	valloc(kmemmap, struct map, ekmempt - kmempt);
	valloc(kmemusage, struct kmemusage, ekmempt - kmempt);
	valloc(useriomap, struct map, nproc);
#ifdef SYSVSHM
	valloc(shmsegs, struct shmid_ds, shminfo.shmmni);
#endif

	/*
	 * Determine how many buffers to allocate.
	 * Since HPs tend to be long on memory and short on disk speed,
	 * we allocate more buffer space than the BSD standard of
	 * use 10% of memory for the first 2 Meg, 5% of remaining.
	 * We just allocate a flat 10%.  Insure a minimum of 16 buffers.
	 * We allocate 1/2 as many swap buffer headers as file i/o buffers.
	 */
	if (bufpages == 0)
		bufpages = physmem / 10 / CLSIZE;
	if (nbuf == 0) {
		nbuf = bufpages;
		if (nbuf < 16)
			nbuf = 16;
	}
	if (nswbuf == 0) {
		nswbuf = (nbuf / 2) &~ 1;	/* force even */
		if (nswbuf > 256)
			nswbuf = 256;		/* sanity */
	}
	valloc(swbuf, struct buf, nswbuf);

	/*
	 * Now the amount of virtual memory remaining for buffers
	 * can be calculated, estimating needs for the cmap.
	 */
	ncmap = (maxmem*NBPG - (firstaddr*NBPG + ((int)v - mapaddr))) /
		(CLBYTES + sizeof(struct cmap)) + 2;
	maxbufs = ((SYSPTSIZE * NBPG) -
		(int)(v + ncmap * sizeof(struct cmap))) /
		(MAXBSIZE + sizeof(struct buf));
	if (maxbufs < 16)
		panic("sys pt too small");
	if (nbuf > maxbufs) {
		printf("SYSPTSIZE limits number of buffers to %d\n", maxbufs);
		nbuf = maxbufs;
	}
	if (bufpages > nbuf * (MAXBSIZE / CLBYTES))
		bufpages = nbuf * (MAXBSIZE / CLBYTES);
	valloc(buf, struct buf, nbuf);

	/*
	 * Allocate space for core map.
	 * Allow space for all of physical memory minus the amount
	 * dedicated to the system. The amount of physical memory
	 * dedicated to the system is the total virtual memory of
	 * the system thus far, plus core map, buffer pages,
	 * and buffer headers not yet allocated.
	 * Add 2: 1 because the 0th entry is unused, 1 for rounding.
	 */
	ncmap = (maxmem*NBPG - (firstaddr * NBPG +
		((int)(v + bufpages*CLBYTES) - mapaddr))) /
		(CLBYTES + sizeof(struct cmap)) + 2;
	valloclim(cmap, struct cmap, ncmap, ecmap);

	/*
	 * Clear space allocated thus far, and make r/w entries
	 * for the space in the kernel map.
	 */
	unixsize = btoc(v);
	mapaddr = btoc(mapaddr);
	while (mapaddr < unixsize) {
		*(int *)(&Sysmap[mapaddr]) = PG_V | PG_KW | ctob(firstaddr);
		clearseg((unsigned)firstaddr);
		firstaddr++;
		mapaddr++;
	}

	/*
	 * Now allocate buffers proper.  They are different than the above
	 * in that they usually occupy more virtual memory than physical.
	 */
	v = (caddr_t) ((int)(v + PGOFSET) &~ PGOFSET);
	valloc(buffers, char, MAXBSIZE * nbuf);
	base = bufpages / nbuf;
	residual = bufpages % nbuf;
	for (i = 0; i < nbuf; i++) {
		n = (i < residual ? base + 1 : base) * CLSIZE;
		for (j = 0; j < n; j++) {
			*(int *)(&Sysmap[mapaddr+j]) =
			    PG_CI | PG_V | PG_KW | ctob(firstaddr);
			clearseg((unsigned)firstaddr);
			firstaddr++;
		}
		mapaddr += MAXBSIZE / NBPG;
	}

	unixsize = btoc(v);
	if (firstaddr - Sysmap[0].pg_pfnum >= physmem - 8*UPAGES)
		panic("no memory");
	TBIA();				/* After we just cleared it all! */

	/*
	 * Initialize callouts
	 */
	callfree = callout;
	for (i = 1; i < ncallout; i++)
		callout[i-1].c_next = &callout[i];

	/*
	 * Initialize memory allocator and swap
	 * and user page table maps.
	 *
	 * THE USER PAGE TABLE MAP IS CALLED ``kernelmap''
	 * WHICH IS A VERY UNDESCRIPTIVE AND INCONSISTENT NAME.
	 */
	meminit(firstaddr, maxmem);
	maxmem = freemem;
	printf("avail mem = %d\n", ctob(maxmem));
	printf("using %d buffers containing %d bytes of memory\n",
		nbuf, bufpages * CLBYTES);
	rminit(kernelmap, (long)&Usrptsize-CLSIZE, (long)1, "usrpt", nproc);
	rminit(useriomap, (long)USRIOSIZE, (long)1, "usrio", nproc);
	rminit(mbmap, (long)(nmbclusters * MCLBYTES / NBPG), (long)CLSIZE,
	    "mbclusters", nmbclusters/4);
	kmeminit();	/* now safe to do malloc/free */

	/*
	 * Set up CPU-specific registers, cache, etc.
	 */
	initcpu();

	/*
	 * Set up buffers, so they can be used to read disk labels.
	 */
	bhinit();
	binit();

	/*
	 * Configure the system.
	 */
	configure();
}

#ifdef PGINPROF
/*
 * Return the difference (in microseconds)
 * between the  current time and a previous
 * time as represented by the arguments.
 */
/*ARGSUSED*/
vmtime(otime, olbolt, oicr)
	register int otime, olbolt, oicr;
{

	return (((time.tv_sec-otime)*100 + lbolt-olbolt)*10000);
}
#endif

/*
 * Clear registers on exec
 */
setregs(entry)
	u_long entry;
{
	u.u_ar0[PC] = entry & ~1;
#ifdef FPCOPROC
	/* restore a null state frame */
	u.u_pcb.pcb_fpregs.fpf_null = 0;
	m68881_restore(&u.u_pcb.pcb_fpregs);
#endif
#ifdef HPUXCOMPAT
	if (u.u_procp->p_flag & SHPUX) {

		u.u_ar0[A0] = 0;	/* not 68010 (bit 31), no FPA (30) */
		u.u_r.r_val1 = 0;	/* no float card */
#ifdef FPCOPROC
		u.u_r.r_val2 = 1;	/* yes 68881 */
#else
		u.u_r.r_val2 = 0;	/* no 68881 */
#endif
	}
	/*
	 * Ensure we perform the right action on traps type 1 and 2:
	 * If our parent is an HPUX process and we are being traced, turn
	 * on HPUX style interpretation.  Else if we were using the HPUX
	 * style interpretation, revert to the BSD interpretation.
	 *
	 * XXX This doesn't have much to do with setting registers but
	 * I didn't want to muck up kern_exec.c with this code, so I
	 * stuck it here.
	 */
	if ((u.u_procp->p_pptr->p_flag & SHPUX) &&
	    (u.u_procp->p_flag & STRC)) {
		tweaksigcode(1);
		u.u_pcb.pcb_flags |= PCB_HPUXTRACE;
	} else if (u.u_pcb.pcb_flags & PCB_HPUXTRACE) {
		tweaksigcode(0);
		u.u_pcb.pcb_flags &= ~PCB_HPUXTRACE;
	}
#endif
}

identifycpu()
{
	printf("HP9000/");
	switch (machineid) {
	case HP_320:
		printf("320 (16.67Mhz");
		break;
	case HP_330:
		printf("318/319/330 (16.67Mhz");
		break;
	case HP_340:
		printf("340 (16.67Mhz");
		break;
	case HP_350:
		printf("350 (25Mhz");
		break;
	case HP_360:
		printf("360 (25Mhz");
		break;
	case HP_370:
		printf("370 (33.33Mhz");
		break;
	case HP_375:
		printf("345/375 (50Mhz");
		break;
	default:
		printf("\nunknown machine type %d\n", machineid);
		panic("startup");
	}
	printf(" MC680%s CPU", mmutype == MMU_68030 ? "30" : "20");
	switch (mmutype) {
	case MMU_68030:
		printf("+MMU");
		break;
	case MMU_68851:
		printf(", MC68851 MMU");
		break;
	case MMU_HP:
		printf(", HP MMU");
		break;
	default:
		printf("\nunknown MMU type %d\n", mmutype);
		panic("startup");
	}
	if (mmutype == MMU_68030)
		printf(", %sMhz MC68882 FPU",
		       machineid == HP_340 ? "16.67" :
		       (machineid == HP_360 ? "25" :
			(machineid == HP_370 ? "33.33" : "50")));
	else
		printf(", %sMhz MC68881 FPU",
		       machineid == HP_350 ? "20" : "16.67");
	switch (ectype) {
	case EC_VIRT:
		printf(", %dK virtual-address cache",
		       machineid == HP_320 ? 16 : 32);
		break;
	case EC_PHYS:
		printf(", %dK physical-address cache",
		       machineid == HP_370 ? 64 : 32);
		break;
	}
	printf(")\n");
	/*
	 * Now that we have told the user what they have,
	 * let them know if that machine type isn't configured.
	 */
	switch (machineid) {
	case -1:		/* keep compilers happy */
#if !defined(HP320) && !defined(HP350)
	case HP_320:
	case HP_350:
#endif
#ifndef HP330
	case HP_330:
#endif
#if !defined(HP360) && !defined(HP370)
	case HP_340:
	case HP_360:
	case HP_370:
#endif
		panic("CPU type not configured");
	default:
		break;
	}
}

#ifdef HPUXCOMPAT
tweaksigcode(ishpux)
{
	static short *sigtrap = NULL;

	/* locate trap instruction in pcb_sigc */
	if (sigtrap == NULL) {
		register struct pcb *pcp = &u.u_pcb;

		sigtrap = &pcp->pcb_sigc[sizeof(pcp->pcb_sigc)/sizeof(short)];
		while (--sigtrap >= pcp->pcb_sigc)
			if ((*sigtrap & 0xFFF0) == 0x4E40)
				break;
		if (sigtrap < pcp->pcb_sigc)
			panic("bogus sigcode\n");
	}
	*sigtrap = ishpux ? 0x4E42 : 0x4E41;
}
#endif

#define SS_RTEFRAME	1
#define SS_FPSTATE	2
#define SS_USERREGS	4

struct sigstate {
	int	ss_flags;		/* which of the following are valid */
	struct	frame ss_frame;		/* original exception frame */
	struct	fpframe ss_fpstate;	/* 68881/68882 state info */
};

/*
 * WARNING: code in locore.s assumes the layout shown for sf_signum
 * thru sf_handler so... don't screw with them!
 */
struct sigframe {
	int	sf_signum;		/* signo for handler */
	int	sf_code;		/* additional info for handler */
	struct	sigcontext *sf_scp;	/* context ptr for handler */
	sig_t	sf_handler;		/* handler addr for u_sigc */
	struct	sigstate sf_state;	/* state of the hardware */
	struct	sigcontext sf_sc;	/* actual context */
};

#ifdef HPUXCOMPAT
struct	hpuxsigcontext {
	int	hsc_syscall;
	char	hsc_action;
	char	hsc_pad1;
	char	hsc_pad2;
	char	hsc_onstack;
	int	hsc_mask;
	int	hsc_sp;
	short	hsc_ps;
	int	hsc_pc;
/* the rest aren't part of the context but are included for our convenience */
	short	hsc_pad;
	u_int	hsc_magic;		/* XXX sigreturn: cookie */
	struct	sigcontext *hsc_realsc;	/* XXX sigreturn: ptr to BSD context */
};

/*
 * For an HP-UX process, a partial hpuxsigframe follows the normal sigframe.
 * Tremendous waste of space, but some HP-UX applications (e.g. LCL) need it.
 */
struct hpuxsigframe {
	int	hsf_signum;
	int	hsf_code;
	struct	sigcontext *hsf_scp;
	struct	hpuxsigcontext hsf_sc;
	int	hsf_regs[15];
};
#endif

#ifdef DEBUG
int sigdebug = 0;
int sigpid = 0;
#define SDB_FOLLOW	0x01
#define SDB_KSTACK	0x02
#define SDB_FPSTATE	0x04
#endif

/*
 * Send an interrupt to process.
 */
sendsig(catcher, sig, mask, code)
	sig_t catcher;
	int sig, mask;
	unsigned code;
{
	register struct proc *p = u.u_procp;
	register struct sigframe *fp, *kfp;
	register struct frame *frame;
	register short ft;
	int oonstack, fsize;

	frame = (struct frame *)u.u_ar0;
	ft = frame->f_format;
	oonstack = u.u_onstack;
	/*
	 * Allocate and validate space for the signal handler
	 * context. Note that if the stack is in P0 space, the
	 * call to grow() is a nop, and the useracc() check
	 * will fail if the process has not already allocated
	 * the space with a `brk'.
	 */
#ifdef HPUXCOMPAT
	if (p->p_flag & SHPUX)
		fsize = sizeof(struct sigframe) + sizeof(struct hpuxsigframe);
	else
#endif
	fsize = sizeof(struct sigframe);
	if (!u.u_onstack && (u.u_sigonstack & sigmask(sig))) {
		fp = (struct sigframe *)(u.u_sigsp - fsize);
		u.u_onstack = 1;
	} else
		fp = (struct sigframe *)(frame->f_regs[SP] - fsize);
	if ((unsigned)fp <= USRSTACK - ctob(u.u_ssize))
		(void)grow((unsigned)fp);
#ifdef DEBUG
	if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid)
		printf("sendsig: pid %d, sig %d ssp %x usp %x scp %x ft %d\n",
		       p->p_pid, sig, &oonstack, fp, &fp->sf_sc, ft);
#endif
	if (useracc((caddr_t)fp, fsize, B_WRITE) == 0) {
#ifdef DEBUG
		if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid)
			printf("sendsig: pid %d, useracc failed on sig %d\n",
			       p->p_pid, sig);
#endif
		/*
		 * Process has trashed its stack; give it an illegal
		 * instruction to halt it in its tracks.
		 */
		SIGACTION(p, SIGILL) = SIG_DFL;
		sig = sigmask(SIGILL);
		p->p_sigignore &= ~sig;
		p->p_sigcatch &= ~sig;
		p->p_sigmask &= ~sig;
		psignal(p, SIGILL);
		return;
	}
	kfp = (struct sigframe *)malloc((u_long)fsize, M_TEMP, M_WAITOK);
	/*
	 * Build the argument list for the signal handler.
	 */
	kfp->sf_signum = sig;
	kfp->sf_code = code;
	kfp->sf_scp = &fp->sf_sc;
	kfp->sf_handler = catcher;
	/*
	 * Save necessary hardware state.  Currently this includes:
	 *	- general registers
	 *	- original exception frame (if not a "normal" frame)
	 *	- FP coprocessor state
	 */
	kfp->sf_state.ss_flags = SS_USERREGS;
	bcopy((caddr_t)frame->f_regs,
	      (caddr_t)kfp->sf_state.ss_frame.f_regs, sizeof frame->f_regs);
	if (ft >= FMT9) {
#ifdef DEBUG
		if (ft != FMT9 && ft != FMTA && ft != FMTB)
			panic("sendsig: bogus frame type");
#endif
		kfp->sf_state.ss_flags |= SS_RTEFRAME;
		kfp->sf_state.ss_frame.f_format = frame->f_format;
		kfp->sf_state.ss_frame.f_vector = frame->f_vector;
		bcopy((caddr_t)&frame->F_u,
		      (caddr_t)&kfp->sf_state.ss_frame.F_u,
		      (ft == FMT9) ? FMT9SIZE :
		      (ft == FMTA) ? FMTASIZE : FMTBSIZE);
		/*
		 * Gag!  Leave an indicator that we need to clean up the
		 * kernel stack.  We do this by setting the "pad word"
		 * above the hardware stack frame.  "bexit" in locore
		 * will then know that it must compress the kernel stack
		 * and create a normal four word stack frame.
		 */
		frame->f_stackadj = -1;
#ifdef DEBUG
		if (sigdebug & SDB_FOLLOW)
			printf("sendsig: pid %d, copy out %d of frame %d\n",
			       p->p_pid,
			       (ft == FMT9) ? FMT9SIZE :
			       (ft == FMTA) ? FMTASIZE : FMTBSIZE, ft);
#endif
	}
#ifdef FPCOPROC
	kfp->sf_state.ss_flags |= SS_FPSTATE;
	m68881_save(&kfp->sf_state.ss_fpstate);
#ifdef DEBUG
	if ((sigdebug & SDB_FPSTATE) && *(char *)&kfp->sf_state.ss_fpstate)
		printf("sendsig: pid %d, copy out FP state (%x) to %x\n",
		       p->p_pid, *(u_int *)&kfp->sf_state.ss_fpstate,
		       &kfp->sf_state.ss_fpstate);
#endif
#endif
	/*
	 * Build the signal context to be used by sigreturn.
	 */
	kfp->sf_sc.sc_onstack = oonstack;
	kfp->sf_sc.sc_mask = mask;
	kfp->sf_sc.sc_sp = frame->f_regs[SP];
	kfp->sf_sc.sc_fp = frame->f_regs[A6];
	kfp->sf_sc.sc_ap = (int)&fp->sf_state;
	kfp->sf_sc.sc_pc = frame->f_pc;
	kfp->sf_sc.sc_ps = frame->f_sr;
#ifdef HPUXCOMPAT
	/*
	 * Create an HP-UX style sigcontext structure and associated goo
	 */
	if (p->p_flag & SHPUX) {
		register struct hpuxsigframe *hkfp;

		hkfp = (struct hpuxsigframe *)&kfp[1];
		hkfp->hsf_signum = bsdtohpuxsig(kfp->sf_signum);
		hkfp->hsf_code = kfp->sf_code;
		hkfp->hsf_scp = (struct sigcontext *)
			&((struct hpuxsigframe *)(&fp[1]))->hsf_sc;
		hkfp->hsf_sc.hsc_syscall = 0;		/* XXX */
		hkfp->hsf_sc.hsc_action = 0;		/* XXX */
		hkfp->hsf_sc.hsc_pad1 = hkfp->hsf_sc.hsc_pad2 = 0;
		hkfp->hsf_sc.hsc_onstack = kfp->sf_sc.sc_onstack;
		hkfp->hsf_sc.hsc_mask = kfp->sf_sc.sc_mask;
		hkfp->hsf_sc.hsc_sp = kfp->sf_sc.sc_sp;
		hkfp->hsf_sc.hsc_ps = kfp->sf_sc.sc_ps;
		hkfp->hsf_sc.hsc_pc = kfp->sf_sc.sc_pc;
		hkfp->hsf_sc.hsc_pad = 0;
		hkfp->hsf_sc.hsc_magic = 0xdeadbeef;
		hkfp->hsf_sc.hsc_realsc = kfp->sf_scp;
		bcopy((caddr_t)frame->f_regs, (caddr_t)hkfp->hsf_regs,
		      sizeof (hkfp->hsf_regs));

		kfp->sf_signum = hkfp->hsf_signum;
		kfp->sf_scp = hkfp->hsf_scp;
	}
#endif
	(void) copyout((caddr_t)kfp, (caddr_t)fp, fsize);
	frame->f_regs[SP] = (int)fp;
#ifdef DEBUG
	if (sigdebug & SDB_FOLLOW)
		printf("sendsig: pid %d, scp %x, fp %x, sc_ap %x\n",
		       p->p_pid, kfp->sf_scp, fp, kfp->sf_sc.sc_ap);
#endif
	/*
	 * User PC is set to signal trampoline code.  The catch is that
	 * it must be set to reference the pcb via the user space address
	 * NOT via u.  Assumption: u-area is at USRSTACK.
	 */
	frame->f_pc = (int)((struct user *)USRSTACK)->u_pcb.pcb_sigc;
#ifdef DEBUG
	if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid)
		printf("sendsig: pid %d, sig %d, returns\n",
		       p->p_pid, sig);
#endif
	free((caddr_t)kfp, M_TEMP);
}

/*
 * System call to cleanup state after a signal
 * has been taken.  Reset signal mask and
 * stack state from context left by sendsig (above).
 * Return to previous pc and psl as specified by
 * context left by sendsig. Check carefully to
 * make sure that the user has not modified the
 * psl to gain improper priviledges or to cause
 * a machine fault.
 */
sigreturn()
{
	struct a {
		struct sigcontext *sigcntxp;
	};
	register struct sigcontext *scp;
	register struct frame *frame;
	register int rf;
	struct sigcontext tsigc;
	struct sigstate tstate;
	int flags;

	scp = ((struct a *)(u.u_ap))->sigcntxp;
#ifdef DEBUG
	if (sigdebug & SDB_FOLLOW)
		printf("sigreturn: pid %d, scp %x\n", u.u_procp->p_pid, scp);
#endif
	if ((int)scp & 1)
		RETURN (EINVAL);
#ifdef HPUXCOMPAT
	/*
	 * Grab context as an HP-UX style context and determine if it
	 * was one that we contructed in sendsig.
	 */
	if (u.u_procp->p_flag & SHPUX) {
		struct hpuxsigcontext *hscp = (struct hpuxsigcontext *)scp;
		struct hpuxsigcontext htsigc;

		if (useracc((caddr_t)hscp, sizeof (*hscp), B_WRITE) == 0 ||
		    copyin((caddr_t)hscp, (caddr_t)&htsigc, sizeof htsigc))
			RETURN (0);
		/*
		 * If not generated by sendsig or we cannot restore the
		 * BSD-style sigcontext, just restore what we can -- state
		 * will be lost, but them's the breaks.
		 */
		hscp = &htsigc;
		if (hscp->hsc_magic != 0xdeadbeef ||
		    (scp = hscp->hsc_realsc) == 0 ||
		    useracc((caddr_t)scp, sizeof (*scp), B_WRITE) == 0 ||
		    copyin((caddr_t)scp, (caddr_t)&tsigc, sizeof tsigc)) {
			u.u_onstack = hscp->hsc_onstack & 01;
			u.u_procp->p_sigmask = hscp->hsc_mask &~ sigcantmask;
			frame = (struct frame *) u.u_ar0;
			frame->f_regs[SP] = hscp->hsc_sp;
			frame->f_pc = hscp->hsc_pc;
			frame->f_sr = hscp->hsc_ps &~ PSL_USERCLR;
			RETURN (EJUSTRETURN);
		}
		/*
		 * Otherwise, overlay BSD context with possibly modified
		 * HP-UX values.
		 */
		tsigc.sc_onstack = hscp->hsc_onstack;
		tsigc.sc_mask = hscp->hsc_mask;
		tsigc.sc_sp = hscp->hsc_sp;
		tsigc.sc_ps = hscp->hsc_ps;
		tsigc.sc_pc = hscp->hsc_pc;
	} else
#endif
	/*
	 * Test and fetch the context structure.
	 * We grab it all at once for speed.
	 */
	if (useracc((caddr_t)scp, sizeof (*scp), B_WRITE) == 0 ||
	    copyin((caddr_t)scp, (caddr_t)&tsigc, sizeof tsigc))
		RETURN (EINVAL);
	scp = &tsigc;
	if ((scp->sc_ps & (PSL_MBZ|PSL_IPL|PSL_S)) != 0)
		RETURN (EINVAL);
	/*
	 * Restore the user supplied information
	 */
	u.u_onstack = scp->sc_onstack & 01;
	u.u_procp->p_sigmask = scp->sc_mask &~ sigcantmask;
	frame = (struct frame *) u.u_ar0;
	frame->f_regs[SP] = scp->sc_sp;
	frame->f_regs[A6] = scp->sc_fp;
	frame->f_pc = scp->sc_pc;
	frame->f_sr = scp->sc_ps;
	/*
	 * Grab pointer to hardware state information.
	 * If zero, the user is probably doing a longjmp.
	 */
	if ((rf = scp->sc_ap) == 0)
		RETURN (JUSTRETURN);
	/*
	 * See if there is anything to do before we go to the
	 * expense of copying in close to 1/2K of data
	 */
	flags = fuword((caddr_t)rf);
#ifdef DEBUG
	if (sigdebug & SDB_FOLLOW)
		printf("sigreturn: pid %d, sc_ap %x, flags %x\n",
		       u.u_procp->p_pid, rf, flags);
#endif
	if (flags == 0 || copyin((caddr_t)rf, (caddr_t)&tstate, sizeof tstate))
		RETURN (JUSTRETURN);
#ifdef DEBUG
	if ((sigdebug & SDB_KSTACK) && u.u_procp->p_pid == sigpid)
		printf("sigreturn: pid %d, ssp %x usp %x scp %x ft %d\n",
		       u.u_procp->p_pid, &flags, scp->sc_sp,
		       ((struct a *)(u.u_ap))->sigcntxp,
		       (flags&SS_RTEFRAME) ? tstate.ss_frame.f_format : -1);
#endif
	/*
	 * Restore most of the users registers except for A6 and SP
	 * which were handled above.
	 */
	if (flags & SS_USERREGS)
		bcopy((caddr_t)tstate.ss_frame.f_regs,
		      (caddr_t)frame->f_regs, sizeof(frame->f_regs)-2*NBPW);
	/*
	 * Restore long stack frames.  Note that we do not copy
	 * back the saved SR or PC, they were picked up above from
	 * the sigcontext structure.
	 */
	if (flags & SS_RTEFRAME) {
		register int sz;

		/* grab frame type and validate */
		sz = tstate.ss_frame.f_format;
		if (sz == FMT9)
			sz = FMT9SIZE;
		else if (sz == FMTA)
			sz = FMTASIZE;
		else if (sz == FMTB) {
			sz = FMTBSIZE;
			/* no k-stack adjustment necessary */
			frame->f_stackadj = 0;
		} else
			RETURN (EINVAL);
		frame->f_format = tstate.ss_frame.f_format;
		frame->f_vector = tstate.ss_frame.f_vector;
		bcopy((caddr_t)&tstate.ss_frame.F_u, (caddr_t)&frame->F_u, sz);
#ifdef DEBUG
		if (sigdebug & SDB_FOLLOW)
			printf("sigreturn: copy in %d of frame type %d\n",
			       sz, tstate.ss_frame.f_format);
#endif
	}
#ifdef FPCOPROC
	/*
	 * Finally we restore the original FP context
	 */
	if (flags & SS_FPSTATE)
		m68881_restore(&tstate.ss_fpstate);
#ifdef DEBUG
	if ((sigdebug & SDB_FPSTATE) && *(char *)&tstate.ss_fpstate)
		printf("sigreturn: pid %d, copied in FP state (%x) at %x\n",
		       u.u_procp->p_pid, *(u_int *)&tstate.ss_fpstate,
		       &tstate.ss_fpstate);
#endif
#endif
#ifdef DEBUG
	if ((sigdebug & SDB_FOLLOW) ||
	    ((sigdebug & SDB_KSTACK) && u.u_procp->p_pid == sigpid))
		printf("sigreturn: pid %d, returns\n", u.u_procp->p_pid);
#endif
	RETURN (JUSTRETURN);
}

int	waittime = -1;

boot(howto)
	register int howto;
{
	/* take a snap shot before clobbering any registers */
	resume((u_int)pcbb(u.u_procp));

	boothowto = howto;
	if ((howto&RB_NOSYNC) == 0 && waittime < 0 && bfreelist[0].b_forw) {
		register struct buf *bp;
		int iter, nbusy;

		waittime = 0;
		(void) spl0();
		printf("syncing disks... ");
		/*
		 * Release vnodes held by texts before sync.
		 */
		if (panicstr == 0)
			xumount(NULL);
#include "fd.h"
#if NFD > 0
		fdshutdown();
#endif
		sync((struct sigcontext *)0);

		for (iter = 0; iter < 20; iter++) {
			nbusy = 0;
			for (bp = &buf[nbuf]; --bp >= buf; )
				if ((bp->b_flags & (B_BUSY|B_INVAL)) == B_BUSY)
					nbusy++;
			if (nbusy == 0)
				break;
			printf("%d ", nbusy);
			DELAY(40000 * iter);
		}
		if (nbusy)
			printf("giving up\n");
		else
			printf("done\n");
		/*
		 * If we've been adjusting the clock, the todr
		 * will be out of synch; adjust it now.
		 */
		resettodr();
	}
	splhigh();			/* extreme priority */
	if (howto&RB_HALT) {
		printf("halted\n\n");
		asm("	stop	#0x2700");
	} else {
		if (howto & RB_DUMP)
			dumpsys();
		doboot();
		/*NOTREACHED*/
	}
	/*NOTREACHED*/
}

int	dumpmag = 0x8fca0101;	/* magic number for savecore */
int	dumpsize = 0;		/* also for savecore */

dumpconf()
{
	int nblks;

	dumpsize = physmem;
	if (dumpdev != NODEV && bdevsw[major(dumpdev)].d_psize) {
		nblks = (*bdevsw[major(dumpdev)].d_psize)(dumpdev);
		if (dumpsize > btoc(dbtob(nblks - dumplo)))
			dumpsize = btoc(dbtob(nblks - dumplo));
		else if (dumplo == 0)
			dumplo = nblks - btodb(ctob(physmem));
	}
	/*
	 * Don't dump on the first CLBYTES (why CLBYTES?)
	 * in case the dump device includes a disk label.
	 */
	if (dumplo < btodb(CLBYTES))
		dumplo = btodb(CLBYTES);
}

/*
 * Doadump comes here after turning off memory management and
 * getting on the dump stack, either when called above, or by
 * the auto-restart code.
 */
dumpsys()
{

	msgbufmapped = 0;
	if (dumpdev == NODEV)
		return;
	/*
	 * For dumps during autoconfiguration,
	 * if dump device has already configured...
	 */
	if (dumpsize == 0)
		dumpconf();
	if (dumplo < 0)
		return;
	printf("\ndumping to dev %x, offset %d\n", dumpdev, dumplo);
	printf("dump ");
	switch ((*bdevsw[major(dumpdev)].d_dump)(dumpdev)) {

	case ENXIO:
		printf("device bad\n");
		break;

	case EFAULT:
		printf("device not ready\n");
		break;

	case EINVAL:
		printf("area improper\n");
		break;

	case EIO:
		printf("i/o error\n");
		break;

	default:
		printf("succeeded\n");
		break;
	}
}

/*
 * Return the best possible estimate of the time in the timeval
 * to which tvp points.  We do this by returning the current time
 * plus the amount of time since the last clock interrupt (clock.c:clkread).
 *
 * Check that this time is no less than any previously-reported time,
 * which could happen around the time of a clock adjustment.  Just for fun,
 * we guarantee that the time will be greater than the value obtained by a
 * previous call.
 */
microtime(tvp)
	register struct timeval *tvp;
{
	int s = splhigh();
	static struct timeval lasttime;

	*tvp = time;
	tvp->tv_usec += clkread();
	while (tvp->tv_usec > 1000000) {
		tvp->tv_sec++;
		tvp->tv_usec -= 1000000;
	}
	if (tvp->tv_sec == lasttime.tv_sec &&
	    tvp->tv_usec <= lasttime.tv_usec &&
	    (tvp->tv_usec = lasttime.tv_usec + 1) > 1000000) {
		tvp->tv_sec++;
		tvp->tv_usec -= 1000000;
	}
	lasttime = *tvp;
	splx(s);
}

initcpu()
{
	parityenable();
}

straytrap(addr)
	register int addr;
{
	printf("stray trap, addr 0x%x\n", addr);
}

int	*nofault;

badaddr(addr)
	register caddr_t addr;
{
	register int i;
	label_t	faultbuf;

#ifdef lint
	i = *addr; if (i) return(0);
#endif
	nofault = (int *) &faultbuf;
	if (setjmp((label_t *)nofault)) {
		nofault = (int *) 0;
		return(1);
	}
	i = *(volatile short *)addr;
	nofault = (int *) 0;
	return(0);
}

badbaddr(addr)
	register caddr_t addr;
{
	register int i;
	label_t	faultbuf;

#ifdef lint
	i = *addr; if (i) return(0);
#endif
	nofault = (int *) &faultbuf;
	if (setjmp((label_t *)nofault)) {
		nofault = (int *) 0;
		return(1);
	}
	i = *(volatile char *)addr;
	nofault = (int *) 0;
	return(0);
}

netintr()
{
#ifdef INET
	if (netisr & (1 << NETISR_IP)) {
		netisr &= ~(1 << NETISR_IP);
		ipintr();
	}
#endif
#ifdef NS
	if (netisr & (1 << NETISR_NS)) {
		netisr &= ~(1 << NETISR_NS);
		nsintr();
	}
#endif
#ifdef ISO
	if (netisr & (1 << NETISR_ISO)) {
		netisr &= ~(1 << NETISR_ISO);
		clnlintr();
	}
#endif
}

intrhand(sr)
	int sr;
{
	register struct isr *isr;
	register int found = 0;
	register int ipl;
	extern struct isr isrqueue[];

	ipl = (sr >> 8) & 7;
	switch (ipl) {

	case 3:
	case 4:
	case 5:
		ipl = ISRIPL(ipl);
		isr = isrqueue[ipl].isr_forw;
		for (; isr != &isrqueue[ipl]; isr = isr->isr_forw) {
			if ((isr->isr_intr)(isr->isr_arg)) {
				found++;
				break;
			}
		}
		if (found == 0)
			printf("stray interrupt, sr 0x%x\n", sr);
		break;

	case 0:
	case 1:
	case 2:
	case 6:
	case 7:
		printf("intrhand: unexpected sr 0x%x\n", sr);
		break;
	}
}

#if defined(DEBUG) && !defined(PANICBUTTON)
#define PANICBUTTON
#endif

#ifdef PANICBUTTON
int panicbutton = 1;	/* non-zero if panic buttons are enabled */
int crashandburn = 0;
int candbdelay = 50;	/* give em half a second */

candbtimer()
{
	crashandburn = 0;
}
#endif

/*
 * Level 7 interrupts can be caused by the keyboard or parity errors.
 */
nmihand(frame)
	struct frame frame;
{
	if (kbdnmi()) {
#ifdef PANICBUTTON
		printf("Got a keyboard NMI\n");
		if (panicbutton) {
			if (crashandburn) {
				crashandburn = 0;
				panic(panicstr ?
				      "forced crash, nosync" : "forced crash");
			}
			crashandburn++;
			timeout(candbtimer, (caddr_t)0, candbdelay);
		}
#endif
		return;
	}
	if (parityerror(&frame))
		return;
	/* panic?? */
	printf("unexpected level 7 interrupt ignored\n");
}

/*
 * Parity error section.  Contains magic.
 */
#define PARREG		((volatile short *)IOV(0x5B0000))
static int gotparmem = 0;
#ifdef DEBUG
int ignorekperr = 0;	/* ignore kernel parity errors */
#endif

/*
 * Enable parity detection
 */
parityenable()
{
	label_t	faultbuf;

	nofault = (int *) &faultbuf;
	if (setjmp((label_t *)nofault)) {
		nofault = (int *) 0;
#ifdef DEBUG
		printf("No parity memory\n");
#endif
		return;
	}
	*PARREG = 1;
	nofault = (int *) 0;
	gotparmem = 1;
#ifdef DEBUG
	printf("Parity detection enabled\n");
#endif
}

/*
 * Determine if level 7 interrupt was caused by a parity error
 * and deal with it if it was.  Returns 1 if it was a parity error.
 */
parityerror(fp)
	struct frame *fp;
{
	if (!gotparmem)
		return(0);
	*PARREG = 0;
	DELAY(10);
	*PARREG = 1;
	if (panicstr) {
		printf("parity error after panic ignored\n");
		return(1);
	}
	if (!findparerror())
		printf("WARNING: transient parity error ignored\n");
	else if (USERMODE(fp->f_sr)) {
		printf("pid %d: parity error\n", u.u_procp->p_pid);
		uprintf("sorry, pid %d killed due to memory parity error\n",
			u.u_procp->p_pid);
		psignal(u.u_procp, SIGKILL);
#ifdef DEBUG
	} else if (ignorekperr) {
		printf("WARNING: kernel parity error ignored\n");
#endif
	} else {
		regdump(fp->f_regs, 128);
		panic("kernel parity error");
	}
	return(1);
}

/*
 * Yuk!  There has got to be a better way to do this!
 * Searching all of memory with interrupts blocked can lead to disaster.
 */
findparerror()
{
	static label_t parcatch;
	static int looking = 0;
	volatile struct pte opte;
	volatile int pg, o, s;
	register volatile int *ip;
	register int i;
	int found;

#ifdef lint
	ip = &found;
	i = o = pg = 0; if (i) return(0);
#endif
	/*
	 * If looking is true we are searching for a known parity error
	 * and it has just occured.  All we do is return to the higher
	 * level invocation.
	 */
	if (looking)
		longjmp(&parcatch);
	s = splhigh();
	/*
	 * If setjmp returns true, the parity error we were searching
	 * for has just occured (longjmp above) at the current pg+o
	 */
	if (setjmp(&parcatch)) {
		printf("Parity error at 0x%x\n", ctob(pg)|o);
		found = 1;
		goto done;
	}
	/*
	 * If we get here, a parity error has occured for the first time
	 * and we need to find it.  We turn off any external caches and
	 * loop thru memory, testing every longword til a fault occurs and
	 * we regain control at setjmp above.  Note that because of the
	 * setjmp, pg and o need to be volatile or their values will be lost.
	 */
	looking = 1;
	ecacheoff();
	opte = mmap[0];
	for (pg = btoc(lowram); pg < btoc(lowram)+physmem; pg++) {
		*(u_int *)mmap = PG_RO|PG_CI|PG_V;
		mmap[0].pg_pfnum = pg;
		TBIS(vmmap);
		ip = (int *)vmmap;
		for (o = 0; o < NBPG; o += sizeof(int))
			i = *ip++;
	}
	/*
	 * Getting here implies no fault was found.  Should never happen.
	 */
	printf("Couldn't locate parity error\n");
	found = 0;
done:
	looking = 0;
	mmap[0] = opte;
	TBIS(vmmap);
	ecacheon();
	splx(s);
	return(found);
}

regdump(rp, sbytes)
  int *rp; /* must not be register */
  int sbytes;
{
	static int doingdump = 0;
	register int i;
	int s;
	extern char *hexstr();

	if (doingdump)
		return;
	s = splhigh();
	doingdump = 1;
	printf("pid = %d, pc = %s, ", u.u_procp->p_pid, hexstr(rp[PC], 8));
	printf("ps = %s, ", hexstr(rp[PS], 4));
	printf("sfc = %s, ", hexstr(getsfc(), 4));
	printf("dfc = %s\n", hexstr(getdfc(), 4));
	printf("p0 = %x@%s, ",
	       u.u_pcb.pcb_p0lr, hexstr((int)u.u_pcb.pcb_p0br, 8));
	printf("p1 = %x@%s\n\n",
	       u.u_pcb.pcb_p1lr, hexstr((int)u.u_pcb.pcb_p1br, 8));
	printf("Registers:\n     ");
	for (i = 0; i < 8; i++)
		printf("        %d", i);
	printf("\ndreg:");
	for (i = 0; i < 8; i++)
		printf(" %s", hexstr(rp[i], 8));
	printf("\nareg:");
	for (i = 0; i < 8; i++)
		printf(" %s", hexstr(rp[i+8], 8));
	if (sbytes > 0) {
		if (rp[PS] & PSL_S) {
			printf("\n\nKernel stack (%s):",
			       hexstr((int)(((int *)&rp)-1), 8));
			dumpmem(((int *)&rp)-1, sbytes, 0);
		} else {
			printf("\n\nUser stack (%s):", hexstr(rp[SP], 8));
			dumpmem((int *)rp[SP], sbytes, 1);
		}
	}
	doingdump = 0;
	splx(s);
}

#define KSADDR	((int *)&(((char *)&u)[(UPAGES-1)*NBPG]))

dumpmem(ptr, sz, ustack)
 register int *ptr;
 int sz;
{
	register int i, val;
	extern char *hexstr();

	for (i = 0; i < sz; i++) {
		if ((i & 7) == 0)
			printf("\n%s: ", hexstr((int)ptr, 6));
		else
			printf(" ");
		if (ustack == 1) {
			if ((val = fuword(ptr++)) == -1)
				break;
		} else {
			if (ustack == 0 && (ptr < KSADDR || ptr > KSADDR+(NBPG/4-1)))
				break;
			val = *ptr++;
		}
		printf("%s", hexstr(val, 8));
	}
	printf("\n");
}

char *
hexstr(val, len)
	register int val;
{
	static char nbuf[9];
	register int x, i;

	if (len > 8)
		return("");
	nbuf[len] = '\0';
	for (i = len-1; i >= 0; --i) {
		x = val & 0xF;
		if (x > 9)
			nbuf[i] = x - 10 + 'A';
		else
			nbuf[i] = x + '0';
		val >>= 4;
	}
	return(nbuf);
}