[unix-history] / usr / src / sys / kern / kern_physio.c

/*	kern_physio.c	4.30	82/05/22	*/

#include "../h/param.h"
#include "../h/systm.h"
#include "../h/dir.h"
#include "../h/user.h"
#include "../h/buf.h"
#include "../h/conf.h"
#include "../h/proc.h"
#include "../h/seg.h"
#include "../h/pte.h"
#include "../h/vm.h"
#include "../h/trace.h"

/*
 * Swap IO headers -
 * They contain the necessary information for the swap I/O.
 * At any given time, a swap header can be in three
 * different lists. When free it is in the free list, 
 * when allocated and the I/O queued, it is on the swap 
 * device list, and finally, if the operation was a dirty 
 * page push, when the I/O completes, it is inserted 
 * in a list of cleaned pages to be processed by the pageout daemon.
 */
struct	buf *swbuf;
short	*swsize;		/* CAN WE JUST USE B_BCOUNT? */
int	*swpf;

/*
 * swap I/O -
 *
 * If the flag indicates a dirty page push initiated
 * by the pageout daemon, we map the page into the i th
 * virtual page of process 2 (the daemon itself) where i is
 * the index of the swap header that has been allocated.
 * We simply initialize the header and queue the I/O but
 * do not wait for completion. When the I/O completes,
 * iodone() will link the header to a list of cleaned
 * pages to be processed by the pageout daemon.
 */
swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
	struct proc *p;
	swblk_t dblkno;
	caddr_t addr;
	int flag, nbytes;
	dev_t dev;
	unsigned pfcent;
{
	register struct buf *bp;
	register int c;
	int p2dp;
	register struct pte *dpte, *vpte;
	int s;

	s = spl6();
	while (bswlist.av_forw == NULL) {
		bswlist.b_flags |= B_WANTED;
		sleep((caddr_t)&bswlist, PSWP+1);
	}
	bp = bswlist.av_forw;
	bswlist.av_forw = bp->av_forw;
	splx(s);

	bp->b_flags = B_BUSY | B_PHYS | rdflg | flag;
	if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0)
		if (rdflg == B_READ)
			sum.v_pswpin += btoc(nbytes);
		else
			sum.v_pswpout += btoc(nbytes);
	bp->b_proc = p;
	if (flag & B_DIRTY) {
		p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
		dpte = dptopte(&proc[2], p2dp);
		vpte = vtopte(p, btop(addr));
		for (c = 0; c < nbytes; c += NBPG) {
			if (vpte->pg_pfnum == 0 || vpte->pg_fod)
				panic("swap bad pte");
			*dpte++ = *vpte++;
		}
		bp->b_un.b_addr = (caddr_t)ctob(p2dp);
	} else
		bp->b_un.b_addr = addr;
	while (nbytes > 0) {
		c = imin(ctob(120), nbytes);
		bp->b_bcount = c;
		bp->b_blkno = dblkno;
		bp->b_dev = dev;
		if (flag & B_DIRTY) {
			swpf[bp - swbuf] = pfcent;
			swsize[bp - swbuf] = nbytes;
		}
#ifdef TRACE
		trace(TR_SWAPIO, dev, bp->b_blkno);
#endif
		(*bdevsw[major(dev)].d_strategy)(bp);
		if (flag & B_DIRTY) {
			if (c < nbytes)
				panic("big push");
			return;
		}
		s = spl6();
		while((bp->b_flags&B_DONE)==0)
			sleep((caddr_t)bp, PSWP);
		splx(s);
		bp->b_un.b_addr += c;
		bp->b_flags &= ~B_DONE;
		if (bp->b_flags & B_ERROR) {
			if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE)
				panic("hard IO err in swap");
			swkill(p, (char *)0);
		}
		nbytes -= c;
		dblkno += btoc(c);
	}
	s = spl6();
	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY);
	bp->av_forw = bswlist.av_forw;
	bswlist.av_forw = bp;
	if (bswlist.b_flags & B_WANTED) {
		bswlist.b_flags &= ~B_WANTED;
		wakeup((caddr_t)&bswlist);
		wakeup((caddr_t)&proc[2]);
	}
	splx(s);
}

/*
 * If rout == 0 then killed on swap error, else
 * rout is the name of the routine where we ran out of
 * swap space.
 */
swkill(p, rout)
	struct proc *p;
	char *rout;
{
	char *mesg;

	printf("pid %d: ", p->p_pid);
	if (rout)
		printf(mesg = "killed due to no swap space\n");
	else
		printf(mesg = "killed on swap error\n");
	uprintf("sorry, pid %d was %s", p->p_pid, mesg);
	/*
	 * To be sure no looping (e.g. in vmsched trying to
	 * swap out) mark process locked in core (as though
	 * done by user) after killing it so noone will try
	 * to swap it out.
	 */
	psignal(p, SIGKILL);
	p->p_flag |= SULOCK;
}

/*
 * Raw I/O. The arguments are
 *	The strategy routine for the device
 *	A buffer, which will always be a special buffer
 *	  header owned exclusively by the device for this purpose
 *	The device number
 *	Read/write flag
 * Essentially all the work is computing physical addresses and
 * validating them.
 * If the user has the proper access privilidges, the process is
 * marked 'delayed unlock' and the pages involved in the I/O are
 * faulted and locked. After the completion of the I/O, the above pages
 * are unlocked.
 */
physio(strat, bp, dev, rw, mincnt)
int (*strat)(); 
register struct buf *bp;
unsigned (*mincnt)();
{
	register int c;
	char *a;
	int s;

	if (useracc(u.u_base,u.u_count,rw==B_READ?B_WRITE:B_READ) == NULL) {
		u.u_error = EFAULT;
		return;
	}
	s = spl6();
	while (bp->b_flags&B_BUSY) {
		bp->b_flags |= B_WANTED;
		sleep((caddr_t)bp, PRIBIO+1);
	}
	splx(s);
	bp->b_error = 0;
	bp->b_proc = u.u_procp;
	bp->b_un.b_addr = u.u_base;
	while (u.u_count != 0) {
		bp->b_flags = B_BUSY | B_PHYS | rw;
		bp->b_dev = dev;
		bp->b_blkno = u.u_offset >> PGSHIFT;
		bp->b_bcount = u.u_count;
		(*mincnt)(bp);
		c = bp->b_bcount;
		u.u_procp->p_flag |= SPHYSIO;
		vslock(a = bp->b_un.b_addr, c);
		(*strat)(bp);
		(void) spl6();
		while ((bp->b_flags&B_DONE) == 0)
			sleep((caddr_t)bp, PRIBIO);
		vsunlock(a, c, rw);
		u.u_procp->p_flag &= ~SPHYSIO;
		if (bp->b_flags&B_WANTED)
			wakeup((caddr_t)bp);
		splx(s);
		bp->b_un.b_addr += c;
		u.u_count -= c;
		u.u_offset += c;
		if (bp->b_flags&B_ERROR)
			break;
	}
	bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS);
	u.u_count = bp->b_resid;
	geterror(bp);
}

/*ARGSUSED*/
unsigned
minphys(bp)
struct buf *bp;
{

	if (bp->b_bcount > 63 * 1024)
		bp->b_bcount = 63 * 1024;
}
Commit	Line	Data
bd841446	1	/* kern_physio.c 4.30 82/05/22 */
663dbc72 BJ	2
	3	#include "../h/param.h"
	4	#include "../h/systm.h"
	5	#include "../h/dir.h"
	6	#include "../h/user.h"
	7	#include "../h/buf.h"
	8	#include "../h/conf.h"
	9	#include "../h/proc.h"
	10	#include "../h/seg.h"
	11	#include "../h/pte.h"
	12	#include "../h/vm.h"
973ecc4f	13	#include "../h/trace.h"
663dbc72	14
663dbc72 BJ	15	/*
	16	* Swap IO headers -
	17	* They contain the necessary information for the swap I/O.
	18	* At any given time, a swap header can be in three
	19	* different lists. When free it is in the free list,
	20	* when allocated and the I/O queued, it is on the swap
	21	* device list, and finally, if the operation was a dirty
	22	* page push, when the I/O completes, it is inserted
	23	* in a list of cleaned pages to be processed by the pageout daemon.
	24	*/
4c05b581 BJ	25	struct buf *swbuf;
	26	short swsize; / CAN WE JUST USE B_BCOUNT? */
	27	int *swpf;
663dbc72	28
663dbc72 BJ	29	/*
	30	* swap I/O -
	31	*
	32	* If the flag indicates a dirty page push initiated
	33	* by the pageout daemon, we map the page into the i th
	34	* virtual page of process 2 (the daemon itself) where i is
	35	* the index of the swap header that has been allocated.
	36	* We simply initialize the header and queue the I/O but
	37	* do not wait for completion. When the I/O completes,
	38	* iodone() will link the header to a list of cleaned
	39	* pages to be processed by the pageout daemon.
	40	*/
	41	swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent)
	42	struct proc *p;
	43	swblk_t dblkno;
	44	caddr_t addr;
	45	int flag, nbytes;
	46	dev_t dev;
	47	unsigned pfcent;
	48	{
	49	register struct buf *bp;
	50	register int c;
	51	int p2dp;
	52	register struct pte dpte, vpte;
530d0032	53	int s;
663dbc72	54
530d0032	55	s = spl6();
663dbc72 BJ	56	while (bswlist.av_forw == NULL) {
	57	bswlist.b_flags \|= B_WANTED;
	58	sleep((caddr_t)&bswlist, PSWP+1);
	59	}
	60	bp = bswlist.av_forw;
	61	bswlist.av_forw = bp->av_forw;
530d0032	62	splx(s);
663dbc72 BJ	63
	64	bp->b_flags = B_BUSY \| B_PHYS \| rdflg \| flag;
	65	if ((bp->b_flags & (B_DIRTY\|B_PGIN)) == 0)
	66	if (rdflg == B_READ)
	67	sum.v_pswpin += btoc(nbytes);
	68	else
	69	sum.v_pswpout += btoc(nbytes);
	70	bp->b_proc = p;
	71	if (flag & B_DIRTY) {
	72	p2dp = ((bp - swbuf) * CLSIZE) * KLMAX;
	73	dpte = dptopte(&proc[2], p2dp);
	74	vpte = vtopte(p, btop(addr));
	75	for (c = 0; c < nbytes; c += NBPG) {
	76	if (vpte->pg_pfnum == 0 \|\| vpte->pg_fod)
	77	panic("swap bad pte");
	78	dpte++ = vpte++;
	79	}
	80	bp->b_un.b_addr = (caddr_t)ctob(p2dp);
	81	} else
	82	bp->b_un.b_addr = addr;
	83	while (nbytes > 0) {
	84	c = imin(ctob(120), nbytes);
	85	bp->b_bcount = c;
	86	bp->b_blkno = dblkno;
	87	bp->b_dev = dev;
d2f87136 BJ	88	if (flag & B_DIRTY) {
	89	swpf[bp - swbuf] = pfcent;
	90	swsize[bp - swbuf] = nbytes;
	91	}
53f9ca20 BJ	92	#ifdef TRACE
	93	trace(TR_SWAPIO, dev, bp->b_blkno);
	94	#endif
663dbc72 BJ	95	(*bdevsw[major(dev)].d_strategy)(bp);
	96	if (flag & B_DIRTY) {
	97	if (c < nbytes)
	98	panic("big push");
663dbc72 BJ	99	return;
663dbc72 BJ	100	}
530d0032	101	s = spl6();
663dbc72 BJ	102	while((bp->b_flags&B_DONE)==0)
663dbc72 BJ	103	sleep((caddr_t)bp, PSWP);
530d0032	104	splx(s);
663dbc72 BJ	105	bp->b_un.b_addr += c;
	106	bp->b_flags &= ~B_DONE;
	107	if (bp->b_flags & B_ERROR) {
	108	if ((flag & (B_UAREA\|B_PAGET)) \|\| rdflg == B_WRITE)
	109	panic("hard IO err in swap");
	110	swkill(p, (char *)0);
	111	}
	112	nbytes -= c;
	113	dblkno += btoc(c);
	114	}
530d0032	115	s = spl6();
663dbc72 BJ	116	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS\|B_PAGET\|B_UAREA\|B_DIRTY);
	117	bp->av_forw = bswlist.av_forw;
	118	bswlist.av_forw = bp;
	119	if (bswlist.b_flags & B_WANTED) {
	120	bswlist.b_flags &= ~B_WANTED;
	121	wakeup((caddr_t)&bswlist);
	122	wakeup((caddr_t)&proc[2]);
	123	}
530d0032	124	splx(s);
663dbc72 BJ	125	}
	126
	127	/*
	128	* If rout == 0 then killed on swap error, else
	129	* rout is the name of the routine where we ran out of
	130	* swap space.
	131	*/
	132	swkill(p, rout)
	133	struct proc *p;
	134	char *rout;
	135	{
444f631c	136	char *mesg;
663dbc72	137
444f631c	138	printf("pid %d: ", p->p_pid);
663dbc72	139	if (rout)
444f631c	140	printf(mesg = "killed due to no swap space\n");
663dbc72	141	else
444f631c BJ	142	printf(mesg = "killed on swap error\n");
444f631c BJ	143	uprintf("sorry, pid %d was %s", p->p_pid, mesg);
663dbc72 BJ	144	/*
	145	* To be sure no looping (e.g. in vmsched trying to
	146	* swap out) mark process locked in core (as though
	147	* done by user) after killing it so noone will try
	148	* to swap it out.
	149	*/
a30d2e97	150	psignal(p, SIGKILL);
663dbc72 BJ	151	p->p_flag \|= SULOCK;
	152	}
	153
663dbc72 BJ	154	/*
	155	* Raw I/O. The arguments are
	156	* The strategy routine for the device
	157	* A buffer, which will always be a special buffer
	158	* header owned exclusively by the device for this purpose
	159	* The device number
	160	* Read/write flag
	161	* Essentially all the work is computing physical addresses and
	162	* validating them.
	163	* If the user has the proper access privilidges, the process is
	164	* marked 'delayed unlock' and the pages involved in the I/O are
	165	* faulted and locked. After the completion of the I/O, the above pages
	166	* are unlocked.
	167	*/
	168	physio(strat, bp, dev, rw, mincnt)
	169	int (*strat)();
	170	register struct buf *bp;
	171	unsigned (*mincnt)();
	172	{
	173	register int c;
	174	char *a;
530d0032	175	int s;
663dbc72 BJ	176
	177	if (useracc(u.u_base,u.u_count,rw==B_READ?B_WRITE:B_READ) == NULL) {
	178	u.u_error = EFAULT;
	179	return;
	180	}
530d0032	181	s = spl6();
663dbc72 BJ	182	while (bp->b_flags&B_BUSY) {
	183	bp->b_flags \|= B_WANTED;
	184	sleep((caddr_t)bp, PRIBIO+1);
	185	}
ef3b3d5a	186	splx(s);
663dbc72 BJ	187	bp->b_error = 0;
	188	bp->b_proc = u.u_procp;
	189	bp->b_un.b_addr = u.u_base;
52a593fa	190	while (u.u_count != 0) {
663dbc72 BJ	191	bp->b_flags = B_BUSY \| B_PHYS \| rw;
	192	bp->b_dev = dev;
	193	bp->b_blkno = u.u_offset >> PGSHIFT;
	194	bp->b_bcount = u.u_count;
	195	(*mincnt)(bp);
	196	c = bp->b_bcount;
	197	u.u_procp->p_flag \|= SPHYSIO;
	198	vslock(a = bp->b_un.b_addr, c);
	199	(*strat)(bp);
81263dba	200	(void) spl6();
663dbc72 BJ	201	while ((bp->b_flags&B_DONE) == 0)
	202	sleep((caddr_t)bp, PRIBIO);
	203	vsunlock(a, c, rw);
	204	u.u_procp->p_flag &= ~SPHYSIO;
	205	if (bp->b_flags&B_WANTED)
	206	wakeup((caddr_t)bp);
530d0032	207	splx(s);
663dbc72 BJ	208	bp->b_un.b_addr += c;
	209	u.u_count -= c;
	210	u.u_offset += c;
52a593fa BJ	211	if (bp->b_flags&B_ERROR)
52a593fa BJ	212	break;
663dbc72 BJ	213	}
	214	bp->b_flags &= ~(B_BUSY\|B_WANTED\|B_PHYS);
	215	u.u_count = bp->b_resid;
	216	geterror(bp);
	217	}
	218
	219	/ARGSUSED/
	220	unsigned
	221	minphys(bp)
	222	struct buf *bp;
	223	{
	224
2ec65c94 BJ	225	if (bp->b_bcount > 63 * 1024)
2ec65c94 BJ	226	bp->b_bcount = 63 * 1024;
663dbc72 BJ	227	}
663dbc72 BJ	228