Commit | Line | Data |
---|---|---|
961945a8 SL |
1 | /* kern_physio.c 4.37 82/12/17 */ |
2 | ||
3 | #include "../machine/pte.h" | |
663dbc72 BJ |
4 | |
5 | #include "../h/param.h" | |
6 | #include "../h/systm.h" | |
7 | #include "../h/dir.h" | |
8 | #include "../h/user.h" | |
9 | #include "../h/buf.h" | |
10 | #include "../h/conf.h" | |
11 | #include "../h/proc.h" | |
12 | #include "../h/seg.h" | |
663dbc72 | 13 | #include "../h/vm.h" |
973ecc4f | 14 | #include "../h/trace.h" |
d6d7360b | 15 | #include "../h/uio.h" |
663dbc72 | 16 | |
663dbc72 BJ |
17 | /* |
18 | * Swap IO headers - | |
19 | * They contain the necessary information for the swap I/O. | |
20 | * At any given time, a swap header can be in three | |
21 | * different lists. When free it is in the free list, | |
22 | * when allocated and the I/O queued, it is on the swap | |
23 | * device list, and finally, if the operation was a dirty | |
24 | * page push, when the I/O completes, it is inserted | |
25 | * in a list of cleaned pages to be processed by the pageout daemon. | |
26 | */ | |
4c05b581 BJ |
27 | struct buf *swbuf; |
28 | short *swsize; /* CAN WE JUST USE B_BCOUNT? */ | |
29 | int *swpf; | |
663dbc72 | 30 | |
663dbc72 BJ |
31 | /* |
32 | * swap I/O - | |
33 | * | |
34 | * If the flag indicates a dirty page push initiated | |
35 | * by the pageout daemon, we map the page into the i th | |
36 | * virtual page of process 2 (the daemon itself) where i is | |
37 | * the index of the swap header that has been allocated. | |
38 | * We simply initialize the header and queue the I/O but | |
39 | * do not wait for completion. When the I/O completes, | |
40 | * iodone() will link the header to a list of cleaned | |
41 | * pages to be processed by the pageout daemon. | |
42 | */ | |
43 | swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent) | |
44 | struct proc *p; | |
45 | swblk_t dblkno; | |
46 | caddr_t addr; | |
39d536e6 | 47 | int nbytes, rdflg, flag; |
663dbc72 | 48 | dev_t dev; |
39d536e6 | 49 | u_int pfcent; |
663dbc72 BJ |
50 | { |
51 | register struct buf *bp; | |
e438ed8e | 52 | register u_int c; |
663dbc72 BJ |
53 | int p2dp; |
54 | register struct pte *dpte, *vpte; | |
530d0032 | 55 | int s; |
663dbc72 | 56 | |
530d0032 | 57 | s = spl6(); |
663dbc72 BJ |
58 | while (bswlist.av_forw == NULL) { |
59 | bswlist.b_flags |= B_WANTED; | |
60 | sleep((caddr_t)&bswlist, PSWP+1); | |
61 | } | |
62 | bp = bswlist.av_forw; | |
63 | bswlist.av_forw = bp->av_forw; | |
530d0032 | 64 | splx(s); |
663dbc72 BJ |
65 | |
66 | bp->b_flags = B_BUSY | B_PHYS | rdflg | flag; | |
67 | if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0) | |
68 | if (rdflg == B_READ) | |
69 | sum.v_pswpin += btoc(nbytes); | |
70 | else | |
71 | sum.v_pswpout += btoc(nbytes); | |
72 | bp->b_proc = p; | |
73 | if (flag & B_DIRTY) { | |
74 | p2dp = ((bp - swbuf) * CLSIZE) * KLMAX; | |
75 | dpte = dptopte(&proc[2], p2dp); | |
76 | vpte = vtopte(p, btop(addr)); | |
77 | for (c = 0; c < nbytes; c += NBPG) { | |
78 | if (vpte->pg_pfnum == 0 || vpte->pg_fod) | |
79 | panic("swap bad pte"); | |
80 | *dpte++ = *vpte++; | |
81 | } | |
82 | bp->b_un.b_addr = (caddr_t)ctob(p2dp); | |
83 | } else | |
84 | bp->b_un.b_addr = addr; | |
85 | while (nbytes > 0) { | |
e438ed8e BJ |
86 | bp->b_bcount = nbytes; |
87 | minphys(bp); | |
88 | c = bp->b_bcount; | |
663dbc72 BJ |
89 | bp->b_blkno = dblkno; |
90 | bp->b_dev = dev; | |
d2f87136 BJ |
91 | if (flag & B_DIRTY) { |
92 | swpf[bp - swbuf] = pfcent; | |
93 | swsize[bp - swbuf] = nbytes; | |
94 | } | |
53f9ca20 BJ |
95 | #ifdef TRACE |
96 | trace(TR_SWAPIO, dev, bp->b_blkno); | |
97 | #endif | |
ca1f746a | 98 | physstrat(bp, bdevsw[major(dev)].d_strategy, PSWP); |
663dbc72 BJ |
99 | if (flag & B_DIRTY) { |
100 | if (c < nbytes) | |
101 | panic("big push"); | |
663dbc72 BJ |
102 | return; |
103 | } | |
663dbc72 BJ |
104 | bp->b_un.b_addr += c; |
105 | bp->b_flags &= ~B_DONE; | |
106 | if (bp->b_flags & B_ERROR) { | |
107 | if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE) | |
108 | panic("hard IO err in swap"); | |
109 | swkill(p, (char *)0); | |
110 | } | |
111 | nbytes -= c; | |
e438ed8e | 112 | dblkno += c / DEV_BSIZE; |
663dbc72 | 113 | } |
530d0032 | 114 | s = spl6(); |
663dbc72 BJ |
115 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); |
116 | bp->av_forw = bswlist.av_forw; | |
117 | bswlist.av_forw = bp; | |
118 | if (bswlist.b_flags & B_WANTED) { | |
119 | bswlist.b_flags &= ~B_WANTED; | |
120 | wakeup((caddr_t)&bswlist); | |
121 | wakeup((caddr_t)&proc[2]); | |
122 | } | |
530d0032 | 123 | splx(s); |
663dbc72 BJ |
124 | } |
125 | ||
126 | /* | |
127 | * If rout == 0 then killed on swap error, else | |
128 | * rout is the name of the routine where we ran out of | |
129 | * swap space. | |
130 | */ | |
131 | swkill(p, rout) | |
132 | struct proc *p; | |
133 | char *rout; | |
134 | { | |
444f631c | 135 | char *mesg; |
663dbc72 | 136 | |
444f631c | 137 | printf("pid %d: ", p->p_pid); |
663dbc72 | 138 | if (rout) |
444f631c | 139 | printf(mesg = "killed due to no swap space\n"); |
663dbc72 | 140 | else |
444f631c BJ |
141 | printf(mesg = "killed on swap error\n"); |
142 | uprintf("sorry, pid %d was %s", p->p_pid, mesg); | |
663dbc72 BJ |
143 | /* |
144 | * To be sure no looping (e.g. in vmsched trying to | |
145 | * swap out) mark process locked in core (as though | |
146 | * done by user) after killing it so noone will try | |
147 | * to swap it out. | |
148 | */ | |
a30d2e97 | 149 | psignal(p, SIGKILL); |
663dbc72 BJ |
150 | p->p_flag |= SULOCK; |
151 | } | |
152 | ||
663dbc72 BJ |
153 | /* |
154 | * Raw I/O. The arguments are | |
155 | * The strategy routine for the device | |
156 | * A buffer, which will always be a special buffer | |
157 | * header owned exclusively by the device for this purpose | |
158 | * The device number | |
159 | * Read/write flag | |
160 | * Essentially all the work is computing physical addresses and | |
161 | * validating them. | |
162 | * If the user has the proper access privilidges, the process is | |
163 | * marked 'delayed unlock' and the pages involved in the I/O are | |
164 | * faulted and locked. After the completion of the I/O, the above pages | |
165 | * are unlocked. | |
166 | */ | |
d6d7360b BJ |
167 | physio(strat, bp, dev, rw, mincnt, uio) |
168 | int (*strat)(); | |
169 | register struct buf *bp; | |
170 | dev_t dev; | |
171 | int rw; | |
172 | unsigned (*mincnt)(); | |
173 | struct uio *uio; | |
663dbc72 | 174 | { |
406ddcbe | 175 | register struct iovec *iov = uio->uio_iov; |
663dbc72 BJ |
176 | register int c; |
177 | char *a; | |
d6d7360b | 178 | int s, error = 0; |
663dbc72 | 179 | |
d6d7360b | 180 | nextiov: |
406ddcbe | 181 | if (uio->uio_iovcnt == 0) |
d6d7360b | 182 | return (0); |
406ddcbe | 183 | if (useracc(iov->iov_base,(u_int)iov->iov_len,rw==B_READ?B_WRITE:B_READ) == NULL) |
d6d7360b | 184 | return (EFAULT); |
530d0032 | 185 | s = spl6(); |
663dbc72 BJ |
186 | while (bp->b_flags&B_BUSY) { |
187 | bp->b_flags |= B_WANTED; | |
188 | sleep((caddr_t)bp, PRIBIO+1); | |
189 | } | |
ef3b3d5a | 190 | splx(s); |
663dbc72 BJ |
191 | bp->b_error = 0; |
192 | bp->b_proc = u.u_procp; | |
d6d7360b BJ |
193 | bp->b_un.b_addr = iov->iov_base; |
194 | while (iov->iov_len > 0) { | |
663dbc72 BJ |
195 | bp->b_flags = B_BUSY | B_PHYS | rw; |
196 | bp->b_dev = dev; | |
e438ed8e | 197 | bp->b_blkno = uio->uio_offset / DEV_BSIZE; |
d6d7360b | 198 | bp->b_bcount = iov->iov_len; |
663dbc72 BJ |
199 | (*mincnt)(bp); |
200 | c = bp->b_bcount; | |
201 | u.u_procp->p_flag |= SPHYSIO; | |
202 | vslock(a = bp->b_un.b_addr, c); | |
e438ed8e | 203 | physstrat(bp, strat, PRIBIO); |
81263dba | 204 | (void) spl6(); |
663dbc72 BJ |
205 | vsunlock(a, c, rw); |
206 | u.u_procp->p_flag &= ~SPHYSIO; | |
207 | if (bp->b_flags&B_WANTED) | |
208 | wakeup((caddr_t)bp); | |
530d0032 | 209 | splx(s); |
d6d7360b | 210 | c -= bp->b_resid; |
663dbc72 | 211 | bp->b_un.b_addr += c; |
d6d7360b BJ |
212 | iov->iov_len -= c; |
213 | uio->uio_resid -= c; | |
214 | uio->uio_offset += c; | |
961945a8 SL |
215 | /* temp kludge for tape drives */ |
216 | if (bp->b_resid || bp->b_flags&B_ERROR) | |
52a593fa | 217 | break; |
663dbc72 BJ |
218 | } |
219 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS); | |
d6d7360b | 220 | error = geterror(bp); |
961945a8 SL |
221 | /* temp kludge for tape drives */ |
222 | if (bp->b_resid || error) | |
d6d7360b | 223 | return (error); |
d6d7360b BJ |
224 | uio->uio_iov++; |
225 | uio->uio_iovcnt--; | |
226 | goto nextiov; | |
663dbc72 BJ |
227 | } |
228 | ||
663dbc72 BJ |
229 | unsigned |
230 | minphys(bp) | |
d6d7360b | 231 | struct buf *bp; |
663dbc72 BJ |
232 | { |
233 | ||
2ec65c94 BJ |
234 | if (bp->b_bcount > 63 * 1024) |
235 | bp->b_bcount = 63 * 1024; | |
663dbc72 BJ |
236 | } |
237 |