2 * Copyright (c) 1993 The Regents of the University of California.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 #include <machine/asmacros.h>
40 #include <machine/cputypes.h>
41 #include <machine/pmap.h>
42 #include <machine/specialreg.h>
58 .globl _copyout_vector
60 .long _generic_copyout
61 .globl _ovbcopy_vector
64 #if defined(I586_CPU) && NNPX > 0
74 * void bzero(void *buf, u_int len)
99 * do 64 byte chunks first
101 * XXX this is probably over-unrolled at least for DX2's
158 * a jump table seems to be faster than a loop or more range reductions
160 * XXX need a const section for non-text
195 #if defined(I586_CPU) && NNPX > 0
201 * The FPU register method is twice as fast as the integer register
202 * method unless the target is in the L1 cache and we pre-allocate a
203 * cache line for it (then the integer register method is 4-5 times
204 * faster). However, we never pre-allocate cache lines, since that
205 * would make the integer method 25% or more slower for the common
206 * case when the target isn't in either the L1 cache or the L2 cache.
207 * Thus we normally use the FPU register method unless the overhead
208 * would be too large.
210 cmpl $256,%ecx /* empirical; clts, fninit, smsw cost a lot */
214 * The FPU registers may belong to an application or to fastmove()
215 * or to another invocation of bcopy() or ourself in a higher level
216 * interrupt or trap handler. Preserving the registers is
217 * complicated since we avoid it if possible at all levels. We
218 * want to localize the complications even when that increases them.
219 * Here the extra work involves preserving CR0_TS in TS.
220 * `npxproc != NULL' is supposed to be the condition that all the
221 * FPU resources belong to an application, but npxproc and CR0_TS
222 * aren't set atomically enough for this condition to work in
223 * interrupt handlers.
225 * Case 1: FPU registers belong to the application: we must preserve
226 * the registers if we use them, so we only use the FPU register
227 * method if the target size is large enough to amortize the extra
228 * overhead for preserving them. CR0_TS must be preserved although
229 * it is very likely to end up as set.
231 * Case 2: FPU registers belong to fastmove(): fastmove() currently
232 * makes the registers look like they belong to an application so
233 * that cpu_switch() and savectx() don't have to know about it, so
234 * this case reduces to case 1.
236 * Case 3: FPU registers belong to the kernel: don't use the FPU
237 * register method. This case is unlikely, and supporting it would
238 * be more complicated and might take too much stack.
240 * Case 4: FPU registers don't belong to anyone: the FPU registers
241 * don't need to be preserved, so we always use the FPU register
242 * method. CR0_TS must be preserved although it is very likely to
243 * always end up as clear.
247 cmpl $256+184,%ecx /* empirical; not quite 2*108 more */
249 sarb $1,kernel_fpu_lock
258 sarb $1,kernel_fpu_lock
262 fninit /* XXX should avoid needing this */
267 * Align to an 8 byte boundary (misalignment in the main loop would
268 * cost a factor of >= 2). Avoid jumps (at little cost if it is
269 * already aligned) by always zeroing 8 bytes and using the part up
270 * to the _next_ alignment position.
273 addl %edx,%ecx /* part of %ecx -= new_%edx - %edx */
279 * Similarly align `len' to a multiple of 8.
286 * This wouldn't be any faster if it were unrolled, since the loop
287 * control instructions are much faster than the fstl and/or done
288 * in parallel with it so their overhead is insignificant.
290 fpureg_i586_bzero_loop:
295 jae fpureg_i586_bzero_loop
302 movb $0xfe,kernel_fpu_lock
308 movb $0xfe,kernel_fpu_lock
313 * `rep stos' seems to be the best method in practice for small
314 * counts. Fancy methods usually take too long to start up due
315 * to cache and BTB misses.
335 #endif /* I586_CPU && NNPX > 0 */
387 /* fillw(pat, base, cnt) */
407 cmpl %ecx,%eax /* overlapping && src < dst? */
409 cld /* nope, copy forwards */
418 addl %ecx,%edi /* copy backwards. */
439 * generic_bcopy(src, dst, cnt)
440 * ws@tools.de (Wolfgang Solfrank, TooLs GmbH) +49-228-985800
451 cmpl %ecx,%eax /* overlapping && src < dst? */
454 shrl $2,%ecx /* copy by 32-bit words */
455 cld /* nope, copy forwards */
459 andl $3,%ecx /* any bytes left? */
468 addl %ecx,%edi /* copy backwards */
472 andl $3,%ecx /* any fractional bytes? */
476 movl 20(%esp),%ecx /* copy remainder by 32-bit words */
487 #if defined(I586_CPU) && NNPX > 0
497 cmpl %ecx,%eax /* overlapping && src < dst? */
503 sarb $1,kernel_fpu_lock
516 fninit /* XXX should avoid needing this */
521 #define DCACHE_SIZE 8192
522 cmpl $(DCACHE_SIZE-512)/2,%ecx
524 movl $(DCACHE_SIZE-512)/2,%ecx
528 jb 5f /* XXX should prefetch if %ecx >= 32 */
549 large_i586_bcopy_loop:
570 jae large_i586_bcopy_loop
582 movb $0xfe,kernel_fpu_lock
585 * This is a duplicate of the main part of generic_bcopy. See the comments
586 * there. Jumping into generic_bcopy would cost a whole 0-1 cycles and
587 * would mess up high resolution profiling.
623 #endif /* I586_CPU && NNPX > 0 */
626 * Note: memcpy does not support overlapping copies
635 shrl $2,%ecx /* copy by 32-bit words */
636 cld /* nope, copy forwards */
640 andl $3,%ecx /* any bytes left? */
648 /*****************************************************************************/
649 /* copyout and fubyte family */
650 /*****************************************************************************/
652 * Access user memory from inside the kernel. These routines and possibly
653 * the math- and DOS emulators should be the only places that do this.
655 * We have to access the memory with user's permissions, so use a segment
656 * selector with RPL 3. For writes to user space we have to additionally
657 * check the PTE for write permission, because the 386 does not check
658 * write permissions when we are executing with EPL 0. The 486 does check
659 * this if the WP bit is set in CR0, so we can use a simpler version here.
661 * These routines set curpcb->onfault for the time they execute. When a
662 * protection violation occurs inside the functions, the trap handler
663 * returns to *curpcb->onfault instead of the function.
666 /* copyout(from_kernel, to_user, len) */
671 ENTRY(generic_copyout)
673 movl $copyout_fault,PCB_ONFAULT(%eax)
680 testl %ebx,%ebx /* anything to do? */
684 * Check explicitly for non-user addresses. If 486 write protection
685 * is being used, this check is essential because we are in kernel
686 * mode so the h/w does not provide any protection against writing
691 * First, prevent address wrapping.
697 * XXX STOP USING VM_MAXUSER_ADDRESS.
698 * It is an end address, not a max, so every time it is used correctly it
699 * looks like there is an off by one error, and of course it caused an off
700 * by one error in several places.
702 cmpl $VM_MAXUSER_ADDRESS,%eax
705 #if defined(I386_CPU)
707 #if defined(I486_CPU) || defined(I586_CPU) || defined(I686_CPU)
708 cmpl $CPUCLASS_386,_cpu_class
712 * We have to check each PTE for user write permission.
713 * The checking may cause a page fault, so it is important to set
714 * up everything for return via copyout_fault before here.
716 /* compute number of pages */
721 shrl $IDXSHIFT+2,%ecx
724 /* compute PTE offset for start address */
730 /* check PTE for each page */
731 leal _PTmap(%edx),%eax
734 testb $PG_V,_PTmap(%eax) /* PTE page must be valid */
736 movb _PTmap(%edx),%al
737 andb $PG_V|PG_RW|PG_U,%al /* page must be valid and user writable */
738 cmpb $PG_V|PG_RW|PG_U,%al
742 /* simulate a trap */
747 call _trapwrite /* trapwrite(addr) */
752 testl %eax,%eax /* if not ok, return EFAULT */
758 jnz 1b /* check next page */
759 #endif /* I386_CPU */
761 /* bcopy(%esi, %edi, %ebx) */
765 #if defined(I586_CPU) && NNPX > 0
784 movl %eax,PCB_ONFAULT(%edx)
793 movl $0,PCB_ONFAULT(%edx)
797 #if defined(I586_CPU) && NNPX > 0
800 * Duplicated from generic_copyout. Could be done a bit better.
803 movl $copyout_fault,PCB_ONFAULT(%eax)
810 testl %ebx,%ebx /* anything to do? */
814 * Check explicitly for non-user addresses. If 486 write protection
815 * is being used, this check is essential because we are in kernel
816 * mode so the h/w does not provide any protection against writing
821 * First, prevent address wrapping.
827 * XXX STOP USING VM_MAXUSER_ADDRESS.
828 * It is an end address, not a max, so every time it is used correctly it
829 * looks like there is an off by one error, and of course it caused an off
830 * by one error in several places.
832 cmpl $VM_MAXUSER_ADDRESS,%eax
835 /* bcopy(%esi, %edi, %ebx) */
839 * End of duplicated code.
849 #endif /* I586_CPU && NNPX > 0 */
851 /* copyin(from_user, to_kernel, len) */
856 ENTRY(generic_copyin)
858 movl $copyin_fault,PCB_ONFAULT(%eax)
861 movl 12(%esp),%esi /* caddr_t from */
862 movl 16(%esp),%edi /* caddr_t to */
863 movl 20(%esp),%ecx /* size_t len */
866 * make sure address is valid
871 cmpl $VM_MAXUSER_ADDRESS,%edx
874 #if defined(I586_CPU) && NNPX > 0
879 shrl $2,%ecx /* copy longword-wise */
884 andb $3,%cl /* copy remaining bytes */
888 #if defined(I586_CPU) && NNPX > 0
896 movl %eax,PCB_ONFAULT(%edx)
904 movl $0,PCB_ONFAULT(%edx)
908 #if defined(I586_CPU) && NNPX > 0
911 * Duplicated from generic_copyin. Could be done a bit better.
914 movl $copyin_fault,PCB_ONFAULT(%eax)
917 movl 12(%esp),%esi /* caddr_t from */
918 movl 16(%esp),%edi /* caddr_t to */
919 movl 20(%esp),%ecx /* size_t len */
922 * make sure address is valid
927 cmpl $VM_MAXUSER_ADDRESS,%edx
930 * End of duplicated code.
936 pushl %ebx /* XXX prepare for fastmove_fault */
941 #endif /* I586_CPU && NNPX > 0 */
943 #if defined(I586_CPU) && NNPX > 0
944 /* fastmove(src, dst, len)
947 len in %ecx XXX changed to on stack for profiling
948 uses %eax and %edx for tmp. storage
950 /* XXX use ENTRY() to get profiling. fastmove() is actually a non-entry. */
954 subl $PCB_SAVEFPU_SIZE+3*4,%esp
960 testl $7,%esi /* check if src addr is multiple of 8 */
963 testl $7,%edi /* check if dst addr is multiple of 8 */
966 /* if (npxproc != NULL) { */
969 /* fnsave(&curpcb->pcb_savefpu); */
971 fnsave PCB_SAVEFPU(%eax)
972 /* npxproc = NULL; */
976 /* now we own the FPU. */
979 * The process' FP state is saved in the pcb, but if we get
980 * switched, the cpu_switch() will store our FP state in the
981 * pcb. It should be possible to avoid all the copying for
982 * this, e.g., by setting a flag to tell cpu_switch() to
983 * save the state somewhere else.
985 /* tmp = curpcb->pcb_savefpu; */
991 addl $PCB_SAVEFPU,%esi
993 movl $PCB_SAVEFPU_SIZE>>2,%ecx
999 /* stop_emulating(); */
1001 /* npxproc = curproc; */
1005 movl $fastmove_fault,PCB_ONFAULT(%eax)
1062 /* curpcb->pcb_savefpu = tmp; */
1067 addl $PCB_SAVEFPU,%edi
1070 movl $PCB_SAVEFPU_SIZE>>2,%ecx
1077 /* start_emulating(); */
1081 /* npxproc = NULL; */
1087 movl $fastmove_tail_fault,PCB_ONFAULT(%eax)
1090 shrl $2,%ecx /* copy longword-wise */
1095 andb $3,%cl /* copy remaining bytes */
1106 addl $PCB_SAVEFPU,%edi
1109 movl $PCB_SAVEFPU_SIZE>>2,%ecx
1118 fastmove_tail_fault:
1126 movl $0,PCB_ONFAULT(%edx)
1129 #endif /* I586_CPU && NNPX > 0 */
1132 * fu{byte,sword,word} : fetch a byte (sword, word) from user memory
1136 movl $fusufault,PCB_ONFAULT(%ecx)
1137 movl 4(%esp),%edx /* from */
1139 cmpl $VM_MAXUSER_ADDRESS-4,%edx /* verify address is valid */
1143 movl $0,PCB_ONFAULT(%ecx)
1147 * These two routines are called from the profiling code, potentially
1148 * at interrupt time. If they fail, that's okay, good things will
1149 * happen later. Fail all the time for now - until the trap code is
1150 * able to deal with this.
1159 movl $fusufault,PCB_ONFAULT(%ecx)
1162 cmpl $VM_MAXUSER_ADDRESS-2,%edx
1166 movl $0,PCB_ONFAULT(%ecx)
1171 movl $fusufault,PCB_ONFAULT(%ecx)
1174 cmpl $VM_MAXUSER_ADDRESS-1,%edx
1178 movl $0,PCB_ONFAULT(%ecx)
1185 movl %eax,PCB_ONFAULT(%ecx)
1190 * su{byte,sword,word}: write a byte (word, longword) to user memory
1194 movl $fusufault,PCB_ONFAULT(%ecx)
1197 #if defined(I386_CPU)
1199 #if defined(I486_CPU) || defined(I586_CPU) || defined(I686_CPU)
1200 cmpl $CPUCLASS_386,_cpu_class
1201 jne 2f /* we only have to set the right segment selector */
1202 #endif /* I486_CPU || I586_CPU || I686_CPU */
1204 /* XXX - page boundary crossing is still not handled */
1209 leal _PTmap(%edx),%ecx
1212 testb $PG_V,_PTmap(%ecx) /* PTE page must be valid */
1214 movb _PTmap(%edx),%dl
1215 andb $PG_V|PG_RW|PG_U,%dl /* page must be valid and user writable */
1216 cmpb $PG_V|PG_RW|PG_U,%dl
1220 /* simulate a trap */
1223 popl %edx /* remove junk parameter from stack */
1231 cmpl $VM_MAXUSER_ADDRESS-4,%edx /* verify address validity */
1238 movl %eax,PCB_ONFAULT(%ecx)
1243 movl $fusufault,PCB_ONFAULT(%ecx)
1246 #if defined(I386_CPU)
1248 #if defined(I486_CPU) || defined(I586_CPU) || defined(I686_CPU)
1249 cmpl $CPUCLASS_386,_cpu_class
1251 #endif /* I486_CPU || I586_CPU || I686_CPU */
1253 /* XXX - page boundary crossing is still not handled */
1258 leal _PTmap(%edx),%ecx
1261 testb $PG_V,_PTmap(%ecx) /* PTE page must be valid */
1263 movb _PTmap(%edx),%dl
1264 andb $PG_V|PG_RW|PG_U,%dl /* page must be valid and user writable */
1265 cmpb $PG_V|PG_RW|PG_U,%dl
1269 /* simulate a trap */
1272 popl %edx /* remove junk parameter from stack */
1280 cmpl $VM_MAXUSER_ADDRESS-2,%edx /* verify address validity */
1286 movl _curpcb,%ecx /* restore trashed register */
1287 movl %eax,PCB_ONFAULT(%ecx)
1293 movl $fusufault,PCB_ONFAULT(%ecx)
1296 #if defined(I386_CPU)
1298 #if defined(I486_CPU) || defined(I586_CPU) || defined(I686_CPU)
1299 cmpl $CPUCLASS_386,_cpu_class
1301 #endif /* I486_CPU || I586_CPU || I686_CPU */
1307 leal _PTmap(%edx),%ecx
1310 testb $PG_V,_PTmap(%ecx) /* PTE page must be valid */
1312 movb _PTmap(%edx),%dl
1313 andb $PG_V|PG_RW|PG_U,%dl /* page must be valid and user writable */
1314 cmpb $PG_V|PG_RW|PG_U,%dl
1318 /* simulate a trap */
1321 popl %edx /* remove junk parameter from stack */
1329 cmpl $VM_MAXUSER_ADDRESS-1,%edx /* verify address validity */
1335 movl _curpcb,%ecx /* restore trashed register */
1336 movl %eax,PCB_ONFAULT(%ecx)
1340 * copyinstr(from, to, maxlen, int *lencopied)
1341 * copy a string from from to to, stop when a 0 character is reached.
1342 * return ENAMETOOLONG if string is longer than maxlen, and
1343 * EFAULT on protection violations. If lencopied is non-zero,
1344 * return the actual length in *lencopied.
1350 movl $cpystrflt,PCB_ONFAULT(%ecx)
1352 movl 12(%esp),%esi /* %esi = from */
1353 movl 16(%esp),%edi /* %edi = to */
1354 movl 20(%esp),%edx /* %edx = maxlen */
1356 movl $VM_MAXUSER_ADDRESS,%eax
1358 /* make sure 'from' is within bounds */
1362 /* restrict maxlen to <= VM_MAXUSER_ADDRESS-from */
1380 /* Success -- 0 byte reached */
1385 /* edx is zero - return ENAMETOOLONG or EFAULT */
1386 cmpl $VM_MAXUSER_ADDRESS,%esi
1389 movl $ENAMETOOLONG,%eax
1396 /* set *lencopied and return %eax */
1398 movl $0,PCB_ONFAULT(%ecx)
1412 * copystr(from, to, maxlen, int *lencopied)
1418 movl 12(%esp),%esi /* %esi = from */
1419 movl 16(%esp),%edi /* %edi = to */
1420 movl 20(%esp),%edx /* %edx = maxlen */
1431 /* Success -- 0 byte reached */
1436 /* edx is zero -- return ENAMETOOLONG */
1437 movl $ENAMETOOLONG,%eax
1440 /* set *lencopied and return %eax */
1462 cld /* compare forwards */
1481 * Handling of special 386 registers and descriptor tables etc
1483 /* void lgdt(struct region_descriptor *rdp); */
1485 /* reload the descriptor table */
1489 /* flush the prefetch q */
1493 /* reload "stale" selectors */
1504 /* reload code selector by turning return into intersegmental return */
1511 * void lidt(struct region_descriptor *rdp);
1519 * void lldt(u_short sel)
1526 * void ltr(u_short sel)
1532 /* ssdtosd(*ssdp,*sdp) */
1569 /* void load_cr3(caddr_t cr3) */
1571 #if defined(SWTCH_OPTIM_STATS)
1572 incl _tlb_flush_count
1583 /* void load_cr4(caddr_t cr4) */
1589 /* void load_dr6(u_int dr6) */
1595 /* void reset_dbregs() */
1598 movl %eax,%dr7 /* disable all breapoints first */
1606 /*****************************************************************************/
1607 /* setjump, longjump */
1608 /*****************************************************************************/
1612 movl %ebx,(%eax) /* save ebx */
1613 movl %esp,4(%eax) /* save esp */
1614 movl %ebp,8(%eax) /* save ebp */
1615 movl %esi,12(%eax) /* save esi */
1616 movl %edi,16(%eax) /* save edi */
1617 movl (%esp),%edx /* get rta */
1618 movl %edx,20(%eax) /* save eip */
1619 xorl %eax,%eax /* return(0); */
1624 movl (%eax),%ebx /* restore ebx */
1625 movl 4(%eax),%esp /* restore esp */
1626 movl 8(%eax),%ebp /* restore ebp */
1627 movl 12(%eax),%esi /* restore esi */
1628 movl 16(%eax),%edi /* restore edi */
1629 movl 20(%eax),%edx /* get rta */
1630 movl %edx,(%esp) /* put in return frame */
1631 xorl %eax,%eax /* return(1); */
1636 * Support for BB-profiling (gcc -a). The kernbb program will extract
1637 * the data from the kernel.
1647 NON_GPROF_ENTRY(__bb_init_func)
1653 .byte 0xc3 /* avoid macro for `ret' */