| /* |
| * User Space Access Routines |
| * |
| * Copyright (C) 2000-2002 Hewlett-Packard (John Marvin) |
| * Copyright (C) 2000 Richard Hirst <rhirst with parisc-linux.org> |
| * Copyright (C) 2001 Matthieu Delahaye <delahaym at esiee.fr> |
| * Copyright (C) 2003 Randolph Chung <tausq with parisc-linux.org> |
| * Copyright (C) 2017 Helge Deller <deller@gmx.de> |
| * Copyright (C) 2017 John David Anglin <dave.anglin@bell.net> |
| * |
| * |
| * This program is free software; you can redistribute it and/or modify |
| * it under the terms of the GNU General Public License as published by |
| * the Free Software Foundation; either version 2, or (at your option) |
| * any later version. |
| * |
| * This program is distributed in the hope that it will be useful, |
| * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| * GNU General Public License for more details. |
| * |
| * You should have received a copy of the GNU General Public License |
| * along with this program; if not, write to the Free Software |
| * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| */ |
| |
| /* |
| * These routines still have plenty of room for optimization |
| * (word & doubleword load/store, dual issue, store hints, etc.). |
| */ |
| |
| /* |
| * The following routines assume that space register 3 (sr3) contains |
| * the space id associated with the current users address space. |
| */ |
| |
| |
| .text |
| |
| #include <asm/assembly.h> |
| #include <asm/errno.h> |
| #include <linux/linkage.h> |
| |
| /* |
| * get_sr gets the appropriate space value into |
| * sr1 for kernel/user space access, depending |
| * on the flag stored in the task structure. |
| */ |
| |
| .macro get_sr |
| mfctl %cr30,%r1 |
| ldw TI_SEGMENT(%r1),%r22 |
| mfsp %sr3,%r1 |
| or,<> %r22,%r0,%r0 |
| copy %r0,%r1 |
| mtsp %r1,%sr1 |
| .endm |
| |
| /* |
| * unsigned long lclear_user(void *to, unsigned long n) |
| * |
| * Returns 0 for success. |
| * otherwise, returns number of bytes not transferred. |
| */ |
| |
| ENTRY_CFI(lclear_user) |
| comib,=,n 0,%r25,$lclu_done |
| get_sr |
| $lclu_loop: |
| addib,<> -1,%r25,$lclu_loop |
| 1: stbs,ma %r0,1(%sr1,%r26) |
| |
| $lclu_done: |
| bv %r0(%r2) |
| copy %r25,%r28 |
| |
| 2: b $lclu_done |
| ldo 1(%r25),%r25 |
| |
| ASM_EXCEPTIONTABLE_ENTRY(1b,2b) |
| ENDPROC_CFI(lclear_user) |
| |
| |
| /* |
| * long lstrnlen_user(char *s, long n) |
| * |
| * Returns 0 if exception before zero byte or reaching N, |
| * N+1 if N would be exceeded, |
| * else strlen + 1 (i.e. includes zero byte). |
| */ |
| |
| ENTRY_CFI(lstrnlen_user) |
| comib,= 0,%r25,$lslen_nzero |
| copy %r26,%r24 |
| get_sr |
| 1: ldbs,ma 1(%sr1,%r26),%r1 |
| $lslen_loop: |
| comib,=,n 0,%r1,$lslen_done |
| addib,<> -1,%r25,$lslen_loop |
| 2: ldbs,ma 1(%sr1,%r26),%r1 |
| $lslen_done: |
| bv %r0(%r2) |
| sub %r26,%r24,%r28 |
| |
| $lslen_nzero: |
| b $lslen_done |
| ldo 1(%r26),%r26 /* special case for N == 0 */ |
| |
| 3: b $lslen_done |
| copy %r24,%r26 /* reset r26 so 0 is returned on fault */ |
| |
| ASM_EXCEPTIONTABLE_ENTRY(1b,3b) |
| ASM_EXCEPTIONTABLE_ENTRY(2b,3b) |
| |
| ENDPROC_CFI(lstrnlen_user) |
| |
| |
| /* |
| * unsigned long pa_memcpy(void *dstp, const void *srcp, unsigned long len) |
| * |
| * Inputs: |
| * - sr1 already contains space of source region |
| * - sr2 already contains space of destination region |
| * |
| * Returns: |
| * - number of bytes that could not be copied. |
| * On success, this will be zero. |
| * |
| * This code is based on a C-implementation of a copy routine written by |
| * Randolph Chung, which in turn was derived from the glibc. |
| * |
| * Several strategies are tried to try to get the best performance for various |
| * conditions. In the optimal case, we copy by loops that copy 32- or 16-bytes |
| * at a time using general registers. Unaligned copies are handled either by |
| * aligning the destination and then using shift-and-write method, or in a few |
| * cases by falling back to a byte-at-a-time copy. |
| * |
| * Testing with various alignments and buffer sizes shows that this code is |
| * often >10x faster than a simple byte-at-a-time copy, even for strangely |
| * aligned operands. It is interesting to note that the glibc version of memcpy |
| * (written in C) is actually quite fast already. This routine is able to beat |
| * it by 30-40% for aligned copies because of the loop unrolling, but in some |
| * cases the glibc version is still slightly faster. This lends more |
| * credibility that gcc can generate very good code as long as we are careful. |
| * |
| * Possible optimizations: |
| * - add cache prefetching |
| * - try not to use the post-increment address modifiers; they may create |
| * additional interlocks. Assumption is that those were only efficient on old |
| * machines (pre PA8000 processors) |
| */ |
| |
| dst = arg0 |
| src = arg1 |
| len = arg2 |
| end = arg3 |
| t1 = r19 |
| t2 = r20 |
| t3 = r21 |
| t4 = r22 |
| srcspc = sr1 |
| dstspc = sr2 |
| |
| t0 = r1 |
| a1 = t1 |
| a2 = t2 |
| a3 = t3 |
| a0 = t4 |
| |
| save_src = ret0 |
| save_dst = ret1 |
| save_len = r31 |
| |
| ENTRY_CFI(pa_memcpy) |
| /* Last destination address */ |
| add dst,len,end |
| |
| /* short copy with less than 16 bytes? */ |
| cmpib,COND(>>=),n 15,len,.Lbyte_loop |
| |
| /* same alignment? */ |
| xor src,dst,t0 |
| extru t0,31,2,t1 |
| cmpib,<>,n 0,t1,.Lunaligned_copy |
| |
| #ifdef CONFIG_64BIT |
| /* only do 64-bit copies if we can get aligned. */ |
| extru t0,31,3,t1 |
| cmpib,<>,n 0,t1,.Lalign_loop32 |
| |
| /* loop until we are 64-bit aligned */ |
| .Lalign_loop64: |
| extru dst,31,3,t1 |
| cmpib,=,n 0,t1,.Lcopy_loop_16_start |
| 20: ldb,ma 1(srcspc,src),t1 |
| 21: stb,ma t1,1(dstspc,dst) |
| b .Lalign_loop64 |
| ldo -1(len),len |
| |
| ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done) |
| |
| .Lcopy_loop_16_start: |
| ldi 31,t0 |
| .Lcopy_loop_16: |
| cmpb,COND(>>=),n t0,len,.Lword_loop |
| |
| 10: ldd 0(srcspc,src),t1 |
| 11: ldd 8(srcspc,src),t2 |
| ldo 16(src),src |
| 12: std,ma t1,8(dstspc,dst) |
| 13: std,ma t2,8(dstspc,dst) |
| 14: ldd 0(srcspc,src),t1 |
| 15: ldd 8(srcspc,src),t2 |
| ldo 16(src),src |
| 16: std,ma t1,8(dstspc,dst) |
| 17: std,ma t2,8(dstspc,dst) |
| |
| ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(11b,.Lcopy16_fault) |
| ASM_EXCEPTIONTABLE_ENTRY(12b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(13b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(14b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(15b,.Lcopy16_fault) |
| ASM_EXCEPTIONTABLE_ENTRY(16b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(17b,.Lcopy_done) |
| |
| b .Lcopy_loop_16 |
| ldo -32(len),len |
| |
| .Lword_loop: |
| cmpib,COND(>>=),n 3,len,.Lbyte_loop |
| 20: ldw,ma 4(srcspc,src),t1 |
| 21: stw,ma t1,4(dstspc,dst) |
| b .Lword_loop |
| ldo -4(len),len |
| |
| ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done) |
| |
| #endif /* CONFIG_64BIT */ |
| |
| /* loop until we are 32-bit aligned */ |
| .Lalign_loop32: |
| extru dst,31,2,t1 |
| cmpib,=,n 0,t1,.Lcopy_loop_8 |
| 20: ldb,ma 1(srcspc,src),t1 |
| 21: stb,ma t1,1(dstspc,dst) |
| b .Lalign_loop32 |
| ldo -1(len),len |
| |
| ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done) |
| |
| |
| .Lcopy_loop_8: |
| cmpib,COND(>>=),n 15,len,.Lbyte_loop |
| |
| 10: ldw 0(srcspc,src),t1 |
| 11: ldw 4(srcspc,src),t2 |
| 12: stw,ma t1,4(dstspc,dst) |
| 13: stw,ma t2,4(dstspc,dst) |
| 14: ldw 8(srcspc,src),t1 |
| 15: ldw 12(srcspc,src),t2 |
| ldo 16(src),src |
| 16: stw,ma t1,4(dstspc,dst) |
| 17: stw,ma t2,4(dstspc,dst) |
| |
| ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(11b,.Lcopy8_fault) |
| ASM_EXCEPTIONTABLE_ENTRY(12b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(13b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(14b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(15b,.Lcopy8_fault) |
| ASM_EXCEPTIONTABLE_ENTRY(16b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(17b,.Lcopy_done) |
| |
| b .Lcopy_loop_8 |
| ldo -16(len),len |
| |
| .Lbyte_loop: |
| cmpclr,COND(<>) len,%r0,%r0 |
| b,n .Lcopy_done |
| 20: ldb 0(srcspc,src),t1 |
| ldo 1(src),src |
| 21: stb,ma t1,1(dstspc,dst) |
| b .Lbyte_loop |
| ldo -1(len),len |
| |
| ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done) |
| |
| .Lcopy_done: |
| bv %r0(%r2) |
| sub end,dst,ret0 |
| |
| |
| /* src and dst are not aligned the same way. */ |
| /* need to go the hard way */ |
| .Lunaligned_copy: |
| /* align until dst is 32bit-word-aligned */ |
| extru dst,31,2,t1 |
| cmpib,=,n 0,t1,.Lcopy_dstaligned |
| 20: ldb 0(srcspc,src),t1 |
| ldo 1(src),src |
| 21: stb,ma t1,1(dstspc,dst) |
| b .Lunaligned_copy |
| ldo -1(len),len |
| |
| ASM_EXCEPTIONTABLE_ENTRY(20b,.Lcopy_done) |
| ASM_EXCEPTIONTABLE_ENTRY(21b,.Lcopy_done) |
| |
| .Lcopy_dstaligned: |
| |
| /* store src, dst and len in safe place */ |
| copy src,save_src |
| copy dst,save_dst |
| copy len,save_len |
| |
| /* len now needs give number of words to copy */ |
| SHRREG len,2,len |
| |
| /* |
| * Copy from a not-aligned src to an aligned dst using shifts. |
| * Handles 4 words per loop. |
| */ |
| |
| depw,z src,28,2,t0 |
| subi 32,t0,t0 |
| mtsar t0 |
| extru len,31,2,t0 |
| cmpib,= 2,t0,.Lcase2 |
| /* Make src aligned by rounding it down. */ |
| depi 0,31,2,src |
| |
| cmpiclr,<> 3,t0,%r0 |
| b,n .Lcase3 |
| cmpiclr,<> 1,t0,%r0 |
| b,n .Lcase1 |
| .Lcase0: |
| cmpb,COND(=) %r0,len,.Lcda_finish |
| nop |
| |
| 1: ldw,ma 4(srcspc,src), a3 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| 1: ldw,ma 4(srcspc,src), a0 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| b,n .Ldo3 |
| .Lcase1: |
| 1: ldw,ma 4(srcspc,src), a2 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| 1: ldw,ma 4(srcspc,src), a3 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| ldo -1(len),len |
| cmpb,COND(=),n %r0,len,.Ldo0 |
| .Ldo4: |
| 1: ldw,ma 4(srcspc,src), a0 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| shrpw a2, a3, %sar, t0 |
| 1: stw,ma t0, 4(dstspc,dst) |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done) |
| .Ldo3: |
| 1: ldw,ma 4(srcspc,src), a1 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| shrpw a3, a0, %sar, t0 |
| 1: stw,ma t0, 4(dstspc,dst) |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done) |
| .Ldo2: |
| 1: ldw,ma 4(srcspc,src), a2 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| shrpw a0, a1, %sar, t0 |
| 1: stw,ma t0, 4(dstspc,dst) |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done) |
| .Ldo1: |
| 1: ldw,ma 4(srcspc,src), a3 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| shrpw a1, a2, %sar, t0 |
| 1: stw,ma t0, 4(dstspc,dst) |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done) |
| ldo -4(len),len |
| cmpb,COND(<>) %r0,len,.Ldo4 |
| nop |
| .Ldo0: |
| shrpw a2, a3, %sar, t0 |
| 1: stw,ma t0, 4(dstspc,dst) |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcopy_done) |
| |
| .Lcda_rdfault: |
| .Lcda_finish: |
| /* calculate new src, dst and len and jump to byte-copy loop */ |
| sub dst,save_dst,t0 |
| add save_src,t0,src |
| b .Lbyte_loop |
| sub save_len,t0,len |
| |
| .Lcase3: |
| 1: ldw,ma 4(srcspc,src), a0 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| 1: ldw,ma 4(srcspc,src), a1 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| b .Ldo2 |
| ldo 1(len),len |
| .Lcase2: |
| 1: ldw,ma 4(srcspc,src), a1 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| 1: ldw,ma 4(srcspc,src), a2 |
| ASM_EXCEPTIONTABLE_ENTRY(1b,.Lcda_rdfault) |
| b .Ldo1 |
| ldo 2(len),len |
| |
| |
| /* fault exception fixup handlers: */ |
| #ifdef CONFIG_64BIT |
| .Lcopy16_fault: |
| b .Lcopy_done |
| 10: std,ma t1,8(dstspc,dst) |
| ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done) |
| #endif |
| |
| .Lcopy8_fault: |
| b .Lcopy_done |
| 10: stw,ma t1,4(dstspc,dst) |
| ASM_EXCEPTIONTABLE_ENTRY(10b,.Lcopy_done) |
| ENDPROC_CFI(pa_memcpy) |
| |
| .end |
