diff options
Diffstat (limited to 'arch/mips/lib/csum_partial.S')
| -rw-r--r-- | arch/mips/lib/csum_partial.S | 275 |
1 files changed, 158 insertions, 117 deletions
diff --git a/arch/mips/lib/csum_partial.S b/arch/mips/lib/csum_partial.S index c0a77fe038be..8d7784122c14 100644 --- a/arch/mips/lib/csum_partial.S +++ b/arch/mips/lib/csum_partial.S @@ -7,6 +7,7 @@ * * Copyright (C) 1998, 1999 Ralf Baechle * Copyright (C) 1999 Silicon Graphics, Inc. + * Copyright (C) 2007 Maciej W. Rozycki */ #include <linux/errno.h> #include <asm/asm.h> @@ -52,9 +53,12 @@ #define UNIT(unit) ((unit)*NBYTES) #define ADDC(sum,reg) \ + .set push; \ + .set noat; \ ADD sum, reg; \ sltu v1, sum, reg; \ - ADD sum, v1 + ADD sum, v1; \ + .set pop #define CSUM_BIGCHUNK1(src, offset, sum, _t0, _t1, _t2, _t3) \ LOAD _t0, (offset + UNIT(0))(src); \ @@ -92,13 +96,13 @@ LEAF(csum_partial) move t7, zero sltiu t8, a1, 0x8 - bnez t8, small_csumcpy /* < 8 bytes to copy */ + bnez t8, .Lsmall_csumcpy /* < 8 bytes to copy */ move t2, a1 andi t7, src, 0x1 /* odd buffer? */ -hword_align: - beqz t7, word_align +.Lhword_align: + beqz t7, .Lword_align andi t8, src, 0x2 lbu t0, (src) @@ -110,8 +114,8 @@ hword_align: PTR_ADDU src, src, 0x1 andi t8, src, 0x2 -word_align: - beqz t8, dword_align +.Lword_align: + beqz t8, .Ldword_align sltiu t8, a1, 56 lhu t0, (src) @@ -120,12 +124,12 @@ word_align: sltiu t8, a1, 56 PTR_ADDU src, src, 0x2 -dword_align: - bnez t8, do_end_words +.Ldword_align: + bnez t8, .Ldo_end_words move t8, a1 andi t8, src, 0x4 - beqz t8, qword_align + beqz t8, .Lqword_align andi t8, src, 0x8 lw t0, 0x00(src) @@ -134,8 +138,8 @@ dword_align: PTR_ADDU src, src, 0x4 andi t8, src, 0x8 -qword_align: - beqz t8, oword_align +.Lqword_align: + beqz t8, .Loword_align andi t8, src, 0x10 #ifdef USE_DOUBLE @@ -152,8 +156,8 @@ qword_align: PTR_ADDU src, src, 0x8 andi t8, src, 0x10 -oword_align: - beqz t8, begin_movement +.Loword_align: + beqz t8, .Lbegin_movement LONG_SRL t8, a1, 0x7 #ifdef USE_DOUBLE @@ -168,51 +172,55 @@ oword_align: PTR_ADDU src, src, 0x10 LONG_SRL t8, a1, 0x7 -begin_movement: +.Lbegin_movement: beqz t8, 1f andi t2, a1, 0x40 -move_128bytes: +.Lmove_128bytes: CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4) CSUM_BIGCHUNK(src, 0x20, sum, t0, t1, t3, t4) CSUM_BIGCHUNK(src, 0x40, sum, t0, t1, t3, t4) CSUM_BIGCHUNK(src, 0x60, sum, t0, t1, t3, t4) LONG_SUBU t8, t8, 0x01 - bnez t8, move_128bytes - PTR_ADDU src, src, 0x80 + .set reorder /* DADDI_WAR */ + PTR_ADDU src, src, 0x80 + bnez t8, .Lmove_128bytes + .set noreorder 1: beqz t2, 1f andi t2, a1, 0x20 -move_64bytes: +.Lmove_64bytes: CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4) CSUM_BIGCHUNK(src, 0x20, sum, t0, t1, t3, t4) PTR_ADDU src, src, 0x40 1: - beqz t2, do_end_words + beqz t2, .Ldo_end_words andi t8, a1, 0x1c -move_32bytes: +.Lmove_32bytes: CSUM_BIGCHUNK(src, 0x00, sum, t0, t1, t3, t4) andi t8, a1, 0x1c PTR_ADDU src, src, 0x20 -do_end_words: - beqz t8, small_csumcpy +.Ldo_end_words: + beqz t8, .Lsmall_csumcpy andi t2, a1, 0x3 LONG_SRL t8, t8, 0x2 -end_words: +.Lend_words: lw t0, (src) LONG_SUBU t8, t8, 0x1 ADDC(sum, t0) - bnez t8, end_words - PTR_ADDU src, src, 0x4 + .set reorder /* DADDI_WAR */ + PTR_ADDU src, src, 0x4 + bnez t8, .Lend_words + .set noreorder /* unknown src alignment and < 8 bytes to go */ -small_csumcpy: +.Lsmall_csumcpy: move a1, t2 andi t0, a1, 4 @@ -246,6 +254,8 @@ small_csumcpy: 1: ADDC(sum, t1) /* fold checksum */ + .set push + .set noat #ifdef USE_DOUBLE dsll32 v1, sum, 0 daddu sum, v1 @@ -266,6 +276,7 @@ small_csumcpy: srl sum, sum, 8 or sum, v1 andi sum, 0xffff + .set pop 1: .set reorder /* Add the passed partial csum. */ @@ -373,7 +384,11 @@ small_csumcpy: #define ADDRMASK (NBYTES-1) +#ifndef CONFIG_CPU_DADDI_WORKAROUNDS .set noat +#else + .set at=v1 +#endif LEAF(__csum_partial_copy_user) PTR_ADDU AT, src, len /* See (1) above. */ @@ -398,95 +413,101 @@ FEXPORT(csum_partial_copy_nocheck) */ sltu t2, len, NBYTES and t1, dst, ADDRMASK - bnez t2, copy_bytes_checklen + bnez t2, .Lcopy_bytes_checklen and t0, src, ADDRMASK andi odd, dst, 0x1 /* odd buffer? */ - bnez t1, dst_unaligned + bnez t1, .Ldst_unaligned nop - bnez t0, src_unaligned_dst_aligned + bnez t0, .Lsrc_unaligned_dst_aligned /* * use delay slot for fall-through * src and dst are aligned; need to compute rem */ -both_aligned: +.Lboth_aligned: SRL t0, len, LOG_NBYTES+3 # +3 for 8 units/iter - beqz t0, cleanup_both_aligned # len < 8*NBYTES + beqz t0, .Lcleanup_both_aligned # len < 8*NBYTES nop SUB len, 8*NBYTES # subtract here for bgez loop .align 4 1: -EXC( LOAD t0, UNIT(0)(src), l_exc) -EXC( LOAD t1, UNIT(1)(src), l_exc_copy) -EXC( LOAD t2, UNIT(2)(src), l_exc_copy) -EXC( LOAD t3, UNIT(3)(src), l_exc_copy) -EXC( LOAD t4, UNIT(4)(src), l_exc_copy) -EXC( LOAD t5, UNIT(5)(src), l_exc_copy) -EXC( LOAD t6, UNIT(6)(src), l_exc_copy) -EXC( LOAD t7, UNIT(7)(src), l_exc_copy) +EXC( LOAD t0, UNIT(0)(src), .Ll_exc) +EXC( LOAD t1, UNIT(1)(src), .Ll_exc_copy) +EXC( LOAD t2, UNIT(2)(src), .Ll_exc_copy) +EXC( LOAD t3, UNIT(3)(src), .Ll_exc_copy) +EXC( LOAD t4, UNIT(4)(src), .Ll_exc_copy) +EXC( LOAD t5, UNIT(5)(src), .Ll_exc_copy) +EXC( LOAD t6, UNIT(6)(src), .Ll_exc_copy) +EXC( LOAD t7, UNIT(7)(src), .Ll_exc_copy) SUB len, len, 8*NBYTES ADD src, src, 8*NBYTES -EXC( STORE t0, UNIT(0)(dst), s_exc) +EXC( STORE t0, UNIT(0)(dst), .Ls_exc) ADDC(sum, t0) -EXC( STORE t1, UNIT(1)(dst), s_exc) +EXC( STORE t1, UNIT(1)(dst), .Ls_exc) ADDC(sum, t1) -EXC( STORE t2, UNIT(2)(dst), s_exc) +EXC( STORE t2, UNIT(2)(dst), .Ls_exc) ADDC(sum, t2) -EXC( STORE t3, UNIT(3)(dst), s_exc) +EXC( STORE t3, UNIT(3)(dst), .Ls_exc) ADDC(sum, t3) -EXC( STORE t4, UNIT(4)(dst), s_exc) +EXC( STORE t4, UNIT(4)(dst), .Ls_exc) ADDC(sum, t4) -EXC( STORE t5, UNIT(5)(dst), s_exc) +EXC( STORE t5, UNIT(5)(dst), .Ls_exc) ADDC(sum, t5) -EXC( STORE t6, UNIT(6)(dst), s_exc) +EXC( STORE t6, UNIT(6)(dst), .Ls_exc) ADDC(sum, t6) -EXC( STORE t7, UNIT(7)(dst), s_exc) +EXC( STORE t7, UNIT(7)(dst), .Ls_exc) ADDC(sum, t7) + .set reorder /* DADDI_WAR */ + ADD dst, dst, 8*NBYTES bgez len, 1b - ADD dst, dst, 8*NBYTES + .set noreorder ADD len, 8*NBYTES # revert len (see above) /* * len == the number of bytes left to copy < 8*NBYTES */ -cleanup_both_aligned: +.Lcleanup_both_aligned: #define rem t7 - beqz len, done + beqz len, .Ldone sltu t0, len, 4*NBYTES - bnez t0, less_than_4units + bnez t0, .Lless_than_4units and rem, len, (NBYTES-1) # rem = len % NBYTES /* * len >= 4*NBYTES */ -EXC( LOAD t0, UNIT(0)(src), l_exc) -EXC( LOAD t1, UNIT(1)(src), l_exc_copy) -EXC( LOAD t2, UNIT(2)(src), l_exc_copy) -EXC( LOAD t3, UNIT(3)(src), l_exc_copy) +EXC( LOAD t0, UNIT(0)(src), .Ll_exc) +EXC( LOAD t1, UNIT(1)(src), .Ll_exc_copy) +EXC( LOAD t2, UNIT(2)(src), .Ll_exc_copy) +EXC( LOAD t3, UNIT(3)(src), .Ll_exc_copy) SUB len, len, 4*NBYTES ADD src, src, 4*NBYTES -EXC( STORE t0, UNIT(0)(dst), s_exc) +EXC( STORE t0, UNIT(0)(dst), .Ls_exc) ADDC(sum, t0) -EXC( STORE t1, UNIT(1)(dst), s_exc) +EXC( STORE t1, UNIT(1)(dst), .Ls_exc) ADDC(sum, t1) -EXC( STORE t2, UNIT(2)(dst), s_exc) +EXC( STORE t2, UNIT(2)(dst), .Ls_exc) ADDC(sum, t2) -EXC( STORE t3, UNIT(3)(dst), s_exc) +EXC( STORE t3, UNIT(3)(dst), .Ls_exc) ADDC(sum, t3) - beqz len, done - ADD dst, dst, 4*NBYTES -less_than_4units: + .set reorder /* DADDI_WAR */ + ADD dst, dst, 4*NBYTES + beqz len, .Ldone + .set noreorder +.Lless_than_4units: /* * rem = len % NBYTES */ - beq rem, len, copy_bytes + beq rem, len, .Lcopy_bytes nop 1: -EXC( LOAD t0, 0(src), l_exc) +EXC( LOAD t0, 0(src), .Ll_exc) ADD src, src, NBYTES SUB len, len, NBYTES -EXC( STORE t0, 0(dst), s_exc) +EXC( STORE t0, 0(dst), .Ls_exc) ADDC(sum, t0) + .set reorder /* DADDI_WAR */ + ADD dst, dst, NBYTES bne rem, len, 1b - ADD dst, dst, NBYTES + .set noreorder /* * src and dst are aligned, need to copy rem bytes (rem < NBYTES) @@ -500,20 +521,20 @@ EXC( STORE t0, 0(dst), s_exc) * more instruction-level parallelism. */ #define bits t2 - beqz len, done + beqz len, .Ldone ADD t1, dst, len # t1 is just past last byte of dst li bits, 8*NBYTES SLL rem, len, 3 # rem = number of bits to keep -EXC( LOAD t0, 0(src), l_exc) +EXC( LOAD t0, 0(src), .Ll_exc) SUB bits, bits, rem # bits = number of bits to discard SHIFT_DISCARD t0, t0, bits -EXC( STREST t0, -1(t1), s_exc) +EXC( STREST t0, -1(t1), .Ls_exc) SHIFT_DISCARD_REVERT t0, t0, bits .set reorder ADDC(sum, t0) - b done + b .Ldone .set noreorder -dst_unaligned: +.Ldst_unaligned: /* * dst is unaligned * t0 = src & ADDRMASK @@ -524,25 +545,25 @@ dst_unaligned: * Set match = (src and dst have same alignment) */ #define match rem -EXC( LDFIRST t3, FIRST(0)(src), l_exc) +EXC( LDFIRST t3, FIRST(0)(src), .Ll_exc) ADD t2, zero, NBYTES -EXC( LDREST t3, REST(0)(src), l_exc_copy) +EXC( LDREST t3, REST(0)(src), .Ll_exc_copy) SUB t2, t2, t1 # t2 = number of bytes copied xor match, t0, t1 -EXC( STFIRST t3, FIRST(0)(dst), s_exc) +EXC( STFIRST t3, FIRST(0)(dst), .Ls_exc) SLL t4, t1, 3 # t4 = number of bits to discard SHIFT_DISCARD t3, t3, t4 /* no SHIFT_DISCARD_REVERT to handle odd buffer properly */ ADDC(sum, t3) - beq len, t2, done + beq len, t2, .Ldone SUB len, len, t2 ADD dst, dst, t2 - beqz match, both_aligned + beqz match, .Lboth_aligned ADD src, src, t2 -src_unaligned_dst_aligned: +.Lsrc_unaligned_dst_aligned: SRL t0, len, LOG_NBYTES+2 # +2 for 4 units/iter - beqz t0, cleanup_src_unaligned + beqz t0, .Lcleanup_src_unaligned and rem, len, (4*NBYTES-1) # rem = len % 4*NBYTES 1: /* @@ -551,49 +572,53 @@ src_unaligned_dst_aligned: * It's OK to load FIRST(N+1) before REST(N) because the two addresses * are to the same unit (unless src is aligned, but it's not). */ -EXC( LDFIRST t0, FIRST(0)(src), l_exc) -EXC( LDFIRST t1, FIRST(1)(src), l_exc_copy) +EXC( LDFIRST t0, FIRST(0)(src), .Ll_exc) +EXC( LDFIRST t1, FIRST(1)(src), .Ll_exc_copy) SUB len, len, 4*NBYTES -EXC( LDREST t0, REST(0)(src), l_exc_copy) -EXC( LDREST t1, REST(1)(src), l_exc_copy) -EXC( LDFIRST t2, FIRST(2)(src), l_exc_copy) -EXC( LDFIRST t3, FIRST(3)(src), l_exc_copy) -EXC( LDREST t2, REST(2)(src), l_exc_copy) -EXC( LDREST t3, REST(3)(src), l_exc_copy) +EXC( LDREST t0, REST(0)(src), .Ll_exc_copy) +EXC( LDREST t1, REST(1)(src), .Ll_exc_copy) +EXC( LDFIRST t2, FIRST(2)(src), .Ll_exc_copy) +EXC( LDFIRST t3, FIRST(3)(src), .Ll_exc_copy) +EXC( LDREST t2, REST(2)(src), .Ll_exc_copy) +EXC( LDREST t3, REST(3)(src), .Ll_exc_copy) ADD src, src, 4*NBYTES #ifdef CONFIG_CPU_SB1 nop # improves slotting #endif -EXC( STORE t0, UNIT(0)(dst), s_exc) +EXC( STORE t0, UNIT(0)(dst), .Ls_exc) ADDC(sum, t0) -EXC( STORE t1, UNIT(1)(dst), s_exc) +EXC( STORE t1, UNIT(1)(dst), .Ls_exc) ADDC(sum, t1) -EXC( STORE t2, UNIT(2)(dst), s_exc) +EXC( STORE t2, UNIT(2)(dst), .Ls_exc) ADDC(sum, t2) -EXC( STORE t3, UNIT(3)(dst), s_exc) +EXC( STORE t3, UNIT(3)(dst), .Ls_exc) ADDC(sum, t3) + .set reorder /* DADDI_WAR */ + ADD dst, dst, 4*NBYTES bne len, rem, 1b - ADD dst, dst, 4*NBYTES + .set noreorder -cleanup_src_unaligned: - beqz len, done +.Lcleanup_src_unaligned: + beqz len, .Ldone and rem, len, NBYTES-1 # rem = len % NBYTES - beq rem, len, copy_bytes + beq rem, len, .Lcopy_bytes nop 1: -EXC( LDFIRST t0, FIRST(0)(src), l_exc) -EXC( LDREST t0, REST(0)(src), l_exc_copy) +EXC( LDFIRST t0, FIRST(0)(src), .Ll_exc) +EXC( LDREST t0, REST(0)(src), .Ll_exc_copy) ADD src, src, NBYTES SUB len, len, NBYTES -EXC( STORE t0, 0(dst), s_exc) +EXC( STORE t0, 0(dst), .Ls_exc) ADDC(sum, t0) + .set reorder /* DADDI_WAR */ + ADD dst, dst, NBYTES bne len, rem, 1b - ADD dst, dst, NBYTES + .set noreorder -copy_bytes_checklen: - beqz len, done +.Lcopy_bytes_checklen: + beqz len, .Ldone nop -copy_bytes: +.Lcopy_bytes: /* 0 < len < NBYTES */ #ifdef CONFIG_CPU_LITTLE_ENDIAN #define SHIFT_START 0 @@ -604,14 +629,14 @@ copy_bytes: #endif move t2, zero # partial word li t3, SHIFT_START # shift -/* use l_exc_copy here to return correct sum on fault */ +/* use .Ll_exc_copy here to return correct sum on fault */ #define COPY_BYTE(N) \ -EXC( lbu t0, N(src), l_exc_copy); \ +EXC( lbu t0, N(src), .Ll_exc_copy); \ SUB len, len, 1; \ -EXC( sb t0, N(dst), s_exc); \ +EXC( sb t0, N(dst), .Ls_exc); \ SLLV t0, t0, t3; \ addu t3, SHIFT_INC; \ - beqz len, copy_bytes_done; \ + beqz len, .Lcopy_bytes_done; \ or t2, t0 COPY_BYTE(0) @@ -622,15 +647,17 @@ EXC( sb t0, N(dst), s_exc); \ COPY_BYTE(4) COPY_BYTE(5) #endif -EXC( lbu t0, NBYTES-2(src), l_exc_copy) +EXC( lbu t0, NBYTES-2(src), .Ll_exc_copy) SUB len, len, 1 -EXC( sb t0, NBYTES-2(dst), s_exc) +EXC( sb t0, NBYTES-2(dst), .Ls_exc) SLLV t0, t0, t3 or t2, t0 -copy_bytes_done: +.Lcopy_bytes_done: ADDC(sum, t2) -done: +.Ldone: /* fold checksum */ + .set push + .set noat #ifdef USE_DOUBLE dsll32 v1, sum, 0 daddu sum, v1 @@ -651,13 +678,14 @@ done: srl sum, sum, 8 or sum, v1 andi sum, 0xffff + .set pop 1: .set reorder ADDC(sum, psum) jr ra .set noreorder -l_exc_copy: +.Ll_exc_copy: /* * Copy bytes from src until faulting load address (or until a * lb faults) @@ -672,15 +700,17 @@ l_exc_copy: li t2, SHIFT_START LOAD t0, THREAD_BUADDR(t0) 1: -EXC( lbu t1, 0(src), l_exc) +EXC( lbu t1, 0(src), .Ll_exc) ADD src, src, 1 sb t1, 0(dst) # can't fault -- we're copy_from_user SLLV t1, t1, t2 addu t2, SHIFT_INC ADDC(sum, t1) + .set reorder /* DADDI_WAR */ + ADD dst, dst, 1 bne src, t0, 1b - ADD dst, dst, 1 -l_exc: + .set noreorder +.Ll_exc: LOAD t0, TI_TASK($28) nop LOAD t0, THREAD_BUADDR(t0) # t0 is just past last good address @@ -697,19 +727,30 @@ l_exc: * Clear len bytes starting at dst. Can't call __bzero because it * might modify len. An inefficient loop for these rare times... */ - beqz len, done - SUB src, len, 1 + .set reorder /* DADDI_WAR */ + SUB src, len, 1 + beqz len, .Ldone + .set noreorder 1: sb zero, 0(dst) ADD dst, dst, 1 + .set push + .set noat +#ifndef CONFIG_CPU_DADDI_WORKAROUNDS bnez src, 1b SUB src, src, 1 +#else + li v1, 1 + bnez src, 1b + SUB src, src, v1 +#endif li v1, -EFAULT - b done + b .Ldone sw v1, (errptr) -s_exc: +.Ls_exc: li v0, -1 /* invalid checksum */ li v1, -EFAULT jr ra sw v1, (errptr) + .set pop END(__csum_partial_copy_user) |