|
@@ -0,0 +1,169 @@
|
|
1
|
+/*
|
|
2
|
+ * Copyright (C) 2012 Michael Brown <mbrown@fensystems.co.uk>.
|
|
3
|
+ *
|
|
4
|
+ * This program is free software; you can redistribute it and/or
|
|
5
|
+ * modify it under the terms of the GNU General Public License as
|
|
6
|
+ * published by the Free Software Foundation; either version 2 of the
|
|
7
|
+ * License, or (at your option) any later version.
|
|
8
|
+ *
|
|
9
|
+ * This program is distributed in the hope that it will be useful, but
|
|
10
|
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
11
|
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
12
|
+ * General Public License for more details.
|
|
13
|
+ *
|
|
14
|
+ * You should have received a copy of the GNU General Public License
|
|
15
|
+ * along with this program; if not, write to the Free Software
|
|
16
|
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
|
17
|
+ * 02110-1301, USA.
|
|
18
|
+ */
|
|
19
|
+
|
|
20
|
+FILE_LICENCE ( GPL2_OR_LATER );
|
|
21
|
+
|
|
22
|
+/** @file
|
|
23
|
+ *
|
|
24
|
+ * TCP/IP checksum
|
|
25
|
+ *
|
|
26
|
+ */
|
|
27
|
+
|
|
28
|
+#include <limits.h>
|
|
29
|
+#include <ipxe/tcpip.h>
|
|
30
|
+
|
|
31
|
+extern char x86_tcpip_loop_end[];
|
|
32
|
+
|
|
33
|
+/**
|
|
34
|
+ * Calculate continued TCP/IP checkum
|
|
35
|
+ *
|
|
36
|
+ * @v partial Checksum of already-summed data, in network byte order
|
|
37
|
+ * @v data Data buffer
|
|
38
|
+ * @v len Length of data buffer
|
|
39
|
+ * @ret cksum Updated checksum, in network byte order
|
|
40
|
+ */
|
|
41
|
+uint16_t x86_tcpip_continue_chksum ( uint16_t partial,
|
|
42
|
+ const void *data, size_t len ) {
|
|
43
|
+ unsigned long sum = ( ( ~partial ) & 0xffff );
|
|
44
|
+ unsigned long initial_word_count;
|
|
45
|
+ unsigned long loop_count;
|
|
46
|
+ unsigned long loop_partial_count;
|
|
47
|
+ unsigned long final_word_count;
|
|
48
|
+ unsigned long final_byte;
|
|
49
|
+ unsigned long discard_S;
|
|
50
|
+ unsigned long discard_c;
|
|
51
|
+ unsigned long discard_a;
|
|
52
|
+ unsigned long discard_r1;
|
|
53
|
+ unsigned long discard_r2;
|
|
54
|
+
|
|
55
|
+ /* Calculate number of initial 16-bit words required to bring
|
|
56
|
+ * the main loop into alignment. (We don't care about the
|
|
57
|
+ * speed for data aligned to less than 16 bits, since this
|
|
58
|
+ * situation won't occur in practice.)
|
|
59
|
+ */
|
|
60
|
+ if ( len >= sizeof ( sum ) ) {
|
|
61
|
+ initial_word_count = ( ( -( ( intptr_t ) data ) &
|
|
62
|
+ ( sizeof ( sum ) - 1 ) ) >> 1 );
|
|
63
|
+ } else {
|
|
64
|
+ initial_word_count = 0;
|
|
65
|
+ }
|
|
66
|
+ len -= ( initial_word_count * 2 );
|
|
67
|
+
|
|
68
|
+ /* Calculate number of iterations of the main loop. This loop
|
|
69
|
+ * processes native machine words (32-bit or 64-bit), and is
|
|
70
|
+ * unrolled 16 times. We calculate an overall iteration
|
|
71
|
+ * count, and a starting point for the first iteration.
|
|
72
|
+ */
|
|
73
|
+ loop_count = ( len / ( sizeof ( sum ) * 16 ) );
|
|
74
|
+ loop_partial_count =
|
|
75
|
+ ( ( len % ( sizeof ( sum ) * 16 ) ) / sizeof ( sum ) );
|
|
76
|
+
|
|
77
|
+ /* Calculate number of 16-bit words remaining after the main
|
|
78
|
+ * loop completes.
|
|
79
|
+ */
|
|
80
|
+ final_word_count = ( ( len % sizeof ( sum ) ) / 2 );
|
|
81
|
+
|
|
82
|
+ /* Calculate whether or not a final byte remains at the end */
|
|
83
|
+ final_byte = ( len & 1 );
|
|
84
|
+
|
|
85
|
+ /* Calculate the checksum */
|
|
86
|
+ __asm__ ( /* Calculate position at which to jump into the
|
|
87
|
+ * unrolled loop.
|
|
88
|
+ */
|
|
89
|
+ "imul $( -x86_tcpip_loop_step_size ), %4\n\t"
|
|
90
|
+ "add %5, %4\n\t"
|
|
91
|
+
|
|
92
|
+ /* Clear carry flag before starting checksumming */
|
|
93
|
+ "clc\n\t"
|
|
94
|
+
|
|
95
|
+ /* Checksum initial words */
|
|
96
|
+ "jmp 2f\n\t"
|
|
97
|
+ "\n1:\n\t"
|
|
98
|
+ "lodsw\n\t"
|
|
99
|
+ "adcw %w2, %w0\n\t"
|
|
100
|
+ "\n2:\n\t"
|
|
101
|
+ "loop 1b\n\t"
|
|
102
|
+
|
|
103
|
+ /* Main "lods;adc" loop, unrolled x16 */
|
|
104
|
+ "mov %12, %3\n\t"
|
|
105
|
+ "jmp *%4\n\t"
|
|
106
|
+ "\nx86_tcpip_loop_start:\n\t"
|
|
107
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
108
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
109
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
110
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
111
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
112
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
113
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
114
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
115
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
116
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
117
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
118
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
119
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
120
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
121
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
122
|
+ "lods%z2\n\tadc %2, %0\n\t"
|
|
123
|
+ "\nx86_tcpip_loop_end:\n\t"
|
|
124
|
+ "loop x86_tcpip_loop_start\n\t"
|
|
125
|
+ ".equ x86_tcpip_loop_step_size, "
|
|
126
|
+ " ( ( x86_tcpip_loop_end - x86_tcpip_loop_start ) / 16 )\n\t"
|
|
127
|
+
|
|
128
|
+ /* Checksum remaining whole words */
|
|
129
|
+ "mov %13, %3\n\t"
|
|
130
|
+ "jmp 2f\n\t"
|
|
131
|
+ "\n1:\n\t"
|
|
132
|
+ "lodsw\n\t"
|
|
133
|
+ "adcw %w2, %w0\n\t"
|
|
134
|
+ "\n2:\n\t"
|
|
135
|
+ "loop 1b\n\t"
|
|
136
|
+
|
|
137
|
+ /* Checksum final byte if applicable */
|
|
138
|
+ "mov %14, %3\n\t"
|
|
139
|
+ "loop 1f\n\t"
|
|
140
|
+ "adcb (%1), %b0\n\t"
|
|
141
|
+ "adcb $0, %h0\n\t"
|
|
142
|
+ "\n1:\n\t"
|
|
143
|
+
|
|
144
|
+ /* Fold down to a uint16_t */
|
|
145
|
+ "push %0\n\t"
|
|
146
|
+ "popw %w0\n\t"
|
|
147
|
+ "popw %w2\n\t"
|
|
148
|
+ "adcw %w2, %w0\n\t"
|
|
149
|
+#if ULONG_MAX > 0xffffffffUL /* 64-bit only */
|
|
150
|
+ "popw %w2\n\t"
|
|
151
|
+ "adcw %w2, %w0\n\t"
|
|
152
|
+ "popw %w2\n\t"
|
|
153
|
+ "adcw %w2, %w0\n\t"
|
|
154
|
+#endif /* 64-bit only */
|
|
155
|
+
|
|
156
|
+ /* Consume CF */
|
|
157
|
+ "adcw $0, %w0\n\t"
|
|
158
|
+ "adcw $0, %w0\n\t"
|
|
159
|
+
|
|
160
|
+ : "=&Q" ( sum ), "=&S" ( discard_S ), "=&a" ( discard_a ),
|
|
161
|
+ "=&c" ( discard_c ), "=&r" ( discard_r1 ),
|
|
162
|
+ "=&r" ( discard_r2 )
|
|
163
|
+ : "0" ( sum ), "1" ( data ), "2" ( 0 ),
|
|
164
|
+ "3" ( initial_word_count + 1 ), "4" ( loop_partial_count ),
|
|
165
|
+ "5" ( x86_tcpip_loop_end ), "g" ( loop_count + 1 ),
|
|
166
|
+ "g" ( final_word_count + 1 ), "g" ( final_byte ) );
|
|
167
|
+
|
|
168
|
+ return ( ~sum & 0xffff );
|
|
169
|
+}
|