xref: /linux/arch/powerpc/include/asm/checksum.h (revision 44f57d78)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 #ifndef _ASM_POWERPC_CHECKSUM_H
3 #define _ASM_POWERPC_CHECKSUM_H
4 #ifdef __KERNEL__
5 
6 /*
7  */
8 
9 #include <linux/bitops.h>
10 #include <linux/in6.h>
11 /*
12  * Computes the checksum of a memory block at src, length len,
13  * and adds in "sum" (32-bit), while copying the block to dst.
14  * If an access exception occurs on src or dst, it stores -EFAULT
15  * to *src_err or *dst_err respectively (if that pointer is not
16  * NULL), and, for an error on src, zeroes the rest of dst.
17  *
18  * Like csum_partial, this must be called with even lengths,
19  * except for the last fragment.
20  */
21 extern __wsum csum_partial_copy_generic(const void *src, void *dst,
22 					      int len, __wsum sum,
23 					      int *src_err, int *dst_err);
24 
25 #define _HAVE_ARCH_COPY_AND_CSUM_FROM_USER
26 extern __wsum csum_and_copy_from_user(const void __user *src, void *dst,
27 				      int len, __wsum sum, int *err_ptr);
28 #define HAVE_CSUM_COPY_USER
29 extern __wsum csum_and_copy_to_user(const void *src, void __user *dst,
30 				    int len, __wsum sum, int *err_ptr);
31 
32 #define csum_partial_copy_nocheck(src, dst, len, sum)   \
33         csum_partial_copy_generic((src), (dst), (len), (sum), NULL, NULL)
34 
35 
36 /*
37  * turns a 32-bit partial checksum (e.g. from csum_partial) into a
38  * 1's complement 16-bit checksum.
39  */
40 static inline __sum16 csum_fold(__wsum sum)
41 {
42 	unsigned int tmp;
43 
44 	/* swap the two 16-bit halves of sum */
45 	__asm__("rlwinm %0,%1,16,0,31" : "=r" (tmp) : "r" (sum));
46 	/* if there is a carry from adding the two 16-bit halves,
47 	   it will carry from the lower half into the upper half,
48 	   giving us the correct sum in the upper half. */
49 	return (__force __sum16)(~((__force u32)sum + tmp) >> 16);
50 }
51 
52 static inline u32 from64to32(u64 x)
53 {
54 	return (x + ror64(x, 32)) >> 32;
55 }
56 
57 static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr, __u32 len,
58 					__u8 proto, __wsum sum)
59 {
60 #ifdef __powerpc64__
61 	u64 s = (__force u32)sum;
62 
63 	s += (__force u32)saddr;
64 	s += (__force u32)daddr;
65 #ifdef __BIG_ENDIAN__
66 	s += proto + len;
67 #else
68 	s += (proto + len) << 8;
69 #endif
70 	return (__force __wsum) from64to32(s);
71 #else
72     __asm__("\n\
73 	addc %0,%0,%1 \n\
74 	adde %0,%0,%2 \n\
75 	adde %0,%0,%3 \n\
76 	addze %0,%0 \n\
77 	"
78 	: "=r" (sum)
79 	: "r" (daddr), "r"(saddr), "r"(proto + len), "0"(sum));
80 	return sum;
81 #endif
82 }
83 
84 /*
85  * computes the checksum of the TCP/UDP pseudo-header
86  * returns a 16-bit checksum, already complemented
87  */
88 static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr, __u32 len,
89 					__u8 proto, __wsum sum)
90 {
91 	return csum_fold(csum_tcpudp_nofold(saddr, daddr, len, proto, sum));
92 }
93 
94 #define HAVE_ARCH_CSUM_ADD
95 static inline __wsum csum_add(__wsum csum, __wsum addend)
96 {
97 #ifdef __powerpc64__
98 	u64 res = (__force u64)csum;
99 #endif
100 	if (__builtin_constant_p(csum) && csum == 0)
101 		return addend;
102 	if (__builtin_constant_p(addend) && addend == 0)
103 		return csum;
104 
105 #ifdef __powerpc64__
106 	res += (__force u64)addend;
107 	return (__force __wsum)((u32)res + (res >> 32));
108 #else
109 	asm("addc %0,%0,%1;"
110 	    "addze %0,%0;"
111 	    : "+r" (csum) : "r" (addend) : "xer");
112 	return csum;
113 #endif
114 }
115 
116 /*
117  * This is a version of ip_compute_csum() optimized for IP headers,
118  * which always checksum on 4 octet boundaries.  ihl is the number
119  * of 32-bit words and is always >= 5.
120  */
121 static inline __wsum ip_fast_csum_nofold(const void *iph, unsigned int ihl)
122 {
123 	const u32 *ptr = (const u32 *)iph + 1;
124 #ifdef __powerpc64__
125 	unsigned int i;
126 	u64 s = *(const u32 *)iph;
127 
128 	for (i = 0; i < ihl - 1; i++, ptr++)
129 		s += *ptr;
130 	return (__force __wsum)from64to32(s);
131 #else
132 	__wsum sum, tmp;
133 
134 	asm("mtctr %3;"
135 	    "addc %0,%4,%5;"
136 	    "1: lwzu %1, 4(%2);"
137 	    "adde %0,%0,%1;"
138 	    "bdnz 1b;"
139 	    "addze %0,%0;"
140 	    : "=r" (sum), "=r" (tmp), "+b" (ptr)
141 	    : "r" (ihl - 2), "r" (*(const u32 *)iph), "r" (*ptr)
142 	    : "ctr", "xer", "memory");
143 
144 	return sum;
145 #endif
146 }
147 
148 static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
149 {
150 	return csum_fold(ip_fast_csum_nofold(iph, ihl));
151 }
152 
153 /*
154  * computes the checksum of a memory block at buff, length len,
155  * and adds in "sum" (32-bit)
156  *
157  * returns a 32-bit number suitable for feeding into itself
158  * or csum_tcpudp_magic
159  *
160  * this function must be called with even lengths, except
161  * for the last fragment, which may be odd
162  *
163  * it's best to have buff aligned on a 32-bit boundary
164  */
165 __wsum __csum_partial(const void *buff, int len, __wsum sum);
166 
167 static inline __wsum csum_partial(const void *buff, int len, __wsum sum)
168 {
169 	if (__builtin_constant_p(len) && len <= 16 && (len & 1) == 0) {
170 		if (len == 2)
171 			sum = csum_add(sum, (__force __wsum)*(const u16 *)buff);
172 		if (len >= 4)
173 			sum = csum_add(sum, (__force __wsum)*(const u32 *)buff);
174 		if (len == 6)
175 			sum = csum_add(sum, (__force __wsum)
176 					    *(const u16 *)(buff + 4));
177 		if (len >= 8)
178 			sum = csum_add(sum, (__force __wsum)
179 					    *(const u32 *)(buff + 4));
180 		if (len == 10)
181 			sum = csum_add(sum, (__force __wsum)
182 					    *(const u16 *)(buff + 8));
183 		if (len >= 12)
184 			sum = csum_add(sum, (__force __wsum)
185 					    *(const u32 *)(buff + 8));
186 		if (len == 14)
187 			sum = csum_add(sum, (__force __wsum)
188 					    *(const u16 *)(buff + 12));
189 		if (len >= 16)
190 			sum = csum_add(sum, (__force __wsum)
191 					    *(const u32 *)(buff + 12));
192 	} else if (__builtin_constant_p(len) && (len & 3) == 0) {
193 		sum = csum_add(sum, ip_fast_csum_nofold(buff, len >> 2));
194 	} else {
195 		sum = __csum_partial(buff, len, sum);
196 	}
197 	return sum;
198 }
199 
200 /*
201  * this routine is used for miscellaneous IP-like checksums, mainly
202  * in icmp.c
203  */
204 static inline __sum16 ip_compute_csum(const void *buff, int len)
205 {
206 	return csum_fold(csum_partial(buff, len, 0));
207 }
208 
209 #define _HAVE_ARCH_IPV6_CSUM
210 __sum16 csum_ipv6_magic(const struct in6_addr *saddr,
211 			const struct in6_addr *daddr,
212 			__u32 len, __u8 proto, __wsum sum);
213 
214 #endif /* __KERNEL__ */
215 #endif
216