cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

checksum.h (5865B)


      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 */
     21extern __wsum csum_partial_copy_generic(const void *src, void *dst, int len);
     22
     23#define _HAVE_ARCH_COPY_AND_CSUM_FROM_USER
     24extern __wsum csum_and_copy_from_user(const void __user *src, void *dst,
     25				      int len);
     26#define HAVE_CSUM_COPY_USER
     27extern __wsum csum_and_copy_to_user(const void *src, void __user *dst,
     28				    int len);
     29
     30#define _HAVE_ARCH_CSUM_AND_COPY
     31#define csum_partial_copy_nocheck(src, dst, len)   \
     32        csum_partial_copy_generic((src), (dst), (len))
     33
     34
     35/*
     36 * turns a 32-bit partial checksum (e.g. from csum_partial) into a
     37 * 1's complement 16-bit checksum.
     38 */
     39static inline __sum16 csum_fold(__wsum sum)
     40{
     41	u32 tmp = (__force u32)sum;
     42
     43	/*
     44	 * swap the two 16-bit halves of sum
     45	 * if there is a carry from adding the two 16-bit halves,
     46	 * it will carry from the lower half into the upper half,
     47	 * giving us the correct sum in the upper half.
     48	 */
     49	return (__force __sum16)(~(tmp + rol32(tmp, 16)) >> 16);
     50}
     51
     52static inline u32 from64to32(u64 x)
     53{
     54	return (x + ror64(x, 32)) >> 32;
     55}
     56
     57static 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 */
     88static 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
     95static __always_inline __wsum csum_add(__wsum csum, __wsum addend)
     96{
     97#ifdef __powerpc64__
     98	u64 res = (__force u64)csum;
     99
    100	res += (__force u64)addend;
    101	return (__force __wsum)((u32)res + (res >> 32));
    102#else
    103	if (__builtin_constant_p(csum) && csum == 0)
    104		return addend;
    105	if (__builtin_constant_p(addend) && addend == 0)
    106		return csum;
    107
    108	asm("addc %0,%0,%1;"
    109	    "addze %0,%0;"
    110	    : "+r" (csum) : "r" (addend) : "xer");
    111	return csum;
    112#endif
    113}
    114
    115#define HAVE_ARCH_CSUM_SHIFT
    116static __always_inline __wsum csum_shift(__wsum sum, int offset)
    117{
    118	/* rotate sum to align it with a 16b boundary */
    119	return (__force __wsum)rol32((__force u32)sum, (offset & 1) << 3);
    120}
    121
    122/*
    123 * This is a version of ip_compute_csum() optimized for IP headers,
    124 * which always checksum on 4 octet boundaries.  ihl is the number
    125 * of 32-bit words and is always >= 5.
    126 */
    127static inline __wsum ip_fast_csum_nofold(const void *iph, unsigned int ihl)
    128{
    129	const u32 *ptr = (const u32 *)iph + 1;
    130#ifdef __powerpc64__
    131	unsigned int i;
    132	u64 s = *(const u32 *)iph;
    133
    134	for (i = 0; i < ihl - 1; i++, ptr++)
    135		s += *ptr;
    136	return (__force __wsum)from64to32(s);
    137#else
    138	__wsum sum, tmp;
    139
    140	asm("mtctr %3;"
    141	    "addc %0,%4,%5;"
    142	    "1: lwzu %1, 4(%2);"
    143	    "adde %0,%0,%1;"
    144	    "bdnz 1b;"
    145	    "addze %0,%0;"
    146	    : "=r" (sum), "=r" (tmp), "+b" (ptr)
    147	    : "r" (ihl - 2), "r" (*(const u32 *)iph), "r" (*ptr)
    148	    : "ctr", "xer", "memory");
    149
    150	return sum;
    151#endif
    152}
    153
    154static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
    155{
    156	return csum_fold(ip_fast_csum_nofold(iph, ihl));
    157}
    158
    159/*
    160 * computes the checksum of a memory block at buff, length len,
    161 * and adds in "sum" (32-bit)
    162 *
    163 * returns a 32-bit number suitable for feeding into itself
    164 * or csum_tcpudp_magic
    165 *
    166 * this function must be called with even lengths, except
    167 * for the last fragment, which may be odd
    168 *
    169 * it's best to have buff aligned on a 32-bit boundary
    170 */
    171__wsum __csum_partial(const void *buff, int len, __wsum sum);
    172
    173static __always_inline __wsum csum_partial(const void *buff, int len, __wsum sum)
    174{
    175	if (__builtin_constant_p(len) && len <= 16 && (len & 1) == 0) {
    176		if (len == 2)
    177			sum = csum_add(sum, (__force __wsum)*(const u16 *)buff);
    178		if (len >= 4)
    179			sum = csum_add(sum, (__force __wsum)*(const u32 *)buff);
    180		if (len == 6)
    181			sum = csum_add(sum, (__force __wsum)
    182					    *(const u16 *)(buff + 4));
    183		if (len >= 8)
    184			sum = csum_add(sum, (__force __wsum)
    185					    *(const u32 *)(buff + 4));
    186		if (len == 10)
    187			sum = csum_add(sum, (__force __wsum)
    188					    *(const u16 *)(buff + 8));
    189		if (len >= 12)
    190			sum = csum_add(sum, (__force __wsum)
    191					    *(const u32 *)(buff + 8));
    192		if (len == 14)
    193			sum = csum_add(sum, (__force __wsum)
    194					    *(const u16 *)(buff + 12));
    195		if (len >= 16)
    196			sum = csum_add(sum, (__force __wsum)
    197					    *(const u32 *)(buff + 12));
    198	} else if (__builtin_constant_p(len) && (len & 3) == 0) {
    199		sum = csum_add(sum, ip_fast_csum_nofold(buff, len >> 2));
    200	} else {
    201		sum = __csum_partial(buff, len, sum);
    202	}
    203	return sum;
    204}
    205
    206/*
    207 * this routine is used for miscellaneous IP-like checksums, mainly
    208 * in icmp.c
    209 */
    210static inline __sum16 ip_compute_csum(const void *buff, int len)
    211{
    212	return csum_fold(csum_partial(buff, len, 0));
    213}
    214
    215#define _HAVE_ARCH_IPV6_CSUM
    216__sum16 csum_ipv6_magic(const struct in6_addr *saddr,
    217			const struct in6_addr *daddr,
    218			__u32 len, __u8 proto, __wsum sum);
    219
    220#endif /* __KERNEL__ */
    221#endif