Makefile + hash.h: remove PPC_SHA1 implementation
Remove the PPC_SHA1 implementation added in a6ef3518f9a ([PATCH] PPC assembly implementation of SHA1, 2005-04-22). When this was added Apple consumer hardware used the PPC architecture, and the implementation was intended to improve SHA-1 speed there. Since it was added we've moved to using sha1collisiondetection by default, and anyone wanting hard-rolled non-DC SHA-1 implementation can use OpenSSL's via the OPENSSL_SHA1 knob. The PPC_SHA1 originally originally targeted 32 bit PPC, and later the 64 bit PPC 970 (a.k.a. Apple PowerPC G5). See 926172c5e48 (block-sha1: improve code on large-register-set machines, 2009-08-10) for a reference about the performance on G5 (a comment in block-sha1/sha1.c being removed here). I can't get it to do anything but segfault on both the BE and LE POWER machines in the GCC compile farm[1]. Anyone who's concerned about performance on PPC these days is likely to be using the IBM POWER processors. There have been proposals to entirely remove non-sha1collisiondetection implementations from the tree[2]. I think per [3] that would be a bit overzealous. I.e. there are various set-ups git's speed is going to be more important than the relatively implausible SHA-1 collision attack, or where such attacks are entirely mitigated by other means (e.g. by incoming objects being checked with DC_SHA1). But that really doesn't apply to PPC_SHA1 in particular, which seems to have outlived its usefulness. As this gets rid of the only in-tree *.S assembly file we can remove the small bits of logic from the Makefile needed to build objects from *.S (as opposed to *.c) The code being removed here was also throwing warnings with the "-pedantic" flag, it could have been fixed as 544d93bc3b4 (block-sha1: remove use of obsolete x86 assembly, 2022-03-10) did for block-sha1/*, but as noted above let's remove it instead. 1. https://cfarm.tetaneutral.net/machines/list/ Tested on gcc{110,112,135,203}, a mixture of POWER [789] ppc64 and ppc64le. All segfault in anything needing object hashing (e.g. t/t1007-hash-object.sh) when compiled with PPC_SHA1=Y. 2. https://lore.kernel.org/git/20200223223758.120941-1-mh@glandium.org/ 3. https://lore.kernel.org/git/20200224044732.GK1018190@coredump.intra.peff.net/ Acked-by: brian m. carlson" <sandals@crustytoothpaste.net> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Ævar Arnfjörð Bjarmason committed
Aug 31, 2022 at 11:18 UTC
9dc523aa0e20271cfe1474bef9fafbe62b7ff603
8 files changed
+8
-347
INSTALL
+1
-2
@@ -135,8 +135,7 @@ Issues of note:
135
136
By default, git uses OpenSSL for SHA1 but it will use its own
137
library (inspired by Mozilla's) with either NO_OPENSSL or
138
- BLK_SHA1. Also included is a version optimized for PowerPC
139
- (PPC_SHA1).
138
+ BLK_SHA1.
139
140
- "libcurl" library is used for fetching and pushing
141
repositories over http:// or https://, as well as by
Makefile
+5
-13
@@ -155,9 +155,6 @@ include shared.mak
155
# Define BLK_SHA1 environment variable to make use of the bundled
156
# optimized C SHA1 routine.
157
#
158
-# Define PPC_SHA1 environment variable when running make to make use of
159
-# a bundled SHA1 routine optimized for PowerPC.
160
-#
158
# Define DC_SHA1 to unconditionally enable the collision-detecting sha1
159
# algorithm. This is slower, but may detect attempted collision attacks.
160
# Takes priority over other *_SHA1 knobs.
@@ -1802,6 +1799,10 @@ ifdef APPLE_COMMON_CRYPTO
1799
SHA1_MAX_BLOCK_SIZE = 1024L*1024L*1024L
1800
endif
1801
1802
+ifdef PPC_SHA1
1803
+$(error the PPC_SHA1 flag has been removed along with the PowerPC-specific SHA-1 implementation.)
1804
+endif
1805
+
1806
ifdef OPENSSL_SHA1
1807
EXTLIBS += $(LIB_4_CRYPTO)
1808
BASIC_CFLAGS += -DSHA1_OPENSSL
@@ -1810,10 +1811,6 @@ ifdef BLK_SHA1
1811
LIB_OBJS += block-sha1/sha1.o
1812
BASIC_CFLAGS += -DSHA1_BLK
1813
else
1813
-ifdef PPC_SHA1
1814
- LIB_OBJS += ppc/sha1.o ppc/sha1ppc.o
1815
- BASIC_CFLAGS += -DSHA1_PPC
1816
-else
1814
ifdef APPLE_COMMON_CRYPTO
1815
COMPAT_CFLAGS += -DCOMMON_DIGEST_FOR_OPENSSL
1816
BASIC_CFLAGS += -DSHA1_APPLE
@@ -1847,7 +1844,6 @@ endif
1844
endif
1845
endif
1846
endif
1850
-endif
1847
1848
ifdef OPENSSL_SHA256
1849
EXTLIBS += $(LIB_4_CRYPTO)
@@ -2594,14 +2590,10 @@ missing_compdb_dir =
2590
compdb_args =
2591
endif
2592
2597
-ASM_SRC := $(wildcard $(OBJECTS:o=S))
2598
-ASM_OBJ := $(ASM_SRC:S=o)
2599
-C_OBJ := $(filter-out $(ASM_OBJ),$(OBJECTS))
2593
+C_OBJ := $(OBJECTS)
2594
2595
$(C_OBJ): %.o: %.c GIT-CFLAGS $(missing_dep_dirs) $(missing_compdb_dir)
2596
$(QUIET_CC)$(CC) -o $*.o -c $(dep_args) $(compdb_args) $(ALL_CFLAGS) $(EXTRA_CPPFLAGS) $<
2603
-$(ASM_OBJ): %.o: %.S GIT-CFLAGS $(missing_dep_dirs) $(missing_compdb_dir)
2604
- $(QUIET_CC)$(CC) -o $*.o -c $(dep_args) $(compdb_args) $(ALL_CFLAGS) $(EXTRA_CPPFLAGS) $<
2597
2598
%.s: %.c GIT-CFLAGS FORCE
2599
$(QUIET_CC)$(CC) -o $@ -S $(ALL_CFLAGS) $(EXTRA_CPPFLAGS) $<
block-sha1/sha1.c
-4
@@ -28,10 +28,6 @@
28
* try to do the silly "optimize away loads" part because it won't
29
* see what the value will be).
30
*
31
- * Ben Herrenschmidt reports that on PPC, the C version comes close
32
- * to the optimized asm with this (ie on PPC you don't want that
33
- * 'volatile', since there are lots of registers).
34
- *
31
* On ARM we get the best code generation by forcing a full memory barrier
32
* between each SHA_ROUND, otherwise gcc happily get wild with spilling and
33
* the stack frame size simply explode and performance goes down the drain.
configure.ac
-3
@@ -237,9 +237,6 @@ AC_MSG_NOTICE([CHECKS for site configuration])
237
# tests. These tests take up a significant amount of the total test time
238
# but are not needed unless you plan to talk to SVN repos.
239
#
240
-# Define PPC_SHA1 environment variable when running make to make use of
241
-# a bundled SHA1 routine optimized for PowerPC.
242
-#
240
# Define NO_OPENSSL environment variable if you do not have OpenSSL.
241
#
242
# Define OPENSSLDIR=/foo/bar if your openssl header and library files are in
hash.h
+2
-4
@@ -4,9 +4,7 @@
4
#include "git-compat-util.h"
5
#include "repository.h"
6
7
-#if defined(SHA1_PPC)
8
-#include "ppc/sha1.h"
9
-#elif defined(SHA1_APPLE)
7
+#if defined(SHA1_APPLE)
8
#include <CommonCrypto/CommonDigest.h>
9
#elif defined(SHA1_OPENSSL)
10
#include <openssl/sha.h>
@@ -32,7 +30,7 @@
30
* platform's underlying implementation of SHA-1; could be OpenSSL,
31
* blk_SHA, Apple CommonCrypto, etc... Note that the relevant
32
* SHA-1 header may have already defined platform_SHA_CTX for our
35
- * own implementations like block-sha1 and ppc-sha1, so we list
33
+ * own implementations like block-sha1, so we list
34
* the default for OpenSSL compatible SHA-1 implementations here.
35
*/
36
#define platform_SHA_CTX SHA_CTX
ppc/sha1.c
deleted
-72
@@ -1,72 +0,0 @@
1
-/*
2
- * SHA-1 implementation.
3
- *
4
- * Copyright (C) 2005 Paul Mackerras <paulus@samba.org>
5
- *
6
- * This version assumes we are running on a big-endian machine.
7
- * It calls an external sha1_core() to process blocks of 64 bytes.
8
- */
9
-#include <stdio.h>
10
-#include <string.h>
11
-#include "sha1.h"
12
-
13
-void ppc_sha1_core(uint32_t *hash, const unsigned char *p,
14
- unsigned int nblocks);
15
-
16
-int ppc_SHA1_Init(ppc_SHA_CTX *c)
17
-{
18
- c->hash[0] = 0x67452301;
19
- c->hash[1] = 0xEFCDAB89;
20
- c->hash[2] = 0x98BADCFE;
21
- c->hash[3] = 0x10325476;
22
- c->hash[4] = 0xC3D2E1F0;
23
- c->len = 0;
24
- c->cnt = 0;
25
- return 0;
26
-}
27
-
28
-int ppc_SHA1_Update(ppc_SHA_CTX *c, const void *ptr, unsigned long n)
29
-{
30
- unsigned long nb;
31
- const unsigned char *p = ptr;
32
-
33
- c->len += (uint64_t) n << 3;
34
- while (n != 0) {
35
- if (c->cnt || n < 64) {
36
- nb = 64 - c->cnt;
37
- if (nb > n)
38
- nb = n;
39
- memcpy(&c->buf.b[c->cnt], p, nb);
40
- if ((c->cnt += nb) == 64) {
41
- ppc_sha1_core(c->hash, c->buf.b, 1);
42
- c->cnt = 0;
43
- }
44
- } else {
45
- nb = n >> 6;
46
- ppc_sha1_core(c->hash, p, nb);
47
- nb <<= 6;
48
- }
49
- n -= nb;
50
- p += nb;
51
- }
52
- return 0;
53
-}
54
-
55
-int ppc_SHA1_Final(unsigned char *hash, ppc_SHA_CTX *c)
56
-{
57
- unsigned int cnt = c->cnt;
58
-
59
- c->buf.b[cnt++] = 0x80;
60
- if (cnt > 56) {
61
- if (cnt < 64)
62
- memset(&c->buf.b[cnt], 0, 64 - cnt);
63
- ppc_sha1_core(c->hash, c->buf.b, 1);
64
- cnt = 0;
65
- }
66
- if (cnt < 56)
67
- memset(&c->buf.b[cnt], 0, 56 - cnt);
68
- c->buf.l[7] = c->len;
69
- ppc_sha1_core(c->hash, c->buf.b, 1);
70
- memcpy(hash, c->hash, 20);
71
- return 0;
72
-}
ppc/sha1.h
deleted
-25
@@ -1,25 +0,0 @@
1
-/*
2
- * SHA-1 implementation.
3
- *
4
- * Copyright (C) 2005 Paul Mackerras <paulus@samba.org>
5
- */
6
-#include <stdint.h>
7
-
8
-typedef struct {
9
- uint32_t hash[5];
10
- uint32_t cnt;
11
- uint64_t len;
12
- union {
13
- unsigned char b[64];
14
- uint64_t l[8];
15
- } buf;
16
-} ppc_SHA_CTX;
17
-
18
-int ppc_SHA1_Init(ppc_SHA_CTX *c);
19
-int ppc_SHA1_Update(ppc_SHA_CTX *c, const void *p, unsigned long n);
20
-int ppc_SHA1_Final(unsigned char *hash, ppc_SHA_CTX *c);
21
-
22
-#define platform_SHA_CTX ppc_SHA_CTX
23
-#define platform_SHA1_Init ppc_SHA1_Init
24
-#define platform_SHA1_Update ppc_SHA1_Update
25
-#define platform_SHA1_Final ppc_SHA1_Final
ppc/sha1ppc.S
deleted
-224
@@ -1,224 +0,0 @@
1
-/*
2
- * SHA-1 implementation for PowerPC.
3
- *
4
- * Copyright (C) 2005 Paul Mackerras <paulus@samba.org>
5
- */
6
-
7
-/*
8
- * PowerPC calling convention:
9
- * %r0 - volatile temp
10
- * %r1 - stack pointer.
11
- * %r2 - reserved
12
- * %r3-%r12 - Incoming arguments & return values; volatile.
13
- * %r13-%r31 - Callee-save registers
14
- * %lr - Return address, volatile
15
- * %ctr - volatile
16
- *
17
- * Register usage in this routine:
18
- * %r0 - temp
19
- * %r3 - argument (pointer to 5 words of SHA state)
20
- * %r4 - argument (pointer to data to hash)
21
- * %r5 - Constant K in SHA round (initially number of blocks to hash)
22
- * %r6-%r10 - Working copies of SHA variables A..E (actually E..A order)
23
- * %r11-%r26 - Data being hashed W[].
24
- * %r27-%r31 - Previous copies of A..E, for final add back.
25
- * %ctr - loop count
26
- */
27
-
28
-
29
-/*
30
- * We roll the registers for A, B, C, D, E around on each
31
- * iteration; E on iteration t is D on iteration t+1, and so on.
32
- * We use registers 6 - 10 for this. (Registers 27 - 31 hold
33
- * the previous values.)
34
- */
35
-#define RA(t) (((t)+4)%5+6)
36
-#define RB(t) (((t)+3)%5+6)
37
-#define RC(t) (((t)+2)%5+6)
38
-#define RD(t) (((t)+1)%5+6)
39
-#define RE(t) (((t)+0)%5+6)
40
-
41
-/* We use registers 11 - 26 for the W values */
42
-#define W(t) ((t)%16+11)
43
-
44
-/* Register 5 is used for the constant k */
45
-
46
-/*
47
- * The basic SHA-1 round function is:
48
- * E += ROTL(A,5) + F(B,C,D) + W[i] + K; B = ROTL(B,30)
49
- * Then the variables are renamed: (A,B,C,D,E) = (E,A,B,C,D).
50
- *
51
- * Every 20 rounds, the function F() and the constant K changes:
52
- * - 20 rounds of f0(b,c,d) = "bit wise b ? c : d" = (^b & d) + (b & c)
53
- * - 20 rounds of f1(b,c,d) = b^c^d = (b^d)^c
54
- * - 20 rounds of f2(b,c,d) = majority(b,c,d) = (b&d) + ((b^d)&c)
55
- * - 20 more rounds of f1(b,c,d)
56
- *
57
- * These are all scheduled for near-optimal performance on a G4.
58
- * The G4 is a 3-issue out-of-order machine with 3 ALUs, but it can only
59
- * *consider* starting the oldest 3 instructions per cycle. So to get
60
- * maximum performance out of it, you have to treat it as an in-order
61
- * machine. Which means interleaving the computation round t with the
62
- * computation of W[t+4].
63
- *
64
- * The first 16 rounds use W values loaded directly from memory, while the
65
- * remaining 64 use values computed from those first 16. We preload
66
- * 4 values before starting, so there are three kinds of rounds:
67
- * - The first 12 (all f0) also load the W values from memory.
68
- * - The next 64 compute W(i+4) in parallel. 8*f0, 20*f1, 20*f2, 16*f1.
69
- * - The last 4 (all f1) do not do anything with W.
70
- *
71
- * Therefore, we have 6 different round functions:
72
- * STEPD0_LOAD(t,s) - Perform round t and load W(s). s < 16
73
- * STEPD0_UPDATE(t,s) - Perform round t and compute W(s). s >= 16.
74
- * STEPD1_UPDATE(t,s)
75
- * STEPD2_UPDATE(t,s)
76
- * STEPD1(t) - Perform round t with no load or update.
77
- *
78
- * The G5 is more fully out-of-order, and can find the parallelism
79
- * by itself. The big limit is that it has a 2-cycle ALU latency, so
80
- * even though it's 2-way, the code has to be scheduled as if it's
81
- * 4-way, which can be a limit. To help it, we try to schedule the
82
- * read of RA(t) as late as possible so it doesn't stall waiting for
83
- * the previous round's RE(t-1), and we try to rotate RB(t) as early
84
- * as possible while reading RC(t) (= RB(t-1)) as late as possible.
85
- */
86
-
87
-/* the initial loads. */
88
-#define LOADW(s) \
89
- lwz W(s),(s)*4(%r4)
90
-
91
-/*
92
- * Perform a step with F0, and load W(s). Uses W(s) as a temporary
93
- * before loading it.
94
- * This is actually 10 instructions, which is an awkward fit.
95
- * It can execute grouped as listed, or delayed one instruction.
96
- * (If delayed two instructions, there is a stall before the start of the
97
- * second line.) Thus, two iterations take 7 cycles, 3.5 cycles per round.
98
- */
99
-#define STEPD0_LOAD(t,s) \
100
-add RE(t),RE(t),W(t); andc %r0,RD(t),RB(t); and W(s),RC(t),RB(t); \
101
-add RE(t),RE(t),%r0; rotlwi %r0,RA(t),5; rotlwi RB(t),RB(t),30; \
102
-add RE(t),RE(t),W(s); add %r0,%r0,%r5; lwz W(s),(s)*4(%r4); \
103
-add RE(t),RE(t),%r0
104
-
105
-/*
106
- * This is likewise awkward, 13 instructions. However, it can also
107
- * execute starting with 2 out of 3 possible moduli, so it does 2 rounds
108
- * in 9 cycles, 4.5 cycles/round.
109
- */
110
-#define STEPD0_UPDATE(t,s,loadk...) \
111
-add RE(t),RE(t),W(t); andc %r0,RD(t),RB(t); xor W(s),W((s)-16),W((s)-3); \
112
-add RE(t),RE(t),%r0; and %r0,RC(t),RB(t); xor W(s),W(s),W((s)-8); \
113
-add RE(t),RE(t),%r0; rotlwi %r0,RA(t),5; xor W(s),W(s),W((s)-14); \
114
-add RE(t),RE(t),%r5; loadk; rotlwi RB(t),RB(t),30; rotlwi W(s),W(s),1; \
115
-add RE(t),RE(t),%r0
116
-
117
-/* Nicely optimal. Conveniently, also the most common. */
118
-#define STEPD1_UPDATE(t,s,loadk...) \
119
-add RE(t),RE(t),W(t); xor %r0,RD(t),RB(t); xor W(s),W((s)-16),W((s)-3); \
120
-add RE(t),RE(t),%r5; loadk; xor %r0,%r0,RC(t); xor W(s),W(s),W((s)-8); \
121
-add RE(t),RE(t),%r0; rotlwi %r0,RA(t),5; xor W(s),W(s),W((s)-14); \
122
-add RE(t),RE(t),%r0; rotlwi RB(t),RB(t),30; rotlwi W(s),W(s),1
123
-
124
-/*
125
- * The naked version, no UPDATE, for the last 4 rounds. 3 cycles per.
126
- * We could use W(s) as a temp register, but we don't need it.
127
- */
128
-#define STEPD1(t) \
129
- add RE(t),RE(t),W(t); xor %r0,RD(t),RB(t); \
130
-rotlwi RB(t),RB(t),30; add RE(t),RE(t),%r5; xor %r0,%r0,RC(t); \
131
-add RE(t),RE(t),%r0; rotlwi %r0,RA(t),5; /* spare slot */ \
132
-add RE(t),RE(t),%r0
133
-
134
-/*
135
- * 14 instructions, 5 cycles per. The majority function is a bit
136
- * awkward to compute. This can execute with a 1-instruction delay,
137
- * but it causes a 2-instruction delay, which triggers a stall.
138
- */
139
-#define STEPD2_UPDATE(t,s,loadk...) \
140
-add RE(t),RE(t),W(t); and %r0,RD(t),RB(t); xor W(s),W((s)-16),W((s)-3); \
141
-add RE(t),RE(t),%r0; xor %r0,RD(t),RB(t); xor W(s),W(s),W((s)-8); \
142
-add RE(t),RE(t),%r5; loadk; and %r0,%r0,RC(t); xor W(s),W(s),W((s)-14); \
143
-add RE(t),RE(t),%r0; rotlwi %r0,RA(t),5; rotlwi W(s),W(s),1; \
144
-add RE(t),RE(t),%r0; rotlwi RB(t),RB(t),30
145
-
146
-#define STEP0_LOAD4(t,s) \
147
- STEPD0_LOAD(t,s); \
148
- STEPD0_LOAD((t+1),(s)+1); \
149
- STEPD0_LOAD((t)+2,(s)+2); \
150
- STEPD0_LOAD((t)+3,(s)+3)
151
-
152
-#define STEPUP4(fn, t, s, loadk...) \
153
- STEP##fn##_UPDATE(t,s,); \
154
- STEP##fn##_UPDATE((t)+1,(s)+1,); \
155
- STEP##fn##_UPDATE((t)+2,(s)+2,); \
156
- STEP##fn##_UPDATE((t)+3,(s)+3,loadk)
157
-
158
-#define STEPUP20(fn, t, s, loadk...) \
159
- STEPUP4(fn, t, s,); \
160
- STEPUP4(fn, (t)+4, (s)+4,); \
161
- STEPUP4(fn, (t)+8, (s)+8,); \
162
- STEPUP4(fn, (t)+12, (s)+12,); \
163
- STEPUP4(fn, (t)+16, (s)+16, loadk)
164
-
165
- .globl ppc_sha1_core
166
-ppc_sha1_core:
167
- stwu %r1,-80(%r1)
168
- stmw %r13,4(%r1)
169
-
170
- /* Load up A - E */
171
- lmw %r27,0(%r3)
172
-
173
- mtctr %r5
174
-
175
-1:
176
- LOADW(0)
177
- lis %r5,0x5a82
178
- mr RE(0),%r31
179
- LOADW(1)
180
- mr RD(0),%r30
181
- mr RC(0),%r29
182
- LOADW(2)
183
- ori %r5,%r5,0x7999 /* K0-19 */
184
- mr RB(0),%r28
185
- LOADW(3)
186
- mr RA(0),%r27
187
-
188
- STEP0_LOAD4(0, 4)
189
- STEP0_LOAD4(4, 8)
190
- STEP0_LOAD4(8, 12)
191
- STEPUP4(D0, 12, 16,)
192
- STEPUP4(D0, 16, 20, lis %r5,0x6ed9)
193
-
194
- ori %r5,%r5,0xeba1 /* K20-39 */
195
- STEPUP20(D1, 20, 24, lis %r5,0x8f1b)
196
-
197
- ori %r5,%r5,0xbcdc /* K40-59 */
198
- STEPUP20(D2, 40, 44, lis %r5,0xca62)
199
-
200
- ori %r5,%r5,0xc1d6 /* K60-79 */
201
- STEPUP4(D1, 60, 64,)
202
- STEPUP4(D1, 64, 68,)
203
- STEPUP4(D1, 68, 72,)
204
- STEPUP4(D1, 72, 76,)
205
- addi %r4,%r4,64
206
- STEPD1(76)
207
- STEPD1(77)
208
- STEPD1(78)
209
- STEPD1(79)
210
-
211
- /* Add results to original values */
212
- add %r31,%r31,RE(0)
213
- add %r30,%r30,RD(0)
214
- add %r29,%r29,RC(0)
215
- add %r28,%r28,RB(0)
216
- add %r27,%r27,RA(0)
217
-
218
- bdnz 1b
219
-
220
- /* Save final hash, restore registers, and return */
221
- stmw %r27,0(%r3)
222
- lmw %r13,4(%r1)
223
- addi %r1,%r1,80
224
- blr