1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
|
// SPDX-License-Identifier: GPL-2.0-only
/*
* Copyright 2017, Gustavo Romero, IBM Corp.
*
* Check if thread endianness is flipped inadvertently to BE on trap
* caught in TM whilst MSR.FP and MSR.VEC are zero (i.e. just after
* load_fp and load_vec overflowed).
*
* The issue can be checked on LE machines simply by zeroing load_fp
* and load_vec and then causing a trap in TM. Since the endianness
* changes to BE on return from the signal handler, 'nop' is
* thread as an illegal instruction in following sequence:
* tbegin.
* beq 1f
* trap
* tend.
* 1: nop
*
* However, although the issue is also present on BE machines, it's a
* bit trickier to check it on BE machines because MSR.LE bit is set
* to zero which determines a BE endianness that is the native
* endianness on BE machines, so nothing notably critical happens,
* i.e. no illegal instruction is observed immediately after returning
* from the signal handler (as it happens on LE machines). Thus to test
* it on BE machines LE endianness is forced after a first trap and then
* the endianness is verified on subsequent traps to determine if the
* endianness "flipped back" to the native endianness (BE).
*/
#define _GNU_SOURCE
#include <error.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <htmintrin.h>
#include <inttypes.h>
#include <pthread.h>
#include <sched.h>
#include <signal.h>
#include <stdbool.h>
#include "tm.h"
#include "utils.h"
#define pr_error(error_code, format, ...) \
error_at_line(1, error_code, __FILE__, __LINE__, format, ##__VA_ARGS__)
#define MSR_LE 1UL
#define LE 1UL
pthread_t t0_ping;
pthread_t t1_pong;
int exit_from_pong;
int trap_event;
int le;
bool success;
void trap_signal_handler(int signo, siginfo_t *si, void *uc)
{
ucontext_t *ucp = uc;
uint64_t thread_endianness;
/* Get thread endianness: extract bit LE from MSR */
thread_endianness = MSR_LE & ucp->uc_mcontext.gp_regs[PT_MSR];
/***
* Little-Endian Machine
*/
if (le) {
/* First trap event */
if (trap_event == 0) {
/* Do nothing. Since it is returning from this trap
* event that endianness is flipped by the bug, so just
* let the process return from the signal handler and
* check on the second trap event if endianness is
* flipped or not.
*/
}
/* Second trap event */
else if (trap_event == 1) {
/*
* Since trap was caught in TM on first trap event, if
* endianness was still LE (not flipped inadvertently)
* after returning from the signal handler instruction
* (1) is executed (basically a 'nop'), as it's located
* at address of tbegin. +4 (rollback addr). As (1) on
* LE endianness does in effect nothing, instruction (2)
* is then executed again as 'trap', generating a second
* trap event (note that in that case 'trap' is caught
* not in transacional mode). On te other hand, if after
* the return from the signal handler the endianness in-
* advertently flipped, instruction (1) is tread as a
* branch instruction, i.e. b .+8, hence instruction (3)
* and (4) are executed (tbegin.; trap;) and we get sim-
* ilaly on the trap signal handler, but now in TM mode.
* Either way, it's now possible to check the MSR LE bit
* once in the trap handler to verify if endianness was
* flipped or not after the return from the second trap
* event. If endianness is flipped, the bug is present.
* Finally, getting a trap in TM mode or not is just
* worth noting because it affects the math to determine
* the offset added to the NIP on return: the NIP for a
* trap caught in TM is the rollback address, i.e. the
* next instruction after 'tbegin.', whilst the NIP for
* a trap caught in non-transactional mode is the very
* same address of the 'trap' instruction that generated
* the trap event.
*/
if (thread_endianness == LE) {
/* Go to 'success', i.e. instruction (6) */
ucp->uc_mcontext.gp_regs[PT_NIP] += 16;
} else {
/*
* Thread endianness is BE, so it flipped
* inadvertently. Thus we flip back to LE and
* set NIP to go to 'failure', instruction (5).
*/
ucp->uc_mcontext.gp_regs[PT_MSR] |= 1UL;
ucp->uc_mcontext.gp_regs[PT_NIP] += 4;
}
}
}
/***
* Big-Endian Machine
*/
else {
/* First trap event */
if (trap_event == 0) {
/*
* Force thread endianness to be LE. Instructions (1),
* (3), and (4) will be executed, generating a second
* trap in TM mode.
*/
ucp->uc_mcontext.gp_regs[PT_MSR] |= 1UL;
}
/* Second trap event */
else if (trap_event == 1) {
/*
* Do nothing. If bug is present on return from this
* second trap event endianness will flip back "automat-
* ically" to BE, otherwise thread endianness will
* continue to be LE, just as it was set above.
*/
}
/* A third trap event */
else {
/*
* Once here it means that after returning from the sec-
* ond trap event instruction (4) (trap) was executed
* as LE, generating a third trap event. In that case
* endianness is still LE as set on return from the
* first trap event, hence no bug. Otherwise, bug
* flipped back to BE on return from the second trap
* event and instruction (4) was executed as 'tdi' (so
* basically a 'nop') and branch to 'failure' in
* instruction (5) was taken to indicate failure and we
* never get here.
*/
/*
* Flip back to BE and go to instruction (6), i.e. go to
* 'success'.
*/
ucp->uc_mcontext.gp_regs[PT_MSR] &= ~1UL;
ucp->uc_mcontext.gp_regs[PT_NIP] += 8;
}
}
trap_event++;
}
void usr1_signal_handler(int signo, siginfo_t *si, void *not_used)
{
/* Got a USR1 signal from ping(), so just tell pong() to exit */
exit_from_pong = 1;
}
void *ping(void *not_used)
{
uint64_t i;
trap_event = 0;
/*
* Wait an amount of context switches so load_fp and load_vec overflows
* and MSR_[FP|VEC|V] is 0.
*/
for (i = 0; i < 1024*1024*512; i++)
;
asm goto(
/*
* [NA] means "Native Endianness", i.e. it tells how a
* instruction is executed on machine's native endianness (in
* other words, native endianness matches kernel endianness).
* [OP] means "Opposite Endianness", i.e. on a BE machine, it
* tells how a instruction is executed as a LE instruction; con-
* versely, on a LE machine, it tells how a instruction is
* executed as a BE instruction. When [NA] is omitted, it means
* that the native interpretation of a given instruction is not
* relevant for the test. Likewise when [OP] is omitted.
*/
" tbegin. ;" /* (0) tbegin. [NA] */
" tdi 0, 0, 0x48;" /* (1) nop [NA]; b (3) [OP] */
" trap ;" /* (2) trap [NA] */
".long 0x1D05007C;" /* (3) tbegin. [OP] */
".long 0x0800E07F;" /* (4) trap [OP]; nop [NA] */
" b %l[failure] ;" /* (5) b [NA]; MSR.LE flipped (bug) */
" b %l[success] ;" /* (6) b [NA]; MSR.LE did not flip (ok)*/
: : : : failure, success);
failure:
success = false;
goto exit_from_ping;
success:
success = true;
exit_from_ping:
/* Tell pong() to exit before leaving */
pthread_kill(t1_pong, SIGUSR1);
return NULL;
}
void *pong(void *not_used)
{
while (!exit_from_pong)
/*
* Induce context switches on ping() thread
* until ping() finishes its job and signs
* to exit from this loop.
*/
sched_yield();
return NULL;
}
int tm_trap_test(void)
{
uint16_t k = 1;
int cpu, rc;
pthread_attr_t attr;
cpu_set_t cpuset;
struct sigaction trap_sa;
SKIP_IF(!have_htm());
trap_sa.sa_flags = SA_SIGINFO;
trap_sa.sa_sigaction = trap_signal_handler;
sigaction(SIGTRAP, &trap_sa, NULL);
struct sigaction usr1_sa;
usr1_sa.sa_flags = SA_SIGINFO;
usr1_sa.sa_sigaction = usr1_signal_handler;
sigaction(SIGUSR1, &usr1_sa, NULL);
cpu = pick_online_cpu();
FAIL_IF(cpu < 0);
// Set only one CPU in the mask. Both threads will be bound to that CPU.
CPU_ZERO(&cpuset);
CPU_SET(cpu, &cpuset);
/* Init pthread attribute */
rc = pthread_attr_init(&attr);
if (rc)
pr_error(rc, "pthread_attr_init()");
/*
* Bind thread ping() and pong() both to CPU 0 so they ping-pong and
* speed up context switches on ping() thread, speeding up the load_fp
* and load_vec overflow.
*/
rc = pthread_attr_setaffinity_np(&attr, sizeof(cpu_set_t), &cpuset);
if (rc)
pr_error(rc, "pthread_attr_setaffinity()");
/* Figure out the machine endianness */
le = (int) *(uint8_t *)&k;
printf("%s machine detected. Checking if endianness flips %s",
le ? "Little-Endian" : "Big-Endian",
"inadvertently on trap in TM... ");
rc = fflush(0);
if (rc)
pr_error(rc, "fflush()");
/* Launch ping() */
rc = pthread_create(&t0_ping, &attr, ping, NULL);
if (rc)
pr_error(rc, "pthread_create()");
exit_from_pong = 0;
/* Launch pong() */
rc = pthread_create(&t1_pong, &attr, pong, NULL);
if (rc)
pr_error(rc, "pthread_create()");
rc = pthread_join(t0_ping, NULL);
if (rc)
pr_error(rc, "pthread_join()");
rc = pthread_join(t1_pong, NULL);
if (rc)
pr_error(rc, "pthread_join()");
if (success) {
printf("no.\n"); /* no, endianness did not flip inadvertently */
return EXIT_SUCCESS;
}
printf("yes!\n"); /* yes, endianness did flip inadvertently */
return EXIT_FAILURE;
}
int main(int argc, char **argv)
{
return test_harness(tm_trap_test, "tm_trap_test");
}
|