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
|
#pragma once
#include "asm.h"
#include "cache_types.h"
#include "util.h"
#include "cachepc_user.h"
void cachepc_init_pmc(uint8_t index, uint8_t event_no, uint8_t event_mask);
cache_ctx *cachepc_get_ctx(cache_level cl);
void cachepc_release_ctx(cache_ctx *ctx);
cacheline *cachepc_prepare_ds(cache_ctx *ctx);
void cachepc_release_ds(cache_ctx *ctx, cacheline *ds);
cacheline *cachepc_prepare_victim(cache_ctx *ctx, uint32_t set);
void cachepc_release_victim(cache_ctx *ctx, cacheline *ptr);
void *cachepc_aligned_alloc(size_t alignment, size_t size);
void cachepc_save_msrmts(cacheline *head);
void cachepc_print_msrmts(cacheline *head);
__attribute__((always_inline))
static inline cacheline *cachepc_prime(cacheline *head);
__attribute__((always_inline))
static inline cacheline *cachepc_prime_rev(cacheline *head);
__attribute__((always_inline))
static inline cacheline *cachepc_probe(cacheline *head);
__attribute__((always_inline))
static inline void cachepc_victim(void *p);
__attribute__((always_inline))
static inline uint64_t cachepc_read_pmc(uint64_t event);
extern uint16_t *cachepc_msrmts;
extern size_t cachepc_msrmts_count;
extern cache_ctx *cachepc_ctx;
extern cacheline *cachepc_ds;
/*
* Prime phase: fill the target cache (encoded in the size of the data structure)
* with the prepared data structure, i.e. with attacker data.
*/
cacheline *
cachepc_prime(cacheline *head)
{
cacheline *curr_cl, *prev_cl;
cachepc_mfence();
cachepc_cpuid();
curr_cl = head;
do {
prev_cl = curr_cl;
curr_cl = curr_cl->next;
} while (curr_cl != head);
cachepc_mfence();
cachepc_cpuid();
return prev_cl;
}
/*
* Same as prime, but in the reverse direction, i.e. the same direction that probe
* uses. This is beneficial for the following scenarios:
* - L1:
* - Trigger collision chain-reaction to amplify an evicted set (but this has
* the downside of more noisy measurements).
* - L2:
* - Always use this for L2, otherwise the first cache sets will still reside
* in L1 unless the victim filled L1 completely. In this case, an eviction
* has randomly (depending on where the cache set is placed in the randomised
* data structure) the following effect:
* A) An evicted set is L2_ACCESS_TIME - L1_ACCESS_TIME slower
* B) An evicted set is L3_ACCESS_TIME - L2_ACCESS_TIME slower
*/
cacheline *
cachepc_prime_rev(cacheline *head)
{
cacheline *curr_cl;
cachepc_mfence();
cachepc_cpuid();
curr_cl = head;
do {
curr_cl = curr_cl->prev;
} while(curr_cl != head);
cachepc_mfence();
cachepc_cpuid();
return curr_cl->prev;
}
cacheline *
cachepc_probe(cacheline *start_cl)
{
uint64_t pre, post;
cacheline *next_cl;
cacheline *curr_cl;
cachepc_mfence();
cachepc_cpuid();
curr_cl = start_cl;
do {
pre = cachepc_read_pmc(0);
asm volatile(
"mov 8(%[curr_cl]), %%rax \n\t" // +8
"mov 8(%%rax), %%rcx \n\t" // +16
"mov 8(%%rcx), %%rax \n\t" // +24
"mov 8(%%rax), %%rcx \n\t" // +32
"mov 8(%%rcx), %%rax \n\t" // +40
"mov 8(%%rax), %%rcx \n\t" // +48
"mov 8(%%rcx), %[curr_cl_out] \n\t" // +56
"mov 8(%[curr_cl_out]), %[next_cl_out] \n\t" // +64
: [next_cl_out] "=r" (next_cl),
[curr_cl_out] "=r" (curr_cl)
: [curr_cl] "r" (curr_cl)
: "rax", "rcx"
);
post = cachepc_read_pmc(0);
/* works across size boundary */
curr_cl->count = post - pre;
curr_cl = next_cl;
} while (__builtin_expect(curr_cl != start_cl, 1));
next_cl = curr_cl->next;
cachepc_mfence();
cachepc_cpuid();
return next_cl;
}
void
cachepc_victim(void *p)
{
cachepc_mfence();
cachepc_cpuid();
cachepc_readq(p);
cachepc_mfence();
cachepc_cpuid();
}
uint64_t
cachepc_read_pmc(uint64_t event)
{
uint32_t lo, hi;
uint64_t res;
cachepc_mfence();
cachepc_cpuid();
event = 0xC0010201 + 2 * event;
asm volatile (
"rdmsr"
: "=a" (lo), "=d" (hi)
: "c"(event)
);
res = ((uint64_t) hi << 32) | (uint64_t) lo;
cachepc_mfence();
cachepc_cpuid();
return res;
}
|
