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
|
// See LICENSE for license details.
#ifndef _RISCV_MMU_H
#define _RISCV_MMU_H
#include "decode.h"
#include "icache.h"
#include "trap.h"
#include "common.h"
#include "config.h"
#include "processor.h"
#include "memtracer.h"
#include <vector>
// virtual memory configuration
typedef reg_t pte_t;
const reg_t LEVELS = sizeof(pte_t) == 8 ? 3 : 2;
const reg_t PTIDXBITS = 10;
const reg_t PGSHIFT = PTIDXBITS + (sizeof(pte_t) == 8 ? 3 : 2);
const reg_t PGSIZE = 1 << PGSHIFT;
const reg_t VPN_BITS = PTIDXBITS * LEVELS;
const reg_t PPN_BITS = 8*sizeof(reg_t) - PGSHIFT;
const reg_t VA_BITS = VPN_BITS + PGSHIFT;
struct insn_fetch_t
{
insn_func_t func;
insn_t insn;
};
struct icache_entry_t {
reg_t tag;
reg_t pad;
insn_fetch_t data;
};
// this class implements a processor's port into the virtual memory system.
// an MMU and instruction cache are maintained for simulator performance.
class mmu_t
{
public:
mmu_t(char* _mem, size_t _memsz);
~mmu_t();
// template for functions that load an aligned value from memory
#define load_func(type) \
type##_t load_##type(reg_t addr) __attribute__((always_inline)) { \
void* paddr = translate(addr, sizeof(type##_t), false, false); \
return *(type##_t*)paddr; \
}
// load value from memory at aligned address; zero extend to register width
load_func(uint8)
load_func(uint16)
load_func(uint32)
load_func(uint64)
// load value from memory at aligned address; sign extend to register width
load_func(int8)
load_func(int16)
load_func(int32)
load_func(int64)
// template for functions that store an aligned value to memory
#define store_func(type) \
void store_##type(reg_t addr, type##_t val) { \
void* paddr = translate(addr, sizeof(type##_t), true, false); \
*(type##_t*)paddr = val; \
}
// store value to memory at aligned address
store_func(uint8)
store_func(uint16)
store_func(uint32)
store_func(uint64)
inline size_t icache_index(reg_t addr)
{
// for instruction sizes != 4, this hash still works but is suboptimal
return (addr / 4) % ICACHE_SIZE;
}
// load instruction from memory at aligned address.
icache_entry_t* access_icache(reg_t addr) __attribute__((always_inline))
{
reg_t idx = icache_index(addr);
icache_entry_t* entry = &icache[idx];
if (likely(entry->tag == addr))
return entry;
char* iaddr = (char*)translate(addr, 2, false, true);
insn_bits_t insn = *(uint16_t*)iaddr;
if (unlikely(insn_length(insn) == 2)) {
insn = (int16_t)insn;
} else if (likely(insn_length(insn) == 4)) {
if (likely((addr & (PGSIZE-1)) < PGSIZE-2))
insn |= (insn_bits_t)*(int16_t*)(iaddr + 2) << 16;
else
insn |= (insn_bits_t)*(int16_t*)translate(addr + 2, 2, false, true) << 16;
} else if (insn_length(insn) == 6) {
insn |= (insn_bits_t)*(int16_t*)translate(addr + 4, 2, false, true) << 32;
insn |= (insn_bits_t)*(uint16_t*)translate(addr + 2, 2, false, true) << 16;
} else {
static_assert(sizeof(insn_bits_t) == 8, "insn_bits_t must be uint64_t");
insn |= (insn_bits_t)*(int16_t*)translate(addr + 6, 2, false, true) << 48;
insn |= (insn_bits_t)*(uint16_t*)translate(addr + 4, 2, false, true) << 32;
insn |= (insn_bits_t)*(uint16_t*)translate(addr + 2, 2, false, true) << 16;
}
insn_fetch_t fetch = {proc->decode_insn(insn), insn};
icache[idx].tag = addr;
icache[idx].data = fetch;
reg_t paddr = iaddr - mem;
if (!tracer.empty() && tracer.interested_in_range(paddr, paddr + 1, false, true))
{
icache[idx].tag = -1;
tracer.trace(paddr, 1, false, true);
}
return &icache[idx];
}
inline insn_fetch_t load_insn(reg_t addr)
{
return access_icache(addr)->data;
}
void set_processor(processor_t* p) { proc = p; flush_tlb(); }
void flush_tlb();
void flush_icache();
void register_memtracer(memtracer_t*);
private:
char* mem;
size_t memsz;
processor_t* proc;
memtracer_list_t tracer;
// implement an instruction cache for simulator performance
icache_entry_t icache[ICACHE_SIZE];
// implement a TLB for simulator performance
static const reg_t TLB_ENTRIES = 256;
char* tlb_data[TLB_ENTRIES];
reg_t tlb_insn_tag[TLB_ENTRIES];
reg_t tlb_load_tag[TLB_ENTRIES];
reg_t tlb_store_tag[TLB_ENTRIES];
// finish translation on a TLB miss and upate the TLB
void* refill_tlb(reg_t addr, reg_t bytes, bool store, bool fetch);
// perform a page table walk for a given virtual address
pte_t walk(reg_t addr);
// translate a virtual address to a physical address
void* translate(reg_t addr, reg_t bytes, bool store, bool fetch)
__attribute__((always_inline))
{
reg_t idx = (addr >> PGSHIFT) % TLB_ENTRIES;
reg_t expected_tag = addr >> PGSHIFT;
reg_t* tags = fetch ? tlb_insn_tag : store ? tlb_store_tag :tlb_load_tag;
reg_t tag = tags[idx];
void* data = tlb_data[idx] + addr;
if (unlikely(addr & (bytes-1)))
store ? throw trap_store_address_misaligned(addr) : throw trap_load_address_misaligned(addr);
if (likely(tag == expected_tag))
return data;
return refill_tlb(addr, bytes, store, fetch);
}
friend class processor_t;
};
#endif
|
