|
|
|
|
@ -125,6 +125,7 @@ Score: 20/20
|
|
|
|
|
|
|
|
|
|
其次，我们现在管理的可用内存是那一部分? 代理内核的本质也是一段程序，他本身是需要内存空间的，而这一段空间自然不能再被分配。除去内核本身占的空间，内核可支配的物理空间从0x80016000开始，大小在PKE设定为8M，2048个页面，故而供内核支配的的内存的范围为（first_free_page~first_free_paddr）。如下图所示。
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
KERNTOP------->+---------------------------------+ 0x80816000
|
|
|
|
|
|
|
|
|
|
(first_free_paddr) | |
|
|
|
|
|
@ -166,22 +167,20 @@ kernel/user
|
|
|
|
|
| |
|
|
|
|
|
|
|
|
|
|
first_free_paddr->+------------------------------------------------------+ 0x80816000
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
最后，我们来看物理内存分配的单位：操作系统中，物理页是物理内存分配的基本单位。一个物理页的大小是4KB，我们使用结构体Page来表示，其结构如图：
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
struct Page {
|
|
|
|
|
|
|
|
|
|
sint_t ref;
|
|
|
|
|
|
|
|
|
|
uint_t flags;
|
|
|
|
|
|
|
|
|
|
uint_t property;
|
|
|
|
|
|
|
|
|
|
list_entry_t page_link;
|
|
|
|
|
|
|
|
|
|
};
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
l ref表示这样页被页表的引用记数
|
|
|
|
|
|
|
|
|
|
@ -193,11 +192,11 @@ l page_link是维持空闲物理页链表的重要结构。
|
|
|
|
|
|
|
|
|
|
Page结构体对应着物理页，我们来看Page结构体同物理地址之间是如何转换的。首先，我们需要先了解一下物理地址。
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<img src="pictures/fig4_1.png" alt="fig4_1" style="zoom:80%;" />
|
|
|
|
|
|
|
|
|
|
图4.1 RISCV64 物理地址
|
|
|
|
|
图4.1 RISCV64 物理地址
|
|
|
|
|
|
|
|
|
|
总的来说，物理地址分为两部分，页号（PPN）+offset
|
|
|
|
|
总的来说，物理地址分为两部分：页号（PPN）和offset
|
|
|
|
|
|
|
|
|
|
页号可以理解为物理页的编码，而offset则为页内偏移量。现在考虑一下12位的offset对应的内存大小是多少呢？
|
|
|
|
|
|
|
|
|
|
@ -207,7 +206,9 @@ Page结构体对应着物理页，我们来看Page结构体同物理地址之间
|
|
|
|
|
|
|
|
|
|
实际上在初始化空闲页链表之前，系统会定义一个Page结构体的数组，而链表的节点也正是来自于这些数组，这个数组的每一项代表着一个物理页，而且它们的数组下标就代表着每一项具体代表的是哪一个物理页，就如下图所示：
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<img src="pictures/fig4_2.png" alt="fig4_2" style="zoom:80%;" />
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@ -217,65 +218,43 @@ Page结构体对应着物理页，我们来看Page结构体同物理地址之间
|
|
|
|
|
|
|
|
|
|
在PK的machine/minit.c中间中，便通过delegate_traps()，将部分中断及同步异常委托给S模式。（同学们可以查看具体是哪些中断及同步异常）
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
43 // send S-mode interrupts and most exceptions straight to S-mode
|
|
|
|
|
|
|
|
|
|
44 static void delegate_traps()
|
|
|
|
|
|
|
|
|
|
45 {
|
|
|
|
|
|
|
|
|
|
46 if (!supports_extension('S'))
|
|
|
|
|
|
|
|
|
|
47 return;
|
|
|
|
|
|
|
|
|
|
48
|
|
|
|
|
|
|
|
|
|
49 uintptr_t interrupts = MIP_SSIP | MIP_STIP | MIP_SEIP;
|
|
|
|
|
|
|
|
|
|
50 uintptr_t exceptions =
|
|
|
|
|
|
|
|
|
|
51 (1U << CAUSE_MISALIGNED_FETCH) |
|
|
|
|
|
|
|
|
|
|
52 (1U << CAUSE_FETCH_PAGE_FAULT) |
|
|
|
|
|
|
|
|
|
|
53 (1U << CAUSE_BREAKPOINT) |
|
|
|
|
|
|
|
|
|
|
54 (1U << CAUSE_LOAD_PAGE_FAULT) |
|
|
|
|
|
|
|
|
|
|
55 (1U << CAUSE_STORE_PAGE_FAULT) |
|
|
|
|
|
|
|
|
|
|
56 (1U << CAUSE_USER_ECALL);
|
|
|
|
|
|
|
|
|
|
57
|
|
|
|
|
|
|
|
|
|
58 write_csr(mideleg, interrupts);
|
|
|
|
|
|
|
|
|
|
59 write_csr(medeleg, exceptions);
|
|
|
|
|
|
|
|
|
|
60 assert(read_csr(mideleg) == interrupts);
|
|
|
|
|
|
|
|
|
|
61 assert(read_csr(medeleg) == exceptions);
|
|
|
|
|
|
|
|
|
|
62 }
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
这里介绍一下RISCV的中断委托机制，在默认的情况下，所有的异常都会被交由机器模式处理。但正如我们知道的那样，大部分的系统调用都是在S模式下处理的，因此RISCV提供了这一委托机制，可以选择性的将中断交由S模式处理，从而完全绕过M模式。
|
|
|
|
|
|
|
|
|
|
接下，我们继续看S模式下的中断处理。在pk目录下的pk.c文件中的boot_loader函数中将&trap_entry写入了stvec寄存器中，stvec保存着发生异常时处理器需要跳转到的地址，也就是说当中断发生，我们将跳转至trap_entry，现在我们继续跟踪trap_entry。trap_entry在pk目录下的entry.S中，其代码如下：
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
60 trap_entry:
|
|
|
|
|
|
|
|
|
|
61 csrrw sp, sscratch, sp
|
|
|
|
|
|
|
|
|
|
62 bnez sp, 1f
|
|
|
|
|
|
|
|
|
|
63 csrr sp, sscratch
|
|
|
|
|
|
|
|
|
|
64 1:addi sp,sp,-320
|
|
|
|
|
|
|
|
|
|
65 save_tf
|
|
|
|
|
|
|
|
|
|
66 move a0,sp
|
|
|
|
|
|
|
|
|
|
67 jal handle_trap
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
在61行，交换了sp与sscratch的值，这里是为了根据sscratch的值判断该中断是来源于U模式还是S模式。
|
|
|
|
|
|
|
|
|
|
@ -283,148 +262,87 @@ Page结构体对应着物理页，我们来看Page结构体同物理地址之间
|
|
|
|
|
|
|
|
|
|
接着在64,65行保存上下文，最后跳转至67行处理trap。handle_trap在pk目录下的handlers.c文件中，代码如下：
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
112 void handle_trap(trapframe_t* tf)
|
|
|
|
|
|
|
|
|
|
113 {
|
|
|
|
|
|
|
|
|
|
114 if ((intptr_t)tf->cause < 0)
|
|
|
|
|
|
|
|
|
|
115 return handle_interrupt(tf);
|
|
|
|
|
|
|
|
|
|
116
|
|
|
|
|
|
|
|
|
|
117 typedef void (*trap_handler)(trapframe_t*);
|
|
|
|
|
|
|
|
|
|
118
|
|
|
|
|
|
|
|
|
|
119 const static trap_handler trap_handlers[] = {
|
|
|
|
|
|
|
|
|
|
120 [CAUSE_MISALIGNED_FETCH] = handle_misaligned_fetch,
|
|
|
|
|
|
|
|
|
|
121 [CAUSE_FETCH_ACCESS] = handle_instruction_access_fault,
|
|
|
|
|
|
|
|
|
|
122 [CAUSE_LOAD_ACCESS] = handle_load_access_fault,
|
|
|
|
|
|
|
|
|
|
123 [CAUSE_STORE_ACCESS] = handle_store_access_fault,
|
|
|
|
|
|
|
|
|
|
124 [CAUSE_FETCH_PAGE_FAULT] = handle_fault_fetch,
|
|
|
|
|
|
|
|
|
|
125 [CAUSE_ILLEGAL_INSTRUCTION] = handle_illegal_instruction,
|
|
|
|
|
|
|
|
|
|
126 [CAUSE_USER_ECALL] = handle_syscall,
|
|
|
|
|
|
|
|
|
|
127 [CAUSE_BREAKPOINT] = handle_breakpoint,
|
|
|
|
|
|
|
|
|
|
128 [CAUSE_MISALIGNED_LOAD] = handle_misaligned_load,
|
|
|
|
|
|
|
|
|
|
129 [CAUSE_MISALIGNED_STORE] = handle_misaligned_store,
|
|
|
|
|
|
|
|
|
|
130 [CAUSE_LOAD_PAGE_FAULT] = handle_fault_load,
|
|
|
|
|
|
|
|
|
|
131 [CAUSE_STORE_PAGE_FAULT] = handle_fault_store,
|
|
|
|
|
|
|
|
|
|
132 };
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
handle_trap函数中实现了S模式下各类中断的处理。可以看到，代码的126行就对应着系统调用的处理，handle_syscall的实现如下：
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
100 static void handle_syscall(trapframe_t* tf)
|
|
|
|
|
|
|
|
|
|
101 {
|
|
|
|
|
|
|
|
|
|
102 tf->gpr[10] = do_syscall(tf->gpr[10], tf->gpr[11], tf->gpr[12], tf->gpr[13],
|
|
|
|
|
|
|
|
|
|
103 tf->gpr[14], tf->gpr[15], tf->gpr[17]);
|
|
|
|
|
|
|
|
|
|
104 tf->epc += 4;
|
|
|
|
|
|
|
|
|
|
105 }
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
还记得我们在例3.1中是将中断号写入x17寄存器嘛？其对应的就是这里do_syscall的最后一个参数，我们跟踪进入do_syscall函数，其代码如下：
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
313 long do_syscall(long a0, long a1, long a2, long a3, long a4, long a5, unsigned long n)
|
|
|
|
|
|
|
|
|
|
314 {
|
|
|
|
|
|
|
|
|
|
315 const static void* syscall_table[] = {
|
|
|
|
|
|
|
|
|
|
316 // your code here:
|
|
|
|
|
|
|
|
|
|
317 // add get_init_memsize syscall
|
|
|
|
|
|
|
|
|
|
318 [SYS_init_memsize ] = sys_get_init_memsize,
|
|
|
|
|
|
|
|
|
|
319 [SYS_exit] = sys_exit,
|
|
|
|
|
|
|
|
|
|
320 [SYS_exit_group] = sys_exit,
|
|
|
|
|
|
|
|
|
|
321 [SYS_read] = sys_read,
|
|
|
|
|
|
|
|
|
|
322 [SYS_pread] = sys_pread,
|
|
|
|
|
|
|
|
|
|
323 [SYS_write] = sys_write,
|
|
|
|
|
|
|
|
|
|
324 [SYS_openat] = sys_openat,
|
|
|
|
|
|
|
|
|
|
325 [SYS_close] = sys_close,
|
|
|
|
|
|
|
|
|
|
326 [SYS_fstat] = sys_fstat,
|
|
|
|
|
|
|
|
|
|
327 [SYS_lseek] = sys_lseek,
|
|
|
|
|
|
|
|
|
|
328 [SYS_renameat] = sys_renameat,
|
|
|
|
|
|
|
|
|
|
329 [SYS_mkdirat] = sys_mkdirat,
|
|
|
|
|
|
|
|
|
|
330 [SYS_getcwd] = sys_getcwd,
|
|
|
|
|
|
|
|
|
|
331 [SYS_brk] = sys_brk,
|
|
|
|
|
|
|
|
|
|
332 [SYS_uname] = sys_uname,
|
|
|
|
|
|
|
|
|
|
333 [SYS_prlimit64] = sys_stub_nosys,
|
|
|
|
|
|
|
|
|
|
334 [SYS_rt_sigaction] = sys_rt_sigaction,
|
|
|
|
|
|
|
|
|
|
335 [SYS_times] = sys_times,
|
|
|
|
|
|
|
|
|
|
336 [SYS_writev] = sys_writev,
|
|
|
|
|
|
|
|
|
|
337 [SYS_readlinkat] = sys_stub_nosys,
|
|
|
|
|
|
|
|
|
|
338 [SYS_rt_sigprocmask] = sys_stub_success,
|
|
|
|
|
|
|
|
|
|
339 [SYS_ioctl] = sys_stub_nosys,
|
|
|
|
|
|
|
|
|
|
340 [SYS_getrusage] = sys_stub_nosys,
|
|
|
|
|
|
|
|
|
|
341 [SYS_getrlimit] = sys_stub_nosys,
|
|
|
|
|
|
|
|
|
|
342 [SYS_setrlimit] = sys_stub_nosys,
|
|
|
|
|
|
|
|
|
|
343 [SYS_set_tid_address] = sys_stub_nosys,
|
|
|
|
|
|
|
|
|
|
344 [SYS_set_robust_list] = sys_stub_nosys,
|
|
|
|
|
|
|
|
|
|
345 };
|
|
|
|
|
|
|
|
|
|
346
|
|
|
|
|
|
|
|
|
|
347 syscall_t f = 0;
|
|
|
|
|
|
|
|
|
|
348
|
|
|
|
|
|
|
|
|
|
349 if (n < ARRAY_SIZE(syscall_table))
|
|
|
|
|
|
|
|
|
|
350 f = syscall_table[n];
|
|
|
|
|
|
|
|
|
|
351 if (!f)
|
|
|
|
|
|
|
|
|
|
352 panic("bad syscall #%ld!",n);
|
|
|
|
|
|
|
|
|
|
353
|
|
|
|
|
|
|
|
|
|
354 return f(a0, a1, a2, a3, a4, a5, n);
|
|
|
|
|
|
|
|
|
|
355 }
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
do_syscall中通过传入的系统调用号n，查询syscall_table得到对应的函数，并最终执行系统调用。
|
Zhiyuan Shao
4 years ago
