中断——中断描述符表的定义和初始化(一) (基于3.16-rc4)
1.中断描述符表的定义(arch/x86/kernel/traps.c)
1 gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
定义的描述符表为一个结构体数组,数组元素类型为gate_desc,大小为8B。NR_VECTORS宏为256,即描述符表大小为256*8B。
2.idt_descr变量的定义(arch/x86/kernel/head_32.S)
1 idt_descr: 2 .word IDT_ENTRIES*8-1 # idt contains 256 entries 3 .long idt_table 4 5 # boot GDT descriptor (later on used by CPU#0): 6 .word 0 # 32 bit align gdt_desc.address
这是内核定义的一个全局变量,存放有中断描述符表的大小和首地址。该变量将存放在idtr寄存器中。
3.中断描述符初步的初始化(arch/x86/kernel/head_32.S)
1 __INIT 2 setup_once: 3 /* 4 * Set up a idt with 256 entries pointing to ignore_int, 5 * interrupt gates. It doesn't actually load idt - that needs 6 * to be done on each CPU. Interrupts are enabled elsewhere, 7 * when we can be relatively sure everything is ok. 8 */ 9 10 movl $idt_table,%edi 11 movl $early_idt_handlers,%eax12 movl $NUM_EXCEPTION_VECTORS,%ecx 13 1: 14 movl %eax,(%edi) 15 movl %eax,4(%edi) 16 /* interrupt gate, dpl=0, present */ 17 movl $(0x8E000000 + __KERNEL_CS),2(%edi) 18 addl $9,%eax 19 addl $8,%edi 20 loop 1b 21 22 movl $256 - NUM_EXCEPTION_VECTORS,%ecx 23 movl $ignore_int,%edx 24 movl $(__KERNEL_CS << 16),%eax 25 movw %dx,%ax /* selector = 0x0010 = cs */ 26 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */ 27 2: 28 movl %eax,(%edi) 29 movl %edx,4(%edi) 30 addl $8,%edi 31 loop 2b 32 ... 33 ...
这段代码是对中断描述符表的初步初始化,14-20行是对前32个中断描述符进行初始化,让所有描述符指向early_idt_handlers处理函数。22-31行是对后256-32=224个中断描述符进行初始化,使之指向ignore_int处理函数。省略号以后是对GDT描述符表的初始化,这里不予讨论。
4.中断描述符表最终的初始化(arch/x86/kernel/traps.c)
1 void __init trap_init(void) 2 { 3 int i; 4 5 #ifdef CONFIG_EISA 6 void __iomem *p = early_ioremap(0x0FFFD9, 4); 7 8 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24)) 9 EISA_bus = 1; 10 early_iounmap(p, 4); 11 #endif 12 13 set_intr_gate(X86_TRAP_DE, divide_error); 14 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK); 15 /* int4 can be called from all */ 16 set_system_intr_gate(X86_TRAP_OF, &overflow); 17 set_intr_gate(X86_TRAP_BR, bounds); 18 set_intr_gate(X86_TRAP_UD, invalid_op); 19 set_intr_gate(X86_TRAP_NM, device_not_available); 20 #ifdef CONFIG_X86_32 21 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS); 22 #else 23 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK); 24 #endif 25 set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun); 26 set_intr_gate(X86_TRAP_TS, invalid_TSS); 27 set_intr_gate(X86_TRAP_NP, segment_not_present); 28 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK); 29 set_intr_gate(X86_TRAP_GP, general_protection); 30 set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug); 31 set_intr_gate(X86_TRAP_MF, coprocessor_error); 32 set_intr_gate(X86_TRAP_AC, alignment_check); 33 #ifdef CONFIG_X86_MCE 34 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK); 35 #endif 36 set_intr_gate(X86_TRAP_XF, simd_coprocessor_error); 37 38 /* Reserve all the builtin and the syscall vector: */ 39 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++) 40 set_bit(i, used_vectors); 41 42 #ifdef CONFIG_IA32_EMULATION 43 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall); 44 set_bit(IA32_SYSCALL_VECTOR, used_vectors); 45 #endif 46 47 #ifdef CONFIG_X86_32 48 set_system_trap_gate(SYSCALL_VECTOR, &system_call);FIRST_EXTERNAL_VECTOR 49 set_bit(SYSCALL_VECTOR, used_vectors); 50 #endif 51 52 /* 53 * Set the IDT descriptor to a fixed read-only location, so that the 54 * "sidt" instruction will not leak the location of the kernel, and 55 * to defend the IDT against arbitrary memory write vulnerabilities. 56 * It will be reloaded in cpu_init() */ 57 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO); 58 idt_descr.address = fix_to_virt(FIX_RO_IDT); 59 60 /* 61 * Should be a barrier for any external CPU state: 62 */ 63 cpu_init(); 64 65 x86_init.irqs.trap_init(); 66 67 #ifdef CONFIG_X86_64 68 memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16); 69 set_nmi_gate(X86_TRAP_DB, &debug); 70 set_nmi_gate(X86_TRAP_BP, &int3); 71 #endif 72 }
该函数对中断描述表的进行了部分初始化,13-36行对系统已分配的异常和非屏蔽中断进行初始化,中断向量号为0-19。接着,39-40行在中断位图表中对已初始化的中断所对应的位进行标记。接着,43和48行又出始化了两个中断,一个是系统中断门,中断向量号为0x80,一个是系统陷阱门,中断向量号为2。
在该函数中,大家可以看出,对中断进行初始化的函数有如下几个:
1 set_intr_gate() 2 set_system_intr_gate() 3 set_system_trap_gate()
4 set_task_gate()
这几个函数也在arch/x86/kernel/traps.c中定义。分别是对中断门,系统中断门,系统陷阱门,任务门描述符的初始化。进一步深入可发现,这几个函数都调用了如下的函数:
1 static inline void _set_gate(int gate, unsigned type, void *addr, 2 unsigned dpl, unsigned ist, unsigned seg) 3 { 4 gate_desc s; 5 6 pack_gate(&s, type, (unsigned long)addr, dpl, ist, seg); 7 /* 8 * does not need to be atomic because it is only done once at 9 * setup time 10 */ 11 write_idt_entry(idt_table, gate, &s); 12 write_trace_idt_entry(gate, &s); 13 }
该函数定义在arch/x86/include/asm/desc.h文件中。在该函数中定义了一个gate_desc类型变量s,并将s的指针传递给pack_gate函数,把要初始化的描述符各个字段的值临时存放在s中。下边分析下pack_gate函数,在分析该函数之前,我们先看下gate_desc结构体。
1 struct desc_struct { 2 union { 3 struct { 4 unsigned int a; 5 unsigned int b; 6 }; 7 struct { 8 u16 limit0; 9 u16 base0; 10 unsigned base1: 8, type: 4, s: 1, dpl: 2, p: 1; 11 unsigned limit: 4, avl: 1, l: 1, d: 1, g: 1, base2: 8; 12 }; 13 }; 14 } __attribute__((packed));
typedef struct desc_struct gate_desc
该结构体定义位于arch/x86/include/asm/desc_defs.h中。该结构体中包含了一个共用体,共用体中又包含了两个结构体。我们知道,共用体在分配内存单元时,并不为每个成员都分配,而是为最大的成员来分配。可以看出该共用体的两个结构体成员大小相等,都是8B,因此整个gate_desc结构体大小就为8B。我们可以使用共用体中的任意一个结构体成员来为这个gate_desc赋值,也就是说我们既可以将gate_desc看成是struct { unsigned int a; unsigned int b; };也可以看成是struct { u16 limit0; u16 base0; .... };下面在分析pack_gate函数过程中将看到赋值过程,我们将gate_desc看作是struct { unsigned int a; unsigned int b; };。
1 static inline void pack_gate(gate_desc *gate, unsigned char type, 2 unsigned long base, unsigned dpl, unsigned flags, 3 unsigned short seg) 4 { 5 gate->a = (seg << 16) | (base & 0xffff); 6 gate->b = (base & 0xffff0000) | (((0x80 | type | (dpl << 5)) & 0xff) << 8); 7 }
该函数也定义在arch/x86/include/asm/desc.h文件中。在该函数中为gate所指向的gate_desc描述符进行初始化。gate->a是描述符的0-31位,gate->b是描述符的32-63位。描述符的如下所示:
接着,我们分析_set_gate()中的11行,write_idt_entry()调用。
1 static inline void native_write_idt_entry(gate_desc *idt, int entry, const gate_desc *gate) 2 { 3 memcpy(&idt[entry], gate, sizeof(*gate)); 4 }
#define write_idt_entry() native_write_idt_entry() //粗略的写了下,大家能明白就行
该函数定义在arch/x86/include/asm/desc.h中。在该函数中,使用memcpy()函数将gate中的字段复制到&idt[entry]所指向的各个字段中。很显然,idt[]数组就是内核中定义的中断描述符表,我们在文章开头给大家看过该定义。gate就是我们在_set_gate()中定义的临时变量s,在这里我们将s中的字段值赋给idt[]数组的对应元素,至此一个描述符的初始化工作就全部完成了,s变量的用途也就结束了,另外,entry变量中存放的是要初始化的中断向量号,用该号来定位idt数组的元素。
最后,再补充说明一点东西,回头看下第4点中的trap_init()函数,在该函数中对中断描述符表进行初始化,使用了很多初始化函数比如set_intr_gate()或set_system_intr_gate()等等,我们拿第一个初始化函数set_intr_gate(X86_TRAP_DE, divide_error)来做说明。X86_TRAP_DE是枚举类型参数,代表的是中断向量号,定义在arch/x86/include/asm/traps.h文件中。这种枚举类型其实有很多。
1 /* Interrupts/Exceptions */ 2 enum { 3 X86_TRAP_DE = 0, /* 0, Divide-by-zero */ 4 X86_TRAP_DB, /* 1, Debug */ 5 X86_TRAP_NMI, /* 2, Non-maskable Interrupt */ 6 X86_TRAP_BP, /* 3, Breakpoint */ 7 X86_TRAP_OF, /* 4, Overflow */ 8 X86_TRAP_BR, /* 5, Bound Range Exceeded */ 9 X86_TRAP_UD, /* 6, Invalid Opcode */ 10 X86_TRAP_NM, /* 7, Device Not Available */ 11 X86_TRAP_DF, /* 8, Double Fault */ 12 X86_TRAP_OLD_MF, /* 9, Coprocessor Segment Overrun */ 13 X86_TRAP_TS, /* 10, Invalid TSS */ 14 X86_TRAP_NP, /* 11, Segment Not Present */ 15 X86_TRAP_SS, /* 12, Stack Segment Fault */ 16 X86_TRAP_GP, /* 13, General Protection Fault */ 17 X86_TRAP_PF, /* 14, Page Fault */ 18 X86_TRAP_SPURIOUS, /* 15, Spurious Interrupt */ 19 X86_TRAP_MF, /* 16, x87 Floating-Point Exception */ 20 X86_TRAP_AC, /* 17, Alignment Check */ 21 X86_TRAP_MC, /* 18, Machine Check */ 22 X86_TRAP_XF, /* 19, SIMD Floating-Point Exception */ 23 X86_TRAP_IRET = 32, /* 32, IRET Exception */ 24 };
第二个参数,是汇编函数的函数名(在这里作为函数指针来使用),该函数为内核原先就定义好的中断或异常处理程序。这种类型的函数有很多,都定义在arch/x86/kernel/entry_32.S文件中,下边我们列举几个给大家看看,有兴趣自己去查。
1 ENTRY(segment_not_present) 2 RING0_EC_FRAME 3 ASM_CLAC 4 pushl_cfi $do_segment_not_present 5 jmp error_code 6 CFI_ENDPROC 7 END(segment_not_present) 8 9 ENTRY(stack_segment) 10 RING0_EC_FRAME 11 ASM_CLAC 12 pushl_cfi $do_stack_segment 13 jmp error_code 14 CFI_ENDPROC 15 END(stack_segment) 16 17 ENTRY(alignment_check) 18 RING0_EC_FRAME 19 ASM_CLAC 20 pushl_cfi $do_alignment_check 21 jmp error_code 22 CFI_ENDPROC 23 END(alignment_check) 24 25 ENTRY(divide_error) 26 RING0_INT_FRAME 27 ASM_CLAC 28 pushl_cfi $0 # no error code 29 pushl_cfi $do_divide_error 30 jmp error_code 31 CFI_ENDPROC 32 END(divide_error)
这些汇编代码只是异常处理程序的开头一部分,可以看到每一个汇编段中,都有一条pushl_cfi $do_***的指令,该$do_***才是真正的异常处理程序(函数名,也是函数指针),现将该函数名压入栈中,然后通过jmp error_code指令跳转到$do_***函数中。error_code其实也是一段汇编代码,如下所示:
1 error_code:
2 /* the function address is in %gs's slot on the stack */
3 pushl_cfi %fs
4 /*CFI_REL_OFFSET fs, 0*/
5 pushl_cfi %es
6 /*CFI_REL_OFFSET es, 0*/
7 pushl_cfi %ds
8 /*CFI_REL_OFFSET ds, 0*/
9 pushl_cfi %eax
10 CFI_REL_OFFSET eax, 0
11 pushl_cfi %ebp
12 CFI_REL_OFFSET ebp, 0
13 pushl_cfi %edi
14 CFI_REL_OFFSET edi, 0
15 pushl_cfi %esi
16 CFI_REL_OFFSET esi, 0
17 pushl_cfi %edx
18 CFI_REL_OFFSET edx, 0
19 pushl_cfi %ecx
20 CFI_REL_OFFSET ecx, 0
21 pushl_cfi %ebx
22 CFI_REL_OFFSET ebx, 0
23 cld
24 movl $(__KERNEL_PERCPU), %ecx
25 movl %ecx, %fs
26 UNWIND_ESPFIX_STACK
27 GS_TO_REG %ecx
28 movl PT_GS(%esp), %edi # get the function address
29 movl PT_ORIG_EAX(%esp), %edx # get the error code
30 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
31 REG_TO_PTGS %ecx
32 SET_KERNEL_GS %ecx
33 movl $(__USER_DS), %ecx
34 movl %ecx, %ds
35 movl %ecx, %es
36 TRACE_IRQS_OFF
37 movl %esp,%eax # pt_regs pointer
38 call *%edi
39 jmp ret_from_exception
40 CFI_ENDPROC
41 END(page_fault)
该片段来自arch/x86/kernel/entry_32.S文件中。代码的开始部分3-22行,对寄存器进行压栈操作,因为这些寄存器将要在随后的异常处理程序中用到,所以事先要保存。最后可以看到在38行,执行了call %edi命令,调用了最终的异常处理程序,在28行可以看到将异常处理程序地址存入了edi寄存器中。第39行通过跳入ret_from_exception中,返回被中断的进程。
至此,中断描述符的初始化工作就告一段落。文中有问题的地方希望大家指正。qq:1193533825