Linux 内核揭密

 主页   资讯   文章   代码   电子书 

Interrupts and Interrupt Handling. Part 3.

Exception Handling

This is the third part of the chapter about an interrupts and an exceptions handling in the Linux kernel and in the previous part we stopped at the setup_arch function from the arch/x86/kernel/setup.c source code file.

We already know that this function executes initialization of architecture-specific stuff. In our case the setup_arch function does x86_64 architecture related initializations. The setup_arch is big function, and in the previous part we stopped on the setting of the two exceptions handlers for the two following exceptions:

  • #DB - debug exception, transfers control from the interrupted process to the debug handler;
  • #BP - breakpoint exception, caused by the int 3 instruction.

These exceptions allow the x86_64 architecture to have early exception processing for the purpose of debugging via the kgdb.

As you can remember we set these exceptions handlers in the early_trap_init function:

void __init early_trap_init(void)
{
        set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
        set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
        load_idt(&idt_descr);
}

from the arch/x86/kernel/traps.c. We already saw implementation of the set_intr_gate_ist and set_system_intr_gate_ist functions in the previous part and now we will look on the implementation of these two exceptions handlers.

Debug and Breakpoint exceptions

Ok, we setup exception handlers in the early_trap_init function for the #DB and #BP exceptions and now time is to consider their implementations. But before we will do this, first of all let's look on details of these exceptions.

The first exceptions - #DB or debug exception occurs when a debug event occurs. For example - attempt to change the contents of a debug register. Debug registers are special registers that were presented in x86 processors starting from the Intel 80386 processor and as you can understand from name of this CPU extension, main purpose of these registers is debugging.

These registers allow to set breakpoints on the code and read or write data to trace it. Debug registers may be accessed only in the privileged mode and an attempt to read or write the debug registers when executing at any other privilege level causes a general protection fault exception. That's why we have used set_intr_gate_ist for the #DB exception, but not the set_system_intr_gate_ist.

The verctor number of the #DB exceptions is 1 (we pass it as X86_TRAP_DB) and as we may read in specification, this exception has no error code:

+-----------------------------------------------------+
|Vector|Mnemonic|Description         |Type |Error Code|
+-----------------------------------------------------+
|1     | #DB    |Reserved            |F/T  |NO        |
+-----------------------------------------------------+

The second exception is #BP or breakpoint exception occurs when processor executes the int 3 instruction. Unlike the DB exception, the #BP exception may occur in userspace. We can add it anywhere in our code, for example let's look on the simple program:

// breakpoint.c
#include <stdio.h>

int main() {
    int i;
    while (i < 6){
        printf("i equal to: %d\n", i);
        __asm__("int3");
        ++i;
    }
}

If we will compile and run this program, we will see following output:

$ gcc breakpoint.c -o breakpoint
i equal to: 0
Trace/breakpoint trap

But if will run it with gdb, we will see our breakpoint and can continue execution of our program:

$ gdb breakpoint
...
...
...
(gdb) run
Starting program: /home/alex/breakpoints 
i equal to: 0

Program received signal SIGTRAP, Trace/breakpoint trap.
0x0000000000400585 in main ()
=> 0x0000000000400585 <main+31>:    83 45 fc 01 add    DWORD PTR [rbp-0x4],0x1
(gdb) c
Continuing.
i equal to: 1

Program received signal SIGTRAP, Trace/breakpoint trap.
0x0000000000400585 in main ()
=> 0x0000000000400585 <main+31>:    83 45 fc 01 add    DWORD PTR [rbp-0x4],0x1
(gdb) c
Continuing.
i equal to: 2

Program received signal SIGTRAP, Trace/breakpoint trap.
0x0000000000400585 in main ()
=> 0x0000000000400585 <main+31>:    83 45 fc 01 add    DWORD PTR [rbp-0x4],0x1
...
...
...

From this moment we know a little about these two exceptions and we can move on to consideration of their handlers.

Preparation before an exception handler

As you may note before, the set_intr_gate_ist and set_system_intr_gate_ist functions takes an addresses of exceptions handlers in theirs second parameter. In or case our two exception handlers will be:

  • debug;
  • int3.

You will not find these functions in the C code. all of that could be found in the kernel's *.c/*.h files only definition of these functions which are located in the arch/x86/include/asm/traps.h kernel header file:

asmlinkage void debug(void);

and

asmlinkage void int3(void);

You may note asmlinkage directive in definitions of these functions. The directive is the special specificator of the gcc. Actually for a C functions which are called from assembly, we need in explicit declaration of the function calling convention. In our case, if function made with asmlinkage descriptor, then gcc will compile the function to retrieve parameters from stack.

So, both handlers are defined in the arch/x86/entry/entry_64.S assembly source code file with the idtentry macro:

idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK

and

idtentry int3 do_int3 has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK

Each exception handler may be consists from two parts. The first part is generic part and it is the same for all exception handlers. An exception handler should to save general purpose registers on the stack, switch to kernel stack if an exception came from userspace and transfer control to the second part of an exception handler. The second part of an exception handler does certain work depends on certain exception. For example page fault exception handler should find virtual page for given address, invalid opcode exception handler should send SIGILL signal and etc.

As we just saw, an exception handler starts from definition of the idtentry macro from the arch/x86/kernel/entry_64.S assembly source code file, so let's look at implementation of this macro. As we may see, the idtentry macro takes five arguments:

  • sym - defines global symbol with the .globl name which will be an an entry of exception handler;
  • do_sym - symbol name which represents a secondary entry of an exception handler;
  • has_error_code - information about existence of an error code of exception.

The last two parameters are optional:

  • paranoid - shows us how we need to check current mode (will see explanation in details later);
  • shift_ist - shows us is an exception running at Interrupt Stack Table.

Definition of the .idtentry macro looks:

.macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
ENTRY(\sym)
...
...
...
END(\sym)
.endm

Before we will consider internals of the idtentry macro, we should to know state of stack when an exception occurs. As we may read in the Intel® 64 and IA-32 Architectures Software Developer’s Manual 3A, the state of stack when an exception occurs is following:

    +------------+
+40 | %SS        |
+32 | %RSP       |
+24 | %RFLAGS    |
+16 | %CS        |
 +8 | %RIP       |
  0 | ERROR CODE | <-- %RSP
    +------------+

Now we may start to consider implementation of the idtmacro. Both #DB and BP exception handlers are defined as:

idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
idtentry int3 do_int3 has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK

If we will look at these definitions, we may know that compiler will generate two routines with debug and int3 names and both of these exception handlers will call do_debug and do_int3 secondary handlers after some preparation. The third parameter defines existence of error code and as we may see both our exception do not have them. As we may see on the diagram above, processor pushes error code on stack if an exception provides it. In our case, the debug and int3 exception do not have error codes. This may bring some difficulties because stack will look differently for exceptions which provides error code and for exceptions which not. That's why implementation of the idtentry macro starts from putting a fake error code to the stack if an exception does not provide it:

.ifeq \has_error_code
    pushq   $-1
.endif

But it is not only fake error-code. Moreover the -1 also represents invalid system call number, so that the system call restart logic will not be triggered.

The last two parameters of the idtentry macro shift_ist and paranoid allow to know do an exception handler runned at stack from Interrupt Stack Table or not. You already may know that each kernel thread in the system has own stack. In addition to these stacks, there are some specialized stacks associated with each processor in the system. One of these stacks is - exception stack. The x86_64 architecture provides special feature which is called - Interrupt Stack Table. This feature allows to switch to a new stack for designated events such as an atomic exceptions like double fault and etc. So the shift_ist parameter allows us to know do we need to switch on IST stack for an exception handler or not.

The second parameter - paranoid defines the method which helps us to know did we come from userspace or not to an exception handler. The easiest way to determine this is to via CPL or Current Privilege Level in CS segment register. If it is equal to 3, we came from userspace, if zero we came from kernel space:

testl $3,CS(%rsp)
jnz userspace
...
...
...
// we are from the kernel space

But unfortunately this method does not give a 100% guarantee. As described in the kernel documentation:

if we are in an NMI/MCE/DEBUG/whatever super-atomic entry context, which might have triggered right after a normal entry wrote CS to the stack but before we executed SWAPGS, then the only safe way to check for GS is the slower method: the RDMSR.

In other words for example NMI could happen inside the critical section of a swapgs instruction. In this way we should check value of the MSR_GS_BASE model specific register which stores pointer to the start of per-cpu area. So to check did we come from userspace or not, we should to check value of the MSR_GS_BASE model specific register and if it is negative we came from kernel space, in other way we came from userspace:

movl $MSR_GS_BASE,%ecx
rdmsr
testl %edx,%edx
js 1f

In first two lines of code we read value of the MSR_GS_BASE model specific register into edx:eax pair. We can't set negative value to the gs from userspace. But from other side we know that direct mapping of the physical memory starts from the 0xffff880000000000 virtual address. In this way, MSR_GS_BASE will contain an address from 0xffff880000000000 to 0xffffc7ffffffffff. After the rdmsr instruction will be executed, the smallest possible value in the %edx register will be - 0xffff8800 which is -30720 in unsigned 4 bytes. That's why kernel space gs which points to start of per-cpu area will contain negative value.

After we pushed fake error code on the stack, we should allocate space for general purpose registers with:

ALLOC_PT_GPREGS_ON_STACK

macro which is defined in the arch/x86/entry/calling.h header file. This macro just allocates 15*8 bytes space on the stack to preserve general purpose registers:

.macro ALLOC_PT_GPREGS_ON_STACK addskip=0
    addq    $-(15*8+\addskip), %rsp
.endm

So the stack will look like this after execution of the ALLOC_PT_GPREGS_ON_STACK:

     +------------+
+160 | %SS        |
+152 | %RSP       |
+144 | %RFLAGS    |
+136 | %CS        |
+128 | %RIP       |
+120 | ERROR CODE |
     |------------|
+112 |            |
+104 |            |
 +96 |            |
 +88 |            |
 +80 |            |
 +72 |            |
 +64 |            |
 +56 |            |
 +48 |            |
 +40 |            |
 +32 |            |
 +24 |            |
 +16 |            |
  +8 |            |
  +0 |            | <- %RSP
     +------------+

After we allocated space for general purpose registers, we do some checks to understand did an exception come from userspace or not and if yes, we should move back to an interrupted process stack or stay on exception stack:

.if \paranoid
    .if \paranoid == 1
        testb   $3, CS(%rsp)
        jnz 1f
    .endif
    call    paranoid_entry
.else
    call    error_entry
.endif

Let's consider all of these there cases in course.

An exception occured in userspace

In the first let's consider a case when an exception has paranoid=1 like our debug and int3 exceptions. In this case we check selector from CS segment register and jump at 1f label if we came from userspace or the paranoid_entry will be called in other way.

Let's consider first case when we came from userspace to an exception handler. As described above we should jump at 1 label. The 1 label starts from the call of the

call    error_entry

routine which saves all general purpose registers in the previously allocated area on the stack:

SAVE_C_REGS 8
SAVE_EXTRA_REGS 8

These both macros are defined in the arch/x86/entry/calling.h header file and just move values of general purpose registers to a certain place at the stack, for example:

.macro SAVE_EXTRA_REGS offset=0
    movq %r15, 0*8+\offset(%rsp)
    movq %r14, 1*8+\offset(%rsp)
    movq %r13, 2*8+\offset(%rsp)
    movq %r12, 3*8+\offset(%rsp)
    movq %rbp, 4*8+\offset(%rsp)
    movq %rbx, 5*8+\offset(%rsp)
.endm

After execution of SAVE_C_REGS and SAVE_EXTRA_REGS the stack will look:

     +------------+
+160 | %SS        |
+152 | %RSP       |
+144 | %RFLAGS    |
+136 | %CS        |
+128 | %RIP       |
+120 | ERROR CODE |
     |------------|
+112 | %RDI       |
+104 | %RSI       |
 +96 | %RDX       |
 +88 | %RCX       |
 +80 | %RAX       |
 +72 | %R8        |
 +64 | %R9        |
 +56 | %R10       |
 +48 | %R11       |
 +40 | %RBX       |
 +32 | %RBP       |
 +24 | %R12       |
 +16 | %R13       |
  +8 | %R14       |
  +0 | %R15       | <- %RSP
     +------------+

After the kernel saved general purpose registers at the stack, we should check that we came from userspace space again with:

testb   $3, CS+8(%rsp)
jz  .Lerror_kernelspace

because we may have potentially fault if as described in documentation truncated %RIP was reported. Anyway, in both cases the SWAPGS instruction will be executed and values from MSR_KERNEL_GS_BASE and MSR_GS_BASE will be swapped. From this moment the %gs register will point to the base address of kernel structures. So, the SWAPGS instruction is called and it was main point of the error_entry routing.

Now we can back to the idtentry macro. We may see following assembler code after the call of error_entry:

movq    %rsp, %rdi
call    sync_regs

Here we put base address of stack pointer %rdi register which will be first argument (according to x86_64 ABI) of the sync_regs function and call this function which is defined in the arch/x86/kernel/traps.c source code file:

asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
{
    struct pt_regs *regs = task_pt_regs(current);
    *regs = *eregs;
    return regs;
}

This function takes the result of the task_ptr_regs macro which is defined in the arch/x86/include/asm/processor.h header file, stores it in the stack pointer and return it. The task_ptr_regs macro expands to the address of thread.sp0 which represents pointer to the normal kernel stack:

#define task_pt_regs(tsk)       ((struct pt_regs *)(tsk)->thread.sp0 - 1)

As we came from userspace, this means that exception handler will run in real process context. After we got stack pointer from the sync_regs we switch stack:

movq    %rax, %rsp

The last two steps before an exception handler will call secondary handler are:

  1. Passing pointer to pt_regs structure which contains preserved general purpose registers to the %rdi register:
movq    %rsp, %rdi

as it will be passed as first parameter of secondary exception handler.

  1. Pass error code to the %rsi register as it will be second argument of an exception handler and set it to -1 on the stack for the same purpose as we did it before - to prevent restart of a system call:
.if \has_error_code
    movq    ORIG_RAX(%rsp), %rsi
    movq    $-1, ORIG_RAX(%rsp)
.else
    xorl    %esi, %esi
.endif

Additionally you may see that we zeroed the %esi register above in a case if an exception does not provide error code.

In the end we just call secondary exception handler:

call    \do_sym

which:

dotraplinkage void do_debug(struct pt_regs *regs, long error_code);

will be for debug exception and:

dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code);

will be for int 3 exception. In this part we will not see implementations of secondary handlers, because of they are very specific, but will see some of them in one of next parts.

We just considered first case when an exception occurred in userspace. Let's consider last two.

An exception with paranoid > 0 occurred in kernelspace

In this case an exception was occurred in kernelspace and idtentry macro is defined with paranoid=1 for this exception. This value of paranoid means that we should use slower way that we saw in the beginning of this part to check do we really came from kernelspace or not. The paranoid_entry routing allows us to know this:

ENTRY(paranoid_entry)
    cld
    SAVE_C_REGS 8
    SAVE_EXTRA_REGS 8
    movl    $1, %ebx
    movl    $MSR_GS_BASE, %ecx
    rdmsr
    testl   %edx, %edx
    js  1f
    SWAPGS
    xorl    %ebx, %ebx
1:  ret
END(paranoid_entry)

As you may see, this function represents the same that we covered before. We use second (slow) method to get information about previous state of an interrupted task. As we checked this and executed SWAPGS in a case if we came from userspace, we should to do the same that we did before: We need to put pointer to a structure which holds general purpose registers to the %rdi (which will be first parameter of a secondary handler) and put error code if an exception provides it to the %rsi (which will be second parameter of a secondary handler):

movq    %rsp, %rdi

.if \has_error_code
    movq    ORIG_RAX(%rsp), %rsi
    movq    $-1, ORIG_RAX(%rsp)
.else
    xorl    %esi, %esi
.endif

The last step before a secondary handler of an exception will be called is cleanup of new IST stack fram:

.if \shift_ist != -1
    subq    $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
.endif

You may remember that we passed the shift_ist as argument of the idtentry macro. Here we check its value and if its not equal to -1, we get pointer to a stack from Interrupt Stack Table by shift_ist index and setup it.

In the end of this second way we just call secondary exception handler as we did it before:

call    \do_sym

The last method is similar to previous both, but an exception occured with paranoid=0 and we may use fast method determination of where we are from.

Exit from an exception handler

After secondary handler will finish its works, we will return to the idtentry macro and the next step will be jump to the error_exit:

jmp error_exit

routine. The error_exit function defined in the same arch/x86/entry/entry_64.S assembly source code file and the main goal of this function is to know where we are from (from userspace or kernelspace) and execute SWPAGS depends on this. Restore registers to previous state and execute iret instruction to transfer control to an interrupted task.

That's all.

Conclusion

It is the end of the third part about interrupts and interrupt handling in the Linux kernel. We saw the initialization of the Interrupt descriptor table in the previous part with the #DB and #BP gates and started to dive into preparation before control will be transferred to an exception handler and implementation of some interrupt handlers in this part. In the next part we will continue to dive into this theme and will go next by the setup_arch function and will try to understand interrupts handling related stuff.

If you have any questions or suggestions write me a comment or ping me at twitter.

Please note that English is not my first language, And I am really sorry for any inconvenience. If you find any mistakes please send me PR to linux-insides.

Links