Reputation: 1488
I'm trying to implement a handler for system calls in Pintos. Before the interrupt is raised the arguments for the system calls are pushed in the following way:
/* Invokes syscall NUMBER, passing argument ARG0, and returns the
return value as an `int'. */
#define syscall1(NUMBER, ARG0) \
({ \
int retval; \
asm volatile \
("pushl %[arg0]; pushl %[number]; int $0x30; addl $8, %%esp" \
: "=a" (retval) \
: [number] "i" (NUMBER), \
[arg0] "g" (ARG0) \
: "memory"); \
retval; \
})
/* Invokes syscall NUMBER, passing arguments ARG0 and ARG1, and
returns the return value as an `int'. */
#define syscall2(NUMBER, ARG0, ARG1) \
({ \
int retval; \
asm volatile \
("pushl %[arg1]; pushl %[arg0]; " \
"pushl %[number]; int $0x30; addl $12, %%esp" \
: "=a" (retval) \
: [number] "i" (NUMBER), \
[arg0] "g" (ARG0), \
[arg1] "g" (ARG1) \
: "memory"); \
retval; \
})
/* Invokes syscall NUMBER, passing arguments ARG0, ARG1, and
ARG2, and returns the return value as an `int'. */
#define syscall3(NUMBER, ARG0, ARG1, ARG2) \
({ \
int retval; \
asm volatile \
("pushl %[arg2]; pushl %[arg1]; pushl %[arg0]; " \
"pushl %[number]; int $0x30; addl $16, %%esp" \
: "=a" (retval) \
: [number] "i" (NUMBER), \
[arg0] "g" (ARG0), \
[arg1] "g" (ARG1), \
[arg2] "g" (ARG2) \
: "memory"); \
retval; \
})
I have available a struct which contains all the registers that were pushed, but also a pointer to the user-level stack (onto which the system call number and the arguments were pushed).
/* Interrupt stack frame. */
struct intr_frame
{
/* Pushed by intr_entry in intr-stubs.S.
These are the interrupted task's saved registers. */
uint32_t edi; /* Saved EDI. */
uint32_t esi; /* Saved ESI. */
uint32_t ebp; /* Saved EBP. */
uint32_t esp_dummy; /* Not used. */
uint32_t ebx; /* Saved EBX. */
uint32_t edx; /* Saved EDX. */
uint32_t ecx; /* Saved ECX. */
uint32_t eax; /* Saved EAX. */
uint16_t gs, :16; /* Saved GS segment register. */
uint16_t fs, :16; /* Saved FS segment register. */
uint16_t es, :16; /* Saved ES segment register. */
uint16_t ds, :16; /* Saved DS segment register. */
/* Pushed by intrNN_stub in intr-stubs.S. */
uint32_t vec_no; /* Interrupt vector number. */
/* Sometimes pushed by the CPU,
otherwise for consistency pushed as 0 by intrNN_stub.
The CPU puts it just under `eip', but we move it here. */
uint32_t error_code; /* Error code. */
/* Pushed by intrNN_stub in intr-stubs.S.
This frame pointer eases interpretation of backtraces. */
void *frame_pointer; /* Saved EBP (frame pointer). */
/* Pushed by the CPU.
These are the interrupted task's saved registers. */
void (*eip) (void); /* Next instruction to execute. */
uint16_t cs, :16; /* Code segment for eip. */
uint32_t eflags; /* Saved CPU flags. */
void *esp; /* Saved stack pointer. */
uint16_t ss, :16; /* Data segment for esp. */
};
I now want to get these arguments. All the pointers on the stack are 4-byte in size, so I thought that I could simply cast an argument (the dereferenced pointer) to the corresponding type, then increase the stack pointer by 4 and cast the next pointer.
I have the following question:
The pushl instruction pushes values onto the stack correct? So I should be able to get those values simply by de-referencing the pointer to the stack? E.g to get the first argument (assuming this is an int) I would use (int) *(f->esp + 4), where f is a pointer to a struct intr_frame and I add 4 because the system call number is the first element on the stack. Now the problem is that pointer arithmetic on void pointers is not allowed in C and the arguments can be of different type, so can anybody give any suggestions on how to pop these arguments from the stack?
Upvotes: 1
Views: 325
Reputation: 12435
Yes, you can get the parameter values by dereferencing the user esp. Just as with any void *, you must cast it to a suitable pointer type before dereferencing it or indexing it. In this case, uint32_t * would be appropriate, so you would use
*(((uint32_t *)f->esp) + 1)
Note the +1 instead of +4, since the index is scaled by the size of the object pointed to. If you want to use the actual byte offset, you would need two casts
*(uint32_t *)(((uint8_t *)f->esp) + 4)
Upvotes: 4