1 /* By Ross Biro 1/23/92 */
3 * Pentium III FXSR, SSE support
4 * Gareth Hughes <gareth@valinux.com>, May 2000
7 #include <linux/kernel.h>
8 #include <linux/sched.h>
10 #include <linux/smp.h>
11 #include <linux/errno.h>
12 #include <linux/ptrace.h>
13 #include <linux/user.h>
14 #include <linux/security.h>
15 #include <linux/audit.h>
16 #include <linux/seccomp.h>
17 #include <linux/signal.h>
19 #include <asm/uaccess.h>
20 #include <asm/pgtable.h>
21 #include <asm/system.h>
22 #include <asm/processor.h>
24 #include <asm/debugreg.h>
29 * does not yet catch signals sent when the child dies.
30 * in exit.c or in signal.c.
34 * Determines which flags the user has access to [1 = access, 0 = no access].
36 #define FLAG_MASK_32 ((unsigned long) \
37 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
38 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
39 X86_EFLAGS_SF | X86_EFLAGS_TF | \
40 X86_EFLAGS_DF | X86_EFLAGS_OF | \
41 X86_EFLAGS_RF | X86_EFLAGS_AC))
43 #define FLAG_MASK FLAG_MASK_32
45 static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
47 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
50 return ®s->bx + regno;
53 static int putreg(struct task_struct *child,
54 unsigned long regno, unsigned long value)
56 struct pt_regs *regs = task_pt_regs(child);
60 if (value && (value & 3) != 3)
62 child->thread.gs = value;
65 * The user-mode %gs is not affected by
66 * kernel entry, so we must update the CPU.
68 loadsegment(gs, value);
73 if (value && (value & 3) != 3)
86 * If the user value contains TF, mark that
87 * it was not "us" (the debugger) that set it.
88 * If not, make sure it stays set if we had.
90 if (value & X86_EFLAGS_TF)
91 clear_tsk_thread_flag(child, TIF_FORCED_TF);
92 else if (test_tsk_thread_flag(child, TIF_FORCED_TF))
93 value |= X86_EFLAGS_TF;
94 value |= regs->flags & ~FLAG_MASK;
97 *pt_regs_access(regs, regno) = value;
101 static unsigned long getreg(struct task_struct *child, unsigned long regno)
103 struct pt_regs *regs = task_pt_regs(child);
104 unsigned long retval = ~0UL;
110 * If the debugger set TF, hide it from the readout.
112 retval = regs->flags;
113 if (test_tsk_thread_flag(child, TIF_FORCED_TF))
114 retval &= ~X86_EFLAGS_TF;
117 retval = child->thread.gs;
118 if (child == current)
119 savesegment(gs, retval);
129 retval &= *pt_regs_access(regs, regno);
135 * This function is trivial and will be inlined by the compiler.
136 * Having it separates the implementation details of debug
137 * registers from the interface details of ptrace.
139 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
142 case 0: return child->thread.debugreg0;
143 case 1: return child->thread.debugreg1;
144 case 2: return child->thread.debugreg2;
145 case 3: return child->thread.debugreg3;
146 case 6: return child->thread.debugreg6;
147 case 7: return child->thread.debugreg7;
152 static int ptrace_set_debugreg(struct task_struct *child,
153 int n, unsigned long data)
157 if (unlikely(n == 4 || n == 5))
160 if (n < 4 && unlikely(data >= TASK_SIZE - 3))
164 case 0: child->thread.debugreg0 = data; break;
165 case 1: child->thread.debugreg1 = data; break;
166 case 2: child->thread.debugreg2 = data; break;
167 case 3: child->thread.debugreg3 = data; break;
170 child->thread.debugreg6 = data;
175 * Sanity-check data. Take one half-byte at once with
176 * check = (val >> (16 + 4*i)) & 0xf. It contains the
177 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
178 * 2 and 3 are LENi. Given a list of invalid values,
179 * we do mask |= 1 << invalid_value, so that
180 * (mask >> check) & 1 is a correct test for invalid
183 * R/Wi contains the type of the breakpoint /
184 * watchpoint, LENi contains the length of the watched
185 * data in the watchpoint case.
187 * The invalid values are:
188 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
189 * - R/Wi == 0x10 (break on I/O reads or writes), so
191 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
194 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
196 * See the Intel Manual "System Programming Guide",
199 * Note that LENi == 0x10 is defined on x86_64 in long
200 * mode (i.e. even for 32-bit userspace software, but
201 * 64-bit kernel), so the x86_64 mask value is 0x5454.
202 * See the AMD manual no. 24593 (AMD64 System Programming)
204 data &= ~DR_CONTROL_RESERVED;
205 for (i = 0; i < 4; i++)
206 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
208 child->thread.debugreg7 = data;
210 set_tsk_thread_flag(child, TIF_DEBUG);
212 clear_tsk_thread_flag(child, TIF_DEBUG);
220 * Called by kernel/ptrace.c when detaching..
222 * Make sure the single step bit is not set.
224 void ptrace_disable(struct task_struct *child)
226 user_disable_single_step(child);
227 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
230 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
232 struct user * dummy = NULL;
234 unsigned long __user *datap = (unsigned long __user *)data;
237 /* when I and D space are separate, these will need to be fixed. */
238 case PTRACE_PEEKTEXT: /* read word at location addr. */
239 case PTRACE_PEEKDATA:
240 ret = generic_ptrace_peekdata(child, addr, data);
243 /* read the word at location addr in the USER area. */
244 case PTRACE_PEEKUSR: {
248 if ((addr & 3) || addr < 0 ||
249 addr > sizeof(struct user) - 3)
252 tmp = 0; /* Default return condition */
253 if(addr < FRAME_SIZE*sizeof(long))
254 tmp = getreg(child, addr);
255 if(addr >= (long) &dummy->u_debugreg[0] &&
256 addr <= (long) &dummy->u_debugreg[7]){
257 addr -= (long) &dummy->u_debugreg[0];
259 tmp = ptrace_get_debugreg(child, addr);
261 ret = put_user(tmp, datap);
265 /* when I and D space are separate, this will have to be fixed. */
266 case PTRACE_POKETEXT: /* write the word at location addr. */
267 case PTRACE_POKEDATA:
268 ret = generic_ptrace_pokedata(child, addr, data);
271 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
273 if ((addr & 3) || addr < 0 ||
274 addr > sizeof(struct user) - 3)
277 if (addr < FRAME_SIZE*sizeof(long)) {
278 ret = putreg(child, addr, data);
281 /* We need to be very careful here. We implicitly
282 want to modify a portion of the task_struct, and we
283 have to be selective about what portions we allow someone
287 if(addr >= (long) &dummy->u_debugreg[0] &&
288 addr <= (long) &dummy->u_debugreg[7]){
289 addr -= (long) &dummy->u_debugreg;
291 ret = ptrace_set_debugreg(child, addr, data);
295 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
296 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
300 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
301 __put_user(getreg(child, i), datap);
308 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
310 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
314 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
315 __get_user(tmp, datap);
316 putreg(child, i, tmp);
323 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
324 if (!access_ok(VERIFY_WRITE, datap,
325 sizeof(struct user_i387_struct))) {
330 if (!tsk_used_math(child))
332 get_fpregs((struct user_i387_struct __user *)data, child);
336 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
337 if (!access_ok(VERIFY_READ, datap,
338 sizeof(struct user_i387_struct))) {
342 set_stopped_child_used_math(child);
343 set_fpregs(child, (struct user_i387_struct __user *)data);
348 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
349 if (!access_ok(VERIFY_WRITE, datap,
350 sizeof(struct user_fxsr_struct))) {
354 if (!tsk_used_math(child))
356 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
360 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
361 if (!access_ok(VERIFY_READ, datap,
362 sizeof(struct user_fxsr_struct))) {
366 set_stopped_child_used_math(child);
367 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
371 case PTRACE_GET_THREAD_AREA:
374 ret = do_get_thread_area(child, addr,
375 (struct user_desc __user *) data);
378 case PTRACE_SET_THREAD_AREA:
381 ret = do_set_thread_area(child, addr,
382 (struct user_desc __user *) data, 0);
386 ret = ptrace_request(child, request, addr, data);
393 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
397 tsk->thread.trap_no = 1;
398 tsk->thread.error_code = error_code;
400 memset(&info, 0, sizeof(info));
401 info.si_signo = SIGTRAP;
402 info.si_code = TRAP_BRKPT;
405 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
407 /* Send us the fake SIGTRAP */
408 force_sig_info(SIGTRAP, &info, tsk);
411 /* notification of system call entry/exit
412 * - triggered by current->work.syscall_trace
414 __attribute__((regparm(3)))
415 int do_syscall_trace(struct pt_regs *regs, int entryexit)
417 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
419 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
422 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
425 /* do the secure computing check first */
427 secure_computing(regs->orig_ax);
429 if (unlikely(current->audit_context)) {
431 audit_syscall_exit(AUDITSC_RESULT(regs->ax),
433 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
434 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
435 * not used, entry.S will call us only on syscall exit, not
436 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
437 * calling send_sigtrap() on syscall entry.
439 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
440 * is_singlestep is false, despite his name, so we will still do
443 else if (is_singlestep)
447 if (!(current->ptrace & PT_PTRACED))
450 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
451 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
452 * here. We have to check this and return */
453 if (is_sysemu && entryexit)
456 /* Fake a debug trap */
458 send_sigtrap(current, regs, 0);
460 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
463 /* the 0x80 provides a way for the tracing parent to distinguish
464 between a syscall stop and SIGTRAP delivery */
465 /* Note that the debugger could change the result of test_thread_flag!*/
466 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
469 * this isn't the same as continuing with a signal, but it will do
470 * for normal use. strace only continues with a signal if the
471 * stopping signal is not SIGTRAP. -brl
473 if (current->exit_code) {
474 send_sig(current->exit_code, current, 1);
475 current->exit_code = 0;
479 if (unlikely(current->audit_context) && !entryexit)
480 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax,
481 regs->bx, regs->cx, regs->dx, regs->si);
485 regs->orig_ax = -1; /* force skip of syscall restarting */
486 if (unlikely(current->audit_context))
487 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);