mm: hugetlb: fix hugepage memory leak caused by wrong reserve count
[linux/fpc-iii.git] / arch / ia64 / kernel / signal.c
blobb3a124da71e5ec7cd960dc5e18096379ed0f860d
1 /*
2 * Architecture-specific signal handling support.
4 * Copyright (C) 1999-2004 Hewlett-Packard Co
5 * David Mosberger-Tang <davidm@hpl.hp.com>
7 * Derived from i386 and Alpha versions.
8 */
10 #include <linux/errno.h>
11 #include <linux/kernel.h>
12 #include <linux/mm.h>
13 #include <linux/ptrace.h>
14 #include <linux/tracehook.h>
15 #include <linux/sched.h>
16 #include <linux/signal.h>
17 #include <linux/smp.h>
18 #include <linux/stddef.h>
19 #include <linux/tty.h>
20 #include <linux/binfmts.h>
21 #include <linux/unistd.h>
22 #include <linux/wait.h>
24 #include <asm/intrinsics.h>
25 #include <asm/uaccess.h>
26 #include <asm/rse.h>
27 #include <asm/sigcontext.h>
29 #include "sigframe.h"
31 #define DEBUG_SIG 0
32 #define STACK_ALIGN 16 /* minimal alignment for stack pointer */
34 #if _NSIG_WORDS > 1
35 # define PUT_SIGSET(k,u) __copy_to_user((u)->sig, (k)->sig, sizeof(sigset_t))
36 # define GET_SIGSET(k,u) __copy_from_user((k)->sig, (u)->sig, sizeof(sigset_t))
37 #else
38 # define PUT_SIGSET(k,u) __put_user((k)->sig[0], &(u)->sig[0])
39 # define GET_SIGSET(k,u) __get_user((k)->sig[0], &(u)->sig[0])
40 #endif
42 static long
43 restore_sigcontext (struct sigcontext __user *sc, struct sigscratch *scr)
45 unsigned long ip, flags, nat, um, cfm, rsc;
46 long err;
48 /* Always make any pending restarted system calls return -EINTR */
49 current->restart_block.fn = do_no_restart_syscall;
51 /* restore scratch that always needs gets updated during signal delivery: */
52 err = __get_user(flags, &sc->sc_flags);
53 err |= __get_user(nat, &sc->sc_nat);
54 err |= __get_user(ip, &sc->sc_ip); /* instruction pointer */
55 err |= __get_user(cfm, &sc->sc_cfm);
56 err |= __get_user(um, &sc->sc_um); /* user mask */
57 err |= __get_user(rsc, &sc->sc_ar_rsc);
58 err |= __get_user(scr->pt.ar_unat, &sc->sc_ar_unat);
59 err |= __get_user(scr->pt.ar_fpsr, &sc->sc_ar_fpsr);
60 err |= __get_user(scr->pt.ar_pfs, &sc->sc_ar_pfs);
61 err |= __get_user(scr->pt.pr, &sc->sc_pr); /* predicates */
62 err |= __get_user(scr->pt.b0, &sc->sc_br[0]); /* b0 (rp) */
63 err |= __get_user(scr->pt.b6, &sc->sc_br[6]); /* b6 */
64 err |= __copy_from_user(&scr->pt.r1, &sc->sc_gr[1], 8); /* r1 */
65 err |= __copy_from_user(&scr->pt.r8, &sc->sc_gr[8], 4*8); /* r8-r11 */
66 err |= __copy_from_user(&scr->pt.r12, &sc->sc_gr[12], 2*8); /* r12-r13 */
67 err |= __copy_from_user(&scr->pt.r15, &sc->sc_gr[15], 8); /* r15 */
69 scr->pt.cr_ifs = cfm | (1UL << 63);
70 scr->pt.ar_rsc = rsc | (3 << 2); /* force PL3 */
72 /* establish new instruction pointer: */
73 scr->pt.cr_iip = ip & ~0x3UL;
74 ia64_psr(&scr->pt)->ri = ip & 0x3;
75 scr->pt.cr_ipsr = (scr->pt.cr_ipsr & ~IA64_PSR_UM) | (um & IA64_PSR_UM);
77 scr->scratch_unat = ia64_put_scratch_nat_bits(&scr->pt, nat);
79 if (!(flags & IA64_SC_FLAG_IN_SYSCALL)) {
80 /* Restore most scratch-state only when not in syscall. */
81 err |= __get_user(scr->pt.ar_ccv, &sc->sc_ar_ccv); /* ar.ccv */
82 err |= __get_user(scr->pt.b7, &sc->sc_br[7]); /* b7 */
83 err |= __get_user(scr->pt.r14, &sc->sc_gr[14]); /* r14 */
84 err |= __copy_from_user(&scr->pt.ar_csd, &sc->sc_ar25, 2*8); /* ar.csd & ar.ssd */
85 err |= __copy_from_user(&scr->pt.r2, &sc->sc_gr[2], 2*8); /* r2-r3 */
86 err |= __copy_from_user(&scr->pt.r16, &sc->sc_gr[16], 16*8); /* r16-r31 */
89 if ((flags & IA64_SC_FLAG_FPH_VALID) != 0) {
90 struct ia64_psr *psr = ia64_psr(&scr->pt);
92 err |= __copy_from_user(current->thread.fph, &sc->sc_fr[32], 96*16);
93 psr->mfh = 0; /* drop signal handler's fph contents... */
94 preempt_disable();
95 if (psr->dfh)
96 ia64_drop_fpu(current);
97 else {
98 /* We already own the local fph, otherwise psr->dfh wouldn't be 0. */
99 __ia64_load_fpu(current->thread.fph);
100 ia64_set_local_fpu_owner(current);
102 preempt_enable();
104 return err;
108 copy_siginfo_to_user (siginfo_t __user *to, const siginfo_t *from)
110 if (!access_ok(VERIFY_WRITE, to, sizeof(siginfo_t)))
111 return -EFAULT;
112 if (from->si_code < 0) {
113 if (__copy_to_user(to, from, sizeof(siginfo_t)))
114 return -EFAULT;
115 return 0;
116 } else {
117 int err;
120 * If you change siginfo_t structure, please be sure this code is fixed
121 * accordingly. It should never copy any pad contained in the structure
122 * to avoid security leaks, but must copy the generic 3 ints plus the
123 * relevant union member.
125 err = __put_user(from->si_signo, &to->si_signo);
126 err |= __put_user(from->si_errno, &to->si_errno);
127 err |= __put_user((short)from->si_code, &to->si_code);
128 switch (from->si_code >> 16) {
129 case __SI_FAULT >> 16:
130 err |= __put_user(from->si_flags, &to->si_flags);
131 err |= __put_user(from->si_isr, &to->si_isr);
132 case __SI_POLL >> 16:
133 err |= __put_user(from->si_addr, &to->si_addr);
134 err |= __put_user(from->si_imm, &to->si_imm);
135 break;
136 case __SI_TIMER >> 16:
137 err |= __put_user(from->si_tid, &to->si_tid);
138 err |= __put_user(from->si_overrun, &to->si_overrun);
139 err |= __put_user(from->si_ptr, &to->si_ptr);
140 break;
141 case __SI_RT >> 16: /* Not generated by the kernel as of now. */
142 case __SI_MESGQ >> 16:
143 err |= __put_user(from->si_uid, &to->si_uid);
144 err |= __put_user(from->si_pid, &to->si_pid);
145 err |= __put_user(from->si_ptr, &to->si_ptr);
146 break;
147 case __SI_CHLD >> 16:
148 err |= __put_user(from->si_utime, &to->si_utime);
149 err |= __put_user(from->si_stime, &to->si_stime);
150 err |= __put_user(from->si_status, &to->si_status);
151 default:
152 err |= __put_user(from->si_uid, &to->si_uid);
153 err |= __put_user(from->si_pid, &to->si_pid);
154 break;
156 return err;
160 long
161 ia64_rt_sigreturn (struct sigscratch *scr)
163 extern char ia64_strace_leave_kernel, ia64_leave_kernel;
164 struct sigcontext __user *sc;
165 struct siginfo si;
166 sigset_t set;
167 long retval;
169 sc = &((struct sigframe __user *) (scr->pt.r12 + 16))->sc;
172 * When we return to the previously executing context, r8 and r10 have already
173 * been setup the way we want them. Indeed, if the signal wasn't delivered while
174 * in a system call, we must not touch r8 or r10 as otherwise user-level state
175 * could be corrupted.
177 retval = (long) &ia64_leave_kernel;
178 if (test_thread_flag(TIF_SYSCALL_TRACE)
179 || test_thread_flag(TIF_SYSCALL_AUDIT))
181 * strace expects to be notified after sigreturn returns even though the
182 * context to which we return may not be in the middle of a syscall.
183 * Thus, the return-value that strace displays for sigreturn is
184 * meaningless.
186 retval = (long) &ia64_strace_leave_kernel;
188 if (!access_ok(VERIFY_READ, sc, sizeof(*sc)))
189 goto give_sigsegv;
191 if (GET_SIGSET(&set, &sc->sc_mask))
192 goto give_sigsegv;
194 set_current_blocked(&set);
196 if (restore_sigcontext(sc, scr))
197 goto give_sigsegv;
199 #if DEBUG_SIG
200 printk("SIG return (%s:%d): sp=%lx ip=%lx\n",
201 current->comm, current->pid, scr->pt.r12, scr->pt.cr_iip);
202 #endif
203 if (restore_altstack(&sc->sc_stack))
204 goto give_sigsegv;
205 return retval;
207 give_sigsegv:
208 si.si_signo = SIGSEGV;
209 si.si_errno = 0;
210 si.si_code = SI_KERNEL;
211 si.si_pid = task_pid_vnr(current);
212 si.si_uid = from_kuid_munged(current_user_ns(), current_uid());
213 si.si_addr = sc;
214 force_sig_info(SIGSEGV, &si, current);
215 return retval;
219 * This does just the minimum required setup of sigcontext.
220 * Specifically, it only installs data that is either not knowable at
221 * the user-level or that gets modified before execution in the
222 * trampoline starts. Everything else is done at the user-level.
224 static long
225 setup_sigcontext (struct sigcontext __user *sc, sigset_t *mask, struct sigscratch *scr)
227 unsigned long flags = 0, ifs, cfm, nat;
228 long err = 0;
230 ifs = scr->pt.cr_ifs;
232 if (on_sig_stack((unsigned long) sc))
233 flags |= IA64_SC_FLAG_ONSTACK;
234 if ((ifs & (1UL << 63)) == 0)
235 /* if cr_ifs doesn't have the valid bit set, we got here through a syscall */
236 flags |= IA64_SC_FLAG_IN_SYSCALL;
237 cfm = ifs & ((1UL << 38) - 1);
238 ia64_flush_fph(current);
239 if ((current->thread.flags & IA64_THREAD_FPH_VALID)) {
240 flags |= IA64_SC_FLAG_FPH_VALID;
241 err = __copy_to_user(&sc->sc_fr[32], current->thread.fph, 96*16);
244 nat = ia64_get_scratch_nat_bits(&scr->pt, scr->scratch_unat);
246 err |= __put_user(flags, &sc->sc_flags);
247 err |= __put_user(nat, &sc->sc_nat);
248 err |= PUT_SIGSET(mask, &sc->sc_mask);
249 err |= __put_user(cfm, &sc->sc_cfm);
250 err |= __put_user(scr->pt.cr_ipsr & IA64_PSR_UM, &sc->sc_um);
251 err |= __put_user(scr->pt.ar_rsc, &sc->sc_ar_rsc);
252 err |= __put_user(scr->pt.ar_unat, &sc->sc_ar_unat); /* ar.unat */
253 err |= __put_user(scr->pt.ar_fpsr, &sc->sc_ar_fpsr); /* ar.fpsr */
254 err |= __put_user(scr->pt.ar_pfs, &sc->sc_ar_pfs);
255 err |= __put_user(scr->pt.pr, &sc->sc_pr); /* predicates */
256 err |= __put_user(scr->pt.b0, &sc->sc_br[0]); /* b0 (rp) */
257 err |= __put_user(scr->pt.b6, &sc->sc_br[6]); /* b6 */
258 err |= __copy_to_user(&sc->sc_gr[1], &scr->pt.r1, 8); /* r1 */
259 err |= __copy_to_user(&sc->sc_gr[8], &scr->pt.r8, 4*8); /* r8-r11 */
260 err |= __copy_to_user(&sc->sc_gr[12], &scr->pt.r12, 2*8); /* r12-r13 */
261 err |= __copy_to_user(&sc->sc_gr[15], &scr->pt.r15, 8); /* r15 */
262 err |= __put_user(scr->pt.cr_iip + ia64_psr(&scr->pt)->ri, &sc->sc_ip);
264 if (!(flags & IA64_SC_FLAG_IN_SYSCALL)) {
265 /* Copy scratch regs to sigcontext if the signal didn't interrupt a syscall. */
266 err |= __put_user(scr->pt.ar_ccv, &sc->sc_ar_ccv); /* ar.ccv */
267 err |= __put_user(scr->pt.b7, &sc->sc_br[7]); /* b7 */
268 err |= __put_user(scr->pt.r14, &sc->sc_gr[14]); /* r14 */
269 err |= __copy_to_user(&sc->sc_ar25, &scr->pt.ar_csd, 2*8); /* ar.csd & ar.ssd */
270 err |= __copy_to_user(&sc->sc_gr[2], &scr->pt.r2, 2*8); /* r2-r3 */
271 err |= __copy_to_user(&sc->sc_gr[16], &scr->pt.r16, 16*8); /* r16-r31 */
273 return err;
277 * Check whether the register-backing store is already on the signal stack.
279 static inline int
280 rbs_on_sig_stack (unsigned long bsp)
282 return (bsp - current->sas_ss_sp < current->sas_ss_size);
285 static long
286 force_sigsegv_info (int sig, void __user *addr)
288 unsigned long flags;
289 struct siginfo si;
291 if (sig == SIGSEGV) {
293 * Acquiring siglock around the sa_handler-update is almost
294 * certainly overkill, but this isn't a
295 * performance-critical path and I'd rather play it safe
296 * here than having to debug a nasty race if and when
297 * something changes in kernel/signal.c that would make it
298 * no longer safe to modify sa_handler without holding the
299 * lock.
301 spin_lock_irqsave(&current->sighand->siglock, flags);
302 current->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
303 spin_unlock_irqrestore(&current->sighand->siglock, flags);
305 si.si_signo = SIGSEGV;
306 si.si_errno = 0;
307 si.si_code = SI_KERNEL;
308 si.si_pid = task_pid_vnr(current);
309 si.si_uid = from_kuid_munged(current_user_ns(), current_uid());
310 si.si_addr = addr;
311 force_sig_info(SIGSEGV, &si, current);
312 return 1;
315 static long
316 setup_frame(struct ksignal *ksig, sigset_t *set, struct sigscratch *scr)
318 extern char __kernel_sigtramp[];
319 unsigned long tramp_addr, new_rbs = 0, new_sp;
320 struct sigframe __user *frame;
321 long err;
323 new_sp = scr->pt.r12;
324 tramp_addr = (unsigned long) __kernel_sigtramp;
325 if (ksig->ka.sa.sa_flags & SA_ONSTACK) {
326 int onstack = sas_ss_flags(new_sp);
328 if (onstack == 0) {
329 new_sp = current->sas_ss_sp + current->sas_ss_size;
331 * We need to check for the register stack being on the
332 * signal stack separately, because it's switched
333 * separately (memory stack is switched in the kernel,
334 * register stack is switched in the signal trampoline).
336 if (!rbs_on_sig_stack(scr->pt.ar_bspstore))
337 new_rbs = ALIGN(current->sas_ss_sp,
338 sizeof(long));
339 } else if (onstack == SS_ONSTACK) {
340 unsigned long check_sp;
343 * If we are on the alternate signal stack and would
344 * overflow it, don't. Return an always-bogus address
345 * instead so we will die with SIGSEGV.
347 check_sp = (new_sp - sizeof(*frame)) & -STACK_ALIGN;
348 if (!likely(on_sig_stack(check_sp)))
349 return force_sigsegv_info(ksig->sig, (void __user *)
350 check_sp);
353 frame = (void __user *) ((new_sp - sizeof(*frame)) & -STACK_ALIGN);
355 if (!access_ok(VERIFY_WRITE, frame, sizeof(*frame)))
356 return force_sigsegv_info(ksig->sig, frame);
358 err = __put_user(ksig->sig, &frame->arg0);
359 err |= __put_user(&frame->info, &frame->arg1);
360 err |= __put_user(&frame->sc, &frame->arg2);
361 err |= __put_user(new_rbs, &frame->sc.sc_rbs_base);
362 err |= __put_user(0, &frame->sc.sc_loadrs); /* initialize to zero */
363 err |= __put_user(ksig->ka.sa.sa_handler, &frame->handler);
365 err |= copy_siginfo_to_user(&frame->info, &ksig->info);
367 err |= __save_altstack(&frame->sc.sc_stack, scr->pt.r12);
368 err |= setup_sigcontext(&frame->sc, set, scr);
370 if (unlikely(err))
371 return force_sigsegv_info(ksig->sig, frame);
373 scr->pt.r12 = (unsigned long) frame - 16; /* new stack pointer */
374 scr->pt.ar_fpsr = FPSR_DEFAULT; /* reset fpsr for signal handler */
375 scr->pt.cr_iip = tramp_addr;
376 ia64_psr(&scr->pt)->ri = 0; /* start executing in first slot */
377 ia64_psr(&scr->pt)->be = 0; /* force little-endian byte-order */
379 * Force the interruption function mask to zero. This has no effect when a
380 * system-call got interrupted by a signal (since, in that case, scr->pt_cr_ifs is
381 * ignored), but it has the desirable effect of making it possible to deliver a
382 * signal with an incomplete register frame (which happens when a mandatory RSE
383 * load faults). Furthermore, it has no negative effect on the getting the user's
384 * dirty partition preserved, because that's governed by scr->pt.loadrs.
386 scr->pt.cr_ifs = (1UL << 63);
389 * Note: this affects only the NaT bits of the scratch regs (the ones saved in
390 * pt_regs), which is exactly what we want.
392 scr->scratch_unat = 0; /* ensure NaT bits of r12 is clear */
394 #if DEBUG_SIG
395 printk("SIG deliver (%s:%d): sig=%d sp=%lx ip=%lx handler=%p\n",
396 current->comm, current->pid, ksig->sig, scr->pt.r12, frame->sc.sc_ip, frame->handler);
397 #endif
398 return 0;
401 static long
402 handle_signal (struct ksignal *ksig, struct sigscratch *scr)
404 int ret = setup_frame(ksig, sigmask_to_save(), scr);
406 if (!ret)
407 signal_setup_done(ret, ksig, test_thread_flag(TIF_SINGLESTEP));
409 return ret;
413 * Note that `init' is a special process: it doesn't get signals it doesn't want to
414 * handle. Thus you cannot kill init even with a SIGKILL even by mistake.
416 void
417 ia64_do_signal (struct sigscratch *scr, long in_syscall)
419 long restart = in_syscall;
420 long errno = scr->pt.r8;
421 struct ksignal ksig;
424 * This only loops in the rare cases of handle_signal() failing, in which case we
425 * need to push through a forced SIGSEGV.
427 while (1) {
428 get_signal(&ksig);
431 * get_signal_to_deliver() may have run a debugger (via notify_parent())
432 * and the debugger may have modified the state (e.g., to arrange for an
433 * inferior call), thus it's important to check for restarting _after_
434 * get_signal_to_deliver().
436 if ((long) scr->pt.r10 != -1)
438 * A system calls has to be restarted only if one of the error codes
439 * ERESTARTNOHAND, ERESTARTSYS, or ERESTARTNOINTR is returned. If r10
440 * isn't -1 then r8 doesn't hold an error code and we don't need to
441 * restart the syscall, so we can clear the "restart" flag here.
443 restart = 0;
445 if (ksig.sig <= 0)
446 break;
448 if (unlikely(restart)) {
449 switch (errno) {
450 case ERESTART_RESTARTBLOCK:
451 case ERESTARTNOHAND:
452 scr->pt.r8 = EINTR;
453 /* note: scr->pt.r10 is already -1 */
454 break;
456 case ERESTARTSYS:
457 if ((ksig.ka.sa.sa_flags & SA_RESTART) == 0) {
458 scr->pt.r8 = EINTR;
459 /* note: scr->pt.r10 is already -1 */
460 break;
462 case ERESTARTNOINTR:
463 ia64_decrement_ip(&scr->pt);
464 restart = 0; /* don't restart twice if handle_signal() fails... */
469 * Whee! Actually deliver the signal. If the delivery failed, we need to
470 * continue to iterate in this loop so we can deliver the SIGSEGV...
472 if (handle_signal(&ksig, scr))
473 return;
476 /* Did we come from a system call? */
477 if (restart) {
478 /* Restart the system call - no handlers present */
479 if (errno == ERESTARTNOHAND || errno == ERESTARTSYS || errno == ERESTARTNOINTR
480 || errno == ERESTART_RESTARTBLOCK)
483 * Note: the syscall number is in r15 which is saved in
484 * pt_regs so all we need to do here is adjust ip so that
485 * the "break" instruction gets re-executed.
487 ia64_decrement_ip(&scr->pt);
488 if (errno == ERESTART_RESTARTBLOCK)
489 scr->pt.r15 = __NR_restart_syscall;
493 /* if there's no signal to deliver, we just put the saved sigmask
494 * back */
495 restore_saved_sigmask();