1 /* arch/sparc64/kernel/process.c
3 * Copyright (C) 1995, 1996, 2008 David S. Miller (davem@davemloft.net)
4 * Copyright (C) 1996 Eddie C. Dost (ecd@skynet.be)
5 * Copyright (C) 1997, 1998 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
9 * This file handles the architecture-dependent parts of process handling..
14 #include <linux/errno.h>
15 #include <linux/export.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
20 #include <linux/smp.h>
21 #include <linux/stddef.h>
22 #include <linux/ptrace.h>
23 #include <linux/slab.h>
24 #include <linux/user.h>
25 #include <linux/delay.h>
26 #include <linux/compat.h>
27 #include <linux/tick.h>
28 #include <linux/init.h>
29 #include <linux/cpu.h>
30 #include <linux/perf_event.h>
31 #include <linux/elfcore.h>
32 #include <linux/sysrq.h>
33 #include <linux/nmi.h>
34 #include <linux/context_tracking.h>
36 #include <asm/uaccess.h>
38 #include <asm/pgalloc.h>
39 #include <asm/pgtable.h>
40 #include <asm/processor.h>
41 #include <asm/pstate.h>
43 #include <asm/fpumacro.h>
45 #include <asm/cpudata.h>
46 #include <asm/mmu_context.h>
47 #include <asm/unistd.h>
48 #include <asm/hypervisor.h>
49 #include <asm/syscalls.h>
50 #include <asm/irq_regs.h>
56 /* Idle loop support on sparc64. */
57 void arch_cpu_idle(void)
59 if (tlb_type != hypervisor) {
67 /* The sun4v sleeping code requires that we have PSTATE.IE cleared over
68 * the cpu sleep hypervisor call.
71 "rdpr %%pstate, %0\n\t"
73 "wrpr %0, %%g0, %%pstate"
77 if (!need_resched() && !cpu_is_offline(smp_processor_id()))
80 /* Re-enable interrupts. */
82 "rdpr %%pstate, %0\n\t"
84 "wrpr %0, %%g0, %%pstate"
90 #ifdef CONFIG_HOTPLUG_CPU
91 void arch_cpu_idle_dead(void)
93 sched_preempt_enable_no_resched();
99 static void show_regwindow32(struct pt_regs *regs)
101 struct reg_window32 __user *rw;
102 struct reg_window32 r_w;
105 __asm__ __volatile__ ("flushw");
106 rw = compat_ptr((unsigned)regs->u_regs[14]);
109 if (copy_from_user (&r_w, rw, sizeof(r_w))) {
115 printk("l0: %08x l1: %08x l2: %08x l3: %08x "
116 "l4: %08x l5: %08x l6: %08x l7: %08x\n",
117 r_w.locals[0], r_w.locals[1], r_w.locals[2], r_w.locals[3],
118 r_w.locals[4], r_w.locals[5], r_w.locals[6], r_w.locals[7]);
119 printk("i0: %08x i1: %08x i2: %08x i3: %08x "
120 "i4: %08x i5: %08x i6: %08x i7: %08x\n",
121 r_w.ins[0], r_w.ins[1], r_w.ins[2], r_w.ins[3],
122 r_w.ins[4], r_w.ins[5], r_w.ins[6], r_w.ins[7]);
125 #define show_regwindow32(regs) do { } while (0)
128 static void show_regwindow(struct pt_regs *regs)
130 struct reg_window __user *rw;
131 struct reg_window *rwk;
132 struct reg_window r_w;
135 if ((regs->tstate & TSTATE_PRIV) || !(test_thread_flag(TIF_32BIT))) {
136 __asm__ __volatile__ ("flushw");
137 rw = (struct reg_window __user *)
138 (regs->u_regs[14] + STACK_BIAS);
139 rwk = (struct reg_window *)
140 (regs->u_regs[14] + STACK_BIAS);
141 if (!(regs->tstate & TSTATE_PRIV)) {
144 if (copy_from_user (&r_w, rw, sizeof(r_w))) {
152 show_regwindow32(regs);
155 printk("l0: %016lx l1: %016lx l2: %016lx l3: %016lx\n",
156 rwk->locals[0], rwk->locals[1], rwk->locals[2], rwk->locals[3]);
157 printk("l4: %016lx l5: %016lx l6: %016lx l7: %016lx\n",
158 rwk->locals[4], rwk->locals[5], rwk->locals[6], rwk->locals[7]);
159 printk("i0: %016lx i1: %016lx i2: %016lx i3: %016lx\n",
160 rwk->ins[0], rwk->ins[1], rwk->ins[2], rwk->ins[3]);
161 printk("i4: %016lx i5: %016lx i6: %016lx i7: %016lx\n",
162 rwk->ins[4], rwk->ins[5], rwk->ins[6], rwk->ins[7]);
163 if (regs->tstate & TSTATE_PRIV)
164 printk("I7: <%pS>\n", (void *) rwk->ins[7]);
167 void show_regs(struct pt_regs *regs)
169 show_regs_print_info(KERN_DEFAULT);
171 printk("TSTATE: %016lx TPC: %016lx TNPC: %016lx Y: %08x %s\n", regs->tstate,
172 regs->tpc, regs->tnpc, regs->y, print_tainted());
173 printk("TPC: <%pS>\n", (void *) regs->tpc);
174 printk("g0: %016lx g1: %016lx g2: %016lx g3: %016lx\n",
175 regs->u_regs[0], regs->u_regs[1], regs->u_regs[2],
177 printk("g4: %016lx g5: %016lx g6: %016lx g7: %016lx\n",
178 regs->u_regs[4], regs->u_regs[5], regs->u_regs[6],
180 printk("o0: %016lx o1: %016lx o2: %016lx o3: %016lx\n",
181 regs->u_regs[8], regs->u_regs[9], regs->u_regs[10],
183 printk("o4: %016lx o5: %016lx sp: %016lx ret_pc: %016lx\n",
184 regs->u_regs[12], regs->u_regs[13], regs->u_regs[14],
186 printk("RPC: <%pS>\n", (void *) regs->u_regs[15]);
187 show_regwindow(regs);
188 show_stack(current, (unsigned long *) regs->u_regs[UREG_FP]);
191 union global_cpu_snapshot global_cpu_snapshot[NR_CPUS];
192 static DEFINE_SPINLOCK(global_cpu_snapshot_lock);
194 static void __global_reg_self(struct thread_info *tp, struct pt_regs *regs,
197 struct global_reg_snapshot *rp;
201 rp = &global_cpu_snapshot[this_cpu].reg;
203 rp->tstate = regs->tstate;
205 rp->tnpc = regs->tnpc;
206 rp->o7 = regs->u_regs[UREG_I7];
208 if (regs->tstate & TSTATE_PRIV) {
209 struct reg_window *rw;
211 rw = (struct reg_window *)
212 (regs->u_regs[UREG_FP] + STACK_BIAS);
213 if (kstack_valid(tp, (unsigned long) rw)) {
215 rw = (struct reg_window *)
216 (rw->ins[6] + STACK_BIAS);
217 if (kstack_valid(tp, (unsigned long) rw))
218 rp->rpc = rw->ins[7];
227 /* In order to avoid hangs we do not try to synchronize with the
228 * global register dump client cpus. The last store they make is to
229 * the thread pointer, so do a short poll waiting for that to become
232 static void __global_reg_poll(struct global_reg_snapshot *gp)
236 while (!gp->thread && ++limit < 100) {
242 void arch_trigger_all_cpu_backtrace(bool include_self)
244 struct thread_info *tp = current_thread_info();
245 struct pt_regs *regs = get_irq_regs();
252 spin_lock_irqsave(&global_cpu_snapshot_lock, flags);
254 this_cpu = raw_smp_processor_id();
256 memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
259 __global_reg_self(tp, regs, this_cpu);
261 smp_fetch_global_regs();
263 for_each_online_cpu(cpu) {
264 struct global_reg_snapshot *gp;
266 if (!include_self && cpu == this_cpu)
269 gp = &global_cpu_snapshot[cpu].reg;
271 __global_reg_poll(gp);
274 printk("%c CPU[%3d]: TSTATE[%016lx] TPC[%016lx] TNPC[%016lx] TASK[%s:%d]\n",
275 (cpu == this_cpu ? '*' : ' '), cpu,
276 gp->tstate, gp->tpc, gp->tnpc,
277 ((tp && tp->task) ? tp->task->comm : "NULL"),
278 ((tp && tp->task) ? tp->task->pid : -1));
280 if (gp->tstate & TSTATE_PRIV) {
281 printk(" TPC[%pS] O7[%pS] I7[%pS] RPC[%pS]\n",
287 printk(" TPC[%lx] O7[%lx] I7[%lx] RPC[%lx]\n",
288 gp->tpc, gp->o7, gp->i7, gp->rpc);
292 memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
294 spin_unlock_irqrestore(&global_cpu_snapshot_lock, flags);
297 #ifdef CONFIG_MAGIC_SYSRQ
299 static void sysrq_handle_globreg(int key)
301 arch_trigger_all_cpu_backtrace(true);
304 static struct sysrq_key_op sparc_globalreg_op = {
305 .handler = sysrq_handle_globreg,
306 .help_msg = "global-regs(y)",
307 .action_msg = "Show Global CPU Regs",
310 static void __global_pmu_self(int this_cpu)
312 struct global_pmu_snapshot *pp;
315 pp = &global_cpu_snapshot[this_cpu].pmu;
318 if (tlb_type == hypervisor &&
319 sun4v_chip_type >= SUN4V_CHIP_NIAGARA4)
322 for (i = 0; i < num; i++) {
323 pp->pcr[i] = pcr_ops->read_pcr(i);
324 pp->pic[i] = pcr_ops->read_pic(i);
328 static void __global_pmu_poll(struct global_pmu_snapshot *pp)
332 while (!pp->pcr[0] && ++limit < 100) {
338 static void pmu_snapshot_all_cpus(void)
343 spin_lock_irqsave(&global_cpu_snapshot_lock, flags);
345 memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
347 this_cpu = raw_smp_processor_id();
349 __global_pmu_self(this_cpu);
351 smp_fetch_global_pmu();
353 for_each_online_cpu(cpu) {
354 struct global_pmu_snapshot *pp = &global_cpu_snapshot[cpu].pmu;
356 __global_pmu_poll(pp);
358 printk("%c CPU[%3d]: PCR[%08lx:%08lx:%08lx:%08lx] PIC[%08lx:%08lx:%08lx:%08lx]\n",
359 (cpu == this_cpu ? '*' : ' '), cpu,
360 pp->pcr[0], pp->pcr[1], pp->pcr[2], pp->pcr[3],
361 pp->pic[0], pp->pic[1], pp->pic[2], pp->pic[3]);
364 memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
366 spin_unlock_irqrestore(&global_cpu_snapshot_lock, flags);
369 static void sysrq_handle_globpmu(int key)
371 pmu_snapshot_all_cpus();
374 static struct sysrq_key_op sparc_globalpmu_op = {
375 .handler = sysrq_handle_globpmu,
376 .help_msg = "global-pmu(x)",
377 .action_msg = "Show Global PMU Regs",
380 static int __init sparc_sysrq_init(void)
382 int ret = register_sysrq_key('y', &sparc_globalreg_op);
385 ret = register_sysrq_key('x', &sparc_globalpmu_op);
389 core_initcall(sparc_sysrq_init);
393 unsigned long thread_saved_pc(struct task_struct *tsk)
395 struct thread_info *ti = task_thread_info(tsk);
396 unsigned long ret = 0xdeadbeefUL;
400 sp = (unsigned long *)(ti->ksp + STACK_BIAS);
401 if (((unsigned long)sp & (sizeof(long) - 1)) == 0UL &&
404 fp = (unsigned long *)(sp[14] + STACK_BIAS);
405 if (((unsigned long)fp & (sizeof(long) - 1)) == 0UL)
412 /* Free current thread data structures etc.. */
413 void exit_thread(void)
415 struct thread_info *t = current_thread_info();
418 if (t->utraps[0] < 2)
425 void flush_thread(void)
427 struct thread_info *t = current_thread_info();
428 struct mm_struct *mm;
432 tsb_context_switch(mm);
434 set_thread_wsaved(0);
436 /* Clear FPU register state. */
440 /* It's a bit more tricky when 64-bit tasks are involved... */
441 static unsigned long clone_stackframe(unsigned long csp, unsigned long psp)
443 bool stack_64bit = test_thread_64bit_stack(psp);
444 unsigned long fp, distance, rval;
449 __get_user(fp, &(((struct reg_window __user *)psp)->ins[6]));
451 if (test_thread_flag(TIF_32BIT))
454 __get_user(fp, &(((struct reg_window32 __user *)psp)->ins[6]));
456 /* Now align the stack as this is mandatory in the Sparc ABI
457 * due to how register windows work. This hides the
458 * restriction from thread libraries etc.
463 rval = (csp - distance);
464 if (copy_in_user((void __user *) rval, (void __user *) psp, distance))
466 else if (!stack_64bit) {
467 if (put_user(((u32)csp),
468 &(((struct reg_window32 __user *)rval)->ins[6])))
471 if (put_user(((u64)csp - STACK_BIAS),
472 &(((struct reg_window __user *)rval)->ins[6])))
475 rval = rval - STACK_BIAS;
481 /* Standard stuff. */
482 static inline void shift_window_buffer(int first_win, int last_win,
483 struct thread_info *t)
487 for (i = first_win; i < last_win; i++) {
488 t->rwbuf_stkptrs[i] = t->rwbuf_stkptrs[i+1];
489 memcpy(&t->reg_window[i], &t->reg_window[i+1],
490 sizeof(struct reg_window));
494 void synchronize_user_stack(void)
496 struct thread_info *t = current_thread_info();
497 unsigned long window;
499 flush_user_windows();
500 if ((window = get_thread_wsaved()) != 0) {
503 struct reg_window *rwin = &t->reg_window[window];
504 int winsize = sizeof(struct reg_window);
507 sp = t->rwbuf_stkptrs[window];
509 if (test_thread_64bit_stack(sp))
512 winsize = sizeof(struct reg_window32);
514 if (!copy_to_user((char __user *)sp, rwin, winsize)) {
515 shift_window_buffer(window, get_thread_wsaved() - 1, t);
516 set_thread_wsaved(get_thread_wsaved() - 1);
522 static void stack_unaligned(unsigned long sp)
526 info.si_signo = SIGBUS;
528 info.si_code = BUS_ADRALN;
529 info.si_addr = (void __user *) sp;
531 force_sig_info(SIGBUS, &info, current);
534 void fault_in_user_windows(void)
536 struct thread_info *t = current_thread_info();
537 unsigned long window;
539 flush_user_windows();
540 window = get_thread_wsaved();
542 if (likely(window != 0)) {
545 struct reg_window *rwin = &t->reg_window[window];
546 int winsize = sizeof(struct reg_window);
549 sp = t->rwbuf_stkptrs[window];
551 if (test_thread_64bit_stack(sp))
554 winsize = sizeof(struct reg_window32);
556 if (unlikely(sp & 0x7UL))
559 if (unlikely(copy_to_user((char __user *)sp,
564 set_thread_wsaved(0);
568 set_thread_wsaved(window + 1);
573 asmlinkage long sparc_do_fork(unsigned long clone_flags,
574 unsigned long stack_start,
575 struct pt_regs *regs,
576 unsigned long stack_size)
578 int __user *parent_tid_ptr, *child_tid_ptr;
579 unsigned long orig_i1 = regs->u_regs[UREG_I1];
583 if (test_thread_flag(TIF_32BIT)) {
584 parent_tid_ptr = compat_ptr(regs->u_regs[UREG_I2]);
585 child_tid_ptr = compat_ptr(regs->u_regs[UREG_I4]);
589 parent_tid_ptr = (int __user *) regs->u_regs[UREG_I2];
590 child_tid_ptr = (int __user *) regs->u_regs[UREG_I4];
593 ret = do_fork(clone_flags, stack_start, stack_size,
594 parent_tid_ptr, child_tid_ptr);
596 /* If we get an error and potentially restart the system
597 * call, we're screwed because copy_thread() clobbered
598 * the parent's %o1. So detect that case and restore it
601 if ((unsigned long)ret >= -ERESTART_RESTARTBLOCK)
602 regs->u_regs[UREG_I1] = orig_i1;
607 /* Copy a Sparc thread. The fork() return value conventions
608 * under SunOS are nothing short of bletcherous:
609 * Parent --> %o0 == childs pid, %o1 == 0
610 * Child --> %o0 == parents pid, %o1 == 1
612 int copy_thread(unsigned long clone_flags, unsigned long sp,
613 unsigned long arg, struct task_struct *p)
615 struct thread_info *t = task_thread_info(p);
616 struct pt_regs *regs = current_pt_regs();
617 struct sparc_stackf *parent_sf;
618 unsigned long child_stack_sz;
619 char *child_trap_frame;
621 /* Calculate offset to stack_frame & pt_regs */
622 child_stack_sz = (STACKFRAME_SZ + TRACEREG_SZ);
623 child_trap_frame = (task_stack_page(p) +
624 (THREAD_SIZE - child_stack_sz));
627 t->ksp = ((unsigned long) child_trap_frame) - STACK_BIAS;
628 t->kregs = (struct pt_regs *) (child_trap_frame +
629 sizeof(struct sparc_stackf));
632 if (unlikely(p->flags & PF_KTHREAD)) {
633 memset(child_trap_frame, 0, child_stack_sz);
634 __thread_flag_byte_ptr(t)[TI_FLAG_BYTE_CWP] =
635 (current_pt_regs()->tstate + 1) & TSTATE_CWP;
636 t->current_ds = ASI_P;
637 t->kregs->u_regs[UREG_G1] = sp; /* function */
638 t->kregs->u_regs[UREG_G2] = arg;
642 parent_sf = ((struct sparc_stackf *) regs) - 1;
643 memcpy(child_trap_frame, parent_sf, child_stack_sz);
644 if (t->flags & _TIF_32BIT) {
645 sp &= 0x00000000ffffffffUL;
646 regs->u_regs[UREG_FP] &= 0x00000000ffffffffUL;
648 t->kregs->u_regs[UREG_FP] = sp;
649 __thread_flag_byte_ptr(t)[TI_FLAG_BYTE_CWP] =
650 (regs->tstate + 1) & TSTATE_CWP;
651 t->current_ds = ASI_AIUS;
652 if (sp != regs->u_regs[UREG_FP]) {
655 csp = clone_stackframe(sp, regs->u_regs[UREG_FP]);
658 t->kregs->u_regs[UREG_FP] = csp;
663 /* Set the return value for the child. */
664 t->kregs->u_regs[UREG_I0] = current->pid;
665 t->kregs->u_regs[UREG_I1] = 1;
667 /* Set the second return value for the parent. */
668 regs->u_regs[UREG_I1] = 0;
670 if (clone_flags & CLONE_SETTLS)
671 t->kregs->u_regs[UREG_G7] = regs->u_regs[UREG_I3];
678 unsigned int pr_regs[32];
679 unsigned long pr_dregs[16];
681 unsigned int __unused;
683 unsigned char pr_qcnt;
684 unsigned char pr_q_entrysize;
686 unsigned int pr_q[64];
690 * fill in the fpu structure for a core dump.
692 int dump_fpu (struct pt_regs * regs, elf_fpregset_t * fpregs)
694 unsigned long *kfpregs = current_thread_info()->fpregs;
695 unsigned long fprs = current_thread_info()->fpsaved[0];
697 if (test_thread_flag(TIF_32BIT)) {
698 elf_fpregset_t32 *fpregs32 = (elf_fpregset_t32 *)fpregs;
701 memcpy(&fpregs32->pr_fr.pr_regs[0], kfpregs,
702 sizeof(unsigned int) * 32);
704 memset(&fpregs32->pr_fr.pr_regs[0], 0,
705 sizeof(unsigned int) * 32);
706 fpregs32->pr_qcnt = 0;
707 fpregs32->pr_q_entrysize = 8;
708 memset(&fpregs32->pr_q[0], 0,
709 (sizeof(unsigned int) * 64));
710 if (fprs & FPRS_FEF) {
711 fpregs32->pr_fsr = (unsigned int) current_thread_info()->xfsr[0];
714 fpregs32->pr_fsr = 0;
719 memcpy(&fpregs->pr_regs[0], kfpregs,
720 sizeof(unsigned int) * 32);
722 memset(&fpregs->pr_regs[0], 0,
723 sizeof(unsigned int) * 32);
725 memcpy(&fpregs->pr_regs[16], kfpregs+16,
726 sizeof(unsigned int) * 32);
728 memset(&fpregs->pr_regs[16], 0,
729 sizeof(unsigned int) * 32);
730 if(fprs & FPRS_FEF) {
731 fpregs->pr_fsr = current_thread_info()->xfsr[0];
732 fpregs->pr_gsr = current_thread_info()->gsr[0];
734 fpregs->pr_fsr = fpregs->pr_gsr = 0;
736 fpregs->pr_fprs = fprs;
740 EXPORT_SYMBOL(dump_fpu);
742 unsigned long get_wchan(struct task_struct *task)
744 unsigned long pc, fp, bias = 0;
745 struct thread_info *tp;
746 struct reg_window *rw;
747 unsigned long ret = 0;
750 if (!task || task == current ||
751 task->state == TASK_RUNNING)
754 tp = task_thread_info(task);
756 fp = task_thread_info(task)->ksp + bias;
759 if (!kstack_valid(tp, fp))
761 rw = (struct reg_window *) fp;
763 if (!in_sched_functions(pc)) {
767 fp = rw->ins[6] + bias;
768 } while (++count < 16);