Linux Audio

Check our new training course

Loading...
v3.5.6
  1/*
  2 * Copyright 2010 Tilera Corporation. All Rights Reserved.
  3 *
  4 *   This program is free software; you can redistribute it and/or
  5 *   modify it under the terms of the GNU General Public License
  6 *   as published by the Free Software Foundation, version 2.
  7 *
  8 *   This program is distributed in the hope that it will be useful, but
  9 *   WITHOUT ANY WARRANTY; without even the implied warranty of
 10 *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 11 *   NON INFRINGEMENT.  See the GNU General Public License for
 12 *   more details.
 13 */
 14
 15#include <linux/sched.h>
 16#include <linux/preempt.h>
 17#include <linux/module.h>
 18#include <linux/fs.h>
 19#include <linux/kprobes.h>
 20#include <linux/elfcore.h>
 21#include <linux/tick.h>
 22#include <linux/init.h>
 23#include <linux/mm.h>
 24#include <linux/compat.h>
 25#include <linux/hardirq.h>
 26#include <linux/syscalls.h>
 27#include <linux/kernel.h>
 28#include <linux/tracehook.h>
 29#include <linux/signal.h>
 
 
 30#include <asm/stack.h>
 31#include <asm/switch_to.h>
 32#include <asm/homecache.h>
 33#include <asm/syscalls.h>
 34#include <asm/traps.h>
 35#include <asm/setup.h>
 
 36#ifdef CONFIG_HARDWALL
 37#include <asm/hardwall.h>
 38#endif
 39#include <arch/chip.h>
 40#include <arch/abi.h>
 41#include <arch/sim_def.h>
 42
 43
 44/*
 45 * Use the (x86) "idle=poll" option to prefer low latency when leaving the
 46 * idle loop over low power while in the idle loop, e.g. if we have
 47 * one thread per core and we want to get threads out of futex waits fast.
 48 */
 49static int no_idle_nap;
 50static int __init idle_setup(char *str)
 51{
 52	if (!str)
 53		return -EINVAL;
 54
 55	if (!strcmp(str, "poll")) {
 56		pr_info("using polling idle threads.\n");
 57		no_idle_nap = 1;
 58	} else if (!strcmp(str, "halt"))
 59		no_idle_nap = 0;
 60	else
 61		return -1;
 62
 63	return 0;
 64}
 65early_param("idle", idle_setup);
 66
 67/*
 68 * The idle thread. There's no useful work to be
 69 * done, so just try to conserve power and have a
 70 * low exit latency (ie sit in a loop waiting for
 71 * somebody to say that they'd like to reschedule)
 72 */
 73void cpu_idle(void)
 74{
 75	int cpu = smp_processor_id();
 76
 77
 78	current_thread_info()->status |= TS_POLLING;
 79
 80	if (no_idle_nap) {
 81		while (1) {
 82			while (!need_resched())
 83				cpu_relax();
 84			schedule();
 85		}
 86	}
 87
 88	/* endless idle loop with no priority at all */
 89	while (1) {
 90		tick_nohz_idle_enter();
 91		rcu_idle_enter();
 92		while (!need_resched()) {
 93			if (cpu_is_offline(cpu))
 94				BUG();  /* no HOTPLUG_CPU */
 95
 96			local_irq_disable();
 97			__get_cpu_var(irq_stat).idle_timestamp = jiffies;
 98			current_thread_info()->status &= ~TS_POLLING;
 99			/*
100			 * TS_POLLING-cleared state must be visible before we
101			 * test NEED_RESCHED:
102			 */
103			smp_mb();
104
105			if (!need_resched())
106				_cpu_idle();
107			else
108				local_irq_enable();
109			current_thread_info()->status |= TS_POLLING;
110		}
111		rcu_idle_exit();
112		tick_nohz_idle_exit();
113		schedule_preempt_disabled();
114	}
115}
116
117/*
118 * Release a thread_info structure
119 */
120void arch_release_thread_info(struct thread_info *info)
121{
122	struct single_step_state *step_state = info->step_state;
123
124#ifdef CONFIG_HARDWALL
125	/*
126	 * We free a thread_info from the context of the task that has
127	 * been scheduled next, so the original task is already dead.
128	 * Calling deactivate here just frees up the data structures.
129	 * If the task we're freeing held the last reference to a
130	 * hardwall fd, it would have been released prior to this point
131	 * anyway via exit_files(), and the hardwall_task.info pointers
132	 * would be NULL by now.
133	 */
134	hardwall_deactivate_all(info->task);
135#endif
136
137	if (step_state) {
138
139		/*
140		 * FIXME: we don't munmap step_state->buffer
141		 * because the mm_struct for this process (info->task->mm)
142		 * has already been zeroed in exit_mm().  Keeping a
143		 * reference to it here seems like a bad move, so this
144		 * means we can't munmap() the buffer, and therefore if we
145		 * ptrace multiple threads in a process, we will slowly
146		 * leak user memory.  (Note that as soon as the last
147		 * thread in a process dies, we will reclaim all user
148		 * memory including single-step buffers in the usual way.)
149		 * We should either assign a kernel VA to this buffer
150		 * somehow, or we should associate the buffer(s) with the
151		 * mm itself so we can clean them up that way.
152		 */
153		kfree(step_state);
154	}
155}
156
157static void save_arch_state(struct thread_struct *t);
158
159int copy_thread(unsigned long clone_flags, unsigned long sp,
160		unsigned long stack_size,
161		struct task_struct *p, struct pt_regs *regs)
162{
163	struct pt_regs *childregs;
164	unsigned long ksp;
 
165
166	/*
167	 * When creating a new kernel thread we pass sp as zero.
168	 * Assign it to a reasonable value now that we have the stack.
 
 
 
 
169	 */
170	if (sp == 0 && regs->ex1 == PL_ICS_EX1(KERNEL_PL, 0))
171		sp = KSTK_TOP(p);
 
 
 
 
 
 
172
173	/*
174	 * Do not clone step state from the parent; each thread
175	 * must make its own lazily.
176	 */
177	task_thread_info(p)->step_state = NULL;
 
 
 
 
 
 
 
 
178
179	/*
180	 * Start new thread in ret_from_fork so it schedules properly
181	 * and then return from interrupt like the parent.
182	 */
183	p->thread.pc = (unsigned long) ret_from_fork;
184
185	/* Save user stack top pointer so we can ID the stack vm area later. */
186	p->thread.usp0 = sp;
 
 
 
187
188	/* Record the pid of the process that created this one. */
189	p->thread.creator_pid = current->pid;
 
 
 
 
 
190
191	/*
192	 * Copy the registers onto the kernel stack so the
193	 * return-from-interrupt code will reload it into registers.
194	 */
195	childregs = task_pt_regs(p);
196	*childregs = *regs;
197	childregs->regs[0] = 0;         /* return value is zero */
198	childregs->sp = sp;  /* override with new user stack pointer */
 
 
 
 
 
 
199
200	/*
201	 * If CLONE_SETTLS is set, set "tp" in the new task to "r4",
202	 * which is passed in as arg #5 to sys_clone().
203	 */
204	if (clone_flags & CLONE_SETTLS)
205		childregs->tp = regs->regs[4];
206
207	/*
208	 * Copy the callee-saved registers from the passed pt_regs struct
209	 * into the context-switch callee-saved registers area.
210	 * This way when we start the interrupt-return sequence, the
211	 * callee-save registers will be correctly in registers, which
212	 * is how we assume the compiler leaves them as we start doing
213	 * the normal return-from-interrupt path after calling C code.
214	 * Zero out the C ABI save area to mark the top of the stack.
215	 */
216	ksp = (unsigned long) childregs;
217	ksp -= C_ABI_SAVE_AREA_SIZE;   /* interrupt-entry save area */
218	((long *)ksp)[0] = ((long *)ksp)[1] = 0;
219	ksp -= CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long);
220	memcpy((void *)ksp, &regs->regs[CALLEE_SAVED_FIRST_REG],
221	       CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long));
222	ksp -= C_ABI_SAVE_AREA_SIZE;   /* __switch_to() save area */
223	((long *)ksp)[0] = ((long *)ksp)[1] = 0;
224	p->thread.ksp = ksp;
225
226#if CHIP_HAS_TILE_DMA()
227	/*
228	 * No DMA in the new thread.  We model this on the fact that
229	 * fork() clears the pending signals, alarms, and aio for the child.
230	 */
231	memset(&p->thread.tile_dma_state, 0, sizeof(struct tile_dma_state));
232	memset(&p->thread.dma_async_tlb, 0, sizeof(struct async_tlb));
233#endif
234
235#if CHIP_HAS_SN_PROC()
236	/* Likewise, the new thread is not running static processor code. */
237	p->thread.sn_proc_running = 0;
238	memset(&p->thread.sn_async_tlb, 0, sizeof(struct async_tlb));
239#endif
240
241#if CHIP_HAS_PROC_STATUS_SPR()
242	/* New thread has its miscellaneous processor state bits clear. */
243	p->thread.proc_status = 0;
244#endif
245
246#ifdef CONFIG_HARDWALL
247	/* New thread does not own any networks. */
248	memset(&p->thread.hardwall[0], 0,
249	       sizeof(struct hardwall_task) * HARDWALL_TYPES);
250#endif
251
252
253	/*
254	 * Start the new thread with the current architecture state
255	 * (user interrupt masks, etc.).
256	 */
257	save_arch_state(&p->thread);
258
259	return 0;
260}
261
 
 
 
 
 
 
 
 
 
 
 
 
 
 
262/*
263 * Return "current" if it looks plausible, or else a pointer to a dummy.
264 * This can be helpful if we are just trying to emit a clean panic.
265 */
266struct task_struct *validate_current(void)
267{
268	static struct task_struct corrupt = { .comm = "<corrupt>" };
269	struct task_struct *tsk = current;
270	if (unlikely((unsigned long)tsk < PAGE_OFFSET ||
271		     (high_memory && (void *)tsk > high_memory) ||
272		     ((unsigned long)tsk & (__alignof__(*tsk) - 1)) != 0)) {
273		pr_err("Corrupt 'current' %p (sp %#lx)\n", tsk, stack_pointer);
274		tsk = &corrupt;
275	}
276	return tsk;
277}
278
279/* Take and return the pointer to the previous task, for schedule_tail(). */
280struct task_struct *sim_notify_fork(struct task_struct *prev)
281{
282	struct task_struct *tsk = current;
283	__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK_PARENT |
284		     (tsk->thread.creator_pid << _SIM_CONTROL_OPERATOR_BITS));
285	__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK |
286		     (tsk->pid << _SIM_CONTROL_OPERATOR_BITS));
287	return prev;
288}
289
290int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
291{
292	struct pt_regs *ptregs = task_pt_regs(tsk);
293	elf_core_copy_regs(regs, ptregs);
294	return 1;
295}
296
297#if CHIP_HAS_TILE_DMA()
298
299/* Allow user processes to access the DMA SPRs */
300void grant_dma_mpls(void)
301{
302#if CONFIG_KERNEL_PL == 2
303	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
304	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
305#else
306	__insn_mtspr(SPR_MPL_DMA_CPL_SET_0, 1);
307	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0, 1);
308#endif
309}
310
311/* Forbid user processes from accessing the DMA SPRs */
312void restrict_dma_mpls(void)
313{
314#if CONFIG_KERNEL_PL == 2
315	__insn_mtspr(SPR_MPL_DMA_CPL_SET_2, 1);
316	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_2, 1);
317#else
318	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
319	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
320#endif
321}
322
323/* Pause the DMA engine, then save off its state registers. */
324static void save_tile_dma_state(struct tile_dma_state *dma)
325{
326	unsigned long state = __insn_mfspr(SPR_DMA_USER_STATUS);
327	unsigned long post_suspend_state;
328
329	/* If we're running, suspend the engine. */
330	if ((state & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK)
331		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__SUSPEND_MASK);
332
333	/*
334	 * Wait for the engine to idle, then save regs.  Note that we
335	 * want to record the "running" bit from before suspension,
336	 * and the "done" bit from after, so that we can properly
337	 * distinguish a case where the user suspended the engine from
338	 * the case where the kernel suspended as part of the context
339	 * swap.
340	 */
341	do {
342		post_suspend_state = __insn_mfspr(SPR_DMA_USER_STATUS);
343	} while (post_suspend_state & SPR_DMA_STATUS__BUSY_MASK);
344
345	dma->src = __insn_mfspr(SPR_DMA_SRC_ADDR);
346	dma->src_chunk = __insn_mfspr(SPR_DMA_SRC_CHUNK_ADDR);
347	dma->dest = __insn_mfspr(SPR_DMA_DST_ADDR);
348	dma->dest_chunk = __insn_mfspr(SPR_DMA_DST_CHUNK_ADDR);
349	dma->strides = __insn_mfspr(SPR_DMA_STRIDE);
350	dma->chunk_size = __insn_mfspr(SPR_DMA_CHUNK_SIZE);
351	dma->byte = __insn_mfspr(SPR_DMA_BYTE);
352	dma->status = (state & SPR_DMA_STATUS__RUNNING_MASK) |
353		(post_suspend_state & SPR_DMA_STATUS__DONE_MASK);
354}
355
356/* Restart a DMA that was running before we were context-switched out. */
357static void restore_tile_dma_state(struct thread_struct *t)
358{
359	const struct tile_dma_state *dma = &t->tile_dma_state;
360
361	/*
362	 * The only way to restore the done bit is to run a zero
363	 * length transaction.
364	 */
365	if ((dma->status & SPR_DMA_STATUS__DONE_MASK) &&
366	    !(__insn_mfspr(SPR_DMA_USER_STATUS) & SPR_DMA_STATUS__DONE_MASK)) {
367		__insn_mtspr(SPR_DMA_BYTE, 0);
368		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK);
369		while (__insn_mfspr(SPR_DMA_USER_STATUS) &
370		       SPR_DMA_STATUS__BUSY_MASK)
371			;
372	}
373
374	__insn_mtspr(SPR_DMA_SRC_ADDR, dma->src);
375	__insn_mtspr(SPR_DMA_SRC_CHUNK_ADDR, dma->src_chunk);
376	__insn_mtspr(SPR_DMA_DST_ADDR, dma->dest);
377	__insn_mtspr(SPR_DMA_DST_CHUNK_ADDR, dma->dest_chunk);
378	__insn_mtspr(SPR_DMA_STRIDE, dma->strides);
379	__insn_mtspr(SPR_DMA_CHUNK_SIZE, dma->chunk_size);
380	__insn_mtspr(SPR_DMA_BYTE, dma->byte);
381
382	/*
383	 * Restart the engine if we were running and not done.
384	 * Clear a pending async DMA fault that we were waiting on return
385	 * to user space to execute, since we expect the DMA engine
386	 * to regenerate those faults for us now.  Note that we don't
387	 * try to clear the TIF_ASYNC_TLB flag, since it's relatively
388	 * harmless if set, and it covers both DMA and the SN processor.
389	 */
390	if ((dma->status & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK) {
391		t->dma_async_tlb.fault_num = 0;
392		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK);
393	}
394}
395
396#endif
397
398static void save_arch_state(struct thread_struct *t)
399{
400#if CHIP_HAS_SPLIT_INTR_MASK()
401	t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0_0) |
402		((u64)__insn_mfspr(SPR_INTERRUPT_MASK_0_1) << 32);
403#else
404	t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0);
405#endif
406	t->ex_context[0] = __insn_mfspr(SPR_EX_CONTEXT_0_0);
407	t->ex_context[1] = __insn_mfspr(SPR_EX_CONTEXT_0_1);
408	t->system_save[0] = __insn_mfspr(SPR_SYSTEM_SAVE_0_0);
409	t->system_save[1] = __insn_mfspr(SPR_SYSTEM_SAVE_0_1);
410	t->system_save[2] = __insn_mfspr(SPR_SYSTEM_SAVE_0_2);
411	t->system_save[3] = __insn_mfspr(SPR_SYSTEM_SAVE_0_3);
412	t->intctrl_0 = __insn_mfspr(SPR_INTCTRL_0_STATUS);
413#if CHIP_HAS_PROC_STATUS_SPR()
414	t->proc_status = __insn_mfspr(SPR_PROC_STATUS);
415#endif
416#if !CHIP_HAS_FIXED_INTVEC_BASE()
417	t->interrupt_vector_base = __insn_mfspr(SPR_INTERRUPT_VECTOR_BASE_0);
418#endif
419#if CHIP_HAS_TILE_RTF_HWM()
420	t->tile_rtf_hwm = __insn_mfspr(SPR_TILE_RTF_HWM);
421#endif
422#if CHIP_HAS_DSTREAM_PF()
423	t->dstream_pf = __insn_mfspr(SPR_DSTREAM_PF);
424#endif
425}
426
427static void restore_arch_state(const struct thread_struct *t)
428{
429#if CHIP_HAS_SPLIT_INTR_MASK()
430	__insn_mtspr(SPR_INTERRUPT_MASK_0_0, (u32) t->interrupt_mask);
431	__insn_mtspr(SPR_INTERRUPT_MASK_0_1, t->interrupt_mask >> 32);
432#else
433	__insn_mtspr(SPR_INTERRUPT_MASK_0, t->interrupt_mask);
434#endif
435	__insn_mtspr(SPR_EX_CONTEXT_0_0, t->ex_context[0]);
436	__insn_mtspr(SPR_EX_CONTEXT_0_1, t->ex_context[1]);
437	__insn_mtspr(SPR_SYSTEM_SAVE_0_0, t->system_save[0]);
438	__insn_mtspr(SPR_SYSTEM_SAVE_0_1, t->system_save[1]);
439	__insn_mtspr(SPR_SYSTEM_SAVE_0_2, t->system_save[2]);
440	__insn_mtspr(SPR_SYSTEM_SAVE_0_3, t->system_save[3]);
441	__insn_mtspr(SPR_INTCTRL_0_STATUS, t->intctrl_0);
442#if CHIP_HAS_PROC_STATUS_SPR()
443	__insn_mtspr(SPR_PROC_STATUS, t->proc_status);
444#endif
445#if !CHIP_HAS_FIXED_INTVEC_BASE()
446	__insn_mtspr(SPR_INTERRUPT_VECTOR_BASE_0, t->interrupt_vector_base);
447#endif
448#if CHIP_HAS_TILE_RTF_HWM()
449	__insn_mtspr(SPR_TILE_RTF_HWM, t->tile_rtf_hwm);
450#endif
451#if CHIP_HAS_DSTREAM_PF()
452	__insn_mtspr(SPR_DSTREAM_PF, t->dstream_pf);
453#endif
454}
455
456
457void _prepare_arch_switch(struct task_struct *next)
458{
459#if CHIP_HAS_SN_PROC()
460	int snctl;
461#endif
462#if CHIP_HAS_TILE_DMA()
463	struct tile_dma_state *dma = &current->thread.tile_dma_state;
464	if (dma->enabled)
465		save_tile_dma_state(dma);
466#endif
467#if CHIP_HAS_SN_PROC()
468	/*
469	 * Suspend the static network processor if it was running.
470	 * We do not suspend the fabric itself, just like we don't
471	 * try to suspend the UDN.
472	 */
473	snctl = __insn_mfspr(SPR_SNCTL);
474	current->thread.sn_proc_running =
475		(snctl & SPR_SNCTL__FRZPROC_MASK) == 0;
476	if (current->thread.sn_proc_running)
477		__insn_mtspr(SPR_SNCTL, snctl | SPR_SNCTL__FRZPROC_MASK);
478#endif
479}
480
481
482struct task_struct *__sched _switch_to(struct task_struct *prev,
483				       struct task_struct *next)
484{
485	/* DMA state is already saved; save off other arch state. */
486	save_arch_state(&prev->thread);
487
488#if CHIP_HAS_TILE_DMA()
489	/*
490	 * Restore DMA in new task if desired.
491	 * Note that it is only safe to restart here since interrupts
492	 * are disabled, so we can't take any DMATLB miss or access
493	 * interrupts before we have finished switching stacks.
494	 */
495	if (next->thread.tile_dma_state.enabled) {
496		restore_tile_dma_state(&next->thread);
497		grant_dma_mpls();
498	} else {
499		restrict_dma_mpls();
500	}
501#endif
502
503	/* Restore other arch state. */
504	restore_arch_state(&next->thread);
505
506#if CHIP_HAS_SN_PROC()
507	/*
508	 * Restart static network processor in the new process
509	 * if it was running before.
510	 */
511	if (next->thread.sn_proc_running) {
512		int snctl = __insn_mfspr(SPR_SNCTL);
513		__insn_mtspr(SPR_SNCTL, snctl & ~SPR_SNCTL__FRZPROC_MASK);
514	}
515#endif
516
517#ifdef CONFIG_HARDWALL
518	/* Enable or disable access to the network registers appropriately. */
519	hardwall_switch_tasks(prev, next);
520#endif
521
 
 
 
 
 
522	/*
523	 * Switch kernel SP, PC, and callee-saved registers.
524	 * In the context of the new task, return the old task pointer
525	 * (i.e. the task that actually called __switch_to).
526	 * Pass the value to use for SYSTEM_SAVE_K_0 when we reset our sp.
527	 */
528	return __switch_to(prev, next, next_current_ksp0(next));
529}
530
531/*
532 * This routine is called on return from interrupt if any of the
533 * TIF_WORK_MASK flags are set in thread_info->flags.  It is
534 * entered with interrupts disabled so we don't miss an event
535 * that modified the thread_info flags.  If any flag is set, we
536 * handle it and return, and the calling assembly code will
537 * re-disable interrupts, reload the thread flags, and call back
538 * if more flags need to be handled.
539 *
540 * We return whether we need to check the thread_info flags again
541 * or not.  Note that we don't clear TIF_SINGLESTEP here, so it's
542 * important that it be tested last, and then claim that we don't
543 * need to recheck the flags.
544 */
545int do_work_pending(struct pt_regs *regs, u32 thread_info_flags)
546{
547	/* If we enter in kernel mode, do nothing and exit the caller loop. */
548	if (!user_mode(regs))
549		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
550
551	if (thread_info_flags & _TIF_NEED_RESCHED) {
552		schedule();
553		return 1;
554	}
555#if CHIP_HAS_TILE_DMA() || CHIP_HAS_SN_PROC()
556	if (thread_info_flags & _TIF_ASYNC_TLB) {
557		do_async_page_fault(regs);
558		return 1;
559	}
560#endif
561	if (thread_info_flags & _TIF_SIGPENDING) {
562		do_signal(regs);
563		return 1;
564	}
565	if (thread_info_flags & _TIF_NOTIFY_RESUME) {
566		clear_thread_flag(TIF_NOTIFY_RESUME);
567		tracehook_notify_resume(regs);
568		return 1;
569	}
570	if (thread_info_flags & _TIF_SINGLESTEP) {
571		single_step_once(regs);
572		return 0;
 
 
 
 
 
 
 
573	}
574	panic("work_pending: bad flags %#x\n", thread_info_flags);
575}
576
577/* Note there is an implicit fifth argument if (clone_flags & CLONE_SETTLS). */
578SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
579		void __user *, parent_tidptr, void __user *, child_tidptr,
580		struct pt_regs *, regs)
581{
582	if (!newsp)
583		newsp = regs->sp;
584	return do_fork(clone_flags, newsp, regs, 0,
585		       parent_tidptr, child_tidptr);
586}
587
588/*
589 * sys_execve() executes a new program.
590 */
591SYSCALL_DEFINE4(execve, const char __user *, path,
592		const char __user *const __user *, argv,
593		const char __user *const __user *, envp,
594		struct pt_regs *, regs)
595{
596	long error;
597	char *filename;
598
599	filename = getname(path);
600	error = PTR_ERR(filename);
601	if (IS_ERR(filename))
602		goto out;
603	error = do_execve(filename, argv, envp, regs);
604	putname(filename);
605	if (error == 0)
606		single_step_execve();
607out:
608	return error;
609}
610
611#ifdef CONFIG_COMPAT
612long compat_sys_execve(const char __user *path,
613		       compat_uptr_t __user *argv,
614		       compat_uptr_t __user *envp,
615		       struct pt_regs *regs)
616{
617	long error;
618	char *filename;
619
620	filename = getname(path);
621	error = PTR_ERR(filename);
622	if (IS_ERR(filename))
623		goto out;
624	error = compat_do_execve(filename, argv, envp, regs);
625	putname(filename);
626	if (error == 0)
627		single_step_execve();
628out:
629	return error;
630}
631#endif
632
633unsigned long get_wchan(struct task_struct *p)
634{
635	struct KBacktraceIterator kbt;
636
637	if (!p || p == current || p->state == TASK_RUNNING)
638		return 0;
639
640	for (KBacktraceIterator_init(&kbt, p, NULL);
641	     !KBacktraceIterator_end(&kbt);
642	     KBacktraceIterator_next(&kbt)) {
643		if (!in_sched_functions(kbt.it.pc))
644			return kbt.it.pc;
645	}
646
647	return 0;
648}
649
650/*
651 * We pass in lr as zero (cleared in kernel_thread) and the caller
652 * part of the backtrace ABI on the stack also zeroed (in copy_thread)
653 * so that backtraces will stop with this function.
654 * Note that we don't use r0, since copy_thread() clears it.
655 */
656static void start_kernel_thread(int dummy, int (*fn)(int), int arg)
657{
658	do_exit(fn(arg));
659}
660
661/*
662 * Create a kernel thread
663 */
664int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
665{
666	struct pt_regs regs;
667
668	memset(&regs, 0, sizeof(regs));
669	regs.ex1 = PL_ICS_EX1(KERNEL_PL, 0);  /* run at kernel PL, no ICS */
670	regs.pc = (long) start_kernel_thread;
671	regs.flags = PT_FLAGS_CALLER_SAVES;   /* need to restore r1 and r2 */
672	regs.regs[1] = (long) fn;             /* function pointer */
673	regs.regs[2] = (long) arg;            /* parameter register */
674
675	/* Ok, create the new process.. */
676	return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs,
677		       0, NULL, NULL);
678}
679EXPORT_SYMBOL(kernel_thread);
680
681/* Flush thread state. */
682void flush_thread(void)
683{
684	/* Nothing */
685}
686
687/*
688 * Free current thread data structures etc..
689 */
690void exit_thread(void)
691{
692	/* Nothing */
 
 
 
 
 
 
 
 
693}
694
695void show_regs(struct pt_regs *regs)
696{
697	struct task_struct *tsk = validate_current();
698	int i;
699
700	pr_err("\n");
701	pr_err(" Pid: %d, comm: %20s, CPU: %d\n",
702	       tsk->pid, tsk->comm, smp_processor_id());
703#ifdef __tilegx__
704	for (i = 0; i < 51; i += 3)
705		pr_err(" r%-2d: "REGFMT" r%-2d: "REGFMT" r%-2d: "REGFMT"\n",
706		       i, regs->regs[i], i+1, regs->regs[i+1],
707		       i+2, regs->regs[i+2]);
708	pr_err(" r51: "REGFMT" r52: "REGFMT" tp : "REGFMT"\n",
709	       regs->regs[51], regs->regs[52], regs->tp);
710	pr_err(" sp : "REGFMT" lr : "REGFMT"\n", regs->sp, regs->lr);
711#else
712	for (i = 0; i < 52; i += 4)
713		pr_err(" r%-2d: "REGFMT" r%-2d: "REGFMT
714		       " r%-2d: "REGFMT" r%-2d: "REGFMT"\n",
715		       i, regs->regs[i], i+1, regs->regs[i+1],
716		       i+2, regs->regs[i+2], i+3, regs->regs[i+3]);
717	pr_err(" r52: "REGFMT" tp : "REGFMT" sp : "REGFMT" lr : "REGFMT"\n",
718	       regs->regs[52], regs->tp, regs->sp, regs->lr);
719#endif
720	pr_err(" pc : "REGFMT" ex1: %ld     faultnum: %ld\n",
721	       regs->pc, regs->ex1, regs->faultnum);
 
 
 
 
 
 
 
 
 
722
723	dump_stack_regs(regs);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
724}
v4.6
  1/*
  2 * Copyright 2010 Tilera Corporation. All Rights Reserved.
  3 *
  4 *   This program is free software; you can redistribute it and/or
  5 *   modify it under the terms of the GNU General Public License
  6 *   as published by the Free Software Foundation, version 2.
  7 *
  8 *   This program is distributed in the hope that it will be useful, but
  9 *   WITHOUT ANY WARRANTY; without even the implied warranty of
 10 *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 11 *   NON INFRINGEMENT.  See the GNU General Public License for
 12 *   more details.
 13 */
 14
 15#include <linux/sched.h>
 16#include <linux/preempt.h>
 17#include <linux/module.h>
 18#include <linux/fs.h>
 19#include <linux/kprobes.h>
 20#include <linux/elfcore.h>
 21#include <linux/tick.h>
 22#include <linux/init.h>
 23#include <linux/mm.h>
 24#include <linux/compat.h>
 25#include <linux/hardirq.h>
 26#include <linux/syscalls.h>
 27#include <linux/kernel.h>
 28#include <linux/tracehook.h>
 29#include <linux/signal.h>
 30#include <linux/delay.h>
 31#include <linux/context_tracking.h>
 32#include <asm/stack.h>
 33#include <asm/switch_to.h>
 34#include <asm/homecache.h>
 35#include <asm/syscalls.h>
 36#include <asm/traps.h>
 37#include <asm/setup.h>
 38#include <asm/uaccess.h>
 39#ifdef CONFIG_HARDWALL
 40#include <asm/hardwall.h>
 41#endif
 42#include <arch/chip.h>
 43#include <arch/abi.h>
 44#include <arch/sim_def.h>
 45
 
 46/*
 47 * Use the (x86) "idle=poll" option to prefer low latency when leaving the
 48 * idle loop over low power while in the idle loop, e.g. if we have
 49 * one thread per core and we want to get threads out of futex waits fast.
 50 */
 
 51static int __init idle_setup(char *str)
 52{
 53	if (!str)
 54		return -EINVAL;
 55
 56	if (!strcmp(str, "poll")) {
 57		pr_info("using polling idle threads\n");
 58		cpu_idle_poll_ctrl(true);
 59		return 0;
 60	} else if (!strcmp(str, "halt")) {
 61		return 0;
 62	}
 63	return -1;
 
 64}
 65early_param("idle", idle_setup);
 66
 67void arch_cpu_idle(void)
 
 
 
 
 
 
 68{
 69	__this_cpu_write(irq_stat.idle_timestamp, jiffies);
 70	_cpu_idle();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 71}
 72
 73/*
 74 * Release a thread_info structure
 75 */
 76void arch_release_thread_info(struct thread_info *info)
 77{
 78	struct single_step_state *step_state = info->step_state;
 79
 
 
 
 
 
 
 
 
 
 
 
 
 
 80	if (step_state) {
 81
 82		/*
 83		 * FIXME: we don't munmap step_state->buffer
 84		 * because the mm_struct for this process (info->task->mm)
 85		 * has already been zeroed in exit_mm().  Keeping a
 86		 * reference to it here seems like a bad move, so this
 87		 * means we can't munmap() the buffer, and therefore if we
 88		 * ptrace multiple threads in a process, we will slowly
 89		 * leak user memory.  (Note that as soon as the last
 90		 * thread in a process dies, we will reclaim all user
 91		 * memory including single-step buffers in the usual way.)
 92		 * We should either assign a kernel VA to this buffer
 93		 * somehow, or we should associate the buffer(s) with the
 94		 * mm itself so we can clean them up that way.
 95		 */
 96		kfree(step_state);
 97	}
 98}
 99
100static void save_arch_state(struct thread_struct *t);
101
102int copy_thread(unsigned long clone_flags, unsigned long sp,
103		unsigned long arg, struct task_struct *p)
 
104{
105	struct pt_regs *childregs = task_pt_regs(p);
106	unsigned long ksp;
107	unsigned long *callee_regs;
108
109	/*
110	 * Set up the stack and stack pointer appropriately for the
111	 * new child to find itself woken up in __switch_to().
112	 * The callee-saved registers must be on the stack to be read;
113	 * the new task will then jump to assembly support to handle
114	 * calling schedule_tail(), etc., and (for userspace tasks)
115	 * returning to the context set up in the pt_regs.
116	 */
117	ksp = (unsigned long) childregs;
118	ksp -= C_ABI_SAVE_AREA_SIZE;   /* interrupt-entry save area */
119	((long *)ksp)[0] = ((long *)ksp)[1] = 0;
120	ksp -= CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long);
121	callee_regs = (unsigned long *)ksp;
122	ksp -= C_ABI_SAVE_AREA_SIZE;   /* __switch_to() save area */
123	((long *)ksp)[0] = ((long *)ksp)[1] = 0;
124	p->thread.ksp = ksp;
125
126	/* Record the pid of the task that created this one. */
127	p->thread.creator_pid = current->pid;
128
129	if (unlikely(p->flags & PF_KTHREAD)) {
130		/* kernel thread */
131		memset(childregs, 0, sizeof(struct pt_regs));
132		memset(&callee_regs[2], 0,
133		       (CALLEE_SAVED_REGS_COUNT - 2) * sizeof(unsigned long));
134		callee_regs[0] = sp;   /* r30 = function */
135		callee_regs[1] = arg;  /* r31 = arg */
136		p->thread.pc = (unsigned long) ret_from_kernel_thread;
137		return 0;
138	}
139
140	/*
141	 * Start new thread in ret_from_fork so it schedules properly
142	 * and then return from interrupt like the parent.
143	 */
144	p->thread.pc = (unsigned long) ret_from_fork;
145
146	/*
147	 * Do not clone step state from the parent; each thread
148	 * must make its own lazily.
149	 */
150	task_thread_info(p)->step_state = NULL;
151
152#ifdef __tilegx__
153	/*
154	 * Do not clone unalign jit fixup from the parent; each thread
155	 * must allocate its own on demand.
156	 */
157	task_thread_info(p)->unalign_jit_base = NULL;
158#endif
159
160	/*
161	 * Copy the registers onto the kernel stack so the
162	 * return-from-interrupt code will reload it into registers.
163	 */
164	*childregs = *current_pt_regs();
 
165	childregs->regs[0] = 0;         /* return value is zero */
166	if (sp)
167		childregs->sp = sp;  /* override with new user stack pointer */
168	memcpy(callee_regs, &childregs->regs[CALLEE_SAVED_FIRST_REG],
169	       CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long));
170
171	/* Save user stack top pointer so we can ID the stack vm area later. */
172	p->thread.usp0 = childregs->sp;
173
174	/*
175	 * If CLONE_SETTLS is set, set "tp" in the new task to "r4",
176	 * which is passed in as arg #5 to sys_clone().
177	 */
178	if (clone_flags & CLONE_SETTLS)
179		childregs->tp = childregs->regs[4];
180
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
181
182#if CHIP_HAS_TILE_DMA()
183	/*
184	 * No DMA in the new thread.  We model this on the fact that
185	 * fork() clears the pending signals, alarms, and aio for the child.
186	 */
187	memset(&p->thread.tile_dma_state, 0, sizeof(struct tile_dma_state));
188	memset(&p->thread.dma_async_tlb, 0, sizeof(struct async_tlb));
189#endif
190
 
 
 
 
 
 
 
191	/* New thread has its miscellaneous processor state bits clear. */
192	p->thread.proc_status = 0;
 
193
194#ifdef CONFIG_HARDWALL
195	/* New thread does not own any networks. */
196	memset(&p->thread.hardwall[0], 0,
197	       sizeof(struct hardwall_task) * HARDWALL_TYPES);
198#endif
199
200
201	/*
202	 * Start the new thread with the current architecture state
203	 * (user interrupt masks, etc.).
204	 */
205	save_arch_state(&p->thread);
206
207	return 0;
208}
209
210int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
211{
212	task_thread_info(tsk)->align_ctl = val;
213	return 0;
214}
215
216int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
217{
218	return put_user(task_thread_info(tsk)->align_ctl,
219			(unsigned int __user *)adr);
220}
221
222static struct task_struct corrupt_current = { .comm = "<corrupt>" };
223
224/*
225 * Return "current" if it looks plausible, or else a pointer to a dummy.
226 * This can be helpful if we are just trying to emit a clean panic.
227 */
228struct task_struct *validate_current(void)
229{
 
230	struct task_struct *tsk = current;
231	if (unlikely((unsigned long)tsk < PAGE_OFFSET ||
232		     (high_memory && (void *)tsk > high_memory) ||
233		     ((unsigned long)tsk & (__alignof__(*tsk) - 1)) != 0)) {
234		pr_err("Corrupt 'current' %p (sp %#lx)\n", tsk, stack_pointer);
235		tsk = &corrupt_current;
236	}
237	return tsk;
238}
239
240/* Take and return the pointer to the previous task, for schedule_tail(). */
241struct task_struct *sim_notify_fork(struct task_struct *prev)
242{
243	struct task_struct *tsk = current;
244	__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK_PARENT |
245		     (tsk->thread.creator_pid << _SIM_CONTROL_OPERATOR_BITS));
246	__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK |
247		     (tsk->pid << _SIM_CONTROL_OPERATOR_BITS));
248	return prev;
249}
250
251int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
252{
253	struct pt_regs *ptregs = task_pt_regs(tsk);
254	elf_core_copy_regs(regs, ptregs);
255	return 1;
256}
257
258#if CHIP_HAS_TILE_DMA()
259
260/* Allow user processes to access the DMA SPRs */
261void grant_dma_mpls(void)
262{
263#if CONFIG_KERNEL_PL == 2
264	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
265	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
266#else
267	__insn_mtspr(SPR_MPL_DMA_CPL_SET_0, 1);
268	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0, 1);
269#endif
270}
271
272/* Forbid user processes from accessing the DMA SPRs */
273void restrict_dma_mpls(void)
274{
275#if CONFIG_KERNEL_PL == 2
276	__insn_mtspr(SPR_MPL_DMA_CPL_SET_2, 1);
277	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_2, 1);
278#else
279	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
280	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
281#endif
282}
283
284/* Pause the DMA engine, then save off its state registers. */
285static void save_tile_dma_state(struct tile_dma_state *dma)
286{
287	unsigned long state = __insn_mfspr(SPR_DMA_USER_STATUS);
288	unsigned long post_suspend_state;
289
290	/* If we're running, suspend the engine. */
291	if ((state & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK)
292		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__SUSPEND_MASK);
293
294	/*
295	 * Wait for the engine to idle, then save regs.  Note that we
296	 * want to record the "running" bit from before suspension,
297	 * and the "done" bit from after, so that we can properly
298	 * distinguish a case where the user suspended the engine from
299	 * the case where the kernel suspended as part of the context
300	 * swap.
301	 */
302	do {
303		post_suspend_state = __insn_mfspr(SPR_DMA_USER_STATUS);
304	} while (post_suspend_state & SPR_DMA_STATUS__BUSY_MASK);
305
306	dma->src = __insn_mfspr(SPR_DMA_SRC_ADDR);
307	dma->src_chunk = __insn_mfspr(SPR_DMA_SRC_CHUNK_ADDR);
308	dma->dest = __insn_mfspr(SPR_DMA_DST_ADDR);
309	dma->dest_chunk = __insn_mfspr(SPR_DMA_DST_CHUNK_ADDR);
310	dma->strides = __insn_mfspr(SPR_DMA_STRIDE);
311	dma->chunk_size = __insn_mfspr(SPR_DMA_CHUNK_SIZE);
312	dma->byte = __insn_mfspr(SPR_DMA_BYTE);
313	dma->status = (state & SPR_DMA_STATUS__RUNNING_MASK) |
314		(post_suspend_state & SPR_DMA_STATUS__DONE_MASK);
315}
316
317/* Restart a DMA that was running before we were context-switched out. */
318static void restore_tile_dma_state(struct thread_struct *t)
319{
320	const struct tile_dma_state *dma = &t->tile_dma_state;
321
322	/*
323	 * The only way to restore the done bit is to run a zero
324	 * length transaction.
325	 */
326	if ((dma->status & SPR_DMA_STATUS__DONE_MASK) &&
327	    !(__insn_mfspr(SPR_DMA_USER_STATUS) & SPR_DMA_STATUS__DONE_MASK)) {
328		__insn_mtspr(SPR_DMA_BYTE, 0);
329		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK);
330		while (__insn_mfspr(SPR_DMA_USER_STATUS) &
331		       SPR_DMA_STATUS__BUSY_MASK)
332			;
333	}
334
335	__insn_mtspr(SPR_DMA_SRC_ADDR, dma->src);
336	__insn_mtspr(SPR_DMA_SRC_CHUNK_ADDR, dma->src_chunk);
337	__insn_mtspr(SPR_DMA_DST_ADDR, dma->dest);
338	__insn_mtspr(SPR_DMA_DST_CHUNK_ADDR, dma->dest_chunk);
339	__insn_mtspr(SPR_DMA_STRIDE, dma->strides);
340	__insn_mtspr(SPR_DMA_CHUNK_SIZE, dma->chunk_size);
341	__insn_mtspr(SPR_DMA_BYTE, dma->byte);
342
343	/*
344	 * Restart the engine if we were running and not done.
345	 * Clear a pending async DMA fault that we were waiting on return
346	 * to user space to execute, since we expect the DMA engine
347	 * to regenerate those faults for us now.  Note that we don't
348	 * try to clear the TIF_ASYNC_TLB flag, since it's relatively
349	 * harmless if set, and it covers both DMA and the SN processor.
350	 */
351	if ((dma->status & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK) {
352		t->dma_async_tlb.fault_num = 0;
353		__insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK);
354	}
355}
356
357#endif
358
359static void save_arch_state(struct thread_struct *t)
360{
361#if CHIP_HAS_SPLIT_INTR_MASK()
362	t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0_0) |
363		((u64)__insn_mfspr(SPR_INTERRUPT_MASK_0_1) << 32);
364#else
365	t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0);
366#endif
367	t->ex_context[0] = __insn_mfspr(SPR_EX_CONTEXT_0_0);
368	t->ex_context[1] = __insn_mfspr(SPR_EX_CONTEXT_0_1);
369	t->system_save[0] = __insn_mfspr(SPR_SYSTEM_SAVE_0_0);
370	t->system_save[1] = __insn_mfspr(SPR_SYSTEM_SAVE_0_1);
371	t->system_save[2] = __insn_mfspr(SPR_SYSTEM_SAVE_0_2);
372	t->system_save[3] = __insn_mfspr(SPR_SYSTEM_SAVE_0_3);
373	t->intctrl_0 = __insn_mfspr(SPR_INTCTRL_0_STATUS);
 
374	t->proc_status = __insn_mfspr(SPR_PROC_STATUS);
 
375#if !CHIP_HAS_FIXED_INTVEC_BASE()
376	t->interrupt_vector_base = __insn_mfspr(SPR_INTERRUPT_VECTOR_BASE_0);
377#endif
 
378	t->tile_rtf_hwm = __insn_mfspr(SPR_TILE_RTF_HWM);
 
379#if CHIP_HAS_DSTREAM_PF()
380	t->dstream_pf = __insn_mfspr(SPR_DSTREAM_PF);
381#endif
382}
383
384static void restore_arch_state(const struct thread_struct *t)
385{
386#if CHIP_HAS_SPLIT_INTR_MASK()
387	__insn_mtspr(SPR_INTERRUPT_MASK_0_0, (u32) t->interrupt_mask);
388	__insn_mtspr(SPR_INTERRUPT_MASK_0_1, t->interrupt_mask >> 32);
389#else
390	__insn_mtspr(SPR_INTERRUPT_MASK_0, t->interrupt_mask);
391#endif
392	__insn_mtspr(SPR_EX_CONTEXT_0_0, t->ex_context[0]);
393	__insn_mtspr(SPR_EX_CONTEXT_0_1, t->ex_context[1]);
394	__insn_mtspr(SPR_SYSTEM_SAVE_0_0, t->system_save[0]);
395	__insn_mtspr(SPR_SYSTEM_SAVE_0_1, t->system_save[1]);
396	__insn_mtspr(SPR_SYSTEM_SAVE_0_2, t->system_save[2]);
397	__insn_mtspr(SPR_SYSTEM_SAVE_0_3, t->system_save[3]);
398	__insn_mtspr(SPR_INTCTRL_0_STATUS, t->intctrl_0);
 
399	__insn_mtspr(SPR_PROC_STATUS, t->proc_status);
 
400#if !CHIP_HAS_FIXED_INTVEC_BASE()
401	__insn_mtspr(SPR_INTERRUPT_VECTOR_BASE_0, t->interrupt_vector_base);
402#endif
 
403	__insn_mtspr(SPR_TILE_RTF_HWM, t->tile_rtf_hwm);
 
404#if CHIP_HAS_DSTREAM_PF()
405	__insn_mtspr(SPR_DSTREAM_PF, t->dstream_pf);
406#endif
407}
408
409
410void _prepare_arch_switch(struct task_struct *next)
411{
 
 
 
412#if CHIP_HAS_TILE_DMA()
413	struct tile_dma_state *dma = &current->thread.tile_dma_state;
414	if (dma->enabled)
415		save_tile_dma_state(dma);
416#endif
 
 
 
 
 
 
 
 
 
 
 
 
417}
418
419
420struct task_struct *__sched _switch_to(struct task_struct *prev,
421				       struct task_struct *next)
422{
423	/* DMA state is already saved; save off other arch state. */
424	save_arch_state(&prev->thread);
425
426#if CHIP_HAS_TILE_DMA()
427	/*
428	 * Restore DMA in new task if desired.
429	 * Note that it is only safe to restart here since interrupts
430	 * are disabled, so we can't take any DMATLB miss or access
431	 * interrupts before we have finished switching stacks.
432	 */
433	if (next->thread.tile_dma_state.enabled) {
434		restore_tile_dma_state(&next->thread);
435		grant_dma_mpls();
436	} else {
437		restrict_dma_mpls();
438	}
439#endif
440
441	/* Restore other arch state. */
442	restore_arch_state(&next->thread);
443
 
 
 
 
 
 
 
 
 
 
 
444#ifdef CONFIG_HARDWALL
445	/* Enable or disable access to the network registers appropriately. */
446	hardwall_switch_tasks(prev, next);
447#endif
448
449	/* Notify the simulator of task exit. */
450	if (unlikely(prev->state == TASK_DEAD))
451		__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_EXIT |
452			     (prev->pid << _SIM_CONTROL_OPERATOR_BITS));
453
454	/*
455	 * Switch kernel SP, PC, and callee-saved registers.
456	 * In the context of the new task, return the old task pointer
457	 * (i.e. the task that actually called __switch_to).
458	 * Pass the value to use for SYSTEM_SAVE_K_0 when we reset our sp.
459	 */
460	return __switch_to(prev, next, next_current_ksp0(next));
461}
462
463/*
464 * This routine is called on return from interrupt if any of the
465 * TIF_ALLWORK_MASK flags are set in thread_info->flags.  It is
466 * entered with interrupts disabled so we don't miss an event that
467 * modified the thread_info flags.  We loop until all the tested flags
468 * are clear.  Note that the function is called on certain conditions
469 * that are not listed in the loop condition here (e.g. SINGLESTEP)
470 * which guarantees we will do those things once, and redo them if any
471 * of the other work items is re-done, but won't continue looping if
472 * all the other work is done.
 
 
 
473 */
474void prepare_exit_to_usermode(struct pt_regs *regs, u32 thread_info_flags)
475{
476	if (WARN_ON(!user_mode(regs)))
477		return;
478
479	do {
480		local_irq_enable();
481
482		if (thread_info_flags & _TIF_NEED_RESCHED)
483			schedule();
484
485#if CHIP_HAS_TILE_DMA()
486		if (thread_info_flags & _TIF_ASYNC_TLB)
487			do_async_page_fault(regs);
488#endif
489
490		if (thread_info_flags & _TIF_SIGPENDING)
491			do_signal(regs);
492
493		if (thread_info_flags & _TIF_NOTIFY_RESUME) {
494			clear_thread_flag(TIF_NOTIFY_RESUME);
495			tracehook_notify_resume(regs);
496		}
497
498		local_irq_disable();
499		thread_info_flags = READ_ONCE(current_thread_info()->flags);
500
501	} while (thread_info_flags & _TIF_WORK_MASK);
502
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
503	if (thread_info_flags & _TIF_SINGLESTEP) {
504		single_step_once(regs);
505#ifndef __tilegx__
506		/*
507		 * FIXME: on tilepro, since we enable interrupts in
508		 * this routine, it's possible that we miss a signal
509		 * or other asynchronous event.
510		 */
511		local_irq_disable();
512#endif
513	}
 
 
514
515	user_enter();
 
 
 
 
 
 
 
 
516}
517
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
518unsigned long get_wchan(struct task_struct *p)
519{
520	struct KBacktraceIterator kbt;
521
522	if (!p || p == current || p->state == TASK_RUNNING)
523		return 0;
524
525	for (KBacktraceIterator_init(&kbt, p, NULL);
526	     !KBacktraceIterator_end(&kbt);
527	     KBacktraceIterator_next(&kbt)) {
528		if (!in_sched_functions(kbt.it.pc))
529			return kbt.it.pc;
530	}
531
532	return 0;
533}
534
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
535/* Flush thread state. */
536void flush_thread(void)
537{
538	/* Nothing */
539}
540
541/*
542 * Free current thread data structures etc..
543 */
544void exit_thread(void)
545{
546#ifdef CONFIG_HARDWALL
547	/*
548	 * Remove the task from the list of tasks that are associated
549	 * with any live hardwalls.  (If the task that is exiting held
550	 * the last reference to a hardwall fd, it would already have
551	 * been released and deactivated at this point.)
552	 */
553	hardwall_deactivate_all(current);
554#endif
555}
556
557void tile_show_regs(struct pt_regs *regs)
558{
 
559	int i;
 
 
 
 
560#ifdef __tilegx__
561	for (i = 0; i < 17; i++)
562		pr_err(" r%-2d: "REGFMT" r%-2d: "REGFMT" r%-2d: "REGFMT"\n",
563		       i, regs->regs[i], i+18, regs->regs[i+18],
564		       i+36, regs->regs[i+36]);
565	pr_err(" r17: "REGFMT" r35: "REGFMT" tp : "REGFMT"\n",
566	       regs->regs[17], regs->regs[35], regs->tp);
567	pr_err(" sp : "REGFMT" lr : "REGFMT"\n", regs->sp, regs->lr);
568#else
569	for (i = 0; i < 13; i++)
570		pr_err(" r%-2d: "REGFMT" r%-2d: "REGFMT
571		       " r%-2d: "REGFMT" r%-2d: "REGFMT"\n",
572		       i, regs->regs[i], i+14, regs->regs[i+14],
573		       i+27, regs->regs[i+27], i+40, regs->regs[i+40]);
574	pr_err(" r13: "REGFMT" tp : "REGFMT" sp : "REGFMT" lr : "REGFMT"\n",
575	       regs->regs[13], regs->tp, regs->sp, regs->lr);
576#endif
577	pr_err(" pc : "REGFMT" ex1: %ld     faultnum: %ld flags:%s%s%s%s\n",
578	       regs->pc, regs->ex1, regs->faultnum,
579	       is_compat_task() ? " compat" : "",
580	       (regs->flags & PT_FLAGS_DISABLE_IRQ) ? " noirq" : "",
581	       !(regs->flags & PT_FLAGS_CALLER_SAVES) ? " nocallersave" : "",
582	       (regs->flags & PT_FLAGS_RESTORE_REGS) ? " restoreregs" : "");
583}
584
585void show_regs(struct pt_regs *regs)
586{
587	struct KBacktraceIterator kbt;
588
589	show_regs_print_info(KERN_DEFAULT);
590	tile_show_regs(regs);
591
592	KBacktraceIterator_init(&kbt, NULL, regs);
593	tile_show_stack(&kbt);
594}
595
596/* To ensure stack dump on tiles occurs one by one. */
597static DEFINE_SPINLOCK(backtrace_lock);
598/* To ensure no backtrace occurs before all of the stack dump are done. */
599static atomic_t backtrace_cpus;
600/* The cpu mask to avoid reentrance. */
601static struct cpumask backtrace_mask;
602
603void do_nmi_dump_stack(struct pt_regs *regs)
604{
605	int is_idle = is_idle_task(current) && !in_interrupt();
606	int cpu;
607
608	nmi_enter();
609	cpu = smp_processor_id();
610	if (WARN_ON_ONCE(!cpumask_test_and_clear_cpu(cpu, &backtrace_mask)))
611		goto done;
612
613	spin_lock(&backtrace_lock);
614	if (is_idle)
615		pr_info("CPU: %d idle\n", cpu);
616	else
617		show_regs(regs);
618	spin_unlock(&backtrace_lock);
619	atomic_dec(&backtrace_cpus);
620done:
621	nmi_exit();
622}
623
624#ifdef __tilegx__
625void arch_trigger_all_cpu_backtrace(bool self)
626{
627	struct cpumask mask;
628	HV_Coord tile;
629	unsigned int timeout;
630	int cpu;
631	int ongoing;
632	HV_NMI_Info info[NR_CPUS];
633
634	ongoing = atomic_cmpxchg(&backtrace_cpus, 0, num_online_cpus() - 1);
635	if (ongoing != 0) {
636		pr_err("Trying to do all-cpu backtrace.\n");
637		pr_err("But another all-cpu backtrace is ongoing (%d cpus left)\n",
638		       ongoing);
639		if (self) {
640			pr_err("Reporting the stack on this cpu only.\n");
641			dump_stack();
642		}
643		return;
644	}
645
646	cpumask_copy(&mask, cpu_online_mask);
647	cpumask_clear_cpu(smp_processor_id(), &mask);
648	cpumask_copy(&backtrace_mask, &mask);
649
650	/* Backtrace for myself first. */
651	if (self)
652		dump_stack();
653
654	/* Tentatively dump stack on remote tiles via NMI. */
655	timeout = 100;
656	while (!cpumask_empty(&mask) && timeout) {
657		for_each_cpu(cpu, &mask) {
658			tile.x = cpu_x(cpu);
659			tile.y = cpu_y(cpu);
660			info[cpu] = hv_send_nmi(tile, TILE_NMI_DUMP_STACK, 0);
661			if (info[cpu].result == HV_NMI_RESULT_OK)
662				cpumask_clear_cpu(cpu, &mask);
663		}
664
665		mdelay(10);
666		timeout--;
667	}
668
669	/* Warn about cpus stuck in ICS and decrement their counts here. */
670	if (!cpumask_empty(&mask)) {
671		for_each_cpu(cpu, &mask) {
672			switch (info[cpu].result) {
673			case HV_NMI_RESULT_FAIL_ICS:
674				pr_warn("Skipping stack dump of cpu %d in ICS at pc %#llx\n",
675					cpu, info[cpu].pc);
676				break;
677			case HV_NMI_RESULT_FAIL_HV:
678				pr_warn("Skipping stack dump of cpu %d in hypervisor\n",
679					cpu);
680				break;
681			case HV_ENOSYS:
682				pr_warn("Hypervisor too old to allow remote stack dumps.\n");
683				goto skip_for_each;
684			default:  /* should not happen */
685				pr_warn("Skipping stack dump of cpu %d [%d,%#llx]\n",
686					cpu, info[cpu].result, info[cpu].pc);
687				break;
688			}
689		}
690skip_for_each:
691		atomic_sub(cpumask_weight(&mask), &backtrace_cpus);
692	}
693}
694#endif /* __tilegx_ */