Loading...
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/system.h>
31#include <asm/stack.h>
32#include <asm/homecache.h>
33#include <asm/syscalls.h>
34#include <asm/traps.h>
35#ifdef CONFIG_HARDWALL
36#include <asm/hardwall.h>
37#endif
38#include <arch/chip.h>
39#include <arch/abi.h>
40
41
42/*
43 * Use the (x86) "idle=poll" option to prefer low latency when leaving the
44 * idle loop over low power while in the idle loop, e.g. if we have
45 * one thread per core and we want to get threads out of futex waits fast.
46 */
47static int no_idle_nap;
48static int __init idle_setup(char *str)
49{
50 if (!str)
51 return -EINVAL;
52
53 if (!strcmp(str, "poll")) {
54 pr_info("using polling idle threads.\n");
55 no_idle_nap = 1;
56 } else if (!strcmp(str, "halt"))
57 no_idle_nap = 0;
58 else
59 return -1;
60
61 return 0;
62}
63early_param("idle", idle_setup);
64
65/*
66 * The idle thread. There's no useful work to be
67 * done, so just try to conserve power and have a
68 * low exit latency (ie sit in a loop waiting for
69 * somebody to say that they'd like to reschedule)
70 */
71void cpu_idle(void)
72{
73 int cpu = smp_processor_id();
74
75
76 current_thread_info()->status |= TS_POLLING;
77
78 if (no_idle_nap) {
79 while (1) {
80 while (!need_resched())
81 cpu_relax();
82 schedule();
83 }
84 }
85
86 /* endless idle loop with no priority at all */
87 while (1) {
88 tick_nohz_stop_sched_tick(1);
89 while (!need_resched()) {
90 if (cpu_is_offline(cpu))
91 BUG(); /* no HOTPLUG_CPU */
92
93 local_irq_disable();
94 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
95 current_thread_info()->status &= ~TS_POLLING;
96 /*
97 * TS_POLLING-cleared state must be visible before we
98 * test NEED_RESCHED:
99 */
100 smp_mb();
101
102 if (!need_resched())
103 _cpu_idle();
104 else
105 local_irq_enable();
106 current_thread_info()->status |= TS_POLLING;
107 }
108 tick_nohz_restart_sched_tick();
109 preempt_enable_no_resched();
110 schedule();
111 preempt_disable();
112 }
113}
114
115struct thread_info *alloc_thread_info_node(struct task_struct *task, int node)
116{
117 struct page *page;
118 gfp_t flags = GFP_KERNEL;
119
120#ifdef CONFIG_DEBUG_STACK_USAGE
121 flags |= __GFP_ZERO;
122#endif
123
124 page = alloc_pages_node(node, flags, THREAD_SIZE_ORDER);
125 if (!page)
126 return NULL;
127
128 return (struct thread_info *)page_address(page);
129}
130
131/*
132 * Free a thread_info node, and all of its derivative
133 * data structures.
134 */
135void free_thread_info(struct thread_info *info)
136{
137 struct single_step_state *step_state = info->step_state;
138
139#ifdef CONFIG_HARDWALL
140 /*
141 * We free a thread_info from the context of the task that has
142 * been scheduled next, so the original task is already dead.
143 * Calling deactivate here just frees up the data structures.
144 * If the task we're freeing held the last reference to a
145 * hardwall fd, it would have been released prior to this point
146 * anyway via exit_files(), and "hardwall" would be NULL by now.
147 */
148 if (info->task->thread.hardwall)
149 hardwall_deactivate(info->task);
150#endif
151
152 if (step_state) {
153
154 /*
155 * FIXME: we don't munmap step_state->buffer
156 * because the mm_struct for this process (info->task->mm)
157 * has already been zeroed in exit_mm(). Keeping a
158 * reference to it here seems like a bad move, so this
159 * means we can't munmap() the buffer, and therefore if we
160 * ptrace multiple threads in a process, we will slowly
161 * leak user memory. (Note that as soon as the last
162 * thread in a process dies, we will reclaim all user
163 * memory including single-step buffers in the usual way.)
164 * We should either assign a kernel VA to this buffer
165 * somehow, or we should associate the buffer(s) with the
166 * mm itself so we can clean them up that way.
167 */
168 kfree(step_state);
169 }
170
171 free_pages((unsigned long)info, THREAD_SIZE_ORDER);
172}
173
174static void save_arch_state(struct thread_struct *t);
175
176int copy_thread(unsigned long clone_flags, unsigned long sp,
177 unsigned long stack_size,
178 struct task_struct *p, struct pt_regs *regs)
179{
180 struct pt_regs *childregs;
181 unsigned long ksp;
182
183 /*
184 * When creating a new kernel thread we pass sp as zero.
185 * Assign it to a reasonable value now that we have the stack.
186 */
187 if (sp == 0 && regs->ex1 == PL_ICS_EX1(KERNEL_PL, 0))
188 sp = KSTK_TOP(p);
189
190 /*
191 * Do not clone step state from the parent; each thread
192 * must make its own lazily.
193 */
194 task_thread_info(p)->step_state = NULL;
195
196 /*
197 * Start new thread in ret_from_fork so it schedules properly
198 * and then return from interrupt like the parent.
199 */
200 p->thread.pc = (unsigned long) ret_from_fork;
201
202 /* Save user stack top pointer so we can ID the stack vm area later. */
203 p->thread.usp0 = sp;
204
205 /* Record the pid of the process that created this one. */
206 p->thread.creator_pid = current->pid;
207
208 /*
209 * Copy the registers onto the kernel stack so the
210 * return-from-interrupt code will reload it into registers.
211 */
212 childregs = task_pt_regs(p);
213 *childregs = *regs;
214 childregs->regs[0] = 0; /* return value is zero */
215 childregs->sp = sp; /* override with new user stack pointer */
216
217 /*
218 * If CLONE_SETTLS is set, set "tp" in the new task to "r4",
219 * which is passed in as arg #5 to sys_clone().
220 */
221 if (clone_flags & CLONE_SETTLS)
222 childregs->tp = regs->regs[4];
223
224 /*
225 * Copy the callee-saved registers from the passed pt_regs struct
226 * into the context-switch callee-saved registers area.
227 * This way when we start the interrupt-return sequence, the
228 * callee-save registers will be correctly in registers, which
229 * is how we assume the compiler leaves them as we start doing
230 * the normal return-from-interrupt path after calling C code.
231 * Zero out the C ABI save area to mark the top of the stack.
232 */
233 ksp = (unsigned long) childregs;
234 ksp -= C_ABI_SAVE_AREA_SIZE; /* interrupt-entry save area */
235 ((long *)ksp)[0] = ((long *)ksp)[1] = 0;
236 ksp -= CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long);
237 memcpy((void *)ksp, ®s->regs[CALLEE_SAVED_FIRST_REG],
238 CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long));
239 ksp -= C_ABI_SAVE_AREA_SIZE; /* __switch_to() save area */
240 ((long *)ksp)[0] = ((long *)ksp)[1] = 0;
241 p->thread.ksp = ksp;
242
243#if CHIP_HAS_TILE_DMA()
244 /*
245 * No DMA in the new thread. We model this on the fact that
246 * fork() clears the pending signals, alarms, and aio for the child.
247 */
248 memset(&p->thread.tile_dma_state, 0, sizeof(struct tile_dma_state));
249 memset(&p->thread.dma_async_tlb, 0, sizeof(struct async_tlb));
250#endif
251
252#if CHIP_HAS_SN_PROC()
253 /* Likewise, the new thread is not running static processor code. */
254 p->thread.sn_proc_running = 0;
255 memset(&p->thread.sn_async_tlb, 0, sizeof(struct async_tlb));
256#endif
257
258#if CHIP_HAS_PROC_STATUS_SPR()
259 /* New thread has its miscellaneous processor state bits clear. */
260 p->thread.proc_status = 0;
261#endif
262
263#ifdef CONFIG_HARDWALL
264 /* New thread does not own any networks. */
265 p->thread.hardwall = NULL;
266#endif
267
268
269 /*
270 * Start the new thread with the current architecture state
271 * (user interrupt masks, etc.).
272 */
273 save_arch_state(&p->thread);
274
275 return 0;
276}
277
278/*
279 * Return "current" if it looks plausible, or else a pointer to a dummy.
280 * This can be helpful if we are just trying to emit a clean panic.
281 */
282struct task_struct *validate_current(void)
283{
284 static struct task_struct corrupt = { .comm = "<corrupt>" };
285 struct task_struct *tsk = current;
286 if (unlikely((unsigned long)tsk < PAGE_OFFSET ||
287 (void *)tsk > high_memory ||
288 ((unsigned long)tsk & (__alignof__(*tsk) - 1)) != 0)) {
289 pr_err("Corrupt 'current' %p (sp %#lx)\n", tsk, stack_pointer);
290 tsk = &corrupt;
291 }
292 return tsk;
293}
294
295/* Take and return the pointer to the previous task, for schedule_tail(). */
296struct task_struct *sim_notify_fork(struct task_struct *prev)
297{
298 struct task_struct *tsk = current;
299 __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK_PARENT |
300 (tsk->thread.creator_pid << _SIM_CONTROL_OPERATOR_BITS));
301 __insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK |
302 (tsk->pid << _SIM_CONTROL_OPERATOR_BITS));
303 return prev;
304}
305
306int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
307{
308 struct pt_regs *ptregs = task_pt_regs(tsk);
309 elf_core_copy_regs(regs, ptregs);
310 return 1;
311}
312
313#if CHIP_HAS_TILE_DMA()
314
315/* Allow user processes to access the DMA SPRs */
316void grant_dma_mpls(void)
317{
318#if CONFIG_KERNEL_PL == 2
319 __insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
320 __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
321#else
322 __insn_mtspr(SPR_MPL_DMA_CPL_SET_0, 1);
323 __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0, 1);
324#endif
325}
326
327/* Forbid user processes from accessing the DMA SPRs */
328void restrict_dma_mpls(void)
329{
330#if CONFIG_KERNEL_PL == 2
331 __insn_mtspr(SPR_MPL_DMA_CPL_SET_2, 1);
332 __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_2, 1);
333#else
334 __insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
335 __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
336#endif
337}
338
339/* Pause the DMA engine, then save off its state registers. */
340static void save_tile_dma_state(struct tile_dma_state *dma)
341{
342 unsigned long state = __insn_mfspr(SPR_DMA_USER_STATUS);
343 unsigned long post_suspend_state;
344
345 /* If we're running, suspend the engine. */
346 if ((state & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK)
347 __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__SUSPEND_MASK);
348
349 /*
350 * Wait for the engine to idle, then save regs. Note that we
351 * want to record the "running" bit from before suspension,
352 * and the "done" bit from after, so that we can properly
353 * distinguish a case where the user suspended the engine from
354 * the case where the kernel suspended as part of the context
355 * swap.
356 */
357 do {
358 post_suspend_state = __insn_mfspr(SPR_DMA_USER_STATUS);
359 } while (post_suspend_state & SPR_DMA_STATUS__BUSY_MASK);
360
361 dma->src = __insn_mfspr(SPR_DMA_SRC_ADDR);
362 dma->src_chunk = __insn_mfspr(SPR_DMA_SRC_CHUNK_ADDR);
363 dma->dest = __insn_mfspr(SPR_DMA_DST_ADDR);
364 dma->dest_chunk = __insn_mfspr(SPR_DMA_DST_CHUNK_ADDR);
365 dma->strides = __insn_mfspr(SPR_DMA_STRIDE);
366 dma->chunk_size = __insn_mfspr(SPR_DMA_CHUNK_SIZE);
367 dma->byte = __insn_mfspr(SPR_DMA_BYTE);
368 dma->status = (state & SPR_DMA_STATUS__RUNNING_MASK) |
369 (post_suspend_state & SPR_DMA_STATUS__DONE_MASK);
370}
371
372/* Restart a DMA that was running before we were context-switched out. */
373static void restore_tile_dma_state(struct thread_struct *t)
374{
375 const struct tile_dma_state *dma = &t->tile_dma_state;
376
377 /*
378 * The only way to restore the done bit is to run a zero
379 * length transaction.
380 */
381 if ((dma->status & SPR_DMA_STATUS__DONE_MASK) &&
382 !(__insn_mfspr(SPR_DMA_USER_STATUS) & SPR_DMA_STATUS__DONE_MASK)) {
383 __insn_mtspr(SPR_DMA_BYTE, 0);
384 __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK);
385 while (__insn_mfspr(SPR_DMA_USER_STATUS) &
386 SPR_DMA_STATUS__BUSY_MASK)
387 ;
388 }
389
390 __insn_mtspr(SPR_DMA_SRC_ADDR, dma->src);
391 __insn_mtspr(SPR_DMA_SRC_CHUNK_ADDR, dma->src_chunk);
392 __insn_mtspr(SPR_DMA_DST_ADDR, dma->dest);
393 __insn_mtspr(SPR_DMA_DST_CHUNK_ADDR, dma->dest_chunk);
394 __insn_mtspr(SPR_DMA_STRIDE, dma->strides);
395 __insn_mtspr(SPR_DMA_CHUNK_SIZE, dma->chunk_size);
396 __insn_mtspr(SPR_DMA_BYTE, dma->byte);
397
398 /*
399 * Restart the engine if we were running and not done.
400 * Clear a pending async DMA fault that we were waiting on return
401 * to user space to execute, since we expect the DMA engine
402 * to regenerate those faults for us now. Note that we don't
403 * try to clear the TIF_ASYNC_TLB flag, since it's relatively
404 * harmless if set, and it covers both DMA and the SN processor.
405 */
406 if ((dma->status & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK) {
407 t->dma_async_tlb.fault_num = 0;
408 __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK);
409 }
410}
411
412#endif
413
414static void save_arch_state(struct thread_struct *t)
415{
416#if CHIP_HAS_SPLIT_INTR_MASK()
417 t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0_0) |
418 ((u64)__insn_mfspr(SPR_INTERRUPT_MASK_0_1) << 32);
419#else
420 t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0);
421#endif
422 t->ex_context[0] = __insn_mfspr(SPR_EX_CONTEXT_0_0);
423 t->ex_context[1] = __insn_mfspr(SPR_EX_CONTEXT_0_1);
424 t->system_save[0] = __insn_mfspr(SPR_SYSTEM_SAVE_0_0);
425 t->system_save[1] = __insn_mfspr(SPR_SYSTEM_SAVE_0_1);
426 t->system_save[2] = __insn_mfspr(SPR_SYSTEM_SAVE_0_2);
427 t->system_save[3] = __insn_mfspr(SPR_SYSTEM_SAVE_0_3);
428 t->intctrl_0 = __insn_mfspr(SPR_INTCTRL_0_STATUS);
429#if CHIP_HAS_PROC_STATUS_SPR()
430 t->proc_status = __insn_mfspr(SPR_PROC_STATUS);
431#endif
432#if !CHIP_HAS_FIXED_INTVEC_BASE()
433 t->interrupt_vector_base = __insn_mfspr(SPR_INTERRUPT_VECTOR_BASE_0);
434#endif
435#if CHIP_HAS_TILE_RTF_HWM()
436 t->tile_rtf_hwm = __insn_mfspr(SPR_TILE_RTF_HWM);
437#endif
438#if CHIP_HAS_DSTREAM_PF()
439 t->dstream_pf = __insn_mfspr(SPR_DSTREAM_PF);
440#endif
441}
442
443static void restore_arch_state(const struct thread_struct *t)
444{
445#if CHIP_HAS_SPLIT_INTR_MASK()
446 __insn_mtspr(SPR_INTERRUPT_MASK_0_0, (u32) t->interrupt_mask);
447 __insn_mtspr(SPR_INTERRUPT_MASK_0_1, t->interrupt_mask >> 32);
448#else
449 __insn_mtspr(SPR_INTERRUPT_MASK_0, t->interrupt_mask);
450#endif
451 __insn_mtspr(SPR_EX_CONTEXT_0_0, t->ex_context[0]);
452 __insn_mtspr(SPR_EX_CONTEXT_0_1, t->ex_context[1]);
453 __insn_mtspr(SPR_SYSTEM_SAVE_0_0, t->system_save[0]);
454 __insn_mtspr(SPR_SYSTEM_SAVE_0_1, t->system_save[1]);
455 __insn_mtspr(SPR_SYSTEM_SAVE_0_2, t->system_save[2]);
456 __insn_mtspr(SPR_SYSTEM_SAVE_0_3, t->system_save[3]);
457 __insn_mtspr(SPR_INTCTRL_0_STATUS, t->intctrl_0);
458#if CHIP_HAS_PROC_STATUS_SPR()
459 __insn_mtspr(SPR_PROC_STATUS, t->proc_status);
460#endif
461#if !CHIP_HAS_FIXED_INTVEC_BASE()
462 __insn_mtspr(SPR_INTERRUPT_VECTOR_BASE_0, t->interrupt_vector_base);
463#endif
464#if CHIP_HAS_TILE_RTF_HWM()
465 __insn_mtspr(SPR_TILE_RTF_HWM, t->tile_rtf_hwm);
466#endif
467#if CHIP_HAS_DSTREAM_PF()
468 __insn_mtspr(SPR_DSTREAM_PF, t->dstream_pf);
469#endif
470}
471
472
473void _prepare_arch_switch(struct task_struct *next)
474{
475#if CHIP_HAS_SN_PROC()
476 int snctl;
477#endif
478#if CHIP_HAS_TILE_DMA()
479 struct tile_dma_state *dma = ¤t->thread.tile_dma_state;
480 if (dma->enabled)
481 save_tile_dma_state(dma);
482#endif
483#if CHIP_HAS_SN_PROC()
484 /*
485 * Suspend the static network processor if it was running.
486 * We do not suspend the fabric itself, just like we don't
487 * try to suspend the UDN.
488 */
489 snctl = __insn_mfspr(SPR_SNCTL);
490 current->thread.sn_proc_running =
491 (snctl & SPR_SNCTL__FRZPROC_MASK) == 0;
492 if (current->thread.sn_proc_running)
493 __insn_mtspr(SPR_SNCTL, snctl | SPR_SNCTL__FRZPROC_MASK);
494#endif
495}
496
497
498struct task_struct *__sched _switch_to(struct task_struct *prev,
499 struct task_struct *next)
500{
501 /* DMA state is already saved; save off other arch state. */
502 save_arch_state(&prev->thread);
503
504#if CHIP_HAS_TILE_DMA()
505 /*
506 * Restore DMA in new task if desired.
507 * Note that it is only safe to restart here since interrupts
508 * are disabled, so we can't take any DMATLB miss or access
509 * interrupts before we have finished switching stacks.
510 */
511 if (next->thread.tile_dma_state.enabled) {
512 restore_tile_dma_state(&next->thread);
513 grant_dma_mpls();
514 } else {
515 restrict_dma_mpls();
516 }
517#endif
518
519 /* Restore other arch state. */
520 restore_arch_state(&next->thread);
521
522#if CHIP_HAS_SN_PROC()
523 /*
524 * Restart static network processor in the new process
525 * if it was running before.
526 */
527 if (next->thread.sn_proc_running) {
528 int snctl = __insn_mfspr(SPR_SNCTL);
529 __insn_mtspr(SPR_SNCTL, snctl & ~SPR_SNCTL__FRZPROC_MASK);
530 }
531#endif
532
533#ifdef CONFIG_HARDWALL
534 /* Enable or disable access to the network registers appropriately. */
535 if (prev->thread.hardwall != NULL) {
536 if (next->thread.hardwall == NULL)
537 restrict_network_mpls();
538 } else if (next->thread.hardwall != NULL) {
539 grant_network_mpls();
540 }
541#endif
542
543 /*
544 * Switch kernel SP, PC, and callee-saved registers.
545 * In the context of the new task, return the old task pointer
546 * (i.e. the task that actually called __switch_to).
547 * Pass the value to use for SYSTEM_SAVE_K_0 when we reset our sp.
548 */
549 return __switch_to(prev, next, next_current_ksp0(next));
550}
551
552/*
553 * This routine is called on return from interrupt if any of the
554 * TIF_WORK_MASK flags are set in thread_info->flags. It is
555 * entered with interrupts disabled so we don't miss an event
556 * that modified the thread_info flags. If any flag is set, we
557 * handle it and return, and the calling assembly code will
558 * re-disable interrupts, reload the thread flags, and call back
559 * if more flags need to be handled.
560 *
561 * We return whether we need to check the thread_info flags again
562 * or not. Note that we don't clear TIF_SINGLESTEP here, so it's
563 * important that it be tested last, and then claim that we don't
564 * need to recheck the flags.
565 */
566int do_work_pending(struct pt_regs *regs, u32 thread_info_flags)
567{
568 if (thread_info_flags & _TIF_NEED_RESCHED) {
569 schedule();
570 return 1;
571 }
572#if CHIP_HAS_TILE_DMA() || CHIP_HAS_SN_PROC()
573 if (thread_info_flags & _TIF_ASYNC_TLB) {
574 do_async_page_fault(regs);
575 return 1;
576 }
577#endif
578 if (thread_info_flags & _TIF_SIGPENDING) {
579 do_signal(regs);
580 return 1;
581 }
582 if (thread_info_flags & _TIF_NOTIFY_RESUME) {
583 clear_thread_flag(TIF_NOTIFY_RESUME);
584 tracehook_notify_resume(regs);
585 if (current->replacement_session_keyring)
586 key_replace_session_keyring();
587 return 1;
588 }
589 if (thread_info_flags & _TIF_SINGLESTEP) {
590 if ((regs->ex1 & SPR_EX_CONTEXT_1_1__PL_MASK) == 0)
591 single_step_once(regs);
592 return 0;
593 }
594 panic("work_pending: bad flags %#x\n", thread_info_flags);
595}
596
597/* Note there is an implicit fifth argument if (clone_flags & CLONE_SETTLS). */
598SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
599 void __user *, parent_tidptr, void __user *, child_tidptr,
600 struct pt_regs *, regs)
601{
602 if (!newsp)
603 newsp = regs->sp;
604 return do_fork(clone_flags, newsp, regs, 0,
605 parent_tidptr, child_tidptr);
606}
607
608/*
609 * sys_execve() executes a new program.
610 */
611SYSCALL_DEFINE4(execve, const char __user *, path,
612 const char __user *const __user *, argv,
613 const char __user *const __user *, envp,
614 struct pt_regs *, regs)
615{
616 long error;
617 char *filename;
618
619 filename = getname(path);
620 error = PTR_ERR(filename);
621 if (IS_ERR(filename))
622 goto out;
623 error = do_execve(filename, argv, envp, regs);
624 putname(filename);
625 if (error == 0)
626 single_step_execve();
627out:
628 return error;
629}
630
631#ifdef CONFIG_COMPAT
632long compat_sys_execve(const char __user *path,
633 compat_uptr_t __user *argv,
634 compat_uptr_t __user *envp,
635 struct pt_regs *regs)
636{
637 long error;
638 char *filename;
639
640 filename = getname(path);
641 error = PTR_ERR(filename);
642 if (IS_ERR(filename))
643 goto out;
644 error = compat_do_execve(filename, argv, envp, regs);
645 putname(filename);
646 if (error == 0)
647 single_step_execve();
648out:
649 return error;
650}
651#endif
652
653unsigned long get_wchan(struct task_struct *p)
654{
655 struct KBacktraceIterator kbt;
656
657 if (!p || p == current || p->state == TASK_RUNNING)
658 return 0;
659
660 for (KBacktraceIterator_init(&kbt, p, NULL);
661 !KBacktraceIterator_end(&kbt);
662 KBacktraceIterator_next(&kbt)) {
663 if (!in_sched_functions(kbt.it.pc))
664 return kbt.it.pc;
665 }
666
667 return 0;
668}
669
670/*
671 * We pass in lr as zero (cleared in kernel_thread) and the caller
672 * part of the backtrace ABI on the stack also zeroed (in copy_thread)
673 * so that backtraces will stop with this function.
674 * Note that we don't use r0, since copy_thread() clears it.
675 */
676static void start_kernel_thread(int dummy, int (*fn)(int), int arg)
677{
678 do_exit(fn(arg));
679}
680
681/*
682 * Create a kernel thread
683 */
684int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
685{
686 struct pt_regs regs;
687
688 memset(®s, 0, sizeof(regs));
689 regs.ex1 = PL_ICS_EX1(KERNEL_PL, 0); /* run at kernel PL, no ICS */
690 regs.pc = (long) start_kernel_thread;
691 regs.flags = PT_FLAGS_CALLER_SAVES; /* need to restore r1 and r2 */
692 regs.regs[1] = (long) fn; /* function pointer */
693 regs.regs[2] = (long) arg; /* parameter register */
694
695 /* Ok, create the new process.. */
696 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s,
697 0, NULL, NULL);
698}
699EXPORT_SYMBOL(kernel_thread);
700
701/* Flush thread state. */
702void flush_thread(void)
703{
704 /* Nothing */
705}
706
707/*
708 * Free current thread data structures etc..
709 */
710void exit_thread(void)
711{
712 /* Nothing */
713}
714
715void show_regs(struct pt_regs *regs)
716{
717 struct task_struct *tsk = validate_current();
718 int i;
719
720 pr_err("\n");
721 pr_err(" Pid: %d, comm: %20s, CPU: %d\n",
722 tsk->pid, tsk->comm, smp_processor_id());
723#ifdef __tilegx__
724 for (i = 0; i < 51; i += 3)
725 pr_err(" r%-2d: "REGFMT" r%-2d: "REGFMT" r%-2d: "REGFMT"\n",
726 i, regs->regs[i], i+1, regs->regs[i+1],
727 i+2, regs->regs[i+2]);
728 pr_err(" r51: "REGFMT" r52: "REGFMT" tp : "REGFMT"\n",
729 regs->regs[51], regs->regs[52], regs->tp);
730 pr_err(" sp : "REGFMT" lr : "REGFMT"\n", regs->sp, regs->lr);
731#else
732 for (i = 0; i < 52; i += 4)
733 pr_err(" r%-2d: "REGFMT" r%-2d: "REGFMT
734 " r%-2d: "REGFMT" r%-2d: "REGFMT"\n",
735 i, regs->regs[i], i+1, regs->regs[i+1],
736 i+2, regs->regs[i+2], i+3, regs->regs[i+3]);
737 pr_err(" r52: "REGFMT" tp : "REGFMT" sp : "REGFMT" lr : "REGFMT"\n",
738 regs->regs[52], regs->tp, regs->sp, regs->lr);
739#endif
740 pr_err(" pc : "REGFMT" ex1: %ld faultnum: %ld\n",
741 regs->pc, regs->ex1, regs->faultnum);
742
743 dump_stack_regs(regs);
744}
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, ®s->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 = ¤t->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(®s, 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, ®s,
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}