2 * Copyright 2010 Tilera Corporation. All Rights Reserved.
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.
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
15 #include <linux/sched.h>
16 #include <linux/preempt.h>
17 #include <linux/module.h>
19 #include <linux/kprobes.h>
20 #include <linux/elfcore.h>
21 #include <linux/tick.h>
22 #include <linux/init.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/context_tracking.h>
31 #include <asm/stack.h>
32 #include <asm/switch_to.h>
33 #include <asm/homecache.h>
34 #include <asm/syscalls.h>
35 #include <asm/traps.h>
36 #include <asm/setup.h>
37 #include <asm/uaccess.h>
38 #ifdef CONFIG_HARDWALL
39 #include <asm/hardwall.h>
41 #include <arch/chip.h>
43 #include <arch/sim_def.h>
46 * Use the (x86) "idle=poll" option to prefer low latency when leaving the
47 * idle loop over low power while in the idle loop, e.g. if we have
48 * one thread per core and we want to get threads out of futex waits fast.
50 static int __init idle_setup(char *str)
55 if (!strcmp(str, "poll")) {
56 pr_info("using polling idle threads\n");
57 cpu_idle_poll_ctrl(true);
59 } else if (!strcmp(str, "halt")) {
64 early_param("idle", idle_setup);
66 void arch_cpu_idle(void)
68 __this_cpu_write(irq_stat.idle_timestamp, jiffies);
73 * Release a thread_info structure
75 void arch_release_thread_info(struct thread_info *info)
77 struct single_step_state *step_state = info->step_state;
82 * FIXME: we don't munmap step_state->buffer
83 * because the mm_struct for this process (info->task->mm)
84 * has already been zeroed in exit_mm(). Keeping a
85 * reference to it here seems like a bad move, so this
86 * means we can't munmap() the buffer, and therefore if we
87 * ptrace multiple threads in a process, we will slowly
88 * leak user memory. (Note that as soon as the last
89 * thread in a process dies, we will reclaim all user
90 * memory including single-step buffers in the usual way.)
91 * We should either assign a kernel VA to this buffer
92 * somehow, or we should associate the buffer(s) with the
93 * mm itself so we can clean them up that way.
99 static void save_arch_state(struct thread_struct *t);
101 int copy_thread(unsigned long clone_flags, unsigned long sp,
102 unsigned long arg, struct task_struct *p)
104 struct pt_regs *childregs = task_pt_regs(p);
106 unsigned long *callee_regs;
109 * Set up the stack and stack pointer appropriately for the
110 * new child to find itself woken up in __switch_to().
111 * The callee-saved registers must be on the stack to be read;
112 * the new task will then jump to assembly support to handle
113 * calling schedule_tail(), etc., and (for userspace tasks)
114 * returning to the context set up in the pt_regs.
116 ksp = (unsigned long) childregs;
117 ksp -= C_ABI_SAVE_AREA_SIZE; /* interrupt-entry save area */
118 ((long *)ksp)[0] = ((long *)ksp)[1] = 0;
119 ksp -= CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long);
120 callee_regs = (unsigned long *)ksp;
121 ksp -= C_ABI_SAVE_AREA_SIZE; /* __switch_to() save area */
122 ((long *)ksp)[0] = ((long *)ksp)[1] = 0;
125 /* Record the pid of the task that created this one. */
126 p->thread.creator_pid = current->pid;
128 if (unlikely(p->flags & PF_KTHREAD)) {
130 memset(childregs, 0, sizeof(struct pt_regs));
131 memset(&callee_regs[2], 0,
132 (CALLEE_SAVED_REGS_COUNT - 2) * sizeof(unsigned long));
133 callee_regs[0] = sp; /* r30 = function */
134 callee_regs[1] = arg; /* r31 = arg */
135 childregs->ex1 = PL_ICS_EX1(KERNEL_PL, 0);
136 p->thread.pc = (unsigned long) ret_from_kernel_thread;
141 * Start new thread in ret_from_fork so it schedules properly
142 * and then return from interrupt like the parent.
144 p->thread.pc = (unsigned long) ret_from_fork;
147 * Do not clone step state from the parent; each thread
148 * must make its own lazily.
150 task_thread_info(p)->step_state = NULL;
154 * Do not clone unalign jit fixup from the parent; each thread
155 * must allocate its own on demand.
157 task_thread_info(p)->unalign_jit_base = NULL;
161 * Copy the registers onto the kernel stack so the
162 * return-from-interrupt code will reload it into registers.
164 *childregs = *current_pt_regs();
165 childregs->regs[0] = 0; /* return value is zero */
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));
171 /* Save user stack top pointer so we can ID the stack vm area later. */
172 p->thread.usp0 = childregs->sp;
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().
178 if (clone_flags & CLONE_SETTLS)
179 childregs->tp = childregs->regs[4];
182 #if CHIP_HAS_TILE_DMA()
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.
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));
191 /* New thread has its miscellaneous processor state bits clear. */
192 p->thread.proc_status = 0;
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);
202 * Start the new thread with the current architecture state
203 * (user interrupt masks, etc.).
205 save_arch_state(&p->thread);
210 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
212 task_thread_info(tsk)->align_ctl = val;
216 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
218 return put_user(task_thread_info(tsk)->align_ctl,
219 (unsigned int __user *)adr);
222 static struct task_struct corrupt_current = { .comm = "<corrupt>" };
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.
228 struct task_struct *validate_current(void)
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;
240 /* Take and return the pointer to the previous task, for schedule_tail(). */
241 struct task_struct *sim_notify_fork(struct task_struct *prev)
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));
251 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
253 struct pt_regs *ptregs = task_pt_regs(tsk);
254 elf_core_copy_regs(regs, ptregs);
258 #if CHIP_HAS_TILE_DMA()
260 /* Allow user processes to access the DMA SPRs */
261 void grant_dma_mpls(void)
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);
267 __insn_mtspr(SPR_MPL_DMA_CPL_SET_0, 1);
268 __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0, 1);
272 /* Forbid user processes from accessing the DMA SPRs */
273 void restrict_dma_mpls(void)
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);
279 __insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
280 __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
284 /* Pause the DMA engine, then save off its state registers. */
285 static void save_tile_dma_state(struct tile_dma_state *dma)
287 unsigned long state = __insn_mfspr(SPR_DMA_USER_STATUS);
288 unsigned long post_suspend_state;
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);
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
303 post_suspend_state = __insn_mfspr(SPR_DMA_USER_STATUS);
304 } while (post_suspend_state & SPR_DMA_STATUS__BUSY_MASK);
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);
317 /* Restart a DMA that was running before we were context-switched out. */
318 static void restore_tile_dma_state(struct thread_struct *t)
320 const struct tile_dma_state *dma = &t->tile_dma_state;
323 * The only way to restore the done bit is to run a zero
324 * length transaction.
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)
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);
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.
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);
359 static void save_arch_state(struct thread_struct *t)
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);
365 t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0);
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);
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);
384 static void restore_arch_state(const struct thread_struct *t)
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);
390 __insn_mtspr(SPR_INTERRUPT_MASK_0, t->interrupt_mask);
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);
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);
410 void _prepare_arch_switch(struct task_struct *next)
412 #if CHIP_HAS_TILE_DMA()
413 struct tile_dma_state *dma = ¤t->thread.tile_dma_state;
415 save_tile_dma_state(dma);
420 struct task_struct *__sched _switch_to(struct task_struct *prev,
421 struct task_struct *next)
423 /* DMA state is already saved; save off other arch state. */
424 save_arch_state(&prev->thread);
426 #if CHIP_HAS_TILE_DMA()
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.
433 if (next->thread.tile_dma_state.enabled) {
434 restore_tile_dma_state(&next->thread);
441 /* Restore other arch state. */
442 restore_arch_state(&next->thread);
444 #ifdef CONFIG_HARDWALL
445 /* Enable or disable access to the network registers appropriately. */
446 hardwall_switch_tasks(prev, next);
450 * Switch kernel SP, PC, and callee-saved registers.
451 * In the context of the new task, return the old task pointer
452 * (i.e. the task that actually called __switch_to).
453 * Pass the value to use for SYSTEM_SAVE_K_0 when we reset our sp.
455 return __switch_to(prev, next, next_current_ksp0(next));
459 * This routine is called on return from interrupt if any of the
460 * TIF_WORK_MASK flags are set in thread_info->flags. It is
461 * entered with interrupts disabled so we don't miss an event
462 * that modified the thread_info flags. If any flag is set, we
463 * handle it and return, and the calling assembly code will
464 * re-disable interrupts, reload the thread flags, and call back
465 * if more flags need to be handled.
467 * We return whether we need to check the thread_info flags again
468 * or not. Note that we don't clear TIF_SINGLESTEP here, so it's
469 * important that it be tested last, and then claim that we don't
470 * need to recheck the flags.
472 int do_work_pending(struct pt_regs *regs, u32 thread_info_flags)
474 /* If we enter in kernel mode, do nothing and exit the caller loop. */
475 if (!user_mode(regs))
480 /* Enable interrupts; they are disabled again on return to caller. */
483 if (thread_info_flags & _TIF_NEED_RESCHED) {
487 #if CHIP_HAS_TILE_DMA()
488 if (thread_info_flags & _TIF_ASYNC_TLB) {
489 do_async_page_fault(regs);
493 if (thread_info_flags & _TIF_SIGPENDING) {
497 if (thread_info_flags & _TIF_NOTIFY_RESUME) {
498 clear_thread_flag(TIF_NOTIFY_RESUME);
499 tracehook_notify_resume(regs);
502 if (thread_info_flags & _TIF_SINGLESTEP)
503 single_step_once(regs);
510 unsigned long get_wchan(struct task_struct *p)
512 struct KBacktraceIterator kbt;
514 if (!p || p == current || p->state == TASK_RUNNING)
517 for (KBacktraceIterator_init(&kbt, p, NULL);
518 !KBacktraceIterator_end(&kbt);
519 KBacktraceIterator_next(&kbt)) {
520 if (!in_sched_functions(kbt.it.pc))
527 /* Flush thread state. */
528 void flush_thread(void)
534 * Free current thread data structures etc..
536 void exit_thread(void)
538 #ifdef CONFIG_HARDWALL
540 * Remove the task from the list of tasks that are associated
541 * with any live hardwalls. (If the task that is exiting held
542 * the last reference to a hardwall fd, it would already have
543 * been released and deactivated at this point.)
545 hardwall_deactivate_all(current);
549 void show_regs(struct pt_regs *regs)
551 struct task_struct *tsk = validate_current();
554 if (tsk != &corrupt_current)
555 show_regs_print_info(KERN_ERR);
557 for (i = 0; i < 17; i++)
558 pr_err(" r%-2d: " REGFMT " r%-2d: " REGFMT " r%-2d: " REGFMT "\n",
559 i, regs->regs[i], i+18, regs->regs[i+18],
560 i+36, regs->regs[i+36]);
561 pr_err(" r17: " REGFMT " r35: " REGFMT " tp : " REGFMT "\n",
562 regs->regs[17], regs->regs[35], regs->tp);
563 pr_err(" sp : " REGFMT " lr : " REGFMT "\n", regs->sp, regs->lr);
565 for (i = 0; i < 13; i++)
566 pr_err(" r%-2d: " REGFMT " r%-2d: " REGFMT " r%-2d: " REGFMT " r%-2d: " REGFMT "\n",
567 i, regs->regs[i], i+14, regs->regs[i+14],
568 i+27, regs->regs[i+27], i+40, regs->regs[i+40]);
569 pr_err(" r13: " REGFMT " tp : " REGFMT " sp : " REGFMT " lr : " REGFMT "\n",
570 regs->regs[13], regs->tp, regs->sp, regs->lr);
572 pr_err(" pc : " REGFMT " ex1: %ld faultnum: %ld\n",
573 regs->pc, regs->ex1, regs->faultnum);
575 dump_stack_regs(regs);