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 <asm/system.h>
29 #include <asm/stack.h>
30 #include <asm/homecache.h>
31 #include <asm/syscalls.h>
32 #ifdef CONFIG_HARDWALL
33 #include <asm/hardwall.h>
35 #include <arch/chip.h>
40 * Use the (x86) "idle=poll" option to prefer low latency when leaving the
41 * idle loop over low power while in the idle loop, e.g. if we have
42 * one thread per core and we want to get threads out of futex waits fast.
44 static int no_idle_nap
;
45 static int __init
idle_setup(char *str
)
50 if (!strcmp(str
, "poll")) {
51 pr_info("using polling idle threads.\n");
53 } else if (!strcmp(str
, "halt"))
60 early_param("idle", idle_setup
);
63 * The idle thread. There's no useful work to be
64 * done, so just try to conserve power and have a
65 * low exit latency (ie sit in a loop waiting for
66 * somebody to say that they'd like to reschedule)
70 int cpu
= smp_processor_id();
73 current_thread_info()->status
|= TS_POLLING
;
77 while (!need_resched())
83 /* endless idle loop with no priority at all */
85 tick_nohz_stop_sched_tick(1);
86 while (!need_resched()) {
87 if (cpu_is_offline(cpu
))
88 BUG(); /* no HOTPLUG_CPU */
91 __get_cpu_var(irq_stat
).idle_timestamp
= jiffies
;
92 current_thread_info()->status
&= ~TS_POLLING
;
94 * TS_POLLING-cleared state must be visible before we
103 current_thread_info()->status
|= TS_POLLING
;
105 tick_nohz_restart_sched_tick();
106 preempt_enable_no_resched();
112 struct thread_info
*alloc_thread_info(struct task_struct
*task
)
115 gfp_t flags
= GFP_KERNEL
;
117 #ifdef CONFIG_DEBUG_STACK_USAGE
121 page
= alloc_pages(flags
, THREAD_SIZE_ORDER
);
125 return (struct thread_info
*)page_address(page
);
129 * Free a thread_info node, and all of its derivative
132 void free_thread_info(struct thread_info
*info
)
134 struct single_step_state
*step_state
= info
->step_state
;
136 #ifdef CONFIG_HARDWALL
138 * We free a thread_info from the context of the task that has
139 * been scheduled next, so the original task is already dead.
140 * Calling deactivate here just frees up the data structures.
141 * If the task we're freeing held the last reference to a
142 * hardwall fd, it would have been released prior to this point
143 * anyway via exit_files(), and "hardwall" would be NULL by now.
145 if (info
->task
->thread
.hardwall
)
146 hardwall_deactivate(info
->task
);
152 * FIXME: we don't munmap step_state->buffer
153 * because the mm_struct for this process (info->task->mm)
154 * has already been zeroed in exit_mm(). Keeping a
155 * reference to it here seems like a bad move, so this
156 * means we can't munmap() the buffer, and therefore if we
157 * ptrace multiple threads in a process, we will slowly
158 * leak user memory. (Note that as soon as the last
159 * thread in a process dies, we will reclaim all user
160 * memory including single-step buffers in the usual way.)
161 * We should either assign a kernel VA to this buffer
162 * somehow, or we should associate the buffer(s) with the
163 * mm itself so we can clean them up that way.
168 free_page((unsigned long)info
);
171 static void save_arch_state(struct thread_struct
*t
);
173 int copy_thread(unsigned long clone_flags
, unsigned long sp
,
174 unsigned long stack_size
,
175 struct task_struct
*p
, struct pt_regs
*regs
)
177 struct pt_regs
*childregs
;
181 * When creating a new kernel thread we pass sp as zero.
182 * Assign it to a reasonable value now that we have the stack.
184 if (sp
== 0 && regs
->ex1
== PL_ICS_EX1(KERNEL_PL
, 0))
188 * Do not clone step state from the parent; each thread
189 * must make its own lazily.
191 task_thread_info(p
)->step_state
= NULL
;
194 * Start new thread in ret_from_fork so it schedules properly
195 * and then return from interrupt like the parent.
197 p
->thread
.pc
= (unsigned long) ret_from_fork
;
199 /* Save user stack top pointer so we can ID the stack vm area later. */
202 /* Record the pid of the process that created this one. */
203 p
->thread
.creator_pid
= current
->pid
;
206 * Copy the registers onto the kernel stack so the
207 * return-from-interrupt code will reload it into registers.
209 childregs
= task_pt_regs(p
);
211 childregs
->regs
[0] = 0; /* return value is zero */
212 childregs
->sp
= sp
; /* override with new user stack pointer */
215 * Copy the callee-saved registers from the passed pt_regs struct
216 * into the context-switch callee-saved registers area.
217 * We have to restore the callee-saved registers since we may
218 * be cloning a userspace task with userspace register state,
219 * and we won't be unwinding the same kernel frames to restore them.
220 * Zero out the C ABI save area to mark the top of the stack.
222 ksp
= (unsigned long) childregs
;
223 ksp
-= C_ABI_SAVE_AREA_SIZE
; /* interrupt-entry save area */
224 ((long *)ksp
)[0] = ((long *)ksp
)[1] = 0;
225 ksp
-= CALLEE_SAVED_REGS_COUNT
* sizeof(unsigned long);
226 memcpy((void *)ksp
, ®s
->regs
[CALLEE_SAVED_FIRST_REG
],
227 CALLEE_SAVED_REGS_COUNT
* sizeof(unsigned long));
228 ksp
-= C_ABI_SAVE_AREA_SIZE
; /* __switch_to() save area */
229 ((long *)ksp
)[0] = ((long *)ksp
)[1] = 0;
232 #if CHIP_HAS_TILE_DMA()
234 * No DMA in the new thread. We model this on the fact that
235 * fork() clears the pending signals, alarms, and aio for the child.
237 memset(&p
->thread
.tile_dma_state
, 0, sizeof(struct tile_dma_state
));
238 memset(&p
->thread
.dma_async_tlb
, 0, sizeof(struct async_tlb
));
241 #if CHIP_HAS_SN_PROC()
242 /* Likewise, the new thread is not running static processor code. */
243 p
->thread
.sn_proc_running
= 0;
244 memset(&p
->thread
.sn_async_tlb
, 0, sizeof(struct async_tlb
));
247 #if CHIP_HAS_PROC_STATUS_SPR()
248 /* New thread has its miscellaneous processor state bits clear. */
249 p
->thread
.proc_status
= 0;
252 #ifdef CONFIG_HARDWALL
253 /* New thread does not own any networks. */
254 p
->thread
.hardwall
= NULL
;
259 * Start the new thread with the current architecture state
260 * (user interrupt masks, etc.).
262 save_arch_state(&p
->thread
);
268 * Return "current" if it looks plausible, or else a pointer to a dummy.
269 * This can be helpful if we are just trying to emit a clean panic.
271 struct task_struct
*validate_current(void)
273 static struct task_struct corrupt
= { .comm
= "<corrupt>" };
274 struct task_struct
*tsk
= current
;
275 if (unlikely((unsigned long)tsk
< PAGE_OFFSET
||
276 (void *)tsk
> high_memory
||
277 ((unsigned long)tsk
& (__alignof__(*tsk
) - 1)) != 0)) {
278 pr_err("Corrupt 'current' %p (sp %#lx)\n", tsk
, stack_pointer
);
284 /* Take and return the pointer to the previous task, for schedule_tail(). */
285 struct task_struct
*sim_notify_fork(struct task_struct
*prev
)
287 struct task_struct
*tsk
= current
;
288 __insn_mtspr(SPR_SIM_CONTROL
, SIM_CONTROL_OS_FORK_PARENT
|
289 (tsk
->thread
.creator_pid
<< _SIM_CONTROL_OPERATOR_BITS
));
290 __insn_mtspr(SPR_SIM_CONTROL
, SIM_CONTROL_OS_FORK
|
291 (tsk
->pid
<< _SIM_CONTROL_OPERATOR_BITS
));
295 int dump_task_regs(struct task_struct
*tsk
, elf_gregset_t
*regs
)
297 struct pt_regs
*ptregs
= task_pt_regs(tsk
);
298 elf_core_copy_regs(regs
, ptregs
);
302 #if CHIP_HAS_TILE_DMA()
304 /* Allow user processes to access the DMA SPRs */
305 void grant_dma_mpls(void)
307 __insn_mtspr(SPR_MPL_DMA_CPL_SET_0
, 1);
308 __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0
, 1);
311 /* Forbid user processes from accessing the DMA SPRs */
312 void restrict_dma_mpls(void)
314 __insn_mtspr(SPR_MPL_DMA_CPL_SET_1
, 1);
315 __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1
, 1);
318 /* Pause the DMA engine, then save off its state registers. */
319 static void save_tile_dma_state(struct tile_dma_state
*dma
)
321 unsigned long state
= __insn_mfspr(SPR_DMA_USER_STATUS
);
322 unsigned long post_suspend_state
;
324 /* If we're running, suspend the engine. */
325 if ((state
& DMA_STATUS_MASK
) == SPR_DMA_STATUS__RUNNING_MASK
)
326 __insn_mtspr(SPR_DMA_CTR
, SPR_DMA_CTR__SUSPEND_MASK
);
329 * Wait for the engine to idle, then save regs. Note that we
330 * want to record the "running" bit from before suspension,
331 * and the "done" bit from after, so that we can properly
332 * distinguish a case where the user suspended the engine from
333 * the case where the kernel suspended as part of the context
337 post_suspend_state
= __insn_mfspr(SPR_DMA_USER_STATUS
);
338 } while (post_suspend_state
& SPR_DMA_STATUS__BUSY_MASK
);
340 dma
->src
= __insn_mfspr(SPR_DMA_SRC_ADDR
);
341 dma
->src_chunk
= __insn_mfspr(SPR_DMA_SRC_CHUNK_ADDR
);
342 dma
->dest
= __insn_mfspr(SPR_DMA_DST_ADDR
);
343 dma
->dest_chunk
= __insn_mfspr(SPR_DMA_DST_CHUNK_ADDR
);
344 dma
->strides
= __insn_mfspr(SPR_DMA_STRIDE
);
345 dma
->chunk_size
= __insn_mfspr(SPR_DMA_CHUNK_SIZE
);
346 dma
->byte
= __insn_mfspr(SPR_DMA_BYTE
);
347 dma
->status
= (state
& SPR_DMA_STATUS__RUNNING_MASK
) |
348 (post_suspend_state
& SPR_DMA_STATUS__DONE_MASK
);
351 /* Restart a DMA that was running before we were context-switched out. */
352 static void restore_tile_dma_state(struct thread_struct
*t
)
354 const struct tile_dma_state
*dma
= &t
->tile_dma_state
;
357 * The only way to restore the done bit is to run a zero
358 * length transaction.
360 if ((dma
->status
& SPR_DMA_STATUS__DONE_MASK
) &&
361 !(__insn_mfspr(SPR_DMA_USER_STATUS
) & SPR_DMA_STATUS__DONE_MASK
)) {
362 __insn_mtspr(SPR_DMA_BYTE
, 0);
363 __insn_mtspr(SPR_DMA_CTR
, SPR_DMA_CTR__REQUEST_MASK
);
364 while (__insn_mfspr(SPR_DMA_USER_STATUS
) &
365 SPR_DMA_STATUS__BUSY_MASK
)
369 __insn_mtspr(SPR_DMA_SRC_ADDR
, dma
->src
);
370 __insn_mtspr(SPR_DMA_SRC_CHUNK_ADDR
, dma
->src_chunk
);
371 __insn_mtspr(SPR_DMA_DST_ADDR
, dma
->dest
);
372 __insn_mtspr(SPR_DMA_DST_CHUNK_ADDR
, dma
->dest_chunk
);
373 __insn_mtspr(SPR_DMA_STRIDE
, dma
->strides
);
374 __insn_mtspr(SPR_DMA_CHUNK_SIZE
, dma
->chunk_size
);
375 __insn_mtspr(SPR_DMA_BYTE
, dma
->byte
);
378 * Restart the engine if we were running and not done.
379 * Clear a pending async DMA fault that we were waiting on return
380 * to user space to execute, since we expect the DMA engine
381 * to regenerate those faults for us now. Note that we don't
382 * try to clear the TIF_ASYNC_TLB flag, since it's relatively
383 * harmless if set, and it covers both DMA and the SN processor.
385 if ((dma
->status
& DMA_STATUS_MASK
) == SPR_DMA_STATUS__RUNNING_MASK
) {
386 t
->dma_async_tlb
.fault_num
= 0;
387 __insn_mtspr(SPR_DMA_CTR
, SPR_DMA_CTR__REQUEST_MASK
);
393 static void save_arch_state(struct thread_struct
*t
)
395 #if CHIP_HAS_SPLIT_INTR_MASK()
396 t
->interrupt_mask
= __insn_mfspr(SPR_INTERRUPT_MASK_0_0
) |
397 ((u64
)__insn_mfspr(SPR_INTERRUPT_MASK_0_1
) << 32);
399 t
->interrupt_mask
= __insn_mfspr(SPR_INTERRUPT_MASK_0
);
401 t
->ex_context
[0] = __insn_mfspr(SPR_EX_CONTEXT_0_0
);
402 t
->ex_context
[1] = __insn_mfspr(SPR_EX_CONTEXT_0_1
);
403 t
->system_save
[0] = __insn_mfspr(SPR_SYSTEM_SAVE_0_0
);
404 t
->system_save
[1] = __insn_mfspr(SPR_SYSTEM_SAVE_0_1
);
405 t
->system_save
[2] = __insn_mfspr(SPR_SYSTEM_SAVE_0_2
);
406 t
->system_save
[3] = __insn_mfspr(SPR_SYSTEM_SAVE_0_3
);
407 t
->intctrl_0
= __insn_mfspr(SPR_INTCTRL_0_STATUS
);
408 #if CHIP_HAS_PROC_STATUS_SPR()
409 t
->proc_status
= __insn_mfspr(SPR_PROC_STATUS
);
411 #if !CHIP_HAS_FIXED_INTVEC_BASE()
412 t
->interrupt_vector_base
= __insn_mfspr(SPR_INTERRUPT_VECTOR_BASE_0
);
414 #if CHIP_HAS_TILE_RTF_HWM()
415 t
->tile_rtf_hwm
= __insn_mfspr(SPR_TILE_RTF_HWM
);
417 #if CHIP_HAS_DSTREAM_PF()
418 t
->dstream_pf
= __insn_mfspr(SPR_DSTREAM_PF
);
422 static void restore_arch_state(const struct thread_struct
*t
)
424 #if CHIP_HAS_SPLIT_INTR_MASK()
425 __insn_mtspr(SPR_INTERRUPT_MASK_0_0
, (u32
) t
->interrupt_mask
);
426 __insn_mtspr(SPR_INTERRUPT_MASK_0_1
, t
->interrupt_mask
>> 32);
428 __insn_mtspr(SPR_INTERRUPT_MASK_0
, t
->interrupt_mask
);
430 __insn_mtspr(SPR_EX_CONTEXT_0_0
, t
->ex_context
[0]);
431 __insn_mtspr(SPR_EX_CONTEXT_0_1
, t
->ex_context
[1]);
432 __insn_mtspr(SPR_SYSTEM_SAVE_0_0
, t
->system_save
[0]);
433 __insn_mtspr(SPR_SYSTEM_SAVE_0_1
, t
->system_save
[1]);
434 __insn_mtspr(SPR_SYSTEM_SAVE_0_2
, t
->system_save
[2]);
435 __insn_mtspr(SPR_SYSTEM_SAVE_0_3
, t
->system_save
[3]);
436 __insn_mtspr(SPR_INTCTRL_0_STATUS
, t
->intctrl_0
);
437 #if CHIP_HAS_PROC_STATUS_SPR()
438 __insn_mtspr(SPR_PROC_STATUS
, t
->proc_status
);
440 #if !CHIP_HAS_FIXED_INTVEC_BASE()
441 __insn_mtspr(SPR_INTERRUPT_VECTOR_BASE_0
, t
->interrupt_vector_base
);
443 #if CHIP_HAS_TILE_RTF_HWM()
444 __insn_mtspr(SPR_TILE_RTF_HWM
, t
->tile_rtf_hwm
);
446 #if CHIP_HAS_DSTREAM_PF()
447 __insn_mtspr(SPR_DSTREAM_PF
, t
->dstream_pf
);
452 void _prepare_arch_switch(struct task_struct
*next
)
454 #if CHIP_HAS_SN_PROC()
457 #if CHIP_HAS_TILE_DMA()
458 struct tile_dma_state
*dma
= ¤t
->thread
.tile_dma_state
;
460 save_tile_dma_state(dma
);
462 #if CHIP_HAS_SN_PROC()
464 * Suspend the static network processor if it was running.
465 * We do not suspend the fabric itself, just like we don't
466 * try to suspend the UDN.
468 snctl
= __insn_mfspr(SPR_SNCTL
);
469 current
->thread
.sn_proc_running
=
470 (snctl
& SPR_SNCTL__FRZPROC_MASK
) == 0;
471 if (current
->thread
.sn_proc_running
)
472 __insn_mtspr(SPR_SNCTL
, snctl
| SPR_SNCTL__FRZPROC_MASK
);
477 struct task_struct
*__sched
_switch_to(struct task_struct
*prev
,
478 struct task_struct
*next
)
480 /* DMA state is already saved; save off other arch state. */
481 save_arch_state(&prev
->thread
);
483 #if CHIP_HAS_TILE_DMA()
485 * Restore DMA in new task if desired.
486 * Note that it is only safe to restart here since interrupts
487 * are disabled, so we can't take any DMATLB miss or access
488 * interrupts before we have finished switching stacks.
490 if (next
->thread
.tile_dma_state
.enabled
) {
491 restore_tile_dma_state(&next
->thread
);
498 /* Restore other arch state. */
499 restore_arch_state(&next
->thread
);
501 #if CHIP_HAS_SN_PROC()
503 * Restart static network processor in the new process
504 * if it was running before.
506 if (next
->thread
.sn_proc_running
) {
507 int snctl
= __insn_mfspr(SPR_SNCTL
);
508 __insn_mtspr(SPR_SNCTL
, snctl
& ~SPR_SNCTL__FRZPROC_MASK
);
512 #ifdef CONFIG_HARDWALL
513 /* Enable or disable access to the network registers appropriately. */
514 if (prev
->thread
.hardwall
!= NULL
) {
515 if (next
->thread
.hardwall
== NULL
)
516 restrict_network_mpls();
517 } else if (next
->thread
.hardwall
!= NULL
) {
518 grant_network_mpls();
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_1_0 when we reset our sp.
528 return __switch_to(prev
, next
, next_current_ksp0(next
));
531 long _sys_fork(struct pt_regs
*regs
)
533 return do_fork(SIGCHLD
, regs
->sp
, regs
, 0, NULL
, NULL
);
536 long _sys_clone(unsigned long clone_flags
, unsigned long newsp
,
537 void __user
*parent_tidptr
, void __user
*child_tidptr
,
538 struct pt_regs
*regs
)
542 return do_fork(clone_flags
, newsp
, regs
, 0,
543 parent_tidptr
, child_tidptr
);
546 long _sys_vfork(struct pt_regs
*regs
)
548 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
->sp
,
549 regs
, 0, NULL
, NULL
);
553 * sys_execve() executes a new program.
555 long _sys_execve(const char __user
*path
,
556 const char __user
*const __user
*argv
,
557 const char __user
*const __user
*envp
, struct pt_regs
*regs
)
562 filename
= getname(path
);
563 error
= PTR_ERR(filename
);
564 if (IS_ERR(filename
))
566 error
= do_execve(filename
, argv
, envp
, regs
);
573 long _compat_sys_execve(const char __user
*path
,
574 const compat_uptr_t __user
*argv
,
575 const compat_uptr_t __user
*envp
, struct pt_regs
*regs
)
580 filename
= getname(path
);
581 error
= PTR_ERR(filename
);
582 if (IS_ERR(filename
))
584 error
= compat_do_execve(filename
, argv
, envp
, regs
);
591 unsigned long get_wchan(struct task_struct
*p
)
593 struct KBacktraceIterator kbt
;
595 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
598 for (KBacktraceIterator_init(&kbt
, p
, NULL
);
599 !KBacktraceIterator_end(&kbt
);
600 KBacktraceIterator_next(&kbt
)) {
601 if (!in_sched_functions(kbt
.it
.pc
))
609 * We pass in lr as zero (cleared in kernel_thread) and the caller
610 * part of the backtrace ABI on the stack also zeroed (in copy_thread)
611 * so that backtraces will stop with this function.
612 * Note that we don't use r0, since copy_thread() clears it.
614 static void start_kernel_thread(int dummy
, int (*fn
)(int), int arg
)
620 * Create a kernel thread
622 int kernel_thread(int (*fn
)(void *), void * arg
, unsigned long flags
)
626 memset(®s
, 0, sizeof(regs
));
627 regs
.ex1
= PL_ICS_EX1(KERNEL_PL
, 0); /* run at kernel PL, no ICS */
628 regs
.pc
= (long) start_kernel_thread
;
629 regs
.flags
= PT_FLAGS_CALLER_SAVES
; /* need to restore r1 and r2 */
630 regs
.regs
[1] = (long) fn
; /* function pointer */
631 regs
.regs
[2] = (long) arg
; /* parameter register */
633 /* Ok, create the new process.. */
634 return do_fork(flags
| CLONE_VM
| CLONE_UNTRACED
, 0, ®s
,
637 EXPORT_SYMBOL(kernel_thread
);
639 /* Flush thread state. */
640 void flush_thread(void)
646 * Free current thread data structures etc..
648 void exit_thread(void)
653 void show_regs(struct pt_regs
*regs
)
655 struct task_struct
*tsk
= validate_current();
659 pr_err(" Pid: %d, comm: %20s, CPU: %d\n",
660 tsk
->pid
, tsk
->comm
, smp_processor_id());
662 for (i
= 0; i
< 51; i
+= 3)
663 pr_err(" r%-2d: "REGFMT
" r%-2d: "REGFMT
" r%-2d: "REGFMT
"\n",
664 i
, regs
->regs
[i
], i
+1, regs
->regs
[i
+1],
665 i
+2, regs
->regs
[i
+2]);
666 pr_err(" r51: "REGFMT
" r52: "REGFMT
" tp : "REGFMT
"\n",
667 regs
->regs
[51], regs
->regs
[52], regs
->tp
);
668 pr_err(" sp : "REGFMT
" lr : "REGFMT
"\n", regs
->sp
, regs
->lr
);
670 for (i
= 0; i
< 52; i
+= 3)
671 pr_err(" r%-2d: "REGFMT
" r%-2d: "REGFMT
672 " r%-2d: "REGFMT
" r%-2d: "REGFMT
"\n",
673 i
, regs
->regs
[i
], i
+1, regs
->regs
[i
+1],
674 i
+2, regs
->regs
[i
+2], i
+3, regs
->regs
[i
+3]);
675 pr_err(" r52: "REGFMT
" tp : "REGFMT
" sp : "REGFMT
" lr : "REGFMT
"\n",
676 regs
->regs
[52], regs
->tp
, regs
->sp
, regs
->lr
);
678 pr_err(" pc : "REGFMT
" ex1: %ld faultnum: %ld\n",
679 regs
->pc
, regs
->ex1
, regs
->faultnum
);
681 dump_stack_regs(regs
);