Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  1 #include <linux/errno.h>
  2 #include <linux/kernel.h>
  3 #include <linux/mm.h>
  4 #include <linux/smp.h>
  5 #include <linux/prctl.h>
  6 #include <linux/slab.h>
  7 #include <linux/sched.h>
  8 #include <linux/module.h>
  9 #include <linux/pm.h>
 10 #include <linux/clockchips.h>
 11 #include <linux/random.h>
 12 #include <trace/power.h>
 13 #include <asm/system.h>
 14 #include <asm/apic.h>
 15 #include <asm/syscalls.h>
 16 #include <asm/idle.h>
 17 #include <asm/uaccess.h>
 18 #include <asm/i387.h>
 19 #include <asm/ds.h>
 20 
 21 unsigned long idle_halt;
 22 EXPORT_SYMBOL(idle_halt);
 23 unsigned long idle_nomwait;
 24 EXPORT_SYMBOL(idle_nomwait);
 25 
 26 struct kmem_cache *task_xstate_cachep;
 27 
 28 DEFINE_TRACE(power_start);
 29 DEFINE_TRACE(power_end);
 30 
 31 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 32 {
 33         *dst = *src;
 34         if (src->thread.xstate) {
 35                 dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
 36                                                       GFP_KERNEL);
 37                 if (!dst->thread.xstate)
 38                         return -ENOMEM;
 39                 WARN_ON((unsigned long)dst->thread.xstate & 15);
 40                 memcpy(dst->thread.xstate, src->thread.xstate, xstate_size);
 41         }
 42         return 0;
 43 }
 44 
 45 void free_thread_xstate(struct task_struct *tsk)
 46 {
 47         if (tsk->thread.xstate) {
 48                 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
 49                 tsk->thread.xstate = NULL;
 50         }
 51 
 52         WARN(tsk->thread.ds_ctx, "leaking DS context\n");
 53 }
 54 
 55 void free_thread_info(struct thread_info *ti)
 56 {
 57         free_thread_xstate(ti->task);
 58         free_pages((unsigned long)ti, get_order(THREAD_SIZE));
 59 }
 60 
 61 void arch_task_cache_init(void)
 62 {
 63         task_xstate_cachep =
 64                 kmem_cache_create("task_xstate", xstate_size,
 65                                   __alignof__(union thread_xstate),
 66                                   SLAB_PANIC | SLAB_NOTRACK, NULL);
 67 }
 68 
 69 /*
 70  * Free current thread data structures etc..
 71  */
 72 void exit_thread(void)
 73 {
 74         struct task_struct *me = current;
 75         struct thread_struct *t = &me->thread;
 76         unsigned long *bp = t->io_bitmap_ptr;
 77 
 78         if (bp) {
 79                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
 80 
 81                 t->io_bitmap_ptr = NULL;
 82                 clear_thread_flag(TIF_IO_BITMAP);
 83                 /*
 84                  * Careful, clear this in the TSS too:
 85                  */
 86                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 87                 t->io_bitmap_max = 0;
 88                 put_cpu();
 89                 kfree(bp);
 90         }
 91 }
 92 
 93 void flush_thread(void)
 94 {
 95         struct task_struct *tsk = current;
 96 
 97         clear_tsk_thread_flag(tsk, TIF_DEBUG);
 98 
 99         tsk->thread.debugreg0 = 0;
100         tsk->thread.debugreg1 = 0;
101         tsk->thread.debugreg2 = 0;
102         tsk->thread.debugreg3 = 0;
103         tsk->thread.debugreg6 = 0;
104         tsk->thread.debugreg7 = 0;
105         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
106         /*
107          * Forget coprocessor state..
108          */
109         tsk->fpu_counter = 0;
110         clear_fpu(tsk);
111         clear_used_math();
112 }
113 
114 static void hard_disable_TSC(void)
115 {
116         write_cr4(read_cr4() | X86_CR4_TSD);
117 }
118 
119 void disable_TSC(void)
120 {
121         preempt_disable();
122         if (!test_and_set_thread_flag(TIF_NOTSC))
123                 /*
124                  * Must flip the CPU state synchronously with
125                  * TIF_NOTSC in the current running context.
126                  */
127                 hard_disable_TSC();
128         preempt_enable();
129 }
130 
131 static void hard_enable_TSC(void)
132 {
133         write_cr4(read_cr4() & ~X86_CR4_TSD);
134 }
135 
136 static void enable_TSC(void)
137 {
138         preempt_disable();
139         if (test_and_clear_thread_flag(TIF_NOTSC))
140                 /*
141                  * Must flip the CPU state synchronously with
142                  * TIF_NOTSC in the current running context.
143                  */
144                 hard_enable_TSC();
145         preempt_enable();
146 }
147 
148 int get_tsc_mode(unsigned long adr)
149 {
150         unsigned int val;
151 
152         if (test_thread_flag(TIF_NOTSC))
153                 val = PR_TSC_SIGSEGV;
154         else
155                 val = PR_TSC_ENABLE;
156 
157         return put_user(val, (unsigned int __user *)adr);
158 }
159 
160 int set_tsc_mode(unsigned int val)
161 {
162         if (val == PR_TSC_SIGSEGV)
163                 disable_TSC();
164         else if (val == PR_TSC_ENABLE)
165                 enable_TSC();
166         else
167                 return -EINVAL;
168 
169         return 0;
170 }
171 
172 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
173                       struct tss_struct *tss)
174 {
175         struct thread_struct *prev, *next;
176 
177         prev = &prev_p->thread;
178         next = &next_p->thread;
179 
180         if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
181             test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
182                 ds_switch_to(prev_p, next_p);
183         else if (next->debugctlmsr != prev->debugctlmsr)
184                 update_debugctlmsr(next->debugctlmsr);
185 
186         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
187                 set_debugreg(next->debugreg0, 0);
188                 set_debugreg(next->debugreg1, 1);
189                 set_debugreg(next->debugreg2, 2);
190                 set_debugreg(next->debugreg3, 3);
191                 /* no 4 and 5 */
192                 set_debugreg(next->debugreg6, 6);
193                 set_debugreg(next->debugreg7, 7);
194         }
195 
196         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
197             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
198                 /* prev and next are different */
199                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
200                         hard_disable_TSC();
201                 else
202                         hard_enable_TSC();
203         }
204 
205         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
206                 /*
207                  * Copy the relevant range of the IO bitmap.
208                  * Normally this is 128 bytes or less:
209                  */
210                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
211                        max(prev->io_bitmap_max, next->io_bitmap_max));
212         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
213                 /*
214                  * Clear any possible leftover bits:
215                  */
216                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
217         }
218 }
219 
220 int sys_fork(struct pt_regs *regs)
221 {
222         return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
223 }
224 
225 /*
226  * This is trivial, and on the face of it looks like it
227  * could equally well be done in user mode.
228  *
229  * Not so, for quite unobvious reasons - register pressure.
230  * In user mode vfork() cannot have a stack frame, and if
231  * done by calling the "clone()" system call directly, you
232  * do not have enough call-clobbered registers to hold all
233  * the information you need.
234  */
235 int sys_vfork(struct pt_regs *regs)
236 {
237         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
238                        NULL, NULL);
239 }
240 
241 
242 /*
243  * Idle related variables and functions
244  */
245 unsigned long boot_option_idle_override = 0;
246 EXPORT_SYMBOL(boot_option_idle_override);
247 
248 /*
249  * Powermanagement idle function, if any..
250  */
251 void (*pm_idle)(void);
252 EXPORT_SYMBOL(pm_idle);
253 
254 #ifdef CONFIG_X86_32
255 /*
256  * This halt magic was a workaround for ancient floppy DMA
257  * wreckage. It should be safe to remove.
258  */
259 static int hlt_counter;
260 void disable_hlt(void)
261 {
262         hlt_counter++;
263 }
264 EXPORT_SYMBOL(disable_hlt);
265 
266 void enable_hlt(void)
267 {
268         hlt_counter--;
269 }
270 EXPORT_SYMBOL(enable_hlt);
271 
272 static inline int hlt_use_halt(void)
273 {
274         return (!hlt_counter && boot_cpu_data.hlt_works_ok);
275 }
276 #else
277 static inline int hlt_use_halt(void)
278 {
279         return 1;
280 }
281 #endif
282 
283 /*
284  * We use this if we don't have any better
285  * idle routine..
286  */
287 void default_idle(void)
288 {
289         if (hlt_use_halt()) {
290                 struct power_trace it;
291 
292                 trace_power_start(&it, POWER_CSTATE, 1);
293                 current_thread_info()->status &= ~TS_POLLING;
294                 /*
295                  * TS_POLLING-cleared state must be visible before we
296                  * test NEED_RESCHED:
297                  */
298                 smp_mb();
299 
300                 if (!need_resched())
301                         safe_halt();    /* enables interrupts racelessly */
302                 else
303                         local_irq_enable();
304                 current_thread_info()->status |= TS_POLLING;
305                 trace_power_end(&it);
306         } else {
307                 local_irq_enable();
308                 /* loop is done by the caller */
309                 cpu_relax();
310         }
311 }
312 #ifdef CONFIG_APM_MODULE
313 EXPORT_SYMBOL(default_idle);
314 #endif
315 
316 void stop_this_cpu(void *dummy)
317 {
318         local_irq_disable();
319         /*
320          * Remove this CPU:
321          */
322         set_cpu_online(smp_processor_id(), false);
323         disable_local_APIC();
324 
325         for (;;) {
326                 if (hlt_works(smp_processor_id()))
327                         halt();
328         }
329 }
330 
331 static void do_nothing(void *unused)
332 {
333 }
334 
335 /*
336  * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
337  * pm_idle and update to new pm_idle value. Required while changing pm_idle
338  * handler on SMP systems.
339  *
340  * Caller must have changed pm_idle to the new value before the call. Old
341  * pm_idle value will not be used by any CPU after the return of this function.
342  */
343 void cpu_idle_wait(void)
344 {
345         smp_mb();
346         /* kick all the CPUs so that they exit out of pm_idle */
347         smp_call_function(do_nothing, NULL, 1);
348 }
349 EXPORT_SYMBOL_GPL(cpu_idle_wait);
350 
351 /*
352  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
353  * which can obviate IPI to trigger checking of need_resched.
354  * We execute MONITOR against need_resched and enter optimized wait state
355  * through MWAIT. Whenever someone changes need_resched, we would be woken
356  * up from MWAIT (without an IPI).
357  *
358  * New with Core Duo processors, MWAIT can take some hints based on CPU
359  * capability.
360  */
361 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
362 {
363         struct power_trace it;
364 
365         trace_power_start(&it, POWER_CSTATE, (ax>>4)+1);
366         if (!need_resched()) {
367                 if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
368                         clflush((void *)&current_thread_info()->flags);
369 
370                 __monitor((void *)&current_thread_info()->flags, 0, 0);
371                 smp_mb();
372                 if (!need_resched())
373                         __mwait(ax, cx);
374         }
375         trace_power_end(&it);
376 }
377 
378 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
379 static void mwait_idle(void)
380 {
381         struct power_trace it;
382         if (!need_resched()) {
383                 trace_power_start(&it, POWER_CSTATE, 1);
384                 if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
385                         clflush((void *)&current_thread_info()->flags);
386 
387                 __monitor((void *)&current_thread_info()->flags, 0, 0);
388                 smp_mb();
389                 if (!need_resched())
390                         __sti_mwait(0, 0);
391                 else
392                         local_irq_enable();
393                 trace_power_end(&it);
394         } else
395                 local_irq_enable();
396 }
397 
398 /*
399  * On SMP it's slightly faster (but much more power-consuming!)
400  * to poll the ->work.need_resched flag instead of waiting for the
401  * cross-CPU IPI to arrive. Use this option with caution.
402  */
403 static void poll_idle(void)
404 {
405         struct power_trace it;
406 
407         trace_power_start(&it, POWER_CSTATE, 0);
408         local_irq_enable();
409         while (!need_resched())
410                 cpu_relax();
411         trace_power_end(&it);
412 }
413 
414 /*
415  * mwait selection logic:
416  *
417  * It depends on the CPU. For AMD CPUs that support MWAIT this is
418  * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
419  * then depend on a clock divisor and current Pstate of the core. If
420  * all cores of a processor are in halt state (C1) the processor can
421  * enter the C1E (C1 enhanced) state. If mwait is used this will never
422  * happen.
423  *
424  * idle=mwait overrides this decision and forces the usage of mwait.
425  */
426 static int __cpuinitdata force_mwait;
427 
428 #define MWAIT_INFO                      0x05
429 #define MWAIT_ECX_EXTENDED_INFO         0x01
430 #define MWAIT_EDX_C1                    0xf0
431 
432 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
433 {
434         u32 eax, ebx, ecx, edx;
435 
436         if (force_mwait)
437                 return 1;
438 
439         if (c->cpuid_level < MWAIT_INFO)
440                 return 0;
441 
442         cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
443         /* Check, whether EDX has extended info about MWAIT */
444         if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
445                 return 1;
446 
447         /*
448          * edx enumeratios MONITOR/MWAIT extensions. Check, whether
449          * C1  supports MWAIT
450          */
451         return (edx & MWAIT_EDX_C1);
452 }
453 
454 /*
455  * Check for AMD CPUs, which have potentially C1E support
456  */
457 static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
458 {
459         if (c->x86_vendor != X86_VENDOR_AMD)
460                 return 0;
461 
462         if (c->x86 < 0x0F)
463                 return 0;
464 
465         /* Family 0x0f models < rev F do not have C1E */
466         if (c->x86 == 0x0f && c->x86_model < 0x40)
467                 return 0;
468 
469         return 1;
470 }
471 
472 static cpumask_var_t c1e_mask;
473 static int c1e_detected;
474 
475 void c1e_remove_cpu(int cpu)
476 {
477         if (c1e_mask != NULL)
478                 cpumask_clear_cpu(cpu, c1e_mask);
479 }
480 
481 /*
482  * C1E aware idle routine. We check for C1E active in the interrupt
483  * pending message MSR. If we detect C1E, then we handle it the same
484  * way as C3 power states (local apic timer and TSC stop)
485  */
486 static void c1e_idle(void)
487 {
488         if (need_resched())
489                 return;
490 
491         if (!c1e_detected) {
492                 u32 lo, hi;
493 
494                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
495                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
496                         c1e_detected = 1;
497                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
498                                 mark_tsc_unstable("TSC halt in AMD C1E");
499                         printk(KERN_INFO "System has AMD C1E enabled\n");
500                         set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E);
501                 }
502         }
503 
504         if (c1e_detected) {
505                 int cpu = smp_processor_id();
506 
507                 if (!cpumask_test_cpu(cpu, c1e_mask)) {
508                         cpumask_set_cpu(cpu, c1e_mask);
509                         /*
510                          * Force broadcast so ACPI can not interfere.
511                          */
512                         clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
513                                            &cpu);
514                         printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
515                                cpu);
516                 }
517                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
518 
519                 default_idle();
520 
521                 /*
522                  * The switch back from broadcast mode needs to be
523                  * called with interrupts disabled.
524                  */
525                  local_irq_disable();
526                  clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
527                  local_irq_enable();
528         } else
529                 default_idle();
530 }
531 
532 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
533 {
534 #ifdef CONFIG_SMP
535         if (pm_idle == poll_idle && smp_num_siblings > 1) {
536                 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
537                         " performance may degrade.\n");
538         }
539 #endif
540         if (pm_idle)
541                 return;
542 
543         if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
544                 /*
545                  * One CPU supports mwait => All CPUs supports mwait
546                  */
547                 printk(KERN_INFO "using mwait in idle threads.\n");
548                 pm_idle = mwait_idle;
549         } else if (check_c1e_idle(c)) {
550                 printk(KERN_INFO "using C1E aware idle routine\n");
551                 pm_idle = c1e_idle;
552         } else
553                 pm_idle = default_idle;
554 }
555 
556 void __init init_c1e_mask(void)
557 {
558         /* If we're using c1e_idle, we need to allocate c1e_mask. */
559         if (pm_idle == c1e_idle) {
560                 alloc_cpumask_var(&c1e_mask, GFP_KERNEL);
561                 cpumask_clear(c1e_mask);
562         }
563 }
564 
565 static int __init idle_setup(char *str)
566 {
567         if (!str)
568                 return -EINVAL;
569 
570         if (!strcmp(str, "poll")) {
571                 printk("using polling idle threads.\n");
572                 pm_idle = poll_idle;
573         } else if (!strcmp(str, "mwait"))
574                 force_mwait = 1;
575         else if (!strcmp(str, "halt")) {
576                 /*
577                  * When the boot option of idle=halt is added, halt is
578                  * forced to be used for CPU idle. In such case CPU C2/C3
579                  * won't be used again.
580                  * To continue to load the CPU idle driver, don't touch
581                  * the boot_option_idle_override.
582                  */
583                 pm_idle = default_idle;
584                 idle_halt = 1;
585                 return 0;
586         } else if (!strcmp(str, "nomwait")) {
587                 /*
588                  * If the boot option of "idle=nomwait" is added,
589                  * it means that mwait will be disabled for CPU C2/C3
590                  * states. In such case it won't touch the variable
591                  * of boot_option_idle_override.
592                  */
593                 idle_nomwait = 1;
594                 return 0;
595         } else
596                 return -1;
597 
598         boot_option_idle_override = 1;
599         return 0;
600 }
601 early_param("idle", idle_setup);
602 
603 unsigned long arch_align_stack(unsigned long sp)
604 {
605         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
606                 sp -= get_random_int() % 8192;
607         return sp & ~0xf;
608 }
609 
610 unsigned long arch_randomize_brk(struct mm_struct *mm)
611 {
612         unsigned long range_end = mm->brk + 0x02000000;
613         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
614 }
615 
616 
  This page was automatically generated by the LXR engine.