perf_counter: Add optional hw_perf_group_sched_in arch function
[linux-2.6] / kernel / relay.c
1 /*
2  * Public API and common code for kernel->userspace relay file support.
3  *
4  * See Documentation/filesystems/relay.txt for an overview.
5  *
6  * Copyright (C) 2002-2005 - Tom Zanussi (zanussi@us.ibm.com), IBM Corp
7  * Copyright (C) 1999-2005 - Karim Yaghmour (karim@opersys.com)
8  *
9  * Moved to kernel/relay.c by Paul Mundt, 2006.
10  * November 2006 - CPU hotplug support by Mathieu Desnoyers
11  *      (mathieu.desnoyers@polymtl.ca)
12  *
13  * This file is released under the GPL.
14  */
15 #include <linux/errno.h>
16 #include <linux/stddef.h>
17 #include <linux/slab.h>
18 #include <linux/module.h>
19 #include <linux/string.h>
20 #include <linux/relay.h>
21 #include <linux/vmalloc.h>
22 #include <linux/mm.h>
23 #include <linux/cpu.h>
24 #include <linux/splice.h>
25
26 /* list of open channels, for cpu hotplug */
27 static DEFINE_MUTEX(relay_channels_mutex);
28 static LIST_HEAD(relay_channels);
29
30 /*
31  * close() vm_op implementation for relay file mapping.
32  */
33 static void relay_file_mmap_close(struct vm_area_struct *vma)
34 {
35         struct rchan_buf *buf = vma->vm_private_data;
36         buf->chan->cb->buf_unmapped(buf, vma->vm_file);
37 }
38
39 /*
40  * fault() vm_op implementation for relay file mapping.
41  */
42 static int relay_buf_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
43 {
44         struct page *page;
45         struct rchan_buf *buf = vma->vm_private_data;
46         pgoff_t pgoff = vmf->pgoff;
47
48         if (!buf)
49                 return VM_FAULT_OOM;
50
51         page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
52         if (!page)
53                 return VM_FAULT_SIGBUS;
54         get_page(page);
55         vmf->page = page;
56
57         return 0;
58 }
59
60 /*
61  * vm_ops for relay file mappings.
62  */
63 static struct vm_operations_struct relay_file_mmap_ops = {
64         .fault = relay_buf_fault,
65         .close = relay_file_mmap_close,
66 };
67
68 /*
69  * allocate an array of pointers of struct page
70  */
71 static struct page **relay_alloc_page_array(unsigned int n_pages)
72 {
73         struct page **array;
74         size_t pa_size = n_pages * sizeof(struct page *);
75
76         if (pa_size > PAGE_SIZE) {
77                 array = vmalloc(pa_size);
78                 if (array)
79                         memset(array, 0, pa_size);
80         } else {
81                 array = kzalloc(pa_size, GFP_KERNEL);
82         }
83         return array;
84 }
85
86 /*
87  * free an array of pointers of struct page
88  */
89 static void relay_free_page_array(struct page **array)
90 {
91         if (is_vmalloc_addr(array))
92                 vfree(array);
93         else
94                 kfree(array);
95 }
96
97 /**
98  *      relay_mmap_buf: - mmap channel buffer to process address space
99  *      @buf: relay channel buffer
100  *      @vma: vm_area_struct describing memory to be mapped
101  *
102  *      Returns 0 if ok, negative on error
103  *
104  *      Caller should already have grabbed mmap_sem.
105  */
106 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
107 {
108         unsigned long length = vma->vm_end - vma->vm_start;
109         struct file *filp = vma->vm_file;
110
111         if (!buf)
112                 return -EBADF;
113
114         if (length != (unsigned long)buf->chan->alloc_size)
115                 return -EINVAL;
116
117         vma->vm_ops = &relay_file_mmap_ops;
118         vma->vm_flags |= VM_DONTEXPAND;
119         vma->vm_private_data = buf;
120         buf->chan->cb->buf_mapped(buf, filp);
121
122         return 0;
123 }
124
125 /**
126  *      relay_alloc_buf - allocate a channel buffer
127  *      @buf: the buffer struct
128  *      @size: total size of the buffer
129  *
130  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
131  *      passed in size will get page aligned, if it isn't already.
132  */
133 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
134 {
135         void *mem;
136         unsigned int i, j, n_pages;
137
138         *size = PAGE_ALIGN(*size);
139         n_pages = *size >> PAGE_SHIFT;
140
141         buf->page_array = relay_alloc_page_array(n_pages);
142         if (!buf->page_array)
143                 return NULL;
144
145         for (i = 0; i < n_pages; i++) {
146                 buf->page_array[i] = alloc_page(GFP_KERNEL);
147                 if (unlikely(!buf->page_array[i]))
148                         goto depopulate;
149                 set_page_private(buf->page_array[i], (unsigned long)buf);
150         }
151         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
152         if (!mem)
153                 goto depopulate;
154
155         memset(mem, 0, *size);
156         buf->page_count = n_pages;
157         return mem;
158
159 depopulate:
160         for (j = 0; j < i; j++)
161                 __free_page(buf->page_array[j]);
162         relay_free_page_array(buf->page_array);
163         return NULL;
164 }
165
166 /**
167  *      relay_create_buf - allocate and initialize a channel buffer
168  *      @chan: the relay channel
169  *
170  *      Returns channel buffer if successful, %NULL otherwise.
171  */
172 static struct rchan_buf *relay_create_buf(struct rchan *chan)
173 {
174         struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
175         if (!buf)
176                 return NULL;
177
178         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
179         if (!buf->padding)
180                 goto free_buf;
181
182         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
183         if (!buf->start)
184                 goto free_buf;
185
186         buf->chan = chan;
187         kref_get(&buf->chan->kref);
188         return buf;
189
190 free_buf:
191         kfree(buf->padding);
192         kfree(buf);
193         return NULL;
194 }
195
196 /**
197  *      relay_destroy_channel - free the channel struct
198  *      @kref: target kernel reference that contains the relay channel
199  *
200  *      Should only be called from kref_put().
201  */
202 static void relay_destroy_channel(struct kref *kref)
203 {
204         struct rchan *chan = container_of(kref, struct rchan, kref);
205         kfree(chan);
206 }
207
208 /**
209  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
210  *      @buf: the buffer struct
211  */
212 static void relay_destroy_buf(struct rchan_buf *buf)
213 {
214         struct rchan *chan = buf->chan;
215         unsigned int i;
216
217         if (likely(buf->start)) {
218                 vunmap(buf->start);
219                 for (i = 0; i < buf->page_count; i++)
220                         __free_page(buf->page_array[i]);
221                 relay_free_page_array(buf->page_array);
222         }
223         chan->buf[buf->cpu] = NULL;
224         kfree(buf->padding);
225         kfree(buf);
226         kref_put(&chan->kref, relay_destroy_channel);
227 }
228
229 /**
230  *      relay_remove_buf - remove a channel buffer
231  *      @kref: target kernel reference that contains the relay buffer
232  *
233  *      Removes the file from the fileystem, which also frees the
234  *      rchan_buf_struct and the channel buffer.  Should only be called from
235  *      kref_put().
236  */
237 static void relay_remove_buf(struct kref *kref)
238 {
239         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
240         buf->chan->cb->remove_buf_file(buf->dentry);
241         relay_destroy_buf(buf);
242 }
243
244 /**
245  *      relay_buf_empty - boolean, is the channel buffer empty?
246  *      @buf: channel buffer
247  *
248  *      Returns 1 if the buffer is empty, 0 otherwise.
249  */
250 static int relay_buf_empty(struct rchan_buf *buf)
251 {
252         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
253 }
254
255 /**
256  *      relay_buf_full - boolean, is the channel buffer full?
257  *      @buf: channel buffer
258  *
259  *      Returns 1 if the buffer is full, 0 otherwise.
260  */
261 int relay_buf_full(struct rchan_buf *buf)
262 {
263         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
264         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
265 }
266 EXPORT_SYMBOL_GPL(relay_buf_full);
267
268 /*
269  * High-level relay kernel API and associated functions.
270  */
271
272 /*
273  * rchan_callback implementations defining default channel behavior.  Used
274  * in place of corresponding NULL values in client callback struct.
275  */
276
277 /*
278  * subbuf_start() default callback.  Does nothing.
279  */
280 static int subbuf_start_default_callback (struct rchan_buf *buf,
281                                           void *subbuf,
282                                           void *prev_subbuf,
283                                           size_t prev_padding)
284 {
285         if (relay_buf_full(buf))
286                 return 0;
287
288         return 1;
289 }
290
291 /*
292  * buf_mapped() default callback.  Does nothing.
293  */
294 static void buf_mapped_default_callback(struct rchan_buf *buf,
295                                         struct file *filp)
296 {
297 }
298
299 /*
300  * buf_unmapped() default callback.  Does nothing.
301  */
302 static void buf_unmapped_default_callback(struct rchan_buf *buf,
303                                           struct file *filp)
304 {
305 }
306
307 /*
308  * create_buf_file_create() default callback.  Does nothing.
309  */
310 static struct dentry *create_buf_file_default_callback(const char *filename,
311                                                        struct dentry *parent,
312                                                        int mode,
313                                                        struct rchan_buf *buf,
314                                                        int *is_global)
315 {
316         return NULL;
317 }
318
319 /*
320  * remove_buf_file() default callback.  Does nothing.
321  */
322 static int remove_buf_file_default_callback(struct dentry *dentry)
323 {
324         return -EINVAL;
325 }
326
327 /* relay channel default callbacks */
328 static struct rchan_callbacks default_channel_callbacks = {
329         .subbuf_start = subbuf_start_default_callback,
330         .buf_mapped = buf_mapped_default_callback,
331         .buf_unmapped = buf_unmapped_default_callback,
332         .create_buf_file = create_buf_file_default_callback,
333         .remove_buf_file = remove_buf_file_default_callback,
334 };
335
336 /**
337  *      wakeup_readers - wake up readers waiting on a channel
338  *      @data: contains the channel buffer
339  *
340  *      This is the timer function used to defer reader waking.
341  */
342 static void wakeup_readers(unsigned long data)
343 {
344         struct rchan_buf *buf = (struct rchan_buf *)data;
345         wake_up_interruptible(&buf->read_wait);
346 }
347
348 /**
349  *      __relay_reset - reset a channel buffer
350  *      @buf: the channel buffer
351  *      @init: 1 if this is a first-time initialization
352  *
353  *      See relay_reset() for description of effect.
354  */
355 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
356 {
357         size_t i;
358
359         if (init) {
360                 init_waitqueue_head(&buf->read_wait);
361                 kref_init(&buf->kref);
362                 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
363         } else
364                 del_timer_sync(&buf->timer);
365
366         buf->subbufs_produced = 0;
367         buf->subbufs_consumed = 0;
368         buf->bytes_consumed = 0;
369         buf->finalized = 0;
370         buf->data = buf->start;
371         buf->offset = 0;
372
373         for (i = 0; i < buf->chan->n_subbufs; i++)
374                 buf->padding[i] = 0;
375
376         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
377 }
378
379 /**
380  *      relay_reset - reset the channel
381  *      @chan: the channel
382  *
383  *      This has the effect of erasing all data from all channel buffers
384  *      and restarting the channel in its initial state.  The buffers
385  *      are not freed, so any mappings are still in effect.
386  *
387  *      NOTE. Care should be taken that the channel isn't actually
388  *      being used by anything when this call is made.
389  */
390 void relay_reset(struct rchan *chan)
391 {
392         unsigned int i;
393
394         if (!chan)
395                 return;
396
397         if (chan->is_global && chan->buf[0]) {
398                 __relay_reset(chan->buf[0], 0);
399                 return;
400         }
401
402         mutex_lock(&relay_channels_mutex);
403         for_each_possible_cpu(i)
404                 if (chan->buf[i])
405                         __relay_reset(chan->buf[i], 0);
406         mutex_unlock(&relay_channels_mutex);
407 }
408 EXPORT_SYMBOL_GPL(relay_reset);
409
410 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
411                                         struct dentry *dentry)
412 {
413         buf->dentry = dentry;
414         buf->dentry->d_inode->i_size = buf->early_bytes;
415 }
416
417 static struct dentry *relay_create_buf_file(struct rchan *chan,
418                                             struct rchan_buf *buf,
419                                             unsigned int cpu)
420 {
421         struct dentry *dentry;
422         char *tmpname;
423
424         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
425         if (!tmpname)
426                 return NULL;
427         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
428
429         /* Create file in fs */
430         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
431                                            S_IRUSR, buf,
432                                            &chan->is_global);
433
434         kfree(tmpname);
435
436         return dentry;
437 }
438
439 /*
440  *      relay_open_buf - create a new relay channel buffer
441  *
442  *      used by relay_open() and CPU hotplug.
443  */
444 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
445 {
446         struct rchan_buf *buf = NULL;
447         struct dentry *dentry;
448
449         if (chan->is_global)
450                 return chan->buf[0];
451
452         buf = relay_create_buf(chan);
453         if (!buf)
454                 return NULL;
455
456         if (chan->has_base_filename) {
457                 dentry = relay_create_buf_file(chan, buf, cpu);
458                 if (!dentry)
459                         goto free_buf;
460                 relay_set_buf_dentry(buf, dentry);
461         }
462
463         buf->cpu = cpu;
464         __relay_reset(buf, 1);
465
466         if(chan->is_global) {
467                 chan->buf[0] = buf;
468                 buf->cpu = 0;
469         }
470
471         return buf;
472
473 free_buf:
474         relay_destroy_buf(buf);
475         return NULL;
476 }
477
478 /**
479  *      relay_close_buf - close a channel buffer
480  *      @buf: channel buffer
481  *
482  *      Marks the buffer finalized and restores the default callbacks.
483  *      The channel buffer and channel buffer data structure are then freed
484  *      automatically when the last reference is given up.
485  */
486 static void relay_close_buf(struct rchan_buf *buf)
487 {
488         buf->finalized = 1;
489         del_timer_sync(&buf->timer);
490         kref_put(&buf->kref, relay_remove_buf);
491 }
492
493 static void setup_callbacks(struct rchan *chan,
494                                    struct rchan_callbacks *cb)
495 {
496         if (!cb) {
497                 chan->cb = &default_channel_callbacks;
498                 return;
499         }
500
501         if (!cb->subbuf_start)
502                 cb->subbuf_start = subbuf_start_default_callback;
503         if (!cb->buf_mapped)
504                 cb->buf_mapped = buf_mapped_default_callback;
505         if (!cb->buf_unmapped)
506                 cb->buf_unmapped = buf_unmapped_default_callback;
507         if (!cb->create_buf_file)
508                 cb->create_buf_file = create_buf_file_default_callback;
509         if (!cb->remove_buf_file)
510                 cb->remove_buf_file = remove_buf_file_default_callback;
511         chan->cb = cb;
512 }
513
514 /**
515  *      relay_hotcpu_callback - CPU hotplug callback
516  *      @nb: notifier block
517  *      @action: hotplug action to take
518  *      @hcpu: CPU number
519  *
520  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
521  */
522 static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
523                                 unsigned long action,
524                                 void *hcpu)
525 {
526         unsigned int hotcpu = (unsigned long)hcpu;
527         struct rchan *chan;
528
529         switch(action) {
530         case CPU_UP_PREPARE:
531         case CPU_UP_PREPARE_FROZEN:
532                 mutex_lock(&relay_channels_mutex);
533                 list_for_each_entry(chan, &relay_channels, list) {
534                         if (chan->buf[hotcpu])
535                                 continue;
536                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
537                         if(!chan->buf[hotcpu]) {
538                                 printk(KERN_ERR
539                                         "relay_hotcpu_callback: cpu %d buffer "
540                                         "creation failed\n", hotcpu);
541                                 mutex_unlock(&relay_channels_mutex);
542                                 return NOTIFY_BAD;
543                         }
544                 }
545                 mutex_unlock(&relay_channels_mutex);
546                 break;
547         case CPU_DEAD:
548         case CPU_DEAD_FROZEN:
549                 /* No need to flush the cpu : will be flushed upon
550                  * final relay_flush() call. */
551                 break;
552         }
553         return NOTIFY_OK;
554 }
555
556 /**
557  *      relay_open - create a new relay channel
558  *      @base_filename: base name of files to create, %NULL for buffering only
559  *      @parent: dentry of parent directory, %NULL for root directory or buffer
560  *      @subbuf_size: size of sub-buffers
561  *      @n_subbufs: number of sub-buffers
562  *      @cb: client callback functions
563  *      @private_data: user-defined data
564  *
565  *      Returns channel pointer if successful, %NULL otherwise.
566  *
567  *      Creates a channel buffer for each cpu using the sizes and
568  *      attributes specified.  The created channel buffer files
569  *      will be named base_filename0...base_filenameN-1.  File
570  *      permissions will be %S_IRUSR.
571  */
572 struct rchan *relay_open(const char *base_filename,
573                          struct dentry *parent,
574                          size_t subbuf_size,
575                          size_t n_subbufs,
576                          struct rchan_callbacks *cb,
577                          void *private_data)
578 {
579         unsigned int i;
580         struct rchan *chan;
581
582         if (!(subbuf_size && n_subbufs))
583                 return NULL;
584
585         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
586         if (!chan)
587                 return NULL;
588
589         chan->version = RELAYFS_CHANNEL_VERSION;
590         chan->n_subbufs = n_subbufs;
591         chan->subbuf_size = subbuf_size;
592         chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
593         chan->parent = parent;
594         chan->private_data = private_data;
595         if (base_filename) {
596                 chan->has_base_filename = 1;
597                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
598         }
599         setup_callbacks(chan, cb);
600         kref_init(&chan->kref);
601
602         mutex_lock(&relay_channels_mutex);
603         for_each_online_cpu(i) {
604                 chan->buf[i] = relay_open_buf(chan, i);
605                 if (!chan->buf[i])
606                         goto free_bufs;
607         }
608         list_add(&chan->list, &relay_channels);
609         mutex_unlock(&relay_channels_mutex);
610
611         return chan;
612
613 free_bufs:
614         for_each_possible_cpu(i) {
615                 if (chan->buf[i])
616                         relay_close_buf(chan->buf[i]);
617         }
618
619         kref_put(&chan->kref, relay_destroy_channel);
620         mutex_unlock(&relay_channels_mutex);
621         return NULL;
622 }
623 EXPORT_SYMBOL_GPL(relay_open);
624
625 struct rchan_percpu_buf_dispatcher {
626         struct rchan_buf *buf;
627         struct dentry *dentry;
628 };
629
630 /* Called in atomic context. */
631 static void __relay_set_buf_dentry(void *info)
632 {
633         struct rchan_percpu_buf_dispatcher *p = info;
634
635         relay_set_buf_dentry(p->buf, p->dentry);
636 }
637
638 /**
639  *      relay_late_setup_files - triggers file creation
640  *      @chan: channel to operate on
641  *      @base_filename: base name of files to create
642  *      @parent: dentry of parent directory, %NULL for root directory
643  *
644  *      Returns 0 if successful, non-zero otherwise.
645  *
646  *      Use to setup files for a previously buffer-only channel.
647  *      Useful to do early tracing in kernel, before VFS is up, for example.
648  */
649 int relay_late_setup_files(struct rchan *chan,
650                            const char *base_filename,
651                            struct dentry *parent)
652 {
653         int err = 0;
654         unsigned int i, curr_cpu;
655         unsigned long flags;
656         struct dentry *dentry;
657         struct rchan_percpu_buf_dispatcher disp;
658
659         if (!chan || !base_filename)
660                 return -EINVAL;
661
662         strlcpy(chan->base_filename, base_filename, NAME_MAX);
663
664         mutex_lock(&relay_channels_mutex);
665         /* Is chan already set up? */
666         if (unlikely(chan->has_base_filename))
667                 return -EEXIST;
668         chan->has_base_filename = 1;
669         chan->parent = parent;
670         curr_cpu = get_cpu();
671         /*
672          * The CPU hotplug notifier ran before us and created buffers with
673          * no files associated. So it's safe to call relay_setup_buf_file()
674          * on all currently online CPUs.
675          */
676         for_each_online_cpu(i) {
677                 if (unlikely(!chan->buf[i])) {
678                         printk(KERN_ERR "relay_late_setup_files: CPU %u "
679                                         "has no buffer, it must have!\n", i);
680                         BUG();
681                         err = -EINVAL;
682                         break;
683                 }
684
685                 dentry = relay_create_buf_file(chan, chan->buf[i], i);
686                 if (unlikely(!dentry)) {
687                         err = -EINVAL;
688                         break;
689                 }
690
691                 if (curr_cpu == i) {
692                         local_irq_save(flags);
693                         relay_set_buf_dentry(chan->buf[i], dentry);
694                         local_irq_restore(flags);
695                 } else {
696                         disp.buf = chan->buf[i];
697                         disp.dentry = dentry;
698                         smp_mb();
699                         /* relay_channels_mutex must be held, so wait. */
700                         err = smp_call_function_single(i,
701                                                        __relay_set_buf_dentry,
702                                                        &disp, 1);
703                 }
704                 if (unlikely(err))
705                         break;
706         }
707         put_cpu();
708         mutex_unlock(&relay_channels_mutex);
709
710         return err;
711 }
712
713 /**
714  *      relay_switch_subbuf - switch to a new sub-buffer
715  *      @buf: channel buffer
716  *      @length: size of current event
717  *
718  *      Returns either the length passed in or 0 if full.
719  *
720  *      Performs sub-buffer-switch tasks such as invoking callbacks,
721  *      updating padding counts, waking up readers, etc.
722  */
723 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
724 {
725         void *old, *new;
726         size_t old_subbuf, new_subbuf;
727
728         if (unlikely(length > buf->chan->subbuf_size))
729                 goto toobig;
730
731         if (buf->offset != buf->chan->subbuf_size + 1) {
732                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
733                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
734                 buf->padding[old_subbuf] = buf->prev_padding;
735                 buf->subbufs_produced++;
736                 if (buf->dentry)
737                         buf->dentry->d_inode->i_size +=
738                                 buf->chan->subbuf_size -
739                                 buf->padding[old_subbuf];
740                 else
741                         buf->early_bytes += buf->chan->subbuf_size -
742                                             buf->padding[old_subbuf];
743                 smp_mb();
744                 if (waitqueue_active(&buf->read_wait))
745                         /*
746                          * Calling wake_up_interruptible() from here
747                          * will deadlock if we happen to be logging
748                          * from the scheduler (trying to re-grab
749                          * rq->lock), so defer it.
750                          */
751                         __mod_timer(&buf->timer, jiffies + 1);
752         }
753
754         old = buf->data;
755         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
756         new = buf->start + new_subbuf * buf->chan->subbuf_size;
757         buf->offset = 0;
758         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
759                 buf->offset = buf->chan->subbuf_size + 1;
760                 return 0;
761         }
762         buf->data = new;
763         buf->padding[new_subbuf] = 0;
764
765         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
766                 goto toobig;
767
768         return length;
769
770 toobig:
771         buf->chan->last_toobig = length;
772         return 0;
773 }
774 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
775
776 /**
777  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
778  *      @chan: the channel
779  *      @cpu: the cpu associated with the channel buffer to update
780  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
781  *
782  *      Adds to the channel buffer's consumed sub-buffer count.
783  *      subbufs_consumed should be the number of sub-buffers newly consumed,
784  *      not the total consumed.
785  *
786  *      NOTE. Kernel clients don't need to call this function if the channel
787  *      mode is 'overwrite'.
788  */
789 void relay_subbufs_consumed(struct rchan *chan,
790                             unsigned int cpu,
791                             size_t subbufs_consumed)
792 {
793         struct rchan_buf *buf;
794
795         if (!chan)
796                 return;
797
798         if (cpu >= NR_CPUS || !chan->buf[cpu])
799                 return;
800
801         buf = chan->buf[cpu];
802         buf->subbufs_consumed += subbufs_consumed;
803         if (buf->subbufs_consumed > buf->subbufs_produced)
804                 buf->subbufs_consumed = buf->subbufs_produced;
805 }
806 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
807
808 /**
809  *      relay_close - close the channel
810  *      @chan: the channel
811  *
812  *      Closes all channel buffers and frees the channel.
813  */
814 void relay_close(struct rchan *chan)
815 {
816         unsigned int i;
817
818         if (!chan)
819                 return;
820
821         mutex_lock(&relay_channels_mutex);
822         if (chan->is_global && chan->buf[0])
823                 relay_close_buf(chan->buf[0]);
824         else
825                 for_each_possible_cpu(i)
826                         if (chan->buf[i])
827                                 relay_close_buf(chan->buf[i]);
828
829         if (chan->last_toobig)
830                 printk(KERN_WARNING "relay: one or more items not logged "
831                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
832                        chan->last_toobig, chan->subbuf_size);
833
834         list_del(&chan->list);
835         kref_put(&chan->kref, relay_destroy_channel);
836         mutex_unlock(&relay_channels_mutex);
837 }
838 EXPORT_SYMBOL_GPL(relay_close);
839
840 /**
841  *      relay_flush - close the channel
842  *      @chan: the channel
843  *
844  *      Flushes all channel buffers, i.e. forces buffer switch.
845  */
846 void relay_flush(struct rchan *chan)
847 {
848         unsigned int i;
849
850         if (!chan)
851                 return;
852
853         if (chan->is_global && chan->buf[0]) {
854                 relay_switch_subbuf(chan->buf[0], 0);
855                 return;
856         }
857
858         mutex_lock(&relay_channels_mutex);
859         for_each_possible_cpu(i)
860                 if (chan->buf[i])
861                         relay_switch_subbuf(chan->buf[i], 0);
862         mutex_unlock(&relay_channels_mutex);
863 }
864 EXPORT_SYMBOL_GPL(relay_flush);
865
866 /**
867  *      relay_file_open - open file op for relay files
868  *      @inode: the inode
869  *      @filp: the file
870  *
871  *      Increments the channel buffer refcount.
872  */
873 static int relay_file_open(struct inode *inode, struct file *filp)
874 {
875         struct rchan_buf *buf = inode->i_private;
876         kref_get(&buf->kref);
877         filp->private_data = buf;
878
879         return nonseekable_open(inode, filp);
880 }
881
882 /**
883  *      relay_file_mmap - mmap file op for relay files
884  *      @filp: the file
885  *      @vma: the vma describing what to map
886  *
887  *      Calls upon relay_mmap_buf() to map the file into user space.
888  */
889 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
890 {
891         struct rchan_buf *buf = filp->private_data;
892         return relay_mmap_buf(buf, vma);
893 }
894
895 /**
896  *      relay_file_poll - poll file op for relay files
897  *      @filp: the file
898  *      @wait: poll table
899  *
900  *      Poll implemention.
901  */
902 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
903 {
904         unsigned int mask = 0;
905         struct rchan_buf *buf = filp->private_data;
906
907         if (buf->finalized)
908                 return POLLERR;
909
910         if (filp->f_mode & FMODE_READ) {
911                 poll_wait(filp, &buf->read_wait, wait);
912                 if (!relay_buf_empty(buf))
913                         mask |= POLLIN | POLLRDNORM;
914         }
915
916         return mask;
917 }
918
919 /**
920  *      relay_file_release - release file op for relay files
921  *      @inode: the inode
922  *      @filp: the file
923  *
924  *      Decrements the channel refcount, as the filesystem is
925  *      no longer using it.
926  */
927 static int relay_file_release(struct inode *inode, struct file *filp)
928 {
929         struct rchan_buf *buf = filp->private_data;
930         kref_put(&buf->kref, relay_remove_buf);
931
932         return 0;
933 }
934
935 /*
936  *      relay_file_read_consume - update the consumed count for the buffer
937  */
938 static void relay_file_read_consume(struct rchan_buf *buf,
939                                     size_t read_pos,
940                                     size_t bytes_consumed)
941 {
942         size_t subbuf_size = buf->chan->subbuf_size;
943         size_t n_subbufs = buf->chan->n_subbufs;
944         size_t read_subbuf;
945
946         if (buf->subbufs_produced == buf->subbufs_consumed &&
947             buf->offset == buf->bytes_consumed)
948                 return;
949
950         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
951                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
952                 buf->bytes_consumed = 0;
953         }
954
955         buf->bytes_consumed += bytes_consumed;
956         if (!read_pos)
957                 read_subbuf = buf->subbufs_consumed % n_subbufs;
958         else
959                 read_subbuf = read_pos / buf->chan->subbuf_size;
960         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
961                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
962                     (buf->offset == subbuf_size))
963                         return;
964                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
965                 buf->bytes_consumed = 0;
966         }
967 }
968
969 /*
970  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
971  */
972 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
973 {
974         size_t subbuf_size = buf->chan->subbuf_size;
975         size_t n_subbufs = buf->chan->n_subbufs;
976         size_t produced = buf->subbufs_produced;
977         size_t consumed = buf->subbufs_consumed;
978
979         relay_file_read_consume(buf, read_pos, 0);
980
981         consumed = buf->subbufs_consumed;
982
983         if (unlikely(buf->offset > subbuf_size)) {
984                 if (produced == consumed)
985                         return 0;
986                 return 1;
987         }
988
989         if (unlikely(produced - consumed >= n_subbufs)) {
990                 consumed = produced - n_subbufs + 1;
991                 buf->subbufs_consumed = consumed;
992                 buf->bytes_consumed = 0;
993         }
994
995         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
996         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
997
998         if (consumed > produced)
999                 produced += n_subbufs * subbuf_size;
1000
1001         if (consumed == produced) {
1002                 if (buf->offset == subbuf_size &&
1003                     buf->subbufs_produced > buf->subbufs_consumed)
1004                         return 1;
1005                 return 0;
1006         }
1007
1008         return 1;
1009 }
1010
1011 /**
1012  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1013  *      @read_pos: file read position
1014  *      @buf: relay channel buffer
1015  */
1016 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1017                                            struct rchan_buf *buf)
1018 {
1019         size_t padding, avail = 0;
1020         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1021         size_t subbuf_size = buf->chan->subbuf_size;
1022
1023         write_subbuf = (buf->data - buf->start) / subbuf_size;
1024         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1025         read_subbuf = read_pos / subbuf_size;
1026         read_offset = read_pos % subbuf_size;
1027         padding = buf->padding[read_subbuf];
1028
1029         if (read_subbuf == write_subbuf) {
1030                 if (read_offset + padding < write_offset)
1031                         avail = write_offset - (read_offset + padding);
1032         } else
1033                 avail = (subbuf_size - padding) - read_offset;
1034
1035         return avail;
1036 }
1037
1038 /**
1039  *      relay_file_read_start_pos - find the first available byte to read
1040  *      @read_pos: file read position
1041  *      @buf: relay channel buffer
1042  *
1043  *      If the @read_pos is in the middle of padding, return the
1044  *      position of the first actually available byte, otherwise
1045  *      return the original value.
1046  */
1047 static size_t relay_file_read_start_pos(size_t read_pos,
1048                                         struct rchan_buf *buf)
1049 {
1050         size_t read_subbuf, padding, padding_start, padding_end;
1051         size_t subbuf_size = buf->chan->subbuf_size;
1052         size_t n_subbufs = buf->chan->n_subbufs;
1053         size_t consumed = buf->subbufs_consumed % n_subbufs;
1054
1055         if (!read_pos)
1056                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1057         read_subbuf = read_pos / subbuf_size;
1058         padding = buf->padding[read_subbuf];
1059         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1060         padding_end = (read_subbuf + 1) * subbuf_size;
1061         if (read_pos >= padding_start && read_pos < padding_end) {
1062                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1063                 read_pos = read_subbuf * subbuf_size;
1064         }
1065
1066         return read_pos;
1067 }
1068
1069 /**
1070  *      relay_file_read_end_pos - return the new read position
1071  *      @read_pos: file read position
1072  *      @buf: relay channel buffer
1073  *      @count: number of bytes to be read
1074  */
1075 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1076                                       size_t read_pos,
1077                                       size_t count)
1078 {
1079         size_t read_subbuf, padding, end_pos;
1080         size_t subbuf_size = buf->chan->subbuf_size;
1081         size_t n_subbufs = buf->chan->n_subbufs;
1082
1083         read_subbuf = read_pos / subbuf_size;
1084         padding = buf->padding[read_subbuf];
1085         if (read_pos % subbuf_size + count + padding == subbuf_size)
1086                 end_pos = (read_subbuf + 1) * subbuf_size;
1087         else
1088                 end_pos = read_pos + count;
1089         if (end_pos >= subbuf_size * n_subbufs)
1090                 end_pos = 0;
1091
1092         return end_pos;
1093 }
1094
1095 /*
1096  *      subbuf_read_actor - read up to one subbuf's worth of data
1097  */
1098 static int subbuf_read_actor(size_t read_start,
1099                              struct rchan_buf *buf,
1100                              size_t avail,
1101                              read_descriptor_t *desc,
1102                              read_actor_t actor)
1103 {
1104         void *from;
1105         int ret = 0;
1106
1107         from = buf->start + read_start;
1108         ret = avail;
1109         if (copy_to_user(desc->arg.buf, from, avail)) {
1110                 desc->error = -EFAULT;
1111                 ret = 0;
1112         }
1113         desc->arg.data += ret;
1114         desc->written += ret;
1115         desc->count -= ret;
1116
1117         return ret;
1118 }
1119
1120 typedef int (*subbuf_actor_t) (size_t read_start,
1121                                struct rchan_buf *buf,
1122                                size_t avail,
1123                                read_descriptor_t *desc,
1124                                read_actor_t actor);
1125
1126 /*
1127  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1128  */
1129 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1130                                         subbuf_actor_t subbuf_actor,
1131                                         read_actor_t actor,
1132                                         read_descriptor_t *desc)
1133 {
1134         struct rchan_buf *buf = filp->private_data;
1135         size_t read_start, avail;
1136         int ret;
1137
1138         if (!desc->count)
1139                 return 0;
1140
1141         mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
1142         do {
1143                 if (!relay_file_read_avail(buf, *ppos))
1144                         break;
1145
1146                 read_start = relay_file_read_start_pos(*ppos, buf);
1147                 avail = relay_file_read_subbuf_avail(read_start, buf);
1148                 if (!avail)
1149                         break;
1150
1151                 avail = min(desc->count, avail);
1152                 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1153                 if (desc->error < 0)
1154                         break;
1155
1156                 if (ret) {
1157                         relay_file_read_consume(buf, read_start, ret);
1158                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1159                 }
1160         } while (desc->count && ret);
1161         mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1162
1163         return desc->written;
1164 }
1165
1166 static ssize_t relay_file_read(struct file *filp,
1167                                char __user *buffer,
1168                                size_t count,
1169                                loff_t *ppos)
1170 {
1171         read_descriptor_t desc;
1172         desc.written = 0;
1173         desc.count = count;
1174         desc.arg.buf = buffer;
1175         desc.error = 0;
1176         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1177                                        NULL, &desc);
1178 }
1179
1180 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1181 {
1182         rbuf->bytes_consumed += bytes_consumed;
1183
1184         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1185                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1186                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1187         }
1188 }
1189
1190 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1191                                    struct pipe_buffer *buf)
1192 {
1193         struct rchan_buf *rbuf;
1194
1195         rbuf = (struct rchan_buf *)page_private(buf->page);
1196         relay_consume_bytes(rbuf, buf->private);
1197 }
1198
1199 static struct pipe_buf_operations relay_pipe_buf_ops = {
1200         .can_merge = 0,
1201         .map = generic_pipe_buf_map,
1202         .unmap = generic_pipe_buf_unmap,
1203         .confirm = generic_pipe_buf_confirm,
1204         .release = relay_pipe_buf_release,
1205         .steal = generic_pipe_buf_steal,
1206         .get = generic_pipe_buf_get,
1207 };
1208
1209 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1210 {
1211 }
1212
1213 /*
1214  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1215  */
1216 static int subbuf_splice_actor(struct file *in,
1217                                loff_t *ppos,
1218                                struct pipe_inode_info *pipe,
1219                                size_t len,
1220                                unsigned int flags,
1221                                int *nonpad_ret)
1222 {
1223         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages, ret;
1224         struct rchan_buf *rbuf = in->private_data;
1225         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1226         uint64_t pos = (uint64_t) *ppos;
1227         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1228         size_t read_start = (size_t) do_div(pos, alloc_size);
1229         size_t read_subbuf = read_start / subbuf_size;
1230         size_t padding = rbuf->padding[read_subbuf];
1231         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1232         struct page *pages[PIPE_BUFFERS];
1233         struct partial_page partial[PIPE_BUFFERS];
1234         struct splice_pipe_desc spd = {
1235                 .pages = pages,
1236                 .nr_pages = 0,
1237                 .partial = partial,
1238                 .flags = flags,
1239                 .ops = &relay_pipe_buf_ops,
1240                 .spd_release = relay_page_release,
1241         };
1242
1243         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1244                 return 0;
1245
1246         /*
1247          * Adjust read len, if longer than what is available
1248          */
1249         if (len > (subbuf_size - read_start % subbuf_size))
1250                 len = subbuf_size - read_start % subbuf_size;
1251
1252         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1253         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1254         poff = read_start & ~PAGE_MASK;
1255         nr_pages = min_t(unsigned int, subbuf_pages, PIPE_BUFFERS);
1256
1257         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1258                 unsigned int this_len, this_end, private;
1259                 unsigned int cur_pos = read_start + total_len;
1260
1261                 if (!len)
1262                         break;
1263
1264                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1265                 private = this_len;
1266
1267                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1268                 spd.partial[spd.nr_pages].offset = poff;
1269
1270                 this_end = cur_pos + this_len;
1271                 if (this_end >= nonpad_end) {
1272                         this_len = nonpad_end - cur_pos;
1273                         private = this_len + padding;
1274                 }
1275                 spd.partial[spd.nr_pages].len = this_len;
1276                 spd.partial[spd.nr_pages].private = private;
1277
1278                 len -= this_len;
1279                 total_len += this_len;
1280                 poff = 0;
1281                 pidx = (pidx + 1) % subbuf_pages;
1282
1283                 if (this_end >= nonpad_end) {
1284                         spd.nr_pages++;
1285                         break;
1286                 }
1287         }
1288
1289         if (!spd.nr_pages)
1290                 return 0;
1291
1292         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1293         if (ret < 0 || ret < total_len)
1294                 return ret;
1295
1296         if (read_start + ret == nonpad_end)
1297                 ret += padding;
1298
1299         return ret;
1300 }
1301
1302 static ssize_t relay_file_splice_read(struct file *in,
1303                                       loff_t *ppos,
1304                                       struct pipe_inode_info *pipe,
1305                                       size_t len,
1306                                       unsigned int flags)
1307 {
1308         ssize_t spliced;
1309         int ret;
1310         int nonpad_ret = 0;
1311
1312         ret = 0;
1313         spliced = 0;
1314
1315         while (len && !spliced) {
1316                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1317                 if (ret < 0)
1318                         break;
1319                 else if (!ret) {
1320                         if (flags & SPLICE_F_NONBLOCK)
1321                                 ret = -EAGAIN;
1322                         break;
1323                 }
1324
1325                 *ppos += ret;
1326                 if (ret > len)
1327                         len = 0;
1328                 else
1329                         len -= ret;
1330                 spliced += nonpad_ret;
1331                 nonpad_ret = 0;
1332         }
1333
1334         if (spliced)
1335                 return spliced;
1336
1337         return ret;
1338 }
1339
1340 const struct file_operations relay_file_operations = {
1341         .open           = relay_file_open,
1342         .poll           = relay_file_poll,
1343         .mmap           = relay_file_mmap,
1344         .read           = relay_file_read,
1345         .llseek         = no_llseek,
1346         .release        = relay_file_release,
1347         .splice_read    = relay_file_splice_read,
1348 };
1349 EXPORT_SYMBOL_GPL(relay_file_operations);
1350
1351 static __init int relay_init(void)
1352 {
1353
1354         hotcpu_notifier(relay_hotcpu_callback, 0);
1355         return 0;
1356 }
1357
1358 early_initcall(relay_init);