Merge branch 'kconfig' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / powerpc / kernel / vio.c
1 /*
2 * IBM PowerPC Virtual I/O Infrastructure Support.
3 *
4 * Copyright (c) 2003,2008 IBM Corp.
5 * Dave Engebretsen engebret@us.ibm.com
6 * Santiago Leon santil@us.ibm.com
7 * Hollis Blanchard <hollisb@us.ibm.com>
8 * Stephen Rothwell
9 * Robert Jennings <rcjenn@us.ibm.com>
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
15 */
16
17 #include <linux/types.h>
18 #include <linux/stat.h>
19 #include <linux/device.h>
20 #include <linux/init.h>
21 #include <linux/slab.h>
22 #include <linux/console.h>
23 #include <linux/export.h>
24 #include <linux/mm.h>
25 #include <linux/dma-mapping.h>
26 #include <linux/kobject.h>
27
28 #include <asm/iommu.h>
29 #include <asm/dma.h>
30 #include <asm/vio.h>
31 #include <asm/prom.h>
32 #include <asm/firmware.h>
33 #include <asm/tce.h>
34 #include <asm/abs_addr.h>
35 #include <asm/page.h>
36 #include <asm/hvcall.h>
37
38 static struct bus_type vio_bus_type;
39
40 static struct vio_dev vio_bus_device = { /* fake "parent" device */
41 .name = "vio",
42 .type = "",
43 .dev.init_name = "vio",
44 .dev.bus = &vio_bus_type,
45 };
46
47 #ifdef CONFIG_PPC_SMLPAR
48 /**
49 * vio_cmo_pool - A pool of IO memory for CMO use
50 *
51 * @size: The size of the pool in bytes
52 * @free: The amount of free memory in the pool
53 */
54 struct vio_cmo_pool {
55 size_t size;
56 size_t free;
57 };
58
59 /* How many ms to delay queued balance work */
60 #define VIO_CMO_BALANCE_DELAY 100
61
62 /* Portion out IO memory to CMO devices by this chunk size */
63 #define VIO_CMO_BALANCE_CHUNK 131072
64
65 /**
66 * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement
67 *
68 * @vio_dev: struct vio_dev pointer
69 * @list: pointer to other devices on bus that are being tracked
70 */
71 struct vio_cmo_dev_entry {
72 struct vio_dev *viodev;
73 struct list_head list;
74 };
75
76 /**
77 * vio_cmo - VIO bus accounting structure for CMO entitlement
78 *
79 * @lock: spinlock for entire structure
80 * @balance_q: work queue for balancing system entitlement
81 * @device_list: list of CMO-enabled devices requiring entitlement
82 * @entitled: total system entitlement in bytes
83 * @reserve: pool of memory from which devices reserve entitlement, incl. spare
84 * @excess: pool of excess entitlement not needed for device reserves or spare
85 * @spare: IO memory for device hotplug functionality
86 * @min: minimum necessary for system operation
87 * @desired: desired memory for system operation
88 * @curr: bytes currently allocated
89 * @high: high water mark for IO data usage
90 */
91 struct vio_cmo {
92 spinlock_t lock;
93 struct delayed_work balance_q;
94 struct list_head device_list;
95 size_t entitled;
96 struct vio_cmo_pool reserve;
97 struct vio_cmo_pool excess;
98 size_t spare;
99 size_t min;
100 size_t desired;
101 size_t curr;
102 size_t high;
103 } vio_cmo;
104
105 /**
106 * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows
107 */
108 static int vio_cmo_num_OF_devs(void)
109 {
110 struct device_node *node_vroot;
111 int count = 0;
112
113 /*
114 * Count the number of vdevice entries with an
115 * ibm,my-dma-window OF property
116 */
117 node_vroot = of_find_node_by_name(NULL, "vdevice");
118 if (node_vroot) {
119 struct device_node *of_node;
120 struct property *prop;
121
122 for_each_child_of_node(node_vroot, of_node) {
123 prop = of_find_property(of_node, "ibm,my-dma-window",
124 NULL);
125 if (prop)
126 count++;
127 }
128 }
129 of_node_put(node_vroot);
130 return count;
131 }
132
133 /**
134 * vio_cmo_alloc - allocate IO memory for CMO-enable devices
135 *
136 * @viodev: VIO device requesting IO memory
137 * @size: size of allocation requested
138 *
139 * Allocations come from memory reserved for the devices and any excess
140 * IO memory available to all devices. The spare pool used to service
141 * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be
142 * made available.
143 *
144 * Return codes:
145 * 0 for successful allocation and -ENOMEM for a failure
146 */
147 static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size)
148 {
149 unsigned long flags;
150 size_t reserve_free = 0;
151 size_t excess_free = 0;
152 int ret = -ENOMEM;
153
154 spin_lock_irqsave(&vio_cmo.lock, flags);
155
156 /* Determine the amount of free entitlement available in reserve */
157 if (viodev->cmo.entitled > viodev->cmo.allocated)
158 reserve_free = viodev->cmo.entitled - viodev->cmo.allocated;
159
160 /* If spare is not fulfilled, the excess pool can not be used. */
161 if (vio_cmo.spare >= VIO_CMO_MIN_ENT)
162 excess_free = vio_cmo.excess.free;
163
164 /* The request can be satisfied */
165 if ((reserve_free + excess_free) >= size) {
166 vio_cmo.curr += size;
167 if (vio_cmo.curr > vio_cmo.high)
168 vio_cmo.high = vio_cmo.curr;
169 viodev->cmo.allocated += size;
170 size -= min(reserve_free, size);
171 vio_cmo.excess.free -= size;
172 ret = 0;
173 }
174
175 spin_unlock_irqrestore(&vio_cmo.lock, flags);
176 return ret;
177 }
178
179 /**
180 * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices
181 * @viodev: VIO device freeing IO memory
182 * @size: size of deallocation
183 *
184 * IO memory is freed by the device back to the correct memory pools.
185 * The spare pool is replenished first from either memory pool, then
186 * the reserve pool is used to reduce device entitlement, the excess
187 * pool is used to increase the reserve pool toward the desired entitlement
188 * target, and then the remaining memory is returned to the pools.
189 *
190 */
191 static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size)
192 {
193 unsigned long flags;
194 size_t spare_needed = 0;
195 size_t excess_freed = 0;
196 size_t reserve_freed = size;
197 size_t tmp;
198 int balance = 0;
199
200 spin_lock_irqsave(&vio_cmo.lock, flags);
201 vio_cmo.curr -= size;
202
203 /* Amount of memory freed from the excess pool */
204 if (viodev->cmo.allocated > viodev->cmo.entitled) {
205 excess_freed = min(reserve_freed, (viodev->cmo.allocated -
206 viodev->cmo.entitled));
207 reserve_freed -= excess_freed;
208 }
209
210 /* Remove allocation from device */
211 viodev->cmo.allocated -= (reserve_freed + excess_freed);
212
213 /* Spare is a subset of the reserve pool, replenish it first. */
214 spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare;
215
216 /*
217 * Replenish the spare in the reserve pool from the excess pool.
218 * This moves entitlement into the reserve pool.
219 */
220 if (spare_needed && excess_freed) {
221 tmp = min(excess_freed, spare_needed);
222 vio_cmo.excess.size -= tmp;
223 vio_cmo.reserve.size += tmp;
224 vio_cmo.spare += tmp;
225 excess_freed -= tmp;
226 spare_needed -= tmp;
227 balance = 1;
228 }
229
230 /*
231 * Replenish the spare in the reserve pool from the reserve pool.
232 * This removes entitlement from the device down to VIO_CMO_MIN_ENT,
233 * if needed, and gives it to the spare pool. The amount of used
234 * memory in this pool does not change.
235 */
236 if (spare_needed && reserve_freed) {
237 tmp = min3(spare_needed, reserve_freed, (viodev->cmo.entitled - VIO_CMO_MIN_ENT));
238
239 vio_cmo.spare += tmp;
240 viodev->cmo.entitled -= tmp;
241 reserve_freed -= tmp;
242 spare_needed -= tmp;
243 balance = 1;
244 }
245
246 /*
247 * Increase the reserve pool until the desired allocation is met.
248 * Move an allocation freed from the excess pool into the reserve
249 * pool and schedule a balance operation.
250 */
251 if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) {
252 tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size));
253
254 vio_cmo.excess.size -= tmp;
255 vio_cmo.reserve.size += tmp;
256 excess_freed -= tmp;
257 balance = 1;
258 }
259
260 /* Return memory from the excess pool to that pool */
261 if (excess_freed)
262 vio_cmo.excess.free += excess_freed;
263
264 if (balance)
265 schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY);
266 spin_unlock_irqrestore(&vio_cmo.lock, flags);
267 }
268
269 /**
270 * vio_cmo_entitlement_update - Manage system entitlement changes
271 *
272 * @new_entitlement: new system entitlement to attempt to accommodate
273 *
274 * Increases in entitlement will be used to fulfill the spare entitlement
275 * and the rest is given to the excess pool. Decreases, if they are
276 * possible, come from the excess pool and from unused device entitlement
277 *
278 * Returns: 0 on success, -ENOMEM when change can not be made
279 */
280 int vio_cmo_entitlement_update(size_t new_entitlement)
281 {
282 struct vio_dev *viodev;
283 struct vio_cmo_dev_entry *dev_ent;
284 unsigned long flags;
285 size_t avail, delta, tmp;
286
287 spin_lock_irqsave(&vio_cmo.lock, flags);
288
289 /* Entitlement increases */
290 if (new_entitlement > vio_cmo.entitled) {
291 delta = new_entitlement - vio_cmo.entitled;
292
293 /* Fulfill spare allocation */
294 if (vio_cmo.spare < VIO_CMO_MIN_ENT) {
295 tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare));
296 vio_cmo.spare += tmp;
297 vio_cmo.reserve.size += tmp;
298 delta -= tmp;
299 }
300
301 /* Remaining new allocation goes to the excess pool */
302 vio_cmo.entitled += delta;
303 vio_cmo.excess.size += delta;
304 vio_cmo.excess.free += delta;
305
306 goto out;
307 }
308
309 /* Entitlement decreases */
310 delta = vio_cmo.entitled - new_entitlement;
311 avail = vio_cmo.excess.free;
312
313 /*
314 * Need to check how much unused entitlement each device can
315 * sacrifice to fulfill entitlement change.
316 */
317 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
318 if (avail >= delta)
319 break;
320
321 viodev = dev_ent->viodev;
322 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
323 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
324 avail += viodev->cmo.entitled -
325 max_t(size_t, viodev->cmo.allocated,
326 VIO_CMO_MIN_ENT);
327 }
328
329 if (delta <= avail) {
330 vio_cmo.entitled -= delta;
331
332 /* Take entitlement from the excess pool first */
333 tmp = min(vio_cmo.excess.free, delta);
334 vio_cmo.excess.size -= tmp;
335 vio_cmo.excess.free -= tmp;
336 delta -= tmp;
337
338 /*
339 * Remove all but VIO_CMO_MIN_ENT bytes from devices
340 * until entitlement change is served
341 */
342 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
343 if (!delta)
344 break;
345
346 viodev = dev_ent->viodev;
347 tmp = 0;
348 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
349 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
350 tmp = viodev->cmo.entitled -
351 max_t(size_t, viodev->cmo.allocated,
352 VIO_CMO_MIN_ENT);
353 viodev->cmo.entitled -= min(tmp, delta);
354 delta -= min(tmp, delta);
355 }
356 } else {
357 spin_unlock_irqrestore(&vio_cmo.lock, flags);
358 return -ENOMEM;
359 }
360
361 out:
362 schedule_delayed_work(&vio_cmo.balance_q, 0);
363 spin_unlock_irqrestore(&vio_cmo.lock, flags);
364 return 0;
365 }
366
367 /**
368 * vio_cmo_balance - Balance entitlement among devices
369 *
370 * @work: work queue structure for this operation
371 *
372 * Any system entitlement above the minimum needed for devices, or
373 * already allocated to devices, can be distributed to the devices.
374 * The list of devices is iterated through to recalculate the desired
375 * entitlement level and to determine how much entitlement above the
376 * minimum entitlement is allocated to devices.
377 *
378 * Small chunks of the available entitlement are given to devices until
379 * their requirements are fulfilled or there is no entitlement left to give.
380 * Upon completion sizes of the reserve and excess pools are calculated.
381 *
382 * The system minimum entitlement level is also recalculated here.
383 * Entitlement will be reserved for devices even after vio_bus_remove to
384 * accommodate reloading the driver. The OF tree is walked to count the
385 * number of devices present and this will remove entitlement for devices
386 * that have actually left the system after having vio_bus_remove called.
387 */
388 static void vio_cmo_balance(struct work_struct *work)
389 {
390 struct vio_cmo *cmo;
391 struct vio_dev *viodev;
392 struct vio_cmo_dev_entry *dev_ent;
393 unsigned long flags;
394 size_t avail = 0, level, chunk, need;
395 int devcount = 0, fulfilled;
396
397 cmo = container_of(work, struct vio_cmo, balance_q.work);
398
399 spin_lock_irqsave(&vio_cmo.lock, flags);
400
401 /* Calculate minimum entitlement and fulfill spare */
402 cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT;
403 BUG_ON(cmo->min > cmo->entitled);
404 cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min));
405 cmo->min += cmo->spare;
406 cmo->desired = cmo->min;
407
408 /*
409 * Determine how much entitlement is available and reset device
410 * entitlements
411 */
412 avail = cmo->entitled - cmo->spare;
413 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
414 viodev = dev_ent->viodev;
415 devcount++;
416 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
417 cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT);
418 avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT);
419 }
420
421 /*
422 * Having provided each device with the minimum entitlement, loop
423 * over the devices portioning out the remaining entitlement
424 * until there is nothing left.
425 */
426 level = VIO_CMO_MIN_ENT;
427 while (avail) {
428 fulfilled = 0;
429 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
430 viodev = dev_ent->viodev;
431
432 if (viodev->cmo.desired <= level) {
433 fulfilled++;
434 continue;
435 }
436
437 /*
438 * Give the device up to VIO_CMO_BALANCE_CHUNK
439 * bytes of entitlement, but do not exceed the
440 * desired level of entitlement for the device.
441 */
442 chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK);
443 chunk = min(chunk, (viodev->cmo.desired -
444 viodev->cmo.entitled));
445 viodev->cmo.entitled += chunk;
446
447 /*
448 * If the memory for this entitlement increase was
449 * already allocated to the device it does not come
450 * from the available pool being portioned out.
451 */
452 need = max(viodev->cmo.allocated, viodev->cmo.entitled)-
453 max(viodev->cmo.allocated, level);
454 avail -= need;
455
456 }
457 if (fulfilled == devcount)
458 break;
459 level += VIO_CMO_BALANCE_CHUNK;
460 }
461
462 /* Calculate new reserve and excess pool sizes */
463 cmo->reserve.size = cmo->min;
464 cmo->excess.free = 0;
465 cmo->excess.size = 0;
466 need = 0;
467 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
468 viodev = dev_ent->viodev;
469 /* Calculated reserve size above the minimum entitlement */
470 if (viodev->cmo.entitled)
471 cmo->reserve.size += (viodev->cmo.entitled -
472 VIO_CMO_MIN_ENT);
473 /* Calculated used excess entitlement */
474 if (viodev->cmo.allocated > viodev->cmo.entitled)
475 need += viodev->cmo.allocated - viodev->cmo.entitled;
476 }
477 cmo->excess.size = cmo->entitled - cmo->reserve.size;
478 cmo->excess.free = cmo->excess.size - need;
479
480 cancel_delayed_work(to_delayed_work(work));
481 spin_unlock_irqrestore(&vio_cmo.lock, flags);
482 }
483
484 static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size,
485 dma_addr_t *dma_handle, gfp_t flag)
486 {
487 struct vio_dev *viodev = to_vio_dev(dev);
488 void *ret;
489
490 if (vio_cmo_alloc(viodev, roundup(size, PAGE_SIZE))) {
491 atomic_inc(&viodev->cmo.allocs_failed);
492 return NULL;
493 }
494
495 ret = dma_iommu_ops.alloc_coherent(dev, size, dma_handle, flag);
496 if (unlikely(ret == NULL)) {
497 vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
498 atomic_inc(&viodev->cmo.allocs_failed);
499 }
500
501 return ret;
502 }
503
504 static void vio_dma_iommu_free_coherent(struct device *dev, size_t size,
505 void *vaddr, dma_addr_t dma_handle)
506 {
507 struct vio_dev *viodev = to_vio_dev(dev);
508
509 dma_iommu_ops.free_coherent(dev, size, vaddr, dma_handle);
510
511 vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
512 }
513
514 static dma_addr_t vio_dma_iommu_map_page(struct device *dev, struct page *page,
515 unsigned long offset, size_t size,
516 enum dma_data_direction direction,
517 struct dma_attrs *attrs)
518 {
519 struct vio_dev *viodev = to_vio_dev(dev);
520 dma_addr_t ret = DMA_ERROR_CODE;
521
522 if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) {
523 atomic_inc(&viodev->cmo.allocs_failed);
524 return ret;
525 }
526
527 ret = dma_iommu_ops.map_page(dev, page, offset, size, direction, attrs);
528 if (unlikely(dma_mapping_error(dev, ret))) {
529 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
530 atomic_inc(&viodev->cmo.allocs_failed);
531 }
532
533 return ret;
534 }
535
536 static void vio_dma_iommu_unmap_page(struct device *dev, dma_addr_t dma_handle,
537 size_t size,
538 enum dma_data_direction direction,
539 struct dma_attrs *attrs)
540 {
541 struct vio_dev *viodev = to_vio_dev(dev);
542
543 dma_iommu_ops.unmap_page(dev, dma_handle, size, direction, attrs);
544
545 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
546 }
547
548 static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
549 int nelems, enum dma_data_direction direction,
550 struct dma_attrs *attrs)
551 {
552 struct vio_dev *viodev = to_vio_dev(dev);
553 struct scatterlist *sgl;
554 int ret, count = 0;
555 size_t alloc_size = 0;
556
557 for (sgl = sglist; count < nelems; count++, sgl++)
558 alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE);
559
560 if (vio_cmo_alloc(viodev, alloc_size)) {
561 atomic_inc(&viodev->cmo.allocs_failed);
562 return 0;
563 }
564
565 ret = dma_iommu_ops.map_sg(dev, sglist, nelems, direction, attrs);
566
567 if (unlikely(!ret)) {
568 vio_cmo_dealloc(viodev, alloc_size);
569 atomic_inc(&viodev->cmo.allocs_failed);
570 return ret;
571 }
572
573 for (sgl = sglist, count = 0; count < ret; count++, sgl++)
574 alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
575 if (alloc_size)
576 vio_cmo_dealloc(viodev, alloc_size);
577
578 return ret;
579 }
580
581 static void vio_dma_iommu_unmap_sg(struct device *dev,
582 struct scatterlist *sglist, int nelems,
583 enum dma_data_direction direction,
584 struct dma_attrs *attrs)
585 {
586 struct vio_dev *viodev = to_vio_dev(dev);
587 struct scatterlist *sgl;
588 size_t alloc_size = 0;
589 int count = 0;
590
591 for (sgl = sglist; count < nelems; count++, sgl++)
592 alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
593
594 dma_iommu_ops.unmap_sg(dev, sglist, nelems, direction, attrs);
595
596 vio_cmo_dealloc(viodev, alloc_size);
597 }
598
599 static int vio_dma_iommu_dma_supported(struct device *dev, u64 mask)
600 {
601 return dma_iommu_ops.dma_supported(dev, mask);
602 }
603
604 static u64 vio_dma_get_required_mask(struct device *dev)
605 {
606 return dma_iommu_ops.get_required_mask(dev);
607 }
608
609 struct dma_map_ops vio_dma_mapping_ops = {
610 .alloc_coherent = vio_dma_iommu_alloc_coherent,
611 .free_coherent = vio_dma_iommu_free_coherent,
612 .map_sg = vio_dma_iommu_map_sg,
613 .unmap_sg = vio_dma_iommu_unmap_sg,
614 .map_page = vio_dma_iommu_map_page,
615 .unmap_page = vio_dma_iommu_unmap_page,
616 .dma_supported = vio_dma_iommu_dma_supported,
617 .get_required_mask = vio_dma_get_required_mask,
618 };
619
620 /**
621 * vio_cmo_set_dev_desired - Set desired entitlement for a device
622 *
623 * @viodev: struct vio_dev for device to alter
624 * @new_desired: new desired entitlement level in bytes
625 *
626 * For use by devices to request a change to their entitlement at runtime or
627 * through sysfs. The desired entitlement level is changed and a balancing
628 * of system resources is scheduled to run in the future.
629 */
630 void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
631 {
632 unsigned long flags;
633 struct vio_cmo_dev_entry *dev_ent;
634 int found = 0;
635
636 if (!firmware_has_feature(FW_FEATURE_CMO))
637 return;
638
639 spin_lock_irqsave(&vio_cmo.lock, flags);
640 if (desired < VIO_CMO_MIN_ENT)
641 desired = VIO_CMO_MIN_ENT;
642
643 /*
644 * Changes will not be made for devices not in the device list.
645 * If it is not in the device list, then no driver is loaded
646 * for the device and it can not receive entitlement.
647 */
648 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
649 if (viodev == dev_ent->viodev) {
650 found = 1;
651 break;
652 }
653 if (!found) {
654 spin_unlock_irqrestore(&vio_cmo.lock, flags);
655 return;
656 }
657
658 /* Increase/decrease in desired device entitlement */
659 if (desired >= viodev->cmo.desired) {
660 /* Just bump the bus and device values prior to a balance*/
661 vio_cmo.desired += desired - viodev->cmo.desired;
662 viodev->cmo.desired = desired;
663 } else {
664 /* Decrease bus and device values for desired entitlement */
665 vio_cmo.desired -= viodev->cmo.desired - desired;
666 viodev->cmo.desired = desired;
667 /*
668 * If less entitlement is desired than current entitlement, move
669 * any reserve memory in the change region to the excess pool.
670 */
671 if (viodev->cmo.entitled > desired) {
672 vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
673 vio_cmo.excess.size += viodev->cmo.entitled - desired;
674 /*
675 * If entitlement moving from the reserve pool to the
676 * excess pool is currently unused, add to the excess
677 * free counter.
678 */
679 if (viodev->cmo.allocated < viodev->cmo.entitled)
680 vio_cmo.excess.free += viodev->cmo.entitled -
681 max(viodev->cmo.allocated, desired);
682 viodev->cmo.entitled = desired;
683 }
684 }
685 schedule_delayed_work(&vio_cmo.balance_q, 0);
686 spin_unlock_irqrestore(&vio_cmo.lock, flags);
687 }
688
689 /**
690 * vio_cmo_bus_probe - Handle CMO specific bus probe activities
691 *
692 * @viodev - Pointer to struct vio_dev for device
693 *
694 * Determine the devices IO memory entitlement needs, attempting
695 * to satisfy the system minimum entitlement at first and scheduling
696 * a balance operation to take care of the rest at a later time.
697 *
698 * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
699 * -ENOMEM when entitlement is not available for device or
700 * device entry.
701 *
702 */
703 static int vio_cmo_bus_probe(struct vio_dev *viodev)
704 {
705 struct vio_cmo_dev_entry *dev_ent;
706 struct device *dev = &viodev->dev;
707 struct vio_driver *viodrv = to_vio_driver(dev->driver);
708 unsigned long flags;
709 size_t size;
710
711 /*
712 * Check to see that device has a DMA window and configure
713 * entitlement for the device.
714 */
715 if (of_get_property(viodev->dev.of_node,
716 "ibm,my-dma-window", NULL)) {
717 /* Check that the driver is CMO enabled and get desired DMA */
718 if (!viodrv->get_desired_dma) {
719 dev_err(dev, "%s: device driver does not support CMO\n",
720 __func__);
721 return -EINVAL;
722 }
723
724 viodev->cmo.desired = IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev));
725 if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
726 viodev->cmo.desired = VIO_CMO_MIN_ENT;
727 size = VIO_CMO_MIN_ENT;
728
729 dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
730 GFP_KERNEL);
731 if (!dev_ent)
732 return -ENOMEM;
733
734 dev_ent->viodev = viodev;
735 spin_lock_irqsave(&vio_cmo.lock, flags);
736 list_add(&dev_ent->list, &vio_cmo.device_list);
737 } else {
738 viodev->cmo.desired = 0;
739 size = 0;
740 spin_lock_irqsave(&vio_cmo.lock, flags);
741 }
742
743 /*
744 * If the needs for vio_cmo.min have not changed since they
745 * were last set, the number of devices in the OF tree has
746 * been constant and the IO memory for this is already in
747 * the reserve pool.
748 */
749 if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
750 VIO_CMO_MIN_ENT)) {
751 /* Updated desired entitlement if device requires it */
752 if (size)
753 vio_cmo.desired += (viodev->cmo.desired -
754 VIO_CMO_MIN_ENT);
755 } else {
756 size_t tmp;
757
758 tmp = vio_cmo.spare + vio_cmo.excess.free;
759 if (tmp < size) {
760 dev_err(dev, "%s: insufficient free "
761 "entitlement to add device. "
762 "Need %lu, have %lu\n", __func__,
763 size, (vio_cmo.spare + tmp));
764 spin_unlock_irqrestore(&vio_cmo.lock, flags);
765 return -ENOMEM;
766 }
767
768 /* Use excess pool first to fulfill request */
769 tmp = min(size, vio_cmo.excess.free);
770 vio_cmo.excess.free -= tmp;
771 vio_cmo.excess.size -= tmp;
772 vio_cmo.reserve.size += tmp;
773
774 /* Use spare if excess pool was insufficient */
775 vio_cmo.spare -= size - tmp;
776
777 /* Update bus accounting */
778 vio_cmo.min += size;
779 vio_cmo.desired += viodev->cmo.desired;
780 }
781 spin_unlock_irqrestore(&vio_cmo.lock, flags);
782 return 0;
783 }
784
785 /**
786 * vio_cmo_bus_remove - Handle CMO specific bus removal activities
787 *
788 * @viodev - Pointer to struct vio_dev for device
789 *
790 * Remove the device from the cmo device list. The minimum entitlement
791 * will be reserved for the device as long as it is in the system. The
792 * rest of the entitlement the device had been allocated will be returned
793 * to the system.
794 */
795 static void vio_cmo_bus_remove(struct vio_dev *viodev)
796 {
797 struct vio_cmo_dev_entry *dev_ent;
798 unsigned long flags;
799 size_t tmp;
800
801 spin_lock_irqsave(&vio_cmo.lock, flags);
802 if (viodev->cmo.allocated) {
803 dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
804 "allocated after remove operation.\n",
805 __func__, viodev->cmo.allocated);
806 BUG();
807 }
808
809 /*
810 * Remove the device from the device list being maintained for
811 * CMO enabled devices.
812 */
813 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
814 if (viodev == dev_ent->viodev) {
815 list_del(&dev_ent->list);
816 kfree(dev_ent);
817 break;
818 }
819
820 /*
821 * Devices may not require any entitlement and they do not need
822 * to be processed. Otherwise, return the device's entitlement
823 * back to the pools.
824 */
825 if (viodev->cmo.entitled) {
826 /*
827 * This device has not yet left the OF tree, it's
828 * minimum entitlement remains in vio_cmo.min and
829 * vio_cmo.desired
830 */
831 vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
832
833 /*
834 * Save min allocation for device in reserve as long
835 * as it exists in OF tree as determined by later
836 * balance operation
837 */
838 viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
839
840 /* Replenish spare from freed reserve pool */
841 if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
842 tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
843 vio_cmo.spare));
844 vio_cmo.spare += tmp;
845 viodev->cmo.entitled -= tmp;
846 }
847
848 /* Remaining reserve goes to excess pool */
849 vio_cmo.excess.size += viodev->cmo.entitled;
850 vio_cmo.excess.free += viodev->cmo.entitled;
851 vio_cmo.reserve.size -= viodev->cmo.entitled;
852
853 /*
854 * Until the device is removed it will keep a
855 * minimum entitlement; this will guarantee that
856 * a module unload/load will result in a success.
857 */
858 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
859 viodev->cmo.desired = VIO_CMO_MIN_ENT;
860 atomic_set(&viodev->cmo.allocs_failed, 0);
861 }
862
863 spin_unlock_irqrestore(&vio_cmo.lock, flags);
864 }
865
866 static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
867 {
868 set_dma_ops(&viodev->dev, &vio_dma_mapping_ops);
869 }
870
871 /**
872 * vio_cmo_bus_init - CMO entitlement initialization at bus init time
873 *
874 * Set up the reserve and excess entitlement pools based on available
875 * system entitlement and the number of devices in the OF tree that
876 * require entitlement in the reserve pool.
877 */
878 static void vio_cmo_bus_init(void)
879 {
880 struct hvcall_mpp_data mpp_data;
881 int err;
882
883 memset(&vio_cmo, 0, sizeof(struct vio_cmo));
884 spin_lock_init(&vio_cmo.lock);
885 INIT_LIST_HEAD(&vio_cmo.device_list);
886 INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
887
888 /* Get current system entitlement */
889 err = h_get_mpp(&mpp_data);
890
891 /*
892 * On failure, continue with entitlement set to 0, will panic()
893 * later when spare is reserved.
894 */
895 if (err != H_SUCCESS) {
896 printk(KERN_ERR "%s: unable to determine system IO "\
897 "entitlement. (%d)\n", __func__, err);
898 vio_cmo.entitled = 0;
899 } else {
900 vio_cmo.entitled = mpp_data.entitled_mem;
901 }
902
903 /* Set reservation and check against entitlement */
904 vio_cmo.spare = VIO_CMO_MIN_ENT;
905 vio_cmo.reserve.size = vio_cmo.spare;
906 vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
907 VIO_CMO_MIN_ENT);
908 if (vio_cmo.reserve.size > vio_cmo.entitled) {
909 printk(KERN_ERR "%s: insufficient system entitlement\n",
910 __func__);
911 panic("%s: Insufficient system entitlement", __func__);
912 }
913
914 /* Set the remaining accounting variables */
915 vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
916 vio_cmo.excess.free = vio_cmo.excess.size;
917 vio_cmo.min = vio_cmo.reserve.size;
918 vio_cmo.desired = vio_cmo.reserve.size;
919 }
920
921 /* sysfs device functions and data structures for CMO */
922
923 #define viodev_cmo_rd_attr(name) \
924 static ssize_t viodev_cmo_##name##_show(struct device *dev, \
925 struct device_attribute *attr, \
926 char *buf) \
927 { \
928 return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \
929 }
930
931 static ssize_t viodev_cmo_allocs_failed_show(struct device *dev,
932 struct device_attribute *attr, char *buf)
933 {
934 struct vio_dev *viodev = to_vio_dev(dev);
935 return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
936 }
937
938 static ssize_t viodev_cmo_allocs_failed_reset(struct device *dev,
939 struct device_attribute *attr, const char *buf, size_t count)
940 {
941 struct vio_dev *viodev = to_vio_dev(dev);
942 atomic_set(&viodev->cmo.allocs_failed, 0);
943 return count;
944 }
945
946 static ssize_t viodev_cmo_desired_set(struct device *dev,
947 struct device_attribute *attr, const char *buf, size_t count)
948 {
949 struct vio_dev *viodev = to_vio_dev(dev);
950 size_t new_desired;
951 int ret;
952
953 ret = strict_strtoul(buf, 10, &new_desired);
954 if (ret)
955 return ret;
956
957 vio_cmo_set_dev_desired(viodev, new_desired);
958 return count;
959 }
960
961 viodev_cmo_rd_attr(desired);
962 viodev_cmo_rd_attr(entitled);
963 viodev_cmo_rd_attr(allocated);
964
965 static ssize_t name_show(struct device *, struct device_attribute *, char *);
966 static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
967 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
968 char *buf);
969 static struct device_attribute vio_cmo_dev_attrs[] = {
970 __ATTR_RO(name),
971 __ATTR_RO(devspec),
972 __ATTR_RO(modalias),
973 __ATTR(cmo_desired, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
974 viodev_cmo_desired_show, viodev_cmo_desired_set),
975 __ATTR(cmo_entitled, S_IRUGO, viodev_cmo_entitled_show, NULL),
976 __ATTR(cmo_allocated, S_IRUGO, viodev_cmo_allocated_show, NULL),
977 __ATTR(cmo_allocs_failed, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
978 viodev_cmo_allocs_failed_show, viodev_cmo_allocs_failed_reset),
979 __ATTR_NULL
980 };
981
982 /* sysfs bus functions and data structures for CMO */
983
984 #define viobus_cmo_rd_attr(name) \
985 static ssize_t \
986 viobus_cmo_##name##_show(struct bus_type *bt, char *buf) \
987 { \
988 return sprintf(buf, "%lu\n", vio_cmo.name); \
989 }
990
991 #define viobus_cmo_pool_rd_attr(name, var) \
992 static ssize_t \
993 viobus_cmo_##name##_pool_show_##var(struct bus_type *bt, char *buf) \
994 { \
995 return sprintf(buf, "%lu\n", vio_cmo.name.var); \
996 }
997
998 static ssize_t viobus_cmo_high_reset(struct bus_type *bt, const char *buf,
999 size_t count)
1000 {
1001 unsigned long flags;
1002
1003 spin_lock_irqsave(&vio_cmo.lock, flags);
1004 vio_cmo.high = vio_cmo.curr;
1005 spin_unlock_irqrestore(&vio_cmo.lock, flags);
1006
1007 return count;
1008 }
1009
1010 viobus_cmo_rd_attr(entitled);
1011 viobus_cmo_pool_rd_attr(reserve, size);
1012 viobus_cmo_pool_rd_attr(excess, size);
1013 viobus_cmo_pool_rd_attr(excess, free);
1014 viobus_cmo_rd_attr(spare);
1015 viobus_cmo_rd_attr(min);
1016 viobus_cmo_rd_attr(desired);
1017 viobus_cmo_rd_attr(curr);
1018 viobus_cmo_rd_attr(high);
1019
1020 static struct bus_attribute vio_cmo_bus_attrs[] = {
1021 __ATTR(cmo_entitled, S_IRUGO, viobus_cmo_entitled_show, NULL),
1022 __ATTR(cmo_reserve_size, S_IRUGO, viobus_cmo_reserve_pool_show_size, NULL),
1023 __ATTR(cmo_excess_size, S_IRUGO, viobus_cmo_excess_pool_show_size, NULL),
1024 __ATTR(cmo_excess_free, S_IRUGO, viobus_cmo_excess_pool_show_free, NULL),
1025 __ATTR(cmo_spare, S_IRUGO, viobus_cmo_spare_show, NULL),
1026 __ATTR(cmo_min, S_IRUGO, viobus_cmo_min_show, NULL),
1027 __ATTR(cmo_desired, S_IRUGO, viobus_cmo_desired_show, NULL),
1028 __ATTR(cmo_curr, S_IRUGO, viobus_cmo_curr_show, NULL),
1029 __ATTR(cmo_high, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
1030 viobus_cmo_high_show, viobus_cmo_high_reset),
1031 __ATTR_NULL
1032 };
1033
1034 static void vio_cmo_sysfs_init(void)
1035 {
1036 vio_bus_type.dev_attrs = vio_cmo_dev_attrs;
1037 vio_bus_type.bus_attrs = vio_cmo_bus_attrs;
1038 }
1039 #else /* CONFIG_PPC_SMLPAR */
1040 int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1041 void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1042 static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1043 static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1044 static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
1045 static void vio_cmo_bus_init(void) {}
1046 static void vio_cmo_sysfs_init(void) { }
1047 #endif /* CONFIG_PPC_SMLPAR */
1048 EXPORT_SYMBOL(vio_cmo_entitlement_update);
1049 EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1050
1051 static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1052 {
1053 const unsigned char *dma_window;
1054 struct iommu_table *tbl;
1055 unsigned long offset, size;
1056
1057 dma_window = of_get_property(dev->dev.of_node,
1058 "ibm,my-dma-window", NULL);
1059 if (!dma_window)
1060 return NULL;
1061
1062 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
1063 if (tbl == NULL)
1064 return NULL;
1065
1066 of_parse_dma_window(dev->dev.of_node, dma_window,
1067 &tbl->it_index, &offset, &size);
1068
1069 /* TCE table size - measured in tce entries */
1070 tbl->it_size = size >> IOMMU_PAGE_SHIFT;
1071 /* offset for VIO should always be 0 */
1072 tbl->it_offset = offset >> IOMMU_PAGE_SHIFT;
1073 tbl->it_busno = 0;
1074 tbl->it_type = TCE_VB;
1075 tbl->it_blocksize = 16;
1076
1077 return iommu_init_table(tbl, -1);
1078 }
1079
1080 /**
1081 * vio_match_device: - Tell if a VIO device has a matching
1082 * VIO device id structure.
1083 * @ids: array of VIO device id structures to search in
1084 * @dev: the VIO device structure to match against
1085 *
1086 * Used by a driver to check whether a VIO device present in the
1087 * system is in its list of supported devices. Returns the matching
1088 * vio_device_id structure or NULL if there is no match.
1089 */
1090 static const struct vio_device_id *vio_match_device(
1091 const struct vio_device_id *ids, const struct vio_dev *dev)
1092 {
1093 while (ids->type[0] != '\0') {
1094 if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
1095 of_device_is_compatible(dev->dev.of_node,
1096 ids->compat))
1097 return ids;
1098 ids++;
1099 }
1100 return NULL;
1101 }
1102
1103 /*
1104 * Convert from struct device to struct vio_dev and pass to driver.
1105 * dev->driver has already been set by generic code because vio_bus_match
1106 * succeeded.
1107 */
1108 static int vio_bus_probe(struct device *dev)
1109 {
1110 struct vio_dev *viodev = to_vio_dev(dev);
1111 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1112 const struct vio_device_id *id;
1113 int error = -ENODEV;
1114
1115 if (!viodrv->probe)
1116 return error;
1117
1118 id = vio_match_device(viodrv->id_table, viodev);
1119 if (id) {
1120 memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1121 if (firmware_has_feature(FW_FEATURE_CMO)) {
1122 error = vio_cmo_bus_probe(viodev);
1123 if (error)
1124 return error;
1125 }
1126 error = viodrv->probe(viodev, id);
1127 if (error && firmware_has_feature(FW_FEATURE_CMO))
1128 vio_cmo_bus_remove(viodev);
1129 }
1130
1131 return error;
1132 }
1133
1134 /* convert from struct device to struct vio_dev and pass to driver. */
1135 static int vio_bus_remove(struct device *dev)
1136 {
1137 struct vio_dev *viodev = to_vio_dev(dev);
1138 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1139 struct device *devptr;
1140 int ret = 1;
1141
1142 /*
1143 * Hold a reference to the device after the remove function is called
1144 * to allow for CMO accounting cleanup for the device.
1145 */
1146 devptr = get_device(dev);
1147
1148 if (viodrv->remove)
1149 ret = viodrv->remove(viodev);
1150
1151 if (!ret && firmware_has_feature(FW_FEATURE_CMO))
1152 vio_cmo_bus_remove(viodev);
1153
1154 put_device(devptr);
1155 return ret;
1156 }
1157
1158 /**
1159 * vio_register_driver: - Register a new vio driver
1160 * @drv: The vio_driver structure to be registered.
1161 */
1162 int __vio_register_driver(struct vio_driver *viodrv, struct module *owner,
1163 const char *mod_name)
1164 {
1165 pr_debug("%s: driver %s registering\n", __func__, viodrv->name);
1166
1167 /* fill in 'struct driver' fields */
1168 viodrv->driver.name = viodrv->name;
1169 viodrv->driver.pm = viodrv->pm;
1170 viodrv->driver.bus = &vio_bus_type;
1171 viodrv->driver.owner = owner;
1172 viodrv->driver.mod_name = mod_name;
1173
1174 return driver_register(&viodrv->driver);
1175 }
1176 EXPORT_SYMBOL(__vio_register_driver);
1177
1178 /**
1179 * vio_unregister_driver - Remove registration of vio driver.
1180 * @driver: The vio_driver struct to be removed form registration
1181 */
1182 void vio_unregister_driver(struct vio_driver *viodrv)
1183 {
1184 driver_unregister(&viodrv->driver);
1185 }
1186 EXPORT_SYMBOL(vio_unregister_driver);
1187
1188 /* vio_dev refcount hit 0 */
1189 static void __devinit vio_dev_release(struct device *dev)
1190 {
1191 struct iommu_table *tbl = get_iommu_table_base(dev);
1192
1193 if (tbl)
1194 iommu_free_table(tbl, dev->of_node ?
1195 dev->of_node->full_name : dev_name(dev));
1196 of_node_put(dev->of_node);
1197 kfree(to_vio_dev(dev));
1198 }
1199
1200 /**
1201 * vio_register_device_node: - Register a new vio device.
1202 * @of_node: The OF node for this device.
1203 *
1204 * Creates and initializes a vio_dev structure from the data in
1205 * of_node and adds it to the list of virtual devices.
1206 * Returns a pointer to the created vio_dev or NULL if node has
1207 * NULL device_type or compatible fields.
1208 */
1209 struct vio_dev *vio_register_device_node(struct device_node *of_node)
1210 {
1211 struct vio_dev *viodev;
1212 const unsigned int *unit_address;
1213
1214 /* we need the 'device_type' property, in order to match with drivers */
1215 if (of_node->type == NULL) {
1216 printk(KERN_WARNING "%s: node %s missing 'device_type'\n",
1217 __func__,
1218 of_node->name ? of_node->name : "<unknown>");
1219 return NULL;
1220 }
1221
1222 unit_address = of_get_property(of_node, "reg", NULL);
1223 if (unit_address == NULL) {
1224 printk(KERN_WARNING "%s: node %s missing 'reg'\n",
1225 __func__,
1226 of_node->name ? of_node->name : "<unknown>");
1227 return NULL;
1228 }
1229
1230 /* allocate a vio_dev for this node */
1231 viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1232 if (viodev == NULL)
1233 return NULL;
1234
1235 viodev->irq = irq_of_parse_and_map(of_node, 0);
1236
1237 dev_set_name(&viodev->dev, "%x", *unit_address);
1238 viodev->name = of_node->name;
1239 viodev->type = of_node->type;
1240 viodev->unit_address = *unit_address;
1241 viodev->dev.of_node = of_node_get(of_node);
1242
1243 if (firmware_has_feature(FW_FEATURE_CMO))
1244 vio_cmo_set_dma_ops(viodev);
1245 else
1246 set_dma_ops(&viodev->dev, &dma_iommu_ops);
1247 set_iommu_table_base(&viodev->dev, vio_build_iommu_table(viodev));
1248 set_dev_node(&viodev->dev, of_node_to_nid(of_node));
1249
1250 /* init generic 'struct device' fields: */
1251 viodev->dev.parent = &vio_bus_device.dev;
1252 viodev->dev.bus = &vio_bus_type;
1253 viodev->dev.release = vio_dev_release;
1254 /* needed to ensure proper operation of coherent allocations
1255 * later, in case driver doesn't set it explicitly */
1256 dma_set_mask(&viodev->dev, DMA_BIT_MASK(64));
1257 dma_set_coherent_mask(&viodev->dev, DMA_BIT_MASK(64));
1258
1259 /* register with generic device framework */
1260 if (device_register(&viodev->dev)) {
1261 printk(KERN_ERR "%s: failed to register device %s\n",
1262 __func__, dev_name(&viodev->dev));
1263 put_device(&viodev->dev);
1264 return NULL;
1265 }
1266
1267 return viodev;
1268 }
1269 EXPORT_SYMBOL(vio_register_device_node);
1270
1271 /**
1272 * vio_bus_init: - Initialize the virtual IO bus
1273 */
1274 static int __init vio_bus_init(void)
1275 {
1276 int err;
1277 struct device_node *node_vroot;
1278
1279 if (firmware_has_feature(FW_FEATURE_CMO))
1280 vio_cmo_sysfs_init();
1281
1282 err = bus_register(&vio_bus_type);
1283 if (err) {
1284 printk(KERN_ERR "failed to register VIO bus\n");
1285 return err;
1286 }
1287
1288 /*
1289 * The fake parent of all vio devices, just to give us
1290 * a nice directory
1291 */
1292 err = device_register(&vio_bus_device.dev);
1293 if (err) {
1294 printk(KERN_WARNING "%s: device_register returned %i\n",
1295 __func__, err);
1296 return err;
1297 }
1298
1299 if (firmware_has_feature(FW_FEATURE_CMO))
1300 vio_cmo_bus_init();
1301
1302 node_vroot = of_find_node_by_name(NULL, "vdevice");
1303 if (node_vroot) {
1304 struct device_node *of_node;
1305
1306 /*
1307 * Create struct vio_devices for each virtual device in
1308 * the device tree. Drivers will associate with them later.
1309 */
1310 for (of_node = node_vroot->child; of_node != NULL;
1311 of_node = of_node->sibling)
1312 vio_register_device_node(of_node);
1313 of_node_put(node_vroot);
1314 }
1315
1316 return 0;
1317 }
1318 __initcall(vio_bus_init);
1319
1320 static ssize_t name_show(struct device *dev,
1321 struct device_attribute *attr, char *buf)
1322 {
1323 return sprintf(buf, "%s\n", to_vio_dev(dev)->name);
1324 }
1325
1326 static ssize_t devspec_show(struct device *dev,
1327 struct device_attribute *attr, char *buf)
1328 {
1329 struct device_node *of_node = dev->of_node;
1330
1331 return sprintf(buf, "%s\n", of_node ? of_node->full_name : "none");
1332 }
1333
1334 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1335 char *buf)
1336 {
1337 const struct vio_dev *vio_dev = to_vio_dev(dev);
1338 struct device_node *dn;
1339 const char *cp;
1340
1341 dn = dev->of_node;
1342 if (!dn)
1343 return -ENODEV;
1344 cp = of_get_property(dn, "compatible", NULL);
1345 if (!cp)
1346 return -ENODEV;
1347
1348 return sprintf(buf, "vio:T%sS%s\n", vio_dev->type, cp);
1349 }
1350
1351 static struct device_attribute vio_dev_attrs[] = {
1352 __ATTR_RO(name),
1353 __ATTR_RO(devspec),
1354 __ATTR_RO(modalias),
1355 __ATTR_NULL
1356 };
1357
1358 void __devinit vio_unregister_device(struct vio_dev *viodev)
1359 {
1360 device_unregister(&viodev->dev);
1361 }
1362 EXPORT_SYMBOL(vio_unregister_device);
1363
1364 static int vio_bus_match(struct device *dev, struct device_driver *drv)
1365 {
1366 const struct vio_dev *vio_dev = to_vio_dev(dev);
1367 struct vio_driver *vio_drv = to_vio_driver(drv);
1368 const struct vio_device_id *ids = vio_drv->id_table;
1369
1370 return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL);
1371 }
1372
1373 static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env)
1374 {
1375 const struct vio_dev *vio_dev = to_vio_dev(dev);
1376 struct device_node *dn;
1377 const char *cp;
1378
1379 dn = dev->of_node;
1380 if (!dn)
1381 return -ENODEV;
1382 cp = of_get_property(dn, "compatible", NULL);
1383 if (!cp)
1384 return -ENODEV;
1385
1386 add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
1387 return 0;
1388 }
1389
1390 static struct bus_type vio_bus_type = {
1391 .name = "vio",
1392 .dev_attrs = vio_dev_attrs,
1393 .uevent = vio_hotplug,
1394 .match = vio_bus_match,
1395 .probe = vio_bus_probe,
1396 .remove = vio_bus_remove,
1397 };
1398
1399 /**
1400 * vio_get_attribute: - get attribute for virtual device
1401 * @vdev: The vio device to get property.
1402 * @which: The property/attribute to be extracted.
1403 * @length: Pointer to length of returned data size (unused if NULL).
1404 *
1405 * Calls prom.c's of_get_property() to return the value of the
1406 * attribute specified by @which
1407 */
1408 const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1409 {
1410 return of_get_property(vdev->dev.of_node, which, length);
1411 }
1412 EXPORT_SYMBOL(vio_get_attribute);
1413
1414 #ifdef CONFIG_PPC_PSERIES
1415 /* vio_find_name() - internal because only vio.c knows how we formatted the
1416 * kobject name
1417 */
1418 static struct vio_dev *vio_find_name(const char *name)
1419 {
1420 struct device *found;
1421
1422 found = bus_find_device_by_name(&vio_bus_type, NULL, name);
1423 if (!found)
1424 return NULL;
1425
1426 return to_vio_dev(found);
1427 }
1428
1429 /**
1430 * vio_find_node - find an already-registered vio_dev
1431 * @vnode: device_node of the virtual device we're looking for
1432 */
1433 struct vio_dev *vio_find_node(struct device_node *vnode)
1434 {
1435 const uint32_t *unit_address;
1436 char kobj_name[20];
1437
1438 /* construct the kobject name from the device node */
1439 unit_address = of_get_property(vnode, "reg", NULL);
1440 if (!unit_address)
1441 return NULL;
1442 snprintf(kobj_name, sizeof(kobj_name), "%x", *unit_address);
1443
1444 return vio_find_name(kobj_name);
1445 }
1446 EXPORT_SYMBOL(vio_find_node);
1447
1448 int vio_enable_interrupts(struct vio_dev *dev)
1449 {
1450 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1451 if (rc != H_SUCCESS)
1452 printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1453 return rc;
1454 }
1455 EXPORT_SYMBOL(vio_enable_interrupts);
1456
1457 int vio_disable_interrupts(struct vio_dev *dev)
1458 {
1459 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1460 if (rc != H_SUCCESS)
1461 printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1462 return rc;
1463 }
1464 EXPORT_SYMBOL(vio_disable_interrupts);
1465 #endif /* CONFIG_PPC_PSERIES */