Loading...
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/cpu.h>
18#include <linux/types.h>
19#include <linux/delay.h>
20#include <linux/stat.h>
21#include <linux/device.h>
22#include <linux/init.h>
23#include <linux/slab.h>
24#include <linux/console.h>
25#include <linux/export.h>
26#include <linux/mm.h>
27#include <linux/dma-mapping.h>
28#include <linux/kobject.h>
29
30#include <asm/iommu.h>
31#include <asm/dma.h>
32#include <asm/vio.h>
33#include <asm/prom.h>
34#include <asm/firmware.h>
35#include <asm/tce.h>
36#include <asm/abs_addr.h>
37#include <asm/page.h>
38#include <asm/hvcall.h>
39
40static struct bus_type vio_bus_type;
41
42static struct vio_dev vio_bus_device = { /* fake "parent" device */
43 .name = "vio",
44 .type = "",
45 .dev.init_name = "vio",
46 .dev.bus = &vio_bus_type,
47};
48
49#ifdef CONFIG_PPC_SMLPAR
50/**
51 * vio_cmo_pool - A pool of IO memory for CMO use
52 *
53 * @size: The size of the pool in bytes
54 * @free: The amount of free memory in the pool
55 */
56struct vio_cmo_pool {
57 size_t size;
58 size_t free;
59};
60
61/* How many ms to delay queued balance work */
62#define VIO_CMO_BALANCE_DELAY 100
63
64/* Portion out IO memory to CMO devices by this chunk size */
65#define VIO_CMO_BALANCE_CHUNK 131072
66
67/**
68 * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement
69 *
70 * @vio_dev: struct vio_dev pointer
71 * @list: pointer to other devices on bus that are being tracked
72 */
73struct vio_cmo_dev_entry {
74 struct vio_dev *viodev;
75 struct list_head list;
76};
77
78/**
79 * vio_cmo - VIO bus accounting structure for CMO entitlement
80 *
81 * @lock: spinlock for entire structure
82 * @balance_q: work queue for balancing system entitlement
83 * @device_list: list of CMO-enabled devices requiring entitlement
84 * @entitled: total system entitlement in bytes
85 * @reserve: pool of memory from which devices reserve entitlement, incl. spare
86 * @excess: pool of excess entitlement not needed for device reserves or spare
87 * @spare: IO memory for device hotplug functionality
88 * @min: minimum necessary for system operation
89 * @desired: desired memory for system operation
90 * @curr: bytes currently allocated
91 * @high: high water mark for IO data usage
92 */
93struct vio_cmo {
94 spinlock_t lock;
95 struct delayed_work balance_q;
96 struct list_head device_list;
97 size_t entitled;
98 struct vio_cmo_pool reserve;
99 struct vio_cmo_pool excess;
100 size_t spare;
101 size_t min;
102 size_t desired;
103 size_t curr;
104 size_t high;
105} vio_cmo;
106
107/**
108 * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows
109 */
110static int vio_cmo_num_OF_devs(void)
111{
112 struct device_node *node_vroot;
113 int count = 0;
114
115 /*
116 * Count the number of vdevice entries with an
117 * ibm,my-dma-window OF property
118 */
119 node_vroot = of_find_node_by_name(NULL, "vdevice");
120 if (node_vroot) {
121 struct device_node *of_node;
122 struct property *prop;
123
124 for_each_child_of_node(node_vroot, of_node) {
125 prop = of_find_property(of_node, "ibm,my-dma-window",
126 NULL);
127 if (prop)
128 count++;
129 }
130 }
131 of_node_put(node_vroot);
132 return count;
133}
134
135/**
136 * vio_cmo_alloc - allocate IO memory for CMO-enable devices
137 *
138 * @viodev: VIO device requesting IO memory
139 * @size: size of allocation requested
140 *
141 * Allocations come from memory reserved for the devices and any excess
142 * IO memory available to all devices. The spare pool used to service
143 * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be
144 * made available.
145 *
146 * Return codes:
147 * 0 for successful allocation and -ENOMEM for a failure
148 */
149static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size)
150{
151 unsigned long flags;
152 size_t reserve_free = 0;
153 size_t excess_free = 0;
154 int ret = -ENOMEM;
155
156 spin_lock_irqsave(&vio_cmo.lock, flags);
157
158 /* Determine the amount of free entitlement available in reserve */
159 if (viodev->cmo.entitled > viodev->cmo.allocated)
160 reserve_free = viodev->cmo.entitled - viodev->cmo.allocated;
161
162 /* If spare is not fulfilled, the excess pool can not be used. */
163 if (vio_cmo.spare >= VIO_CMO_MIN_ENT)
164 excess_free = vio_cmo.excess.free;
165
166 /* The request can be satisfied */
167 if ((reserve_free + excess_free) >= size) {
168 vio_cmo.curr += size;
169 if (vio_cmo.curr > vio_cmo.high)
170 vio_cmo.high = vio_cmo.curr;
171 viodev->cmo.allocated += size;
172 size -= min(reserve_free, size);
173 vio_cmo.excess.free -= size;
174 ret = 0;
175 }
176
177 spin_unlock_irqrestore(&vio_cmo.lock, flags);
178 return ret;
179}
180
181/**
182 * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices
183 * @viodev: VIO device freeing IO memory
184 * @size: size of deallocation
185 *
186 * IO memory is freed by the device back to the correct memory pools.
187 * The spare pool is replenished first from either memory pool, then
188 * the reserve pool is used to reduce device entitlement, the excess
189 * pool is used to increase the reserve pool toward the desired entitlement
190 * target, and then the remaining memory is returned to the pools.
191 *
192 */
193static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size)
194{
195 unsigned long flags;
196 size_t spare_needed = 0;
197 size_t excess_freed = 0;
198 size_t reserve_freed = size;
199 size_t tmp;
200 int balance = 0;
201
202 spin_lock_irqsave(&vio_cmo.lock, flags);
203 vio_cmo.curr -= size;
204
205 /* Amount of memory freed from the excess pool */
206 if (viodev->cmo.allocated > viodev->cmo.entitled) {
207 excess_freed = min(reserve_freed, (viodev->cmo.allocated -
208 viodev->cmo.entitled));
209 reserve_freed -= excess_freed;
210 }
211
212 /* Remove allocation from device */
213 viodev->cmo.allocated -= (reserve_freed + excess_freed);
214
215 /* Spare is a subset of the reserve pool, replenish it first. */
216 spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare;
217
218 /*
219 * Replenish the spare in the reserve pool from the excess pool.
220 * This moves entitlement into the reserve pool.
221 */
222 if (spare_needed && excess_freed) {
223 tmp = min(excess_freed, spare_needed);
224 vio_cmo.excess.size -= tmp;
225 vio_cmo.reserve.size += tmp;
226 vio_cmo.spare += tmp;
227 excess_freed -= tmp;
228 spare_needed -= tmp;
229 balance = 1;
230 }
231
232 /*
233 * Replenish the spare in the reserve pool from the reserve pool.
234 * This removes entitlement from the device down to VIO_CMO_MIN_ENT,
235 * if needed, and gives it to the spare pool. The amount of used
236 * memory in this pool does not change.
237 */
238 if (spare_needed && reserve_freed) {
239 tmp = min3(spare_needed, reserve_freed, (viodev->cmo.entitled - VIO_CMO_MIN_ENT));
240
241 vio_cmo.spare += tmp;
242 viodev->cmo.entitled -= tmp;
243 reserve_freed -= tmp;
244 spare_needed -= tmp;
245 balance = 1;
246 }
247
248 /*
249 * Increase the reserve pool until the desired allocation is met.
250 * Move an allocation freed from the excess pool into the reserve
251 * pool and schedule a balance operation.
252 */
253 if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) {
254 tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size));
255
256 vio_cmo.excess.size -= tmp;
257 vio_cmo.reserve.size += tmp;
258 excess_freed -= tmp;
259 balance = 1;
260 }
261
262 /* Return memory from the excess pool to that pool */
263 if (excess_freed)
264 vio_cmo.excess.free += excess_freed;
265
266 if (balance)
267 schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY);
268 spin_unlock_irqrestore(&vio_cmo.lock, flags);
269}
270
271/**
272 * vio_cmo_entitlement_update - Manage system entitlement changes
273 *
274 * @new_entitlement: new system entitlement to attempt to accommodate
275 *
276 * Increases in entitlement will be used to fulfill the spare entitlement
277 * and the rest is given to the excess pool. Decreases, if they are
278 * possible, come from the excess pool and from unused device entitlement
279 *
280 * Returns: 0 on success, -ENOMEM when change can not be made
281 */
282int vio_cmo_entitlement_update(size_t new_entitlement)
283{
284 struct vio_dev *viodev;
285 struct vio_cmo_dev_entry *dev_ent;
286 unsigned long flags;
287 size_t avail, delta, tmp;
288
289 spin_lock_irqsave(&vio_cmo.lock, flags);
290
291 /* Entitlement increases */
292 if (new_entitlement > vio_cmo.entitled) {
293 delta = new_entitlement - vio_cmo.entitled;
294
295 /* Fulfill spare allocation */
296 if (vio_cmo.spare < VIO_CMO_MIN_ENT) {
297 tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare));
298 vio_cmo.spare += tmp;
299 vio_cmo.reserve.size += tmp;
300 delta -= tmp;
301 }
302
303 /* Remaining new allocation goes to the excess pool */
304 vio_cmo.entitled += delta;
305 vio_cmo.excess.size += delta;
306 vio_cmo.excess.free += delta;
307
308 goto out;
309 }
310
311 /* Entitlement decreases */
312 delta = vio_cmo.entitled - new_entitlement;
313 avail = vio_cmo.excess.free;
314
315 /*
316 * Need to check how much unused entitlement each device can
317 * sacrifice to fulfill entitlement change.
318 */
319 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
320 if (avail >= delta)
321 break;
322
323 viodev = dev_ent->viodev;
324 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
325 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
326 avail += viodev->cmo.entitled -
327 max_t(size_t, viodev->cmo.allocated,
328 VIO_CMO_MIN_ENT);
329 }
330
331 if (delta <= avail) {
332 vio_cmo.entitled -= delta;
333
334 /* Take entitlement from the excess pool first */
335 tmp = min(vio_cmo.excess.free, delta);
336 vio_cmo.excess.size -= tmp;
337 vio_cmo.excess.free -= tmp;
338 delta -= tmp;
339
340 /*
341 * Remove all but VIO_CMO_MIN_ENT bytes from devices
342 * until entitlement change is served
343 */
344 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
345 if (!delta)
346 break;
347
348 viodev = dev_ent->viodev;
349 tmp = 0;
350 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
351 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
352 tmp = viodev->cmo.entitled -
353 max_t(size_t, viodev->cmo.allocated,
354 VIO_CMO_MIN_ENT);
355 viodev->cmo.entitled -= min(tmp, delta);
356 delta -= min(tmp, delta);
357 }
358 } else {
359 spin_unlock_irqrestore(&vio_cmo.lock, flags);
360 return -ENOMEM;
361 }
362
363out:
364 schedule_delayed_work(&vio_cmo.balance_q, 0);
365 spin_unlock_irqrestore(&vio_cmo.lock, flags);
366 return 0;
367}
368
369/**
370 * vio_cmo_balance - Balance entitlement among devices
371 *
372 * @work: work queue structure for this operation
373 *
374 * Any system entitlement above the minimum needed for devices, or
375 * already allocated to devices, can be distributed to the devices.
376 * The list of devices is iterated through to recalculate the desired
377 * entitlement level and to determine how much entitlement above the
378 * minimum entitlement is allocated to devices.
379 *
380 * Small chunks of the available entitlement are given to devices until
381 * their requirements are fulfilled or there is no entitlement left to give.
382 * Upon completion sizes of the reserve and excess pools are calculated.
383 *
384 * The system minimum entitlement level is also recalculated here.
385 * Entitlement will be reserved for devices even after vio_bus_remove to
386 * accommodate reloading the driver. The OF tree is walked to count the
387 * number of devices present and this will remove entitlement for devices
388 * that have actually left the system after having vio_bus_remove called.
389 */
390static void vio_cmo_balance(struct work_struct *work)
391{
392 struct vio_cmo *cmo;
393 struct vio_dev *viodev;
394 struct vio_cmo_dev_entry *dev_ent;
395 unsigned long flags;
396 size_t avail = 0, level, chunk, need;
397 int devcount = 0, fulfilled;
398
399 cmo = container_of(work, struct vio_cmo, balance_q.work);
400
401 spin_lock_irqsave(&vio_cmo.lock, flags);
402
403 /* Calculate minimum entitlement and fulfill spare */
404 cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT;
405 BUG_ON(cmo->min > cmo->entitled);
406 cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min));
407 cmo->min += cmo->spare;
408 cmo->desired = cmo->min;
409
410 /*
411 * Determine how much entitlement is available and reset device
412 * entitlements
413 */
414 avail = cmo->entitled - cmo->spare;
415 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
416 viodev = dev_ent->viodev;
417 devcount++;
418 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
419 cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT);
420 avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT);
421 }
422
423 /*
424 * Having provided each device with the minimum entitlement, loop
425 * over the devices portioning out the remaining entitlement
426 * until there is nothing left.
427 */
428 level = VIO_CMO_MIN_ENT;
429 while (avail) {
430 fulfilled = 0;
431 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
432 viodev = dev_ent->viodev;
433
434 if (viodev->cmo.desired <= level) {
435 fulfilled++;
436 continue;
437 }
438
439 /*
440 * Give the device up to VIO_CMO_BALANCE_CHUNK
441 * bytes of entitlement, but do not exceed the
442 * desired level of entitlement for the device.
443 */
444 chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK);
445 chunk = min(chunk, (viodev->cmo.desired -
446 viodev->cmo.entitled));
447 viodev->cmo.entitled += chunk;
448
449 /*
450 * If the memory for this entitlement increase was
451 * already allocated to the device it does not come
452 * from the available pool being portioned out.
453 */
454 need = max(viodev->cmo.allocated, viodev->cmo.entitled)-
455 max(viodev->cmo.allocated, level);
456 avail -= need;
457
458 }
459 if (fulfilled == devcount)
460 break;
461 level += VIO_CMO_BALANCE_CHUNK;
462 }
463
464 /* Calculate new reserve and excess pool sizes */
465 cmo->reserve.size = cmo->min;
466 cmo->excess.free = 0;
467 cmo->excess.size = 0;
468 need = 0;
469 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
470 viodev = dev_ent->viodev;
471 /* Calculated reserve size above the minimum entitlement */
472 if (viodev->cmo.entitled)
473 cmo->reserve.size += (viodev->cmo.entitled -
474 VIO_CMO_MIN_ENT);
475 /* Calculated used excess entitlement */
476 if (viodev->cmo.allocated > viodev->cmo.entitled)
477 need += viodev->cmo.allocated - viodev->cmo.entitled;
478 }
479 cmo->excess.size = cmo->entitled - cmo->reserve.size;
480 cmo->excess.free = cmo->excess.size - need;
481
482 cancel_delayed_work(to_delayed_work(work));
483 spin_unlock_irqrestore(&vio_cmo.lock, flags);
484}
485
486static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size,
487 dma_addr_t *dma_handle, gfp_t flag,
488 struct dma_attrs *attrs)
489{
490 struct vio_dev *viodev = to_vio_dev(dev);
491 void *ret;
492
493 if (vio_cmo_alloc(viodev, roundup(size, PAGE_SIZE))) {
494 atomic_inc(&viodev->cmo.allocs_failed);
495 return NULL;
496 }
497
498 ret = dma_iommu_ops.alloc(dev, size, dma_handle, flag, attrs);
499 if (unlikely(ret == NULL)) {
500 vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
501 atomic_inc(&viodev->cmo.allocs_failed);
502 }
503
504 return ret;
505}
506
507static void vio_dma_iommu_free_coherent(struct device *dev, size_t size,
508 void *vaddr, dma_addr_t dma_handle,
509 struct dma_attrs *attrs)
510{
511 struct vio_dev *viodev = to_vio_dev(dev);
512
513 dma_iommu_ops.free(dev, size, vaddr, dma_handle, attrs);
514
515 vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
516}
517
518static dma_addr_t vio_dma_iommu_map_page(struct device *dev, struct page *page,
519 unsigned long offset, size_t size,
520 enum dma_data_direction direction,
521 struct dma_attrs *attrs)
522{
523 struct vio_dev *viodev = to_vio_dev(dev);
524 dma_addr_t ret = DMA_ERROR_CODE;
525
526 if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) {
527 atomic_inc(&viodev->cmo.allocs_failed);
528 return ret;
529 }
530
531 ret = dma_iommu_ops.map_page(dev, page, offset, size, direction, attrs);
532 if (unlikely(dma_mapping_error(dev, ret))) {
533 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
534 atomic_inc(&viodev->cmo.allocs_failed);
535 }
536
537 return ret;
538}
539
540static void vio_dma_iommu_unmap_page(struct device *dev, dma_addr_t dma_handle,
541 size_t size,
542 enum dma_data_direction direction,
543 struct dma_attrs *attrs)
544{
545 struct vio_dev *viodev = to_vio_dev(dev);
546
547 dma_iommu_ops.unmap_page(dev, dma_handle, size, direction, attrs);
548
549 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
550}
551
552static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
553 int nelems, enum dma_data_direction direction,
554 struct dma_attrs *attrs)
555{
556 struct vio_dev *viodev = to_vio_dev(dev);
557 struct scatterlist *sgl;
558 int ret, count = 0;
559 size_t alloc_size = 0;
560
561 for (sgl = sglist; count < nelems; count++, sgl++)
562 alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE);
563
564 if (vio_cmo_alloc(viodev, alloc_size)) {
565 atomic_inc(&viodev->cmo.allocs_failed);
566 return 0;
567 }
568
569 ret = dma_iommu_ops.map_sg(dev, sglist, nelems, direction, attrs);
570
571 if (unlikely(!ret)) {
572 vio_cmo_dealloc(viodev, alloc_size);
573 atomic_inc(&viodev->cmo.allocs_failed);
574 return ret;
575 }
576
577 for (sgl = sglist, count = 0; count < ret; count++, sgl++)
578 alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
579 if (alloc_size)
580 vio_cmo_dealloc(viodev, alloc_size);
581
582 return ret;
583}
584
585static void vio_dma_iommu_unmap_sg(struct device *dev,
586 struct scatterlist *sglist, int nelems,
587 enum dma_data_direction direction,
588 struct dma_attrs *attrs)
589{
590 struct vio_dev *viodev = to_vio_dev(dev);
591 struct scatterlist *sgl;
592 size_t alloc_size = 0;
593 int count = 0;
594
595 for (sgl = sglist; count < nelems; count++, sgl++)
596 alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
597
598 dma_iommu_ops.unmap_sg(dev, sglist, nelems, direction, attrs);
599
600 vio_cmo_dealloc(viodev, alloc_size);
601}
602
603static int vio_dma_iommu_dma_supported(struct device *dev, u64 mask)
604{
605 return dma_iommu_ops.dma_supported(dev, mask);
606}
607
608static u64 vio_dma_get_required_mask(struct device *dev)
609{
610 return dma_iommu_ops.get_required_mask(dev);
611}
612
613struct dma_map_ops vio_dma_mapping_ops = {
614 .alloc = vio_dma_iommu_alloc_coherent,
615 .free = vio_dma_iommu_free_coherent,
616 .map_sg = vio_dma_iommu_map_sg,
617 .unmap_sg = vio_dma_iommu_unmap_sg,
618 .map_page = vio_dma_iommu_map_page,
619 .unmap_page = vio_dma_iommu_unmap_page,
620 .dma_supported = vio_dma_iommu_dma_supported,
621 .get_required_mask = vio_dma_get_required_mask,
622};
623
624/**
625 * vio_cmo_set_dev_desired - Set desired entitlement for a device
626 *
627 * @viodev: struct vio_dev for device to alter
628 * @new_desired: new desired entitlement level in bytes
629 *
630 * For use by devices to request a change to their entitlement at runtime or
631 * through sysfs. The desired entitlement level is changed and a balancing
632 * of system resources is scheduled to run in the future.
633 */
634void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
635{
636 unsigned long flags;
637 struct vio_cmo_dev_entry *dev_ent;
638 int found = 0;
639
640 if (!firmware_has_feature(FW_FEATURE_CMO))
641 return;
642
643 spin_lock_irqsave(&vio_cmo.lock, flags);
644 if (desired < VIO_CMO_MIN_ENT)
645 desired = VIO_CMO_MIN_ENT;
646
647 /*
648 * Changes will not be made for devices not in the device list.
649 * If it is not in the device list, then no driver is loaded
650 * for the device and it can not receive entitlement.
651 */
652 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
653 if (viodev == dev_ent->viodev) {
654 found = 1;
655 break;
656 }
657 if (!found) {
658 spin_unlock_irqrestore(&vio_cmo.lock, flags);
659 return;
660 }
661
662 /* Increase/decrease in desired device entitlement */
663 if (desired >= viodev->cmo.desired) {
664 /* Just bump the bus and device values prior to a balance*/
665 vio_cmo.desired += desired - viodev->cmo.desired;
666 viodev->cmo.desired = desired;
667 } else {
668 /* Decrease bus and device values for desired entitlement */
669 vio_cmo.desired -= viodev->cmo.desired - desired;
670 viodev->cmo.desired = desired;
671 /*
672 * If less entitlement is desired than current entitlement, move
673 * any reserve memory in the change region to the excess pool.
674 */
675 if (viodev->cmo.entitled > desired) {
676 vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
677 vio_cmo.excess.size += viodev->cmo.entitled - desired;
678 /*
679 * If entitlement moving from the reserve pool to the
680 * excess pool is currently unused, add to the excess
681 * free counter.
682 */
683 if (viodev->cmo.allocated < viodev->cmo.entitled)
684 vio_cmo.excess.free += viodev->cmo.entitled -
685 max(viodev->cmo.allocated, desired);
686 viodev->cmo.entitled = desired;
687 }
688 }
689 schedule_delayed_work(&vio_cmo.balance_q, 0);
690 spin_unlock_irqrestore(&vio_cmo.lock, flags);
691}
692
693/**
694 * vio_cmo_bus_probe - Handle CMO specific bus probe activities
695 *
696 * @viodev - Pointer to struct vio_dev for device
697 *
698 * Determine the devices IO memory entitlement needs, attempting
699 * to satisfy the system minimum entitlement at first and scheduling
700 * a balance operation to take care of the rest at a later time.
701 *
702 * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
703 * -ENOMEM when entitlement is not available for device or
704 * device entry.
705 *
706 */
707static int vio_cmo_bus_probe(struct vio_dev *viodev)
708{
709 struct vio_cmo_dev_entry *dev_ent;
710 struct device *dev = &viodev->dev;
711 struct vio_driver *viodrv = to_vio_driver(dev->driver);
712 unsigned long flags;
713 size_t size;
714 bool dma_capable = false;
715
716 /* A device requires entitlement if it has a DMA window property */
717 switch (viodev->family) {
718 case VDEVICE:
719 if (of_get_property(viodev->dev.of_node,
720 "ibm,my-dma-window", NULL))
721 dma_capable = true;
722 break;
723 case PFO:
724 dma_capable = false;
725 break;
726 default:
727 dev_warn(dev, "unknown device family: %d\n", viodev->family);
728 BUG();
729 break;
730 }
731
732 /* Configure entitlement for the device. */
733 if (dma_capable) {
734 /* Check that the driver is CMO enabled and get desired DMA */
735 if (!viodrv->get_desired_dma) {
736 dev_err(dev, "%s: device driver does not support CMO\n",
737 __func__);
738 return -EINVAL;
739 }
740
741 viodev->cmo.desired = IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev));
742 if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
743 viodev->cmo.desired = VIO_CMO_MIN_ENT;
744 size = VIO_CMO_MIN_ENT;
745
746 dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
747 GFP_KERNEL);
748 if (!dev_ent)
749 return -ENOMEM;
750
751 dev_ent->viodev = viodev;
752 spin_lock_irqsave(&vio_cmo.lock, flags);
753 list_add(&dev_ent->list, &vio_cmo.device_list);
754 } else {
755 viodev->cmo.desired = 0;
756 size = 0;
757 spin_lock_irqsave(&vio_cmo.lock, flags);
758 }
759
760 /*
761 * If the needs for vio_cmo.min have not changed since they
762 * were last set, the number of devices in the OF tree has
763 * been constant and the IO memory for this is already in
764 * the reserve pool.
765 */
766 if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
767 VIO_CMO_MIN_ENT)) {
768 /* Updated desired entitlement if device requires it */
769 if (size)
770 vio_cmo.desired += (viodev->cmo.desired -
771 VIO_CMO_MIN_ENT);
772 } else {
773 size_t tmp;
774
775 tmp = vio_cmo.spare + vio_cmo.excess.free;
776 if (tmp < size) {
777 dev_err(dev, "%s: insufficient free "
778 "entitlement to add device. "
779 "Need %lu, have %lu\n", __func__,
780 size, (vio_cmo.spare + tmp));
781 spin_unlock_irqrestore(&vio_cmo.lock, flags);
782 return -ENOMEM;
783 }
784
785 /* Use excess pool first to fulfill request */
786 tmp = min(size, vio_cmo.excess.free);
787 vio_cmo.excess.free -= tmp;
788 vio_cmo.excess.size -= tmp;
789 vio_cmo.reserve.size += tmp;
790
791 /* Use spare if excess pool was insufficient */
792 vio_cmo.spare -= size - tmp;
793
794 /* Update bus accounting */
795 vio_cmo.min += size;
796 vio_cmo.desired += viodev->cmo.desired;
797 }
798 spin_unlock_irqrestore(&vio_cmo.lock, flags);
799 return 0;
800}
801
802/**
803 * vio_cmo_bus_remove - Handle CMO specific bus removal activities
804 *
805 * @viodev - Pointer to struct vio_dev for device
806 *
807 * Remove the device from the cmo device list. The minimum entitlement
808 * will be reserved for the device as long as it is in the system. The
809 * rest of the entitlement the device had been allocated will be returned
810 * to the system.
811 */
812static void vio_cmo_bus_remove(struct vio_dev *viodev)
813{
814 struct vio_cmo_dev_entry *dev_ent;
815 unsigned long flags;
816 size_t tmp;
817
818 spin_lock_irqsave(&vio_cmo.lock, flags);
819 if (viodev->cmo.allocated) {
820 dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
821 "allocated after remove operation.\n",
822 __func__, viodev->cmo.allocated);
823 BUG();
824 }
825
826 /*
827 * Remove the device from the device list being maintained for
828 * CMO enabled devices.
829 */
830 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
831 if (viodev == dev_ent->viodev) {
832 list_del(&dev_ent->list);
833 kfree(dev_ent);
834 break;
835 }
836
837 /*
838 * Devices may not require any entitlement and they do not need
839 * to be processed. Otherwise, return the device's entitlement
840 * back to the pools.
841 */
842 if (viodev->cmo.entitled) {
843 /*
844 * This device has not yet left the OF tree, it's
845 * minimum entitlement remains in vio_cmo.min and
846 * vio_cmo.desired
847 */
848 vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
849
850 /*
851 * Save min allocation for device in reserve as long
852 * as it exists in OF tree as determined by later
853 * balance operation
854 */
855 viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
856
857 /* Replenish spare from freed reserve pool */
858 if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
859 tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
860 vio_cmo.spare));
861 vio_cmo.spare += tmp;
862 viodev->cmo.entitled -= tmp;
863 }
864
865 /* Remaining reserve goes to excess pool */
866 vio_cmo.excess.size += viodev->cmo.entitled;
867 vio_cmo.excess.free += viodev->cmo.entitled;
868 vio_cmo.reserve.size -= viodev->cmo.entitled;
869
870 /*
871 * Until the device is removed it will keep a
872 * minimum entitlement; this will guarantee that
873 * a module unload/load will result in a success.
874 */
875 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
876 viodev->cmo.desired = VIO_CMO_MIN_ENT;
877 atomic_set(&viodev->cmo.allocs_failed, 0);
878 }
879
880 spin_unlock_irqrestore(&vio_cmo.lock, flags);
881}
882
883static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
884{
885 set_dma_ops(&viodev->dev, &vio_dma_mapping_ops);
886}
887
888/**
889 * vio_cmo_bus_init - CMO entitlement initialization at bus init time
890 *
891 * Set up the reserve and excess entitlement pools based on available
892 * system entitlement and the number of devices in the OF tree that
893 * require entitlement in the reserve pool.
894 */
895static void vio_cmo_bus_init(void)
896{
897 struct hvcall_mpp_data mpp_data;
898 int err;
899
900 memset(&vio_cmo, 0, sizeof(struct vio_cmo));
901 spin_lock_init(&vio_cmo.lock);
902 INIT_LIST_HEAD(&vio_cmo.device_list);
903 INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
904
905 /* Get current system entitlement */
906 err = h_get_mpp(&mpp_data);
907
908 /*
909 * On failure, continue with entitlement set to 0, will panic()
910 * later when spare is reserved.
911 */
912 if (err != H_SUCCESS) {
913 printk(KERN_ERR "%s: unable to determine system IO "\
914 "entitlement. (%d)\n", __func__, err);
915 vio_cmo.entitled = 0;
916 } else {
917 vio_cmo.entitled = mpp_data.entitled_mem;
918 }
919
920 /* Set reservation and check against entitlement */
921 vio_cmo.spare = VIO_CMO_MIN_ENT;
922 vio_cmo.reserve.size = vio_cmo.spare;
923 vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
924 VIO_CMO_MIN_ENT);
925 if (vio_cmo.reserve.size > vio_cmo.entitled) {
926 printk(KERN_ERR "%s: insufficient system entitlement\n",
927 __func__);
928 panic("%s: Insufficient system entitlement", __func__);
929 }
930
931 /* Set the remaining accounting variables */
932 vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
933 vio_cmo.excess.free = vio_cmo.excess.size;
934 vio_cmo.min = vio_cmo.reserve.size;
935 vio_cmo.desired = vio_cmo.reserve.size;
936}
937
938/* sysfs device functions and data structures for CMO */
939
940#define viodev_cmo_rd_attr(name) \
941static ssize_t viodev_cmo_##name##_show(struct device *dev, \
942 struct device_attribute *attr, \
943 char *buf) \
944{ \
945 return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \
946}
947
948static ssize_t viodev_cmo_allocs_failed_show(struct device *dev,
949 struct device_attribute *attr, char *buf)
950{
951 struct vio_dev *viodev = to_vio_dev(dev);
952 return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
953}
954
955static ssize_t viodev_cmo_allocs_failed_reset(struct device *dev,
956 struct device_attribute *attr, const char *buf, size_t count)
957{
958 struct vio_dev *viodev = to_vio_dev(dev);
959 atomic_set(&viodev->cmo.allocs_failed, 0);
960 return count;
961}
962
963static ssize_t viodev_cmo_desired_set(struct device *dev,
964 struct device_attribute *attr, const char *buf, size_t count)
965{
966 struct vio_dev *viodev = to_vio_dev(dev);
967 size_t new_desired;
968 int ret;
969
970 ret = strict_strtoul(buf, 10, &new_desired);
971 if (ret)
972 return ret;
973
974 vio_cmo_set_dev_desired(viodev, new_desired);
975 return count;
976}
977
978viodev_cmo_rd_attr(desired);
979viodev_cmo_rd_attr(entitled);
980viodev_cmo_rd_attr(allocated);
981
982static ssize_t name_show(struct device *, struct device_attribute *, char *);
983static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
984static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
985 char *buf);
986static struct device_attribute vio_cmo_dev_attrs[] = {
987 __ATTR_RO(name),
988 __ATTR_RO(devspec),
989 __ATTR_RO(modalias),
990 __ATTR(cmo_desired, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
991 viodev_cmo_desired_show, viodev_cmo_desired_set),
992 __ATTR(cmo_entitled, S_IRUGO, viodev_cmo_entitled_show, NULL),
993 __ATTR(cmo_allocated, S_IRUGO, viodev_cmo_allocated_show, NULL),
994 __ATTR(cmo_allocs_failed, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
995 viodev_cmo_allocs_failed_show, viodev_cmo_allocs_failed_reset),
996 __ATTR_NULL
997};
998
999/* sysfs bus functions and data structures for CMO */
1000
1001#define viobus_cmo_rd_attr(name) \
1002static ssize_t \
1003viobus_cmo_##name##_show(struct bus_type *bt, char *buf) \
1004{ \
1005 return sprintf(buf, "%lu\n", vio_cmo.name); \
1006}
1007
1008#define viobus_cmo_pool_rd_attr(name, var) \
1009static ssize_t \
1010viobus_cmo_##name##_pool_show_##var(struct bus_type *bt, char *buf) \
1011{ \
1012 return sprintf(buf, "%lu\n", vio_cmo.name.var); \
1013}
1014
1015static ssize_t viobus_cmo_high_reset(struct bus_type *bt, const char *buf,
1016 size_t count)
1017{
1018 unsigned long flags;
1019
1020 spin_lock_irqsave(&vio_cmo.lock, flags);
1021 vio_cmo.high = vio_cmo.curr;
1022 spin_unlock_irqrestore(&vio_cmo.lock, flags);
1023
1024 return count;
1025}
1026
1027viobus_cmo_rd_attr(entitled);
1028viobus_cmo_pool_rd_attr(reserve, size);
1029viobus_cmo_pool_rd_attr(excess, size);
1030viobus_cmo_pool_rd_attr(excess, free);
1031viobus_cmo_rd_attr(spare);
1032viobus_cmo_rd_attr(min);
1033viobus_cmo_rd_attr(desired);
1034viobus_cmo_rd_attr(curr);
1035viobus_cmo_rd_attr(high);
1036
1037static struct bus_attribute vio_cmo_bus_attrs[] = {
1038 __ATTR(cmo_entitled, S_IRUGO, viobus_cmo_entitled_show, NULL),
1039 __ATTR(cmo_reserve_size, S_IRUGO, viobus_cmo_reserve_pool_show_size, NULL),
1040 __ATTR(cmo_excess_size, S_IRUGO, viobus_cmo_excess_pool_show_size, NULL),
1041 __ATTR(cmo_excess_free, S_IRUGO, viobus_cmo_excess_pool_show_free, NULL),
1042 __ATTR(cmo_spare, S_IRUGO, viobus_cmo_spare_show, NULL),
1043 __ATTR(cmo_min, S_IRUGO, viobus_cmo_min_show, NULL),
1044 __ATTR(cmo_desired, S_IRUGO, viobus_cmo_desired_show, NULL),
1045 __ATTR(cmo_curr, S_IRUGO, viobus_cmo_curr_show, NULL),
1046 __ATTR(cmo_high, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
1047 viobus_cmo_high_show, viobus_cmo_high_reset),
1048 __ATTR_NULL
1049};
1050
1051static void vio_cmo_sysfs_init(void)
1052{
1053 vio_bus_type.dev_attrs = vio_cmo_dev_attrs;
1054 vio_bus_type.bus_attrs = vio_cmo_bus_attrs;
1055}
1056#else /* CONFIG_PPC_SMLPAR */
1057int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1058void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1059static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1060static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1061static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
1062static void vio_cmo_bus_init(void) {}
1063static void vio_cmo_sysfs_init(void) { }
1064#endif /* CONFIG_PPC_SMLPAR */
1065EXPORT_SYMBOL(vio_cmo_entitlement_update);
1066EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1067
1068
1069/*
1070 * Platform Facilities Option (PFO) support
1071 */
1072
1073/**
1074 * vio_h_cop_sync - Perform a synchronous PFO co-processor operation
1075 *
1076 * @vdev - Pointer to a struct vio_dev for device
1077 * @op - Pointer to a struct vio_pfo_op for the operation parameters
1078 *
1079 * Calls the hypervisor to synchronously perform the PFO operation
1080 * described in @op. In the case of a busy response from the hypervisor,
1081 * the operation will be re-submitted indefinitely unless a non-zero timeout
1082 * is specified or an error occurs. The timeout places a limit on when to
1083 * stop re-submitting a operation, the total time can be exceeded if an
1084 * operation is in progress.
1085 *
1086 * If op->hcall_ret is not NULL, this will be set to the return from the
1087 * last h_cop_op call or it will be 0 if an error not involving the h_call
1088 * was encountered.
1089 *
1090 * Returns:
1091 * 0 on success,
1092 * -EINVAL if the h_call fails due to an invalid parameter,
1093 * -E2BIG if the h_call can not be performed synchronously,
1094 * -EBUSY if a timeout is specified and has elapsed,
1095 * -EACCES if the memory area for data/status has been rescinded, or
1096 * -EPERM if a hardware fault has been indicated
1097 */
1098int vio_h_cop_sync(struct vio_dev *vdev, struct vio_pfo_op *op)
1099{
1100 struct device *dev = &vdev->dev;
1101 unsigned long deadline = 0;
1102 long hret = 0;
1103 int ret = 0;
1104
1105 if (op->timeout)
1106 deadline = jiffies + msecs_to_jiffies(op->timeout);
1107
1108 while (true) {
1109 hret = plpar_hcall_norets(H_COP, op->flags,
1110 vdev->resource_id,
1111 op->in, op->inlen, op->out,
1112 op->outlen, op->csbcpb);
1113
1114 if (hret == H_SUCCESS ||
1115 (hret != H_NOT_ENOUGH_RESOURCES &&
1116 hret != H_BUSY && hret != H_RESOURCE) ||
1117 (op->timeout && time_after(deadline, jiffies)))
1118 break;
1119
1120 dev_dbg(dev, "%s: hcall ret(%ld), retrying.\n", __func__, hret);
1121 }
1122
1123 switch (hret) {
1124 case H_SUCCESS:
1125 ret = 0;
1126 break;
1127 case H_OP_MODE:
1128 case H_TOO_BIG:
1129 ret = -E2BIG;
1130 break;
1131 case H_RESCINDED:
1132 ret = -EACCES;
1133 break;
1134 case H_HARDWARE:
1135 ret = -EPERM;
1136 break;
1137 case H_NOT_ENOUGH_RESOURCES:
1138 case H_RESOURCE:
1139 case H_BUSY:
1140 ret = -EBUSY;
1141 break;
1142 default:
1143 ret = -EINVAL;
1144 break;
1145 }
1146
1147 if (ret)
1148 dev_dbg(dev, "%s: Sync h_cop_op failure (ret:%d) (hret:%ld)\n",
1149 __func__, ret, hret);
1150
1151 op->hcall_err = hret;
1152 return ret;
1153}
1154EXPORT_SYMBOL(vio_h_cop_sync);
1155
1156static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1157{
1158 const unsigned char *dma_window;
1159 struct iommu_table *tbl;
1160 unsigned long offset, size;
1161
1162 dma_window = of_get_property(dev->dev.of_node,
1163 "ibm,my-dma-window", NULL);
1164 if (!dma_window)
1165 return NULL;
1166
1167 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
1168 if (tbl == NULL)
1169 return NULL;
1170
1171 of_parse_dma_window(dev->dev.of_node, dma_window,
1172 &tbl->it_index, &offset, &size);
1173
1174 /* TCE table size - measured in tce entries */
1175 tbl->it_size = size >> IOMMU_PAGE_SHIFT;
1176 /* offset for VIO should always be 0 */
1177 tbl->it_offset = offset >> IOMMU_PAGE_SHIFT;
1178 tbl->it_busno = 0;
1179 tbl->it_type = TCE_VB;
1180 tbl->it_blocksize = 16;
1181
1182 return iommu_init_table(tbl, -1);
1183}
1184
1185/**
1186 * vio_match_device: - Tell if a VIO device has a matching
1187 * VIO device id structure.
1188 * @ids: array of VIO device id structures to search in
1189 * @dev: the VIO device structure to match against
1190 *
1191 * Used by a driver to check whether a VIO device present in the
1192 * system is in its list of supported devices. Returns the matching
1193 * vio_device_id structure or NULL if there is no match.
1194 */
1195static const struct vio_device_id *vio_match_device(
1196 const struct vio_device_id *ids, const struct vio_dev *dev)
1197{
1198 while (ids->type[0] != '\0') {
1199 if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
1200 of_device_is_compatible(dev->dev.of_node,
1201 ids->compat))
1202 return ids;
1203 ids++;
1204 }
1205 return NULL;
1206}
1207
1208/*
1209 * Convert from struct device to struct vio_dev and pass to driver.
1210 * dev->driver has already been set by generic code because vio_bus_match
1211 * succeeded.
1212 */
1213static int vio_bus_probe(struct device *dev)
1214{
1215 struct vio_dev *viodev = to_vio_dev(dev);
1216 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1217 const struct vio_device_id *id;
1218 int error = -ENODEV;
1219
1220 if (!viodrv->probe)
1221 return error;
1222
1223 id = vio_match_device(viodrv->id_table, viodev);
1224 if (id) {
1225 memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1226 if (firmware_has_feature(FW_FEATURE_CMO)) {
1227 error = vio_cmo_bus_probe(viodev);
1228 if (error)
1229 return error;
1230 }
1231 error = viodrv->probe(viodev, id);
1232 if (error && firmware_has_feature(FW_FEATURE_CMO))
1233 vio_cmo_bus_remove(viodev);
1234 }
1235
1236 return error;
1237}
1238
1239/* convert from struct device to struct vio_dev and pass to driver. */
1240static int vio_bus_remove(struct device *dev)
1241{
1242 struct vio_dev *viodev = to_vio_dev(dev);
1243 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1244 struct device *devptr;
1245 int ret = 1;
1246
1247 /*
1248 * Hold a reference to the device after the remove function is called
1249 * to allow for CMO accounting cleanup for the device.
1250 */
1251 devptr = get_device(dev);
1252
1253 if (viodrv->remove)
1254 ret = viodrv->remove(viodev);
1255
1256 if (!ret && firmware_has_feature(FW_FEATURE_CMO))
1257 vio_cmo_bus_remove(viodev);
1258
1259 put_device(devptr);
1260 return ret;
1261}
1262
1263/**
1264 * vio_register_driver: - Register a new vio driver
1265 * @drv: The vio_driver structure to be registered.
1266 */
1267int __vio_register_driver(struct vio_driver *viodrv, struct module *owner,
1268 const char *mod_name)
1269{
1270 pr_debug("%s: driver %s registering\n", __func__, viodrv->name);
1271
1272 /* fill in 'struct driver' fields */
1273 viodrv->driver.name = viodrv->name;
1274 viodrv->driver.pm = viodrv->pm;
1275 viodrv->driver.bus = &vio_bus_type;
1276 viodrv->driver.owner = owner;
1277 viodrv->driver.mod_name = mod_name;
1278
1279 return driver_register(&viodrv->driver);
1280}
1281EXPORT_SYMBOL(__vio_register_driver);
1282
1283/**
1284 * vio_unregister_driver - Remove registration of vio driver.
1285 * @driver: The vio_driver struct to be removed form registration
1286 */
1287void vio_unregister_driver(struct vio_driver *viodrv)
1288{
1289 driver_unregister(&viodrv->driver);
1290}
1291EXPORT_SYMBOL(vio_unregister_driver);
1292
1293/* vio_dev refcount hit 0 */
1294static void __devinit vio_dev_release(struct device *dev)
1295{
1296 struct iommu_table *tbl = get_iommu_table_base(dev);
1297
1298 if (tbl)
1299 iommu_free_table(tbl, dev->of_node ?
1300 dev->of_node->full_name : dev_name(dev));
1301 of_node_put(dev->of_node);
1302 kfree(to_vio_dev(dev));
1303}
1304
1305/**
1306 * vio_register_device_node: - Register a new vio device.
1307 * @of_node: The OF node for this device.
1308 *
1309 * Creates and initializes a vio_dev structure from the data in
1310 * of_node and adds it to the list of virtual devices.
1311 * Returns a pointer to the created vio_dev or NULL if node has
1312 * NULL device_type or compatible fields.
1313 */
1314struct vio_dev *vio_register_device_node(struct device_node *of_node)
1315{
1316 struct vio_dev *viodev;
1317 struct device_node *parent_node;
1318 const unsigned int *unit_address;
1319 const unsigned int *pfo_resid = NULL;
1320 enum vio_dev_family family;
1321 const char *of_node_name = of_node->name ? of_node->name : "<unknown>";
1322
1323 /*
1324 * Determine if this node is a under the /vdevice node or under the
1325 * /ibm,platform-facilities node. This decides the device's family.
1326 */
1327 parent_node = of_get_parent(of_node);
1328 if (parent_node) {
1329 if (!strcmp(parent_node->full_name, "/ibm,platform-facilities"))
1330 family = PFO;
1331 else if (!strcmp(parent_node->full_name, "/vdevice"))
1332 family = VDEVICE;
1333 else {
1334 pr_warn("%s: parent(%s) of %s not recognized.\n",
1335 __func__,
1336 parent_node->full_name,
1337 of_node_name);
1338 of_node_put(parent_node);
1339 return NULL;
1340 }
1341 of_node_put(parent_node);
1342 } else {
1343 pr_warn("%s: could not determine the parent of node %s.\n",
1344 __func__, of_node_name);
1345 return NULL;
1346 }
1347
1348 if (family == PFO) {
1349 if (of_get_property(of_node, "interrupt-controller", NULL)) {
1350 pr_debug("%s: Skipping the interrupt controller %s.\n",
1351 __func__, of_node_name);
1352 return NULL;
1353 }
1354 }
1355
1356 /* allocate a vio_dev for this node */
1357 viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1358 if (viodev == NULL) {
1359 pr_warn("%s: allocation failure for VIO device.\n", __func__);
1360 return NULL;
1361 }
1362
1363 /* we need the 'device_type' property, in order to match with drivers */
1364 viodev->family = family;
1365 if (viodev->family == VDEVICE) {
1366 if (of_node->type != NULL)
1367 viodev->type = of_node->type;
1368 else {
1369 pr_warn("%s: node %s is missing the 'device_type' "
1370 "property.\n", __func__, of_node_name);
1371 goto out;
1372 }
1373
1374 unit_address = of_get_property(of_node, "reg", NULL);
1375 if (unit_address == NULL) {
1376 pr_warn("%s: node %s missing 'reg'\n",
1377 __func__, of_node_name);
1378 goto out;
1379 }
1380 dev_set_name(&viodev->dev, "%x", *unit_address);
1381 viodev->irq = irq_of_parse_and_map(of_node, 0);
1382 viodev->unit_address = *unit_address;
1383 } else {
1384 /* PFO devices need their resource_id for submitting COP_OPs
1385 * This is an optional field for devices, but is required when
1386 * performing synchronous ops */
1387 pfo_resid = of_get_property(of_node, "ibm,resource-id", NULL);
1388 if (pfo_resid != NULL)
1389 viodev->resource_id = *pfo_resid;
1390
1391 unit_address = NULL;
1392 dev_set_name(&viodev->dev, "%s", of_node_name);
1393 viodev->type = of_node_name;
1394 viodev->irq = 0;
1395 }
1396
1397 viodev->name = of_node->name;
1398 viodev->dev.of_node = of_node_get(of_node);
1399
1400 if (firmware_has_feature(FW_FEATURE_CMO))
1401 vio_cmo_set_dma_ops(viodev);
1402 else
1403 set_dma_ops(&viodev->dev, &dma_iommu_ops);
1404 set_iommu_table_base(&viodev->dev, vio_build_iommu_table(viodev));
1405 set_dev_node(&viodev->dev, of_node_to_nid(of_node));
1406
1407 /* init generic 'struct device' fields: */
1408 viodev->dev.parent = &vio_bus_device.dev;
1409 viodev->dev.bus = &vio_bus_type;
1410 viodev->dev.release = vio_dev_release;
1411 /* needed to ensure proper operation of coherent allocations
1412 * later, in case driver doesn't set it explicitly */
1413 dma_set_mask(&viodev->dev, DMA_BIT_MASK(64));
1414 dma_set_coherent_mask(&viodev->dev, DMA_BIT_MASK(64));
1415
1416 /* register with generic device framework */
1417 if (device_register(&viodev->dev)) {
1418 printk(KERN_ERR "%s: failed to register device %s\n",
1419 __func__, dev_name(&viodev->dev));
1420 put_device(&viodev->dev);
1421 return NULL;
1422 }
1423
1424 return viodev;
1425
1426out: /* Use this exit point for any return prior to device_register */
1427 kfree(viodev);
1428
1429 return NULL;
1430}
1431EXPORT_SYMBOL(vio_register_device_node);
1432
1433/*
1434 * vio_bus_scan_for_devices - Scan OF and register each child device
1435 * @root_name - OF node name for the root of the subtree to search.
1436 * This must be non-NULL
1437 *
1438 * Starting from the root node provide, register the device node for
1439 * each child beneath the root.
1440 */
1441static void vio_bus_scan_register_devices(char *root_name)
1442{
1443 struct device_node *node_root, *node_child;
1444
1445 if (!root_name)
1446 return;
1447
1448 node_root = of_find_node_by_name(NULL, root_name);
1449 if (node_root) {
1450
1451 /*
1452 * Create struct vio_devices for each virtual device in
1453 * the device tree. Drivers will associate with them later.
1454 */
1455 node_child = of_get_next_child(node_root, NULL);
1456 while (node_child) {
1457 vio_register_device_node(node_child);
1458 node_child = of_get_next_child(node_root, node_child);
1459 }
1460 of_node_put(node_root);
1461 }
1462}
1463
1464/**
1465 * vio_bus_init: - Initialize the virtual IO bus
1466 */
1467static int __init vio_bus_init(void)
1468{
1469 int err;
1470
1471 if (firmware_has_feature(FW_FEATURE_CMO))
1472 vio_cmo_sysfs_init();
1473
1474 err = bus_register(&vio_bus_type);
1475 if (err) {
1476 printk(KERN_ERR "failed to register VIO bus\n");
1477 return err;
1478 }
1479
1480 /*
1481 * The fake parent of all vio devices, just to give us
1482 * a nice directory
1483 */
1484 err = device_register(&vio_bus_device.dev);
1485 if (err) {
1486 printk(KERN_WARNING "%s: device_register returned %i\n",
1487 __func__, err);
1488 return err;
1489 }
1490
1491 if (firmware_has_feature(FW_FEATURE_CMO))
1492 vio_cmo_bus_init();
1493
1494 vio_bus_scan_register_devices("vdevice");
1495 vio_bus_scan_register_devices("ibm,platform-facilities");
1496
1497 return 0;
1498}
1499__initcall(vio_bus_init);
1500
1501static ssize_t name_show(struct device *dev,
1502 struct device_attribute *attr, char *buf)
1503{
1504 return sprintf(buf, "%s\n", to_vio_dev(dev)->name);
1505}
1506
1507static ssize_t devspec_show(struct device *dev,
1508 struct device_attribute *attr, char *buf)
1509{
1510 struct device_node *of_node = dev->of_node;
1511
1512 return sprintf(buf, "%s\n", of_node ? of_node->full_name : "none");
1513}
1514
1515static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1516 char *buf)
1517{
1518 const struct vio_dev *vio_dev = to_vio_dev(dev);
1519 struct device_node *dn;
1520 const char *cp;
1521
1522 dn = dev->of_node;
1523 if (!dn)
1524 return -ENODEV;
1525 cp = of_get_property(dn, "compatible", NULL);
1526 if (!cp)
1527 return -ENODEV;
1528
1529 return sprintf(buf, "vio:T%sS%s\n", vio_dev->type, cp);
1530}
1531
1532static struct device_attribute vio_dev_attrs[] = {
1533 __ATTR_RO(name),
1534 __ATTR_RO(devspec),
1535 __ATTR_RO(modalias),
1536 __ATTR_NULL
1537};
1538
1539void __devinit vio_unregister_device(struct vio_dev *viodev)
1540{
1541 device_unregister(&viodev->dev);
1542}
1543EXPORT_SYMBOL(vio_unregister_device);
1544
1545static int vio_bus_match(struct device *dev, struct device_driver *drv)
1546{
1547 const struct vio_dev *vio_dev = to_vio_dev(dev);
1548 struct vio_driver *vio_drv = to_vio_driver(drv);
1549 const struct vio_device_id *ids = vio_drv->id_table;
1550
1551 return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL);
1552}
1553
1554static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env)
1555{
1556 const struct vio_dev *vio_dev = to_vio_dev(dev);
1557 struct device_node *dn;
1558 const char *cp;
1559
1560 dn = dev->of_node;
1561 if (!dn)
1562 return -ENODEV;
1563 cp = of_get_property(dn, "compatible", NULL);
1564 if (!cp)
1565 return -ENODEV;
1566
1567 add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
1568 return 0;
1569}
1570
1571static struct bus_type vio_bus_type = {
1572 .name = "vio",
1573 .dev_attrs = vio_dev_attrs,
1574 .uevent = vio_hotplug,
1575 .match = vio_bus_match,
1576 .probe = vio_bus_probe,
1577 .remove = vio_bus_remove,
1578};
1579
1580/**
1581 * vio_get_attribute: - get attribute for virtual device
1582 * @vdev: The vio device to get property.
1583 * @which: The property/attribute to be extracted.
1584 * @length: Pointer to length of returned data size (unused if NULL).
1585 *
1586 * Calls prom.c's of_get_property() to return the value of the
1587 * attribute specified by @which
1588*/
1589const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1590{
1591 return of_get_property(vdev->dev.of_node, which, length);
1592}
1593EXPORT_SYMBOL(vio_get_attribute);
1594
1595#ifdef CONFIG_PPC_PSERIES
1596/* vio_find_name() - internal because only vio.c knows how we formatted the
1597 * kobject name
1598 */
1599static struct vio_dev *vio_find_name(const char *name)
1600{
1601 struct device *found;
1602
1603 found = bus_find_device_by_name(&vio_bus_type, NULL, name);
1604 if (!found)
1605 return NULL;
1606
1607 return to_vio_dev(found);
1608}
1609
1610/**
1611 * vio_find_node - find an already-registered vio_dev
1612 * @vnode: device_node of the virtual device we're looking for
1613 */
1614struct vio_dev *vio_find_node(struct device_node *vnode)
1615{
1616 const uint32_t *unit_address;
1617 char kobj_name[20];
1618 struct device_node *vnode_parent;
1619 const char *dev_type;
1620
1621 vnode_parent = of_get_parent(vnode);
1622 if (!vnode_parent)
1623 return NULL;
1624
1625 dev_type = of_get_property(vnode_parent, "device_type", NULL);
1626 of_node_put(vnode_parent);
1627 if (!dev_type)
1628 return NULL;
1629
1630 /* construct the kobject name from the device node */
1631 if (!strcmp(dev_type, "vdevice")) {
1632 unit_address = of_get_property(vnode, "reg", NULL);
1633 if (!unit_address)
1634 return NULL;
1635 snprintf(kobj_name, sizeof(kobj_name), "%x", *unit_address);
1636 } else if (!strcmp(dev_type, "ibm,platform-facilities"))
1637 snprintf(kobj_name, sizeof(kobj_name), "%s", vnode->name);
1638 else
1639 return NULL;
1640
1641 return vio_find_name(kobj_name);
1642}
1643EXPORT_SYMBOL(vio_find_node);
1644
1645int vio_enable_interrupts(struct vio_dev *dev)
1646{
1647 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1648 if (rc != H_SUCCESS)
1649 printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1650 return rc;
1651}
1652EXPORT_SYMBOL(vio_enable_interrupts);
1653
1654int vio_disable_interrupts(struct vio_dev *dev)
1655{
1656 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1657 if (rc != H_SUCCESS)
1658 printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1659 return rc;
1660}
1661EXPORT_SYMBOL(vio_disable_interrupts);
1662#endif /* CONFIG_PPC_PSERIES */