PM: Drop generic_subsys_pm_ops
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / powerpc / kernel / vio.c
CommitLineData
1da177e4
LT
1/*
2 * IBM PowerPC Virtual I/O Infrastructure Support.
3 *
a90ab95a 4 * Copyright (c) 2003,2008 IBM Corp.
1da177e4
LT
5 * Dave Engebretsen engebret@us.ibm.com
6 * Santiago Leon santil@us.ibm.com
7 * Hollis Blanchard <hollisb@us.ibm.com>
19dbd0f6 8 * Stephen Rothwell
a90ab95a 9 * Robert Jennings <rcjenn@us.ibm.com>
1da177e4
LT
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
c7f0e8cb 17#include <linux/types.h>
b56eade5 18#include <linux/stat.h>
c7f0e8cb 19#include <linux/device.h>
1da177e4 20#include <linux/init.h>
5a0e3ad6 21#include <linux/slab.h>
1da177e4 22#include <linux/console.h>
4b16f8e2 23#include <linux/export.h>
1da177e4
LT
24#include <linux/mm.h>
25#include <linux/dma-mapping.h>
c7f0e8cb
SR
26#include <linux/kobject.h>
27
1da177e4
LT
28#include <asm/iommu.h>
29#include <asm/dma.h>
1da177e4 30#include <asm/vio.h>
143dcec2 31#include <asm/prom.h>
e10fa773 32#include <asm/firmware.h>
c7f0e8cb
SR
33#include <asm/tce.h>
34#include <asm/abs_addr.h>
35#include <asm/page.h>
36#include <asm/hvcall.h>
37#include <asm/iseries/vio.h>
38#include <asm/iseries/hv_types.h>
39#include <asm/iseries/hv_lp_config.h>
40#include <asm/iseries/hv_call_xm.h>
41#include <asm/iseries/iommu.h>
42
6fccab26
SR
43static struct bus_type vio_bus_type;
44
c7f0e8cb 45static struct vio_dev vio_bus_device = { /* fake "parent" device */
aab0d375 46 .name = "vio",
ac5b33c9 47 .type = "",
aab0d375 48 .dev.init_name = "vio",
ac5b33c9 49 .dev.bus = &vio_bus_type,
1da177e4 50};
ac5b33c9 51
a90ab95a
RJ
52#ifdef CONFIG_PPC_SMLPAR
53/**
54 * vio_cmo_pool - A pool of IO memory for CMO use
55 *
56 * @size: The size of the pool in bytes
57 * @free: The amount of free memory in the pool
58 */
59struct vio_cmo_pool {
60 size_t size;
61 size_t free;
62};
63
64/* How many ms to delay queued balance work */
65#define VIO_CMO_BALANCE_DELAY 100
66
67/* Portion out IO memory to CMO devices by this chunk size */
68#define VIO_CMO_BALANCE_CHUNK 131072
69
70/**
71 * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement
72 *
73 * @vio_dev: struct vio_dev pointer
74 * @list: pointer to other devices on bus that are being tracked
75 */
76struct vio_cmo_dev_entry {
77 struct vio_dev *viodev;
78 struct list_head list;
79};
80
81/**
82 * vio_cmo - VIO bus accounting structure for CMO entitlement
83 *
84 * @lock: spinlock for entire structure
85 * @balance_q: work queue for balancing system entitlement
86 * @device_list: list of CMO-enabled devices requiring entitlement
87 * @entitled: total system entitlement in bytes
88 * @reserve: pool of memory from which devices reserve entitlement, incl. spare
89 * @excess: pool of excess entitlement not needed for device reserves or spare
90 * @spare: IO memory for device hotplug functionality
91 * @min: minimum necessary for system operation
92 * @desired: desired memory for system operation
93 * @curr: bytes currently allocated
94 * @high: high water mark for IO data usage
95 */
96struct vio_cmo {
97 spinlock_t lock;
98 struct delayed_work balance_q;
99 struct list_head device_list;
100 size_t entitled;
101 struct vio_cmo_pool reserve;
102 struct vio_cmo_pool excess;
103 size_t spare;
104 size_t min;
105 size_t desired;
106 size_t curr;
107 size_t high;
108} vio_cmo;
109
110/**
111 * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows
112 */
113static int vio_cmo_num_OF_devs(void)
114{
115 struct device_node *node_vroot;
116 int count = 0;
117
118 /*
119 * Count the number of vdevice entries with an
120 * ibm,my-dma-window OF property
121 */
122 node_vroot = of_find_node_by_name(NULL, "vdevice");
123 if (node_vroot) {
124 struct device_node *of_node;
125 struct property *prop;
126
127 for_each_child_of_node(node_vroot, of_node) {
128 prop = of_find_property(of_node, "ibm,my-dma-window",
129 NULL);
130 if (prop)
131 count++;
132 }
133 }
134 of_node_put(node_vroot);
135 return count;
136}
137
138/**
139 * vio_cmo_alloc - allocate IO memory for CMO-enable devices
140 *
141 * @viodev: VIO device requesting IO memory
142 * @size: size of allocation requested
143 *
144 * Allocations come from memory reserved for the devices and any excess
145 * IO memory available to all devices. The spare pool used to service
146 * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be
147 * made available.
148 *
149 * Return codes:
150 * 0 for successful allocation and -ENOMEM for a failure
151 */
152static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size)
153{
154 unsigned long flags;
155 size_t reserve_free = 0;
156 size_t excess_free = 0;
157 int ret = -ENOMEM;
158
159 spin_lock_irqsave(&vio_cmo.lock, flags);
160
161 /* Determine the amount of free entitlement available in reserve */
162 if (viodev->cmo.entitled > viodev->cmo.allocated)
163 reserve_free = viodev->cmo.entitled - viodev->cmo.allocated;
164
165 /* If spare is not fulfilled, the excess pool can not be used. */
166 if (vio_cmo.spare >= VIO_CMO_MIN_ENT)
167 excess_free = vio_cmo.excess.free;
168
169 /* The request can be satisfied */
170 if ((reserve_free + excess_free) >= size) {
171 vio_cmo.curr += size;
172 if (vio_cmo.curr > vio_cmo.high)
173 vio_cmo.high = vio_cmo.curr;
174 viodev->cmo.allocated += size;
175 size -= min(reserve_free, size);
176 vio_cmo.excess.free -= size;
177 ret = 0;
178 }
179
180 spin_unlock_irqrestore(&vio_cmo.lock, flags);
181 return ret;
182}
183
184/**
185 * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices
186 * @viodev: VIO device freeing IO memory
187 * @size: size of deallocation
188 *
189 * IO memory is freed by the device back to the correct memory pools.
190 * The spare pool is replenished first from either memory pool, then
191 * the reserve pool is used to reduce device entitlement, the excess
192 * pool is used to increase the reserve pool toward the desired entitlement
193 * target, and then the remaining memory is returned to the pools.
194 *
195 */
196static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size)
197{
198 unsigned long flags;
199 size_t spare_needed = 0;
200 size_t excess_freed = 0;
201 size_t reserve_freed = size;
202 size_t tmp;
203 int balance = 0;
204
205 spin_lock_irqsave(&vio_cmo.lock, flags);
206 vio_cmo.curr -= size;
207
208 /* Amount of memory freed from the excess pool */
209 if (viodev->cmo.allocated > viodev->cmo.entitled) {
210 excess_freed = min(reserve_freed, (viodev->cmo.allocated -
211 viodev->cmo.entitled));
212 reserve_freed -= excess_freed;
213 }
214
215 /* Remove allocation from device */
216 viodev->cmo.allocated -= (reserve_freed + excess_freed);
217
218 /* Spare is a subset of the reserve pool, replenish it first. */
219 spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare;
220
221 /*
222 * Replenish the spare in the reserve pool from the excess pool.
223 * This moves entitlement into the reserve pool.
224 */
225 if (spare_needed && excess_freed) {
226 tmp = min(excess_freed, spare_needed);
227 vio_cmo.excess.size -= tmp;
228 vio_cmo.reserve.size += tmp;
229 vio_cmo.spare += tmp;
230 excess_freed -= tmp;
231 spare_needed -= tmp;
232 balance = 1;
233 }
234
235 /*
236 * Replenish the spare in the reserve pool from the reserve pool.
237 * This removes entitlement from the device down to VIO_CMO_MIN_ENT,
238 * if needed, and gives it to the spare pool. The amount of used
239 * memory in this pool does not change.
240 */
241 if (spare_needed && reserve_freed) {
732eacc0 242 tmp = min3(spare_needed, reserve_freed, (viodev->cmo.entitled - VIO_CMO_MIN_ENT));
a90ab95a
RJ
243
244 vio_cmo.spare += tmp;
245 viodev->cmo.entitled -= tmp;
246 reserve_freed -= tmp;
247 spare_needed -= tmp;
248 balance = 1;
249 }
250
251 /*
252 * Increase the reserve pool until the desired allocation is met.
253 * Move an allocation freed from the excess pool into the reserve
254 * pool and schedule a balance operation.
255 */
256 if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) {
257 tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size));
258
259 vio_cmo.excess.size -= tmp;
260 vio_cmo.reserve.size += tmp;
261 excess_freed -= tmp;
262 balance = 1;
263 }
264
265 /* Return memory from the excess pool to that pool */
266 if (excess_freed)
267 vio_cmo.excess.free += excess_freed;
268
269 if (balance)
270 schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY);
271 spin_unlock_irqrestore(&vio_cmo.lock, flags);
272}
273
274/**
275 * vio_cmo_entitlement_update - Manage system entitlement changes
276 *
277 * @new_entitlement: new system entitlement to attempt to accommodate
278 *
279 * Increases in entitlement will be used to fulfill the spare entitlement
280 * and the rest is given to the excess pool. Decreases, if they are
281 * possible, come from the excess pool and from unused device entitlement
282 *
283 * Returns: 0 on success, -ENOMEM when change can not be made
284 */
285int vio_cmo_entitlement_update(size_t new_entitlement)
286{
287 struct vio_dev *viodev;
288 struct vio_cmo_dev_entry *dev_ent;
289 unsigned long flags;
290 size_t avail, delta, tmp;
291
292 spin_lock_irqsave(&vio_cmo.lock, flags);
293
294 /* Entitlement increases */
295 if (new_entitlement > vio_cmo.entitled) {
296 delta = new_entitlement - vio_cmo.entitled;
297
298 /* Fulfill spare allocation */
299 if (vio_cmo.spare < VIO_CMO_MIN_ENT) {
300 tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare));
301 vio_cmo.spare += tmp;
302 vio_cmo.reserve.size += tmp;
303 delta -= tmp;
304 }
305
306 /* Remaining new allocation goes to the excess pool */
307 vio_cmo.entitled += delta;
308 vio_cmo.excess.size += delta;
309 vio_cmo.excess.free += delta;
310
311 goto out;
312 }
313
314 /* Entitlement decreases */
315 delta = vio_cmo.entitled - new_entitlement;
316 avail = vio_cmo.excess.free;
317
318 /*
319 * Need to check how much unused entitlement each device can
320 * sacrifice to fulfill entitlement change.
321 */
322 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
323 if (avail >= delta)
324 break;
325
326 viodev = dev_ent->viodev;
327 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
328 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
329 avail += viodev->cmo.entitled -
330 max_t(size_t, viodev->cmo.allocated,
331 VIO_CMO_MIN_ENT);
332 }
333
334 if (delta <= avail) {
335 vio_cmo.entitled -= delta;
336
337 /* Take entitlement from the excess pool first */
338 tmp = min(vio_cmo.excess.free, delta);
339 vio_cmo.excess.size -= tmp;
340 vio_cmo.excess.free -= tmp;
341 delta -= tmp;
342
343 /*
344 * Remove all but VIO_CMO_MIN_ENT bytes from devices
345 * until entitlement change is served
346 */
347 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
348 if (!delta)
349 break;
350
351 viodev = dev_ent->viodev;
352 tmp = 0;
353 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
354 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
355 tmp = viodev->cmo.entitled -
356 max_t(size_t, viodev->cmo.allocated,
357 VIO_CMO_MIN_ENT);
358 viodev->cmo.entitled -= min(tmp, delta);
359 delta -= min(tmp, delta);
360 }
361 } else {
362 spin_unlock_irqrestore(&vio_cmo.lock, flags);
363 return -ENOMEM;
364 }
365
366out:
367 schedule_delayed_work(&vio_cmo.balance_q, 0);
368 spin_unlock_irqrestore(&vio_cmo.lock, flags);
369 return 0;
370}
371
372/**
373 * vio_cmo_balance - Balance entitlement among devices
374 *
375 * @work: work queue structure for this operation
376 *
377 * Any system entitlement above the minimum needed for devices, or
378 * already allocated to devices, can be distributed to the devices.
379 * The list of devices is iterated through to recalculate the desired
380 * entitlement level and to determine how much entitlement above the
381 * minimum entitlement is allocated to devices.
382 *
383 * Small chunks of the available entitlement are given to devices until
384 * their requirements are fulfilled or there is no entitlement left to give.
385 * Upon completion sizes of the reserve and excess pools are calculated.
386 *
387 * The system minimum entitlement level is also recalculated here.
388 * Entitlement will be reserved for devices even after vio_bus_remove to
389 * accommodate reloading the driver. The OF tree is walked to count the
390 * number of devices present and this will remove entitlement for devices
391 * that have actually left the system after having vio_bus_remove called.
392 */
393static void vio_cmo_balance(struct work_struct *work)
394{
395 struct vio_cmo *cmo;
396 struct vio_dev *viodev;
397 struct vio_cmo_dev_entry *dev_ent;
398 unsigned long flags;
399 size_t avail = 0, level, chunk, need;
400 int devcount = 0, fulfilled;
401
402 cmo = container_of(work, struct vio_cmo, balance_q.work);
403
404 spin_lock_irqsave(&vio_cmo.lock, flags);
405
406 /* Calculate minimum entitlement and fulfill spare */
407 cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT;
408 BUG_ON(cmo->min > cmo->entitled);
409 cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min));
410 cmo->min += cmo->spare;
411 cmo->desired = cmo->min;
412
413 /*
414 * Determine how much entitlement is available and reset device
415 * entitlements
416 */
417 avail = cmo->entitled - cmo->spare;
418 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
419 viodev = dev_ent->viodev;
420 devcount++;
421 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
422 cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT);
423 avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT);
424 }
425
426 /*
427 * Having provided each device with the minimum entitlement, loop
428 * over the devices portioning out the remaining entitlement
429 * until there is nothing left.
430 */
431 level = VIO_CMO_MIN_ENT;
432 while (avail) {
433 fulfilled = 0;
434 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
435 viodev = dev_ent->viodev;
436
437 if (viodev->cmo.desired <= level) {
438 fulfilled++;
439 continue;
440 }
441
442 /*
443 * Give the device up to VIO_CMO_BALANCE_CHUNK
444 * bytes of entitlement, but do not exceed the
445 * desired level of entitlement for the device.
446 */
447 chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK);
448 chunk = min(chunk, (viodev->cmo.desired -
449 viodev->cmo.entitled));
450 viodev->cmo.entitled += chunk;
451
452 /*
453 * If the memory for this entitlement increase was
454 * already allocated to the device it does not come
455 * from the available pool being portioned out.
456 */
457 need = max(viodev->cmo.allocated, viodev->cmo.entitled)-
458 max(viodev->cmo.allocated, level);
459 avail -= need;
460
461 }
462 if (fulfilled == devcount)
463 break;
464 level += VIO_CMO_BALANCE_CHUNK;
465 }
466
467 /* Calculate new reserve and excess pool sizes */
468 cmo->reserve.size = cmo->min;
469 cmo->excess.free = 0;
470 cmo->excess.size = 0;
471 need = 0;
472 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
473 viodev = dev_ent->viodev;
474 /* Calculated reserve size above the minimum entitlement */
475 if (viodev->cmo.entitled)
476 cmo->reserve.size += (viodev->cmo.entitled -
477 VIO_CMO_MIN_ENT);
478 /* Calculated used excess entitlement */
479 if (viodev->cmo.allocated > viodev->cmo.entitled)
480 need += viodev->cmo.allocated - viodev->cmo.entitled;
481 }
482 cmo->excess.size = cmo->entitled - cmo->reserve.size;
483 cmo->excess.free = cmo->excess.size - need;
484
bf6aede7 485 cancel_delayed_work(to_delayed_work(work));
a90ab95a
RJ
486 spin_unlock_irqrestore(&vio_cmo.lock, flags);
487}
488
489static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size,
490 dma_addr_t *dma_handle, gfp_t flag)
491{
492 struct vio_dev *viodev = to_vio_dev(dev);
493 void *ret;
494
69b052e8 495 if (vio_cmo_alloc(viodev, roundup(size, PAGE_SIZE))) {
a90ab95a
RJ
496 atomic_inc(&viodev->cmo.allocs_failed);
497 return NULL;
498 }
499
500 ret = dma_iommu_ops.alloc_coherent(dev, size, dma_handle, flag);
501 if (unlikely(ret == NULL)) {
69b052e8 502 vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
a90ab95a
RJ
503 atomic_inc(&viodev->cmo.allocs_failed);
504 }
505
506 return ret;
507}
508
509static void vio_dma_iommu_free_coherent(struct device *dev, size_t size,
510 void *vaddr, dma_addr_t dma_handle)
511{
512 struct vio_dev *viodev = to_vio_dev(dev);
513
514 dma_iommu_ops.free_coherent(dev, size, vaddr, dma_handle);
515
69b052e8 516 vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
a90ab95a
RJ
517}
518
f9226d57
MN
519static dma_addr_t vio_dma_iommu_map_page(struct device *dev, struct page *page,
520 unsigned long offset, size_t size,
521 enum dma_data_direction direction,
522 struct dma_attrs *attrs)
a90ab95a
RJ
523{
524 struct vio_dev *viodev = to_vio_dev(dev);
525 dma_addr_t ret = DMA_ERROR_CODE;
526
527 if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE))) {
528 atomic_inc(&viodev->cmo.allocs_failed);
529 return ret;
530 }
531
f9226d57 532 ret = dma_iommu_ops.map_page(dev, page, offset, size, direction, attrs);
0764bf63 533 if (unlikely(dma_mapping_error(dev, ret))) {
a90ab95a
RJ
534 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
535 atomic_inc(&viodev->cmo.allocs_failed);
536 }
537
538 return ret;
539}
540
f9226d57
MN
541static void vio_dma_iommu_unmap_page(struct device *dev, dma_addr_t dma_handle,
542 size_t size,
543 enum dma_data_direction direction,
544 struct dma_attrs *attrs)
a90ab95a
RJ
545{
546 struct vio_dev *viodev = to_vio_dev(dev);
547
f9226d57 548 dma_iommu_ops.unmap_page(dev, dma_handle, size, direction, attrs);
a90ab95a
RJ
549
550 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE));
551}
552
553static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
554 int nelems, enum dma_data_direction direction,
555 struct dma_attrs *attrs)
556{
557 struct vio_dev *viodev = to_vio_dev(dev);
558 struct scatterlist *sgl;
559 int ret, count = 0;
560 size_t alloc_size = 0;
561
562 for (sgl = sglist; count < nelems; count++, sgl++)
563 alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE);
564
565 if (vio_cmo_alloc(viodev, alloc_size)) {
566 atomic_inc(&viodev->cmo.allocs_failed);
567 return 0;
568 }
569
570 ret = dma_iommu_ops.map_sg(dev, sglist, nelems, direction, attrs);
571
572 if (unlikely(!ret)) {
573 vio_cmo_dealloc(viodev, alloc_size);
574 atomic_inc(&viodev->cmo.allocs_failed);
69b052e8 575 return ret;
a90ab95a
RJ
576 }
577
578 for (sgl = sglist, count = 0; count < ret; count++, sgl++)
579 alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
580 if (alloc_size)
581 vio_cmo_dealloc(viodev, alloc_size);
582
583 return ret;
584}
585
586static void vio_dma_iommu_unmap_sg(struct device *dev,
587 struct scatterlist *sglist, int nelems,
588 enum dma_data_direction direction,
589 struct dma_attrs *attrs)
590{
591 struct vio_dev *viodev = to_vio_dev(dev);
592 struct scatterlist *sgl;
593 size_t alloc_size = 0;
594 int count = 0;
595
596 for (sgl = sglist; count < nelems; count++, sgl++)
597 alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE);
598
599 dma_iommu_ops.unmap_sg(dev, sglist, nelems, direction, attrs);
600
601 vio_cmo_dealloc(viodev, alloc_size);
602}
603
6d283d78
NA
604static int vio_dma_iommu_dma_supported(struct device *dev, u64 mask)
605{
606 return dma_iommu_ops.dma_supported(dev, mask);
607}
608
d24f9c69
MM
609static u64 vio_dma_get_required_mask(struct device *dev)
610{
611 return dma_iommu_ops.get_required_mask(dev);
612}
613
45223c54 614struct dma_map_ops vio_dma_mapping_ops = {
2eccacd0
MM
615 .alloc_coherent = vio_dma_iommu_alloc_coherent,
616 .free_coherent = vio_dma_iommu_free_coherent,
617 .map_sg = vio_dma_iommu_map_sg,
618 .unmap_sg = vio_dma_iommu_unmap_sg,
619 .map_page = vio_dma_iommu_map_page,
620 .unmap_page = vio_dma_iommu_unmap_page,
621 .dma_supported = vio_dma_iommu_dma_supported,
d24f9c69 622 .get_required_mask = vio_dma_get_required_mask,
a90ab95a
RJ
623};
624
625/**
626 * vio_cmo_set_dev_desired - Set desired entitlement for a device
627 *
628 * @viodev: struct vio_dev for device to alter
629 * @new_desired: new desired entitlement level in bytes
630 *
631 * For use by devices to request a change to their entitlement at runtime or
632 * through sysfs. The desired entitlement level is changed and a balancing
633 * of system resources is scheduled to run in the future.
634 */
635void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
636{
637 unsigned long flags;
638 struct vio_cmo_dev_entry *dev_ent;
639 int found = 0;
640
641 if (!firmware_has_feature(FW_FEATURE_CMO))
642 return;
643
644 spin_lock_irqsave(&vio_cmo.lock, flags);
645 if (desired < VIO_CMO_MIN_ENT)
646 desired = VIO_CMO_MIN_ENT;
647
648 /*
649 * Changes will not be made for devices not in the device list.
650 * If it is not in the device list, then no driver is loaded
651 * for the device and it can not receive entitlement.
652 */
653 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
654 if (viodev == dev_ent->viodev) {
655 found = 1;
656 break;
657 }
f6d8c8bb
JL
658 if (!found) {
659 spin_unlock_irqrestore(&vio_cmo.lock, flags);
a90ab95a 660 return;
f6d8c8bb 661 }
a90ab95a
RJ
662
663 /* Increase/decrease in desired device entitlement */
664 if (desired >= viodev->cmo.desired) {
665 /* Just bump the bus and device values prior to a balance*/
666 vio_cmo.desired += desired - viodev->cmo.desired;
667 viodev->cmo.desired = desired;
668 } else {
669 /* Decrease bus and device values for desired entitlement */
670 vio_cmo.desired -= viodev->cmo.desired - desired;
671 viodev->cmo.desired = desired;
672 /*
673 * If less entitlement is desired than current entitlement, move
674 * any reserve memory in the change region to the excess pool.
675 */
676 if (viodev->cmo.entitled > desired) {
677 vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
678 vio_cmo.excess.size += viodev->cmo.entitled - desired;
679 /*
680 * If entitlement moving from the reserve pool to the
681 * excess pool is currently unused, add to the excess
682 * free counter.
683 */
684 if (viodev->cmo.allocated < viodev->cmo.entitled)
685 vio_cmo.excess.free += viodev->cmo.entitled -
686 max(viodev->cmo.allocated, desired);
687 viodev->cmo.entitled = desired;
688 }
689 }
690 schedule_delayed_work(&vio_cmo.balance_q, 0);
691 spin_unlock_irqrestore(&vio_cmo.lock, flags);
692}
693
694/**
695 * vio_cmo_bus_probe - Handle CMO specific bus probe activities
696 *
697 * @viodev - Pointer to struct vio_dev for device
698 *
699 * Determine the devices IO memory entitlement needs, attempting
700 * to satisfy the system minimum entitlement at first and scheduling
701 * a balance operation to take care of the rest at a later time.
702 *
703 * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
704 * -ENOMEM when entitlement is not available for device or
705 * device entry.
706 *
707 */
708static int vio_cmo_bus_probe(struct vio_dev *viodev)
709{
710 struct vio_cmo_dev_entry *dev_ent;
711 struct device *dev = &viodev->dev;
712 struct vio_driver *viodrv = to_vio_driver(dev->driver);
713 unsigned long flags;
714 size_t size;
715
716 /*
717 * Check to see that device has a DMA window and configure
718 * entitlement for the device.
719 */
58f9b0b0 720 if (of_get_property(viodev->dev.of_node,
a90ab95a
RJ
721 "ibm,my-dma-window", NULL)) {
722 /* Check that the driver is CMO enabled and get desired DMA */
723 if (!viodrv->get_desired_dma) {
724 dev_err(dev, "%s: device driver does not support CMO\n",
725 __func__);
726 return -EINVAL;
727 }
728
729 viodev->cmo.desired = IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev));
730 if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
731 viodev->cmo.desired = VIO_CMO_MIN_ENT;
732 size = VIO_CMO_MIN_ENT;
733
734 dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
735 GFP_KERNEL);
736 if (!dev_ent)
737 return -ENOMEM;
738
739 dev_ent->viodev = viodev;
740 spin_lock_irqsave(&vio_cmo.lock, flags);
741 list_add(&dev_ent->list, &vio_cmo.device_list);
742 } else {
743 viodev->cmo.desired = 0;
744 size = 0;
745 spin_lock_irqsave(&vio_cmo.lock, flags);
746 }
747
748 /*
749 * If the needs for vio_cmo.min have not changed since they
750 * were last set, the number of devices in the OF tree has
751 * been constant and the IO memory for this is already in
752 * the reserve pool.
753 */
754 if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
755 VIO_CMO_MIN_ENT)) {
756 /* Updated desired entitlement if device requires it */
757 if (size)
758 vio_cmo.desired += (viodev->cmo.desired -
759 VIO_CMO_MIN_ENT);
760 } else {
761 size_t tmp;
762
763 tmp = vio_cmo.spare + vio_cmo.excess.free;
764 if (tmp < size) {
765 dev_err(dev, "%s: insufficient free "
766 "entitlement to add device. "
767 "Need %lu, have %lu\n", __func__,
768 size, (vio_cmo.spare + tmp));
769 spin_unlock_irqrestore(&vio_cmo.lock, flags);
770 return -ENOMEM;
771 }
772
773 /* Use excess pool first to fulfill request */
774 tmp = min(size, vio_cmo.excess.free);
775 vio_cmo.excess.free -= tmp;
776 vio_cmo.excess.size -= tmp;
777 vio_cmo.reserve.size += tmp;
778
779 /* Use spare if excess pool was insufficient */
780 vio_cmo.spare -= size - tmp;
781
782 /* Update bus accounting */
783 vio_cmo.min += size;
784 vio_cmo.desired += viodev->cmo.desired;
785 }
786 spin_unlock_irqrestore(&vio_cmo.lock, flags);
787 return 0;
788}
789
790/**
791 * vio_cmo_bus_remove - Handle CMO specific bus removal activities
792 *
793 * @viodev - Pointer to struct vio_dev for device
794 *
795 * Remove the device from the cmo device list. The minimum entitlement
796 * will be reserved for the device as long as it is in the system. The
797 * rest of the entitlement the device had been allocated will be returned
798 * to the system.
799 */
800static void vio_cmo_bus_remove(struct vio_dev *viodev)
801{
802 struct vio_cmo_dev_entry *dev_ent;
803 unsigned long flags;
804 size_t tmp;
805
806 spin_lock_irqsave(&vio_cmo.lock, flags);
807 if (viodev->cmo.allocated) {
808 dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
809 "allocated after remove operation.\n",
810 __func__, viodev->cmo.allocated);
811 BUG();
812 }
813
814 /*
815 * Remove the device from the device list being maintained for
816 * CMO enabled devices.
817 */
818 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
819 if (viodev == dev_ent->viodev) {
820 list_del(&dev_ent->list);
821 kfree(dev_ent);
822 break;
823 }
824
825 /*
826 * Devices may not require any entitlement and they do not need
827 * to be processed. Otherwise, return the device's entitlement
828 * back to the pools.
829 */
830 if (viodev->cmo.entitled) {
831 /*
832 * This device has not yet left the OF tree, it's
833 * minimum entitlement remains in vio_cmo.min and
834 * vio_cmo.desired
835 */
836 vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
837
838 /*
839 * Save min allocation for device in reserve as long
840 * as it exists in OF tree as determined by later
841 * balance operation
842 */
843 viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
844
845 /* Replenish spare from freed reserve pool */
846 if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
847 tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
848 vio_cmo.spare));
849 vio_cmo.spare += tmp;
850 viodev->cmo.entitled -= tmp;
851 }
852
853 /* Remaining reserve goes to excess pool */
854 vio_cmo.excess.size += viodev->cmo.entitled;
855 vio_cmo.excess.free += viodev->cmo.entitled;
856 vio_cmo.reserve.size -= viodev->cmo.entitled;
857
858 /*
859 * Until the device is removed it will keep a
860 * minimum entitlement; this will guarantee that
861 * a module unload/load will result in a success.
862 */
863 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
864 viodev->cmo.desired = VIO_CMO_MIN_ENT;
865 atomic_set(&viodev->cmo.allocs_failed, 0);
866 }
867
868 spin_unlock_irqrestore(&vio_cmo.lock, flags);
869}
870
871static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
872{
6d283d78 873 set_dma_ops(&viodev->dev, &vio_dma_mapping_ops);
a90ab95a
RJ
874}
875
876/**
877 * vio_cmo_bus_init - CMO entitlement initialization at bus init time
878 *
879 * Set up the reserve and excess entitlement pools based on available
880 * system entitlement and the number of devices in the OF tree that
881 * require entitlement in the reserve pool.
882 */
883static void vio_cmo_bus_init(void)
884{
885 struct hvcall_mpp_data mpp_data;
886 int err;
887
888 memset(&vio_cmo, 0, sizeof(struct vio_cmo));
889 spin_lock_init(&vio_cmo.lock);
890 INIT_LIST_HEAD(&vio_cmo.device_list);
891 INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
892
893 /* Get current system entitlement */
894 err = h_get_mpp(&mpp_data);
895
896 /*
897 * On failure, continue with entitlement set to 0, will panic()
898 * later when spare is reserved.
899 */
900 if (err != H_SUCCESS) {
901 printk(KERN_ERR "%s: unable to determine system IO "\
902 "entitlement. (%d)\n", __func__, err);
903 vio_cmo.entitled = 0;
904 } else {
905 vio_cmo.entitled = mpp_data.entitled_mem;
906 }
907
908 /* Set reservation and check against entitlement */
909 vio_cmo.spare = VIO_CMO_MIN_ENT;
910 vio_cmo.reserve.size = vio_cmo.spare;
911 vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
912 VIO_CMO_MIN_ENT);
913 if (vio_cmo.reserve.size > vio_cmo.entitled) {
914 printk(KERN_ERR "%s: insufficient system entitlement\n",
915 __func__);
916 panic("%s: Insufficient system entitlement", __func__);
917 }
918
919 /* Set the remaining accounting variables */
920 vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
921 vio_cmo.excess.free = vio_cmo.excess.size;
922 vio_cmo.min = vio_cmo.reserve.size;
923 vio_cmo.desired = vio_cmo.reserve.size;
924}
925
926/* sysfs device functions and data structures for CMO */
927
928#define viodev_cmo_rd_attr(name) \
929static ssize_t viodev_cmo_##name##_show(struct device *dev, \
930 struct device_attribute *attr, \
931 char *buf) \
932{ \
933 return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \
934}
935
936static ssize_t viodev_cmo_allocs_failed_show(struct device *dev,
937 struct device_attribute *attr, char *buf)
938{
939 struct vio_dev *viodev = to_vio_dev(dev);
940 return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
941}
942
943static ssize_t viodev_cmo_allocs_failed_reset(struct device *dev,
944 struct device_attribute *attr, const char *buf, size_t count)
945{
946 struct vio_dev *viodev = to_vio_dev(dev);
947 atomic_set(&viodev->cmo.allocs_failed, 0);
948 return count;
949}
950
951static ssize_t viodev_cmo_desired_set(struct device *dev,
952 struct device_attribute *attr, const char *buf, size_t count)
953{
954 struct vio_dev *viodev = to_vio_dev(dev);
955 size_t new_desired;
956 int ret;
957
958 ret = strict_strtoul(buf, 10, &new_desired);
959 if (ret)
960 return ret;
961
962 vio_cmo_set_dev_desired(viodev, new_desired);
963 return count;
964}
965
966viodev_cmo_rd_attr(desired);
967viodev_cmo_rd_attr(entitled);
968viodev_cmo_rd_attr(allocated);
969
970static ssize_t name_show(struct device *, struct device_attribute *, char *);
971static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
578b7cd1
BH
972static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
973 char *buf);
a90ab95a
RJ
974static struct device_attribute vio_cmo_dev_attrs[] = {
975 __ATTR_RO(name),
976 __ATTR_RO(devspec),
578b7cd1 977 __ATTR_RO(modalias),
a90ab95a
RJ
978 __ATTR(cmo_desired, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
979 viodev_cmo_desired_show, viodev_cmo_desired_set),
980 __ATTR(cmo_entitled, S_IRUGO, viodev_cmo_entitled_show, NULL),
981 __ATTR(cmo_allocated, S_IRUGO, viodev_cmo_allocated_show, NULL),
982 __ATTR(cmo_allocs_failed, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
983 viodev_cmo_allocs_failed_show, viodev_cmo_allocs_failed_reset),
984 __ATTR_NULL
985};
986
987/* sysfs bus functions and data structures for CMO */
988
989#define viobus_cmo_rd_attr(name) \
990static ssize_t \
991viobus_cmo_##name##_show(struct bus_type *bt, char *buf) \
992{ \
993 return sprintf(buf, "%lu\n", vio_cmo.name); \
994}
995
996#define viobus_cmo_pool_rd_attr(name, var) \
997static ssize_t \
998viobus_cmo_##name##_pool_show_##var(struct bus_type *bt, char *buf) \
999{ \
1000 return sprintf(buf, "%lu\n", vio_cmo.name.var); \
1001}
1002
1003static ssize_t viobus_cmo_high_reset(struct bus_type *bt, const char *buf,
1004 size_t count)
1005{
1006 unsigned long flags;
1007
1008 spin_lock_irqsave(&vio_cmo.lock, flags);
1009 vio_cmo.high = vio_cmo.curr;
1010 spin_unlock_irqrestore(&vio_cmo.lock, flags);
1011
1012 return count;
1013}
1014
1015viobus_cmo_rd_attr(entitled);
1016viobus_cmo_pool_rd_attr(reserve, size);
1017viobus_cmo_pool_rd_attr(excess, size);
1018viobus_cmo_pool_rd_attr(excess, free);
1019viobus_cmo_rd_attr(spare);
1020viobus_cmo_rd_attr(min);
1021viobus_cmo_rd_attr(desired);
1022viobus_cmo_rd_attr(curr);
1023viobus_cmo_rd_attr(high);
1024
1025static struct bus_attribute vio_cmo_bus_attrs[] = {
1026 __ATTR(cmo_entitled, S_IRUGO, viobus_cmo_entitled_show, NULL),
1027 __ATTR(cmo_reserve_size, S_IRUGO, viobus_cmo_reserve_pool_show_size, NULL),
1028 __ATTR(cmo_excess_size, S_IRUGO, viobus_cmo_excess_pool_show_size, NULL),
1029 __ATTR(cmo_excess_free, S_IRUGO, viobus_cmo_excess_pool_show_free, NULL),
1030 __ATTR(cmo_spare, S_IRUGO, viobus_cmo_spare_show, NULL),
1031 __ATTR(cmo_min, S_IRUGO, viobus_cmo_min_show, NULL),
1032 __ATTR(cmo_desired, S_IRUGO, viobus_cmo_desired_show, NULL),
1033 __ATTR(cmo_curr, S_IRUGO, viobus_cmo_curr_show, NULL),
1034 __ATTR(cmo_high, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
1035 viobus_cmo_high_show, viobus_cmo_high_reset),
1036 __ATTR_NULL
1037};
1038
1039static void vio_cmo_sysfs_init(void)
1040{
1041 vio_bus_type.dev_attrs = vio_cmo_dev_attrs;
1042 vio_bus_type.bus_attrs = vio_cmo_bus_attrs;
1043}
1044#else /* CONFIG_PPC_SMLPAR */
1045/* Dummy functions for iSeries platform */
1046int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1047void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1048static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1049static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1050static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
b9fa49a9
NL
1051static void vio_cmo_bus_init(void) {}
1052static void vio_cmo_sysfs_init(void) { }
a90ab95a
RJ
1053#endif /* CONFIG_PPC_SMLPAR */
1054EXPORT_SYMBOL(vio_cmo_entitlement_update);
1055EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1056
c7f0e8cb
SR
1057static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1058{
dd9b67ab
SR
1059 const unsigned char *dma_window;
1060 struct iommu_table *tbl;
1061 unsigned long offset, size;
1062
1063 if (firmware_has_feature(FW_FEATURE_ISERIES))
1064 return vio_build_iommu_table_iseries(dev);
c7f0e8cb 1065
58f9b0b0 1066 dma_window = of_get_property(dev->dev.of_node,
dd9b67ab
SR
1067 "ibm,my-dma-window", NULL);
1068 if (!dma_window)
1069 return NULL;
c7f0e8cb 1070
7aa241fd 1071 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
0f337274 1072 if (tbl == NULL)
1073 return NULL;
c7f0e8cb 1074
58f9b0b0 1075 of_parse_dma_window(dev->dev.of_node, dma_window,
dd9b67ab 1076 &tbl->it_index, &offset, &size);
c7f0e8cb 1077
dd9b67ab
SR
1078 /* TCE table size - measured in tce entries */
1079 tbl->it_size = size >> IOMMU_PAGE_SHIFT;
1080 /* offset for VIO should always be 0 */
1081 tbl->it_offset = offset >> IOMMU_PAGE_SHIFT;
1082 tbl->it_busno = 0;
1083 tbl->it_type = TCE_VB;
7aa241fd 1084 tbl->it_blocksize = 16;
c7f0e8cb 1085
dd9b67ab 1086 return iommu_init_table(tbl, -1);
c7f0e8cb 1087}
1da177e4 1088
e10fa773
SR
1089/**
1090 * vio_match_device: - Tell if a VIO device has a matching
1091 * VIO device id structure.
1092 * @ids: array of VIO device id structures to search in
1093 * @dev: the VIO device structure to match against
1094 *
1095 * Used by a driver to check whether a VIO device present in the
1096 * system is in its list of supported devices. Returns the matching
1097 * vio_device_id structure or NULL if there is no match.
1098 */
1099static const struct vio_device_id *vio_match_device(
1100 const struct vio_device_id *ids, const struct vio_dev *dev)
1101{
1102 while (ids->type[0] != '\0') {
dd721ffd 1103 if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
58f9b0b0 1104 of_device_is_compatible(dev->dev.of_node,
12d04eef 1105 ids->compat))
e10fa773
SR
1106 return ids;
1107 ids++;
1108 }
1109 return NULL;
1110}
1111
5c0b4b87
SR
1112/*
1113 * Convert from struct device to struct vio_dev and pass to driver.
1da177e4 1114 * dev->driver has already been set by generic code because vio_bus_match
5c0b4b87
SR
1115 * succeeded.
1116 */
1da177e4
LT
1117static int vio_bus_probe(struct device *dev)
1118{
1119 struct vio_dev *viodev = to_vio_dev(dev);
1120 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1121 const struct vio_device_id *id;
1122 int error = -ENODEV;
1123
1da177e4
LT
1124 if (!viodrv->probe)
1125 return error;
1126
1127 id = vio_match_device(viodrv->id_table, viodev);
a90ab95a
RJ
1128 if (id) {
1129 memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1130 if (firmware_has_feature(FW_FEATURE_CMO)) {
1131 error = vio_cmo_bus_probe(viodev);
1132 if (error)
1133 return error;
1134 }
1da177e4 1135 error = viodrv->probe(viodev, id);
cd5aeb9f 1136 if (error && firmware_has_feature(FW_FEATURE_CMO))
a90ab95a
RJ
1137 vio_cmo_bus_remove(viodev);
1138 }
1da177e4
LT
1139
1140 return error;
1141}
1142
1143/* convert from struct device to struct vio_dev and pass to driver. */
1144static int vio_bus_remove(struct device *dev)
1145{
1146 struct vio_dev *viodev = to_vio_dev(dev);
1147 struct vio_driver *viodrv = to_vio_driver(dev->driver);
a90ab95a
RJ
1148 struct device *devptr;
1149 int ret = 1;
1150
1151 /*
1152 * Hold a reference to the device after the remove function is called
1153 * to allow for CMO accounting cleanup for the device.
1154 */
1155 devptr = get_device(dev);
1da177e4 1156
5c0b4b87 1157 if (viodrv->remove)
a90ab95a 1158 ret = viodrv->remove(viodev);
1da177e4 1159
a90ab95a
RJ
1160 if (!ret && firmware_has_feature(FW_FEATURE_CMO))
1161 vio_cmo_bus_remove(viodev);
1162
1163 put_device(devptr);
1164 return ret;
1da177e4
LT
1165}
1166
1167/**
1168 * vio_register_driver: - Register a new vio driver
1169 * @drv: The vio_driver structure to be registered.
1170 */
1171int vio_register_driver(struct vio_driver *viodrv)
1172{
e48b1b45 1173 printk(KERN_DEBUG "%s: driver %s registering\n", __func__,
6fdf5392 1174 viodrv->driver.name);
1da177e4
LT
1175
1176 /* fill in 'struct driver' fields */
1da177e4 1177 viodrv->driver.bus = &vio_bus_type;
1da177e4
LT
1178
1179 return driver_register(&viodrv->driver);
1180}
1181EXPORT_SYMBOL(vio_register_driver);
1182
1183/**
1184 * vio_unregister_driver - Remove registration of vio driver.
1185 * @driver: The vio_driver struct to be removed form registration
1186 */
1187void vio_unregister_driver(struct vio_driver *viodrv)
1188{
1189 driver_unregister(&viodrv->driver);
1190}
1191EXPORT_SYMBOL(vio_unregister_driver);
1192
c7f0e8cb
SR
1193/* vio_dev refcount hit 0 */
1194static void __devinit vio_dev_release(struct device *dev)
1195{
45848e0f
NA
1196 struct iommu_table *tbl = get_iommu_table_base(dev);
1197
1198 /* iSeries uses a common table for all vio devices */
1199 if (!firmware_has_feature(FW_FEATURE_ISERIES) && tbl)
1200 iommu_free_table(tbl, dev->of_node ?
1201 dev->of_node->full_name : dev_name(dev));
58f9b0b0 1202 of_node_put(dev->of_node);
c7f0e8cb
SR
1203 kfree(to_vio_dev(dev));
1204}
1205
1da177e4 1206/**
e10fa773
SR
1207 * vio_register_device_node: - Register a new vio device.
1208 * @of_node: The OF node for this device.
1da177e4 1209 *
e10fa773 1210 * Creates and initializes a vio_dev structure from the data in
12d04eef 1211 * of_node and adds it to the list of virtual devices.
e10fa773
SR
1212 * Returns a pointer to the created vio_dev or NULL if node has
1213 * NULL device_type or compatible fields.
1da177e4 1214 */
de7d812d 1215struct vio_dev *vio_register_device_node(struct device_node *of_node)
1da177e4 1216{
e10fa773 1217 struct vio_dev *viodev;
a7f67bdf 1218 const unsigned int *unit_address;
e10fa773
SR
1219
1220 /* we need the 'device_type' property, in order to match with drivers */
1221 if (of_node->type == NULL) {
1222 printk(KERN_WARNING "%s: node %s missing 'device_type'\n",
e48b1b45 1223 __func__,
e10fa773
SR
1224 of_node->name ? of_node->name : "<unknown>");
1225 return NULL;
1da177e4 1226 }
e10fa773 1227
e2eb6392 1228 unit_address = of_get_property(of_node, "reg", NULL);
e10fa773
SR
1229 if (unit_address == NULL) {
1230 printk(KERN_WARNING "%s: node %s missing 'reg'\n",
e48b1b45 1231 __func__,
e10fa773
SR
1232 of_node->name ? of_node->name : "<unknown>");
1233 return NULL;
1234 }
1235
1236 /* allocate a vio_dev for this node */
1237 viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1238 if (viodev == NULL)
1239 return NULL;
1240
0ebfff14 1241 viodev->irq = irq_of_parse_and_map(of_node, 0);
e10fa773 1242
aab0d375 1243 dev_set_name(&viodev->dev, "%x", *unit_address);
e10fa773
SR
1244 viodev->name = of_node->name;
1245 viodev->type = of_node->type;
1246 viodev->unit_address = *unit_address;
1247 if (firmware_has_feature(FW_FEATURE_ISERIES)) {
e2eb6392 1248 unit_address = of_get_property(of_node,
e10fa773
SR
1249 "linux,unit_address", NULL);
1250 if (unit_address != NULL)
1251 viodev->unit_address = *unit_address;
1252 }
d706c1b0 1253 viodev->dev.of_node = of_node_get(of_node);
a90ab95a
RJ
1254
1255 if (firmware_has_feature(FW_FEATURE_CMO))
1256 vio_cmo_set_dma_ops(viodev);
1257 else
6d283d78 1258 set_dma_ops(&viodev->dev, &dma_iommu_ops);
738ef42e 1259 set_iommu_table_base(&viodev->dev, vio_build_iommu_table(viodev));
8fae0353 1260 set_dev_node(&viodev->dev, of_node_to_nid(of_node));
c7f0e8cb
SR
1261
1262 /* init generic 'struct device' fields: */
1263 viodev->dev.parent = &vio_bus_device.dev;
1264 viodev->dev.bus = &vio_bus_type;
1265 viodev->dev.release = vio_dev_release;
b3c73856
NA
1266 /* needed to ensure proper operation of coherent allocations
1267 * later, in case driver doesn't set it explicitly */
1268 dma_set_mask(&viodev->dev, DMA_BIT_MASK(64));
1269 dma_set_coherent_mask(&viodev->dev, DMA_BIT_MASK(64));
e10fa773
SR
1270
1271 /* register with generic device framework */
c7f0e8cb
SR
1272 if (device_register(&viodev->dev)) {
1273 printk(KERN_ERR "%s: failed to register device %s\n",
aab0d375 1274 __func__, dev_name(&viodev->dev));
edea8f6f 1275 put_device(&viodev->dev);
e10fa773
SR
1276 return NULL;
1277 }
1278
1279 return viodev;
1da177e4 1280}
e10fa773 1281EXPORT_SYMBOL(vio_register_device_node);
1da177e4 1282
1da177e4
LT
1283/**
1284 * vio_bus_init: - Initialize the virtual IO bus
1285 */
c7f0e8cb 1286static int __init vio_bus_init(void)
1da177e4
LT
1287{
1288 int err;
e10fa773 1289 struct device_node *node_vroot;
1da177e4 1290
a90ab95a
RJ
1291 if (firmware_has_feature(FW_FEATURE_CMO))
1292 vio_cmo_sysfs_init();
1293
1da177e4
LT
1294 err = bus_register(&vio_bus_type);
1295 if (err) {
1296 printk(KERN_ERR "failed to register VIO bus\n");
1297 return err;
1298 }
1299
5c0b4b87
SR
1300 /*
1301 * The fake parent of all vio devices, just to give us
3e494c80
SR
1302 * a nice directory
1303 */
ac5b33c9 1304 err = device_register(&vio_bus_device.dev);
1da177e4 1305 if (err) {
3e494c80 1306 printk(KERN_WARNING "%s: device_register returned %i\n",
e48b1b45 1307 __func__, err);
1da177e4
LT
1308 return err;
1309 }
1310
a90ab95a
RJ
1311 if (firmware_has_feature(FW_FEATURE_CMO))
1312 vio_cmo_bus_init();
1313
30686ba6 1314 node_vroot = of_find_node_by_name(NULL, "vdevice");
e10fa773
SR
1315 if (node_vroot) {
1316 struct device_node *of_node;
1317
1318 /*
1319 * Create struct vio_devices for each virtual device in
1320 * the device tree. Drivers will associate with them later.
1321 */
1322 for (of_node = node_vroot->child; of_node != NULL;
c5467262 1323 of_node = of_node->sibling)
e10fa773 1324 vio_register_device_node(of_node);
30686ba6 1325 of_node_put(node_vroot);
e10fa773
SR
1326 }
1327
3e494c80
SR
1328 return 0;
1329}
c7f0e8cb 1330__initcall(vio_bus_init);
1da177e4 1331
e10fa773 1332static ssize_t name_show(struct device *dev,
5c0b4b87 1333 struct device_attribute *attr, char *buf)
1da177e4
LT
1334{
1335 return sprintf(buf, "%s\n", to_vio_dev(dev)->name);
1336}
e10fa773
SR
1337
1338static ssize_t devspec_show(struct device *dev,
1339 struct device_attribute *attr, char *buf)
1340{
58f9b0b0 1341 struct device_node *of_node = dev->of_node;
e10fa773
SR
1342
1343 return sprintf(buf, "%s\n", of_node ? of_node->full_name : "none");
1344}
1345
578b7cd1
BH
1346static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1347 char *buf)
1348{
1349 const struct vio_dev *vio_dev = to_vio_dev(dev);
1350 struct device_node *dn;
1351 const char *cp;
1352
cf9b59e9 1353 dn = dev->of_node;
578b7cd1
BH
1354 if (!dn)
1355 return -ENODEV;
1356 cp = of_get_property(dn, "compatible", NULL);
1357 if (!cp)
1358 return -ENODEV;
1359
1360 return sprintf(buf, "vio:T%sS%s\n", vio_dev->type, cp);
1361}
1362
e10fa773
SR
1363static struct device_attribute vio_dev_attrs[] = {
1364 __ATTR_RO(name),
1365 __ATTR_RO(devspec),
578b7cd1 1366 __ATTR_RO(modalias),
e10fa773
SR
1367 __ATTR_NULL
1368};
1da177e4 1369
1da177e4
LT
1370void __devinit vio_unregister_device(struct vio_dev *viodev)
1371{
1da177e4
LT
1372 device_unregister(&viodev->dev);
1373}
1374EXPORT_SYMBOL(vio_unregister_device);
1375
1da177e4
LT
1376static int vio_bus_match(struct device *dev, struct device_driver *drv)
1377{
1378 const struct vio_dev *vio_dev = to_vio_dev(dev);
1379 struct vio_driver *vio_drv = to_vio_driver(drv);
1380 const struct vio_device_id *ids = vio_drv->id_table;
1da177e4 1381
5c0b4b87 1382 return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL);
1da177e4
LT
1383}
1384
7eff2e7a 1385static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env)
143dcec2
OH
1386{
1387 const struct vio_dev *vio_dev = to_vio_dev(dev);
12d04eef 1388 struct device_node *dn;
a7f67bdf 1389 const char *cp;
143dcec2 1390
58f9b0b0 1391 dn = dev->of_node;
e10fa773 1392 if (!dn)
143dcec2 1393 return -ENODEV;
7eff2e7a 1394 cp = of_get_property(dn, "compatible", NULL);
143dcec2
OH
1395 if (!cp)
1396 return -ENODEV;
1397
7eff2e7a 1398 add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
143dcec2
OH
1399 return 0;
1400}
1401
6fccab26 1402static struct bus_type vio_bus_type = {
1da177e4 1403 .name = "vio",
e10fa773 1404 .dev_attrs = vio_dev_attrs,
312c004d 1405 .uevent = vio_hotplug,
1da177e4 1406 .match = vio_bus_match,
2f53a80f
RK
1407 .probe = vio_bus_probe,
1408 .remove = vio_bus_remove,
1da177e4 1409};
e10fa773
SR
1410
1411/**
1412 * vio_get_attribute: - get attribute for virtual device
1413 * @vdev: The vio device to get property.
1414 * @which: The property/attribute to be extracted.
1415 * @length: Pointer to length of returned data size (unused if NULL).
1416 *
e2eb6392 1417 * Calls prom.c's of_get_property() to return the value of the
e10fa773
SR
1418 * attribute specified by @which
1419*/
1420const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1421{
58f9b0b0 1422 return of_get_property(vdev->dev.of_node, which, length);
e10fa773
SR
1423}
1424EXPORT_SYMBOL(vio_get_attribute);
c7f0e8cb
SR
1425
1426#ifdef CONFIG_PPC_PSERIES
1427/* vio_find_name() - internal because only vio.c knows how we formatted the
1428 * kobject name
c7f0e8cb 1429 */
c847c853 1430static struct vio_dev *vio_find_name(const char *name)
c7f0e8cb 1431{
c847c853 1432 struct device *found;
c7f0e8cb 1433
c847c853 1434 found = bus_find_device_by_name(&vio_bus_type, NULL, name);
c7f0e8cb
SR
1435 if (!found)
1436 return NULL;
1437
c847c853 1438 return to_vio_dev(found);
c7f0e8cb
SR
1439}
1440
1441/**
1442 * vio_find_node - find an already-registered vio_dev
1443 * @vnode: device_node of the virtual device we're looking for
1444 */
1445struct vio_dev *vio_find_node(struct device_node *vnode)
1446{
a7f67bdf 1447 const uint32_t *unit_address;
aab0d375 1448 char kobj_name[20];
c7f0e8cb
SR
1449
1450 /* construct the kobject name from the device node */
e2eb6392 1451 unit_address = of_get_property(vnode, "reg", NULL);
c7f0e8cb
SR
1452 if (!unit_address)
1453 return NULL;
aab0d375 1454 snprintf(kobj_name, sizeof(kobj_name), "%x", *unit_address);
c7f0e8cb
SR
1455
1456 return vio_find_name(kobj_name);
1457}
1458EXPORT_SYMBOL(vio_find_node);
1459
1460int vio_enable_interrupts(struct vio_dev *dev)
1461{
1462 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1463 if (rc != H_SUCCESS)
1464 printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1465 return rc;
1466}
1467EXPORT_SYMBOL(vio_enable_interrupts);
1468
1469int vio_disable_interrupts(struct vio_dev *dev)
1470{
1471 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1472 if (rc != H_SUCCESS)
1473 printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1474 return rc;
1475}
1476EXPORT_SYMBOL(vio_disable_interrupts);
1477#endif /* CONFIG_PPC_PSERIES */