]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - arch/powerpc/kernel/vio.c
powerpc: various straight conversions from module.h --> export.h
[mirror_ubuntu-zesty-kernel.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
45223c54 609struct dma_map_ops vio_dma_mapping_ops = {
a90ab95a
RJ
610 .alloc_coherent = vio_dma_iommu_alloc_coherent,
611 .free_coherent = vio_dma_iommu_free_coherent,
a90ab95a
RJ
612 .map_sg = vio_dma_iommu_map_sg,
613 .unmap_sg = vio_dma_iommu_unmap_sg,
f9226d57
MN
614 .map_page = vio_dma_iommu_map_page,
615 .unmap_page = vio_dma_iommu_unmap_page,
6d283d78 616 .dma_supported = vio_dma_iommu_dma_supported,
f9226d57 617
a90ab95a
RJ
618};
619
620/**
621 * vio_cmo_set_dev_desired - Set desired entitlement for a device
622 *
623 * @viodev: struct vio_dev for device to alter
624 * @new_desired: new desired entitlement level in bytes
625 *
626 * For use by devices to request a change to their entitlement at runtime or
627 * through sysfs. The desired entitlement level is changed and a balancing
628 * of system resources is scheduled to run in the future.
629 */
630void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
631{
632 unsigned long flags;
633 struct vio_cmo_dev_entry *dev_ent;
634 int found = 0;
635
636 if (!firmware_has_feature(FW_FEATURE_CMO))
637 return;
638
639 spin_lock_irqsave(&vio_cmo.lock, flags);
640 if (desired < VIO_CMO_MIN_ENT)
641 desired = VIO_CMO_MIN_ENT;
642
643 /*
644 * Changes will not be made for devices not in the device list.
645 * If it is not in the device list, then no driver is loaded
646 * for the device and it can not receive entitlement.
647 */
648 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
649 if (viodev == dev_ent->viodev) {
650 found = 1;
651 break;
652 }
f6d8c8bb
JL
653 if (!found) {
654 spin_unlock_irqrestore(&vio_cmo.lock, flags);
a90ab95a 655 return;
f6d8c8bb 656 }
a90ab95a
RJ
657
658 /* Increase/decrease in desired device entitlement */
659 if (desired >= viodev->cmo.desired) {
660 /* Just bump the bus and device values prior to a balance*/
661 vio_cmo.desired += desired - viodev->cmo.desired;
662 viodev->cmo.desired = desired;
663 } else {
664 /* Decrease bus and device values for desired entitlement */
665 vio_cmo.desired -= viodev->cmo.desired - desired;
666 viodev->cmo.desired = desired;
667 /*
668 * If less entitlement is desired than current entitlement, move
669 * any reserve memory in the change region to the excess pool.
670 */
671 if (viodev->cmo.entitled > desired) {
672 vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
673 vio_cmo.excess.size += viodev->cmo.entitled - desired;
674 /*
675 * If entitlement moving from the reserve pool to the
676 * excess pool is currently unused, add to the excess
677 * free counter.
678 */
679 if (viodev->cmo.allocated < viodev->cmo.entitled)
680 vio_cmo.excess.free += viodev->cmo.entitled -
681 max(viodev->cmo.allocated, desired);
682 viodev->cmo.entitled = desired;
683 }
684 }
685 schedule_delayed_work(&vio_cmo.balance_q, 0);
686 spin_unlock_irqrestore(&vio_cmo.lock, flags);
687}
688
689/**
690 * vio_cmo_bus_probe - Handle CMO specific bus probe activities
691 *
692 * @viodev - Pointer to struct vio_dev for device
693 *
694 * Determine the devices IO memory entitlement needs, attempting
695 * to satisfy the system minimum entitlement at first and scheduling
696 * a balance operation to take care of the rest at a later time.
697 *
698 * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
699 * -ENOMEM when entitlement is not available for device or
700 * device entry.
701 *
702 */
703static int vio_cmo_bus_probe(struct vio_dev *viodev)
704{
705 struct vio_cmo_dev_entry *dev_ent;
706 struct device *dev = &viodev->dev;
707 struct vio_driver *viodrv = to_vio_driver(dev->driver);
708 unsigned long flags;
709 size_t size;
710
711 /*
712 * Check to see that device has a DMA window and configure
713 * entitlement for the device.
714 */
58f9b0b0 715 if (of_get_property(viodev->dev.of_node,
a90ab95a
RJ
716 "ibm,my-dma-window", NULL)) {
717 /* Check that the driver is CMO enabled and get desired DMA */
718 if (!viodrv->get_desired_dma) {
719 dev_err(dev, "%s: device driver does not support CMO\n",
720 __func__);
721 return -EINVAL;
722 }
723
724 viodev->cmo.desired = IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev));
725 if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
726 viodev->cmo.desired = VIO_CMO_MIN_ENT;
727 size = VIO_CMO_MIN_ENT;
728
729 dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
730 GFP_KERNEL);
731 if (!dev_ent)
732 return -ENOMEM;
733
734 dev_ent->viodev = viodev;
735 spin_lock_irqsave(&vio_cmo.lock, flags);
736 list_add(&dev_ent->list, &vio_cmo.device_list);
737 } else {
738 viodev->cmo.desired = 0;
739 size = 0;
740 spin_lock_irqsave(&vio_cmo.lock, flags);
741 }
742
743 /*
744 * If the needs for vio_cmo.min have not changed since they
745 * were last set, the number of devices in the OF tree has
746 * been constant and the IO memory for this is already in
747 * the reserve pool.
748 */
749 if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
750 VIO_CMO_MIN_ENT)) {
751 /* Updated desired entitlement if device requires it */
752 if (size)
753 vio_cmo.desired += (viodev->cmo.desired -
754 VIO_CMO_MIN_ENT);
755 } else {
756 size_t tmp;
757
758 tmp = vio_cmo.spare + vio_cmo.excess.free;
759 if (tmp < size) {
760 dev_err(dev, "%s: insufficient free "
761 "entitlement to add device. "
762 "Need %lu, have %lu\n", __func__,
763 size, (vio_cmo.spare + tmp));
764 spin_unlock_irqrestore(&vio_cmo.lock, flags);
765 return -ENOMEM;
766 }
767
768 /* Use excess pool first to fulfill request */
769 tmp = min(size, vio_cmo.excess.free);
770 vio_cmo.excess.free -= tmp;
771 vio_cmo.excess.size -= tmp;
772 vio_cmo.reserve.size += tmp;
773
774 /* Use spare if excess pool was insufficient */
775 vio_cmo.spare -= size - tmp;
776
777 /* Update bus accounting */
778 vio_cmo.min += size;
779 vio_cmo.desired += viodev->cmo.desired;
780 }
781 spin_unlock_irqrestore(&vio_cmo.lock, flags);
782 return 0;
783}
784
785/**
786 * vio_cmo_bus_remove - Handle CMO specific bus removal activities
787 *
788 * @viodev - Pointer to struct vio_dev for device
789 *
790 * Remove the device from the cmo device list. The minimum entitlement
791 * will be reserved for the device as long as it is in the system. The
792 * rest of the entitlement the device had been allocated will be returned
793 * to the system.
794 */
795static void vio_cmo_bus_remove(struct vio_dev *viodev)
796{
797 struct vio_cmo_dev_entry *dev_ent;
798 unsigned long flags;
799 size_t tmp;
800
801 spin_lock_irqsave(&vio_cmo.lock, flags);
802 if (viodev->cmo.allocated) {
803 dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
804 "allocated after remove operation.\n",
805 __func__, viodev->cmo.allocated);
806 BUG();
807 }
808
809 /*
810 * Remove the device from the device list being maintained for
811 * CMO enabled devices.
812 */
813 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
814 if (viodev == dev_ent->viodev) {
815 list_del(&dev_ent->list);
816 kfree(dev_ent);
817 break;
818 }
819
820 /*
821 * Devices may not require any entitlement and they do not need
822 * to be processed. Otherwise, return the device's entitlement
823 * back to the pools.
824 */
825 if (viodev->cmo.entitled) {
826 /*
827 * This device has not yet left the OF tree, it's
828 * minimum entitlement remains in vio_cmo.min and
829 * vio_cmo.desired
830 */
831 vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
832
833 /*
834 * Save min allocation for device in reserve as long
835 * as it exists in OF tree as determined by later
836 * balance operation
837 */
838 viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
839
840 /* Replenish spare from freed reserve pool */
841 if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
842 tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
843 vio_cmo.spare));
844 vio_cmo.spare += tmp;
845 viodev->cmo.entitled -= tmp;
846 }
847
848 /* Remaining reserve goes to excess pool */
849 vio_cmo.excess.size += viodev->cmo.entitled;
850 vio_cmo.excess.free += viodev->cmo.entitled;
851 vio_cmo.reserve.size -= viodev->cmo.entitled;
852
853 /*
854 * Until the device is removed it will keep a
855 * minimum entitlement; this will guarantee that
856 * a module unload/load will result in a success.
857 */
858 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
859 viodev->cmo.desired = VIO_CMO_MIN_ENT;
860 atomic_set(&viodev->cmo.allocs_failed, 0);
861 }
862
863 spin_unlock_irqrestore(&vio_cmo.lock, flags);
864}
865
866static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
867{
6d283d78 868 set_dma_ops(&viodev->dev, &vio_dma_mapping_ops);
a90ab95a
RJ
869}
870
871/**
872 * vio_cmo_bus_init - CMO entitlement initialization at bus init time
873 *
874 * Set up the reserve and excess entitlement pools based on available
875 * system entitlement and the number of devices in the OF tree that
876 * require entitlement in the reserve pool.
877 */
878static void vio_cmo_bus_init(void)
879{
880 struct hvcall_mpp_data mpp_data;
881 int err;
882
883 memset(&vio_cmo, 0, sizeof(struct vio_cmo));
884 spin_lock_init(&vio_cmo.lock);
885 INIT_LIST_HEAD(&vio_cmo.device_list);
886 INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
887
888 /* Get current system entitlement */
889 err = h_get_mpp(&mpp_data);
890
891 /*
892 * On failure, continue with entitlement set to 0, will panic()
893 * later when spare is reserved.
894 */
895 if (err != H_SUCCESS) {
896 printk(KERN_ERR "%s: unable to determine system IO "\
897 "entitlement. (%d)\n", __func__, err);
898 vio_cmo.entitled = 0;
899 } else {
900 vio_cmo.entitled = mpp_data.entitled_mem;
901 }
902
903 /* Set reservation and check against entitlement */
904 vio_cmo.spare = VIO_CMO_MIN_ENT;
905 vio_cmo.reserve.size = vio_cmo.spare;
906 vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
907 VIO_CMO_MIN_ENT);
908 if (vio_cmo.reserve.size > vio_cmo.entitled) {
909 printk(KERN_ERR "%s: insufficient system entitlement\n",
910 __func__);
911 panic("%s: Insufficient system entitlement", __func__);
912 }
913
914 /* Set the remaining accounting variables */
915 vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
916 vio_cmo.excess.free = vio_cmo.excess.size;
917 vio_cmo.min = vio_cmo.reserve.size;
918 vio_cmo.desired = vio_cmo.reserve.size;
919}
920
921/* sysfs device functions and data structures for CMO */
922
923#define viodev_cmo_rd_attr(name) \
924static ssize_t viodev_cmo_##name##_show(struct device *dev, \
925 struct device_attribute *attr, \
926 char *buf) \
927{ \
928 return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \
929}
930
931static ssize_t viodev_cmo_allocs_failed_show(struct device *dev,
932 struct device_attribute *attr, char *buf)
933{
934 struct vio_dev *viodev = to_vio_dev(dev);
935 return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
936}
937
938static ssize_t viodev_cmo_allocs_failed_reset(struct device *dev,
939 struct device_attribute *attr, const char *buf, size_t count)
940{
941 struct vio_dev *viodev = to_vio_dev(dev);
942 atomic_set(&viodev->cmo.allocs_failed, 0);
943 return count;
944}
945
946static ssize_t viodev_cmo_desired_set(struct device *dev,
947 struct device_attribute *attr, const char *buf, size_t count)
948{
949 struct vio_dev *viodev = to_vio_dev(dev);
950 size_t new_desired;
951 int ret;
952
953 ret = strict_strtoul(buf, 10, &new_desired);
954 if (ret)
955 return ret;
956
957 vio_cmo_set_dev_desired(viodev, new_desired);
958 return count;
959}
960
961viodev_cmo_rd_attr(desired);
962viodev_cmo_rd_attr(entitled);
963viodev_cmo_rd_attr(allocated);
964
965static ssize_t name_show(struct device *, struct device_attribute *, char *);
966static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
578b7cd1
BH
967static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
968 char *buf);
a90ab95a
RJ
969static struct device_attribute vio_cmo_dev_attrs[] = {
970 __ATTR_RO(name),
971 __ATTR_RO(devspec),
578b7cd1 972 __ATTR_RO(modalias),
a90ab95a
RJ
973 __ATTR(cmo_desired, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
974 viodev_cmo_desired_show, viodev_cmo_desired_set),
975 __ATTR(cmo_entitled, S_IRUGO, viodev_cmo_entitled_show, NULL),
976 __ATTR(cmo_allocated, S_IRUGO, viodev_cmo_allocated_show, NULL),
977 __ATTR(cmo_allocs_failed, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
978 viodev_cmo_allocs_failed_show, viodev_cmo_allocs_failed_reset),
979 __ATTR_NULL
980};
981
982/* sysfs bus functions and data structures for CMO */
983
984#define viobus_cmo_rd_attr(name) \
985static ssize_t \
986viobus_cmo_##name##_show(struct bus_type *bt, char *buf) \
987{ \
988 return sprintf(buf, "%lu\n", vio_cmo.name); \
989}
990
991#define viobus_cmo_pool_rd_attr(name, var) \
992static ssize_t \
993viobus_cmo_##name##_pool_show_##var(struct bus_type *bt, char *buf) \
994{ \
995 return sprintf(buf, "%lu\n", vio_cmo.name.var); \
996}
997
998static ssize_t viobus_cmo_high_reset(struct bus_type *bt, const char *buf,
999 size_t count)
1000{
1001 unsigned long flags;
1002
1003 spin_lock_irqsave(&vio_cmo.lock, flags);
1004 vio_cmo.high = vio_cmo.curr;
1005 spin_unlock_irqrestore(&vio_cmo.lock, flags);
1006
1007 return count;
1008}
1009
1010viobus_cmo_rd_attr(entitled);
1011viobus_cmo_pool_rd_attr(reserve, size);
1012viobus_cmo_pool_rd_attr(excess, size);
1013viobus_cmo_pool_rd_attr(excess, free);
1014viobus_cmo_rd_attr(spare);
1015viobus_cmo_rd_attr(min);
1016viobus_cmo_rd_attr(desired);
1017viobus_cmo_rd_attr(curr);
1018viobus_cmo_rd_attr(high);
1019
1020static struct bus_attribute vio_cmo_bus_attrs[] = {
1021 __ATTR(cmo_entitled, S_IRUGO, viobus_cmo_entitled_show, NULL),
1022 __ATTR(cmo_reserve_size, S_IRUGO, viobus_cmo_reserve_pool_show_size, NULL),
1023 __ATTR(cmo_excess_size, S_IRUGO, viobus_cmo_excess_pool_show_size, NULL),
1024 __ATTR(cmo_excess_free, S_IRUGO, viobus_cmo_excess_pool_show_free, NULL),
1025 __ATTR(cmo_spare, S_IRUGO, viobus_cmo_spare_show, NULL),
1026 __ATTR(cmo_min, S_IRUGO, viobus_cmo_min_show, NULL),
1027 __ATTR(cmo_desired, S_IRUGO, viobus_cmo_desired_show, NULL),
1028 __ATTR(cmo_curr, S_IRUGO, viobus_cmo_curr_show, NULL),
1029 __ATTR(cmo_high, S_IWUSR|S_IRUSR|S_IWGRP|S_IRGRP|S_IROTH,
1030 viobus_cmo_high_show, viobus_cmo_high_reset),
1031 __ATTR_NULL
1032};
1033
1034static void vio_cmo_sysfs_init(void)
1035{
1036 vio_bus_type.dev_attrs = vio_cmo_dev_attrs;
1037 vio_bus_type.bus_attrs = vio_cmo_bus_attrs;
1038}
1039#else /* CONFIG_PPC_SMLPAR */
1040/* Dummy functions for iSeries platform */
1041int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1042void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1043static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1044static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1045static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
b9fa49a9
NL
1046static void vio_cmo_bus_init(void) {}
1047static void vio_cmo_sysfs_init(void) { }
a90ab95a
RJ
1048#endif /* CONFIG_PPC_SMLPAR */
1049EXPORT_SYMBOL(vio_cmo_entitlement_update);
1050EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1051
c7f0e8cb
SR
1052static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1053{
dd9b67ab
SR
1054 const unsigned char *dma_window;
1055 struct iommu_table *tbl;
1056 unsigned long offset, size;
1057
1058 if (firmware_has_feature(FW_FEATURE_ISERIES))
1059 return vio_build_iommu_table_iseries(dev);
c7f0e8cb 1060
58f9b0b0 1061 dma_window = of_get_property(dev->dev.of_node,
dd9b67ab
SR
1062 "ibm,my-dma-window", NULL);
1063 if (!dma_window)
1064 return NULL;
c7f0e8cb 1065
7aa241fd 1066 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
0f337274 1067 if (tbl == NULL)
1068 return NULL;
c7f0e8cb 1069
58f9b0b0 1070 of_parse_dma_window(dev->dev.of_node, dma_window,
dd9b67ab 1071 &tbl->it_index, &offset, &size);
c7f0e8cb 1072
dd9b67ab
SR
1073 /* TCE table size - measured in tce entries */
1074 tbl->it_size = size >> IOMMU_PAGE_SHIFT;
1075 /* offset for VIO should always be 0 */
1076 tbl->it_offset = offset >> IOMMU_PAGE_SHIFT;
1077 tbl->it_busno = 0;
1078 tbl->it_type = TCE_VB;
7aa241fd 1079 tbl->it_blocksize = 16;
c7f0e8cb 1080
dd9b67ab 1081 return iommu_init_table(tbl, -1);
c7f0e8cb 1082}
1da177e4 1083
e10fa773
SR
1084/**
1085 * vio_match_device: - Tell if a VIO device has a matching
1086 * VIO device id structure.
1087 * @ids: array of VIO device id structures to search in
1088 * @dev: the VIO device structure to match against
1089 *
1090 * Used by a driver to check whether a VIO device present in the
1091 * system is in its list of supported devices. Returns the matching
1092 * vio_device_id structure or NULL if there is no match.
1093 */
1094static const struct vio_device_id *vio_match_device(
1095 const struct vio_device_id *ids, const struct vio_dev *dev)
1096{
1097 while (ids->type[0] != '\0') {
dd721ffd 1098 if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
58f9b0b0 1099 of_device_is_compatible(dev->dev.of_node,
12d04eef 1100 ids->compat))
e10fa773
SR
1101 return ids;
1102 ids++;
1103 }
1104 return NULL;
1105}
1106
5c0b4b87
SR
1107/*
1108 * Convert from struct device to struct vio_dev and pass to driver.
1da177e4 1109 * dev->driver has already been set by generic code because vio_bus_match
5c0b4b87
SR
1110 * succeeded.
1111 */
1da177e4
LT
1112static int vio_bus_probe(struct device *dev)
1113{
1114 struct vio_dev *viodev = to_vio_dev(dev);
1115 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1116 const struct vio_device_id *id;
1117 int error = -ENODEV;
1118
1da177e4
LT
1119 if (!viodrv->probe)
1120 return error;
1121
1122 id = vio_match_device(viodrv->id_table, viodev);
a90ab95a
RJ
1123 if (id) {
1124 memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1125 if (firmware_has_feature(FW_FEATURE_CMO)) {
1126 error = vio_cmo_bus_probe(viodev);
1127 if (error)
1128 return error;
1129 }
1da177e4 1130 error = viodrv->probe(viodev, id);
cd5aeb9f 1131 if (error && firmware_has_feature(FW_FEATURE_CMO))
a90ab95a
RJ
1132 vio_cmo_bus_remove(viodev);
1133 }
1da177e4
LT
1134
1135 return error;
1136}
1137
1138/* convert from struct device to struct vio_dev and pass to driver. */
1139static int vio_bus_remove(struct device *dev)
1140{
1141 struct vio_dev *viodev = to_vio_dev(dev);
1142 struct vio_driver *viodrv = to_vio_driver(dev->driver);
a90ab95a
RJ
1143 struct device *devptr;
1144 int ret = 1;
1145
1146 /*
1147 * Hold a reference to the device after the remove function is called
1148 * to allow for CMO accounting cleanup for the device.
1149 */
1150 devptr = get_device(dev);
1da177e4 1151
5c0b4b87 1152 if (viodrv->remove)
a90ab95a 1153 ret = viodrv->remove(viodev);
1da177e4 1154
a90ab95a
RJ
1155 if (!ret && firmware_has_feature(FW_FEATURE_CMO))
1156 vio_cmo_bus_remove(viodev);
1157
1158 put_device(devptr);
1159 return ret;
1da177e4
LT
1160}
1161
1162/**
1163 * vio_register_driver: - Register a new vio driver
1164 * @drv: The vio_driver structure to be registered.
1165 */
1166int vio_register_driver(struct vio_driver *viodrv)
1167{
e48b1b45 1168 printk(KERN_DEBUG "%s: driver %s registering\n", __func__,
6fdf5392 1169 viodrv->driver.name);
1da177e4
LT
1170
1171 /* fill in 'struct driver' fields */
1da177e4 1172 viodrv->driver.bus = &vio_bus_type;
1da177e4
LT
1173
1174 return driver_register(&viodrv->driver);
1175}
1176EXPORT_SYMBOL(vio_register_driver);
1177
1178/**
1179 * vio_unregister_driver - Remove registration of vio driver.
1180 * @driver: The vio_driver struct to be removed form registration
1181 */
1182void vio_unregister_driver(struct vio_driver *viodrv)
1183{
1184 driver_unregister(&viodrv->driver);
1185}
1186EXPORT_SYMBOL(vio_unregister_driver);
1187
c7f0e8cb
SR
1188/* vio_dev refcount hit 0 */
1189static void __devinit vio_dev_release(struct device *dev)
1190{
45848e0f
NA
1191 struct iommu_table *tbl = get_iommu_table_base(dev);
1192
1193 /* iSeries uses a common table for all vio devices */
1194 if (!firmware_has_feature(FW_FEATURE_ISERIES) && tbl)
1195 iommu_free_table(tbl, dev->of_node ?
1196 dev->of_node->full_name : dev_name(dev));
58f9b0b0 1197 of_node_put(dev->of_node);
c7f0e8cb
SR
1198 kfree(to_vio_dev(dev));
1199}
1200
1da177e4 1201/**
e10fa773
SR
1202 * vio_register_device_node: - Register a new vio device.
1203 * @of_node: The OF node for this device.
1da177e4 1204 *
e10fa773 1205 * Creates and initializes a vio_dev structure from the data in
12d04eef 1206 * of_node and adds it to the list of virtual devices.
e10fa773
SR
1207 * Returns a pointer to the created vio_dev or NULL if node has
1208 * NULL device_type or compatible fields.
1da177e4 1209 */
de7d812d 1210struct vio_dev *vio_register_device_node(struct device_node *of_node)
1da177e4 1211{
e10fa773 1212 struct vio_dev *viodev;
a7f67bdf 1213 const unsigned int *unit_address;
e10fa773
SR
1214
1215 /* we need the 'device_type' property, in order to match with drivers */
1216 if (of_node->type == NULL) {
1217 printk(KERN_WARNING "%s: node %s missing 'device_type'\n",
e48b1b45 1218 __func__,
e10fa773
SR
1219 of_node->name ? of_node->name : "<unknown>");
1220 return NULL;
1da177e4 1221 }
e10fa773 1222
e2eb6392 1223 unit_address = of_get_property(of_node, "reg", NULL);
e10fa773
SR
1224 if (unit_address == NULL) {
1225 printk(KERN_WARNING "%s: node %s missing 'reg'\n",
e48b1b45 1226 __func__,
e10fa773
SR
1227 of_node->name ? of_node->name : "<unknown>");
1228 return NULL;
1229 }
1230
1231 /* allocate a vio_dev for this node */
1232 viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1233 if (viodev == NULL)
1234 return NULL;
1235
0ebfff14 1236 viodev->irq = irq_of_parse_and_map(of_node, 0);
e10fa773 1237
aab0d375 1238 dev_set_name(&viodev->dev, "%x", *unit_address);
e10fa773
SR
1239 viodev->name = of_node->name;
1240 viodev->type = of_node->type;
1241 viodev->unit_address = *unit_address;
1242 if (firmware_has_feature(FW_FEATURE_ISERIES)) {
e2eb6392 1243 unit_address = of_get_property(of_node,
e10fa773
SR
1244 "linux,unit_address", NULL);
1245 if (unit_address != NULL)
1246 viodev->unit_address = *unit_address;
1247 }
d706c1b0 1248 viodev->dev.of_node = of_node_get(of_node);
a90ab95a
RJ
1249
1250 if (firmware_has_feature(FW_FEATURE_CMO))
1251 vio_cmo_set_dma_ops(viodev);
1252 else
6d283d78 1253 set_dma_ops(&viodev->dev, &dma_iommu_ops);
738ef42e 1254 set_iommu_table_base(&viodev->dev, vio_build_iommu_table(viodev));
8fae0353 1255 set_dev_node(&viodev->dev, of_node_to_nid(of_node));
c7f0e8cb
SR
1256
1257 /* init generic 'struct device' fields: */
1258 viodev->dev.parent = &vio_bus_device.dev;
1259 viodev->dev.bus = &vio_bus_type;
1260 viodev->dev.release = vio_dev_release;
b3c73856
NA
1261 /* needed to ensure proper operation of coherent allocations
1262 * later, in case driver doesn't set it explicitly */
1263 dma_set_mask(&viodev->dev, DMA_BIT_MASK(64));
1264 dma_set_coherent_mask(&viodev->dev, DMA_BIT_MASK(64));
e10fa773
SR
1265
1266 /* register with generic device framework */
c7f0e8cb
SR
1267 if (device_register(&viodev->dev)) {
1268 printk(KERN_ERR "%s: failed to register device %s\n",
aab0d375 1269 __func__, dev_name(&viodev->dev));
edea8f6f 1270 put_device(&viodev->dev);
e10fa773
SR
1271 return NULL;
1272 }
1273
1274 return viodev;
1da177e4 1275}
e10fa773 1276EXPORT_SYMBOL(vio_register_device_node);
1da177e4 1277
1da177e4
LT
1278/**
1279 * vio_bus_init: - Initialize the virtual IO bus
1280 */
c7f0e8cb 1281static int __init vio_bus_init(void)
1da177e4
LT
1282{
1283 int err;
e10fa773 1284 struct device_node *node_vroot;
1da177e4 1285
a90ab95a
RJ
1286 if (firmware_has_feature(FW_FEATURE_CMO))
1287 vio_cmo_sysfs_init();
1288
1da177e4
LT
1289 err = bus_register(&vio_bus_type);
1290 if (err) {
1291 printk(KERN_ERR "failed to register VIO bus\n");
1292 return err;
1293 }
1294
5c0b4b87
SR
1295 /*
1296 * The fake parent of all vio devices, just to give us
3e494c80
SR
1297 * a nice directory
1298 */
ac5b33c9 1299 err = device_register(&vio_bus_device.dev);
1da177e4 1300 if (err) {
3e494c80 1301 printk(KERN_WARNING "%s: device_register returned %i\n",
e48b1b45 1302 __func__, err);
1da177e4
LT
1303 return err;
1304 }
1305
a90ab95a
RJ
1306 if (firmware_has_feature(FW_FEATURE_CMO))
1307 vio_cmo_bus_init();
1308
30686ba6 1309 node_vroot = of_find_node_by_name(NULL, "vdevice");
e10fa773
SR
1310 if (node_vroot) {
1311 struct device_node *of_node;
1312
1313 /*
1314 * Create struct vio_devices for each virtual device in
1315 * the device tree. Drivers will associate with them later.
1316 */
1317 for (of_node = node_vroot->child; of_node != NULL;
c5467262 1318 of_node = of_node->sibling)
e10fa773 1319 vio_register_device_node(of_node);
30686ba6 1320 of_node_put(node_vroot);
e10fa773
SR
1321 }
1322
3e494c80
SR
1323 return 0;
1324}
c7f0e8cb 1325__initcall(vio_bus_init);
1da177e4 1326
e10fa773 1327static ssize_t name_show(struct device *dev,
5c0b4b87 1328 struct device_attribute *attr, char *buf)
1da177e4
LT
1329{
1330 return sprintf(buf, "%s\n", to_vio_dev(dev)->name);
1331}
e10fa773
SR
1332
1333static ssize_t devspec_show(struct device *dev,
1334 struct device_attribute *attr, char *buf)
1335{
58f9b0b0 1336 struct device_node *of_node = dev->of_node;
e10fa773
SR
1337
1338 return sprintf(buf, "%s\n", of_node ? of_node->full_name : "none");
1339}
1340
578b7cd1
BH
1341static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1342 char *buf)
1343{
1344 const struct vio_dev *vio_dev = to_vio_dev(dev);
1345 struct device_node *dn;
1346 const char *cp;
1347
cf9b59e9 1348 dn = dev->of_node;
578b7cd1
BH
1349 if (!dn)
1350 return -ENODEV;
1351 cp = of_get_property(dn, "compatible", NULL);
1352 if (!cp)
1353 return -ENODEV;
1354
1355 return sprintf(buf, "vio:T%sS%s\n", vio_dev->type, cp);
1356}
1357
e10fa773
SR
1358static struct device_attribute vio_dev_attrs[] = {
1359 __ATTR_RO(name),
1360 __ATTR_RO(devspec),
578b7cd1 1361 __ATTR_RO(modalias),
e10fa773
SR
1362 __ATTR_NULL
1363};
1da177e4 1364
1da177e4
LT
1365void __devinit vio_unregister_device(struct vio_dev *viodev)
1366{
1da177e4
LT
1367 device_unregister(&viodev->dev);
1368}
1369EXPORT_SYMBOL(vio_unregister_device);
1370
1da177e4
LT
1371static int vio_bus_match(struct device *dev, struct device_driver *drv)
1372{
1373 const struct vio_dev *vio_dev = to_vio_dev(dev);
1374 struct vio_driver *vio_drv = to_vio_driver(drv);
1375 const struct vio_device_id *ids = vio_drv->id_table;
1da177e4 1376
5c0b4b87 1377 return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL);
1da177e4
LT
1378}
1379
7eff2e7a 1380static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env)
143dcec2
OH
1381{
1382 const struct vio_dev *vio_dev = to_vio_dev(dev);
12d04eef 1383 struct device_node *dn;
a7f67bdf 1384 const char *cp;
143dcec2 1385
58f9b0b0 1386 dn = dev->of_node;
e10fa773 1387 if (!dn)
143dcec2 1388 return -ENODEV;
7eff2e7a 1389 cp = of_get_property(dn, "compatible", NULL);
143dcec2
OH
1390 if (!cp)
1391 return -ENODEV;
1392
7eff2e7a 1393 add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
143dcec2
OH
1394 return 0;
1395}
1396
6fccab26 1397static struct bus_type vio_bus_type = {
1da177e4 1398 .name = "vio",
e10fa773 1399 .dev_attrs = vio_dev_attrs,
312c004d 1400 .uevent = vio_hotplug,
1da177e4 1401 .match = vio_bus_match,
2f53a80f
RK
1402 .probe = vio_bus_probe,
1403 .remove = vio_bus_remove,
a1263c71 1404 .pm = GENERIC_SUBSYS_PM_OPS,
1da177e4 1405};
e10fa773
SR
1406
1407/**
1408 * vio_get_attribute: - get attribute for virtual device
1409 * @vdev: The vio device to get property.
1410 * @which: The property/attribute to be extracted.
1411 * @length: Pointer to length of returned data size (unused if NULL).
1412 *
e2eb6392 1413 * Calls prom.c's of_get_property() to return the value of the
e10fa773
SR
1414 * attribute specified by @which
1415*/
1416const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1417{
58f9b0b0 1418 return of_get_property(vdev->dev.of_node, which, length);
e10fa773
SR
1419}
1420EXPORT_SYMBOL(vio_get_attribute);
c7f0e8cb
SR
1421
1422#ifdef CONFIG_PPC_PSERIES
1423/* vio_find_name() - internal because only vio.c knows how we formatted the
1424 * kobject name
c7f0e8cb 1425 */
c847c853 1426static struct vio_dev *vio_find_name(const char *name)
c7f0e8cb 1427{
c847c853 1428 struct device *found;
c7f0e8cb 1429
c847c853 1430 found = bus_find_device_by_name(&vio_bus_type, NULL, name);
c7f0e8cb
SR
1431 if (!found)
1432 return NULL;
1433
c847c853 1434 return to_vio_dev(found);
c7f0e8cb
SR
1435}
1436
1437/**
1438 * vio_find_node - find an already-registered vio_dev
1439 * @vnode: device_node of the virtual device we're looking for
1440 */
1441struct vio_dev *vio_find_node(struct device_node *vnode)
1442{
a7f67bdf 1443 const uint32_t *unit_address;
aab0d375 1444 char kobj_name[20];
c7f0e8cb
SR
1445
1446 /* construct the kobject name from the device node */
e2eb6392 1447 unit_address = of_get_property(vnode, "reg", NULL);
c7f0e8cb
SR
1448 if (!unit_address)
1449 return NULL;
aab0d375 1450 snprintf(kobj_name, sizeof(kobj_name), "%x", *unit_address);
c7f0e8cb
SR
1451
1452 return vio_find_name(kobj_name);
1453}
1454EXPORT_SYMBOL(vio_find_node);
1455
1456int vio_enable_interrupts(struct vio_dev *dev)
1457{
1458 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1459 if (rc != H_SUCCESS)
1460 printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1461 return rc;
1462}
1463EXPORT_SYMBOL(vio_enable_interrupts);
1464
1465int vio_disable_interrupts(struct vio_dev *dev)
1466{
1467 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1468 if (rc != H_SUCCESS)
1469 printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1470 return rc;
1471}
1472EXPORT_SYMBOL(vio_disable_interrupts);
1473#endif /* CONFIG_PPC_PSERIES */