1 /*
2  * IBM PowerPC Virtual I/O Infrastructure Support.
3  *
4  *    Copyright (c) 2003,2008 IBM Corp.
5  *     Dave Engebretsen engebret@us.ibm.com
6  *     Santiago Leon santil@us.ibm.com
7  *     Hollis Blanchard <hollisb@us.ibm.com>
8  *     Stephen Rothwell
9  *     Robert Jennings <rcjenn@us.ibm.com>
10  *
11  *      This program is free software; you can redistribute it and/or
12  *      modify it under the terms of the GNU General Public License
13  *      as published by the Free Software Foundation; either version
14  *      2 of the License, or (at your option) any later version.
15  */
16 
17 #include <linux/cpu.h>
18 #include <linux/types.h>
19 #include <linux/delay.h>
20 #include <linux/stat.h>
21 #include <linux/device.h>
22 #include <linux/init.h>
23 #include <linux/slab.h>
24 #include <linux/console.h>
25 #include <linux/export.h>
26 #include <linux/mm.h>
27 #include <linux/dma-mapping.h>
28 #include <linux/kobject.h>
29 
30 #include <asm/iommu.h>
31 #include <asm/dma.h>
32 #include <asm/vio.h>
33 #include <asm/prom.h>
34 #include <asm/firmware.h>
35 #include <asm/tce.h>
36 #include <asm/page.h>
37 #include <asm/hvcall.h>
38 
39 static struct vio_dev vio_bus_device  = { /* fake "parent" device */
40 	.name = "vio",
41 	.type = "",
42 	.dev.init_name = "vio",
43 	.dev.bus = &vio_bus_type,
44 };
45 
46 #ifdef CONFIG_PPC_SMLPAR
47 /**
48  * vio_cmo_pool - A pool of IO memory for CMO use
49  *
50  * @size: The size of the pool in bytes
51  * @free: The amount of free memory in the pool
52  */
53 struct vio_cmo_pool {
54 	size_t size;
55 	size_t free;
56 };
57 
58 /* How many ms to delay queued balance work */
59 #define VIO_CMO_BALANCE_DELAY 100
60 
61 /* Portion out IO memory to CMO devices by this chunk size */
62 #define VIO_CMO_BALANCE_CHUNK 131072
63 
64 /**
65  * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement
66  *
67  * @vio_dev: struct vio_dev pointer
68  * @list: pointer to other devices on bus that are being tracked
69  */
70 struct vio_cmo_dev_entry {
71 	struct vio_dev *viodev;
72 	struct list_head list;
73 };
74 
75 /**
76  * vio_cmo - VIO bus accounting structure for CMO entitlement
77  *
78  * @lock: spinlock for entire structure
79  * @balance_q: work queue for balancing system entitlement
80  * @device_list: list of CMO-enabled devices requiring entitlement
81  * @entitled: total system entitlement in bytes
82  * @reserve: pool of memory from which devices reserve entitlement, incl. spare
83  * @excess: pool of excess entitlement not needed for device reserves or spare
84  * @spare: IO memory for device hotplug functionality
85  * @min: minimum necessary for system operation
86  * @desired: desired memory for system operation
87  * @curr: bytes currently allocated
88  * @high: high water mark for IO data usage
89  */
90 static struct vio_cmo {
91 	spinlock_t lock;
92 	struct delayed_work balance_q;
93 	struct list_head device_list;
94 	size_t entitled;
95 	struct vio_cmo_pool reserve;
96 	struct vio_cmo_pool excess;
97 	size_t spare;
98 	size_t min;
99 	size_t desired;
100 	size_t curr;
101 	size_t high;
102 } vio_cmo;
103 
104 /**
105  * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows
106  */
107 static int vio_cmo_num_OF_devs(void)
108 {
109 	struct device_node *node_vroot;
110 	int count = 0;
111 
112 	/*
113 	 * Count the number of vdevice entries with an
114 	 * ibm,my-dma-window OF property
115 	 */
116 	node_vroot = of_find_node_by_name(NULL, "vdevice");
117 	if (node_vroot) {
118 		struct device_node *of_node;
119 		struct property *prop;
120 
121 		for_each_child_of_node(node_vroot, of_node) {
122 			prop = of_find_property(of_node, "ibm,my-dma-window",
123 			                       NULL);
124 			if (prop)
125 				count++;
126 		}
127 	}
128 	of_node_put(node_vroot);
129 	return count;
130 }
131 
132 /**
133  * vio_cmo_alloc - allocate IO memory for CMO-enable devices
134  *
135  * @viodev: VIO device requesting IO memory
136  * @size: size of allocation requested
137  *
138  * Allocations come from memory reserved for the devices and any excess
139  * IO memory available to all devices.  The spare pool used to service
140  * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be
141  * made available.
142  *
143  * Return codes:
144  *  0 for successful allocation and -ENOMEM for a failure
145  */
146 static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size)
147 {
148 	unsigned long flags;
149 	size_t reserve_free = 0;
150 	size_t excess_free = 0;
151 	int ret = -ENOMEM;
152 
153 	spin_lock_irqsave(&vio_cmo.lock, flags);
154 
155 	/* Determine the amount of free entitlement available in reserve */
156 	if (viodev->cmo.entitled > viodev->cmo.allocated)
157 		reserve_free = viodev->cmo.entitled - viodev->cmo.allocated;
158 
159 	/* If spare is not fulfilled, the excess pool can not be used. */
160 	if (vio_cmo.spare >= VIO_CMO_MIN_ENT)
161 		excess_free = vio_cmo.excess.free;
162 
163 	/* The request can be satisfied */
164 	if ((reserve_free + excess_free) >= size) {
165 		vio_cmo.curr += size;
166 		if (vio_cmo.curr > vio_cmo.high)
167 			vio_cmo.high = vio_cmo.curr;
168 		viodev->cmo.allocated += size;
169 		size -= min(reserve_free, size);
170 		vio_cmo.excess.free -= size;
171 		ret = 0;
172 	}
173 
174 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
175 	return ret;
176 }
177 
178 /**
179  * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices
180  * @viodev: VIO device freeing IO memory
181  * @size: size of deallocation
182  *
183  * IO memory is freed by the device back to the correct memory pools.
184  * The spare pool is replenished first from either memory pool, then
185  * the reserve pool is used to reduce device entitlement, the excess
186  * pool is used to increase the reserve pool toward the desired entitlement
187  * target, and then the remaining memory is returned to the pools.
188  *
189  */
190 static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size)
191 {
192 	unsigned long flags;
193 	size_t spare_needed = 0;
194 	size_t excess_freed = 0;
195 	size_t reserve_freed = size;
196 	size_t tmp;
197 	int balance = 0;
198 
199 	spin_lock_irqsave(&vio_cmo.lock, flags);
200 	vio_cmo.curr -= size;
201 
202 	/* Amount of memory freed from the excess pool */
203 	if (viodev->cmo.allocated > viodev->cmo.entitled) {
204 		excess_freed = min(reserve_freed, (viodev->cmo.allocated -
205 		                                   viodev->cmo.entitled));
206 		reserve_freed -= excess_freed;
207 	}
208 
209 	/* Remove allocation from device */
210 	viodev->cmo.allocated -= (reserve_freed + excess_freed);
211 
212 	/* Spare is a subset of the reserve pool, replenish it first. */
213 	spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare;
214 
215 	/*
216 	 * Replenish the spare in the reserve pool from the excess pool.
217 	 * This moves entitlement into the reserve pool.
218 	 */
219 	if (spare_needed && excess_freed) {
220 		tmp = min(excess_freed, spare_needed);
221 		vio_cmo.excess.size -= tmp;
222 		vio_cmo.reserve.size += tmp;
223 		vio_cmo.spare += tmp;
224 		excess_freed -= tmp;
225 		spare_needed -= tmp;
226 		balance = 1;
227 	}
228 
229 	/*
230 	 * Replenish the spare in the reserve pool from the reserve pool.
231 	 * This removes entitlement from the device down to VIO_CMO_MIN_ENT,
232 	 * if needed, and gives it to the spare pool. The amount of used
233 	 * memory in this pool does not change.
234 	 */
235 	if (spare_needed && reserve_freed) {
236 		tmp = min3(spare_needed, reserve_freed, (viodev->cmo.entitled - VIO_CMO_MIN_ENT));
237 
238 		vio_cmo.spare += tmp;
239 		viodev->cmo.entitled -= tmp;
240 		reserve_freed -= tmp;
241 		spare_needed -= tmp;
242 		balance = 1;
243 	}
244 
245 	/*
246 	 * Increase the reserve pool until the desired allocation is met.
247 	 * Move an allocation freed from the excess pool into the reserve
248 	 * pool and schedule a balance operation.
249 	 */
250 	if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) {
251 		tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size));
252 
253 		vio_cmo.excess.size -= tmp;
254 		vio_cmo.reserve.size += tmp;
255 		excess_freed -= tmp;
256 		balance = 1;
257 	}
258 
259 	/* Return memory from the excess pool to that pool */
260 	if (excess_freed)
261 		vio_cmo.excess.free += excess_freed;
262 
263 	if (balance)
264 		schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY);
265 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
266 }
267 
268 /**
269  * vio_cmo_entitlement_update - Manage system entitlement changes
270  *
271  * @new_entitlement: new system entitlement to attempt to accommodate
272  *
273  * Increases in entitlement will be used to fulfill the spare entitlement
274  * and the rest is given to the excess pool.  Decreases, if they are
275  * possible, come from the excess pool and from unused device entitlement
276  *
277  * Returns: 0 on success, -ENOMEM when change can not be made
278  */
279 int vio_cmo_entitlement_update(size_t new_entitlement)
280 {
281 	struct vio_dev *viodev;
282 	struct vio_cmo_dev_entry *dev_ent;
283 	unsigned long flags;
284 	size_t avail, delta, tmp;
285 
286 	spin_lock_irqsave(&vio_cmo.lock, flags);
287 
288 	/* Entitlement increases */
289 	if (new_entitlement > vio_cmo.entitled) {
290 		delta = new_entitlement - vio_cmo.entitled;
291 
292 		/* Fulfill spare allocation */
293 		if (vio_cmo.spare < VIO_CMO_MIN_ENT) {
294 			tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare));
295 			vio_cmo.spare += tmp;
296 			vio_cmo.reserve.size += tmp;
297 			delta -= tmp;
298 		}
299 
300 		/* Remaining new allocation goes to the excess pool */
301 		vio_cmo.entitled += delta;
302 		vio_cmo.excess.size += delta;
303 		vio_cmo.excess.free += delta;
304 
305 		goto out;
306 	}
307 
308 	/* Entitlement decreases */
309 	delta = vio_cmo.entitled - new_entitlement;
310 	avail = vio_cmo.excess.free;
311 
312 	/*
313 	 * Need to check how much unused entitlement each device can
314 	 * sacrifice to fulfill entitlement change.
315 	 */
316 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
317 		if (avail >= delta)
318 			break;
319 
320 		viodev = dev_ent->viodev;
321 		if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
322 		    (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
323 				avail += viodev->cmo.entitled -
324 				         max_t(size_t, viodev->cmo.allocated,
325 				               VIO_CMO_MIN_ENT);
326 	}
327 
328 	if (delta <= avail) {
329 		vio_cmo.entitled -= delta;
330 
331 		/* Take entitlement from the excess pool first */
332 		tmp = min(vio_cmo.excess.free, delta);
333 		vio_cmo.excess.size -= tmp;
334 		vio_cmo.excess.free -= tmp;
335 		delta -= tmp;
336 
337 		/*
338 		 * Remove all but VIO_CMO_MIN_ENT bytes from devices
339 		 * until entitlement change is served
340 		 */
341 		list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
342 			if (!delta)
343 				break;
344 
345 			viodev = dev_ent->viodev;
346 			tmp = 0;
347 			if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
348 			    (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
349 				tmp = viodev->cmo.entitled -
350 				      max_t(size_t, viodev->cmo.allocated,
351 				            VIO_CMO_MIN_ENT);
352 			viodev->cmo.entitled -= min(tmp, delta);
353 			delta -= min(tmp, delta);
354 		}
355 	} else {
356 		spin_unlock_irqrestore(&vio_cmo.lock, flags);
357 		return -ENOMEM;
358 	}
359 
360 out:
361 	schedule_delayed_work(&vio_cmo.balance_q, 0);
362 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
363 	return 0;
364 }
365 
366 /**
367  * vio_cmo_balance - Balance entitlement among devices
368  *
369  * @work: work queue structure for this operation
370  *
371  * Any system entitlement above the minimum needed for devices, or
372  * already allocated to devices, can be distributed to the devices.
373  * The list of devices is iterated through to recalculate the desired
374  * entitlement level and to determine how much entitlement above the
375  * minimum entitlement is allocated to devices.
376  *
377  * Small chunks of the available entitlement are given to devices until
378  * their requirements are fulfilled or there is no entitlement left to give.
379  * Upon completion sizes of the reserve and excess pools are calculated.
380  *
381  * The system minimum entitlement level is also recalculated here.
382  * Entitlement will be reserved for devices even after vio_bus_remove to
383  * accommodate reloading the driver.  The OF tree is walked to count the
384  * number of devices present and this will remove entitlement for devices
385  * that have actually left the system after having vio_bus_remove called.
386  */
387 static void vio_cmo_balance(struct work_struct *work)
388 {
389 	struct vio_cmo *cmo;
390 	struct vio_dev *viodev;
391 	struct vio_cmo_dev_entry *dev_ent;
392 	unsigned long flags;
393 	size_t avail = 0, level, chunk, need;
394 	int devcount = 0, fulfilled;
395 
396 	cmo = container_of(work, struct vio_cmo, balance_q.work);
397 
398 	spin_lock_irqsave(&vio_cmo.lock, flags);
399 
400 	/* Calculate minimum entitlement and fulfill spare */
401 	cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT;
402 	BUG_ON(cmo->min > cmo->entitled);
403 	cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min));
404 	cmo->min += cmo->spare;
405 	cmo->desired = cmo->min;
406 
407 	/*
408 	 * Determine how much entitlement is available and reset device
409 	 * entitlements
410 	 */
411 	avail = cmo->entitled - cmo->spare;
412 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
413 		viodev = dev_ent->viodev;
414 		devcount++;
415 		viodev->cmo.entitled = VIO_CMO_MIN_ENT;
416 		cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT);
417 		avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT);
418 	}
419 
420 	/*
421 	 * Having provided each device with the minimum entitlement, loop
422 	 * over the devices portioning out the remaining entitlement
423 	 * until there is nothing left.
424 	 */
425 	level = VIO_CMO_MIN_ENT;
426 	while (avail) {
427 		fulfilled = 0;
428 		list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
429 			viodev = dev_ent->viodev;
430 
431 			if (viodev->cmo.desired <= level) {
432 				fulfilled++;
433 				continue;
434 			}
435 
436 			/*
437 			 * Give the device up to VIO_CMO_BALANCE_CHUNK
438 			 * bytes of entitlement, but do not exceed the
439 			 * desired level of entitlement for the device.
440 			 */
441 			chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK);
442 			chunk = min(chunk, (viodev->cmo.desired -
443 			                    viodev->cmo.entitled));
444 			viodev->cmo.entitled += chunk;
445 
446 			/*
447 			 * If the memory for this entitlement increase was
448 			 * already allocated to the device it does not come
449 			 * from the available pool being portioned out.
450 			 */
451 			need = max(viodev->cmo.allocated, viodev->cmo.entitled)-
452 			       max(viodev->cmo.allocated, level);
453 			avail -= need;
454 
455 		}
456 		if (fulfilled == devcount)
457 			break;
458 		level += VIO_CMO_BALANCE_CHUNK;
459 	}
460 
461 	/* Calculate new reserve and excess pool sizes */
462 	cmo->reserve.size = cmo->min;
463 	cmo->excess.free = 0;
464 	cmo->excess.size = 0;
465 	need = 0;
466 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
467 		viodev = dev_ent->viodev;
468 		/* Calculated reserve size above the minimum entitlement */
469 		if (viodev->cmo.entitled)
470 			cmo->reserve.size += (viodev->cmo.entitled -
471 			                      VIO_CMO_MIN_ENT);
472 		/* Calculated used excess entitlement */
473 		if (viodev->cmo.allocated > viodev->cmo.entitled)
474 			need += viodev->cmo.allocated - viodev->cmo.entitled;
475 	}
476 	cmo->excess.size = cmo->entitled - cmo->reserve.size;
477 	cmo->excess.free = cmo->excess.size - need;
478 
479 	cancel_delayed_work(to_delayed_work(work));
480 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
481 }
482 
483 static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size,
484 					  dma_addr_t *dma_handle, gfp_t flag,
485 					  unsigned long attrs)
486 {
487 	struct vio_dev *viodev = to_vio_dev(dev);
488 	void *ret;
489 
490 	if (vio_cmo_alloc(viodev, roundup(size, PAGE_SIZE))) {
491 		atomic_inc(&viodev->cmo.allocs_failed);
492 		return NULL;
493 	}
494 
495 	ret = dma_iommu_ops.alloc(dev, size, dma_handle, flag, attrs);
496 	if (unlikely(ret == NULL)) {
497 		vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
498 		atomic_inc(&viodev->cmo.allocs_failed);
499 	}
500 
501 	return ret;
502 }
503 
504 static void vio_dma_iommu_free_coherent(struct device *dev, size_t size,
505 					void *vaddr, dma_addr_t dma_handle,
506 					unsigned long attrs)
507 {
508 	struct vio_dev *viodev = to_vio_dev(dev);
509 
510 	dma_iommu_ops.free(dev, size, vaddr, dma_handle, attrs);
511 
512 	vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
513 }
514 
515 static dma_addr_t vio_dma_iommu_map_page(struct device *dev, struct page *page,
516                                          unsigned long offset, size_t size,
517                                          enum dma_data_direction direction,
518                                          unsigned long attrs)
519 {
520 	struct vio_dev *viodev = to_vio_dev(dev);
521 	struct iommu_table *tbl;
522 	dma_addr_t ret = IOMMU_MAPPING_ERROR;
523 
524 	tbl = get_iommu_table_base(dev);
525 	if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)))) {
526 		atomic_inc(&viodev->cmo.allocs_failed);
527 		return ret;
528 	}
529 
530 	ret = dma_iommu_ops.map_page(dev, page, offset, size, direction, attrs);
531 	if (unlikely(dma_mapping_error(dev, ret))) {
532 		vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)));
533 		atomic_inc(&viodev->cmo.allocs_failed);
534 	}
535 
536 	return ret;
537 }
538 
539 static void vio_dma_iommu_unmap_page(struct device *dev, dma_addr_t dma_handle,
540 				     size_t size,
541 				     enum dma_data_direction direction,
542 				     unsigned long attrs)
543 {
544 	struct vio_dev *viodev = to_vio_dev(dev);
545 	struct iommu_table *tbl;
546 
547 	tbl = get_iommu_table_base(dev);
548 	dma_iommu_ops.unmap_page(dev, dma_handle, size, direction, attrs);
549 
550 	vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)));
551 }
552 
553 static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
554                                 int nelems, enum dma_data_direction direction,
555                                 unsigned long attrs)
556 {
557 	struct vio_dev *viodev = to_vio_dev(dev);
558 	struct iommu_table *tbl;
559 	struct scatterlist *sgl;
560 	int ret, count;
561 	size_t alloc_size = 0;
562 
563 	tbl = get_iommu_table_base(dev);
564 	for_each_sg(sglist, sgl, nelems, count)
565 		alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE(tbl));
566 
567 	if (vio_cmo_alloc(viodev, alloc_size)) {
568 		atomic_inc(&viodev->cmo.allocs_failed);
569 		return 0;
570 	}
571 
572 	ret = dma_iommu_ops.map_sg(dev, sglist, nelems, direction, attrs);
573 
574 	if (unlikely(!ret)) {
575 		vio_cmo_dealloc(viodev, alloc_size);
576 		atomic_inc(&viodev->cmo.allocs_failed);
577 		return ret;
578 	}
579 
580 	for_each_sg(sglist, sgl, ret, count)
581 		alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE(tbl));
582 	if (alloc_size)
583 		vio_cmo_dealloc(viodev, alloc_size);
584 
585 	return ret;
586 }
587 
588 static void vio_dma_iommu_unmap_sg(struct device *dev,
589 		struct scatterlist *sglist, int nelems,
590 		enum dma_data_direction direction,
591 		unsigned long attrs)
592 {
593 	struct vio_dev *viodev = to_vio_dev(dev);
594 	struct iommu_table *tbl;
595 	struct scatterlist *sgl;
596 	size_t alloc_size = 0;
597 	int count;
598 
599 	tbl = get_iommu_table_base(dev);
600 	for_each_sg(sglist, sgl, nelems, count)
601 		alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE(tbl));
602 
603 	dma_iommu_ops.unmap_sg(dev, sglist, nelems, direction, attrs);
604 
605 	vio_cmo_dealloc(viodev, alloc_size);
606 }
607 
608 static int vio_dma_iommu_dma_supported(struct device *dev, u64 mask)
609 {
610         return dma_iommu_ops.dma_supported(dev, mask);
611 }
612 
613 static u64 vio_dma_get_required_mask(struct device *dev)
614 {
615         return dma_iommu_ops.get_required_mask(dev);
616 }
617 
618 static const struct dma_map_ops vio_dma_mapping_ops = {
619 	.alloc             = vio_dma_iommu_alloc_coherent,
620 	.free              = vio_dma_iommu_free_coherent,
621 	.mmap		   = dma_nommu_mmap_coherent,
622 	.map_sg            = vio_dma_iommu_map_sg,
623 	.unmap_sg          = vio_dma_iommu_unmap_sg,
624 	.map_page          = vio_dma_iommu_map_page,
625 	.unmap_page        = vio_dma_iommu_unmap_page,
626 	.dma_supported     = vio_dma_iommu_dma_supported,
627 	.get_required_mask = vio_dma_get_required_mask,
628 	.mapping_error	   = dma_iommu_mapping_error,
629 };
630 
631 /**
632  * vio_cmo_set_dev_desired - Set desired entitlement for a device
633  *
634  * @viodev: struct vio_dev for device to alter
635  * @desired: new desired entitlement level in bytes
636  *
637  * For use by devices to request a change to their entitlement at runtime or
638  * through sysfs.  The desired entitlement level is changed and a balancing
639  * of system resources is scheduled to run in the future.
640  */
641 void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
642 {
643 	unsigned long flags;
644 	struct vio_cmo_dev_entry *dev_ent;
645 	int found = 0;
646 
647 	if (!firmware_has_feature(FW_FEATURE_CMO))
648 		return;
649 
650 	spin_lock_irqsave(&vio_cmo.lock, flags);
651 	if (desired < VIO_CMO_MIN_ENT)
652 		desired = VIO_CMO_MIN_ENT;
653 
654 	/*
655 	 * Changes will not be made for devices not in the device list.
656 	 * If it is not in the device list, then no driver is loaded
657 	 * for the device and it can not receive entitlement.
658 	 */
659 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
660 		if (viodev == dev_ent->viodev) {
661 			found = 1;
662 			break;
663 		}
664 	if (!found) {
665 		spin_unlock_irqrestore(&vio_cmo.lock, flags);
666 		return;
667 	}
668 
669 	/* Increase/decrease in desired device entitlement */
670 	if (desired >= viodev->cmo.desired) {
671 		/* Just bump the bus and device values prior to a balance*/
672 		vio_cmo.desired += desired - viodev->cmo.desired;
673 		viodev->cmo.desired = desired;
674 	} else {
675 		/* Decrease bus and device values for desired entitlement */
676 		vio_cmo.desired -= viodev->cmo.desired - desired;
677 		viodev->cmo.desired = desired;
678 		/*
679 		 * If less entitlement is desired than current entitlement, move
680 		 * any reserve memory in the change region to the excess pool.
681 		 */
682 		if (viodev->cmo.entitled > desired) {
683 			vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
684 			vio_cmo.excess.size += viodev->cmo.entitled - desired;
685 			/*
686 			 * If entitlement moving from the reserve pool to the
687 			 * excess pool is currently unused, add to the excess
688 			 * free counter.
689 			 */
690 			if (viodev->cmo.allocated < viodev->cmo.entitled)
691 				vio_cmo.excess.free += viodev->cmo.entitled -
692 				                       max(viodev->cmo.allocated, desired);
693 			viodev->cmo.entitled = desired;
694 		}
695 	}
696 	schedule_delayed_work(&vio_cmo.balance_q, 0);
697 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
698 }
699 
700 /**
701  * vio_cmo_bus_probe - Handle CMO specific bus probe activities
702  *
703  * @viodev - Pointer to struct vio_dev for device
704  *
705  * Determine the devices IO memory entitlement needs, attempting
706  * to satisfy the system minimum entitlement at first and scheduling
707  * a balance operation to take care of the rest at a later time.
708  *
709  * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
710  *          -ENOMEM when entitlement is not available for device or
711  *          device entry.
712  *
713  */
714 static int vio_cmo_bus_probe(struct vio_dev *viodev)
715 {
716 	struct vio_cmo_dev_entry *dev_ent;
717 	struct device *dev = &viodev->dev;
718 	struct iommu_table *tbl;
719 	struct vio_driver *viodrv = to_vio_driver(dev->driver);
720 	unsigned long flags;
721 	size_t size;
722 	bool dma_capable = false;
723 
724 	tbl = get_iommu_table_base(dev);
725 
726 	/* A device requires entitlement if it has a DMA window property */
727 	switch (viodev->family) {
728 	case VDEVICE:
729 		if (of_get_property(viodev->dev.of_node,
730 					"ibm,my-dma-window", NULL))
731 			dma_capable = true;
732 		break;
733 	case PFO:
734 		dma_capable = false;
735 		break;
736 	default:
737 		dev_warn(dev, "unknown device family: %d\n", viodev->family);
738 		BUG();
739 		break;
740 	}
741 
742 	/* Configure entitlement for the device. */
743 	if (dma_capable) {
744 		/* Check that the driver is CMO enabled and get desired DMA */
745 		if (!viodrv->get_desired_dma) {
746 			dev_err(dev, "%s: device driver does not support CMO\n",
747 			        __func__);
748 			return -EINVAL;
749 		}
750 
751 		viodev->cmo.desired =
752 			IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev), tbl);
753 		if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
754 			viodev->cmo.desired = VIO_CMO_MIN_ENT;
755 		size = VIO_CMO_MIN_ENT;
756 
757 		dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
758 		                  GFP_KERNEL);
759 		if (!dev_ent)
760 			return -ENOMEM;
761 
762 		dev_ent->viodev = viodev;
763 		spin_lock_irqsave(&vio_cmo.lock, flags);
764 		list_add(&dev_ent->list, &vio_cmo.device_list);
765 	} else {
766 		viodev->cmo.desired = 0;
767 		size = 0;
768 		spin_lock_irqsave(&vio_cmo.lock, flags);
769 	}
770 
771 	/*
772 	 * If the needs for vio_cmo.min have not changed since they
773 	 * were last set, the number of devices in the OF tree has
774 	 * been constant and the IO memory for this is already in
775 	 * the reserve pool.
776 	 */
777 	if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
778 	                    VIO_CMO_MIN_ENT)) {
779 		/* Updated desired entitlement if device requires it */
780 		if (size)
781 			vio_cmo.desired += (viodev->cmo.desired -
782 		                        VIO_CMO_MIN_ENT);
783 	} else {
784 		size_t tmp;
785 
786 		tmp = vio_cmo.spare + vio_cmo.excess.free;
787 		if (tmp < size) {
788 			dev_err(dev, "%s: insufficient free "
789 			        "entitlement to add device. "
790 			        "Need %lu, have %lu\n", __func__,
791 				size, (vio_cmo.spare + tmp));
792 			spin_unlock_irqrestore(&vio_cmo.lock, flags);
793 			return -ENOMEM;
794 		}
795 
796 		/* Use excess pool first to fulfill request */
797 		tmp = min(size, vio_cmo.excess.free);
798 		vio_cmo.excess.free -= tmp;
799 		vio_cmo.excess.size -= tmp;
800 		vio_cmo.reserve.size += tmp;
801 
802 		/* Use spare if excess pool was insufficient */
803 		vio_cmo.spare -= size - tmp;
804 
805 		/* Update bus accounting */
806 		vio_cmo.min += size;
807 		vio_cmo.desired += viodev->cmo.desired;
808 	}
809 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
810 	return 0;
811 }
812 
813 /**
814  * vio_cmo_bus_remove - Handle CMO specific bus removal activities
815  *
816  * @viodev - Pointer to struct vio_dev for device
817  *
818  * Remove the device from the cmo device list.  The minimum entitlement
819  * will be reserved for the device as long as it is in the system.  The
820  * rest of the entitlement the device had been allocated will be returned
821  * to the system.
822  */
823 static void vio_cmo_bus_remove(struct vio_dev *viodev)
824 {
825 	struct vio_cmo_dev_entry *dev_ent;
826 	unsigned long flags;
827 	size_t tmp;
828 
829 	spin_lock_irqsave(&vio_cmo.lock, flags);
830 	if (viodev->cmo.allocated) {
831 		dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
832 		        "allocated after remove operation.\n",
833 		        __func__, viodev->cmo.allocated);
834 		BUG();
835 	}
836 
837 	/*
838 	 * Remove the device from the device list being maintained for
839 	 * CMO enabled devices.
840 	 */
841 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
842 		if (viodev == dev_ent->viodev) {
843 			list_del(&dev_ent->list);
844 			kfree(dev_ent);
845 			break;
846 		}
847 
848 	/*
849 	 * Devices may not require any entitlement and they do not need
850 	 * to be processed.  Otherwise, return the device's entitlement
851 	 * back to the pools.
852 	 */
853 	if (viodev->cmo.entitled) {
854 		/*
855 		 * This device has not yet left the OF tree, it's
856 		 * minimum entitlement remains in vio_cmo.min and
857 		 * vio_cmo.desired
858 		 */
859 		vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
860 
861 		/*
862 		 * Save min allocation for device in reserve as long
863 		 * as it exists in OF tree as determined by later
864 		 * balance operation
865 		 */
866 		viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
867 
868 		/* Replenish spare from freed reserve pool */
869 		if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
870 			tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
871 			                                 vio_cmo.spare));
872 			vio_cmo.spare += tmp;
873 			viodev->cmo.entitled -= tmp;
874 		}
875 
876 		/* Remaining reserve goes to excess pool */
877 		vio_cmo.excess.size += viodev->cmo.entitled;
878 		vio_cmo.excess.free += viodev->cmo.entitled;
879 		vio_cmo.reserve.size -= viodev->cmo.entitled;
880 
881 		/*
882 		 * Until the device is removed it will keep a
883 		 * minimum entitlement; this will guarantee that
884 		 * a module unload/load will result in a success.
885 		 */
886 		viodev->cmo.entitled = VIO_CMO_MIN_ENT;
887 		viodev->cmo.desired = VIO_CMO_MIN_ENT;
888 		atomic_set(&viodev->cmo.allocs_failed, 0);
889 	}
890 
891 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
892 }
893 
894 static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
895 {
896 	set_dma_ops(&viodev->dev, &vio_dma_mapping_ops);
897 }
898 
899 /**
900  * vio_cmo_bus_init - CMO entitlement initialization at bus init time
901  *
902  * Set up the reserve and excess entitlement pools based on available
903  * system entitlement and the number of devices in the OF tree that
904  * require entitlement in the reserve pool.
905  */
906 static void vio_cmo_bus_init(void)
907 {
908 	struct hvcall_mpp_data mpp_data;
909 	int err;
910 
911 	memset(&vio_cmo, 0, sizeof(struct vio_cmo));
912 	spin_lock_init(&vio_cmo.lock);
913 	INIT_LIST_HEAD(&vio_cmo.device_list);
914 	INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
915 
916 	/* Get current system entitlement */
917 	err = h_get_mpp(&mpp_data);
918 
919 	/*
920 	 * On failure, continue with entitlement set to 0, will panic()
921 	 * later when spare is reserved.
922 	 */
923 	if (err != H_SUCCESS) {
924 		printk(KERN_ERR "%s: unable to determine system IO "\
925 		       "entitlement. (%d)\n", __func__, err);
926 		vio_cmo.entitled = 0;
927 	} else {
928 		vio_cmo.entitled = mpp_data.entitled_mem;
929 	}
930 
931 	/* Set reservation and check against entitlement */
932 	vio_cmo.spare = VIO_CMO_MIN_ENT;
933 	vio_cmo.reserve.size = vio_cmo.spare;
934 	vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
935 	                         VIO_CMO_MIN_ENT);
936 	if (vio_cmo.reserve.size > vio_cmo.entitled) {
937 		printk(KERN_ERR "%s: insufficient system entitlement\n",
938 		       __func__);
939 		panic("%s: Insufficient system entitlement", __func__);
940 	}
941 
942 	/* Set the remaining accounting variables */
943 	vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
944 	vio_cmo.excess.free = vio_cmo.excess.size;
945 	vio_cmo.min = vio_cmo.reserve.size;
946 	vio_cmo.desired = vio_cmo.reserve.size;
947 }
948 
949 /* sysfs device functions and data structures for CMO */
950 
951 #define viodev_cmo_rd_attr(name)                                        \
952 static ssize_t cmo_##name##_show(struct device *dev,                    \
953                                         struct device_attribute *attr,  \
954                                          char *buf)                     \
955 {                                                                       \
956 	return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name);        \
957 }
958 
959 static ssize_t cmo_allocs_failed_show(struct device *dev,
960 		struct device_attribute *attr, char *buf)
961 {
962 	struct vio_dev *viodev = to_vio_dev(dev);
963 	return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
964 }
965 
966 static ssize_t cmo_allocs_failed_store(struct device *dev,
967 		struct device_attribute *attr, const char *buf, size_t count)
968 {
969 	struct vio_dev *viodev = to_vio_dev(dev);
970 	atomic_set(&viodev->cmo.allocs_failed, 0);
971 	return count;
972 }
973 
974 static ssize_t cmo_desired_store(struct device *dev,
975 		struct device_attribute *attr, const char *buf, size_t count)
976 {
977 	struct vio_dev *viodev = to_vio_dev(dev);
978 	size_t new_desired;
979 	int ret;
980 
981 	ret = kstrtoul(buf, 10, &new_desired);
982 	if (ret)
983 		return ret;
984 
985 	vio_cmo_set_dev_desired(viodev, new_desired);
986 	return count;
987 }
988 
989 viodev_cmo_rd_attr(desired);
990 viodev_cmo_rd_attr(entitled);
991 viodev_cmo_rd_attr(allocated);
992 
993 static ssize_t name_show(struct device *, struct device_attribute *, char *);
994 static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
995 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
996 			     char *buf);
997 
998 static struct device_attribute dev_attr_name;
999 static struct device_attribute dev_attr_devspec;
1000 static struct device_attribute dev_attr_modalias;
1001 
1002 static DEVICE_ATTR_RO(cmo_entitled);
1003 static DEVICE_ATTR_RO(cmo_allocated);
1004 static DEVICE_ATTR_RW(cmo_desired);
1005 static DEVICE_ATTR_RW(cmo_allocs_failed);
1006 
1007 static struct attribute *vio_cmo_dev_attrs[] = {
1008 	&dev_attr_name.attr,
1009 	&dev_attr_devspec.attr,
1010 	&dev_attr_modalias.attr,
1011 	&dev_attr_cmo_entitled.attr,
1012 	&dev_attr_cmo_allocated.attr,
1013 	&dev_attr_cmo_desired.attr,
1014 	&dev_attr_cmo_allocs_failed.attr,
1015 	NULL,
1016 };
1017 ATTRIBUTE_GROUPS(vio_cmo_dev);
1018 
1019 /* sysfs bus functions and data structures for CMO */
1020 
1021 #define viobus_cmo_rd_attr(name)                                        \
1022 static ssize_t cmo_bus_##name##_show(struct bus_type *bt, char *buf)    \
1023 {                                                                       \
1024 	return sprintf(buf, "%lu\n", vio_cmo.name);                     \
1025 }                                                                       \
1026 static struct bus_attribute bus_attr_cmo_bus_##name =			\
1027 	__ATTR(cmo_##name, S_IRUGO, cmo_bus_##name##_show, NULL)
1028 
1029 #define viobus_cmo_pool_rd_attr(name, var)                              \
1030 static ssize_t                                                          \
1031 cmo_##name##_##var##_show(struct bus_type *bt, char *buf)               \
1032 {                                                                       \
1033 	return sprintf(buf, "%lu\n", vio_cmo.name.var);                 \
1034 }                                                                       \
1035 static BUS_ATTR_RO(cmo_##name##_##var)
1036 
1037 viobus_cmo_rd_attr(entitled);
1038 viobus_cmo_rd_attr(spare);
1039 viobus_cmo_rd_attr(min);
1040 viobus_cmo_rd_attr(desired);
1041 viobus_cmo_rd_attr(curr);
1042 viobus_cmo_pool_rd_attr(reserve, size);
1043 viobus_cmo_pool_rd_attr(excess, size);
1044 viobus_cmo_pool_rd_attr(excess, free);
1045 
1046 static ssize_t cmo_high_show(struct bus_type *bt, char *buf)
1047 {
1048 	return sprintf(buf, "%lu\n", vio_cmo.high);
1049 }
1050 
1051 static ssize_t cmo_high_store(struct bus_type *bt, const char *buf,
1052 			      size_t count)
1053 {
1054 	unsigned long flags;
1055 
1056 	spin_lock_irqsave(&vio_cmo.lock, flags);
1057 	vio_cmo.high = vio_cmo.curr;
1058 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
1059 
1060 	return count;
1061 }
1062 static BUS_ATTR_RW(cmo_high);
1063 
1064 static struct attribute *vio_bus_attrs[] = {
1065 	&bus_attr_cmo_bus_entitled.attr,
1066 	&bus_attr_cmo_bus_spare.attr,
1067 	&bus_attr_cmo_bus_min.attr,
1068 	&bus_attr_cmo_bus_desired.attr,
1069 	&bus_attr_cmo_bus_curr.attr,
1070 	&bus_attr_cmo_high.attr,
1071 	&bus_attr_cmo_reserve_size.attr,
1072 	&bus_attr_cmo_excess_size.attr,
1073 	&bus_attr_cmo_excess_free.attr,
1074 	NULL,
1075 };
1076 ATTRIBUTE_GROUPS(vio_bus);
1077 
1078 static void vio_cmo_sysfs_init(void)
1079 {
1080 	vio_bus_type.dev_groups = vio_cmo_dev_groups;
1081 	vio_bus_type.bus_groups = vio_bus_groups;
1082 }
1083 #else /* CONFIG_PPC_SMLPAR */
1084 int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1085 void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1086 static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1087 static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1088 static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
1089 static void vio_cmo_bus_init(void) {}
1090 static void vio_cmo_sysfs_init(void) { }
1091 #endif /* CONFIG_PPC_SMLPAR */
1092 EXPORT_SYMBOL(vio_cmo_entitlement_update);
1093 EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1094 
1095 
1096 /*
1097  * Platform Facilities Option (PFO) support
1098  */
1099 
1100 /**
1101  * vio_h_cop_sync - Perform a synchronous PFO co-processor operation
1102  *
1103  * @vdev - Pointer to a struct vio_dev for device
1104  * @op - Pointer to a struct vio_pfo_op for the operation parameters
1105  *
1106  * Calls the hypervisor to synchronously perform the PFO operation
1107  * described in @op.  In the case of a busy response from the hypervisor,
1108  * the operation will be re-submitted indefinitely unless a non-zero timeout
1109  * is specified or an error occurs. The timeout places a limit on when to
1110  * stop re-submitting a operation, the total time can be exceeded if an
1111  * operation is in progress.
1112  *
1113  * If op->hcall_ret is not NULL, this will be set to the return from the
1114  * last h_cop_op call or it will be 0 if an error not involving the h_call
1115  * was encountered.
1116  *
1117  * Returns:
1118  *	0 on success,
1119  *	-EINVAL if the h_call fails due to an invalid parameter,
1120  *	-E2BIG if the h_call can not be performed synchronously,
1121  *	-EBUSY if a timeout is specified and has elapsed,
1122  *	-EACCES if the memory area for data/status has been rescinded, or
1123  *	-EPERM if a hardware fault has been indicated
1124  */
1125 int vio_h_cop_sync(struct vio_dev *vdev, struct vio_pfo_op *op)
1126 {
1127 	struct device *dev = &vdev->dev;
1128 	unsigned long deadline = 0;
1129 	long hret = 0;
1130 	int ret = 0;
1131 
1132 	if (op->timeout)
1133 		deadline = jiffies + msecs_to_jiffies(op->timeout);
1134 
1135 	while (true) {
1136 		hret = plpar_hcall_norets(H_COP, op->flags,
1137 				vdev->resource_id,
1138 				op->in, op->inlen, op->out,
1139 				op->outlen, op->csbcpb);
1140 
1141 		if (hret == H_SUCCESS ||
1142 		    (hret != H_NOT_ENOUGH_RESOURCES &&
1143 		     hret != H_BUSY && hret != H_RESOURCE) ||
1144 		    (op->timeout && time_after(deadline, jiffies)))
1145 			break;
1146 
1147 		dev_dbg(dev, "%s: hcall ret(%ld), retrying.\n", __func__, hret);
1148 	}
1149 
1150 	switch (hret) {
1151 	case H_SUCCESS:
1152 		ret = 0;
1153 		break;
1154 	case H_OP_MODE:
1155 	case H_TOO_BIG:
1156 		ret = -E2BIG;
1157 		break;
1158 	case H_RESCINDED:
1159 		ret = -EACCES;
1160 		break;
1161 	case H_HARDWARE:
1162 		ret = -EPERM;
1163 		break;
1164 	case H_NOT_ENOUGH_RESOURCES:
1165 	case H_RESOURCE:
1166 	case H_BUSY:
1167 		ret = -EBUSY;
1168 		break;
1169 	default:
1170 		ret = -EINVAL;
1171 		break;
1172 	}
1173 
1174 	if (ret)
1175 		dev_dbg(dev, "%s: Sync h_cop_op failure (ret:%d) (hret:%ld)\n",
1176 				__func__, ret, hret);
1177 
1178 	op->hcall_err = hret;
1179 	return ret;
1180 }
1181 EXPORT_SYMBOL(vio_h_cop_sync);
1182 
1183 static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1184 {
1185 	const __be32 *dma_window;
1186 	struct iommu_table *tbl;
1187 	unsigned long offset, size;
1188 
1189 	dma_window = of_get_property(dev->dev.of_node,
1190 				  "ibm,my-dma-window", NULL);
1191 	if (!dma_window)
1192 		return NULL;
1193 
1194 	tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
1195 	if (tbl == NULL)
1196 		return NULL;
1197 
1198 	of_parse_dma_window(dev->dev.of_node, dma_window,
1199 			    &tbl->it_index, &offset, &size);
1200 
1201 	/* TCE table size - measured in tce entries */
1202 	tbl->it_page_shift = IOMMU_PAGE_SHIFT_4K;
1203 	tbl->it_size = size >> tbl->it_page_shift;
1204 	/* offset for VIO should always be 0 */
1205 	tbl->it_offset = offset >> tbl->it_page_shift;
1206 	tbl->it_busno = 0;
1207 	tbl->it_type = TCE_VB;
1208 	tbl->it_blocksize = 16;
1209 
1210 	if (firmware_has_feature(FW_FEATURE_LPAR))
1211 		tbl->it_ops = &iommu_table_lpar_multi_ops;
1212 	else
1213 		tbl->it_ops = &iommu_table_pseries_ops;
1214 
1215 	return iommu_init_table(tbl, -1);
1216 }
1217 
1218 /**
1219  * vio_match_device: - Tell if a VIO device has a matching
1220  *			VIO device id structure.
1221  * @ids:	array of VIO device id structures to search in
1222  * @dev:	the VIO device structure to match against
1223  *
1224  * Used by a driver to check whether a VIO device present in the
1225  * system is in its list of supported devices. Returns the matching
1226  * vio_device_id structure or NULL if there is no match.
1227  */
1228 static const struct vio_device_id *vio_match_device(
1229 		const struct vio_device_id *ids, const struct vio_dev *dev)
1230 {
1231 	while (ids->type[0] != '\0') {
1232 		if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
1233 		    of_device_is_compatible(dev->dev.of_node,
1234 					 ids->compat))
1235 			return ids;
1236 		ids++;
1237 	}
1238 	return NULL;
1239 }
1240 
1241 /*
1242  * Convert from struct device to struct vio_dev and pass to driver.
1243  * dev->driver has already been set by generic code because vio_bus_match
1244  * succeeded.
1245  */
1246 static int vio_bus_probe(struct device *dev)
1247 {
1248 	struct vio_dev *viodev = to_vio_dev(dev);
1249 	struct vio_driver *viodrv = to_vio_driver(dev->driver);
1250 	const struct vio_device_id *id;
1251 	int error = -ENODEV;
1252 
1253 	if (!viodrv->probe)
1254 		return error;
1255 
1256 	id = vio_match_device(viodrv->id_table, viodev);
1257 	if (id) {
1258 		memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1259 		if (firmware_has_feature(FW_FEATURE_CMO)) {
1260 			error = vio_cmo_bus_probe(viodev);
1261 			if (error)
1262 				return error;
1263 		}
1264 		error = viodrv->probe(viodev, id);
1265 		if (error && firmware_has_feature(FW_FEATURE_CMO))
1266 			vio_cmo_bus_remove(viodev);
1267 	}
1268 
1269 	return error;
1270 }
1271 
1272 /* convert from struct device to struct vio_dev and pass to driver. */
1273 static int vio_bus_remove(struct device *dev)
1274 {
1275 	struct vio_dev *viodev = to_vio_dev(dev);
1276 	struct vio_driver *viodrv = to_vio_driver(dev->driver);
1277 	struct device *devptr;
1278 	int ret = 1;
1279 
1280 	/*
1281 	 * Hold a reference to the device after the remove function is called
1282 	 * to allow for CMO accounting cleanup for the device.
1283 	 */
1284 	devptr = get_device(dev);
1285 
1286 	if (viodrv->remove)
1287 		ret = viodrv->remove(viodev);
1288 
1289 	if (!ret && firmware_has_feature(FW_FEATURE_CMO))
1290 		vio_cmo_bus_remove(viodev);
1291 
1292 	put_device(devptr);
1293 	return ret;
1294 }
1295 
1296 /**
1297  * vio_register_driver: - Register a new vio driver
1298  * @viodrv:	The vio_driver structure to be registered.
1299  */
1300 int __vio_register_driver(struct vio_driver *viodrv, struct module *owner,
1301 			  const char *mod_name)
1302 {
1303 	pr_debug("%s: driver %s registering\n", __func__, viodrv->name);
1304 
1305 	/* fill in 'struct driver' fields */
1306 	viodrv->driver.name = viodrv->name;
1307 	viodrv->driver.pm = viodrv->pm;
1308 	viodrv->driver.bus = &vio_bus_type;
1309 	viodrv->driver.owner = owner;
1310 	viodrv->driver.mod_name = mod_name;
1311 
1312 	return driver_register(&viodrv->driver);
1313 }
1314 EXPORT_SYMBOL(__vio_register_driver);
1315 
1316 /**
1317  * vio_unregister_driver - Remove registration of vio driver.
1318  * @viodrv:	The vio_driver struct to be removed form registration
1319  */
1320 void vio_unregister_driver(struct vio_driver *viodrv)
1321 {
1322 	driver_unregister(&viodrv->driver);
1323 }
1324 EXPORT_SYMBOL(vio_unregister_driver);
1325 
1326 /* vio_dev refcount hit 0 */
1327 static void vio_dev_release(struct device *dev)
1328 {
1329 	struct iommu_table *tbl = get_iommu_table_base(dev);
1330 
1331 	if (tbl)
1332 		iommu_tce_table_put(tbl);
1333 	of_node_put(dev->of_node);
1334 	kfree(to_vio_dev(dev));
1335 }
1336 
1337 /**
1338  * vio_register_device_node: - Register a new vio device.
1339  * @of_node:	The OF node for this device.
1340  *
1341  * Creates and initializes a vio_dev structure from the data in
1342  * of_node and adds it to the list of virtual devices.
1343  * Returns a pointer to the created vio_dev or NULL if node has
1344  * NULL device_type or compatible fields.
1345  */
1346 struct vio_dev *vio_register_device_node(struct device_node *of_node)
1347 {
1348 	struct vio_dev *viodev;
1349 	struct device_node *parent_node;
1350 	const __be32 *prop;
1351 	enum vio_dev_family family;
1352 
1353 	/*
1354 	 * Determine if this node is a under the /vdevice node or under the
1355 	 * /ibm,platform-facilities node.  This decides the device's family.
1356 	 */
1357 	parent_node = of_get_parent(of_node);
1358 	if (parent_node) {
1359 		if (!strcmp(parent_node->type, "ibm,platform-facilities"))
1360 			family = PFO;
1361 		else if (!strcmp(parent_node->type, "vdevice"))
1362 			family = VDEVICE;
1363 		else {
1364 			pr_warn("%s: parent(%pOF) of %pOFn not recognized.\n",
1365 					__func__,
1366 					parent_node,
1367 					of_node);
1368 			of_node_put(parent_node);
1369 			return NULL;
1370 		}
1371 		of_node_put(parent_node);
1372 	} else {
1373 		pr_warn("%s: could not determine the parent of node %pOFn.\n",
1374 				__func__, of_node);
1375 		return NULL;
1376 	}
1377 
1378 	if (family == PFO) {
1379 		if (of_get_property(of_node, "interrupt-controller", NULL)) {
1380 			pr_debug("%s: Skipping the interrupt controller %pOFn.\n",
1381 					__func__, of_node);
1382 			return NULL;
1383 		}
1384 	}
1385 
1386 	/* allocate a vio_dev for this node */
1387 	viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1388 	if (viodev == NULL) {
1389 		pr_warn("%s: allocation failure for VIO device.\n", __func__);
1390 		return NULL;
1391 	}
1392 
1393 	/* we need the 'device_type' property, in order to match with drivers */
1394 	viodev->family = family;
1395 	if (viodev->family == VDEVICE) {
1396 		unsigned int unit_address;
1397 
1398 		if (of_node->type != NULL)
1399 			viodev->type = of_node->type;
1400 		else {
1401 			pr_warn("%s: node %pOFn is missing the 'device_type' "
1402 					"property.\n", __func__, of_node);
1403 			goto out;
1404 		}
1405 
1406 		prop = of_get_property(of_node, "reg", NULL);
1407 		if (prop == NULL) {
1408 			pr_warn("%s: node %pOFn missing 'reg'\n",
1409 					__func__, of_node);
1410 			goto out;
1411 		}
1412 		unit_address = of_read_number(prop, 1);
1413 		dev_set_name(&viodev->dev, "%x", unit_address);
1414 		viodev->irq = irq_of_parse_and_map(of_node, 0);
1415 		viodev->unit_address = unit_address;
1416 	} else {
1417 		/* PFO devices need their resource_id for submitting COP_OPs
1418 		 * This is an optional field for devices, but is required when
1419 		 * performing synchronous ops */
1420 		prop = of_get_property(of_node, "ibm,resource-id", NULL);
1421 		if (prop != NULL)
1422 			viodev->resource_id = of_read_number(prop, 1);
1423 
1424 		dev_set_name(&viodev->dev, "%pOFn", of_node);
1425 		viodev->type = dev_name(&viodev->dev);
1426 		viodev->irq = 0;
1427 	}
1428 
1429 	viodev->name = of_node->name;
1430 	viodev->dev.of_node = of_node_get(of_node);
1431 
1432 	set_dev_node(&viodev->dev, of_node_to_nid(of_node));
1433 
1434 	/* init generic 'struct device' fields: */
1435 	viodev->dev.parent = &vio_bus_device.dev;
1436 	viodev->dev.bus = &vio_bus_type;
1437 	viodev->dev.release = vio_dev_release;
1438 
1439 	if (of_get_property(viodev->dev.of_node, "ibm,my-dma-window", NULL)) {
1440 		if (firmware_has_feature(FW_FEATURE_CMO))
1441 			vio_cmo_set_dma_ops(viodev);
1442 		else
1443 			set_dma_ops(&viodev->dev, &dma_iommu_ops);
1444 
1445 		set_iommu_table_base(&viodev->dev,
1446 				     vio_build_iommu_table(viodev));
1447 
1448 		/* needed to ensure proper operation of coherent allocations
1449 		 * later, in case driver doesn't set it explicitly */
1450 		viodev->dev.coherent_dma_mask = DMA_BIT_MASK(64);
1451 		viodev->dev.dma_mask = &viodev->dev.coherent_dma_mask;
1452 	}
1453 
1454 	/* register with generic device framework */
1455 	if (device_register(&viodev->dev)) {
1456 		printk(KERN_ERR "%s: failed to register device %s\n",
1457 				__func__, dev_name(&viodev->dev));
1458 		put_device(&viodev->dev);
1459 		return NULL;
1460 	}
1461 
1462 	return viodev;
1463 
1464 out:	/* Use this exit point for any return prior to device_register */
1465 	kfree(viodev);
1466 
1467 	return NULL;
1468 }
1469 EXPORT_SYMBOL(vio_register_device_node);
1470 
1471 /*
1472  * vio_bus_scan_for_devices - Scan OF and register each child device
1473  * @root_name - OF node name for the root of the subtree to search.
1474  *		This must be non-NULL
1475  *
1476  * Starting from the root node provide, register the device node for
1477  * each child beneath the root.
1478  */
1479 static void vio_bus_scan_register_devices(char *root_name)
1480 {
1481 	struct device_node *node_root, *node_child;
1482 
1483 	if (!root_name)
1484 		return;
1485 
1486 	node_root = of_find_node_by_name(NULL, root_name);
1487 	if (node_root) {
1488 
1489 		/*
1490 		 * Create struct vio_devices for each virtual device in
1491 		 * the device tree. Drivers will associate with them later.
1492 		 */
1493 		node_child = of_get_next_child(node_root, NULL);
1494 		while (node_child) {
1495 			vio_register_device_node(node_child);
1496 			node_child = of_get_next_child(node_root, node_child);
1497 		}
1498 		of_node_put(node_root);
1499 	}
1500 }
1501 
1502 /**
1503  * vio_bus_init: - Initialize the virtual IO bus
1504  */
1505 static int __init vio_bus_init(void)
1506 {
1507 	int err;
1508 
1509 	if (firmware_has_feature(FW_FEATURE_CMO))
1510 		vio_cmo_sysfs_init();
1511 
1512 	err = bus_register(&vio_bus_type);
1513 	if (err) {
1514 		printk(KERN_ERR "failed to register VIO bus\n");
1515 		return err;
1516 	}
1517 
1518 	/*
1519 	 * The fake parent of all vio devices, just to give us
1520 	 * a nice directory
1521 	 */
1522 	err = device_register(&vio_bus_device.dev);
1523 	if (err) {
1524 		printk(KERN_WARNING "%s: device_register returned %i\n",
1525 				__func__, err);
1526 		return err;
1527 	}
1528 
1529 	if (firmware_has_feature(FW_FEATURE_CMO))
1530 		vio_cmo_bus_init();
1531 
1532 	return 0;
1533 }
1534 postcore_initcall(vio_bus_init);
1535 
1536 static int __init vio_device_init(void)
1537 {
1538 	vio_bus_scan_register_devices("vdevice");
1539 	vio_bus_scan_register_devices("ibm,platform-facilities");
1540 
1541 	return 0;
1542 }
1543 device_initcall(vio_device_init);
1544 
1545 static ssize_t name_show(struct device *dev,
1546 		struct device_attribute *attr, char *buf)
1547 {
1548 	return sprintf(buf, "%s\n", to_vio_dev(dev)->name);
1549 }
1550 static DEVICE_ATTR_RO(name);
1551 
1552 static ssize_t devspec_show(struct device *dev,
1553 		struct device_attribute *attr, char *buf)
1554 {
1555 	struct device_node *of_node = dev->of_node;
1556 
1557 	return sprintf(buf, "%pOF\n", of_node);
1558 }
1559 static DEVICE_ATTR_RO(devspec);
1560 
1561 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1562 			     char *buf)
1563 {
1564 	const struct vio_dev *vio_dev = to_vio_dev(dev);
1565 	struct device_node *dn;
1566 	const char *cp;
1567 
1568 	dn = dev->of_node;
1569 	if (!dn) {
1570 		strcpy(buf, "\n");
1571 		return strlen(buf);
1572 	}
1573 	cp = of_get_property(dn, "compatible", NULL);
1574 	if (!cp) {
1575 		strcpy(buf, "\n");
1576 		return strlen(buf);
1577 	}
1578 
1579 	return sprintf(buf, "vio:T%sS%s\n", vio_dev->type, cp);
1580 }
1581 static DEVICE_ATTR_RO(modalias);
1582 
1583 static struct attribute *vio_dev_attrs[] = {
1584 	&dev_attr_name.attr,
1585 	&dev_attr_devspec.attr,
1586 	&dev_attr_modalias.attr,
1587 	NULL,
1588 };
1589 ATTRIBUTE_GROUPS(vio_dev);
1590 
1591 void vio_unregister_device(struct vio_dev *viodev)
1592 {
1593 	device_unregister(&viodev->dev);
1594 	if (viodev->family == VDEVICE)
1595 		irq_dispose_mapping(viodev->irq);
1596 }
1597 EXPORT_SYMBOL(vio_unregister_device);
1598 
1599 static int vio_bus_match(struct device *dev, struct device_driver *drv)
1600 {
1601 	const struct vio_dev *vio_dev = to_vio_dev(dev);
1602 	struct vio_driver *vio_drv = to_vio_driver(drv);
1603 	const struct vio_device_id *ids = vio_drv->id_table;
1604 
1605 	return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL);
1606 }
1607 
1608 static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env)
1609 {
1610 	const struct vio_dev *vio_dev = to_vio_dev(dev);
1611 	struct device_node *dn;
1612 	const char *cp;
1613 
1614 	dn = dev->of_node;
1615 	if (!dn)
1616 		return -ENODEV;
1617 	cp = of_get_property(dn, "compatible", NULL);
1618 	if (!cp)
1619 		return -ENODEV;
1620 
1621 	add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
1622 	return 0;
1623 }
1624 
1625 struct bus_type vio_bus_type = {
1626 	.name = "vio",
1627 	.dev_groups = vio_dev_groups,
1628 	.uevent = vio_hotplug,
1629 	.match = vio_bus_match,
1630 	.probe = vio_bus_probe,
1631 	.remove = vio_bus_remove,
1632 };
1633 
1634 /**
1635  * vio_get_attribute: - get attribute for virtual device
1636  * @vdev:	The vio device to get property.
1637  * @which:	The property/attribute to be extracted.
1638  * @length:	Pointer to length of returned data size (unused if NULL).
1639  *
1640  * Calls prom.c's of_get_property() to return the value of the
1641  * attribute specified by @which
1642 */
1643 const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1644 {
1645 	return of_get_property(vdev->dev.of_node, which, length);
1646 }
1647 EXPORT_SYMBOL(vio_get_attribute);
1648 
1649 #ifdef CONFIG_PPC_PSERIES
1650 /* vio_find_name() - internal because only vio.c knows how we formatted the
1651  * kobject name
1652  */
1653 static struct vio_dev *vio_find_name(const char *name)
1654 {
1655 	struct device *found;
1656 
1657 	found = bus_find_device_by_name(&vio_bus_type, NULL, name);
1658 	if (!found)
1659 		return NULL;
1660 
1661 	return to_vio_dev(found);
1662 }
1663 
1664 /**
1665  * vio_find_node - find an already-registered vio_dev
1666  * @vnode: device_node of the virtual device we're looking for
1667  *
1668  * Takes a reference to the embedded struct device which needs to be dropped
1669  * after use.
1670  */
1671 struct vio_dev *vio_find_node(struct device_node *vnode)
1672 {
1673 	char kobj_name[20];
1674 	struct device_node *vnode_parent;
1675 	const char *dev_type;
1676 
1677 	vnode_parent = of_get_parent(vnode);
1678 	if (!vnode_parent)
1679 		return NULL;
1680 
1681 	dev_type = of_get_property(vnode_parent, "device_type", NULL);
1682 	of_node_put(vnode_parent);
1683 	if (!dev_type)
1684 		return NULL;
1685 
1686 	/* construct the kobject name from the device node */
1687 	if (!strcmp(dev_type, "vdevice")) {
1688 		const __be32 *prop;
1689 
1690 		prop = of_get_property(vnode, "reg", NULL);
1691 		if (!prop)
1692 			return NULL;
1693 		snprintf(kobj_name, sizeof(kobj_name), "%x",
1694 			 (uint32_t)of_read_number(prop, 1));
1695 	} else if (!strcmp(dev_type, "ibm,platform-facilities"))
1696 		snprintf(kobj_name, sizeof(kobj_name), "%pOFn", vnode);
1697 	else
1698 		return NULL;
1699 
1700 	return vio_find_name(kobj_name);
1701 }
1702 EXPORT_SYMBOL(vio_find_node);
1703 
1704 int vio_enable_interrupts(struct vio_dev *dev)
1705 {
1706 	int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1707 	if (rc != H_SUCCESS)
1708 		printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1709 	return rc;
1710 }
1711 EXPORT_SYMBOL(vio_enable_interrupts);
1712 
1713 int vio_disable_interrupts(struct vio_dev *dev)
1714 {
1715 	int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1716 	if (rc != H_SUCCESS)
1717 		printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1718 	return rc;
1719 }
1720 EXPORT_SYMBOL(vio_disable_interrupts);
1721 #endif /* CONFIG_PPC_PSERIES */
1722