1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * IBM PowerPC Virtual I/O Infrastructure Support.
4  *
5  *    Copyright (c) 2003,2008 IBM Corp.
6  *     Dave Engebretsen engebret@us.ibm.com
7  *     Santiago Leon santil@us.ibm.com
8  *     Hollis Blanchard <hollisb@us.ibm.com>
9  *     Stephen Rothwell
10  *     Robert Jennings <rcjenn@us.ibm.com>
11  */
12 
13 #include <linux/cpu.h>
14 #include <linux/types.h>
15 #include <linux/delay.h>
16 #include <linux/stat.h>
17 #include <linux/device.h>
18 #include <linux/init.h>
19 #include <linux/slab.h>
20 #include <linux/console.h>
21 #include <linux/export.h>
22 #include <linux/mm.h>
23 #include <linux/dma-map-ops.h>
24 #include <linux/kobject.h>
25 #include <linux/kexec.h>
26 
27 #include <asm/iommu.h>
28 #include <asm/dma.h>
29 #include <asm/vio.h>
30 #include <asm/prom.h>
31 #include <asm/firmware.h>
32 #include <asm/tce.h>
33 #include <asm/page.h>
34 #include <asm/hvcall.h>
35 #include <asm/machdep.h>
36 
37 static struct vio_dev vio_bus_device  = { /* fake "parent" device */
38 	.name = "vio",
39 	.type = "",
40 	.dev.init_name = "vio",
41 	.dev.bus = &vio_bus_type,
42 };
43 
44 #ifdef CONFIG_PPC_SMLPAR
45 /**
46  * vio_cmo_pool - A pool of IO memory for CMO use
47  *
48  * @size: The size of the pool in bytes
49  * @free: The amount of free memory in the pool
50  */
51 struct vio_cmo_pool {
52 	size_t size;
53 	size_t free;
54 };
55 
56 /* How many ms to delay queued balance work */
57 #define VIO_CMO_BALANCE_DELAY 100
58 
59 /* Portion out IO memory to CMO devices by this chunk size */
60 #define VIO_CMO_BALANCE_CHUNK 131072
61 
62 /**
63  * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement
64  *
65  * @vio_dev: struct vio_dev pointer
66  * @list: pointer to other devices on bus that are being tracked
67  */
68 struct vio_cmo_dev_entry {
69 	struct vio_dev *viodev;
70 	struct list_head list;
71 };
72 
73 /**
74  * vio_cmo - VIO bus accounting structure for CMO entitlement
75  *
76  * @lock: spinlock for entire structure
77  * @balance_q: work queue for balancing system entitlement
78  * @device_list: list of CMO-enabled devices requiring entitlement
79  * @entitled: total system entitlement in bytes
80  * @reserve: pool of memory from which devices reserve entitlement, incl. spare
81  * @excess: pool of excess entitlement not needed for device reserves or spare
82  * @spare: IO memory for device hotplug functionality
83  * @min: minimum necessary for system operation
84  * @desired: desired memory for system operation
85  * @curr: bytes currently allocated
86  * @high: high water mark for IO data usage
87  */
88 static struct vio_cmo {
89 	spinlock_t lock;
90 	struct delayed_work balance_q;
91 	struct list_head device_list;
92 	size_t entitled;
93 	struct vio_cmo_pool reserve;
94 	struct vio_cmo_pool excess;
95 	size_t spare;
96 	size_t min;
97 	size_t desired;
98 	size_t curr;
99 	size_t high;
100 } vio_cmo;
101 
102 /**
103  * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows
104  */
105 static int vio_cmo_num_OF_devs(void)
106 {
107 	struct device_node *node_vroot;
108 	int count = 0;
109 
110 	/*
111 	 * Count the number of vdevice entries with an
112 	 * ibm,my-dma-window OF property
113 	 */
114 	node_vroot = of_find_node_by_name(NULL, "vdevice");
115 	if (node_vroot) {
116 		struct device_node *of_node;
117 		struct property *prop;
118 
119 		for_each_child_of_node(node_vroot, of_node) {
120 			prop = of_find_property(of_node, "ibm,my-dma-window",
121 			                       NULL);
122 			if (prop)
123 				count++;
124 		}
125 	}
126 	of_node_put(node_vroot);
127 	return count;
128 }
129 
130 /**
131  * vio_cmo_alloc - allocate IO memory for CMO-enable devices
132  *
133  * @viodev: VIO device requesting IO memory
134  * @size: size of allocation requested
135  *
136  * Allocations come from memory reserved for the devices and any excess
137  * IO memory available to all devices.  The spare pool used to service
138  * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be
139  * made available.
140  *
141  * Return codes:
142  *  0 for successful allocation and -ENOMEM for a failure
143  */
144 static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size)
145 {
146 	unsigned long flags;
147 	size_t reserve_free = 0;
148 	size_t excess_free = 0;
149 	int ret = -ENOMEM;
150 
151 	spin_lock_irqsave(&vio_cmo.lock, flags);
152 
153 	/* Determine the amount of free entitlement available in reserve */
154 	if (viodev->cmo.entitled > viodev->cmo.allocated)
155 		reserve_free = viodev->cmo.entitled - viodev->cmo.allocated;
156 
157 	/* If spare is not fulfilled, the excess pool can not be used. */
158 	if (vio_cmo.spare >= VIO_CMO_MIN_ENT)
159 		excess_free = vio_cmo.excess.free;
160 
161 	/* The request can be satisfied */
162 	if ((reserve_free + excess_free) >= size) {
163 		vio_cmo.curr += size;
164 		if (vio_cmo.curr > vio_cmo.high)
165 			vio_cmo.high = vio_cmo.curr;
166 		viodev->cmo.allocated += size;
167 		size -= min(reserve_free, size);
168 		vio_cmo.excess.free -= size;
169 		ret = 0;
170 	}
171 
172 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
173 	return ret;
174 }
175 
176 /**
177  * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices
178  * @viodev: VIO device freeing IO memory
179  * @size: size of deallocation
180  *
181  * IO memory is freed by the device back to the correct memory pools.
182  * The spare pool is replenished first from either memory pool, then
183  * the reserve pool is used to reduce device entitlement, the excess
184  * pool is used to increase the reserve pool toward the desired entitlement
185  * target, and then the remaining memory is returned to the pools.
186  *
187  */
188 static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size)
189 {
190 	unsigned long flags;
191 	size_t spare_needed = 0;
192 	size_t excess_freed = 0;
193 	size_t reserve_freed = size;
194 	size_t tmp;
195 	int balance = 0;
196 
197 	spin_lock_irqsave(&vio_cmo.lock, flags);
198 	vio_cmo.curr -= size;
199 
200 	/* Amount of memory freed from the excess pool */
201 	if (viodev->cmo.allocated > viodev->cmo.entitled) {
202 		excess_freed = min(reserve_freed, (viodev->cmo.allocated -
203 		                                   viodev->cmo.entitled));
204 		reserve_freed -= excess_freed;
205 	}
206 
207 	/* Remove allocation from device */
208 	viodev->cmo.allocated -= (reserve_freed + excess_freed);
209 
210 	/* Spare is a subset of the reserve pool, replenish it first. */
211 	spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare;
212 
213 	/*
214 	 * Replenish the spare in the reserve pool from the excess pool.
215 	 * This moves entitlement into the reserve pool.
216 	 */
217 	if (spare_needed && excess_freed) {
218 		tmp = min(excess_freed, spare_needed);
219 		vio_cmo.excess.size -= tmp;
220 		vio_cmo.reserve.size += tmp;
221 		vio_cmo.spare += tmp;
222 		excess_freed -= tmp;
223 		spare_needed -= tmp;
224 		balance = 1;
225 	}
226 
227 	/*
228 	 * Replenish the spare in the reserve pool from the reserve pool.
229 	 * This removes entitlement from the device down to VIO_CMO_MIN_ENT,
230 	 * if needed, and gives it to the spare pool. The amount of used
231 	 * memory in this pool does not change.
232 	 */
233 	if (spare_needed && reserve_freed) {
234 		tmp = min3(spare_needed, reserve_freed, (viodev->cmo.entitled - VIO_CMO_MIN_ENT));
235 
236 		vio_cmo.spare += tmp;
237 		viodev->cmo.entitled -= tmp;
238 		reserve_freed -= tmp;
239 		spare_needed -= tmp;
240 		balance = 1;
241 	}
242 
243 	/*
244 	 * Increase the reserve pool until the desired allocation is met.
245 	 * Move an allocation freed from the excess pool into the reserve
246 	 * pool and schedule a balance operation.
247 	 */
248 	if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) {
249 		tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size));
250 
251 		vio_cmo.excess.size -= tmp;
252 		vio_cmo.reserve.size += tmp;
253 		excess_freed -= tmp;
254 		balance = 1;
255 	}
256 
257 	/* Return memory from the excess pool to that pool */
258 	if (excess_freed)
259 		vio_cmo.excess.free += excess_freed;
260 
261 	if (balance)
262 		schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY);
263 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
264 }
265 
266 /**
267  * vio_cmo_entitlement_update - Manage system entitlement changes
268  *
269  * @new_entitlement: new system entitlement to attempt to accommodate
270  *
271  * Increases in entitlement will be used to fulfill the spare entitlement
272  * and the rest is given to the excess pool.  Decreases, if they are
273  * possible, come from the excess pool and from unused device entitlement
274  *
275  * Returns: 0 on success, -ENOMEM when change can not be made
276  */
277 int vio_cmo_entitlement_update(size_t new_entitlement)
278 {
279 	struct vio_dev *viodev;
280 	struct vio_cmo_dev_entry *dev_ent;
281 	unsigned long flags;
282 	size_t avail, delta, tmp;
283 
284 	spin_lock_irqsave(&vio_cmo.lock, flags);
285 
286 	/* Entitlement increases */
287 	if (new_entitlement > vio_cmo.entitled) {
288 		delta = new_entitlement - vio_cmo.entitled;
289 
290 		/* Fulfill spare allocation */
291 		if (vio_cmo.spare < VIO_CMO_MIN_ENT) {
292 			tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare));
293 			vio_cmo.spare += tmp;
294 			vio_cmo.reserve.size += tmp;
295 			delta -= tmp;
296 		}
297 
298 		/* Remaining new allocation goes to the excess pool */
299 		vio_cmo.entitled += delta;
300 		vio_cmo.excess.size += delta;
301 		vio_cmo.excess.free += delta;
302 
303 		goto out;
304 	}
305 
306 	/* Entitlement decreases */
307 	delta = vio_cmo.entitled - new_entitlement;
308 	avail = vio_cmo.excess.free;
309 
310 	/*
311 	 * Need to check how much unused entitlement each device can
312 	 * sacrifice to fulfill entitlement change.
313 	 */
314 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
315 		if (avail >= delta)
316 			break;
317 
318 		viodev = dev_ent->viodev;
319 		if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
320 		    (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
321 				avail += viodev->cmo.entitled -
322 				         max_t(size_t, viodev->cmo.allocated,
323 				               VIO_CMO_MIN_ENT);
324 	}
325 
326 	if (delta <= avail) {
327 		vio_cmo.entitled -= delta;
328 
329 		/* Take entitlement from the excess pool first */
330 		tmp = min(vio_cmo.excess.free, delta);
331 		vio_cmo.excess.size -= tmp;
332 		vio_cmo.excess.free -= tmp;
333 		delta -= tmp;
334 
335 		/*
336 		 * Remove all but VIO_CMO_MIN_ENT bytes from devices
337 		 * until entitlement change is served
338 		 */
339 		list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
340 			if (!delta)
341 				break;
342 
343 			viodev = dev_ent->viodev;
344 			tmp = 0;
345 			if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
346 			    (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
347 				tmp = viodev->cmo.entitled -
348 				      max_t(size_t, viodev->cmo.allocated,
349 				            VIO_CMO_MIN_ENT);
350 			viodev->cmo.entitled -= min(tmp, delta);
351 			delta -= min(tmp, delta);
352 		}
353 	} else {
354 		spin_unlock_irqrestore(&vio_cmo.lock, flags);
355 		return -ENOMEM;
356 	}
357 
358 out:
359 	schedule_delayed_work(&vio_cmo.balance_q, 0);
360 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
361 	return 0;
362 }
363 
364 /**
365  * vio_cmo_balance - Balance entitlement among devices
366  *
367  * @work: work queue structure for this operation
368  *
369  * Any system entitlement above the minimum needed for devices, or
370  * already allocated to devices, can be distributed to the devices.
371  * The list of devices is iterated through to recalculate the desired
372  * entitlement level and to determine how much entitlement above the
373  * minimum entitlement is allocated to devices.
374  *
375  * Small chunks of the available entitlement are given to devices until
376  * their requirements are fulfilled or there is no entitlement left to give.
377  * Upon completion sizes of the reserve and excess pools are calculated.
378  *
379  * The system minimum entitlement level is also recalculated here.
380  * Entitlement will be reserved for devices even after vio_bus_remove to
381  * accommodate reloading the driver.  The OF tree is walked to count the
382  * number of devices present and this will remove entitlement for devices
383  * that have actually left the system after having vio_bus_remove called.
384  */
385 static void vio_cmo_balance(struct work_struct *work)
386 {
387 	struct vio_cmo *cmo;
388 	struct vio_dev *viodev;
389 	struct vio_cmo_dev_entry *dev_ent;
390 	unsigned long flags;
391 	size_t avail = 0, level, chunk, need;
392 	int devcount = 0, fulfilled;
393 
394 	cmo = container_of(work, struct vio_cmo, balance_q.work);
395 
396 	spin_lock_irqsave(&vio_cmo.lock, flags);
397 
398 	/* Calculate minimum entitlement and fulfill spare */
399 	cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT;
400 	BUG_ON(cmo->min > cmo->entitled);
401 	cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min));
402 	cmo->min += cmo->spare;
403 	cmo->desired = cmo->min;
404 
405 	/*
406 	 * Determine how much entitlement is available and reset device
407 	 * entitlements
408 	 */
409 	avail = cmo->entitled - cmo->spare;
410 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
411 		viodev = dev_ent->viodev;
412 		devcount++;
413 		viodev->cmo.entitled = VIO_CMO_MIN_ENT;
414 		cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT);
415 		avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT);
416 	}
417 
418 	/*
419 	 * Having provided each device with the minimum entitlement, loop
420 	 * over the devices portioning out the remaining entitlement
421 	 * until there is nothing left.
422 	 */
423 	level = VIO_CMO_MIN_ENT;
424 	while (avail) {
425 		fulfilled = 0;
426 		list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
427 			viodev = dev_ent->viodev;
428 
429 			if (viodev->cmo.desired <= level) {
430 				fulfilled++;
431 				continue;
432 			}
433 
434 			/*
435 			 * Give the device up to VIO_CMO_BALANCE_CHUNK
436 			 * bytes of entitlement, but do not exceed the
437 			 * desired level of entitlement for the device.
438 			 */
439 			chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK);
440 			chunk = min(chunk, (viodev->cmo.desired -
441 			                    viodev->cmo.entitled));
442 			viodev->cmo.entitled += chunk;
443 
444 			/*
445 			 * If the memory for this entitlement increase was
446 			 * already allocated to the device it does not come
447 			 * from the available pool being portioned out.
448 			 */
449 			need = max(viodev->cmo.allocated, viodev->cmo.entitled)-
450 			       max(viodev->cmo.allocated, level);
451 			avail -= need;
452 
453 		}
454 		if (fulfilled == devcount)
455 			break;
456 		level += VIO_CMO_BALANCE_CHUNK;
457 	}
458 
459 	/* Calculate new reserve and excess pool sizes */
460 	cmo->reserve.size = cmo->min;
461 	cmo->excess.free = 0;
462 	cmo->excess.size = 0;
463 	need = 0;
464 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
465 		viodev = dev_ent->viodev;
466 		/* Calculated reserve size above the minimum entitlement */
467 		if (viodev->cmo.entitled)
468 			cmo->reserve.size += (viodev->cmo.entitled -
469 			                      VIO_CMO_MIN_ENT);
470 		/* Calculated used excess entitlement */
471 		if (viodev->cmo.allocated > viodev->cmo.entitled)
472 			need += viodev->cmo.allocated - viodev->cmo.entitled;
473 	}
474 	cmo->excess.size = cmo->entitled - cmo->reserve.size;
475 	cmo->excess.free = cmo->excess.size - need;
476 
477 	cancel_delayed_work(to_delayed_work(work));
478 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
479 }
480 
481 static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size,
482 					  dma_addr_t *dma_handle, gfp_t flag,
483 					  unsigned long attrs)
484 {
485 	struct vio_dev *viodev = to_vio_dev(dev);
486 	void *ret;
487 
488 	if (vio_cmo_alloc(viodev, roundup(size, PAGE_SIZE))) {
489 		atomic_inc(&viodev->cmo.allocs_failed);
490 		return NULL;
491 	}
492 
493 	ret = iommu_alloc_coherent(dev, get_iommu_table_base(dev), size,
494 				    dma_handle, dev->coherent_dma_mask, flag,
495 				    dev_to_node(dev));
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 	iommu_free_coherent(get_iommu_table_base(dev), size, vaddr, dma_handle);
511 	vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
512 }
513 
514 static dma_addr_t vio_dma_iommu_map_page(struct device *dev, struct page *page,
515                                          unsigned long offset, size_t size,
516                                          enum dma_data_direction direction,
517                                          unsigned long attrs)
518 {
519 	struct vio_dev *viodev = to_vio_dev(dev);
520 	struct iommu_table *tbl = get_iommu_table_base(dev);
521 	dma_addr_t ret = DMA_MAPPING_ERROR;
522 
523 	if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl))))
524 		goto out_fail;
525 	ret = iommu_map_page(dev, tbl, page, offset, size, dma_get_mask(dev),
526 			direction, attrs);
527 	if (unlikely(ret == DMA_MAPPING_ERROR))
528 		goto out_deallocate;
529 	return ret;
530 
531 out_deallocate:
532 	vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)));
533 out_fail:
534 	atomic_inc(&viodev->cmo.allocs_failed);
535 	return DMA_MAPPING_ERROR;
536 }
537 
538 static void vio_dma_iommu_unmap_page(struct device *dev, dma_addr_t dma_handle,
539 				     size_t size,
540 				     enum dma_data_direction direction,
541 				     unsigned long attrs)
542 {
543 	struct vio_dev *viodev = to_vio_dev(dev);
544 	struct iommu_table *tbl = get_iommu_table_base(dev);
545 
546 	iommu_unmap_page(tbl, dma_handle, size, direction, attrs);
547 	vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)));
548 }
549 
550 static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
551                                 int nelems, enum dma_data_direction direction,
552                                 unsigned long attrs)
553 {
554 	struct vio_dev *viodev = to_vio_dev(dev);
555 	struct iommu_table *tbl = get_iommu_table_base(dev);
556 	struct scatterlist *sgl;
557 	int ret, count;
558 	size_t alloc_size = 0;
559 
560 	for_each_sg(sglist, sgl, nelems, count)
561 		alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE(tbl));
562 
563 	if (vio_cmo_alloc(viodev, alloc_size))
564 		goto out_fail;
565 	ret = ppc_iommu_map_sg(dev, tbl, sglist, nelems, dma_get_mask(dev),
566 			direction, attrs);
567 	if (unlikely(!ret))
568 		goto out_deallocate;
569 
570 	for_each_sg(sglist, sgl, ret, count)
571 		alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE(tbl));
572 	if (alloc_size)
573 		vio_cmo_dealloc(viodev, alloc_size);
574 	return ret;
575 
576 out_deallocate:
577 	vio_cmo_dealloc(viodev, alloc_size);
578 out_fail:
579 	atomic_inc(&viodev->cmo.allocs_failed);
580 	return 0;
581 }
582 
583 static void vio_dma_iommu_unmap_sg(struct device *dev,
584 		struct scatterlist *sglist, int nelems,
585 		enum dma_data_direction direction,
586 		unsigned long attrs)
587 {
588 	struct vio_dev *viodev = to_vio_dev(dev);
589 	struct iommu_table *tbl = get_iommu_table_base(dev);
590 	struct scatterlist *sgl;
591 	size_t alloc_size = 0;
592 	int count;
593 
594 	for_each_sg(sglist, sgl, nelems, count)
595 		alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE(tbl));
596 
597 	ppc_iommu_unmap_sg(tbl, sglist, nelems, direction, attrs);
598 	vio_cmo_dealloc(viodev, alloc_size);
599 }
600 
601 static const struct dma_map_ops vio_dma_mapping_ops = {
602 	.alloc             = vio_dma_iommu_alloc_coherent,
603 	.free              = vio_dma_iommu_free_coherent,
604 	.map_sg            = vio_dma_iommu_map_sg,
605 	.unmap_sg          = vio_dma_iommu_unmap_sg,
606 	.map_page          = vio_dma_iommu_map_page,
607 	.unmap_page        = vio_dma_iommu_unmap_page,
608 	.dma_supported     = dma_iommu_dma_supported,
609 	.get_required_mask = dma_iommu_get_required_mask,
610 	.mmap		   = dma_common_mmap,
611 	.get_sgtable	   = dma_common_get_sgtable,
612 	.alloc_pages	   = dma_common_alloc_pages,
613 	.free_pages	   = dma_common_free_pages,
614 };
615 
616 /**
617  * vio_cmo_set_dev_desired - Set desired entitlement for a device
618  *
619  * @viodev: struct vio_dev for device to alter
620  * @desired: new desired entitlement level in bytes
621  *
622  * For use by devices to request a change to their entitlement at runtime or
623  * through sysfs.  The desired entitlement level is changed and a balancing
624  * of system resources is scheduled to run in the future.
625  */
626 void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
627 {
628 	unsigned long flags;
629 	struct vio_cmo_dev_entry *dev_ent;
630 	int found = 0;
631 
632 	if (!firmware_has_feature(FW_FEATURE_CMO))
633 		return;
634 
635 	spin_lock_irqsave(&vio_cmo.lock, flags);
636 	if (desired < VIO_CMO_MIN_ENT)
637 		desired = VIO_CMO_MIN_ENT;
638 
639 	/*
640 	 * Changes will not be made for devices not in the device list.
641 	 * If it is not in the device list, then no driver is loaded
642 	 * for the device and it can not receive entitlement.
643 	 */
644 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
645 		if (viodev == dev_ent->viodev) {
646 			found = 1;
647 			break;
648 		}
649 	if (!found) {
650 		spin_unlock_irqrestore(&vio_cmo.lock, flags);
651 		return;
652 	}
653 
654 	/* Increase/decrease in desired device entitlement */
655 	if (desired >= viodev->cmo.desired) {
656 		/* Just bump the bus and device values prior to a balance*/
657 		vio_cmo.desired += desired - viodev->cmo.desired;
658 		viodev->cmo.desired = desired;
659 	} else {
660 		/* Decrease bus and device values for desired entitlement */
661 		vio_cmo.desired -= viodev->cmo.desired - desired;
662 		viodev->cmo.desired = desired;
663 		/*
664 		 * If less entitlement is desired than current entitlement, move
665 		 * any reserve memory in the change region to the excess pool.
666 		 */
667 		if (viodev->cmo.entitled > desired) {
668 			vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
669 			vio_cmo.excess.size += viodev->cmo.entitled - desired;
670 			/*
671 			 * If entitlement moving from the reserve pool to the
672 			 * excess pool is currently unused, add to the excess
673 			 * free counter.
674 			 */
675 			if (viodev->cmo.allocated < viodev->cmo.entitled)
676 				vio_cmo.excess.free += viodev->cmo.entitled -
677 				                       max(viodev->cmo.allocated, desired);
678 			viodev->cmo.entitled = desired;
679 		}
680 	}
681 	schedule_delayed_work(&vio_cmo.balance_q, 0);
682 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
683 }
684 
685 /**
686  * vio_cmo_bus_probe - Handle CMO specific bus probe activities
687  *
688  * @viodev - Pointer to struct vio_dev for device
689  *
690  * Determine the devices IO memory entitlement needs, attempting
691  * to satisfy the system minimum entitlement at first and scheduling
692  * a balance operation to take care of the rest at a later time.
693  *
694  * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
695  *          -ENOMEM when entitlement is not available for device or
696  *          device entry.
697  *
698  */
699 static int vio_cmo_bus_probe(struct vio_dev *viodev)
700 {
701 	struct vio_cmo_dev_entry *dev_ent;
702 	struct device *dev = &viodev->dev;
703 	struct iommu_table *tbl;
704 	struct vio_driver *viodrv = to_vio_driver(dev->driver);
705 	unsigned long flags;
706 	size_t size;
707 	bool dma_capable = false;
708 
709 	tbl = get_iommu_table_base(dev);
710 
711 	/* A device requires entitlement if it has a DMA window property */
712 	switch (viodev->family) {
713 	case VDEVICE:
714 		if (of_get_property(viodev->dev.of_node,
715 					"ibm,my-dma-window", NULL))
716 			dma_capable = true;
717 		break;
718 	case PFO:
719 		dma_capable = false;
720 		break;
721 	default:
722 		dev_warn(dev, "unknown device family: %d\n", viodev->family);
723 		BUG();
724 		break;
725 	}
726 
727 	/* Configure entitlement for the device. */
728 	if (dma_capable) {
729 		/* Check that the driver is CMO enabled and get desired DMA */
730 		if (!viodrv->get_desired_dma) {
731 			dev_err(dev, "%s: device driver does not support CMO\n",
732 			        __func__);
733 			return -EINVAL;
734 		}
735 
736 		viodev->cmo.desired =
737 			IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev), tbl);
738 		if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
739 			viodev->cmo.desired = VIO_CMO_MIN_ENT;
740 		size = VIO_CMO_MIN_ENT;
741 
742 		dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
743 		                  GFP_KERNEL);
744 		if (!dev_ent)
745 			return -ENOMEM;
746 
747 		dev_ent->viodev = viodev;
748 		spin_lock_irqsave(&vio_cmo.lock, flags);
749 		list_add(&dev_ent->list, &vio_cmo.device_list);
750 	} else {
751 		viodev->cmo.desired = 0;
752 		size = 0;
753 		spin_lock_irqsave(&vio_cmo.lock, flags);
754 	}
755 
756 	/*
757 	 * If the needs for vio_cmo.min have not changed since they
758 	 * were last set, the number of devices in the OF tree has
759 	 * been constant and the IO memory for this is already in
760 	 * the reserve pool.
761 	 */
762 	if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
763 	                    VIO_CMO_MIN_ENT)) {
764 		/* Updated desired entitlement if device requires it */
765 		if (size)
766 			vio_cmo.desired += (viodev->cmo.desired -
767 		                        VIO_CMO_MIN_ENT);
768 	} else {
769 		size_t tmp;
770 
771 		tmp = vio_cmo.spare + vio_cmo.excess.free;
772 		if (tmp < size) {
773 			dev_err(dev, "%s: insufficient free "
774 			        "entitlement to add device. "
775 			        "Need %lu, have %lu\n", __func__,
776 				size, (vio_cmo.spare + tmp));
777 			spin_unlock_irqrestore(&vio_cmo.lock, flags);
778 			return -ENOMEM;
779 		}
780 
781 		/* Use excess pool first to fulfill request */
782 		tmp = min(size, vio_cmo.excess.free);
783 		vio_cmo.excess.free -= tmp;
784 		vio_cmo.excess.size -= tmp;
785 		vio_cmo.reserve.size += tmp;
786 
787 		/* Use spare if excess pool was insufficient */
788 		vio_cmo.spare -= size - tmp;
789 
790 		/* Update bus accounting */
791 		vio_cmo.min += size;
792 		vio_cmo.desired += viodev->cmo.desired;
793 	}
794 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
795 	return 0;
796 }
797 
798 /**
799  * vio_cmo_bus_remove - Handle CMO specific bus removal activities
800  *
801  * @viodev - Pointer to struct vio_dev for device
802  *
803  * Remove the device from the cmo device list.  The minimum entitlement
804  * will be reserved for the device as long as it is in the system.  The
805  * rest of the entitlement the device had been allocated will be returned
806  * to the system.
807  */
808 static void vio_cmo_bus_remove(struct vio_dev *viodev)
809 {
810 	struct vio_cmo_dev_entry *dev_ent;
811 	unsigned long flags;
812 	size_t tmp;
813 
814 	spin_lock_irqsave(&vio_cmo.lock, flags);
815 	if (viodev->cmo.allocated) {
816 		dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
817 		        "allocated after remove operation.\n",
818 		        __func__, viodev->cmo.allocated);
819 		BUG();
820 	}
821 
822 	/*
823 	 * Remove the device from the device list being maintained for
824 	 * CMO enabled devices.
825 	 */
826 	list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
827 		if (viodev == dev_ent->viodev) {
828 			list_del(&dev_ent->list);
829 			kfree(dev_ent);
830 			break;
831 		}
832 
833 	/*
834 	 * Devices may not require any entitlement and they do not need
835 	 * to be processed.  Otherwise, return the device's entitlement
836 	 * back to the pools.
837 	 */
838 	if (viodev->cmo.entitled) {
839 		/*
840 		 * This device has not yet left the OF tree, it's
841 		 * minimum entitlement remains in vio_cmo.min and
842 		 * vio_cmo.desired
843 		 */
844 		vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
845 
846 		/*
847 		 * Save min allocation for device in reserve as long
848 		 * as it exists in OF tree as determined by later
849 		 * balance operation
850 		 */
851 		viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
852 
853 		/* Replenish spare from freed reserve pool */
854 		if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
855 			tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
856 			                                 vio_cmo.spare));
857 			vio_cmo.spare += tmp;
858 			viodev->cmo.entitled -= tmp;
859 		}
860 
861 		/* Remaining reserve goes to excess pool */
862 		vio_cmo.excess.size += viodev->cmo.entitled;
863 		vio_cmo.excess.free += viodev->cmo.entitled;
864 		vio_cmo.reserve.size -= viodev->cmo.entitled;
865 
866 		/*
867 		 * Until the device is removed it will keep a
868 		 * minimum entitlement; this will guarantee that
869 		 * a module unload/load will result in a success.
870 		 */
871 		viodev->cmo.entitled = VIO_CMO_MIN_ENT;
872 		viodev->cmo.desired = VIO_CMO_MIN_ENT;
873 		atomic_set(&viodev->cmo.allocs_failed, 0);
874 	}
875 
876 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
877 }
878 
879 static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
880 {
881 	set_dma_ops(&viodev->dev, &vio_dma_mapping_ops);
882 }
883 
884 /**
885  * vio_cmo_bus_init - CMO entitlement initialization at bus init time
886  *
887  * Set up the reserve and excess entitlement pools based on available
888  * system entitlement and the number of devices in the OF tree that
889  * require entitlement in the reserve pool.
890  */
891 static void vio_cmo_bus_init(void)
892 {
893 	struct hvcall_mpp_data mpp_data;
894 	int err;
895 
896 	memset(&vio_cmo, 0, sizeof(struct vio_cmo));
897 	spin_lock_init(&vio_cmo.lock);
898 	INIT_LIST_HEAD(&vio_cmo.device_list);
899 	INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
900 
901 	/* Get current system entitlement */
902 	err = h_get_mpp(&mpp_data);
903 
904 	/*
905 	 * On failure, continue with entitlement set to 0, will panic()
906 	 * later when spare is reserved.
907 	 */
908 	if (err != H_SUCCESS) {
909 		printk(KERN_ERR "%s: unable to determine system IO "\
910 		       "entitlement. (%d)\n", __func__, err);
911 		vio_cmo.entitled = 0;
912 	} else {
913 		vio_cmo.entitled = mpp_data.entitled_mem;
914 	}
915 
916 	/* Set reservation and check against entitlement */
917 	vio_cmo.spare = VIO_CMO_MIN_ENT;
918 	vio_cmo.reserve.size = vio_cmo.spare;
919 	vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
920 	                         VIO_CMO_MIN_ENT);
921 	if (vio_cmo.reserve.size > vio_cmo.entitled) {
922 		printk(KERN_ERR "%s: insufficient system entitlement\n",
923 		       __func__);
924 		panic("%s: Insufficient system entitlement", __func__);
925 	}
926 
927 	/* Set the remaining accounting variables */
928 	vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
929 	vio_cmo.excess.free = vio_cmo.excess.size;
930 	vio_cmo.min = vio_cmo.reserve.size;
931 	vio_cmo.desired = vio_cmo.reserve.size;
932 }
933 
934 /* sysfs device functions and data structures for CMO */
935 
936 #define viodev_cmo_rd_attr(name)                                        \
937 static ssize_t cmo_##name##_show(struct device *dev,                    \
938                                         struct device_attribute *attr,  \
939                                          char *buf)                     \
940 {                                                                       \
941 	return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name);        \
942 }
943 
944 static ssize_t cmo_allocs_failed_show(struct device *dev,
945 		struct device_attribute *attr, char *buf)
946 {
947 	struct vio_dev *viodev = to_vio_dev(dev);
948 	return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
949 }
950 
951 static ssize_t cmo_allocs_failed_store(struct device *dev,
952 		struct device_attribute *attr, const char *buf, size_t count)
953 {
954 	struct vio_dev *viodev = to_vio_dev(dev);
955 	atomic_set(&viodev->cmo.allocs_failed, 0);
956 	return count;
957 }
958 
959 static ssize_t cmo_desired_store(struct device *dev,
960 		struct device_attribute *attr, const char *buf, size_t count)
961 {
962 	struct vio_dev *viodev = to_vio_dev(dev);
963 	size_t new_desired;
964 	int ret;
965 
966 	ret = kstrtoul(buf, 10, &new_desired);
967 	if (ret)
968 		return ret;
969 
970 	vio_cmo_set_dev_desired(viodev, new_desired);
971 	return count;
972 }
973 
974 viodev_cmo_rd_attr(desired);
975 viodev_cmo_rd_attr(entitled);
976 viodev_cmo_rd_attr(allocated);
977 
978 static ssize_t name_show(struct device *, struct device_attribute *, char *);
979 static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
980 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
981 			     char *buf);
982 
983 static struct device_attribute dev_attr_name;
984 static struct device_attribute dev_attr_devspec;
985 static struct device_attribute dev_attr_modalias;
986 
987 static DEVICE_ATTR_RO(cmo_entitled);
988 static DEVICE_ATTR_RO(cmo_allocated);
989 static DEVICE_ATTR_RW(cmo_desired);
990 static DEVICE_ATTR_RW(cmo_allocs_failed);
991 
992 static struct attribute *vio_cmo_dev_attrs[] = {
993 	&dev_attr_name.attr,
994 	&dev_attr_devspec.attr,
995 	&dev_attr_modalias.attr,
996 	&dev_attr_cmo_entitled.attr,
997 	&dev_attr_cmo_allocated.attr,
998 	&dev_attr_cmo_desired.attr,
999 	&dev_attr_cmo_allocs_failed.attr,
1000 	NULL,
1001 };
1002 ATTRIBUTE_GROUPS(vio_cmo_dev);
1003 
1004 /* sysfs bus functions and data structures for CMO */
1005 
1006 #define viobus_cmo_rd_attr(name)                                        \
1007 static ssize_t cmo_bus_##name##_show(struct bus_type *bt, char *buf)    \
1008 {                                                                       \
1009 	return sprintf(buf, "%lu\n", vio_cmo.name);                     \
1010 }                                                                       \
1011 static struct bus_attribute bus_attr_cmo_bus_##name =			\
1012 	__ATTR(cmo_##name, S_IRUGO, cmo_bus_##name##_show, NULL)
1013 
1014 #define viobus_cmo_pool_rd_attr(name, var)                              \
1015 static ssize_t                                                          \
1016 cmo_##name##_##var##_show(struct bus_type *bt, char *buf)               \
1017 {                                                                       \
1018 	return sprintf(buf, "%lu\n", vio_cmo.name.var);                 \
1019 }                                                                       \
1020 static BUS_ATTR_RO(cmo_##name##_##var)
1021 
1022 viobus_cmo_rd_attr(entitled);
1023 viobus_cmo_rd_attr(spare);
1024 viobus_cmo_rd_attr(min);
1025 viobus_cmo_rd_attr(desired);
1026 viobus_cmo_rd_attr(curr);
1027 viobus_cmo_pool_rd_attr(reserve, size);
1028 viobus_cmo_pool_rd_attr(excess, size);
1029 viobus_cmo_pool_rd_attr(excess, free);
1030 
1031 static ssize_t cmo_high_show(struct bus_type *bt, char *buf)
1032 {
1033 	return sprintf(buf, "%lu\n", vio_cmo.high);
1034 }
1035 
1036 static ssize_t cmo_high_store(struct bus_type *bt, const char *buf,
1037 			      size_t count)
1038 {
1039 	unsigned long flags;
1040 
1041 	spin_lock_irqsave(&vio_cmo.lock, flags);
1042 	vio_cmo.high = vio_cmo.curr;
1043 	spin_unlock_irqrestore(&vio_cmo.lock, flags);
1044 
1045 	return count;
1046 }
1047 static BUS_ATTR_RW(cmo_high);
1048 
1049 static struct attribute *vio_bus_attrs[] = {
1050 	&bus_attr_cmo_bus_entitled.attr,
1051 	&bus_attr_cmo_bus_spare.attr,
1052 	&bus_attr_cmo_bus_min.attr,
1053 	&bus_attr_cmo_bus_desired.attr,
1054 	&bus_attr_cmo_bus_curr.attr,
1055 	&bus_attr_cmo_high.attr,
1056 	&bus_attr_cmo_reserve_size.attr,
1057 	&bus_attr_cmo_excess_size.attr,
1058 	&bus_attr_cmo_excess_free.attr,
1059 	NULL,
1060 };
1061 ATTRIBUTE_GROUPS(vio_bus);
1062 
1063 static void vio_cmo_sysfs_init(void)
1064 {
1065 	vio_bus_type.dev_groups = vio_cmo_dev_groups;
1066 	vio_bus_type.bus_groups = vio_bus_groups;
1067 }
1068 #else /* CONFIG_PPC_SMLPAR */
1069 int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1070 void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1071 static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1072 static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1073 static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
1074 static void vio_cmo_bus_init(void) {}
1075 static void vio_cmo_sysfs_init(void) { }
1076 #endif /* CONFIG_PPC_SMLPAR */
1077 EXPORT_SYMBOL(vio_cmo_entitlement_update);
1078 EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1079 
1080 
1081 /*
1082  * Platform Facilities Option (PFO) support
1083  */
1084 
1085 /**
1086  * vio_h_cop_sync - Perform a synchronous PFO co-processor operation
1087  *
1088  * @vdev - Pointer to a struct vio_dev for device
1089  * @op - Pointer to a struct vio_pfo_op for the operation parameters
1090  *
1091  * Calls the hypervisor to synchronously perform the PFO operation
1092  * described in @op.  In the case of a busy response from the hypervisor,
1093  * the operation will be re-submitted indefinitely unless a non-zero timeout
1094  * is specified or an error occurs. The timeout places a limit on when to
1095  * stop re-submitting a operation, the total time can be exceeded if an
1096  * operation is in progress.
1097  *
1098  * If op->hcall_ret is not NULL, this will be set to the return from the
1099  * last h_cop_op call or it will be 0 if an error not involving the h_call
1100  * was encountered.
1101  *
1102  * Returns:
1103  *	0 on success,
1104  *	-EINVAL if the h_call fails due to an invalid parameter,
1105  *	-E2BIG if the h_call can not be performed synchronously,
1106  *	-EBUSY if a timeout is specified and has elapsed,
1107  *	-EACCES if the memory area for data/status has been rescinded, or
1108  *	-EPERM if a hardware fault has been indicated
1109  */
1110 int vio_h_cop_sync(struct vio_dev *vdev, struct vio_pfo_op *op)
1111 {
1112 	struct device *dev = &vdev->dev;
1113 	unsigned long deadline = 0;
1114 	long hret = 0;
1115 	int ret = 0;
1116 
1117 	if (op->timeout)
1118 		deadline = jiffies + msecs_to_jiffies(op->timeout);
1119 
1120 	while (true) {
1121 		hret = plpar_hcall_norets(H_COP, op->flags,
1122 				vdev->resource_id,
1123 				op->in, op->inlen, op->out,
1124 				op->outlen, op->csbcpb);
1125 
1126 		if (hret == H_SUCCESS ||
1127 		    (hret != H_NOT_ENOUGH_RESOURCES &&
1128 		     hret != H_BUSY && hret != H_RESOURCE) ||
1129 		    (op->timeout && time_after(deadline, jiffies)))
1130 			break;
1131 
1132 		dev_dbg(dev, "%s: hcall ret(%ld), retrying.\n", __func__, hret);
1133 	}
1134 
1135 	switch (hret) {
1136 	case H_SUCCESS:
1137 		ret = 0;
1138 		break;
1139 	case H_OP_MODE:
1140 	case H_TOO_BIG:
1141 		ret = -E2BIG;
1142 		break;
1143 	case H_RESCINDED:
1144 		ret = -EACCES;
1145 		break;
1146 	case H_HARDWARE:
1147 		ret = -EPERM;
1148 		break;
1149 	case H_NOT_ENOUGH_RESOURCES:
1150 	case H_RESOURCE:
1151 	case H_BUSY:
1152 		ret = -EBUSY;
1153 		break;
1154 	default:
1155 		ret = -EINVAL;
1156 		break;
1157 	}
1158 
1159 	if (ret)
1160 		dev_dbg(dev, "%s: Sync h_cop_op failure (ret:%d) (hret:%ld)\n",
1161 				__func__, ret, hret);
1162 
1163 	op->hcall_err = hret;
1164 	return ret;
1165 }
1166 EXPORT_SYMBOL(vio_h_cop_sync);
1167 
1168 static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1169 {
1170 	const __be32 *dma_window;
1171 	struct iommu_table *tbl;
1172 	unsigned long offset, size;
1173 
1174 	dma_window = of_get_property(dev->dev.of_node,
1175 				  "ibm,my-dma-window", NULL);
1176 	if (!dma_window)
1177 		return NULL;
1178 
1179 	tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
1180 	if (tbl == NULL)
1181 		return NULL;
1182 
1183 	kref_init(&tbl->it_kref);
1184 
1185 	of_parse_dma_window(dev->dev.of_node, dma_window,
1186 			    &tbl->it_index, &offset, &size);
1187 
1188 	/* TCE table size - measured in tce entries */
1189 	tbl->it_page_shift = IOMMU_PAGE_SHIFT_4K;
1190 	tbl->it_size = size >> tbl->it_page_shift;
1191 	/* offset for VIO should always be 0 */
1192 	tbl->it_offset = offset >> tbl->it_page_shift;
1193 	tbl->it_busno = 0;
1194 	tbl->it_type = TCE_VB;
1195 	tbl->it_blocksize = 16;
1196 
1197 	if (firmware_has_feature(FW_FEATURE_LPAR))
1198 		tbl->it_ops = &iommu_table_lpar_multi_ops;
1199 	else
1200 		tbl->it_ops = &iommu_table_pseries_ops;
1201 
1202 	return iommu_init_table(tbl, -1, 0, 0);
1203 }
1204 
1205 /**
1206  * vio_match_device: - Tell if a VIO device has a matching
1207  *			VIO device id structure.
1208  * @ids:	array of VIO device id structures to search in
1209  * @dev:	the VIO device structure to match against
1210  *
1211  * Used by a driver to check whether a VIO device present in the
1212  * system is in its list of supported devices. Returns the matching
1213  * vio_device_id structure or NULL if there is no match.
1214  */
1215 static const struct vio_device_id *vio_match_device(
1216 		const struct vio_device_id *ids, const struct vio_dev *dev)
1217 {
1218 	while (ids->type[0] != '\0') {
1219 		if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
1220 		    of_device_is_compatible(dev->dev.of_node,
1221 					 ids->compat))
1222 			return ids;
1223 		ids++;
1224 	}
1225 	return NULL;
1226 }
1227 
1228 /*
1229  * Convert from struct device to struct vio_dev and pass to driver.
1230  * dev->driver has already been set by generic code because vio_bus_match
1231  * succeeded.
1232  */
1233 static int vio_bus_probe(struct device *dev)
1234 {
1235 	struct vio_dev *viodev = to_vio_dev(dev);
1236 	struct vio_driver *viodrv = to_vio_driver(dev->driver);
1237 	const struct vio_device_id *id;
1238 	int error = -ENODEV;
1239 
1240 	if (!viodrv->probe)
1241 		return error;
1242 
1243 	id = vio_match_device(viodrv->id_table, viodev);
1244 	if (id) {
1245 		memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1246 		if (firmware_has_feature(FW_FEATURE_CMO)) {
1247 			error = vio_cmo_bus_probe(viodev);
1248 			if (error)
1249 				return error;
1250 		}
1251 		error = viodrv->probe(viodev, id);
1252 		if (error && firmware_has_feature(FW_FEATURE_CMO))
1253 			vio_cmo_bus_remove(viodev);
1254 	}
1255 
1256 	return error;
1257 }
1258 
1259 /* convert from struct device to struct vio_dev and pass to driver. */
1260 static int vio_bus_remove(struct device *dev)
1261 {
1262 	struct vio_dev *viodev = to_vio_dev(dev);
1263 	struct vio_driver *viodrv = to_vio_driver(dev->driver);
1264 	struct device *devptr;
1265 
1266 	/*
1267 	 * Hold a reference to the device after the remove function is called
1268 	 * to allow for CMO accounting cleanup for the device.
1269 	 */
1270 	devptr = get_device(dev);
1271 
1272 	if (viodrv->remove)
1273 		viodrv->remove(viodev);
1274 
1275 	if (firmware_has_feature(FW_FEATURE_CMO))
1276 		vio_cmo_bus_remove(viodev);
1277 
1278 	put_device(devptr);
1279 	return 0;
1280 }
1281 
1282 static void vio_bus_shutdown(struct device *dev)
1283 {
1284 	struct vio_dev *viodev = to_vio_dev(dev);
1285 	struct vio_driver *viodrv;
1286 
1287 	if (dev->driver) {
1288 		viodrv = to_vio_driver(dev->driver);
1289 		if (viodrv->shutdown)
1290 			viodrv->shutdown(viodev);
1291 		else if (kexec_in_progress)
1292 			vio_bus_remove(dev);
1293 	}
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 	// vio_bus_type is only initialised for pseries
1304 	if (!machine_is(pseries))
1305 		return -ENODEV;
1306 
1307 	pr_debug("%s: driver %s registering\n", __func__, viodrv->name);
1308 
1309 	/* fill in 'struct driver' fields */
1310 	viodrv->driver.name = viodrv->name;
1311 	viodrv->driver.pm = viodrv->pm;
1312 	viodrv->driver.bus = &vio_bus_type;
1313 	viodrv->driver.owner = owner;
1314 	viodrv->driver.mod_name = mod_name;
1315 
1316 	return driver_register(&viodrv->driver);
1317 }
1318 EXPORT_SYMBOL(__vio_register_driver);
1319 
1320 /**
1321  * vio_unregister_driver - Remove registration of vio driver.
1322  * @viodrv:	The vio_driver struct to be removed form registration
1323  */
1324 void vio_unregister_driver(struct vio_driver *viodrv)
1325 {
1326 	driver_unregister(&viodrv->driver);
1327 }
1328 EXPORT_SYMBOL(vio_unregister_driver);
1329 
1330 /* vio_dev refcount hit 0 */
1331 static void vio_dev_release(struct device *dev)
1332 {
1333 	struct iommu_table *tbl = get_iommu_table_base(dev);
1334 
1335 	if (tbl)
1336 		iommu_tce_table_put(tbl);
1337 	of_node_put(dev->of_node);
1338 	kfree(to_vio_dev(dev));
1339 }
1340 
1341 /**
1342  * vio_register_device_node: - Register a new vio device.
1343  * @of_node:	The OF node for this device.
1344  *
1345  * Creates and initializes a vio_dev structure from the data in
1346  * of_node and adds it to the list of virtual devices.
1347  * Returns a pointer to the created vio_dev or NULL if node has
1348  * NULL device_type or compatible fields.
1349  */
1350 struct vio_dev *vio_register_device_node(struct device_node *of_node)
1351 {
1352 	struct vio_dev *viodev;
1353 	struct device_node *parent_node;
1354 	const __be32 *prop;
1355 	enum vio_dev_family family;
1356 
1357 	/*
1358 	 * Determine if this node is a under the /vdevice node or under the
1359 	 * /ibm,platform-facilities node.  This decides the device's family.
1360 	 */
1361 	parent_node = of_get_parent(of_node);
1362 	if (parent_node) {
1363 		if (of_node_is_type(parent_node, "ibm,platform-facilities"))
1364 			family = PFO;
1365 		else if (of_node_is_type(parent_node, "vdevice"))
1366 			family = VDEVICE;
1367 		else {
1368 			pr_warn("%s: parent(%pOF) of %pOFn not recognized.\n",
1369 					__func__,
1370 					parent_node,
1371 					of_node);
1372 			of_node_put(parent_node);
1373 			return NULL;
1374 		}
1375 		of_node_put(parent_node);
1376 	} else {
1377 		pr_warn("%s: could not determine the parent of node %pOFn.\n",
1378 				__func__, of_node);
1379 		return NULL;
1380 	}
1381 
1382 	if (family == PFO) {
1383 		if (of_get_property(of_node, "interrupt-controller", NULL)) {
1384 			pr_debug("%s: Skipping the interrupt controller %pOFn.\n",
1385 					__func__, of_node);
1386 			return NULL;
1387 		}
1388 	}
1389 
1390 	/* allocate a vio_dev for this node */
1391 	viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1392 	if (viodev == NULL) {
1393 		pr_warn("%s: allocation failure for VIO device.\n", __func__);
1394 		return NULL;
1395 	}
1396 
1397 	/* we need the 'device_type' property, in order to match with drivers */
1398 	viodev->family = family;
1399 	if (viodev->family == VDEVICE) {
1400 		unsigned int unit_address;
1401 
1402 		viodev->type = of_node_get_device_type(of_node);
1403 		if (!viodev->type) {
1404 			pr_warn("%s: node %pOFn is missing the 'device_type' "
1405 					"property.\n", __func__, of_node);
1406 			goto out;
1407 		}
1408 
1409 		prop = of_get_property(of_node, "reg", NULL);
1410 		if (prop == NULL) {
1411 			pr_warn("%s: node %pOFn missing 'reg'\n",
1412 					__func__, of_node);
1413 			goto out;
1414 		}
1415 		unit_address = of_read_number(prop, 1);
1416 		dev_set_name(&viodev->dev, "%x", unit_address);
1417 		viodev->irq = irq_of_parse_and_map(of_node, 0);
1418 		viodev->unit_address = unit_address;
1419 	} else {
1420 		/* PFO devices need their resource_id for submitting COP_OPs
1421 		 * This is an optional field for devices, but is required when
1422 		 * performing synchronous ops */
1423 		prop = of_get_property(of_node, "ibm,resource-id", NULL);
1424 		if (prop != NULL)
1425 			viodev->resource_id = of_read_number(prop, 1);
1426 
1427 		dev_set_name(&viodev->dev, "%pOFn", of_node);
1428 		viodev->type = dev_name(&viodev->dev);
1429 		viodev->irq = 0;
1430 	}
1431 
1432 	viodev->name = of_node->name;
1433 	viodev->dev.of_node = of_node_get(of_node);
1434 
1435 	set_dev_node(&viodev->dev, of_node_to_nid(of_node));
1436 
1437 	/* init generic 'struct device' fields: */
1438 	viodev->dev.parent = &vio_bus_device.dev;
1439 	viodev->dev.bus = &vio_bus_type;
1440 	viodev->dev.release = vio_dev_release;
1441 
1442 	if (of_get_property(viodev->dev.of_node, "ibm,my-dma-window", NULL)) {
1443 		if (firmware_has_feature(FW_FEATURE_CMO))
1444 			vio_cmo_set_dma_ops(viodev);
1445 		else
1446 			set_dma_ops(&viodev->dev, &dma_iommu_ops);
1447 
1448 		set_iommu_table_base(&viodev->dev,
1449 				     vio_build_iommu_table(viodev));
1450 
1451 		/* needed to ensure proper operation of coherent allocations
1452 		 * later, in case driver doesn't set it explicitly */
1453 		viodev->dev.coherent_dma_mask = DMA_BIT_MASK(64);
1454 		viodev->dev.dma_mask = &viodev->dev.coherent_dma_mask;
1455 	}
1456 
1457 	/* register with generic device framework */
1458 	if (device_register(&viodev->dev)) {
1459 		printk(KERN_ERR "%s: failed to register device %s\n",
1460 				__func__, dev_name(&viodev->dev));
1461 		put_device(&viodev->dev);
1462 		return NULL;
1463 	}
1464 
1465 	return viodev;
1466 
1467 out:	/* Use this exit point for any return prior to device_register */
1468 	kfree(viodev);
1469 
1470 	return NULL;
1471 }
1472 EXPORT_SYMBOL(vio_register_device_node);
1473 
1474 /*
1475  * vio_bus_scan_for_devices - Scan OF and register each child device
1476  * @root_name - OF node name for the root of the subtree to search.
1477  *		This must be non-NULL
1478  *
1479  * Starting from the root node provide, register the device node for
1480  * each child beneath the root.
1481  */
1482 static void vio_bus_scan_register_devices(char *root_name)
1483 {
1484 	struct device_node *node_root, *node_child;
1485 
1486 	if (!root_name)
1487 		return;
1488 
1489 	node_root = of_find_node_by_name(NULL, root_name);
1490 	if (node_root) {
1491 
1492 		/*
1493 		 * Create struct vio_devices for each virtual device in
1494 		 * the device tree. Drivers will associate with them later.
1495 		 */
1496 		node_child = of_get_next_child(node_root, NULL);
1497 		while (node_child) {
1498 			vio_register_device_node(node_child);
1499 			node_child = of_get_next_child(node_root, node_child);
1500 		}
1501 		of_node_put(node_root);
1502 	}
1503 }
1504 
1505 /**
1506  * vio_bus_init: - Initialize the virtual IO bus
1507  */
1508 static int __init vio_bus_init(void)
1509 {
1510 	int err;
1511 
1512 	if (firmware_has_feature(FW_FEATURE_CMO))
1513 		vio_cmo_sysfs_init();
1514 
1515 	err = bus_register(&vio_bus_type);
1516 	if (err) {
1517 		printk(KERN_ERR "failed to register VIO bus\n");
1518 		return err;
1519 	}
1520 
1521 	/*
1522 	 * The fake parent of all vio devices, just to give us
1523 	 * a nice directory
1524 	 */
1525 	err = device_register(&vio_bus_device.dev);
1526 	if (err) {
1527 		printk(KERN_WARNING "%s: device_register returned %i\n",
1528 				__func__, err);
1529 		return err;
1530 	}
1531 
1532 	if (firmware_has_feature(FW_FEATURE_CMO))
1533 		vio_cmo_bus_init();
1534 
1535 	return 0;
1536 }
1537 machine_postcore_initcall(pseries, vio_bus_init);
1538 
1539 static int __init vio_device_init(void)
1540 {
1541 	vio_bus_scan_register_devices("vdevice");
1542 	vio_bus_scan_register_devices("ibm,platform-facilities");
1543 
1544 	return 0;
1545 }
1546 machine_device_initcall(pseries, vio_device_init);
1547 
1548 static ssize_t name_show(struct device *dev,
1549 		struct device_attribute *attr, char *buf)
1550 {
1551 	return sprintf(buf, "%s\n", to_vio_dev(dev)->name);
1552 }
1553 static DEVICE_ATTR_RO(name);
1554 
1555 static ssize_t devspec_show(struct device *dev,
1556 		struct device_attribute *attr, char *buf)
1557 {
1558 	struct device_node *of_node = dev->of_node;
1559 
1560 	return sprintf(buf, "%pOF\n", of_node);
1561 }
1562 static DEVICE_ATTR_RO(devspec);
1563 
1564 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1565 			     char *buf)
1566 {
1567 	const struct vio_dev *vio_dev = to_vio_dev(dev);
1568 	struct device_node *dn;
1569 	const char *cp;
1570 
1571 	dn = dev->of_node;
1572 	if (!dn) {
1573 		strcpy(buf, "\n");
1574 		return strlen(buf);
1575 	}
1576 	cp = of_get_property(dn, "compatible", NULL);
1577 	if (!cp) {
1578 		strcpy(buf, "\n");
1579 		return strlen(buf);
1580 	}
1581 
1582 	return sprintf(buf, "vio:T%sS%s\n", vio_dev->type, cp);
1583 }
1584 static DEVICE_ATTR_RO(modalias);
1585 
1586 static struct attribute *vio_dev_attrs[] = {
1587 	&dev_attr_name.attr,
1588 	&dev_attr_devspec.attr,
1589 	&dev_attr_modalias.attr,
1590 	NULL,
1591 };
1592 ATTRIBUTE_GROUPS(vio_dev);
1593 
1594 void vio_unregister_device(struct vio_dev *viodev)
1595 {
1596 	device_unregister(&viodev->dev);
1597 	if (viodev->family == VDEVICE)
1598 		irq_dispose_mapping(viodev->irq);
1599 }
1600 EXPORT_SYMBOL(vio_unregister_device);
1601 
1602 static int vio_bus_match(struct device *dev, struct device_driver *drv)
1603 {
1604 	const struct vio_dev *vio_dev = to_vio_dev(dev);
1605 	struct vio_driver *vio_drv = to_vio_driver(drv);
1606 	const struct vio_device_id *ids = vio_drv->id_table;
1607 
1608 	return (ids != NULL) && (vio_match_device(ids, vio_dev) != NULL);
1609 }
1610 
1611 static int vio_hotplug(struct device *dev, struct kobj_uevent_env *env)
1612 {
1613 	const struct vio_dev *vio_dev = to_vio_dev(dev);
1614 	struct device_node *dn;
1615 	const char *cp;
1616 
1617 	dn = dev->of_node;
1618 	if (!dn)
1619 		return -ENODEV;
1620 	cp = of_get_property(dn, "compatible", NULL);
1621 	if (!cp)
1622 		return -ENODEV;
1623 
1624 	add_uevent_var(env, "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
1625 	return 0;
1626 }
1627 
1628 struct bus_type vio_bus_type = {
1629 	.name = "vio",
1630 	.dev_groups = vio_dev_groups,
1631 	.uevent = vio_hotplug,
1632 	.match = vio_bus_match,
1633 	.probe = vio_bus_probe,
1634 	.remove = vio_bus_remove,
1635 	.shutdown = vio_bus_shutdown,
1636 };
1637 
1638 /**
1639  * vio_get_attribute: - get attribute for virtual device
1640  * @vdev:	The vio device to get property.
1641  * @which:	The property/attribute to be extracted.
1642  * @length:	Pointer to length of returned data size (unused if NULL).
1643  *
1644  * Calls prom.c's of_get_property() to return the value of the
1645  * attribute specified by @which
1646 */
1647 const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1648 {
1649 	return of_get_property(vdev->dev.of_node, which, length);
1650 }
1651 EXPORT_SYMBOL(vio_get_attribute);
1652 
1653 /* vio_find_name() - internal because only vio.c knows how we formatted the
1654  * kobject name
1655  */
1656 static struct vio_dev *vio_find_name(const char *name)
1657 {
1658 	struct device *found;
1659 
1660 	found = bus_find_device_by_name(&vio_bus_type, NULL, name);
1661 	if (!found)
1662 		return NULL;
1663 
1664 	return to_vio_dev(found);
1665 }
1666 
1667 /**
1668  * vio_find_node - find an already-registered vio_dev
1669  * @vnode: device_node of the virtual device we're looking for
1670  *
1671  * Takes a reference to the embedded struct device which needs to be dropped
1672  * after use.
1673  */
1674 struct vio_dev *vio_find_node(struct device_node *vnode)
1675 {
1676 	char kobj_name[20];
1677 	struct device_node *vnode_parent;
1678 
1679 	vnode_parent = of_get_parent(vnode);
1680 	if (!vnode_parent)
1681 		return NULL;
1682 
1683 	/* construct the kobject name from the device node */
1684 	if (of_node_is_type(vnode_parent, "vdevice")) {
1685 		const __be32 *prop;
1686 
1687 		prop = of_get_property(vnode, "reg", NULL);
1688 		if (!prop)
1689 			goto out;
1690 		snprintf(kobj_name, sizeof(kobj_name), "%x",
1691 			 (uint32_t)of_read_number(prop, 1));
1692 	} else if (of_node_is_type(vnode_parent, "ibm,platform-facilities"))
1693 		snprintf(kobj_name, sizeof(kobj_name), "%pOFn", vnode);
1694 	else
1695 		goto out;
1696 
1697 	of_node_put(vnode_parent);
1698 	return vio_find_name(kobj_name);
1699 out:
1700 	of_node_put(vnode_parent);
1701 	return NULL;
1702 }
1703 EXPORT_SYMBOL(vio_find_node);
1704 
1705 int vio_enable_interrupts(struct vio_dev *dev)
1706 {
1707 	int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1708 	if (rc != H_SUCCESS)
1709 		printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1710 	return rc;
1711 }
1712 EXPORT_SYMBOL(vio_enable_interrupts);
1713 
1714 int vio_disable_interrupts(struct vio_dev *dev)
1715 {
1716 	int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1717 	if (rc != H_SUCCESS)
1718 		printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1719 	return rc;
1720 }
1721 EXPORT_SYMBOL(vio_disable_interrupts);
1722 
1723 static int __init vio_init(void)
1724 {
1725 	dma_debug_add_bus(&vio_bus_type);
1726 	return 0;
1727 }
1728 machine_fs_initcall(pseries, vio_init);
1729