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#include <linux/of_irq.h>
27
28#include <asm/iommu.h>
29#include <asm/dma.h>
30#include <asm/vio.h>
31#include <asm/prom.h>
32#include <asm/firmware.h>
33#include <asm/tce.h>
34#include <asm/page.h>
35#include <asm/hvcall.h>
36#include <asm/machdep.h>
37
38static struct vio_dev vio_bus_device = { /* fake "parent" device */
39 .name = "vio",
40 .type = "",
41 .dev.init_name = "vio",
42 .dev.bus = &vio_bus_type,
43};
44
45#ifdef CONFIG_PPC_SMLPAR
46/**
47 * vio_cmo_pool - A pool of IO memory for CMO use
48 *
49 * @size: The size of the pool in bytes
50 * @free: The amount of free memory in the pool
51 */
52struct vio_cmo_pool {
53 size_t size;
54 size_t free;
55};
56
57/* How many ms to delay queued balance work */
58#define VIO_CMO_BALANCE_DELAY 100
59
60/* Portion out IO memory to CMO devices by this chunk size */
61#define VIO_CMO_BALANCE_CHUNK 131072
62
63/**
64 * vio_cmo_dev_entry - A device that is CMO-enabled and requires entitlement
65 *
66 * @vio_dev: struct vio_dev pointer
67 * @list: pointer to other devices on bus that are being tracked
68 */
69struct vio_cmo_dev_entry {
70 struct vio_dev *viodev;
71 struct list_head list;
72};
73
74/**
75 * vio_cmo - VIO bus accounting structure for CMO entitlement
76 *
77 * @lock: spinlock for entire structure
78 * @balance_q: work queue for balancing system entitlement
79 * @device_list: list of CMO-enabled devices requiring entitlement
80 * @entitled: total system entitlement in bytes
81 * @reserve: pool of memory from which devices reserve entitlement, incl. spare
82 * @excess: pool of excess entitlement not needed for device reserves or spare
83 * @spare: IO memory for device hotplug functionality
84 * @min: minimum necessary for system operation
85 * @desired: desired memory for system operation
86 * @curr: bytes currently allocated
87 * @high: high water mark for IO data usage
88 */
89static struct vio_cmo {
90 spinlock_t lock;
91 struct delayed_work balance_q;
92 struct list_head device_list;
93 size_t entitled;
94 struct vio_cmo_pool reserve;
95 struct vio_cmo_pool excess;
96 size_t spare;
97 size_t min;
98 size_t desired;
99 size_t curr;
100 size_t high;
101} vio_cmo;
102
103/**
104 * vio_cmo_OF_devices - Count the number of OF devices that have DMA windows
105 */
106static int vio_cmo_num_OF_devs(void)
107{
108 struct device_node *node_vroot;
109 int count = 0;
110
111 /*
112 * Count the number of vdevice entries with an
113 * ibm,my-dma-window OF property
114 */
115 node_vroot = of_find_node_by_name(NULL, "vdevice");
116 if (node_vroot) {
117 struct device_node *of_node;
118 struct property *prop;
119
120 for_each_child_of_node(node_vroot, of_node) {
121 prop = of_find_property(of_node, "ibm,my-dma-window",
122 NULL);
123 if (prop)
124 count++;
125 }
126 }
127 of_node_put(node_vroot);
128 return count;
129}
130
131/**
132 * vio_cmo_alloc - allocate IO memory for CMO-enable devices
133 *
134 * @viodev: VIO device requesting IO memory
135 * @size: size of allocation requested
136 *
137 * Allocations come from memory reserved for the devices and any excess
138 * IO memory available to all devices. The spare pool used to service
139 * hotplug must be equal to %VIO_CMO_MIN_ENT for the excess pool to be
140 * made available.
141 *
142 * Return codes:
143 * 0 for successful allocation and -ENOMEM for a failure
144 */
145static inline int vio_cmo_alloc(struct vio_dev *viodev, size_t size)
146{
147 unsigned long flags;
148 size_t reserve_free = 0;
149 size_t excess_free = 0;
150 int ret = -ENOMEM;
151
152 spin_lock_irqsave(&vio_cmo.lock, flags);
153
154 /* Determine the amount of free entitlement available in reserve */
155 if (viodev->cmo.entitled > viodev->cmo.allocated)
156 reserve_free = viodev->cmo.entitled - viodev->cmo.allocated;
157
158 /* If spare is not fulfilled, the excess pool can not be used. */
159 if (vio_cmo.spare >= VIO_CMO_MIN_ENT)
160 excess_free = vio_cmo.excess.free;
161
162 /* The request can be satisfied */
163 if ((reserve_free + excess_free) >= size) {
164 vio_cmo.curr += size;
165 if (vio_cmo.curr > vio_cmo.high)
166 vio_cmo.high = vio_cmo.curr;
167 viodev->cmo.allocated += size;
168 size -= min(reserve_free, size);
169 vio_cmo.excess.free -= size;
170 ret = 0;
171 }
172
173 spin_unlock_irqrestore(&vio_cmo.lock, flags);
174 return ret;
175}
176
177/**
178 * vio_cmo_dealloc - deallocate IO memory from CMO-enable devices
179 * @viodev: VIO device freeing IO memory
180 * @size: size of deallocation
181 *
182 * IO memory is freed by the device back to the correct memory pools.
183 * The spare pool is replenished first from either memory pool, then
184 * the reserve pool is used to reduce device entitlement, the excess
185 * pool is used to increase the reserve pool toward the desired entitlement
186 * target, and then the remaining memory is returned to the pools.
187 *
188 */
189static inline void vio_cmo_dealloc(struct vio_dev *viodev, size_t size)
190{
191 unsigned long flags;
192 size_t spare_needed = 0;
193 size_t excess_freed = 0;
194 size_t reserve_freed = size;
195 size_t tmp;
196 int balance = 0;
197
198 spin_lock_irqsave(&vio_cmo.lock, flags);
199 vio_cmo.curr -= size;
200
201 /* Amount of memory freed from the excess pool */
202 if (viodev->cmo.allocated > viodev->cmo.entitled) {
203 excess_freed = min(reserve_freed, (viodev->cmo.allocated -
204 viodev->cmo.entitled));
205 reserve_freed -= excess_freed;
206 }
207
208 /* Remove allocation from device */
209 viodev->cmo.allocated -= (reserve_freed + excess_freed);
210
211 /* Spare is a subset of the reserve pool, replenish it first. */
212 spare_needed = VIO_CMO_MIN_ENT - vio_cmo.spare;
213
214 /*
215 * Replenish the spare in the reserve pool from the excess pool.
216 * This moves entitlement into the reserve pool.
217 */
218 if (spare_needed && excess_freed) {
219 tmp = min(excess_freed, spare_needed);
220 vio_cmo.excess.size -= tmp;
221 vio_cmo.reserve.size += tmp;
222 vio_cmo.spare += tmp;
223 excess_freed -= tmp;
224 spare_needed -= tmp;
225 balance = 1;
226 }
227
228 /*
229 * Replenish the spare in the reserve pool from the reserve pool.
230 * This removes entitlement from the device down to VIO_CMO_MIN_ENT,
231 * if needed, and gives it to the spare pool. The amount of used
232 * memory in this pool does not change.
233 */
234 if (spare_needed && reserve_freed) {
235 tmp = min3(spare_needed, reserve_freed, (viodev->cmo.entitled - VIO_CMO_MIN_ENT));
236
237 vio_cmo.spare += tmp;
238 viodev->cmo.entitled -= tmp;
239 reserve_freed -= tmp;
240 spare_needed -= tmp;
241 balance = 1;
242 }
243
244 /*
245 * Increase the reserve pool until the desired allocation is met.
246 * Move an allocation freed from the excess pool into the reserve
247 * pool and schedule a balance operation.
248 */
249 if (excess_freed && (vio_cmo.desired > vio_cmo.reserve.size)) {
250 tmp = min(excess_freed, (vio_cmo.desired - vio_cmo.reserve.size));
251
252 vio_cmo.excess.size -= tmp;
253 vio_cmo.reserve.size += tmp;
254 excess_freed -= tmp;
255 balance = 1;
256 }
257
258 /* Return memory from the excess pool to that pool */
259 if (excess_freed)
260 vio_cmo.excess.free += excess_freed;
261
262 if (balance)
263 schedule_delayed_work(&vio_cmo.balance_q, VIO_CMO_BALANCE_DELAY);
264 spin_unlock_irqrestore(&vio_cmo.lock, flags);
265}
266
267/**
268 * vio_cmo_entitlement_update - Manage system entitlement changes
269 *
270 * @new_entitlement: new system entitlement to attempt to accommodate
271 *
272 * Increases in entitlement will be used to fulfill the spare entitlement
273 * and the rest is given to the excess pool. Decreases, if they are
274 * possible, come from the excess pool and from unused device entitlement
275 *
276 * Returns: 0 on success, -ENOMEM when change can not be made
277 */
278int vio_cmo_entitlement_update(size_t new_entitlement)
279{
280 struct vio_dev *viodev;
281 struct vio_cmo_dev_entry *dev_ent;
282 unsigned long flags;
283 size_t avail, delta, tmp;
284
285 spin_lock_irqsave(&vio_cmo.lock, flags);
286
287 /* Entitlement increases */
288 if (new_entitlement > vio_cmo.entitled) {
289 delta = new_entitlement - vio_cmo.entitled;
290
291 /* Fulfill spare allocation */
292 if (vio_cmo.spare < VIO_CMO_MIN_ENT) {
293 tmp = min(delta, (VIO_CMO_MIN_ENT - vio_cmo.spare));
294 vio_cmo.spare += tmp;
295 vio_cmo.reserve.size += tmp;
296 delta -= tmp;
297 }
298
299 /* Remaining new allocation goes to the excess pool */
300 vio_cmo.entitled += delta;
301 vio_cmo.excess.size += delta;
302 vio_cmo.excess.free += delta;
303
304 goto out;
305 }
306
307 /* Entitlement decreases */
308 delta = vio_cmo.entitled - new_entitlement;
309 avail = vio_cmo.excess.free;
310
311 /*
312 * Need to check how much unused entitlement each device can
313 * sacrifice to fulfill entitlement change.
314 */
315 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
316 if (avail >= delta)
317 break;
318
319 viodev = dev_ent->viodev;
320 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
321 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
322 avail += viodev->cmo.entitled -
323 max_t(size_t, viodev->cmo.allocated,
324 VIO_CMO_MIN_ENT);
325 }
326
327 if (delta <= avail) {
328 vio_cmo.entitled -= delta;
329
330 /* Take entitlement from the excess pool first */
331 tmp = min(vio_cmo.excess.free, delta);
332 vio_cmo.excess.size -= tmp;
333 vio_cmo.excess.free -= tmp;
334 delta -= tmp;
335
336 /*
337 * Remove all but VIO_CMO_MIN_ENT bytes from devices
338 * until entitlement change is served
339 */
340 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
341 if (!delta)
342 break;
343
344 viodev = dev_ent->viodev;
345 tmp = 0;
346 if ((viodev->cmo.entitled > viodev->cmo.allocated) &&
347 (viodev->cmo.entitled > VIO_CMO_MIN_ENT))
348 tmp = viodev->cmo.entitled -
349 max_t(size_t, viodev->cmo.allocated,
350 VIO_CMO_MIN_ENT);
351 viodev->cmo.entitled -= min(tmp, delta);
352 delta -= min(tmp, delta);
353 }
354 } else {
355 spin_unlock_irqrestore(&vio_cmo.lock, flags);
356 return -ENOMEM;
357 }
358
359out:
360 schedule_delayed_work(&vio_cmo.balance_q, 0);
361 spin_unlock_irqrestore(&vio_cmo.lock, flags);
362 return 0;
363}
364
365/**
366 * vio_cmo_balance - Balance entitlement among devices
367 *
368 * @work: work queue structure for this operation
369 *
370 * Any system entitlement above the minimum needed for devices, or
371 * already allocated to devices, can be distributed to the devices.
372 * The list of devices is iterated through to recalculate the desired
373 * entitlement level and to determine how much entitlement above the
374 * minimum entitlement is allocated to devices.
375 *
376 * Small chunks of the available entitlement are given to devices until
377 * their requirements are fulfilled or there is no entitlement left to give.
378 * Upon completion sizes of the reserve and excess pools are calculated.
379 *
380 * The system minimum entitlement level is also recalculated here.
381 * Entitlement will be reserved for devices even after vio_bus_remove to
382 * accommodate reloading the driver. The OF tree is walked to count the
383 * number of devices present and this will remove entitlement for devices
384 * that have actually left the system after having vio_bus_remove called.
385 */
386static void vio_cmo_balance(struct work_struct *work)
387{
388 struct vio_cmo *cmo;
389 struct vio_dev *viodev;
390 struct vio_cmo_dev_entry *dev_ent;
391 unsigned long flags;
392 size_t avail = 0, level, chunk, need;
393 int devcount = 0, fulfilled;
394
395 cmo = container_of(work, struct vio_cmo, balance_q.work);
396
397 spin_lock_irqsave(&vio_cmo.lock, flags);
398
399 /* Calculate minimum entitlement and fulfill spare */
400 cmo->min = vio_cmo_num_OF_devs() * VIO_CMO_MIN_ENT;
401 BUG_ON(cmo->min > cmo->entitled);
402 cmo->spare = min_t(size_t, VIO_CMO_MIN_ENT, (cmo->entitled - cmo->min));
403 cmo->min += cmo->spare;
404 cmo->desired = cmo->min;
405
406 /*
407 * Determine how much entitlement is available and reset device
408 * entitlements
409 */
410 avail = cmo->entitled - cmo->spare;
411 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
412 viodev = dev_ent->viodev;
413 devcount++;
414 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
415 cmo->desired += (viodev->cmo.desired - VIO_CMO_MIN_ENT);
416 avail -= max_t(size_t, viodev->cmo.allocated, VIO_CMO_MIN_ENT);
417 }
418
419 /*
420 * Having provided each device with the minimum entitlement, loop
421 * over the devices portioning out the remaining entitlement
422 * until there is nothing left.
423 */
424 level = VIO_CMO_MIN_ENT;
425 while (avail) {
426 fulfilled = 0;
427 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
428 viodev = dev_ent->viodev;
429
430 if (viodev->cmo.desired <= level) {
431 fulfilled++;
432 continue;
433 }
434
435 /*
436 * Give the device up to VIO_CMO_BALANCE_CHUNK
437 * bytes of entitlement, but do not exceed the
438 * desired level of entitlement for the device.
439 */
440 chunk = min_t(size_t, avail, VIO_CMO_BALANCE_CHUNK);
441 chunk = min(chunk, (viodev->cmo.desired -
442 viodev->cmo.entitled));
443 viodev->cmo.entitled += chunk;
444
445 /*
446 * If the memory for this entitlement increase was
447 * already allocated to the device it does not come
448 * from the available pool being portioned out.
449 */
450 need = max(viodev->cmo.allocated, viodev->cmo.entitled)-
451 max(viodev->cmo.allocated, level);
452 avail -= need;
453
454 }
455 if (fulfilled == devcount)
456 break;
457 level += VIO_CMO_BALANCE_CHUNK;
458 }
459
460 /* Calculate new reserve and excess pool sizes */
461 cmo->reserve.size = cmo->min;
462 cmo->excess.free = 0;
463 cmo->excess.size = 0;
464 need = 0;
465 list_for_each_entry(dev_ent, &vio_cmo.device_list, list) {
466 viodev = dev_ent->viodev;
467 /* Calculated reserve size above the minimum entitlement */
468 if (viodev->cmo.entitled)
469 cmo->reserve.size += (viodev->cmo.entitled -
470 VIO_CMO_MIN_ENT);
471 /* Calculated used excess entitlement */
472 if (viodev->cmo.allocated > viodev->cmo.entitled)
473 need += viodev->cmo.allocated - viodev->cmo.entitled;
474 }
475 cmo->excess.size = cmo->entitled - cmo->reserve.size;
476 cmo->excess.free = cmo->excess.size - need;
477
478 cancel_delayed_work(to_delayed_work(work));
479 spin_unlock_irqrestore(&vio_cmo.lock, flags);
480}
481
482static void *vio_dma_iommu_alloc_coherent(struct device *dev, size_t size,
483 dma_addr_t *dma_handle, gfp_t flag,
484 unsigned long attrs)
485{
486 struct vio_dev *viodev = to_vio_dev(dev);
487 void *ret;
488
489 if (vio_cmo_alloc(viodev, roundup(size, PAGE_SIZE))) {
490 atomic_inc(&viodev->cmo.allocs_failed);
491 return NULL;
492 }
493
494 ret = iommu_alloc_coherent(dev, get_iommu_table_base(dev), size,
495 dma_handle, dev->coherent_dma_mask, flag,
496 dev_to_node(dev));
497 if (unlikely(ret == NULL)) {
498 vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
499 atomic_inc(&viodev->cmo.allocs_failed);
500 }
501
502 return ret;
503}
504
505static void vio_dma_iommu_free_coherent(struct device *dev, size_t size,
506 void *vaddr, dma_addr_t dma_handle,
507 unsigned long attrs)
508{
509 struct vio_dev *viodev = to_vio_dev(dev);
510
511 iommu_free_coherent(get_iommu_table_base(dev), size, vaddr, dma_handle);
512 vio_cmo_dealloc(viodev, roundup(size, PAGE_SIZE));
513}
514
515static 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 = get_iommu_table_base(dev);
522 dma_addr_t ret = DMA_MAPPING_ERROR;
523
524 if (vio_cmo_alloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl))))
525 goto out_fail;
526 ret = iommu_map_page(dev, tbl, page, offset, size, dma_get_mask(dev),
527 direction, attrs);
528 if (unlikely(ret == DMA_MAPPING_ERROR))
529 goto out_deallocate;
530 return ret;
531
532out_deallocate:
533 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)));
534out_fail:
535 atomic_inc(&viodev->cmo.allocs_failed);
536 return DMA_MAPPING_ERROR;
537}
538
539static 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 = get_iommu_table_base(dev);
546
547 iommu_unmap_page(tbl, dma_handle, size, direction, attrs);
548 vio_cmo_dealloc(viodev, roundup(size, IOMMU_PAGE_SIZE(tbl)));
549}
550
551static int vio_dma_iommu_map_sg(struct device *dev, struct scatterlist *sglist,
552 int nelems, enum dma_data_direction direction,
553 unsigned long attrs)
554{
555 struct vio_dev *viodev = to_vio_dev(dev);
556 struct iommu_table *tbl = get_iommu_table_base(dev);
557 struct scatterlist *sgl;
558 int ret, count;
559 size_t alloc_size = 0;
560
561 for_each_sg(sglist, sgl, nelems, count)
562 alloc_size += roundup(sgl->length, IOMMU_PAGE_SIZE(tbl));
563
564 ret = vio_cmo_alloc(viodev, alloc_size);
565 if (ret)
566 goto out_fail;
567 ret = ppc_iommu_map_sg(dev, tbl, sglist, nelems, dma_get_mask(dev),
568 direction, attrs);
569 if (unlikely(!ret))
570 goto out_deallocate;
571
572 for_each_sg(sglist, sgl, ret, count)
573 alloc_size -= roundup(sgl->dma_length, IOMMU_PAGE_SIZE(tbl));
574 if (alloc_size)
575 vio_cmo_dealloc(viodev, alloc_size);
576 return ret;
577
578out_deallocate:
579 vio_cmo_dealloc(viodev, alloc_size);
580out_fail:
581 atomic_inc(&viodev->cmo.allocs_failed);
582 return ret;
583}
584
585static void vio_dma_iommu_unmap_sg(struct device *dev,
586 struct scatterlist *sglist, int nelems,
587 enum dma_data_direction direction,
588 unsigned long attrs)
589{
590 struct vio_dev *viodev = to_vio_dev(dev);
591 struct iommu_table *tbl = get_iommu_table_base(dev);
592 struct scatterlist *sgl;
593 size_t alloc_size = 0;
594 int count;
595
596 for_each_sg(sglist, sgl, nelems, count)
597 alloc_size += roundup(sgl->dma_length, IOMMU_PAGE_SIZE(tbl));
598
599 ppc_iommu_unmap_sg(tbl, sglist, nelems, direction, attrs);
600 vio_cmo_dealloc(viodev, alloc_size);
601}
602
603static const struct dma_map_ops vio_dma_mapping_ops = {
604 .alloc = vio_dma_iommu_alloc_coherent,
605 .free = vio_dma_iommu_free_coherent,
606 .map_sg = vio_dma_iommu_map_sg,
607 .unmap_sg = vio_dma_iommu_unmap_sg,
608 .map_page = vio_dma_iommu_map_page,
609 .unmap_page = vio_dma_iommu_unmap_page,
610 .dma_supported = dma_iommu_dma_supported,
611 .get_required_mask = dma_iommu_get_required_mask,
612 .mmap = dma_common_mmap,
613 .get_sgtable = dma_common_get_sgtable,
614 .alloc_pages = dma_common_alloc_pages,
615 .free_pages = dma_common_free_pages,
616};
617
618/**
619 * vio_cmo_set_dev_desired - Set desired entitlement for a device
620 *
621 * @viodev: struct vio_dev for device to alter
622 * @desired: new desired entitlement level in bytes
623 *
624 * For use by devices to request a change to their entitlement at runtime or
625 * through sysfs. The desired entitlement level is changed and a balancing
626 * of system resources is scheduled to run in the future.
627 */
628void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired)
629{
630 unsigned long flags;
631 struct vio_cmo_dev_entry *dev_ent;
632 int found = 0;
633
634 if (!firmware_has_feature(FW_FEATURE_CMO))
635 return;
636
637 spin_lock_irqsave(&vio_cmo.lock, flags);
638 if (desired < VIO_CMO_MIN_ENT)
639 desired = VIO_CMO_MIN_ENT;
640
641 /*
642 * Changes will not be made for devices not in the device list.
643 * If it is not in the device list, then no driver is loaded
644 * for the device and it can not receive entitlement.
645 */
646 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
647 if (viodev == dev_ent->viodev) {
648 found = 1;
649 break;
650 }
651 if (!found) {
652 spin_unlock_irqrestore(&vio_cmo.lock, flags);
653 return;
654 }
655
656 /* Increase/decrease in desired device entitlement */
657 if (desired >= viodev->cmo.desired) {
658 /* Just bump the bus and device values prior to a balance*/
659 vio_cmo.desired += desired - viodev->cmo.desired;
660 viodev->cmo.desired = desired;
661 } else {
662 /* Decrease bus and device values for desired entitlement */
663 vio_cmo.desired -= viodev->cmo.desired - desired;
664 viodev->cmo.desired = desired;
665 /*
666 * If less entitlement is desired than current entitlement, move
667 * any reserve memory in the change region to the excess pool.
668 */
669 if (viodev->cmo.entitled > desired) {
670 vio_cmo.reserve.size -= viodev->cmo.entitled - desired;
671 vio_cmo.excess.size += viodev->cmo.entitled - desired;
672 /*
673 * If entitlement moving from the reserve pool to the
674 * excess pool is currently unused, add to the excess
675 * free counter.
676 */
677 if (viodev->cmo.allocated < viodev->cmo.entitled)
678 vio_cmo.excess.free += viodev->cmo.entitled -
679 max(viodev->cmo.allocated, desired);
680 viodev->cmo.entitled = desired;
681 }
682 }
683 schedule_delayed_work(&vio_cmo.balance_q, 0);
684 spin_unlock_irqrestore(&vio_cmo.lock, flags);
685}
686
687/**
688 * vio_cmo_bus_probe - Handle CMO specific bus probe activities
689 *
690 * @viodev - Pointer to struct vio_dev for device
691 *
692 * Determine the devices IO memory entitlement needs, attempting
693 * to satisfy the system minimum entitlement at first and scheduling
694 * a balance operation to take care of the rest at a later time.
695 *
696 * Returns: 0 on success, -EINVAL when device doesn't support CMO, and
697 * -ENOMEM when entitlement is not available for device or
698 * device entry.
699 *
700 */
701static int vio_cmo_bus_probe(struct vio_dev *viodev)
702{
703 struct vio_cmo_dev_entry *dev_ent;
704 struct device *dev = &viodev->dev;
705 struct iommu_table *tbl;
706 struct vio_driver *viodrv = to_vio_driver(dev->driver);
707 unsigned long flags;
708 size_t size;
709 bool dma_capable = false;
710
711 tbl = get_iommu_table_base(dev);
712
713 /* A device requires entitlement if it has a DMA window property */
714 switch (viodev->family) {
715 case VDEVICE:
716 if (of_get_property(viodev->dev.of_node,
717 "ibm,my-dma-window", NULL))
718 dma_capable = true;
719 break;
720 case PFO:
721 dma_capable = false;
722 break;
723 default:
724 dev_warn(dev, "unknown device family: %d\n", viodev->family);
725 BUG();
726 break;
727 }
728
729 /* Configure entitlement for the device. */
730 if (dma_capable) {
731 /* Check that the driver is CMO enabled and get desired DMA */
732 if (!viodrv->get_desired_dma) {
733 dev_err(dev, "%s: device driver does not support CMO\n",
734 __func__);
735 return -EINVAL;
736 }
737
738 viodev->cmo.desired =
739 IOMMU_PAGE_ALIGN(viodrv->get_desired_dma(viodev), tbl);
740 if (viodev->cmo.desired < VIO_CMO_MIN_ENT)
741 viodev->cmo.desired = VIO_CMO_MIN_ENT;
742 size = VIO_CMO_MIN_ENT;
743
744 dev_ent = kmalloc(sizeof(struct vio_cmo_dev_entry),
745 GFP_KERNEL);
746 if (!dev_ent)
747 return -ENOMEM;
748
749 dev_ent->viodev = viodev;
750 spin_lock_irqsave(&vio_cmo.lock, flags);
751 list_add(&dev_ent->list, &vio_cmo.device_list);
752 } else {
753 viodev->cmo.desired = 0;
754 size = 0;
755 spin_lock_irqsave(&vio_cmo.lock, flags);
756 }
757
758 /*
759 * If the needs for vio_cmo.min have not changed since they
760 * were last set, the number of devices in the OF tree has
761 * been constant and the IO memory for this is already in
762 * the reserve pool.
763 */
764 if (vio_cmo.min == ((vio_cmo_num_OF_devs() + 1) *
765 VIO_CMO_MIN_ENT)) {
766 /* Updated desired entitlement if device requires it */
767 if (size)
768 vio_cmo.desired += (viodev->cmo.desired -
769 VIO_CMO_MIN_ENT);
770 } else {
771 size_t tmp;
772
773 tmp = vio_cmo.spare + vio_cmo.excess.free;
774 if (tmp < size) {
775 dev_err(dev, "%s: insufficient free "
776 "entitlement to add device. "
777 "Need %lu, have %lu\n", __func__,
778 size, (vio_cmo.spare + tmp));
779 spin_unlock_irqrestore(&vio_cmo.lock, flags);
780 return -ENOMEM;
781 }
782
783 /* Use excess pool first to fulfill request */
784 tmp = min(size, vio_cmo.excess.free);
785 vio_cmo.excess.free -= tmp;
786 vio_cmo.excess.size -= tmp;
787 vio_cmo.reserve.size += tmp;
788
789 /* Use spare if excess pool was insufficient */
790 vio_cmo.spare -= size - tmp;
791
792 /* Update bus accounting */
793 vio_cmo.min += size;
794 vio_cmo.desired += viodev->cmo.desired;
795 }
796 spin_unlock_irqrestore(&vio_cmo.lock, flags);
797 return 0;
798}
799
800/**
801 * vio_cmo_bus_remove - Handle CMO specific bus removal activities
802 *
803 * @viodev - Pointer to struct vio_dev for device
804 *
805 * Remove the device from the cmo device list. The minimum entitlement
806 * will be reserved for the device as long as it is in the system. The
807 * rest of the entitlement the device had been allocated will be returned
808 * to the system.
809 */
810static void vio_cmo_bus_remove(struct vio_dev *viodev)
811{
812 struct vio_cmo_dev_entry *dev_ent;
813 unsigned long flags;
814 size_t tmp;
815
816 spin_lock_irqsave(&vio_cmo.lock, flags);
817 if (viodev->cmo.allocated) {
818 dev_err(&viodev->dev, "%s: device had %lu bytes of IO "
819 "allocated after remove operation.\n",
820 __func__, viodev->cmo.allocated);
821 BUG();
822 }
823
824 /*
825 * Remove the device from the device list being maintained for
826 * CMO enabled devices.
827 */
828 list_for_each_entry(dev_ent, &vio_cmo.device_list, list)
829 if (viodev == dev_ent->viodev) {
830 list_del(&dev_ent->list);
831 kfree(dev_ent);
832 break;
833 }
834
835 /*
836 * Devices may not require any entitlement and they do not need
837 * to be processed. Otherwise, return the device's entitlement
838 * back to the pools.
839 */
840 if (viodev->cmo.entitled) {
841 /*
842 * This device has not yet left the OF tree, it's
843 * minimum entitlement remains in vio_cmo.min and
844 * vio_cmo.desired
845 */
846 vio_cmo.desired -= (viodev->cmo.desired - VIO_CMO_MIN_ENT);
847
848 /*
849 * Save min allocation for device in reserve as long
850 * as it exists in OF tree as determined by later
851 * balance operation
852 */
853 viodev->cmo.entitled -= VIO_CMO_MIN_ENT;
854
855 /* Replenish spare from freed reserve pool */
856 if (viodev->cmo.entitled && (vio_cmo.spare < VIO_CMO_MIN_ENT)) {
857 tmp = min(viodev->cmo.entitled, (VIO_CMO_MIN_ENT -
858 vio_cmo.spare));
859 vio_cmo.spare += tmp;
860 viodev->cmo.entitled -= tmp;
861 }
862
863 /* Remaining reserve goes to excess pool */
864 vio_cmo.excess.size += viodev->cmo.entitled;
865 vio_cmo.excess.free += viodev->cmo.entitled;
866 vio_cmo.reserve.size -= viodev->cmo.entitled;
867
868 /*
869 * Until the device is removed it will keep a
870 * minimum entitlement; this will guarantee that
871 * a module unload/load will result in a success.
872 */
873 viodev->cmo.entitled = VIO_CMO_MIN_ENT;
874 viodev->cmo.desired = VIO_CMO_MIN_ENT;
875 atomic_set(&viodev->cmo.allocs_failed, 0);
876 }
877
878 spin_unlock_irqrestore(&vio_cmo.lock, flags);
879}
880
881static void vio_cmo_set_dma_ops(struct vio_dev *viodev)
882{
883 set_dma_ops(&viodev->dev, &vio_dma_mapping_ops);
884}
885
886/**
887 * vio_cmo_bus_init - CMO entitlement initialization at bus init time
888 *
889 * Set up the reserve and excess entitlement pools based on available
890 * system entitlement and the number of devices in the OF tree that
891 * require entitlement in the reserve pool.
892 */
893static void vio_cmo_bus_init(void)
894{
895 struct hvcall_mpp_data mpp_data;
896 int err;
897
898 memset(&vio_cmo, 0, sizeof(struct vio_cmo));
899 spin_lock_init(&vio_cmo.lock);
900 INIT_LIST_HEAD(&vio_cmo.device_list);
901 INIT_DELAYED_WORK(&vio_cmo.balance_q, vio_cmo_balance);
902
903 /* Get current system entitlement */
904 err = h_get_mpp(&mpp_data);
905
906 /*
907 * On failure, continue with entitlement set to 0, will panic()
908 * later when spare is reserved.
909 */
910 if (err != H_SUCCESS) {
911 printk(KERN_ERR "%s: unable to determine system IO "\
912 "entitlement. (%d)\n", __func__, err);
913 vio_cmo.entitled = 0;
914 } else {
915 vio_cmo.entitled = mpp_data.entitled_mem;
916 }
917
918 /* Set reservation and check against entitlement */
919 vio_cmo.spare = VIO_CMO_MIN_ENT;
920 vio_cmo.reserve.size = vio_cmo.spare;
921 vio_cmo.reserve.size += (vio_cmo_num_OF_devs() *
922 VIO_CMO_MIN_ENT);
923 if (vio_cmo.reserve.size > vio_cmo.entitled) {
924 printk(KERN_ERR "%s: insufficient system entitlement\n",
925 __func__);
926 panic("%s: Insufficient system entitlement", __func__);
927 }
928
929 /* Set the remaining accounting variables */
930 vio_cmo.excess.size = vio_cmo.entitled - vio_cmo.reserve.size;
931 vio_cmo.excess.free = vio_cmo.excess.size;
932 vio_cmo.min = vio_cmo.reserve.size;
933 vio_cmo.desired = vio_cmo.reserve.size;
934}
935
936/* sysfs device functions and data structures for CMO */
937
938#define viodev_cmo_rd_attr(name) \
939static ssize_t cmo_##name##_show(struct device *dev, \
940 struct device_attribute *attr, \
941 char *buf) \
942{ \
943 return sprintf(buf, "%lu\n", to_vio_dev(dev)->cmo.name); \
944}
945
946static ssize_t cmo_allocs_failed_show(struct device *dev,
947 struct device_attribute *attr, char *buf)
948{
949 struct vio_dev *viodev = to_vio_dev(dev);
950 return sprintf(buf, "%d\n", atomic_read(&viodev->cmo.allocs_failed));
951}
952
953static ssize_t cmo_allocs_failed_store(struct device *dev,
954 struct device_attribute *attr, const char *buf, size_t count)
955{
956 struct vio_dev *viodev = to_vio_dev(dev);
957 atomic_set(&viodev->cmo.allocs_failed, 0);
958 return count;
959}
960
961static ssize_t cmo_desired_store(struct device *dev,
962 struct device_attribute *attr, const char *buf, size_t count)
963{
964 struct vio_dev *viodev = to_vio_dev(dev);
965 size_t new_desired;
966 int ret;
967
968 ret = kstrtoul(buf, 10, &new_desired);
969 if (ret)
970 return ret;
971
972 vio_cmo_set_dev_desired(viodev, new_desired);
973 return count;
974}
975
976viodev_cmo_rd_attr(desired);
977viodev_cmo_rd_attr(entitled);
978viodev_cmo_rd_attr(allocated);
979
980static ssize_t name_show(struct device *, struct device_attribute *, char *);
981static ssize_t devspec_show(struct device *, struct device_attribute *, char *);
982static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
983 char *buf);
984
985static struct device_attribute dev_attr_name;
986static struct device_attribute dev_attr_devspec;
987static struct device_attribute dev_attr_modalias;
988
989static DEVICE_ATTR_RO(cmo_entitled);
990static DEVICE_ATTR_RO(cmo_allocated);
991static DEVICE_ATTR_RW(cmo_desired);
992static DEVICE_ATTR_RW(cmo_allocs_failed);
993
994/* sysfs bus functions and data structures for CMO */
995
996#define viobus_cmo_rd_attr(name) \
997static ssize_t cmo_bus_##name##_show(const struct bus_type *bt, char *buf) \
998{ \
999 return sprintf(buf, "%lu\n", vio_cmo.name); \
1000} \
1001static struct bus_attribute bus_attr_cmo_bus_##name = \
1002 __ATTR(cmo_##name, S_IRUGO, cmo_bus_##name##_show, NULL)
1003
1004#define viobus_cmo_pool_rd_attr(name, var) \
1005static ssize_t \
1006cmo_##name##_##var##_show(const struct bus_type *bt, char *buf) \
1007{ \
1008 return sprintf(buf, "%lu\n", vio_cmo.name.var); \
1009} \
1010static BUS_ATTR_RO(cmo_##name##_##var)
1011
1012viobus_cmo_rd_attr(entitled);
1013viobus_cmo_rd_attr(spare);
1014viobus_cmo_rd_attr(min);
1015viobus_cmo_rd_attr(desired);
1016viobus_cmo_rd_attr(curr);
1017viobus_cmo_pool_rd_attr(reserve, size);
1018viobus_cmo_pool_rd_attr(excess, size);
1019viobus_cmo_pool_rd_attr(excess, free);
1020
1021static ssize_t cmo_high_show(const struct bus_type *bt, char *buf)
1022{
1023 return sprintf(buf, "%lu\n", vio_cmo.high);
1024}
1025
1026static ssize_t cmo_high_store(const struct bus_type *bt, const char *buf,
1027 size_t count)
1028{
1029 unsigned long flags;
1030
1031 spin_lock_irqsave(&vio_cmo.lock, flags);
1032 vio_cmo.high = vio_cmo.curr;
1033 spin_unlock_irqrestore(&vio_cmo.lock, flags);
1034
1035 return count;
1036}
1037static BUS_ATTR_RW(cmo_high);
1038
1039static struct attribute *vio_bus_attrs[] = {
1040 &bus_attr_cmo_bus_entitled.attr,
1041 &bus_attr_cmo_bus_spare.attr,
1042 &bus_attr_cmo_bus_min.attr,
1043 &bus_attr_cmo_bus_desired.attr,
1044 &bus_attr_cmo_bus_curr.attr,
1045 &bus_attr_cmo_high.attr,
1046 &bus_attr_cmo_reserve_size.attr,
1047 &bus_attr_cmo_excess_size.attr,
1048 &bus_attr_cmo_excess_free.attr,
1049 NULL,
1050};
1051ATTRIBUTE_GROUPS(vio_bus);
1052
1053static void __init vio_cmo_sysfs_init(void) { }
1054#else /* CONFIG_PPC_SMLPAR */
1055int vio_cmo_entitlement_update(size_t new_entitlement) { return 0; }
1056void vio_cmo_set_dev_desired(struct vio_dev *viodev, size_t desired) {}
1057static int vio_cmo_bus_probe(struct vio_dev *viodev) { return 0; }
1058static void vio_cmo_bus_remove(struct vio_dev *viodev) {}
1059static void vio_cmo_set_dma_ops(struct vio_dev *viodev) {}
1060static void vio_cmo_bus_init(void) {}
1061static void __init vio_cmo_sysfs_init(void) { }
1062#endif /* CONFIG_PPC_SMLPAR */
1063EXPORT_SYMBOL(vio_cmo_entitlement_update);
1064EXPORT_SYMBOL(vio_cmo_set_dev_desired);
1065
1066
1067/*
1068 * Platform Facilities Option (PFO) support
1069 */
1070
1071/**
1072 * vio_h_cop_sync - Perform a synchronous PFO co-processor operation
1073 *
1074 * @vdev - Pointer to a struct vio_dev for device
1075 * @op - Pointer to a struct vio_pfo_op for the operation parameters
1076 *
1077 * Calls the hypervisor to synchronously perform the PFO operation
1078 * described in @op. In the case of a busy response from the hypervisor,
1079 * the operation will be re-submitted indefinitely unless a non-zero timeout
1080 * is specified or an error occurs. The timeout places a limit on when to
1081 * stop re-submitting a operation, the total time can be exceeded if an
1082 * operation is in progress.
1083 *
1084 * If op->hcall_ret is not NULL, this will be set to the return from the
1085 * last h_cop_op call or it will be 0 if an error not involving the h_call
1086 * was encountered.
1087 *
1088 * Returns:
1089 * 0 on success,
1090 * -EINVAL if the h_call fails due to an invalid parameter,
1091 * -E2BIG if the h_call can not be performed synchronously,
1092 * -EBUSY if a timeout is specified and has elapsed,
1093 * -EACCES if the memory area for data/status has been rescinded, or
1094 * -EPERM if a hardware fault has been indicated
1095 */
1096int vio_h_cop_sync(struct vio_dev *vdev, struct vio_pfo_op *op)
1097{
1098 struct device *dev = &vdev->dev;
1099 unsigned long deadline = 0;
1100 long hret = 0;
1101 int ret = 0;
1102
1103 if (op->timeout)
1104 deadline = jiffies + msecs_to_jiffies(m: op->timeout);
1105
1106 while (true) {
1107 hret = plpar_hcall_norets(H_COP, op->flags,
1108 vdev->resource_id,
1109 op->in, op->inlen, op->out,
1110 op->outlen, op->csbcpb);
1111
1112 if (hret == H_SUCCESS ||
1113 (hret != H_NOT_ENOUGH_RESOURCES &&
1114 hret != H_BUSY && hret != H_RESOURCE) ||
1115 (op->timeout && time_after(deadline, jiffies)))
1116 break;
1117
1118 dev_dbg(dev, "%s: hcall ret(%ld), retrying.\n", __func__, hret);
1119 }
1120
1121 switch (hret) {
1122 case H_SUCCESS:
1123 ret = 0;
1124 break;
1125 case H_OP_MODE:
1126 case H_TOO_BIG:
1127 ret = -E2BIG;
1128 break;
1129 case H_RESCINDED:
1130 ret = -EACCES;
1131 break;
1132 case H_HARDWARE:
1133 ret = -EPERM;
1134 break;
1135 case H_NOT_ENOUGH_RESOURCES:
1136 case H_RESOURCE:
1137 case H_BUSY:
1138 ret = -EBUSY;
1139 break;
1140 default:
1141 ret = -EINVAL;
1142 break;
1143 }
1144
1145 if (ret)
1146 dev_dbg(dev, "%s: Sync h_cop_op failure (ret:%d) (hret:%ld)\n",
1147 __func__, ret, hret);
1148
1149 op->hcall_err = hret;
1150 return ret;
1151}
1152EXPORT_SYMBOL(vio_h_cop_sync);
1153
1154static struct iommu_table *vio_build_iommu_table(struct vio_dev *dev)
1155{
1156 const __be32 *dma_window;
1157 struct iommu_table *tbl;
1158 unsigned long offset, size;
1159
1160 dma_window = of_get_property(node: dev->dev.of_node,
1161 name: "ibm,my-dma-window", NULL);
1162 if (!dma_window)
1163 return NULL;
1164
1165 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
1166 if (tbl == NULL)
1167 return NULL;
1168
1169 kref_init(kref: &tbl->it_kref);
1170
1171 of_parse_dma_window(dev->dev.of_node, dma_window,
1172 &tbl->it_index, &offset, &size);
1173
1174 /* TCE table size - measured in tce entries */
1175 tbl->it_page_shift = IOMMU_PAGE_SHIFT_4K;
1176 tbl->it_size = size >> tbl->it_page_shift;
1177 /* offset for VIO should always be 0 */
1178 tbl->it_offset = offset >> tbl->it_page_shift;
1179 tbl->it_busno = 0;
1180 tbl->it_type = TCE_VB;
1181 tbl->it_blocksize = 16;
1182
1183 if (firmware_has_feature(FW_FEATURE_LPAR))
1184 tbl->it_ops = &iommu_table_lpar_multi_ops;
1185 else
1186 tbl->it_ops = &iommu_table_pseries_ops;
1187
1188 return iommu_init_table(tbl, -1, 0, 0);
1189}
1190
1191/**
1192 * vio_match_device: - Tell if a VIO device has a matching
1193 * VIO device id structure.
1194 * @ids: array of VIO device id structures to search in
1195 * @dev: the VIO device structure to match against
1196 *
1197 * Used by a driver to check whether a VIO device present in the
1198 * system is in its list of supported devices. Returns the matching
1199 * vio_device_id structure or NULL if there is no match.
1200 */
1201static const struct vio_device_id *vio_match_device(
1202 const struct vio_device_id *ids, const struct vio_dev *dev)
1203{
1204 while (ids->type[0] != '\0') {
1205 if ((strncmp(dev->type, ids->type, strlen(ids->type)) == 0) &&
1206 of_device_is_compatible(device: dev->dev.of_node,
1207 ids->compat))
1208 return ids;
1209 ids++;
1210 }
1211 return NULL;
1212}
1213
1214/*
1215 * Convert from struct device to struct vio_dev and pass to driver.
1216 * dev->driver has already been set by generic code because vio_bus_match
1217 * succeeded.
1218 */
1219static int vio_bus_probe(struct device *dev)
1220{
1221 struct vio_dev *viodev = to_vio_dev(dev);
1222 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1223 const struct vio_device_id *id;
1224 int error = -ENODEV;
1225
1226 if (!viodrv->probe)
1227 return error;
1228
1229 id = vio_match_device(ids: viodrv->id_table, dev: viodev);
1230 if (id) {
1231 memset(&viodev->cmo, 0, sizeof(viodev->cmo));
1232 if (firmware_has_feature(FW_FEATURE_CMO)) {
1233 error = vio_cmo_bus_probe(viodev);
1234 if (error)
1235 return error;
1236 }
1237 error = viodrv->probe(viodev, id);
1238 if (error && firmware_has_feature(FW_FEATURE_CMO))
1239 vio_cmo_bus_remove(viodev);
1240 }
1241
1242 return error;
1243}
1244
1245/* convert from struct device to struct vio_dev and pass to driver. */
1246static void vio_bus_remove(struct device *dev)
1247{
1248 struct vio_dev *viodev = to_vio_dev(dev);
1249 struct vio_driver *viodrv = to_vio_driver(dev->driver);
1250 struct device *devptr;
1251
1252 /*
1253 * Hold a reference to the device after the remove function is called
1254 * to allow for CMO accounting cleanup for the device.
1255 */
1256 devptr = get_device(dev);
1257
1258 if (viodrv->remove)
1259 viodrv->remove(viodev);
1260
1261 if (firmware_has_feature(FW_FEATURE_CMO))
1262 vio_cmo_bus_remove(viodev);
1263
1264 put_device(dev: devptr);
1265}
1266
1267static void vio_bus_shutdown(struct device *dev)
1268{
1269 struct vio_dev *viodev = to_vio_dev(dev);
1270 struct vio_driver *viodrv;
1271
1272 if (dev->driver) {
1273 viodrv = to_vio_driver(dev->driver);
1274 if (viodrv->shutdown)
1275 viodrv->shutdown(viodev);
1276 else if (kexec_in_progress)
1277 vio_bus_remove(dev);
1278 }
1279}
1280
1281/**
1282 * vio_register_driver: - Register a new vio driver
1283 * @viodrv: The vio_driver structure to be registered.
1284 */
1285int __vio_register_driver(struct vio_driver *viodrv, struct module *owner,
1286 const char *mod_name)
1287{
1288 // vio_bus_type is only initialised for pseries
1289 if (!machine_is(pseries))
1290 return -ENODEV;
1291
1292 pr_debug("%s: driver %s registering\n", __func__, viodrv->name);
1293
1294 /* fill in 'struct driver' fields */
1295 viodrv->driver.name = viodrv->name;
1296 viodrv->driver.pm = viodrv->pm;
1297 viodrv->driver.bus = &vio_bus_type;
1298 viodrv->driver.owner = owner;
1299 viodrv->driver.mod_name = mod_name;
1300
1301 return driver_register(drv: &viodrv->driver);
1302}
1303EXPORT_SYMBOL(__vio_register_driver);
1304
1305/**
1306 * vio_unregister_driver - Remove registration of vio driver.
1307 * @viodrv: The vio_driver struct to be removed form registration
1308 */
1309void vio_unregister_driver(struct vio_driver *viodrv)
1310{
1311 driver_unregister(drv: &viodrv->driver);
1312}
1313EXPORT_SYMBOL(vio_unregister_driver);
1314
1315/* vio_dev refcount hit 0 */
1316static void vio_dev_release(struct device *dev)
1317{
1318 struct iommu_table *tbl = get_iommu_table_base(dev);
1319
1320 if (tbl)
1321 iommu_tce_table_put(tbl);
1322 of_node_put(node: dev->of_node);
1323 kfree(objp: to_vio_dev(dev));
1324}
1325
1326/**
1327 * vio_register_device_node: - Register a new vio device.
1328 * @of_node: The OF node for this device.
1329 *
1330 * Creates and initializes a vio_dev structure from the data in
1331 * of_node and adds it to the list of virtual devices.
1332 * Returns a pointer to the created vio_dev or NULL if node has
1333 * NULL device_type or compatible fields.
1334 */
1335struct vio_dev *vio_register_device_node(struct device_node *of_node)
1336{
1337 struct vio_dev *viodev;
1338 struct device_node *parent_node;
1339 const __be32 *prop;
1340 enum vio_dev_family family;
1341
1342 /*
1343 * Determine if this node is a under the /vdevice node or under the
1344 * /ibm,platform-facilities node. This decides the device's family.
1345 */
1346 parent_node = of_get_parent(node: of_node);
1347 if (parent_node) {
1348 if (of_node_is_type(np: parent_node, type: "ibm,platform-facilities"))
1349 family = PFO;
1350 else if (of_node_is_type(np: parent_node, type: "vdevice"))
1351 family = VDEVICE;
1352 else {
1353 pr_warn("%s: parent(%pOF) of %pOFn not recognized.\n",
1354 __func__,
1355 parent_node,
1356 of_node);
1357 of_node_put(node: parent_node);
1358 return NULL;
1359 }
1360 of_node_put(node: parent_node);
1361 } else {
1362 pr_warn("%s: could not determine the parent of node %pOFn.\n",
1363 __func__, of_node);
1364 return NULL;
1365 }
1366
1367 if (family == PFO) {
1368 if (of_property_read_bool(np: of_node, propname: "interrupt-controller")) {
1369 pr_debug("%s: Skipping the interrupt controller %pOFn.\n",
1370 __func__, of_node);
1371 return NULL;
1372 }
1373 }
1374
1375 /* allocate a vio_dev for this node */
1376 viodev = kzalloc(sizeof(struct vio_dev), GFP_KERNEL);
1377 if (viodev == NULL) {
1378 pr_warn("%s: allocation failure for VIO device.\n", __func__);
1379 return NULL;
1380 }
1381
1382 /* we need the 'device_type' property, in order to match with drivers */
1383 viodev->family = family;
1384 if (viodev->family == VDEVICE) {
1385 unsigned int unit_address;
1386
1387 viodev->type = of_node_get_device_type(np: of_node);
1388 if (!viodev->type) {
1389 pr_warn("%s: node %pOFn is missing the 'device_type' "
1390 "property.\n", __func__, of_node);
1391 goto out;
1392 }
1393
1394 prop = of_get_property(node: of_node, name: "reg", NULL);
1395 if (prop == NULL) {
1396 pr_warn("%s: node %pOFn missing 'reg'\n",
1397 __func__, of_node);
1398 goto out;
1399 }
1400 unit_address = of_read_number(cell: prop, size: 1);
1401 dev_set_name(dev: &viodev->dev, name: "%x", unit_address);
1402 viodev->irq = irq_of_parse_and_map(node: of_node, index: 0);
1403 viodev->unit_address = unit_address;
1404 } else {
1405 /* PFO devices need their resource_id for submitting COP_OPs
1406 * This is an optional field for devices, but is required when
1407 * performing synchronous ops */
1408 prop = of_get_property(node: of_node, name: "ibm,resource-id", NULL);
1409 if (prop != NULL)
1410 viodev->resource_id = of_read_number(cell: prop, size: 1);
1411
1412 dev_set_name(dev: &viodev->dev, name: "%pOFn", of_node);
1413 viodev->type = dev_name(dev: &viodev->dev);
1414 viodev->irq = 0;
1415 }
1416
1417 viodev->name = of_node->name;
1418 viodev->dev.of_node = of_node_get(node: of_node);
1419
1420 set_dev_node(dev: &viodev->dev, node: of_node_to_nid(np: of_node));
1421
1422 /* init generic 'struct device' fields: */
1423 viodev->dev.parent = &vio_bus_device.dev;
1424 viodev->dev.bus = &vio_bus_type;
1425 viodev->dev.release = vio_dev_release;
1426
1427 if (of_property_present(np: viodev->dev.of_node, propname: "ibm,my-dma-window")) {
1428 if (firmware_has_feature(FW_FEATURE_CMO))
1429 vio_cmo_set_dma_ops(viodev);
1430 else
1431 set_dma_ops(dev: &viodev->dev, dma_ops: &dma_iommu_ops);
1432
1433 set_iommu_table_base(&viodev->dev,
1434 vio_build_iommu_table(dev: viodev));
1435
1436 /* needed to ensure proper operation of coherent allocations
1437 * later, in case driver doesn't set it explicitly */
1438 viodev->dev.coherent_dma_mask = DMA_BIT_MASK(64);
1439 viodev->dev.dma_mask = &viodev->dev.coherent_dma_mask;
1440 }
1441
1442 /* register with generic device framework */
1443 if (device_register(dev: &viodev->dev)) {
1444 printk(KERN_ERR "%s: failed to register device %s\n",
1445 __func__, dev_name(&viodev->dev));
1446 put_device(dev: &viodev->dev);
1447 return NULL;
1448 }
1449
1450 return viodev;
1451
1452out: /* Use this exit point for any return prior to device_register */
1453 kfree(objp: viodev);
1454
1455 return NULL;
1456}
1457EXPORT_SYMBOL(vio_register_device_node);
1458
1459/*
1460 * vio_bus_scan_for_devices - Scan OF and register each child device
1461 * @root_name - OF node name for the root of the subtree to search.
1462 * This must be non-NULL
1463 *
1464 * Starting from the root node provide, register the device node for
1465 * each child beneath the root.
1466 */
1467static void __init vio_bus_scan_register_devices(char *root_name)
1468{
1469 struct device_node *node_root, *node_child;
1470
1471 if (!root_name)
1472 return;
1473
1474 node_root = of_find_node_by_name(NULL, name: root_name);
1475 if (node_root) {
1476
1477 /*
1478 * Create struct vio_devices for each virtual device in
1479 * the device tree. Drivers will associate with them later.
1480 */
1481 node_child = of_get_next_child(node: node_root, NULL);
1482 while (node_child) {
1483 vio_register_device_node(node_child);
1484 node_child = of_get_next_child(node: node_root, prev: node_child);
1485 }
1486 of_node_put(node: node_root);
1487 }
1488}
1489
1490/**
1491 * vio_bus_init: - Initialize the virtual IO bus
1492 */
1493static int __init vio_bus_init(void)
1494{
1495 int err;
1496
1497 if (firmware_has_feature(FW_FEATURE_CMO))
1498 vio_cmo_sysfs_init();
1499
1500 err = bus_register(bus: &vio_bus_type);
1501 if (err) {
1502 printk(KERN_ERR "failed to register VIO bus\n");
1503 return err;
1504 }
1505
1506 /*
1507 * The fake parent of all vio devices, just to give us
1508 * a nice directory
1509 */
1510 err = device_register(dev: &vio_bus_device.dev);
1511 if (err) {
1512 printk(KERN_WARNING "%s: device_register returned %i\n",
1513 __func__, err);
1514 return err;
1515 }
1516
1517 if (firmware_has_feature(FW_FEATURE_CMO))
1518 vio_cmo_bus_init();
1519
1520 return 0;
1521}
1522machine_postcore_initcall(pseries, vio_bus_init);
1523
1524static int __init vio_device_init(void)
1525{
1526 vio_bus_scan_register_devices(root_name: "vdevice");
1527 vio_bus_scan_register_devices(root_name: "ibm,platform-facilities");
1528
1529 return 0;
1530}
1531machine_device_initcall(pseries, vio_device_init);
1532
1533static ssize_t name_show(struct device *dev,
1534 struct device_attribute *attr, char *buf)
1535{
1536 return sprintf(buf, fmt: "%s\n", to_vio_dev(dev)->name);
1537}
1538static DEVICE_ATTR_RO(name);
1539
1540static ssize_t devspec_show(struct device *dev,
1541 struct device_attribute *attr, char *buf)
1542{
1543 struct device_node *of_node = dev->of_node;
1544
1545 return sprintf(buf, fmt: "%pOF\n", of_node);
1546}
1547static DEVICE_ATTR_RO(devspec);
1548
1549static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1550 char *buf)
1551{
1552 const struct vio_dev *vio_dev = to_vio_dev(dev);
1553 struct device_node *dn;
1554 const char *cp;
1555
1556 dn = dev->of_node;
1557 if (!dn) {
1558 strcpy(p: buf, q: "\n");
1559 return strlen(buf);
1560 }
1561 cp = of_get_property(node: dn, name: "compatible", NULL);
1562 if (!cp) {
1563 strcpy(p: buf, q: "\n");
1564 return strlen(buf);
1565 }
1566
1567 return sprintf(buf, fmt: "vio:T%sS%s\n", vio_dev->type, cp);
1568}
1569static DEVICE_ATTR_RO(modalias);
1570
1571void vio_unregister_device(struct vio_dev *viodev)
1572{
1573 device_unregister(dev: &viodev->dev);
1574 if (viodev->family == VDEVICE)
1575 irq_dispose_mapping(virq: viodev->irq);
1576}
1577EXPORT_SYMBOL(vio_unregister_device);
1578
1579static int vio_bus_match(struct device *dev, struct device_driver *drv)
1580{
1581 const struct vio_dev *vio_dev = to_vio_dev(dev);
1582 struct vio_driver *vio_drv = to_vio_driver(drv);
1583 const struct vio_device_id *ids = vio_drv->id_table;
1584
1585 return (ids != NULL) && (vio_match_device(ids, dev: vio_dev) != NULL);
1586}
1587
1588static int vio_hotplug(const struct device *dev, struct kobj_uevent_env *env)
1589{
1590 const struct vio_dev *vio_dev = to_vio_dev(dev);
1591 const struct device_node *dn;
1592 const char *cp;
1593
1594 dn = dev->of_node;
1595 if (!dn)
1596 return -ENODEV;
1597 cp = of_get_property(node: dn, name: "compatible", NULL);
1598 if (!cp)
1599 return -ENODEV;
1600
1601 add_uevent_var(env, format: "MODALIAS=vio:T%sS%s", vio_dev->type, cp);
1602 return 0;
1603}
1604
1605#ifdef CONFIG_PPC_SMLPAR
1606static struct attribute *vio_cmo_dev_attrs[] = {
1607 &dev_attr_name.attr,
1608 &dev_attr_devspec.attr,
1609 &dev_attr_modalias.attr,
1610 &dev_attr_cmo_entitled.attr,
1611 &dev_attr_cmo_allocated.attr,
1612 &dev_attr_cmo_desired.attr,
1613 &dev_attr_cmo_allocs_failed.attr,
1614 NULL,
1615};
1616ATTRIBUTE_GROUPS(vio_cmo_dev);
1617
1618const struct bus_type vio_bus_type = {
1619 .name = "vio",
1620 .dev_groups = vio_cmo_dev_groups,
1621 .bus_groups = vio_bus_groups,
1622 .uevent = vio_hotplug,
1623 .match = vio_bus_match,
1624 .probe = vio_bus_probe,
1625 .remove = vio_bus_remove,
1626 .shutdown = vio_bus_shutdown,
1627};
1628#else /* CONFIG_PPC_SMLPAR */
1629static struct attribute *vio_dev_attrs[] = {
1630 &dev_attr_name.attr,
1631 &dev_attr_devspec.attr,
1632 &dev_attr_modalias.attr,
1633 NULL,
1634};
1635ATTRIBUTE_GROUPS(vio_dev);
1636
1637const struct bus_type vio_bus_type = {
1638 .name = "vio",
1639 .dev_groups = vio_dev_groups,
1640 .uevent = vio_hotplug,
1641 .match = vio_bus_match,
1642 .probe = vio_bus_probe,
1643 .remove = vio_bus_remove,
1644 .shutdown = vio_bus_shutdown,
1645};
1646#endif /* CONFIG_PPC_SMLPAR */
1647
1648/**
1649 * vio_get_attribute: - get attribute for virtual device
1650 * @vdev: The vio device to get property.
1651 * @which: The property/attribute to be extracted.
1652 * @length: Pointer to length of returned data size (unused if NULL).
1653 *
1654 * Calls prom.c's of_get_property() to return the value of the
1655 * attribute specified by @which
1656*/
1657const void *vio_get_attribute(struct vio_dev *vdev, char *which, int *length)
1658{
1659 return of_get_property(node: vdev->dev.of_node, name: which, lenp: length);
1660}
1661EXPORT_SYMBOL(vio_get_attribute);
1662
1663/* vio_find_name() - internal because only vio.c knows how we formatted the
1664 * kobject name
1665 */
1666static struct vio_dev *vio_find_name(const char *name)
1667{
1668 struct device *found;
1669
1670 found = bus_find_device_by_name(bus: &vio_bus_type, NULL, name);
1671 if (!found)
1672 return NULL;
1673
1674 return to_vio_dev(found);
1675}
1676
1677/**
1678 * vio_find_node - find an already-registered vio_dev
1679 * @vnode: device_node of the virtual device we're looking for
1680 *
1681 * Takes a reference to the embedded struct device which needs to be dropped
1682 * after use.
1683 */
1684struct vio_dev *vio_find_node(struct device_node *vnode)
1685{
1686 char kobj_name[20];
1687 struct device_node *vnode_parent;
1688
1689 vnode_parent = of_get_parent(node: vnode);
1690 if (!vnode_parent)
1691 return NULL;
1692
1693 /* construct the kobject name from the device node */
1694 if (of_node_is_type(np: vnode_parent, type: "vdevice")) {
1695 const __be32 *prop;
1696
1697 prop = of_get_property(node: vnode, name: "reg", NULL);
1698 if (!prop)
1699 goto out;
1700 snprintf(buf: kobj_name, size: sizeof(kobj_name), fmt: "%x",
1701 (uint32_t)of_read_number(cell: prop, size: 1));
1702 } else if (of_node_is_type(np: vnode_parent, type: "ibm,platform-facilities"))
1703 snprintf(buf: kobj_name, size: sizeof(kobj_name), fmt: "%pOFn", vnode);
1704 else
1705 goto out;
1706
1707 of_node_put(node: vnode_parent);
1708 return vio_find_name(name: kobj_name);
1709out:
1710 of_node_put(node: vnode_parent);
1711 return NULL;
1712}
1713EXPORT_SYMBOL(vio_find_node);
1714
1715int vio_enable_interrupts(struct vio_dev *dev)
1716{
1717 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_ENABLE);
1718 if (rc != H_SUCCESS)
1719 printk(KERN_ERR "vio: Error 0x%x enabling interrupts\n", rc);
1720 return rc;
1721}
1722EXPORT_SYMBOL(vio_enable_interrupts);
1723
1724int vio_disable_interrupts(struct vio_dev *dev)
1725{
1726 int rc = h_vio_signal(dev->unit_address, VIO_IRQ_DISABLE);
1727 if (rc != H_SUCCESS)
1728 printk(KERN_ERR "vio: Error 0x%x disabling interrupts\n", rc);
1729 return rc;
1730}
1731EXPORT_SYMBOL(vio_disable_interrupts);
1732
1733static int __init vio_init(void)
1734{
1735 dma_debug_add_bus(bus: &vio_bus_type);
1736 return 0;
1737}
1738machine_fs_initcall(pseries, vio_init);
1739

source code of linux/arch/powerpc/platforms/pseries/vio.c